VI. Construction of the fitted family and its cohomological inputs

Abstract. The height argument needs a single family at each cutoff which has both the required prime distribution and small cohomology groups. We trace how the choice of two auxiliary primes produces the field and how the localization kernels enter the error in the height. For ordinary zeta, the equivariant Coates–Sinnott theorem bounds the odd minus contribution by [E:Q]t2log⁡(tNE), including the ramified coefficient primes. For the odd plus contribution, class field theory eliminates two cases, and an integral estimate for the second algebraic K-group settles the bounded-degree cases at 3 and 7. In the three surviving cases the same two rational primes, 11 and 23, suffice for the numerical trace comparisons. Pan's theorem supplies the modular forms for the split rank-two blocks of Count I. The higher-rank constructions and the deformation-ring estimates enter the remaining comparisons.

Let 1 be the primitive Dirichlet character of conductor one. At each sufficiently late cutoff in a specified set of real cutoffs, the construction must supply geometric objects, periods and estimates for one host. Different cutoffs can use different hosts. The quantifier of interest is one family for each sufficiently small positive ε, with a single choice of the family constants before the cutoff varies. The following proof reconstructs the assembly from stated inputs. Proved here marks a prose deduction. Classical marks a named theorem used as input. Supplied by Papers II–V refers to the indicated interface of the other papers in this series, including any hypotheses carried by that interface. Release input marks a substantive assertion whose proof in the pinned release still needs a prose translation in this series. All source links are to commit e7c8972c701d3396947808ab20cc16dc14623d33. Family class and host requirement Quantifiers of the connected class

1. The theorem and its construction

The entry point riemannHypothesis calls rhOfAll, which applies the ordinary zeta reduction to allChi. The latter calls everyCharacter_of_row on the release's integral local–global Galois row. This row has two uses: it goes into the representation and reciprocity inputs for the geometric construction, and it goes into the cohomological-window construction via vwin_of_row. The final assembly specializes a method for all primitive characters. Its full window requirement is ∀ψ VWinAt(ψ,401). When tracing the ordinary specialization, only VWinAt(1,401) is needed by the host supplier for that character. Export Ordinary reduction from all characters Final conclusion Galois row construction Supply for a fixed character

Assembly theorem (proved here from the stated inputs). Fix the target ψ=1, put λ=3/2, and assume the pair and cohomological supply of §§2–3 and the object, representation and comparison inputs of §§4–5. There is an ε∗>0 such that for every 0<ε<ε∗ there exist a family F, constants K,cHH,eK,CL, and an onset x∗ with the following properties.

  1. Its cutoff set S⊂R contains arbitrarily late power intervals, in the sense that for every real u there is Z≥u with [Z,Z3/2]⊂S.
  2. Every x∈S with x≥x∗ has a nonempty finite index set Ix. All members at x have the scale x,ε,J and the same arithmetic host data specified in §4. They carry the host analytic and cohomological estimates of §3.
  3. For every such x the same F has ChainTied(F;2,0,K,cHH,eK,CL,x). This is the comparison data at x, with its field, ranks, weights and selected objects taken from F there. Its constants are independent of x. Window definition Actual shape call Ties at a cutoff

In the proof below e is a dyadic tower index, j is a positive cohomological twist, and ℓ is its coefficient prime. These are distinct from the two auxiliary rational primes q,d, from the canonical extra level prime q0, and from the cutoff x. The notation n denotes the order of the arithmetic frame, which also indexes the initial point choices. Window field and tower index Level prime and point indices

2. Choosing the primes, field and scale

Pair selection (proved for the ordinary character in Paper II). Given an integer M≥1, 0<ε<1/52, and A≥1, there is Y0 so that for every Y≥Y0 one can take primes q,d with (qd,M)=1, q≡3 mod 8, d≡7 mod 16, and 12z<log⁡q≤z,12z′<log⁡d≤z′,z=Yε/5,z′=z(log⁡Y)2. Here q∣d+1 and p∣d+1 for every odd prime p≤z. For each character modulo M, twist by one of Jq,Jd,JqJd, where Jr(a)=(a/r), and use the inducing primitive character ρ of the resulting function on units. The centered ramp and log derivative at one satisfy ∣gρ(y)∣≤y1/2+ε/2 and ∣L′(ρ,1)/L(ρ,1)∣≤4log⁡y for all Y≤y≤YA. A further assertion applies to real primitive bases of conductor dividing 8M. For each of the three nonempty twists the primitive L-function has no zero in ℜs>1−1/400, ∣ℑs∣≤8Mezejdz′, where jd is 1 when Jd occurs and 0 otherwise. At M=1 the ramp tests have one base, while the real box tests still include bases of conductor dividing 8. Definition of the pair and sizes Analytic conditions and quantifiers

The pair theorem selects q first and then d in a reduced class modulo 16q∏3≤p≤zp, with the product over primes. The inequality q>z makes the congruences compatible by the Chinese remainder theorem. Its estimates include a progression count, a zero-box discard and a log-derivative and centered-ramp discard using short character sums. For example, proved here from the counting bounds, if U1=ez, U2=ez′, a zero-box discard of size OM(U11/4) among ≥U19/10 first choices costs OM(U1−13/20) of those choices. The corresponding second discard has proportion OM(U2−13/20+1/(4(log⁡Y)2)) since U1=U21/(log⁡Y)2. The release also needs the short-sum discard on each stage and its estimates uniform over the relevant bases. Paper II, §2 derives these estimates at conductor one from the published progression and zero-density theorems. Paper I, §6 supplies the two required contour estimates. Prime selection · Contour estimates Two-stage assembly Combined analytic discards Prior density trace

Proved here. For odd p≤z the congruence d≡−1 mod p gives (−d/p)=1. At 2 the same splitting follows from −d≡1 mod 8. The congruence d≡−1 mod q also gives (−d/q)=1. For any fixed J≥1 and Y≥x1/J, the lower bounds on log⁡q,log⁡d eventually exceed Blog⁡x for each fixed B>0. This gives q,d>xB uniformly on these scales, including all fixed-prime exclusions and q≠3 after increasing the onset. Host congruences Size exclusions used by the supply

Let Ka=Q(ζ2a) and let Ka+ be its maximal real subfield. For the ordinary route the field specified by IsChiField(1,e,q,d,E) is, up to Q-isomorphism, E=Ke+2+(−q,−d). Proved here. In a cyclotomic field containing these three fields, fixing Ke+2+ is equivalent to a≡±1 mod 2e+2 for the cyclotomic automorphism with exponent a. Since q,d≡3 mod 4, the discriminants of the quadratic fields are −q,−d. Quadratic reciprocity, or the quadratic Gauss sum, gives their two fixing conditions Jq(a)=Jd(a)=1. The subgroup fixing a compositum is the intersection of its fixing subgroups. The fourth condition ψ(a)=1 in the general definition imposes no restriction for 1. Thus the field is CM, of degree 4⋅2e, since the two quadratic radicals are independent over Ke+2+ by their disjoint odd ramification. The tower conductor parameter QS in the host equals 2e+2 at conductor one, by the two inclusions of divisors in the frame construction. The frame order is n=c 2e, with 1≤c≤∣⟨ψ⟩∣ eventually. Hence c=1 for 1. In the general construction the integer factor is the eventual order of a descending sequence of subgroups of a finite character quotient. Fixing subgroup Tower factor and stabilization Host field construction

Proved here (rescaling a cohomological interval). For x≥1 put e(x)=⌈(2+8ε)log⁡2x⌉, so x2+8ε≤2e(x)≤2x2+8ε. Given the window input defined in §3, take J from that input, and take the pair theorem with A=J at Y=X1/J. It gives one q,d for all cutoffs up to X, and the cohomological input gives [W,W2]⊂[Y,X] for that pair. For the ordinary case c=1 use [W,W3/2]. In general set k=c1/(2+8ε) and use x∈[kW,(kW)3/2] with tower index e(x/k). Once W≥k3, this interval lies in [Y,W2] and x/k∈[W,W2]. Also c 2e(x/k) is in the required dyadic bracket for x2+8ε. Once W≥k400, (x/k)401≥x400. The bound exp⁡((x/k)ε/2) may be enlarged to exp⁡(xε/2). The output WinDatum at x consists of x1/J≤Y≤x, the analytic pair up to x, the bracket on n=c2e, and the cohomological estimate with twist limit ⌊x400⌋. Taking X arbitrarily large makes the starting point kW arbitrarily large. Dyadic bracket and monotonicity Window rescaling Output at one cutoff

3. Cohomological length and the choice of a common interval

For a number field E, a rational prime ℓ, and an integer j>0, let Sℓ be the places above ℓ and let GE,Sℓ be the Galois group unramified outside those finite places. Put Шℓ(E,j)=ker⁡ ⁣(H1(GE,Sℓ,(Qℓ/Zℓ)(−j))⟶∏v∈SℓH1(Ev,(Qℓ/Zℓ)(−j))). Here the twist (−j) uses the inverse j-th power of the cyclotomic character. Write λℓ(E,j) for its Zℓ-length, allowing +∞. The release's VVle(E,t,B) requires, for every choice of the closures and localization data in this definition, finite length for every ℓ and 1≤j<t, and ∑ℓ∈Plog⁡ℓ∑1≤j<tλℓ(E,j)≤Bfor every finite set of primes P. In particular an infinite length cannot be hidden by a convention for converting extended integers to ordinary integers. For a CM E the release also separates the odd-prime lengths into the two complex-conjugation parts. Coefficient and localization definitions Quantifiers of the length budget Combination of the parts

Release input (the window to be consumed). For every 0<ε<1/52 and every function e(x) in the dyadic bracket for all x≥1, there are J≥1 and X0 such that for every X≥X0 and every pair satisfying the size and congruence conditions of §2 at Y=X1/J, there is W≥Y with W2≤X. For all x∈[W,W2] and all E satisfying IsChiField(ψ,e(x),q,d,E), VVle(E,⌊x401⌋,exp⁡(xε/2)). The pair is fixed before W is selected. The onset and J are uniform in the pair. The cohomology result does not demand the zero boxes or ramp tests, so the pair chosen with those tests in §2 also qualifies. This is VWinAt(ψ,401). Window quantifiers Use of pair conditions in the window proof

3.1. Exceptional twists and uniformity

Classical. Class field theory describes the unramified abelian pro-ℓ module of the cyclotomic tower by norm limits of class groups. Ferrero–Washington gives μ=0 for the abelian base fields used here. Release input. The comparison in the release uses the quotient with the primes above ℓ split and identifies the length at layer m with a twisted cokernel on that module. With cyclotomic generator γ and its cyclotomic value v, the operator is (vjγ)ℓm−1. The release proves generator bounds and a length bound with at most R ℓ-adic centers for a module with at most R generators. At the moving dyadic base R is uniform in q,d once the target is fixed. For a fixed odd ℓ the base is the cyclotomic field of conductor ℓM with M=ψ.N. The degree, cyclotomic embedding, and layer comparison of the dyadic base are part of the input. In particular the comparison for a host of index e≥2 uses some m≤e+2 and adds at most one to the dyadic length. Cokernel descent Center and layer estimates Uniform rank and host comparison interfaces Ferrero–Washington in the construction

Proved here (finite exceptional set). For an integer cutoff T≥1, put b=⌊log⁡ℓT⌋. Take the exceptional integers 1≤j≤T to be those strictly closer than ℓ−b to a center. Distinct integers i,j in this range have vℓ(i−j)≤b, so the ultrametric inequality allows at most one integer per center. The center estimate bounds the other lengths by R(max⁡(m,⌊log⁡ℓR⌋+3)+c+b) with the cyclotomic constant c of the base. In the release one chooses T=⌊X401⌋ and unions the exceptional sets at 2 and the finitely many odd reference primes. The exceptional sets may vary with X,q,d. Their total cardinality is bounded by a single Rtot fixed before X,q,d. Exceptional-set estimate Union and order of choices

Release input (small exceptional twists). The pointwise estimate called LemmaK takes a totally imaginary field L containing a primitive ℓ-th root of unity for odd ℓ or a primitive fourth root of unity for ℓ=2, with local data as above, and any j≥1. For some absolute CK, H2(GL,Sℓ,Zℓ(j+1)) is finite and the log of its order is at most exp⁡ ⁣((2j[L:Q])CKlog⁡(2∣disc⁡L∣)). The release applies this at a bounded reference layer and uses its second layer comparison for the remaining growth. The discriminant estimate at that layer, the local and global duality estimates, and LemmaK are inputs here. Together the release obtains, on cutoffs satisfying j≤xη for each exceptional j≤x401, where η=ε/(2CK), individual length bounds of size C(1+log⁡x)+2exp⁡(Cx3ε/4) (with the host comparison absorbed into constants). Multiplying by at most x401 still gives an eventual bound exp⁡(x4ε/5) at each tower comparison, since 3ε/4<4ε/5. The cutoff j≤xη permits the use of the discriminant bound in LemmaK. Large exceptional twists must be avoided in selecting the interval. Statement of the H2 estimate Two layer bounds and pointwise target Dyadic application

Proved here (common interval). Take CK≥1, Ct=401, and try the Rtot+1 starts Wa=X1/[2(2Ct/η)a],a=0,…,Rtot. A start can fail only if the union of exceptional sets meets (Waη,Wa2Ct]. Indeed for x∈[Wa,Wa2], a forbidden integer has xη<j≤xCt. The bands are pairwise disjoint since Wa+12Ct=Waη and the lower endpoint is open. At least one start therefore works on its entire interval. All starts satisfy Wa2≤X and Wa≥X1/J for J=2Rtot+2(Ct/η)Rtot. With Y=X1/J, the size bounds on q,d give log⁡(qd)≪Yε/5(1+log⁡2Y)≤Yε/4 eventually. Also q,d>2X401+1 eventually for fixed J. Thus the growth and size conditions apply uniformly for the one selected interval. General interval lemma Uniform window exponent Pair size hypotheses in the comparison

3.2. What the automorphic representations supply

The phrase “Vandiver-like” in this route refers to the finite-length and height estimates for positive twists. The window proof uses quantitative estimates and allowances for exceptional positive integers. Release input. Its pointwise odd-prime plus bound PointHV has the following form. Fix ψ and the reference layer m(ℓ) at each ℓ∈Bψ. There are a polynomial exponent C and fixed primes r1,r2 such that for all qualifying hosts, t≥2, and all B≥0, the bound VVat(ℓ,t,B) on those reference layers gives VVplusLe(E,t,(cond⁡(E)t)C+CB). Here the host has the prescribed character, both q,d exceed r1,r2,2t+1 and the fixed data in the definition of Host.Big, and E is the CM field of that host. VVat is finiteness and a sum of lengths at one prime. The minus bound ThmBcFree applies to all these CM composita for e,t≥2 with the prime-pair conditions. Section 3.4 proves the conductor-one budget and traces its character estimate to the reciprocity and Gauss-sum inputs. Thus the reference-layer bound at a finite set of primes is just one input to a bound over all coefficient primes of the host. Pointwise plus input Host size and conductor conversion Minus input Odd-residue assembly

The pointwise plus bound uses character summands of H2 over Q. The characters are those of the host field with the parity prescribed by j. The following are release inputs. For fixed target, CountI bounds the weighted total at odd ℓ∉Bψ with ℓ−1∤2j by a fixed power of the host conductor parameter times t, under the two fixed-prime exclusions above. CountII and CountIII give individual polynomial bounds in ℓNj, with N the host modulus, for the case EV(ℓ,ρ) at j≥2 under the case-II divisibility conditions, and for nontrivial odd-order ρ outside Bψ with ℓ−1∣2j, respectively. The EV condition requires even order of χP(ℓ,ρ), the factor of the prime-to-ℓ part left after removing its ℓ-conductor component. CountIV treats odd-order prime-to-ℓ part at ℓ∈Bψ using the reference layer. CountV treats j=1 at the reference primes with R1(ℓ,ρ)≤2c2(ρ), and CountVs treats the complementary bound using duality and LemmaK. The trivial-character outside case, with ℓ−1∣2j, uses CountZ from the release's vanishing input PropZ. These conditions concern characters of the host, which continue to vary even for target 1. Character cases Bounds and hypotheses Count IV Assembly into the pointwise estimate

In the exported route, the just-mentioned bounds are obtained through the following representation inputs. Release input (automorphic and Galois construction). The integral Galois row and isobaric base change feed the package LiteratureRankNHypothesis. The package uses local GLn segment classification and irreducibility, depth-zero induction, local converse, regular algebraic purity, Galois compatibility and CM coefficient and period comparisons. The patching step produces HypothesesH′, using descWeak for the weak descent input. From these hypotheses the proof constructs unramified automorphy ends and a reduced-unitary end, and uses the Kato two-place input under the unramified hypotheses for CountI. The Sp4In0 comparison gives CountII and CountIII, while the reduced unitary lifting bound gives CountV. The same package feeds lemmaK_litN_alt. These assertions are required in the applicable conductors, weights, and local types of the count statements. Classical base change and local representation theorems occur in the package, but their names alone do not supply the patching, descent, Kato or quantitative conclusions. Local, automorphic and CM interfaces Patching, the two ends and the four counts Duality, towers, odd minus estimate and assembly

Proved here (size of the combined budget given the inputs). With B=exp⁡(x4ε/5), the dyadic contribution and reference-prime allowance have this exponential scale. The fixed-conductor estimate cond⁡(E)∣lcm⁡(2e+2,M)qd and t≤x401 give log⁡(cond⁡(E)t)=O(xε/4+log⁡x) on the selected interval. Thus the polynomial term of PointHV has log O(xε/4+log⁡x). The release's minus-bound comparison gives an allowance exp⁡(K′xε/4) under these sizes for fixed target and K′ independent of x,q,d. Each of these allowances and their sum are o(exp⁡(xε/2)). All terms are nonnegative, so the sum bounds every finite set of primes with the required finiteness. This explains the slack in the output VVle, and the smaller twist limit ⌊x400⌋ survives rescaling as in §2. Conductor, height and sum of parts Rescaled limit

3.3. Conductor-one tower checks

For 1, pK=7, h(Q(−7))=1, p1=3, and B1={3,7}. The two odd reference bases are Q(ζ3) and Q(ζ7). Proved here using classical Minkowski and Chevalley. The first base is the Eisenstein field. For the second the Minkowski bound is less than 5, while the residue degrees of 2,3 are 3,6. This gives no prime ideal of norm at most the bound. The same bound for Q(−7) is less than 2. At each degree-ℓ step of either odd tower, exactly one prime ramifies and is totally ramified. Chevalley's ambiguous class number formula gives that the fixed class group has order prime to ℓ if this holds at the preceding layer. A nontrivial finite ℓ-group of classes under a cyclic action of order ℓ has at least ℓ fixed elements, by counting orbits. Induction gives zero ℓ-primary groups throughout these reference towers. The reference bounds can therefore use zero lengths. The plus bound for the host still has its separate polynomial term. Definition of the reference primes Reference fields and pointwise host estimate Chevalley's formula

Proved here (character cases at conductor one). The host Galois group in §2 is a 2-group. Every character in the odd-prime counts therefore has 2-power order, and case III, which asks for a nontrivial character of odd order, is empty. At ℓ∈{3,7} and on sufficiently late windows, ℓ is coprime to the host modulus QS qd. There χP(ℓ,ρ)=ρ, so case IV only uses the trivial character. The even-order condition of case II can still hold, for instance for a nontrivial character at an odd prime coprime to the host modulus. Section 3.4 handles the minus eigenspace. Character factors Host modulus and cases Minus bound

Proposition (dyadic bound for the favorable pair). Let q,d be the pair of Paper II, Proposition 2, so q≡3 mod 8, d≡7 mod 16, and v2h(−4qd)=v2h(−8qd)=3. Here h(D) is the quadratic class number of discriminant D. For the host E=Ke+2+(−q,−d) of degree 4⋅2e, every e≥0 and j≥1 satisfy λ2(E,j)≤3e+4+v2(j). Paper II, Proposition 2 · Full dyadic proof

Write Sa=Ka(qd), La=Ka(−q,−d), and XS,XL for their ordinary 2-class groups in the inverse limit under norms. The base degrees are [S2:Q]=4 and [L2:Q]=8. At finite layers Chevalley's formula bounds the 2-rank of Sa by two. The base Ka has odd class number, and for a≥4 the six primes ramified in Sa/Ka come in three conjugate pairs. The unit 1+ζ2a+ζ2a−1, a conjugate interchanging the pairs above d, and ζ2a give independent residue sign vectors (1,1,0),(1,0,1),(1,1,1) on those pairs. The same rank bound follows at a=2,3 from the smaller ramification count and, at a=3, the nonsquare residue of ζ8 above q. The unramified quadratic extension La/Sa gives the bound three for La by Schreier's inequality on the unramified pro-2 groups. These generator bounds pass to the norm limits. Chems-Eddin and Müller, Theorems 4 and 5, give λ−(S2)=2 and odd class numbers for the real tower. Ferrero–Washington and the Iwasawa structure theorem then give rank two for XS.

For k=Q(−d) the two local unit directions at 2 give a Z22-extension unramified outside 2. Each inertia direction maps injectively to the cyclotomic direction, so over the cyclotomic Z2-extension of k the remaining Z2-extension is unramified. Base change gives a rank-one quotient of XL with trivial cyclotomic action. The involution of L∞/S∞ exchanges the local unit directions of k and acts by −1 on this quotient. Norm and extension of ideals identify the +1 rational subspace of XL with XS⊗Q2, so this subspace is the rational kernel of the quotient. Thus the integral generator bounds are attained by the free ranks, and XS≃Z22, XL≃Z23. Finite-layer ranks and the norm maps · Chems-Eddin and Müller, Theorems 4 and 5 · Ferrero–Washington

Take the cyclotomic generator γ of value 5, and let M be its action on XS. The unique prime above 2 in Sa is totally ramified up the tower. Killing its inertia after abelianizing over Sa gives the ordinary 2-class group XS/(M2a−2−1)XS. In layers 2,3, Hasse unit index two would give a unit u with u/uˉ=−ζ2a after multiplication by a root of unity. Then u/(1−ζ2a) would be real with odd dyadic valuation, whereas real elements have even valuation in Sa. The Hasse indices are therefore one. The relative class number formula gives h−(S2)=h(−4qd)/2 and h−(S3)=h(−4qd)h(−8qd)/4. The real class numbers are odd. The two coinvariant orders give the valuation of det⁡(M−I) and the valuation of det⁡(M2−I), so the favorable valuations give v2det⁡(M−I)=v2det⁡(M+I)=2. Put s=tr⁡(M−I), p=det⁡(M−I). From det⁡(M+I)=4+2s+p both s/2 and p/4 are odd. The characteristic polynomial of (M−I)/2 is therefore integral and reduces to Z2+Z+1. Each eigenvalue of M has the form 1+2z with residue zˉ∈F4∖F2. For b=5j(1+2z), the two quantities b−1,b+1 both have valuation one, since 5j≡1 mod 4. The first squaring gives v2(b2−1)=2, and subsequent squarings add one each. Thus the two eigenvalues contribute 2(e+1) to the valuation of the determinant of Tj,e=(5jγ)2e−1. The trivial cyclotomic direction contributes v2(5j2e−1)=e+2+v2(j). On the free lattice XL the sum is the cokernel length 3e+4+v2(j). Low-layer class numbers and determinant calculation

For localization, each of E and Le+2=E(i) has two dyadic primes. Locally −q generates the unramified quadratic extension and −d is already in Q2. Each dyadic prime of E has a unique prime above it, totally ramified of degree two, in Le+2. At either prime the local cyclotomic image equals the global image, namely {±1}(1+2e+2Z2) for E and 1+2e+2Z2 for Le+2. At every finite step Lb/La, b≥a≥2, the index 2b−a is also the local degree and the ramification index at each of the two primes. The local and global cyclotomic quotients above Le+2 thus identify with the same procyclic group generated topologically by γ2e. In degree one its cohomology is zero, since multiplication by 5−j2e−1 on the divisible coefficients is onto. Its degree-two cohomology is zero as well. Inflation–restriction therefore identifies the strict group on Le+2 with invariant characters of the quotient XP of XL that imposes splitting at 2. Odd inertia already lies in the infinite-level group, and at 2 the local degree-one vanishing makes the split condition equivalent to the original local vanishing. Dualizing gives XP/Tj,eXP, an integral quotient of the ordinary cokernel, including any finite torsion of XP.

For Le+2/E the global and each dyadic local quotient have order two and the same fixed coefficient group B=(Q2/Z2)[2e+2+v2(j)], with involution acting by (−1)j. The kernel of restriction before localization is H1(C2,B), of order two. Its localization to either prime is the same inflation map and is injective. Thus restriction on the strict group has zero kernel. For a strict invariant class upstairs the global transgression vanishes by comparison with either local transgression. Its lift downstairs has two local obstructions in H1(C2,B), defined up to simultaneous translation. Their difference gives the cokernel into the invariants as a subgroup of H1(C2,B). This proves the bound with layer m=e and removes the +1 allowance in the general restriction estimate of §3.1. Exact localization comparison · Quadratic restriction sequence

The sum for 1≤j<t is at most (t−1)(3e+4)+v2((t−1)!), hence O(t(1+log⁡x)) when e=O(log⁡x). The favorable choice bypasses the exceptional-twist bound at 2. With Y=X1/J, J≥2 fixed, the same analytically admissible pair supplies the dyadic and odd reference-layer inputs on [X,X] at all sufficiently large X. The pointwise odd-prime estimates for the host supply the remaining terms in the budget. Prime selection with the prescribed sign

3.3a. Vanishing in the trivial and quadratic residual cases

One count in §3.2 needs an exact zero for the trivial host character. It has an elementary class-field proof. Fix a prime ℓ≥5 and an even integer j≥2 with ℓ−1∣2j. Let O be the integers in a finite extension K/Qℓ and let A=K/O, with action α=κℓ−j of GQ, where κℓ is the cyclotomic character. Let S consist of the continuous cohomology classes in H1(Q,A) whose restriction to each inertia group away from ℓ and to the decomposition group at ℓ is zero. Then S=0. In the release this is PropZ for every allowed coefficient ring. Definition of the strict group · Statement of the vanishing

Proof. The image of each continuous cocycle in the discrete group A is finite because GQ is compact. Every cohomology class is consequently killed by a power of ℓ. It suffices to show that every s∈S[ℓ] is zero. Take a cocycle z for s. Write ℓz=∂a and choose a′ with ℓa′=a, using divisibility of A. Replacing z by z−∂a′ puts all its values in A[ℓ]. A coboundary is ∂b(g)=(α(g)−1)b. For inertia away from ℓ the full cyclotomic action is trivial, so the strict restriction forces z itself to vanish there.

If ℓ−1∣j, the action on A[ℓ] is trivial, so z is a continuous homomorphism. Its image cuts out a finite abelian ℓ-extension of Q unramified outside ℓ. By Kronecker–Weber it factors through Q(μℓ∞). Indeed the conductor has no prime divisor away from ℓ; the extension has odd degree and hence has no ramification at infinity. The action on A factors through the same cyclotomic extension, which is totally ramified at ℓ. Its full Galois group is therefore the image of the decomposition group at ℓ. On that decomposition group z=∂b for some b∈A by the strict local condition. Both sides factor through the same cyclotomic quotient, so the equality holds globally. Trivial residual character in the release · Class field theory

Otherwise j≡(ℓ−1)/2(modℓ−1), and evenness of j gives ℓ≡1(mod4). On the additive ℓ-torsion A[ℓ], the action is the quadratic character associated to k=Q(ℓ) with discriminant ℓ. In fact the action factors through (Z/ℓ)× and is its unique character of order two. This remains valid for ramified O, since an integer divisible by ℓ acts as zero on A[ℓ]. At every place above ℓ a decomposition group surjects onto (Z/ℓ)×, so contains h with α(h)−1 an O-unit. Write the local cocycle as z=∂b in A. Evaluating at h shows b∈A[ℓ], since ℓz(h)=0 and α(h)−1 is a unit. It follows that on restriction to Gk the local cocycle vanishes at each such place. The global restriction z∣Gk is a homomorphism to A[ℓ] unramified at every finite place. It therefore factors through the class group of k on the odd-primary part.

By Minkowski’s theorem every ideal class of this real quadratic field contains an integral ideal of norm at most ℓ. For each n the number of ideals of norm n is at most the divisor function d(n), by prime decomposition in a quadratic field. With v=ℓ>1, we have h(k)≤∑n≤vd(n)≤v(1+log⁡v)<v2=ℓ. The narrow and ordinary class numbers differ only by a power of two. Hence z∣Gk=0. Now z descends to a cocycle on the order-two group Gal(k/Q) with values in A[ℓ]. Since multiplication by two is invertible on A[ℓ], averaging on this finite group makes that cocycle a coboundary. This finishes the proof. Quadratic class-number estimate · Quadratic case and its use

3.4. The conductor-one odd minus bound

Proposition (odd minus bound for the ordinary host). Let e,t≥2 be integers and let q,d be primes with q≡3 mod 8 and d≡7 mod 16. Let E=Q(ζ2e+2)+(−q,−d), NE=2e+2qd and D=[E:Q]=2e+2. The conductor calculation is Paper II, Proposition 1. It only uses the coprime moduli and the indicated discriminants, so the bound below covers all of these pairs, including the ones with q∣d+1 used in the final host construction. For every odd ℓ and 1≤j<t, the minus length λℓ−(E,j) is finite. For every finite set P of odd primes, ∑ℓ∈Plog⁡ℓ∑1≤j<tλℓ−(E,j)≤Dt2log⁡(tNE).(VI.1) Both assertions hold for every permitted choice of algebraic closure and local data. For fixed E,t only finitely many odd primes have a nonzero term. In particular this gives the budget in the conductor-one specialization of ThmBcFree. Paper II, Proposition 1 Exact interface Budget and domain General statement

3.4.1. Complex conjugation, coefficient signs and descent

Classical alternative for ordinary zeta. For the conductor-one field above, Poitou–Tate duality, Greither–Popescu's equivariant Coates–Sinnott theorem and Ferrero–Washington give an unconditional proof of (VI.1). This applies for all the indicated e,t,q,d and every odd coefficient prime, including q,d. Section 3.4.3 records the released derivation and the inputs it uses.

Let c be complex conjugation on E, and fix a complex-conjugation lift c~ to an algebraic closure Ω⊂C. On ordinary conductor-one Artin coordinates of Paper II it has bits (1,1) and trivial dyadic coordinate. The inertia involution at d, with bits (1,0) in Paper II's order, acts on just −d. The cohomological projection uses c. On cocycles write Cargz(g)=z(c~−1gc~) and Ctwz(g)=κℓ(c~)−jz(c~−1gc~), where κℓ is the cyclotomic character. These are the argument and coefficient-twisted involutions. Conjugation preserves the Galois quotient unramified outside ℓ and permutes the local conditions. It is compatible with the cocycle relation because κℓ is invariant under conjugation. Since c~ inverts roots of unity, Ctw=(−1)jCarg. Thus Шℓ−(E,j) is the −1 eigenspace for Carg and the (−1)j+1 eigenspace for Ctw. The plus signs are +1 and (−1)j, respectively. At odd ℓ the two projections exist over Zℓ by division by 2. Write λℓ−(E,j)=length⁡ZℓШℓ−(E,j). Artin coordinates Conjugation on cocycles Argument sign modulo a coboundary Eigenspaces and addition of lengths

Let G=Gal⁡(E/Q) and choose the integers O of a finite unramified extension of Qℓ containing the values of every character of G. Such an O exists because ∣G∣ is prime to ℓ and roots of unity of order prime to ℓ occur in unramified extensions. Let KO=Frac⁡(O). Flat base change commutes with the localization kernel and preserves length with this choice of O. Indeed O is free of finite rank over Zℓ, so continuous cochains commute with base change by using a finite basis, and O/ℓO has O-length one. In the release's general choice of O, lengths multiply by vO(ℓ). Since ∣G∣ is a unit, the projectors ∣G∣−1∑g∈Gψ(g)−1g split the scalar-extended module into character summands. Precisely those with ψ(c)=(−1)j+1 occur in the minus part. The resulting full sum has D/2 characters, because c≠1 and evaluation at c is onto {±1} on the character group. The release bounds this count by D. Scalar extension and eigenspace comparison Character family at one prime

Proved here using classical restriction and corestriction. Inflate each ψ to GQ, and write χ=ψ∘ρ on the units modulo NE, where ρ is the cyclotomic quotient onto G. The ψ-summand equals the strict Selmer group Sℓ(α) over Q of KO/O with character α=ψ−1κℓ−j. Here “strict” means zero restriction to inertia at every finite prime and to the full decomposition group at ℓ. The exponent −j is the original coefficient twist, and the factor ψ−1 cancels the ψ-action when taking invariants. The positive-degree cohomology of G on ℓ-primary modules vanishes by averaging. Inflation–restriction identifies the invariant classes after this twist with the rational classes. The local restrictions to the inertia groups and the decomposition group above ℓ detect the required vanishing downstairs: their subgroups over E have indices dividing D, and restriction is injective in degree one because corestriction multiplies by that index. This proves the descent with the indicated local conditions. Its Tate dual character is β=κℓα−1=ψκℓj+1. In particular α(c~)=−1, while β(c~)=1. The Bernoulli index in the bound is k=j+1 and the Bernoulli character is χ, with the same parity as k. Strict local family Character twists in the release Eigenspace descent Cyclotomic and Dirichlet character identification

Strict duality. Let Sℓ,E consist of the places of E over ℓ and its archimedean places. Put A=(Qℓ/Zℓ)(−j) and T=Zℓ(k) with k=j+1. The finite Tate dual of A[ℓm] is T/ℓmT. Finite-coefficient Poitou–Tate pairs the degree-one localization kernel for A[ℓm] with the degree-two kernel for T/ℓmT. The archimedean terms in these degrees vanish. Filtered colimits commute with the finite localization diagram, and the finite-coefficient global and local cohomology groups are finite, so the inverse systems satisfy Mittag–Leffler. Taking limits gives the G-equivariant identification Шℓ(E,j)∨≃ШSℓ,E2(E,T)=ker⁡ ⁣(H2(GE,Sℓ,E,T)⟶∏w∣ℓH2(Ew,T)), where the Pontryagin dual has the contragredient action. The identification is natural under transport of closures and localization data. In particular the character ψ on the strict group corresponds to η=ψ−1 on its dual, with η(c)=ψ(c)=(−1)k for the minus part. Milne, Arithmetic Duality Theorems, I.4.10(a) and I.5.2(b)

Put U0=Spec⁡OE[1/ℓ]. The comparison map H2(GE,Sℓ,E,T)→Het2(U0,T) is injective. For T/ℓmT this is the degree-two edge injection of the Cartan–Leray spectral sequence over finite étale covers of U0. The direct limit of their degree-one cohomology vanishes, since each class is a torsor for a finite locally constant sheaf and trivializes on a finite étale cover. The étale fundamental group is GE,Sℓ,E by the unramified extension correspondence. The finite-coefficient groups in degree one on both sides are finite, so passage to ℓ-adic coefficients gives the injection by inverse limits. We therefore have length⁡OSℓ(ψ−1κℓ−j)≤length⁡O(Het2(U0,T)⊗ZℓO)ψ−1. Here subscripts denote character components. The ψ-part of the strict group and the ψ−1-part of its dual have the same length once finite. Finite locally constant sheaves and the fundamental group · Geisser–Schmidt, Poitou–Tate duality for arithmetic schemes, Theorem B

3.4.2. A single integer controls the weighted sum

Put Πk=∏p≤k+1p, with p prime. For a character χ of G pulled back to the units modulo NE, write χ∗ for its primitive character of conductor fχ∣NE, extended by zero at nonunits. For k=j+1 define Bk,χ∗=fχ k−1∑u=1fχχ∗(u)Bk(u/fχ),ANE,k(χ)=NEΠkBk,χ∗,b(NE,k)=16NE k+1k!. The common character estimate is length⁡OSℓ(ψ−1κℓ−j)≤vO(ANE,k(χ)).(VI.2) Here vO counts powers of a uniformizer. In our unramified O, vO agrees with the valuation vℓ on the chosen embedding into Q‾ℓ, normalized by vℓ(ℓ)=1. The release calls the Bernoulli factor alN. The construction of ThmBcFree takes the finite weighted sum from PrimeBound, which uses PsiBound at each character. The entry ML.iv applies vChar with the reciprocity input. Normalizations Character estimate Length to norm normalization Construction of the two wrappers

Coates–Sinnott and the torsion correction. Apply Greither–Popescu, An Equivariant Main Conjecture in Iwasawa Theory and Applications, J. Algebraic Geom. 24 (2015), 629–692, Theorem 6.11, to E/Q, S={∞,2,q,d,ℓ}, p=ℓ>2 and n=k≥2. The set S contains the ramified, infinite and ℓ-adic places. The field E(μℓ) is abelian over Q and is already CM, so Ferrero–Washington gives the required μE(μℓ)CM,ℓ=0 for every odd ℓ. In particular the theorem applies when ℓ∣NE. Write U=Spec⁡OE[1/(NEℓ)], W=H0(E,(Qℓ/Zℓ)(k)) and ek=(1+(−1)kc)/2. Lemma 6.9(1) identifies W with Het1(U,T)tors, and Het2(U,T) is finite by §6.2 of the same paper. The theorem gives ekFitt⁡Zℓ[G],0Het2(U,T)=Ann⁡Zℓ[G](W) ΘS,E/Q(1−k). In their convention ΘS(1−k) evaluated at η is LS(1−k,η−1). Thus η=ψ−1 from duality gives LS(1−k,ψ). Greither–Popescu, definition of ΘS · Lemma 6.9 and Theorem 6.11 · Published paper · Ferrero–Washington, Ann. of Math. 109 (1979), 377–395

The invariant group W is cyclic of finite order at most ℓvℓ(kNEΠk). To see this, set s=vℓ(NE). Since E⊆Q(μNE), the image of GE on the ℓ-power roots of unity contains 1+ℓmax⁡(1,s)Zℓ. Any element of this subgroup lifts to a cyclotomic automorphism with other components 1 that fixes Q(μNE). The element 1+ℓmax⁡(1,s) acts on the twist by its k-th power. For odd ℓ the binomial formula gives vℓ((1+ℓmax⁡(1,s))k−1)=max⁡(1,s)+vℓ(k). Thus W is contained in the cyclic torsion layer of that exponent, which proves the bound for s≥1. If s=0, the same lifting argument gives the full image Zℓ×. If ℓ−1∤k, a lift of a generator modulo ℓ has k-th power differing from 1 by a unit, so W=0. Otherwise ℓ≤k+1, and Πk supplies the extra power of ℓ.

Fitting ideals commute with base change to each factor O of O[G]. The annihilator of W does as well, by flat base change to the splitting ring followed by its direct-factor projections. In the η-component it is (ℓaℓ,η), where aℓ,η=length⁡O(W⊗ZℓO)η≤vℓ(kNEΠk), since each component is cyclic. For a finite module over O, its zeroth Fitting ideal is generated by ℓ to the power of its length, as follows by decomposing into cyclic modules. For η=ψ−1 with η(c)=(−1)k, we obtain length⁡O(Het2(U,T)⊗ZℓO)η=aℓ,η+vℓ(LS(1−k,ψ)).

The removed Euler factors. For f=fχ, the function LS(1−k,ψ) is the primitive L(1−k,χ∗) multiplied by 1−χ∗(r)rk−1 at every finite r∈S with r∤f. Each such factor is nonzero since ∣χ∗(r)rk−1∣>1. At r∣f the primitive Euler factor is already 1. At r=ℓ the correcting factor, if present, is a unit since k≥2. For the other removed primes, étale localization gives an exact sequence of G-modules 0⟶Het2(U0,T)⟶Het2(U,T)⟶⨁r∣NEr≠ℓ ⨁w∣rH1(κ(w),Zℓ(k−1))⟶0. Indeed, U0 is regular, the complement of U is a disjoint union of regular closed points of codimension one, and ℓ is invertible on U0. Gabber's absolute purity theorem identifies the supported term at w with residue-field cohomology shifted by two degrees and twisted down by one, first for T/ℓmT. All the finite-coefficient groups in this localization sequence are finite, so inverse limits are exact here and compute continuous ℓ-adic cohomology. At the left, H0(κ(w),Zℓ(k−1))=0 since arithmetic Frobenius acts by ∣κ(w)∣k−1≠1. At the right, Het3(U0,T)=0 by the relative-dimension-zero case of Geisser–Schmidt (0.4) for U0 over itself, with the set of places Sℓ,E and each coefficient T/ℓmT. All infinite places are complex and their degree-three groups vanish. Finiteness of the degree-two groups removes the lim←⁡1 term when passing to T. Fujiwara, A Proof of the Absolute Purity Conjecture (after Gabber), Adv. Stud. Pure Math. 36 (2002), 153–183 · Geisser–Schmidt, Theorem B and (0.4)

For w∣r, write Dw,Iw⊂G for the decomposition and inertia groups, Fr∈Dw/Iw for arithmetic Frobenius and fw=[κ(w):Fr]. Procyclic cohomology identifies the residue term at w with Zℓ/(rfw(k−1)−1). Inertia acts trivially and Fr acts on this quotient by rk−1, from its action on the twist and the trivial conjugation action on the absolute Galois group of the residue field. The sum over w∣r is induced from Dw. The η-component vanishes if η∣Iw≠1, which is precisely r∣f. Otherwise projection imposes rk−1=η(Fr), giving O/(rk−1−η(Fr)) since η(Fr)fw=1. Its length is vℓ(1−χ∗(r)rk−1), as χ∗(r)=η(Fr)−1. This accounts for the factors at 2,q,d even when the ambient field ramifies and the character does not. If ℓ=q or d, that prime is already inverted in U0 and its Euler factor was dealt with by the unit calculation at ℓ.

Subtracting the residue lengths in the exact sequence gives length⁡O(Het2(U0,T)⊗ZℓO)ψ−1=aℓ,ψ−1+vℓ(L(1−k,χ∗)). The primitive value is L(1−k,χ∗)=−Bk,χ∗/k by the Hurwitz-zeta formula, including the zeta case f=1. In the matching parity χ∗(−1)=(−1)k, the functional equation relates it by a nonzero factor to L(k,χ∗‾), which is nonzero by the Euler product. The torsion bound pays for the division by k, giving aℓ,ψ−1+vℓ(L(1−k,χ∗))≤vℓ(NEΠkBk,χ∗). This proves (VI.2) by §3.4.1. The primitive L-value can have negative valuation, with the torsion correction supplying the integral Fitting generator. Hurwitz expansion and Dirichlet functional equation · Hurwitz special value

Proved here using classical Bernoulli identities. Each ANE,k(χ) is integral. Expand fχk−1Bk(u/fχ) as ∑i=0k(ki)Bifχi−1uk−i. The factor NE clears the i=0 denominator, and Πk clears all denominators of Bi by the von Staudt–Clausen theorem, B1=−1/2, and vanishing of the odd Bi for i>1. The values of the character are integral as roots of unity or zero. The preceding nonvanishing argument applies under every embedding, since it preserves the parity. Von Staudt–Clausen Release denominator calculation Release nonvanishing

The classical Fourier series of Bk(u) for k≥2 gives sup⁡0≤u≤1∣Bk(u)∣≤2k!ζ(k)/(2π)k≤4k!/(2π)k. Therefore ∣Bk,χ∗∣≤4fχkk!/(2π)k. The elementary inequality ∏p≤mp≤4m follows by induction. For odd m=2u+1≥3, the primes in (u+1,2u+1] divide (2u+1u)≤22u, the latter bound using the two equal middle coefficients. At even m≥4 the product is the previous odd product. Thus NEΠk∣Bk,χ∗∣≤b(NE,k), using 4/(2π)<1. The same bound holds under every complex embedding, since it depends only on fχ≤NE and k. In particular the absolute norm from Q(χ∗) is a positive integer bounded by b(NE,k)[Q(χ∗):Q], uniformly in ℓ. Fourier series of Bernoulli polynomials Archimedean bound in the release

Take the product Mj=∏χ(c)=(−1)j+1ANE,j+1(χ), where χ(c) denotes the value of the corresponding character of G. Construct it in a cyclotomic field containing all character values. Galois automorphisms permute these characters because they preserve the rational parity sign, and preserve the primitive conductor and Bernoulli formula under conjugation. Thus Mj is a rational algebraic integer, hence a nonzero integer. The product computed in any Q‾ℓ is its image there. Add (VI.2) over the character summands to obtain λℓ−(E,j)≤vℓ(Mj),∑ℓ∈Plog⁡ℓ λℓ−(E,j)≤log⁡∣Mj∣≤D2log⁡b(NE,j+1).(VI.3) Here integrality gives nonnegative valuations for all factors, and the first inequality establishes finiteness before any weighting. Unique factorization gives the middle inequality and shows that only primes dividing Mj can contribute. One may take P to contain all odd prime divisors of ∏j=1t−1Mj. Rational integer product Valuation comparison and prime sum Release prime bound

For 2≤k≤t, use 16≤tNE and k!≤tk+1 to bound log⁡b(NE,k) by (k+2)log⁡(tNE). Interchanging the finite sums in (VI.3) gives the unconditional bound ∑ℓ∈Plog⁡ℓ∑1≤j<tλℓ−(E,j)≤D4(t−1)(t+6)log⁡(tNE)≤Dt2log⁡(tNE), as asserted in (VI.1). Every odd ℓ is prime to D here. For a general target the release also bounds the odd primes dividing D by its separate StickBound. Exact inequalities in the release Complementary degree-dividing terms

3.4.3. From reciprocity to the character bound

The released derivation of (VI.2) uses the Gauss-sum and Kolyvagin derivative system stated below, together with the reciprocity and length-descent steps. The dependence on this system is specific to that derivation.

For the strict group in (VI.2) let f=fχ and use the primitive character χ∗. Inflating it gives the same ψ on GQ. In particular α=ψ−1κℓ−j and all its local conditions are unchanged by this passage. Since ψ has order prime to ℓ, its reduction is injective on its values. At a prime p∣f with p≠ℓ, some inertia element has α(g)−1 a unit, by primitivity at p. Outside fℓ, α is unramified. The release applies the character argument at modulus f and then uses f∣NE to enlarge the Bernoulli factor from fΠk to NEΠk. Primitive inertia and the character estimate Passage to the ambient conductor

Proved here using Kummer theory (residually cyclotomic case). Suppose αˉ=κˉℓ, where bars denote reduction modulo the uniformizer of O. Then ψ is unramified outside ℓ, by injectivity on its values and the equality ψˉ−1κˉℓ−j=κˉℓ. Put W=KO/O with action α and let π be a uniformizer. The sequence 0→W[π]→W→πW→0 identifies the strict π-torsion classes with a subgroup of the residual strict group. In detail, globally and on the decomposition group at ℓ, the invariants in W vanish because some inertia element at ℓ has κℓ(g)≡−1 mod ℓ, and then α(g)−1 is a unit. On inertia away from ℓ, W is trivial and multiplication by π on invariants is surjective. The long exact sequence therefore gives the required injectivity on these local groups in degree one. The residual group for Fℓ(1) is zero: by Kummer theory, a rational class unramified away from ℓ has valuations divisible by ℓ away from ℓ, hence is represented by ℓa modulo ℓ-th powers, since ℓ is odd. Local triviality at ℓ forces ℓ∣a as well. Extending coefficients to the residue field of O preserves this argument by linearity. Each W-valued cocycle has finite image by continuity and compactness, so its class is killed by a power of π. Vanishing of the strict π-torsion forces the entire strict group to vanish. This gives (VI.2) in the branch called exc. Classical Kummer theory Residual case in the main proof Residual vanishing

Proved here (the Bernoulli valuation in the other branch). If αˉ≠κˉℓ, choose g where α(g)−κℓ(g) is a unit. For an integer m≥max⁡(1,vℓ(f)), put f0=f/ℓvℓ(f), L=f0ℓm, and choose b∈[1,L−1] representing the action of g on μL. Then χ∗(b)bk−1 is a unit. Indeed modulo π, α(g)−κℓ(g) is (χ∗(b)bj)−1−b. For each unit u mod L use its representative in [1,L−1], write y≡bu mod L with 1≤y<L, and put hu=(bu−y)/L. The integer θb(u)=⌊buL⌋−⌊b(L−u)L⌋=2hu−(b−1) is the difference of the two Stickelberger floor coefficients. Choose cyclotomic lifts σu acting by u on μL, and define zb=∑uθb(u)α(σu)−1. Thus zb≡Sb=∑uθb(u)ujχ∗(u)(modℓmO). On the subgroup fixing μL both α,β are 1 mod ℓmO. Use δ=c~ at all depths. It acts by −1 on μL, and α(δ)−1=−2 is a unit. Floor coefficients Congruences on lifts Choice of the unit and sum

Write H(u)=Lk−1Bk(u/L)/k and dk=2kΠk. Expansion in Bernoulli numbers gives dk(H(u)−uk/(kL)+uk−1/2)∈LZ because all terms with Bernoulli index i≥2 contain Li−1 and dk clears the remaining denominators. Substituting y=bu−Lhu and reducing the binomial expansion of yk modulo L2 now gives dk((bu)jθb(u)+2H(y)−2bkH(u))∈LZ. Multiply by χ∗(bu) and sum over units. Since u↦y permutes them, if χL denotes the character modulo L induced by χ∗, the result is the congruence dk(bjχ∗(b)Sb−2(χ∗(b)bk−1)Bk,χLk)∈ℓmO.(VI.4) In this formula Bk,χL is evaluated at the imprimitive modulus L. Removing the Euler factors at primes dividing L but absent from f multiplies Bk,χ∗ by ∏p∣L, p∤f(1−χ∗(p)pk−1). This follows from the classical Dirichlet Euler product and the formula at 1−k. Here the only possible extra prime is ℓ, and its factor is an ℓ-adic unit because k≥2. Hence Bk,χL and Bk,χ∗ have the same valuation. Coefficient expansion in the release Summed congruence Extra Euler factor

Choose m large enough that ∣ℓm∣ℓ<∣a0dkBk,χ∗/k∣ℓ, for a specified 0≠a0∈O. The right side is positive by §3.4.2, and ∣a0∣ℓ,∣dk∣ℓ≤1. Equation (VI.4), the unit χ∗(b)bk−1, and the oddness of ℓ give ∣Sb∣ℓ=∣Bk,χ∗/k∣ℓ by the strong triangle inequality after dividing the error by dk. The same equality holds for zb by its congruence with Sb. In particular a0zb∣ℓm and zb∣fΠkBk,χ∗ in O, since fΠkk is integral. This is the numerical input to the cocycle bound. There is no factor polynomial in the auxiliary modulus L in the final valuation. Choice of depth and divisibility outputs Norm comparison and final divisibilities

Classical reciprocity input and release assertions. The remaining step is a bound on lengths of normalized cocycles, using an auxiliary family of Gauss sums and Kolyvagin derivatives. The release obtains its reciprocity statement Rec from Poitou–Tate reciprocity and local unramified orthogonality in the Tate pairing at primes away from ℓ. Its needed consequence pairs an α-cocycle z with a β-cocycle ν killed by a common nonzero a. Both are unramified off a finite allowed set. At the allowed primes z is locally a coboundary in W[a] except at one auxiliary prime r′≠ℓ, and both characters are trivial modulo a on the decomposition group at r′. There z vanishes on inertia. If ν(τ)=yw0 for an inertia element τ at r′, with w0 of annihilator (a), then yz(ϕ)=0 for Frobenius ϕ. The cup product uses αβ=κℓ. Its local invariants at the unramified places vanish by orthogonality, and the places where z is a coboundary contribute zero. Global reciprocity leaves the unramified–inertia pairing at r′. Poitou–Tate and local duality, Milne, Arithmetic Duality Theorems, Chapter I Unramified orthogonality Reciprocity interface Reduction to finite coefficients and duality Input wiring

Proved here using classical duality (the finite-coefficient step). Identify W[a] with B=O/(a) by choosing w0. The case of a unit a is trivial. Otherwise, because O is unramified over Zℓ, write a=uℓh with u∈O× and h=vO(a)≥1, and identify B with O/(ℓh). Thus ℓhB=0, and this exponent is exact. Write z,ν in B-coordinates as zB,νB, so that their product cup class has coefficients in B(1). For each Z/ℓh-linear functional λ:B→Z/ℓh, push the cup class to μℓh by applying λ to the underlying coefficient and retaining its Tate twist. The global sum of the local invariants is zero. Outside the allowed finite set both cocycles are unramified, so the cup class factors through the procyclic unramified quotient, whose ℓ-cohomological dimension is one. At the other allowed primes the class vanishes because zB is a coboundary there in B. At the infinite place the positive-degree cohomology is zero for coefficients of odd order. Thus the pushed local cup class at r′ has zero invariant and is zero by local Tate duality.

At r′ both cocycles are additive because α and β are trivial on the decomposition group modulo a, so κℓ=αβ is also trivial on B(1). For an unramified Frobenius ϕ, write zB(s)=η(s)zB(ϕ), where η is the unramified additive class with η(ϕ)=1 in Z/ℓh. Consequently the vanishing just proved is the vanishing of the pairing of η with s↦λ(νB(s)zB(ϕ)), interpreted with the cyclotomic action on μℓh. This map is a cocycle there because the cyclotomic action on B(1) is trivial at r′. The unramified classes with trivial Z/ℓh coefficients are generated by η. Unramified orthogonality for the dual coefficient forces λ(νB(τ)zB(ϕ))=0 on inertia. The functionals separate points of B, by the decomposition of finite abelian ℓ-groups into cyclic groups. Thus νB(τ)zB(ϕ)=0, which is the assertion used in the length descent. Global cup class and removal of the other places Local argument with the unramified generator Separation by finite-coefficient functionals

Release input (Gauss-sum and Kolyvagin derivative system). Fix ℓ,m,L,α,β,b as above. Take a prime r≡1 mod L by classical Dirichlet, choose a place above r, and take w∈KO/O with annihilator (ℓm). Put U=Gal⁡(Ω/Q(μL)). For every list I of distinct admissible primes r′, where r′≡1 mod L, r′≠r and ℓ∤ord⁡r′(r), the system gives a continuous additive map νI:U→KO/O killed by ℓm. These maps are β-equivariant under conjugation and vanish on inertia outside Lr∏I. The initial map takes some inertia element at the chosen place over r to a unit times zbw. For each fresh admissible r′, at every place above r′ and every Frobenius lift ϕ there, some inertia element has new value equal to a unit times νI(ϕ). Here adjoining a prime means prepending it to the list. The decomposition groups at these primes lie in U by splitting. Admissibility The derived system and its local identities Existence statement

The release chooses a degree-one prime in an inverse ideal class, uses a Jacobi sum at that prime, and divides out its contribution from the Stickelberger element for a generator of the product ideal. The resulting elements in Q(μL∏I) satisfy norm and reduction relations up to roots of unity. The release corrects the relations by applying X↦g0(X)/Xa∗ and taking a 2f0∏I-th power, where a∗≡κℓ(g0) mod ℓm and α(g0)−κℓ(g0) is a unit. It then takes the iterated Kolyvagin derivatives, extracts the associated Kummer maps, and projects by a sum weighted by β. The initial value is zbw times the unit 2f0(α(g0)−a∗)2. This is the valuation passed to length descent. The Jacobi-sum identities, the correction of the norm and reduction relations, and the Kummer derivative construction with its local identities are the untranslated inputs. Jacobi-sum and ideal-class construction Norm and reduction up to roots of unity Correction and derivatives Successor value Initial unit and projected floor sum

Proved here using classical Chebotarev (simultaneous auxiliary prime choice). Let zU:U→KO/O be a continuous nonzero additive map killed by ℓm and equivariant for α. Let VI be the O-span of νI(U), a cyclic submodule of W[ℓm]. Choose ρℓm=r and let H⊂U be the kernel of the homomorphism ϑ(u)=u(ρ)/ρ into μℓm. For g∈GQ, ϑ(gug−1)=g(ϑ(u)), since g−1(ρ)/ρ is fixed by u. If an additive map F on U is equivariant for γ, killed by ℓm, and zero on H, it factors through the cyclic image of ϑ. Conjugation by g on that image is the κℓ(g)-power map, so (γ(g)−κℓ(g))F=0. Equivariance on the radical quotient Power action computation

Let Bz=ker⁡zU. The unit difference α(g0)−κℓ(g0) implies H⊄Bz. If VI≠0, let K be the preimage of ℓVI under νI. If H⊆K, apply the same quotient argument to νI mod ℓVI, with γ=β and g=c~. The difference β(c~)−κℓ(c~)=2 is a unit, so all values would lie in ℓVI, contradicting that they span VI. Each value outside ℓVI generates VI. If VI=0, use K=Bz. Finally let C=Gal⁡(Ω/Q(μLℓ)). Take u1∈U acting by 1+L on μLℓ. The element c~u1c~−1u1 lies in H by inversion of ϑ(u1). It acts by (1+L)2≡1+2L≢1 mod Lℓ, since ℓ∣L is odd. Hence H⊄C. Next cyclotomic layer Nonzero and generating values on the radical stabilizer

Suppose Bz,K,C cover H. A group cannot be the union of two proper subgroups, since multiplying elements exclusive to each gives an element outside both. Choose x∈H outside Bz,C, and y∈H outside K,C. Then x∈K, y∈Bz, xy∈C. By cancellation x2y belongs to neither K nor C, so x2y∈Bz. It follows that 2zU(x)=0, contradicting odd-order coefficients and x∉Bz. Thus the open set H∖(Bz∪K∪C) is nonempty. Take a coset of an open normal subgroup contained in it and apply Chebotarev to the class of its image in the finite quotient. Choose the place and Frobenius lift ϕ realizing that coset, avoiding the primes dividing Lr∏I and any other given finite set of primes. At the resulting prime r′, fixing μL and moving μLℓ gives r′=1+Ls with ℓ∤s. The reduction of ρ is a unit in Fr′ because ϕ fixes ρ. Hence rf0s≡1 mod r′ by ρℓm=r, and ℓ∤ord⁡r′(r). This gives a fresh admissible prime with zU(ϕ)≠0 and νI(ϕ) generating VI. Subgroup avoidance Admissibility Open Frobenius conditions Simultaneous choice

Proved here (extension of the derived maps). The derived maps are additive on U. To pair them with z globally, choose δ′ with β(δ′)−1 a unit, which exists since αˉ≠κˉℓ. Define nI(g)=(β(δ′)−1)−1νI(δ′gδ′−1g−1). The commutator is in U because GQ/U is abelian. The commutator of δ′ with gh is the product of its commutator with g and the conjugate by g of its commutator with h. Additivity and equivariance give the β-cocycle identity for nI. For g∈U, equivariance and additivity give nI(g)=νI(g). Continuity follows from that of νI. This extension takes values in the cyclic O-span of νI(U), and it vanishes on inertia outside Lr∏I because that inertia lies in U. Thus the annihilator used in the reciprocity step kills the full cocycle. Global extension and annihilator

Proved here from these inputs (length descent). Normalize a strict cocycle by subtracting the coboundary of (α(δ)−1)−1z(δ), so that its new value at δ is zero. Each normalized class is represented at some finite level a0≠0, since it has finite image. Include a local primitive at ℓ among the elements killed by a0. At each p∣f away from ℓ, zero restriction to inertia and a unit value of α−1 there force the same local primitive on the whole decomposition group by the cocycle identity and normality of inertia. That primitive is killed by a0, by evaluating at the unit difference. This gives precisely the normalized a0-torsion cocycle conditions of ShT at the primes dividing L. These conditions form a directed union as vO(a0) increases. A common finite bound on all these modules bounds the Selmer group, since every finite strictly increasing chain in it is witnessed in some finite level. Normalized cocycles and local conditions Directed limit comparison

Fix a0 and choose m as above. Put J=vO(ℓm) and u=ℓm/zb∈O. Let AI generate the annihilator of νI(U), and let SI be the normalized module with extra vanishing on decomposition groups at r and the primes of I. Thus vO(AI)≤J. A normalized cocycle vanishing on U descends to the abelian quotient GQ/U on W[a0]. Commuting any element with δ in the cocycle identity gives (α(δ)−1)z(g)=0. Hence a nonzero normalized cocycle is nonzero on U. On U it is additive and α-equivariant because α acts trivially on W[a0] there. At a prime split modulo L and away from L, vanishing on inertia and at Frobenius gives vanishing on the decomposition group. This holds at all primes above it by the cocycle conjugation formula, since α acts trivially on W[a0] at the split decomposition groups. For a fresh prime from the simultaneous choice, the derived system makes the new inertia value have the same annihilator as νI(ϕ). This annihilator is (AI). Write AI∪{r′}=AIB. The initial inertia value has annihilator (ℓm/zb), so a0∣A∅. The successive divisibilities then ensure that the new annihilator kills both cocycles. In a cyclic coefficient submodule with generator w0 of annihilator (AIB), this inertia value equals a unit times Bw0, by the DVR classification. Reciprocity therefore kills the evaluation z(ϕ) by B. Evaluation has kernel contained in SI∪{r′} and image in W[B]≃O/(B). Hence length⁡OSI+vO(AI)≤length⁡OSI∪{r′}+vO(AI∪{r′}). Whenever SI≠0, the chosen evaluation is nonzero, so vO(B)≥1. The initial inertia value zbw has annihilator (u), with a0∣u. The same reciprocity argument at r bounds length⁡OShT by length⁡OS∅+vO(A∅)−vO(u). Along the iteration the annihilator exponents grow strictly and are bounded by J, so a module SI=0 is reached. The inequalities telescope to length⁡OShT≤J−vO(u)=vO(zb). Extended lengths pose no problem because the terminal bound is finite. As vO(zb)=vO(Bk,χ∗/k)≤vO(ANE,k(χ)), this gives (VI.2) for the second branch. Cocycle restriction and vanishing criterion Reciprocity step One length increment Initial term and length descent

3.4.4. Placement in the ordinary window

In Paper II's notation the cutoff before rescaling is v, the small positive parameter is ε, and t=⌊v401⌋. The field above has D=2e(v)+2≤8v2+8ε. On the scale in Paper II, log⁡(NEt)≤2vε/4 eventually, uniformly in the pair. The unconditional minus estimate Dt2log⁡(tNE) is therefore at most 16v804+33ε/4, within the allowance C′exp⁡(8vε/4) for sufficiently large v at each fixed ε>0. The bound holds pointwise on every such host and contributes no exceptional twists to the interval choice. The total weight is B2log⁡2+B−+B+, with B2 the unweighted dyadic budget and B+ the odd plus budget on the same host. The host plus eigenspace has the cohomological sign (−1)j, whereas the idempotent in the Fitting theorem selects (−1)j+1. The dyadic and host odd plus bounds are still required under their separate hypotheses. With those bounds supplied, the total is absorbed into B∗=exp⁡(vε/2), which becomes b=exp⁡(xε/2) after rescaling. Paper II, cohomological budgets Assembly of weighted and unweighted parts Final exponential absorption

3.5. Which odd plus estimates does ordinary zeta need?

Write a=e+2≥4, N=2aqd, and D=[E:Q]=2a. A character of the host is uniquely ρ=ηJquJdv, where η is a character of the real dyadic factor and u,v∈{0,1}. The three factors are independent by their ramification. Complex conjugation acts on the character by (−1)u+v. Thus the plus part in twist j uses u+v≡j mod 2. There are D/2 such characters over a splitting coefficient ring. Throughout this section 1≤j<t and q,d>max⁡(2t+1,7). Host character factors and cases

Count Coefficient prime, twist and character Contribution for ordinary zeta
I ℓ∉{2,3,7}, ℓ−1∤2j Weighted sum over all such ℓ and all D/2 characters
II j≥2, ℓ odd, ℓ∈{3,7} or ℓ−1∣2j, and ρ≠1 Individual length estimate
III Nontrivial character of odd order Empty
IV ℓ=3,7, j≥2 even, ρ=1 Zero
V ℓ=3,7, j=1, v2(cond⁡η)≥bℓ Individual length estimate in growing degree
Vs ℓ=3,7, j=1, v2(cond⁡η)<bℓ Fixed-degree estimate proved below
Z ℓ∉{2,3,7}, ℓ−1∣2j, j even, ρ=1 Zero by §3.3a

Here b3=14 and b7=18. Indeed the prime-to-ℓ part of ρ is all of ρ. At every prime in II or IV the ℓ-component of its conductor is trivial, because q,d>max⁡(7,2t+1) and an exterior prime with ℓ−1∣2j satisfies ℓ≤2j+1. Hence the release's even-order test in II selects precisely the nontrivial characters. The reference threshold at j=1 is R1=16ℓ24v2(ℓ2−1). The odd-conductor part of ρ has order at most two and the order of ℓ modulo four is two. Substitution gives 9216 and 200704, respectively. The number of Vs characters is 2min⁡(a,bℓ−1)−1, since there are 2m−2 real dyadic characters of conductor dividing 2m for 2≤m≤a, and two choices of odd quadratic factor at j=1. The outside sum in I includes ℓ=q,d. Thresholds · Weighted and individual bounds

The rational strict group. Let O be a splitting discrete valuation ring of residue characteristic ℓ, with fraction field K, uniformizer π, residue field k, and eO=vO(ℓ). Set W=(K/O)(ρ−1κℓ−j). Write S(W) for the classes in H1(Q,W) trivial on finite inertia away from ℓ and on the full decomposition group at ℓ. The extra decomposition conditions at p∣N, p≠ℓ, are automatic. If inertia acts nontrivially, its image has order prime to ℓ and WIp=0. Otherwise Frobenius minus one acts on W by the nonzero scalar ρ(Frobp)−1p−j−1. Multiplication by this scalar is surjective. In both cases the unramified local cohomology is zero. Taking the ρ-part over E gives S(W) by restriction and corestriction, globally and on the local conditions, since [E:Q] is invertible in O. Its contribution to the sum is (log⁡ℓ)length⁡OS(W)/eO. Rational strict groups and case bounds

For IV it suffices to treat ρ=1. If ℓ−1∣j, the trivial residual-character proof of §3.3a applies also at ℓ=3. Otherwise the residual character αˉ=κˉℓ−j is nontrivial on inertia at ℓ. A class killed by π lifts to a cocycle with values in W[π]. On inertia away from ℓ this lift vanishes, since inertia acts trivially on the entire module. At ℓ its restriction is a coboundary ∂b in W. Evaluating at an element where α−1 is a unit gives b∈W[π]. Over the field cut out by αˉ, the lifted cocycle is therefore an everywhere unramified additive character, trivial above ℓ. Its field of definition lies in Q(ζℓ), which has class number one for ℓ=3,7 by §3.3. Extension of ideals followed by norm is multiplication by a divisor of ℓ−1, so the residual field also has trivial ℓ-class group. Inflation–restriction gives zero over Q, and the π-primary torsion property gives S(W)=0. This covers every even twist used by IV with the same two reference primes.

The bounded-conductor case at 3 and 7. Let F be the field fixed by ker⁡ρ. Its degree is the order of ρ. In Vs that order is at most 2bℓ−3, since the real dyadic group at conductor 2bℓ−1 is cyclic of that order and the quadratic factor has order two. The degree caps are 2048 and 32768, even as q,d grow. Write ΔF for the absolute discriminant. Every character of Gal⁡(F/Q) has conductor dividing N, and conductor–discriminant gives ΔF≤N[F:Q]. Restriction to F embeds the strict group into the one with ρ trivialized. For j=1 Poitou–Tate duality identifies the dual of the latter strict group, before extending scalars to O, with the kernel of the localization map on H2(GF,Sℓ,Zℓ(2)). Here Sℓ contains the infinite places and the primes above ℓ, and modified real cohomology vanishes at odd ℓ. Apply duality first to finite coefficients and then pass to the direct limit on strict degree-one classes and inverse limit on degree-two classes. The finite-level cohomology over the ring of Sℓ-integers is finite, so the inverse systems satisfy the Mittag–Leffler condition. The odd-primary étale K-theory comparison for that ring identifies the ambient H2 with the ℓ-completion of K2(OF[1/ℓ]). In total degree two the only other possible descent term is H0(F,Zℓ(1))=0, since a number field contains only finitely many roots of unity. Localization preserves the ℓ-part of K2(OF) because residue-field K2 vanishes and the residue-field units above ℓ have order prime to ℓ. A summand of order ℓs acquires O-length seO. We have proved

log⁡ℓeOlength⁡OS(W)≤log⁡∣K2(OF)[ℓ∞]∣.(VI.5)

This argument uses prime-to-ℓ restriction and works for ramified coefficient rings. At ℓ=7, F is totally imaginary by parity and unramified at 7 by its conductor. Its group of roots of unity has order wF prime to 7. Emery's bound on K2 away from 6wF gives a bound O[F:Q](ΔF2(log⁡ΔF)[F:Q]−1) on the right of (VI.5). For 3 the following integral estimate suffices. Poitou–Tate duality, Ch. I §4 · Tate on K2 of global fields · Weibel, Chapter VI §8, étale comparison for number rings · Emery, Theorem 1.3

Lemma. For each fixed d0 there are A,C>0 depending only on d0 such that log⁡∣K2(OF)∣≤A(2ΔF)C for [F:Q]≤d0.

Proof. Put m=[F:Q], r=6, L=OFr and Γ=GLr(OF). Use Ash's contractible well-rounded retract X of positive forms, with their common scale fixed by making the shortest nonzero lattice vectors have length one. Each open cell is specified by the set of shortest vectors in L, which spans Fr. Stabilizers are finite, and passing to a boundary cell enlarges that set. We give the count that allows small-prime torsion to be retained. All constants until the end of the proof depend only on m,r. Well-rounded cells and stable homology, Soulé §§3–4

Choose an F-basis x1,…,xr of shortest vectors and put L′=∑iOFxi. Packing balls of radius 1/2 in the real normed lattice shows that its covolume vol⁡(L) is bounded below. In the chosen basis vol⁡(L′) is a signature constant times ΔFr/2∏σdet⁡(Hσ)1/2, with conjugate embeddings counted twice. Here Hσ is the local Gram matrix. Its diagonal entries are at most one, so each determinant is at most one. Hence the index u=[L:L′] is O(ΔFr/2). The lower bound on vol⁡(L′) gives det⁡(Hσ)−1=O(ΔFr) at every embedding. For any other shortest vector ∑bixi, we have ubi∈OF and ∣σbi∣=O(ΔFr/2) by the bound on Hσ−1. There are polynomially many possible coordinate vectors at each u. Indeed an integral element has an embedding of absolute value at least one unless it is zero, by its norm. Packing in the box prescribed for ubi proves the count.

The coordinate lattice lies between OFr and u−1OFr. There are at most u(mr)2 possibilities, since a subgroup of (Z/u)mr has at most mr generators. In a fixed coordinate lattice a shortest-vector set has at most 3mr elements, by disjoint balls of radius 1/2 around them inside the ball of radius 3/2. Thus the number of distinct sets in the polynomially bounded pool of coordinates is polynomial in 2ΔF. The lattice and the set determine the original cell up to Γ. For each face of this cell the original chosen basis stays shortest, so its enlarged set satisfies the same coordinate and packing bounds. The length of a flag of faces is bounded by the dimension of the space of forms plus one. Barycentric subdivision therefore gives polynomially many simplex orbits in every fixed dimension.

Let Q be the image of Γ modulo 3OF, so ∣Q∣≤3mr2. The kernel is torsion-free. Indeed restriction of scalars embeds it in 1+3Mmr(Z3), where cubing an element of exact depth s≥1 raises its depth by exactly one, and a power prime to three preserves depth. Let EQ be the homogeneous bar construction, the contractible free Q-complex with ordered p-simplices in Qp+1. The diagonal action of Γ on X×EQ is free. A stabilizer must lie both in the torsion-free kernel and in a finite stabilizer on X. After subdivision the product cells give a free cellular resolution. In total degree three the quotient has O((2ΔF)C) cells, since at most ∣Q∣p+1 choices from EQ accompany any orbit from X. Each product boundary column has Euclidean norm at most five, by the two simplex boundary formulas, also when faces coincide in the quotient. The torsion of H2(Γ,Z) injects into the torsion of the cokernel of this boundary. Hadamard's inequality on its nonzero maximal minors bounds the logarithm of its torsion order by the number of columns times log⁡5.

Finally the Bass stable rank of a ring of integers is at most two. Rank six lies in the integral isomorphism range for homological stability in degree two, so this H2 identifies with stable H2. The elementary stable group is perfect and equals the stable special linear group over a ring of integers. Its universal central extension has kernel K2(OF). In the split determinant sequence for the stable general linear group, units act trivially on elementary-group homology. Indeed on a finite bar cycle conjugation by g agrees with conjugation by diag⁡(g,det⁡(g)−1) in one larger rank. In the homology spectral sequence, perfectness kills the incoming second differential to bidegree (0,2) and the determinant section kills the incoming third differential. Consequently K2(OF) embeds integrally into stable H2. This K2 is finite by finite generation and its rational rank. Combining the estimates and taking the maximum over m≤d0 proves the lemma. Stability range, Soulé §4.2 · Stable elementary groups and K2, Weibel, Chapters III–IV

Taking d0=2048 at 3 in (VI.5) gives log⁡3eOlength⁡OS(W)≤(3N)C′ for a fixed C′. There are at most 4096 such characters at 3 and 65536 at 7. These factors can be absorbed into one fixed power of 7N. This proves the full contribution of Vs at conductor one.

Remaining odd plus estimates. Count I still requires a bound by (Nt)C on the sum over each finite set of exterior primes at fixed j. Counts II and V require individual bounds by (ℓNj)C with an absolute exponent for this host family. In II the exterior primes are at most 2j+1, so summing those bounds over j<t, the eligible primes and at most D/2 characters costs only a further fixed power of Nt. The two reference primes in II and V have the same property. These three estimates would complete the odd plus term of §3.2. Degree grows in V, and both degree and twist may grow in I and II. The exponents of the fixed-degree lemma above depend on degree. Assembly of the exterior sum

3.6. Why two Frobenius traces can bound a sum over all exterior primes

Count I asks for an estimate on a sum over an arbitrary finite set of coefficient primes ℓ, at fixed twist j. On the ordinary family the comparison field in this count is the fixed field K0=Q(−7). Its class number is one. An exterior prime in Case I is prime to 7h(K0) and to every host character order. Its local modulus component in a host character is trivial except possibly at q,d, where it is quadratic at most. Inertia at 7 acts nontrivially on K0 and trivially on the two cyclotomic fields of conductors N and ℓ. These facts give the field, order and ramification hypotheses in the release's one-character estimate. Parity and the condition ℓ−1∤2j are exactly those of §3.5. Hypotheses on the comparison field and character

The modular form at a split point of Count I

Proposition. Let U be the rational rank-two block of a type-II point in Count I with the Galois conditions of the release. Put M=7N, k=2j+1 and θ=θQ(−7). For the chosen coefficient embedding there is a primitive cuspidal newform f of weight k, nebentype θ and level 7cond⁡(ρ)2 dividing 7N2, attached to U(−j) in the inverse arithmetic Frobenius convention. Its polarized rank-two adjoint block has zero rational finite Selmer group. Rank-two block conditions · Local and global level conditions · Pan, Theorem 1.0.2

Proof. The representation U is continuous, absolutely irreducible already on the cyclotomic kernel over K0, odd by its determinant θ, and unramified outside ℓM. Its residual semisimplification has characters θΨ‾,Ψ‾−1, where Ψ=ρκℓj. These follow from the block conditions, the identity of residual traces and Brauer–Nesbitt (the determinant is fixed and ℓ>2). One may work over a finite coefficient extension. For instance the image of the compact global group is covered by its closed subgroups of matrices in the countably many finite extensions of Qℓ. One such subgroup has interior by Baire, and adjoining the entries of finitely many coset representatives suffices. The same argument applies to the finite list of local characters. Rank-two block conditions

The ordinary flag at ℓ has a line and a quotient whose inertial characters, in that order, are ξκℓj,ξκℓ−j, with ξ factoring through the action on μℓ. Untwist by ξ. The two characters are now unramified twists of cyclotomic powers and hence crystalline. Their ratio D, line divided by quotient, has Hodge–Tate weight −2j in the convention HT(κℓ)=−1. Both H0(Qℓ,D) and H0(Qℓ,D∨(1)) vanish. The local Euler characteristic gives dim⁡H1(Qℓ,D)=1. The crystalline dimension formula gives dim⁡Hf1(Qℓ,D)=1, since Fil0DdR=0. Thus every extension with this flag is crystalline after the finite twist. This proves the potentially semistable hypothesis in Pan's theorem and gives the two distinct weights. Bloch–Kato, §3 · Pan, Theorem 1.0.2

Apply Pan's theorem to U(−j)∨. Count I has ℓ≥5, and its residual representation remains a sum of two characters. The Hodge–Tate weights of this dual are {1−k,0}, so a modular attachment in Deligne's arithmetic Frobenius convention has weight k. More explicitly, a one-dimensional geometric twist over Q is a Tate power times a finite Dirichlet character by class field theory. Absorb the latter into the primitive newform. The gap of the two weights fixes k, and their endpoint fixes the Tate normalization. The determinant of the dual is θκℓ2j, fixing the nebentype. Dualizing gives exactly the polynomial of f at inverse arithmetic Frobenius on U(−j). Changing the coefficient embedding replaces f by its conjugate. Pan, Theorem 1.0.2 · Newform convention

At p≠ℓ inertia acts diagonally on U through two characters of Gal⁡(Q(μM)/Q). Each has local conductor exponent at most vp(M), also at 2, by local reciprocity. The monodromy operator is zero and the Artin conductor exponents add. At p=ℓ the finite inertial part of the potentially crystalline Weil–Deligne representation is the scalar ξ, with monodromy zero. Its conductor exponent is zero if ℓ∤M. Indeed the level condition gives two finite inertial characters of modulus dividing M, and their cyclotomic exponents j,−j distinguish them on every open inertia subgroup. Thus ξ agrees on inertia with the corresponding finite character. At ℓ∣M its tame conductor exponent is at most one. The exponent of the two-dimensional representation at every p is consequently at most 2vp(M). Tate twist and duality preserve these conductor exponents. Local–global compatibility for classical newforms identifies them with the newform conductor exponents, giving the asserted divisibility. Local and global level conditions · Carayol on local compatibility away from the coefficient prime · Saito on compatibility at the coefficient prime

For the prescribed inertia of the marked point one can compute the exact conductor. At p≠7,ℓ, the restriction to inertia is conjugate to characters ρp,ρp−1 on the outer block, giving exponent 2a(ρp), where a(ρp) is the conductor exponent. At 7, inertia over K0 acts trivially and the quotient has order two. Its determinant is the nontrivial quadratic character, so the two eigenvalues are 1,−1 and the exponent is one. At ℓ the scalar ξ on Weil–Deligne inertia equals ρ∣Iℓ, by the prescribed ordinary characters and their distinct cyclotomic exponents. This again gives exponent 2a(ρℓ), including ℓ=q,d. Since 7∤N, the conductor is 7cond⁡(ρ)2. Inertia comparison in the marked frame · Automorphic attachments and the level calculation

The form is non-CM. Otherwise U is induced from a character of a quadratic field L, and its trace vanishes on the nonidentity coset of GL. The ratio of the two residual characters is then −1 on that entire coset and 1 on GL, by taking quotients of two elements in the coset. This ratio restricts on Iℓ to κ‾ℓ 2j, since θ is unramified there and the square of the finite inertia character of ρ is trivial. This restriction is nontrivial by the defining condition of Count I. On the other hand the decomposition group at ℓ preserves the ordinary line. In its intersection with GL, an inertia element has two distinct eigenvalues by the infinite cyclotomic image and the two different powers. Its eigenlines must be the two inducing lines. The ordinary line is one of them, so the entire decomposition group fixes it and belongs to GL. This contradicts the nontrivial residual ratio on Iℓ. Rank-two block conditions

On the scalar matrices the polarized block has character θ. On the trace-zero matrices its action agrees with ad⁡0U. Over GK0 it is conjugation. At complex conjugation the action is −JXtJ for the antidiagonal identity J, which on trace-zero matrices equals conjugation by diag⁡(−1,1)=U(c) in the marked basis. Newton–Thorne's Theorem B(1) applies to the non-CM form over Q and kills the trace-zero adjoint Selmer group. Passing to the dual or a Tate twist preserves this adjoint. A finite Selmer class for the scalar character θ restricts to an everywhere unramified additive character on GK0. At ℓ the crystalline degree-one classes for the trivial coefficient are the unramified ones, by the crystalline dimension formula. The class group of K0 is finite and restriction detects the rational class by corestriction. This kills the scalar contribution. Action on the rank-two block · Newton–Thorne, Theorem B(1)

The forms in this proposition supply the rank-two portion of the finite list in the trace calculation. There are O(kM4) of them. All embed as normalized forms in Sk(Γ1(M2)), and distinct newforms give linearly independent simultaneous eigensystems away from M by multiplicity one. The index of Γ1(M2) is at most M4. The valence bound in weight k bounds the dimension by 1+k[SL2(Z):Γ1(M2)]/12, by bounding the vanishing order at infinity of a nonzero form. The list is stable under conjugation. At either split probe p=11,23, for ℓ≠p, its rank-three trace values are 1+p−jap(f). Deligne's bound gives absolute value at most three in every embedding, and multiplication by pj clears the denominator. If two points give the same form, their restrictions 1⊕U∣GK0 are isomorphic. One may therefore use all these forms, keeping repeated trace values with their system multiplicities, uniformly across coefficient primes. The irreducible rank-three points require their own list. Newform convention · Deligne, Weil I

The two off-block Selmer groups

The two vanishings from Kato's Theorem 14.2(1). At a split type-II point, let U be the rational rank-two block of determinant θ, the quadratic character of K0. The preceding proposition attaches a primitive newform f of weight k=2j+1 to U(−j). At good primes the attachment compares the Hecke polynomial with inverse arithmetic Frobenius. Kato uses this convention in (14.10.3), and (14.10.2) gives determinant κℓ1−kωf−1 for nebentype ωf. Hence U≃Vλ(f)(j) after extending the coefficient field at the place λ∣ℓ selected by the attachment, by comparison of the unramified polynomials and semisimplicity. Apply Theorem 14.2(1) over the finite abelian extension K0/Q with r=j. We have 1≤j≤k−1 and 2j≠k. The finiteness holds at every coefficient place and for every stable lattice in Kato's representation with this twist. Type-II attachment · Attachment convention · Kato, Astérisque 295 (2004), Theorem 14.2(1) · Kato (14.10.2)–(14.10.3)

At every finite v∤ℓ the local condition is the kernel of restriction to inertia, including at the bad primes. At v∣ℓ it is the kernel of the map to cohomology with coefficients in Bcris⊗Vλ(f)(j), and the degree-one condition at infinity is zero. For a stable lattice T in V=Vλ(f)(j), Kato's torsion Selmer group in §14.1 uses the images of these local subspaces in H1(K0,v,V/T). Reduction therefore maps Hf1(K0,V) into that finite group. Every class in the rational Selmer group is unramified outside a fixed finite set S containing the primes above ℓ and the ramification of V. In the exact sequence on GK0,S for 0→T→V→V/T→0, the kernel of reduction is the image of H1(GK0,S,T). Inflation into full global cohomology is injective in degree one for V/T. The integral group is finitely generated over the coefficient DVR. Its image cannot have finite index in a positive-dimensional vector space over the fraction field, so finiteness of the reduction image gives Hf1(K0,V)=0. Finite coefficient extension preserves these kernels. Kato §14.1, pp. 234–235 · Kato §§14.8–14.9

Restriction and averaging over the quadratic group identify this Selmer group, after coefficient extension, with Hf1(Q,U)⊕Hf1(Q,U⊗θ). Indeed the invariant and anti-invariant projections correspond to the two summands of Ind⁡GK0GQ1. The local identifications preserve the finite conditions. Away from ℓ restriction to finite-index inertia detects the kernel by restriction and corestriction. At ℓ the same restriction test applies after tensoring with Bcris. Thus both off-block summands vanish. Equivalently, apply 14.2(1) over Q to f and the primitive quadratic twist of f at the same twist j. Kato §14.6 · The two off-block projections

The ordinary argument in the release. For either V=U or U⊗θ, the Galois hypotheses give odd determinant δ=θ, absolute irreducibility on H0=GK0(μℓ∞), and a local line A and quotient B with inertia characters ξκℓj and ξκℓ−j, where ξ factors through the action on μℓ. The automorphic witness is an additional input to the local comparison. It gives a spherical source after twisting by ϕ=δ−1ξ−1 and twisting down by j. Here δ is unramified at ℓ. The normalized character on A evaluates on Frobenius to the unit root, up to inversion, of the good Hecke polynomial of this source. Both roots have complex absolute value ℓj, by purity at good primes. In particular the character of A cannot be κℓ. Equality would force j=1 and trivial finite inertia twist, and would make the normalized Frobenius value a root of unity. The quadratic twist preserves the inertia flag and multiplies the unit root by an unramified sign. Galois hypotheses and witness · Ordinary source comparison · Kato (14.10.5) and Proposition 17.1

For the global class use the newform h attached to V(−j), so h=f or its primitive quadratic twist. Kato's Theorem 12.5(1), at cyclotomic level zero with r=k−j=j+1 and trivial Dirichlet character, identifies the signed period of the local dual exponential with (2πi)j−1L{ℓ}(h∗,j+1) γ(−1)j−1. Here h∗ is the dual newform, the Euler factor at ℓ is omitted, and the period vector has nonzero indicated sign component. Kato's Theorem 13.5(1) (Jacquet–Shalika) gives nonvanishing at j+1=(k+1)/2, the boundary of the absolute-convergence half-plane. The omitted Euler factor causes no zero there, by the root bound (14.10.5). The geometric reciprocity statement for the smoothed constituents after newform projection is Theorem 9.7. By Theorem 12.6 the integral smoothed families span the rational zeta submodule, so at least one fixed family has nonzero dual exponential after specialization. Proposition 8.12 and Example 13.3 give its integral norm relation (13.1.1). The resulting class y is unramified away from ℓ, and its local projection yB∈H1(Qℓ,B) is nonzero since the dual exponential on H1(A) lands in Fil0DdR(A)=0. For every n≥1 and every q≡1 mod ℓn outside a fixed finite bad set, corestriction from conductor qℓa to the degree-ℓn subfield of Q(μq) gives the tame class with further corestriction Pqy. Here Pq=det⁡(1−q−1V(Frob⁡q))≡det⁡(V(Frob⁡q)−1) mod ℓn, with arithmetic Frobenius. The same family supplies all depths and the classes have the integral unramified condition away from ℓ,q. Kato Theorem 9.7 · Theorems 12.5(1) and 12.6 · (13.1.1) and Example 13.3 · Theorem 13.5(1) · (14.10.5)

Write VD=V∗(1)≃V⊗κℓδ−1 and identify its pairing with V with the wedge pairing in this model. A strict dual class z is unramified away from ℓ and zero locally at ℓ. Scale y,z to make their cocycles and the required local primitives integral. The pinned kernel tests give y∣N0,z∣N0≠0 for a nonzero strict z, where N0=ker⁡V∩ker⁡κℓ. The first test uses the nonzero quotient class and the unequal ℓ-inertia weights on the quotient and the unipotent upper entry. For the second, a class vanishing on N0 descends to the quotient where an element of ker⁡V with nontrivial cyclotomic value is central and acts on VD by a nonidentity scalar. The cocycle identity then makes it a coboundary. The H0-stable span of the two additive restrictions projects onto both copies of V. If the alternating pairing of the two restrictions vanished, that span would be the graph of a scalar by nondegeneracy and absolute irreducibility. On that graph the pairing is twice a nonzero scalar times the wedge. Thus there are g,h∈N0 with b=⟨y(h),z(g)⟩−⟨y(g),z(h)⟩≠0. There is γ∈[H0,H0] with Dγ=det⁡(V(γ)−1)≠0. A subgroup of SL2 consisting entirely of unipotents is either trivial or fixes a unique line. Applied to the normal commutator image, either case would contradict absolute irreducibility. Pinned kernel and commutator tests

At each depth n, the cup y∪z has a primitive u modulo ℓn on the global group unramified outside a finite set S containing ℓ and the representation's ramification. Its local classes vanish by strictness at ℓ and the two unramified conditions elsewhere. Poitou–Tate gives the global primitive since the dual obstruction is an everywhere unramified character of Q with constant coefficients. Put Ds=det⁡(V(s)−1) and H(s)=Dsu(s)−⟨adj⁡(V(s)−1)y(s),z(s)⟩. The tame norm relation gives DsH(s)=0 mod ℓn for s∈[H0,H0]. In this step Chebotarev approximates s by a Frobenius τ at a good q whose degree-ℓn cyclotomic quotient ψq at conductor q is trivial at S. The splitting requirement comes from adjoining ℓn-th roots of the primes in S and μℓn, an abelian extension after base change to the cyclotomic tower. The first moment m of the tame Shapiro cocycle, corrected by the coboundary in its augmentation, satisfies dm=−ψq∪Pqy. The cup cocycle m∪z−ψq∪Pqu is locally zero away from q, hence at q by global reciprocity. Evaluating at a tame inertia element t with ψq(t)=1 and its conjugate by τ gives (V(τ)−1)m(t)=Pqy(τ) and ⟨m(t),z(τ)⟩=Pqu(τ). The adjugate and the congruence for Pq give DτH(τ)=0. Applying the resulting congruence at γ and γ[g,h], whose matrices agree, gives Dγ2b=0 mod ℓn because the cup identity gives H(γ[g,h])−H(γ)=Dγb. As n is arbitrary, this contradicts Dγb≠0. Kato §14.9 on duality · Pinned tame and commutator calculation

For an ordinary dual class, its localization at ℓ comes from H1(Qℓ,B∗(1)). The reciprocity sum for its cup with y gives zero at ℓ since the other local terms vanish. At ℓ this pairs the lift with yB in the perfect pairing of the two one-dimensional local groups for B and B∗(1). The lift is zero and the dual class is strict, hence zero. The primal condition at ℓ is the image of H1(A), exactly orthogonal to the dual condition. It has dimension one by local Euler characteristic since H0(A)=H0(B)=H0(A∗(1))=0. In the Poitou–Tate difference of the ordinary Selmer dimensions, global invariants vanish, unramified local dimensions equal invariant dimensions away from ℓ, and the contribution +1 at ℓ cancels −1 from the single real invariant. The primal ordinary Selmer group therefore vanishes too. The finite condition lands in the primal ordinary one since Hf1(Qℓ,B)=0 by the Bloch–Kato dimension formula and Fil0DdR(B)=DdR(B). Kato §14.9 and Proposition 14.12 · Pinned ordinary dual argument

The one-character cotangent comparison

Fix a good prime p≠ℓ and let g be geometric Frobenius at w∣p in K0. Set Ψ=ρκℓj and Tρ=Ψ(g)+1+Ψ(g)−1.

Write Eℓ=Frac⁡O, let ϖ be a uniformizer, and put eO=vϖ(ℓ). Alpöge’s one-character comparison uses the normalized polarized lift on GQ with matrix diag⁡(Ψ,1,Ψ−1) on GK0. Complex conjugation has matrix part 1, and its action on the Lie algebra is X↦−JXtJ with J the reversing matrix. The two marked entries at the separating element and its conjugate inverse are fixed by the normalization, leaving conjugations by diag⁡(t,1,t−1). The graded presentation A/I of the normalized problem is taken at the split augmentation x with kernel P in A. It comes with lifts of matrix entries, descent of points and comparison of tangents with the complete local ring. The defining ideal I=(F,G,H,H′,Z) has equations of weights −1,−2,1,2,0, respectively. The model and its integral lifting property are inputs to this calculation. Normalized split lift · Graded model · Integral lifting clause

Let C=(P/P2,sat)−1/⟨[Fi]⟩O, where P2,sat={z∈P:bz∈P2 for some 0≠b∈O}. The strict coefficient (Eℓ/O)(Ψ−1) uses the character of the adjoint lattice line O(E10−E21) in indices 0,1,2. Conjugation by the diagonal acts on both entries through Ψ−1 and the sign action −J(⋅)tJ fixes their difference. The source imposes zero decomposition restrictions above ℓN and zero inertia restrictions at every finite prime. At p∣N, p≠ℓ, the unramified local group vanishes by the local calculation for S(W), so the two strict kernels agree. Strict lengths

For 0≠b∈O, normalize a class in S(W)[b] by subtracting the unique coboundary that makes its cocycle z zero at the marked g1, where Ψ(g1)2−1 is a unit. Then bz=0 and z(c)=0 by plus parity and ℓ≠2. The cocycle also vanishes at cg1−1c. Its value on the killed inertia away from ℓ is zero. At ℓ its local primitive already lies in W[b], by evaluation where Ψ2−1 is a unit on inertia. Identifying W[b] with O/(b), the cocycle defines a weight −1 deformation in O⊕ϵO/(b) through multiplication of the reference lift by 1+ϵz(E10−E21). Locally it is an infinitesimal conjugate of the reference lift. Lift the local primitive integrally and use conjugation stability of the ordinary ideal at ℓ. The integral lifting clause of the model then lifts the global derivation on the ambient A to an O-valued derivation with that local compatibility. It kills P2,sat by torsion-freeness, and its reduction kills the Fi. This gives S(W)[b]↪Hom⁡O(C,O/(b)), since the functional recovers the derivative of each lifted matrix entry. Every class is killed by some ϖn by continuity with discrete coefficients. The length of each Hom⁡O(C,O/ϖn) is at most length⁡OC once C has finite length, so the directed union gives the same bound for S(W). Cocycle comparison · Integral lift · Length bound

The graded calculation requires degree-zero finiteness of A0/(A0∩I). The special-fibre argument below supplies this input from the linked ordinary support hypotheses.

In characteristic zero the prescribed ordinary ideal at ℓ must hold on dual-number lifts, so their adjoint cocycles satisfy the finite local condition by potential crystallinity. The away-from-ℓ inertia equations give the other finite local conditions. At the split augmentation, rational vanishing for the nonpositive-weight character summands gives cotangent spanning in weights −2,−1,0. This uses class field theory in weight zero and the local finite condition, Poitou–Tate and positive-twist H2 finiteness in negative weights. The source’s point tests also exclude opposite one-sided branches over the same invariant value away from x. At two-sided points, where both positive and negative weights are nonzero, adjoint rigidity reduces the normalized tangent to the Euler direction of the torus. For the split type-II points this uses the modular block and its scalar, trace-zero and off-block Selmer vanishings. For the type-I points it uses a polarized cuspidal attachment and the adjoint vanishing at the chosen coefficient place. The trace comparison uses attachments over K0 at the Count I levels N in type I and 7N in type II, with the local conditions of the respective admissible families. Finite tangent conditions · Split cotangent spanning · Ordinary coefficient wrapper · Admissible families

The graded calculation uses a three-graded Rees algebra R generated over A0 in the three unit degrees by A−2u0, A2u1, and (A0F∗2+A−22A2)u2. Here Smith form on the weight −1 linear relations gives r classes ϖciyi and an extra equation F∗ with zero linear part, so length⁡OC=∑ici. Write r′ for the number of weight +1 equations and s,s′ for the numbers in weights −2,+2. Cut the weight-zero equations in the split formal chart, let JR be the coefficientwise kernel of that chart map, and use the sections t=(Fi2u0,Giu0,Hi2u1,Hi′u1,F∗2u2). For a=tr⁡r(g)−Tρ in degree zero, choose a representative regular in the cut chart, adding a weight-zero term hFi with h∈A1 if needed. This preserves all point values. In large multidegrees d, the leading terms in R/(JR,a) occur in bidegree (2d0+4d2,2d1+2d2) in the negative and positive variables. The Hilbert series for their ranks is (1−X)−r(1−X2)−s(1−Y)−r′−1(1−Y2)−s′−1. Removing one squared section for the ϖ-layer calculation takes the finite differences with factors (1−X2)r−1+s(1−Y2)r′+s′(1−X4Y2). The resulting series is (1+X)r−1(1+Y)r′(1+X2)/((1−X)(1−Y2)), with eventual coefficient 2r−1+r′ at the specified even positive degree. Each Fi2 contributes 2ci layers. Additivity and eventual nonnegativity of the Koszul Euler characteristics thus give the lower bound 2r+r′length⁡OC≤χ(a,t)d. Leading-term quotient · Euler calculation

For the upper bound, let Md=(R/(JR,t))d. Replacing each section by its cokernel drops a shifted kernel with nonnegative Euler characteristic. Hence χ(a,t)d≤θa(Md), where θa(Md)=length⁡O(Md/aMd)−length⁡OMd[a]. In large degrees the generic support lies on the two-sided locus. Indeed a homogeneous prime on the relevant Rees locus with all A−2A2 products zero would give opposite one-sided points by restriction to the two Rees axes. Away from x this is excluded by the point tests, and over x a negative-only point is excluded by the rational spanning in weights −1,−2. Over the two-sided locus one can invert fh with f∈A−2 and h∈A2. The slice f=1 kills the Euler tangent, so it is finite and reduced over the algebraic closure. Its residual μ2 action gives at most one dimension per weight at each invariant value. Squaring the r+1+r′ equations increases this length by a factor at most 2 per equation, by filtering with its first power and using the shifted weight piece. The local multiplicity of Md[1/ϖ] is therefore at most 2r+1+r′. Generic support · Torus slices · Squared equations · Euler positivity

With a nonzero at the counted points, the kernel and cokernel lengths on finite torsion cancel in θa(Md). On the free part θa is the valuation of the determinant of multiplication by a. Each geometric invariant point contributes the valuation of its trace difference with the generic multiplicity just bounded. The marked-frame comparison injects the two-sided orbit representatives into distinct admissible isomorphism classes on GK0. Dividing the upper bound by 2r+r′ leaves the factor 2. Dividing by eO converts vϖ to vℓ, with vℓ(ℓ)=1. Finite extension of coefficient DVRs multiplies the Selmer length and eO by the same relative ramification index. The source invokes the one-character theorem first over an unramified coefficient DVR and then uses this invariance. At geometric Frobenius, with Q=Nw and z=ρ(Q)−1, the probe is tr⁡σ(g)−(Q−jz+1+Qjz−1). Multiplying by the ℓ-adic unit Qj clears denominators without changing valuations. Distinct systems remain distinct indices even if they have identical trace values. This one-character comparison is separate from the support implication on the two Count II links. One-character theorem · Marked-frame comparison · Coefficient independence · Frobenius application

The details of the Rees support and Euler comparisons are given in the one-character source note. Ordinary cotangent length

Special-fibre finiteness in Count I

The degree-zero finiteness deduction uses one abelian CM field L containing K0, fixed for the marked Count I problem. Put M=7ℓN, write η for the sign of K0, and ω=κˉℓ. The link takes GL+=U=H0∩Hth(w)∩Hlay(mlay) and GL=U∩ker⁡η. Here the index-two H0 contains complex conjugation and agrees with ker⁡η on decomposition groups above M and ramifies somewhere outside M, Hth(w) cuts out the degree-ℓ cyclotomic subfield of conductor w, and Hlay(mlay) cuts out the degree-ℓmlay cyclotomic layer. The auxiliary w∤M splits in the fields of η,Ψˉ,ω. The layer mlay=3C0+7 uses a uniform class-group rank bound C0 and gives [L+:Q]>3C0+7 and local degrees above ℓ greater than five. The frame also uses the strict residual classes of characters Ψˉ−1η and Ψˉ−2η. Its quotient U↠Φ.Γ kills the closed normal subgroup generated by away-from-ℓ inertia in ker⁡η∩ker⁡Ψˉ, and restricts to a surjection GL↠Φ.G0 onto the sign kernel. Linked field · Layer and frame · Common quotient

Work over an unramified coefficient DVR containing the host character values. A local map from the normalized ring to a complete characteristic-ℓ DVR A′ with finite residue field kills the coefficient maximal ideal. Its lift has the marked normalization and is ordinary by the local ideal itself. At ℓ the spectral projectors of an inertia element with Ψˉ2≠1 transport the prescribed flag to A′. The Count I conditions ℓ−1∤2j and quadratic-or-trivial finite inertia give that separation. The prescribed kernel away from ℓ gives descent through Φ.Γ, since reduction preserves prime-to-ℓ inertia kernels. For every linked v∣ℓ the three diagonal characters of the flag on inertia are the powers of Ψˉ with weights (1,0,−1). They equal 1 on Iv∩ker⁡Ψˉ. Integral flag · Descent and local comparison

For each rank n=2,3, write Bn for the degree-zero algebra of the linked matrix space, ϕn:Rndet↠Bn for the twisted determinant comparison and Un=ϕn(Jn)Bn for the image of the ordinary Hecke ideal. The residual tuples on GL before scalar twisting are (Ψˉ,Ψˉ−1) and (Ψˉ,1,Ψˉ−1). The rank-two twist restricts the algebraic character κ with κκc=κℓ−1, and the rank-three twist is κℓ−1. Let ln be the coefficient maximal ideal together with the equations setting the ordered local characters to 1 on Iv∩ker⁡Ψˉ. On a prime containing Un+ln the pro-ℓ weight characters are trivial, since the inertial characters then have order dividing the order of the multiplicative group of the coefficient residue field. Finiteness of Rndet/Jn over weight therefore gives mBn⊆Un+ln. Torsion-freeness gives the expected component dimension 1+n[L+:Q]. The separate rank-two support problem uses a rational seed of determinant ηθ with residual characters (Ψˉη,Ψˉ−1), where θ has order ℓ and conductor w and is trivial on GL. Scalar twists · Weight-fibre finiteness · Seed and dimension inputs · Seed characters

The rank-three seed. Write ρ∣Iℓ=ω~aℓ∣Iℓ, with 2aℓ≡0 mod ℓ−1, and put b=j+aℓ and χ′=[Ψˉω−b], using the Teichmüller lift. Thus χ′ is unramified at ℓ and has the host's inertia away from ℓ. Use the fixed ℓ-adic comparison place throughout. The auxiliary character θ of conductor w has order ℓ and is trivial on GL. Set χ=χ′ηθ and choose m≥2 sufficiently large with m≡b mod ℓ−1. The primitive character χ has conductor 7wcond⁡(χ′) and parity (−1)m+1. The Eisenstein series Em+1(1,χ) has constant term zero modulo the coefficient maximal ideal. Reducing the order-ℓ factor deletes the multiples of w and introduces the factor 1−(χ′η)(w)wm≡0. Subtract the constant term times the integral correction form of the same weight and character, then apply Deligne–Serre lifting. The resulting newform f has this exact prime-to-ℓ level by primitivity, with ap(f)≡1+χ(p)pm for every prime p using zero on nonunits. In particular aℓ(f)≡1. Its arithmetic Frobenius attachment Vf has determinant χκℓm and Vˉfss=1⊕Ψˉηˉ. Eisenstein congruence and rational attachment · Integral correction and lift · Deligne–Serre, Lemme 6.11 · BLGGT, Theorem 2.1.1

The seed cusp on GL3(AL) is the Gelbart–Jacquet adjoint of the solvable base change of f to L. Indeed, in a dihedral projective image the trace-zero coset would force the residual ratio Ψˉηˉ to have square 1, contrary to the marked element of GL. The ordinary weight gap makes the projective image infinite, hence Zariski dense in PGL2 by irreducibility and the exclusion of a torus normalizer. This persists on open subgroups and supplies cuspidality and the irreducibility of the adjoint on the selected L. In geometric normalization the attachment is R=κℓ−1Ad⁡0(Vf∣GL),Rˉss=ω−1(Ψˉ⊕1⊕Ψˉ−1). The common Tate twist supplies the factor ∣kz∣ on the adjoint Satake roots at geometric Frobenius at z∤ℓ. BLGGT's rank-three comparison at ℓ applies by Shin-regularity. Pinned seed and comparison · Arthur–Clozel, Ch. 3 · Gelbart–Jacquet, Theorem 9.3 · BLGGT II, Theorem 1.1

At ℓ let e be the unramified character with arithmetic Frobenius value the unit root of X2−aℓ(f)X+χ(ℓ)ℓm. It reduces to 1 on the decomposition group. The crystalline ordinary line and quotient of Vf have characters χκℓme−1 and e. On the marked local coordinate of L, put λ=χ′κℓme−2. The first adjoint line consists of maps from the quotient to the line, so the order for R is (ξ0,ξ1,ξ2)=κℓ−1(λ,1,λ−1). This reduces in that order to ω−1(Ψˉ,1,Ψˉ−1) on the full local group. Its inertia exponents are (m−1,−1,−m−1), with labeled Hodge–Tate weights (1−m,1,1+m). The gaps exclude equality of two ordered characters up to one cyclotomic power. The smooth inducing characters after removing the labeled algebraic factors are unramified. The arithmetic unit-torus character in coordinate i is κℓhi−i on inertia, with hi the corresponding labeled weight. At a lift gx of the uniformizer in geometric inverse Artin coordinates the normalized ordinary steps are ∏i<sξi(gx)κℓ(gx)i for s=1,2,3. Their reductions are (ω−1Ψˉ,ω−1Ψˉ,1)(gx). Seed point checks · Ordinary principal-series operators

Away from ℓ the conductor of Vf equals that of its determinant. At a ramified place the conductor sum forces the codimension of inertia invariants to be one, giving inertia 1⊕χ. Monodromy commutes with the distinct inertia characters and is therefore zero. On GL both η and θ are trivial, giving the inertia multiset (χ′,1,χ′−1) for R and conductor exponent 2az(χ′) at z∤ℓ, with az computed after restriction to Lz. At z∣ℓ the crystalline parameter has conductor exponent zero. The multiset is the Teichmüller lift of the prescribed residual inertia, and injectivity of reduction on prime-to-ℓ roots kills the linked inertia kernel. These checks apply at both base levels of the definite U(I3) over L+. They have empty Taylor–Wiles sets, the split tame type at every marked v∤ℓ over M and the ordinary Iwahori condition at ℓ, spherical level away from the marked places, and Kc(1) at the respective split scalar place v0 or v0′ above distinct rational primes outside M. The seed is spherical at both of those places. The compact highest weight is (m−1,0,1−m). The simple generic packet has trivial component group modulo the fixed dual center, and the pure-inner-twist product formula gives its compact member with the required split local components. Selected realizations · Two runs · Local newform conductor · KMSW, Theorems 1.6.1 and 5.0.5

Proposition (the split coefficient type). At a marked split place z∤ℓ, let F=Lz, Kz=GL3(OF) and tz=(χ′,1,χ′−1)∣IF, viewed over C via ι. There is a smooth finite free coefficient representation Tz over the given coefficient DVR O such that, for every smooth irreducible π of GL3(F), ((Tz⊗O,ιC)⊗Cπ)Kz≠0⟺rπ∣IF≃tzandNπ=0, where (rπ,Nπ) is the Frobenius-semisimplified local Langlands parameter in the algebraic normalization. The external tensor product over the marked places satisfies TypeForAlt. Pinned type test · Integral existence and normalization

Proof. Use local reciprocity to transport tz to unit characters (α,1,α−1). Group each distinct character μa in a single block of size its multiplicity. Thus the trivial case has one block of size three and the nontrivial quadratic case has blocks of sizes two and one. In the distinct case order the characters (α,α−1,1), since aF(α2)≤aF(α) for the conductor exponent aF. In the compact subgroup Jz⊂Kz the lower block (b,a) is divisible by pFaF(μa/μb). The character λz of Jz is the product of (μa/μa+1)(Δa) over the leading block minors Δa, with μs+1=1 for s blocks. On the integral Levi it is ∏aμa∘det⁡. In particular the quadratic block uses the full integral GL2 with its determinant character. The exponents refer to the actual local unit filtration, including wild characters at dyadic places. Compact model · Multiplicativity on the conductor subgroup

For the upper block parabolic P, the function fz=1Jzλz belongs to the compact model of induction from this determinant character of P∩Kz, with trivial action on its unipotent radical. When the parameter on the right of the equivalence holds, Frobenius preserves each inertia block because its character extends to the Weil group. Frobenius semisimplification therefore splits it into Weil characters even in a repeated block. Local Langlands and the segment classification express π as the parabolic product of twisted spherical representations on the blocks. Each block supplies a determinant eigenvector, and Iwasawa decomposition embeds the compact function model in π∣Kz by multiplying its functions by the tensor of these vectors. Conversely, any nonzero map from the cyclic Kz-span of fz sends fz to a nonzero λz-vector. The conductor calculation gives that vector nonzero image in the upper Jacquet module. The geometric lemma and the separation of the distinct unit-character lines then give the same product of twisted spherical blocks. This argument applies to all smooth irreducibles, including spherical Langlands quotients at reducibility points. The seed itself is a unitary principal series with unlinked inducing characters, so the compact-model embedding applies directly. Induced-model embedding · Jacquet argument · Parameter criterion for the block product · Inertial local Langlands, Theorem 3.7

The Teichmüller values of λz lie in O×. The O-span of the translates of fz is finite free, since it is a submodule of the functions on a finite congruence quotient of Kz. Take its O-linear dual for Tz. The tensor invariant is equivalently a Kz-map from the complexified cyclic span into π. This is the dual convention in the pinned test. The unramified twist ∣det⁡∣−1 in algebraic normalization leaves the test unchanged. The pinned construction obtains its own integral detector and combines the local tests over the product of integral maximals. Integral detector in the release · Product criterion

The seed on the projected ordinary module. In each run fix the enlarged coefficient DVR O=c1.O, the chosen integral type τ0, and the base level used to define J3. Write A for the weight algebra, Θ for its torus character, Mord for the ordinary module before residual projection, and T0 for the commutative A-algebra generated by the Hecke operators on it. The fixed type and the split local vectors above give a nonzero classical eigenvector at the seed weight xA. Arithmetic control identifies the characteristic-zero specialization of Mord at xA with the classical ordinary space at that level, intertwining the spherical operators and all three normalized ordinary steps. The eigenline therefore defines a character of T0, since every operator relation on Mord acts on its specialization. The character is integral over the arithmetic coefficient DVR by finiteness over A. Its generator reductions agree with the prescribed residual profile, by the good split Frobenius polynomials and the ordered step congruences above. Thus it sends the idempotent ea of that residual block to 1. It descends to a character y of the image algebra T on eaMord, because the kernel of restriction from T0 is (1−ea)T0. Coefficient DVR · Arithmetic control · Residual projector · Eigensystem on the projected module

The given map f:R3det↠T defining J3 has the spherical polynomial and ordered unit and uniformizer identities. After applying y, its determinant agrees with that of R at every good split geometric Frobenius. These classes are dense up to conjugacy on the level quotient, so continuity and characteristic-zero determinant uniqueness give equality of the laws. At z∣ℓ write uz,s=y(Uz,s) for the normalized step values and uz,0=1. The torus identity gives the i-th character on inertia as κℓ(g)−ixA(Θ(ez,i(az(g))))−1, where az is inverse geometric Artin and ez,i inserts a unit in coordinate i. At the chosen uniformizer lift gx the character is κℓ(gx)−iuz,i+1/uz,i for i=0,1,2. The inertia formula and the three ratios give exactly ξi on the Weil group and hence, by continuity, on the full decomposition group. The determinant and ordered characters determine the point, so yf=x and J3⊆ker⁡x for each of the two base levels. The seed determinant is absolutely irreducible and ker⁡x is a stable characteristic-zero prime. Hecke identities · Frobenius density · Chenevier, Proposition 1.27 and Theorem 2.12 · Ordered comparison and seed inclusion

This deduction uses one arithmetic weight. The existence of f on the whole image in each run uses the following fixed-level construction.

Proposition (the rank-three determinant on the projected image). Let X=eaMord with the fixed integral type, weight algebra and level of the run, and let T⊂End⁡A(X) be its Hecke image. Use a set of arithmetic weights x:A→Q‾ℓ whose kernels intersect in zero. Assume equivariant ordinary control on these fibers and a spanning family of classical parts on each projected fiber on which all the Hecke generators act by scalars. For each contributing system y:T→Q‾ℓ, assume a continuous rank-three attachment ry on the same level quotient Γ=V.frame.G0 of GL. Its good split polynomials specialize from the formula below. It has conjugate duality with multiplier κℓ−2, and trivial action on Iukill=Iu∩⋂iker⁡χˉi at each marked tame place u, where χˉi are the linked residual characters. At each marked z∣ℓ it has a full stable flag whose ordered characters are the specialized torus and step characters below. Then there is a unique continuous degree-three determinant law D on Γ with the spherical polynomials, specializing to det⁡ry at these evaluations. Together with the three ordered characters at each z, it gives the weight-compatible surjection f:R3det↠T with J3=ker⁡f. Hecke kernel construction · Universal pair

Proof. Ordinary freeness and the residual idempotent make X finite free over A. For a nonzero operator in T, some matrix entry in an A-basis survives at a separating weight. Equivariant control transports its action to the projected fiber. Since scalar parts span that fiber, the operator is nonzero at some character y. Thus the product of the y is injective on the actual image and the image is reduced. This argument only uses the induced action on the fiber, which can have a kernel as an action of the abstract scalar extension of T. The image is finite over the complete Noetherian local ring A, being a submodule of End⁡A(X). The prescribed generator reductions select one local factor with residue field kO. Finiteness over A makes the maximal-ideal topology equivalent to the mA-adic topology and gives completeness and finite adic quotients. Hence T is compact Hausdorff. Each arithmetic weight takes values in a finite coefficient DVR, and integrality over the weight scalars puts y(T) in the integers of a finite extension of that DVR's fraction field. The preimage of the coefficient maximal ideal is mT, so y is continuous. Consequently the diagonal map into ∏yQ‾ℓ with the product of the ℓ-adic topologies is a closed embedding, since its source is compact and its target is Hausdorff. Projected freeness · Arithmetic control · Local image and units · Separation on the image · Compactness and continuity

Write p:GL→Γ for the quotient, [g] for a group-ring basis element and ci(g) for the tuple of i-th elementary characteristic-polynomial coefficients of the ry(g). At a good split w, these tuples are the images of the elementary coefficients, with the alternating signs as in the spherical polynomial D(Z−[p(Frob⁡wgeom)])=∑s=03(−1)sqws(s−1)/2Tw,sZ3−s,Tw,0=1,qw=∣kw∣. They are invariant under all conjugations of the Frobenius lifts. The primes above nonsplit places of L+, apart from finitely many ramified places, have relative degree two and density zero among the primes of L. Chebotarev on finite quotients gives density of the union of the good split classes in Γ. By continuity and closedness each ci(g) lies in T for every g and varies continuously there. This extends the coefficients on single group elements, including any products of group elements viewed as single elements. Good split density · Determinant construction

Put t=c1. To obtain the law on a sum ∑gZg[g] of finite support, use the following coefficients for the cubic, with distinct group elements where indicated. Zg3c3(g)Zg2Zh, g≠hc2(g)t(h)−t(g)t(gh)+t(g2h)ZgZhZk, g,h,k distinctt(g)t(h)t(k)−t(gh)t(k)−t(gk)t(h)−t(hk)t(g)+t(ghk)+t(gkh). Use each ordered pair for the second row and each unordered triple once for the third. At each y, the second row is tr⁡(adj⁡(ry(g))ry(h)) by the cubic adjugate identity. For the third row, sum over ordered triples of distinct row indices (a,b,c), taking row a from ry(g), row b from ry(h) and row c from ry(k). Permute rows and columns simultaneously to that order. The resulting determinants have columns indexed by (a,b,c), so repeating an index contributes zero and the sum can run over all triples. Expanding these determinants and summing gives the product of three traces from the identity permutation, the three transposition terms, and the two oriented three-cycle terms. Thus at every y the cubic is the matrix determinant of the sum. Comparing coefficients in polynomial rings via the injective product of the y gives the unit and multiplicativity identities and compatibility when supports are enlarged or terms are combined. Evaluating these polynomials over every commutative T-algebra gives the determinant law. Continuity is the adic continuity of the group-element coefficients. Uniqueness follows from density and the integral determinant-law identity expressing all mixed coefficients in the group-element coefficients (Amitsur's formula). The displayed cubic keeps all multiplicities with integer coefficients, including in residue characteristics two and three. Integral determinant descent · Uniqueness from dense classes

At z∣ℓ put Uz,0=1. The character ψz,iκℓi on the decomposition group is defined via inverse geometric Artin by the unit character Θ(ez,i(−))−1 mapped to T and the uniformizer value Uz,i+1/Uz,i for i=0,1,2. Each step is a unit. The powers of each ratio extend to Z^ through the finite adic quotients, so reciprocity gives the continuous characters, uniquely determined on the dense Weil group. The torus identity identifies their inertia restrictions with the universal weight characters. It also gives their prescribed values on the preimage of ℓ-primary torsion in the local topological abelianization, since that torsion has zero image in the unramified Z^. Reduction on inertia and the three prescribed residual step ratios give the ordered residual characters on the full decomposition group. For D, reduction of the good spherical polynomials gives the prescribed residual determinant of ω−1(Ψˉ⊕1⊕Ψˉ−1) on dense classes. Continuity and the same integral cubic give equality as laws. The global congruence is a determinant identity, so it allows nonsplit residual representations with that semisimplification. The local congruences retain the prescribed order. The tame kernel condition is on the marked typed inertia subgroups. At the separate depth-one place the base construction uses the specified level and scalar residual polynomial. Weight and residual identities · Local torsion subgroup · Ordered inertia, reduction and torsion · Residual determinant law

Conjugate duality identifies κℓ(g)−2ry(cg−1c−1) with ry(g)T by one change of basis for all g, where c is the frame's complex conjugation. Applied to group-ring sums this proves determinant polarization at each y. On its ordered flag 0=F0⊂F1⊂F2⊂F3, the factor ry(gi)−y(ψz,i(gi)) 1 sends Fi+1 into Fi. For independently chosen gi in the local decomposition group, written in ry via the designated map to Γ, the product in order 0,1,2 is zero, acting from right to left. Its trace after left multiplication by any ry(h) is zero. The same trace statement holds for ry(g)−1 with g∈Iukill. Separation gives the polarized, ordered and tame relations in T. Universality gives f, and equality on the universal inertia characters makes it compatible with A. Since qw is a unit at each good w∤ℓ, the spherical coefficients recover every Tw,s in the image of f. The ordered products of ψz,i(gx)κℓ(gx)i recover all three Uz,s there. These operators and A generate T. Descent of the deformation relations · Surjectivity on generators

The release uses good weights at fixed admitted algebraic weight lam=0. Local reciprocity gives power-series coordinates from the free inertia generators with the stipulated torsion values fixed. One can vary the smooth characters through independent infinite sets of torsion twists in the open unit discs, choosing disjoint ranges of ℓ-power orders to distinguish the three unit characters at each coefficient place. Weierstrass preparation in each coordinate gives weight separation. The pinned control and decomposition give scalar action on the good projected fibers at the same tame type, residual profile and away-from-ℓ level. The spectral and local comparison input covers every contributing constituent, including endoscopic sums and any Speh strings before applying a local exclusion. Once full split transfer is available at a coefficient place, a longer Speh string repeats supercuspidal support under unramified shifts and cannot match the good torus support with distinct unit characters. The reverse implication of the split type proposition and local–global compatibility give the exact Teichmüller inertia multiset with zero monodromy at the typed places. At ℓ the toral zero-monodromy comparison, labeled Hodge–Tate weights and normalized ordinary steps give the ordered stable flag. Almost-everywhere spherical transfer by itself leaves these local comparisons open. The release takes HypothesesH′ as input. The parameter-refined trace route below uses stabilization and the quasi-split stable multiplicity formula to obtain the specified local transfer. Good-weight separation · Projected scalar action · Classical covering · Tame comparison · Ordered flag comparison · Hecke kernel construction

Propagation additionally requires separating classical evaluations of the actual projected modules at all required auxiliary levels and LevPtAlt at the actual split patch completions. Its comparison is with the Frobenius-semisimplified Weil–Deligne parameter, with equality required for toral parameters of zero monodromy. It uses the arithmetic spherical annihilator comparison and classicality at irreducible arithmetic points. The patch modules retain the type, weights and residual data, with ordered residual Frobenius refinements at the patch places. They require freeness over the prescribed diamond group rings, determinant-compatible comparison after augmentation with the same base module, and Frobenius and ordered tame relations after clearing the Frobenius difference factors. The source uses the exponent 3!=6 in the integral annihilation relations at rank three. These are the same-module inputs to the passage from LevelAlt to HasAut. Classical auxiliary-level contract · Arithmetic annihilator comparison · Same-module levels · Determinant-compatible comparison · Split patch identities · Augmentation and annihilation · Passage to patching · Integral rank-three interpolation

Count I rank-three seed calculation · Split type calculation · Base-level eigencharacter and determinant

The arithmetic support condition P1 concerns every generization of a supported nice type-I point. More precisely, if f:Rndet→Ax is the determinant map of a characteristic-ℓ curve point with the required irreducibility, distinct ordered local characters and no cyclotomic intertwiner, then Jn⊆ker⁡f implies Jn⊆q for every prime q⊆ker⁡f. For the intertwiner test, at the realization's auxiliary place choose g with ω(g)≠1 and residual characteristic polynomial (X−c)n, c≠0. The characteristic polynomial of r(g) evaluated at κℓ(g)r(g) is invertible by reduction, so any cyclotomic intertwiner is zero by Cayley–Hamilton. The support theorem uses ordinary auxiliary-level modules at all Taylor–Wiles depths, with group-ring freeness, maps to the prescribed base level and the uniform tame inertia and Frobenius relations, together with the cohomological generator and obstruction estimates. In the patched completion the support and ordinary dimension comparisons give propagation. For P1 the chain has length zero and the stable discriminant stays nonzero at every prime below ker⁡f. The chart statement follows by normalizing the one-dimensional quotient to a DVR and contracting the primes through the determinant comparison. The companion chain condition P3 makes the corresponding propagation across nice endoscopic points joined by thick primes, where a nonsplit extension of the middle constituent by the outer block witnesses thickness. Determinant support tests · Absence of the intertwiner · All auxiliary levels · Patching and zero-length chain · Transfer to charts

Rank-two propagation puts every off-triangular prime over U2. The outer two-by-two corner of a rank-three endoscopic prime is a rank-two point with the matching weight equations. The middle character reduces to the reference on the fixed-weight tests. This transfers the rank-two radical bound to an ideal I3⊂B3 controlling all endoscopic off-triangular primes. The rank-three seed, the connected chart comparison with intersections of dimension at least 3[L+:Q]+1, and P1 and P3 give support over U3 on the remaining off-triangular primes. Thus OffFin holds with I∗=U3∩I3. Every off-triangular prime contracts to one containing I∗, and mB3⊆I∗+l3 follows by primality from the two component bounds. Rank-two propagation · Corner comparison · Rank-three propagation

For a linked DVR point the fixed-weight ideal l3 vanishes. In the off-triangular case OffFin makes its invariant map factor through the residue field. The diagonal entries become the reference entries and the three two-cycles and two three-cycles of off-diagonal entries in the determinant all vanish. In the triangular case divide the three diagonal characters by their reference characters. The ratios take values in principal units. Each finite reduction kills inertia above ℓ by the frame's local generator test, and away from ℓ by prime-to-ℓ inertia in the link. Class field theory bounds all their orders by one exponent, for example 2hL. A finite-order principal unit in a characteristic-ℓ domain is 1, by Frobenius for its ℓ-part and reduction for its prime-to-ℓ part. These two cases give the exact polynomial ∏i=−11(X−Ψˉ(g)i) on GL. An injective finite DVR extension handles the coefficient enlargement in the linked realization, and the identity descends coefficient by coefficient. One arithmetic realization in each rank suffices for this finite-fibre test. This is FinEnd with the same subgroup GL for every test point. Off-triangular case · Triangular case · Coefficient descent · Common subgroup

For any g∈GK0 a power lies in GL, so the eigenvalues are algebraic over the coefficient residue field. Reduction is injective on algebraic constants in A′, which recovers the reference polynomial at g. The marked diagonal difference has three distinct eigenvalues, so a stable subspace is spanned by coordinate lines. The polarized transpose pairs the adjacent entries. Opposite signs with an adjacent entry make the directed graph of entries strongly connected, contradicting the split semisimplification given by the full polynomial. If only the two opposite corner entries occur, the middle line has character 1 and removing it gives the same contradiction on the outer block. The fixed complex conjugation extends this one-sign triangularity to the polarized matrix parts. Each augmented invariant monomial therefore vanishes on every test. A monomial surviving at a prime over ϖ would survive on a one-dimensional complete domain quotient and its normalized DVR. Projection to degree zero then gives ϖC0=mC0 for the completed invariant ring C0. Its special fibre is Artinian and topological Nakayama makes C0 finite over O, hence also its entry-monomial subalgebra. Algebraic constants and marked triangularity · DVR detection · Weight-zero finiteness

This derives FinA from the linked arithmetic support hypotheses. At either base level the rank-three seed inclusion follows by the single-specialization argument once the projected determinant pair is constructed. The remaining finiteness inputs include the rank-two seed inclusion and the same-family spectral and Galois comparisons needed to apply ordinary patching in both ranks for P1 and P3. The rank-four argument below derives auxiliary-level freeness and the augmentation comparison on the selected Count II links from their fixed-level hypotheses. The characteristic-zero attachments and the controlled-level trace lists enter the one-character inequality separately. Ordinary Hecke family · Linked arithmetic input

Count I special-fibre finiteness

The trace product

Here are the two arithmetic inputs to the deduction. Write vℓ(ρ)=length⁡OS(W)/eO for the strict length in §3.5 with ρ and j understood. For a good rational prime p≥7, choose w∣p in K0, put Q=Nw, and let g be inverse arithmetic Frobenius at w.

  1. For ℓ≠p, there is a finite set Xρ of representatives of distinct admissible Galois representations over K0. Write Ψ=ρκℓj and Tρ=Ψ(g)+1+Ψ(g)−1. If the traces of all admissible representations at g differ from Tρ, the strict length is finite and satisfies vℓ(ρ)≤2∑σ∈Xρord⁡ℓ(tr⁡σ(g)−Tρ),ord⁡ℓ(ℓ)=1.(VI.6) The normalized model and arithmetic hypotheses for this comparison are given above. The inequality applies to the two-sided orbit representatives once the admissible systems separate their traces from the diagonal. Application of the one-character bound
  2. There is a Galois-stable list of complex traces us of full Hecke eigensystems at g. Its two parts are the cuspidal rank-three systems and the systems with trace 1+tr⁡U. For the chosen K0, assume ∣Sp∣≤C[N(j+1)]A, ∣us∣≤Q+2 at every embedding, and Qjus algebraically integral. The constants precede the host, twist and test prime. For each coefficient embedding and each ρ, the members of Xρ have their traces at distinct system indices s in this list. Repeated trace values are retained with their system multiplicities. The release obtains this input by local–global compatibility and counting systems at controlled levels. Combination of the two sets of systems

Proposition. These two inputs imply Count I on the ordinary host, with constants independent of the finite set of coefficient primes.

Proof. Take two fixed primes p=11,23 and enlarge the common threshold so q,d>23. Both split in K0 since their classes 4,2 mod 7 are squares. In particular Q=p. Work first at one of the two primes. Let Xj be the complex characters of the host of parity (−1)j. Its cardinality is D/2 and it is stable under all automorphisms of the algebraic numbers. Set R=pj and zρ=ρ(p)−1. For each ρ and each system index define Fρ,s=R(us−1)−zρ−R2zρ−1. This factor is the integral multiple R of the difference between the trace us and the diagonal trace R−1zρ+1+Rzρ−1. The last summand in Fρ,s has absolute value R2, whereas the first two together have absolute value at most R(p+3)+1<R2 for p=11,23 and R≥p. Hence every factor is a nonzero algebraic integer of absolute value at most 2R2. Taking their product over both index sets gives a nonzero rational integer np, since every field automorphism permutes the factors. Thus log⁡∣np∣≤(D/2)∣Sp∣(2jlog⁡p+1).(VI.7) Fix an exterior ℓ≠p and an isomorphism from the chosen algebraic closure of Qℓ to C. It identifies the ℓ-adic host characters with the complex characters of the same parity after adjoining their values. For a nonsplitting coefficient field, extend it to include the host character values. The normalized length is invariant under this extension. Now each difference in (VI.6) occurs amongst the factors indexed by that character. The factor R is an ℓ-adic unit. Unused factors have nonnegative valuation, and the indexing keeps every occurrence even if two different characters have identical traces. We obtain ∑ρ∈Xjvℓ(ρ)≤2ord⁡ℓ(np).(VI.8) For ℓ=11 use n23, and for all other ℓ use n11. The weighted sum over any finite set in Count I is at most 2log⁡∣n11n23∣ by unique factorization in Z. Equation (VI.7), D≤N and j+1≤t give the asserted fixed power of Nt. In the release the construction is made for the characters of each exact kernel. The complex fibers have sizes φ([GN:H]) and are disjoint, so their total on the host is at most D. The product proof above uses the union of the parity fibers. Product and subproduct comparison · Two-prime estimate

An explicit form of the two integers. Put m=2a−2, and for p=11,23 set h11=m and h23=m/2. The signs Jq(p),Jd(p) disappear from the character products in (VI.7). To see this, take the representative −p≡1 mod 4 of the class of p in the cyclic real dyadic group. For b≡1 mod 4, repeated squaring gives v2(b2k−1)=v2(b−1)+k if b≠1 and b≠−1. Here v2(−p−1) is two at 11 and three at 23. Thus the dyadic evaluations at p run over the hp-th roots of unity, each m/hp times. In either parity two pairs of quadratic exponents occur. Each multiplies the dyadic values by a sign, which permutes those roots since a≥4 makes both orders even. The multiplicities for the full set are therefore two at 11 and four at 23.

For R=pj define integral polynomials C0=2, C1=X, Cd+1=XCd−R2Cd−1. If X=U+R2/U, induction gives Cd(X)=Ud+(R2/U)d. The monic polynomial Ch(X)−(1+R2h) has the h roots z+R2/z, zh=1. They are distinct since equality for two unequal roots would imply zz′=R2, whereas ∣zz′∣=1. We obtain the formula np=∏s∈Sp[Chp ⁣(R(us−1))−1−R2hp]2m/hp.(VI.9) The eigensystems in this expression still depend on the host level and twist. For any Case I prime other than 11 and 23, (VI.8) can in fact be applied twice. Its length is bounded by twice the valuation of gcd⁡(∣n11∣,∣n23∣). At 11 use the valuation of n23 alone, and at 23 that of n11 alone. This also identifies a finite set containing every Case I prime of positive length for the specified host and twist. Classification by system indices · Trace estimate for a Galois-stable family of roots

Count I therefore uses the conditional one-character comparison above and the two controlled-level lists. The split coefficient type gives local occurrence for the linked rank-three seed, and the one-weight comparison gives its support at each base level with the projected determinant pair. The auxiliary-level comparisons for special-fibre finiteness and the propagation and descent giving the controlled attachments for all two-sided type-I points remain part of the Count I input. The normalized model with its ordinary ideal and cohomological hypotheses also enters the comparison. The rank-two proposition and Kato's theorem supply the pointwise modular-block ingredients under their stated Galois conditions. For the rank-three adjoint vanishing given attachment, see §3.7. Automorphy and finiteness hypotheses

3.7. Trace products for Counts II and V

Use the notation of §3.5 and write Lℓ,j,ρ=(log⁡ℓ)length⁡OS(W)/eO. The bounds below give the two individual estimates once the source's deformation and automorphy inputs hold on this host. Count statements

Counting the systems. For a fixed imaginary quadratic field F, the number of cohomological cuspidal systems in rank r at principal level M with coefficient system V is OF,r(M2r2dim⁡V). At fixed level 3 the finitely many arithmetic quotients have finite cell models by Borel–Serre compactification. Passing to level 3M lifts each cell at most (3M)2r2 times, by the index bound from reduction into GLr(OF/3M). The cochain dimension is bounded by the cell count times dim⁡V. Cuspidal cohomology injects into arithmetic cohomology, with each distinct cohomological constituent contributing at least one dimension. Level M invariants are also invariant at level 3M. In rank four the Weyl formula at each embedding, for m=(j+1,j,1,0), gives ∏i<kmi−mkk−i=j2(j2−1)12. The two embeddings give O((j+1)8) for the coefficient dimension. The count is therefore O(M32(j+1)8). Rank three over Q(i) has count O(M18) at j=1 because the weight (1,0,−1) is fixed. Level index and cohomology Cuspidal injection and weight criterion Weyl construction Borel–Serre, §§8–11 Franke, Ann. Sci. ENS 31 (1998), 181–279

To count rank-four points over Q, the source fixes 18 imaginary quadratic fields with pairwise coprime discriminants. At most one is ramified at the test prime. At most 16 sign characters can be self-twists of a rank-four irreducible representation, since their intertwining lines in its endomorphisms are independent. Thus some remaining restriction is irreducible. A restriction has at most two lifts up to isomorphism, differing by a sign twist. At the test prime one recovers candidate rational Frobenius eigenvalues by taking roots of order at most two, at bounded cost. The resulting system bound has the same powers of M,j. This step uses base change at the stated level, local–global compatibility and Galois conjugation of algebraically normalized Satake data. Equality of full system indices implies isomorphism by Chebotarev and Brauer–Nesbitt. Equal traces at one prime may belong to different systems, and the product retains their multiplicities. CM system classification Rational classification Arthur–Clozel, Ch. 3

Count II. Put c=v2(cond⁡η) and B=2nII. The source chooses an odd primitive character of conductor B and uses the trivial odd-modulus auxiliary character. Its level parameter is M=ℓNB, and the three finite characters of the split lift factor through modulus M and have orders dividing k=2(ℓ−1)B. The odd-conductor part of ρ has order at most two, so n0=v2(ℓf−1) for f=ord⁡(Z/4)×(ℓ). If ℓ≡1 mod 4, then 2n0≤ℓ−1. Otherwise 2n0≤2(ℓ+1). Since 2c≤2a and ℓ2<qd on II, the ceiling in nII gives B<2max⁡{ℓN4n0,64⋅2cℓ3}≤8ℓN(ℓ+1)2,M≪ℓ4N2,k≪ℓ4N. Thus the dyadic cyclotomic field has degree at most B. The dyadic roots required in the coefficient ring lie in an unramified extension of Qℓ. Even the source's combined exponent nII+nIII costs degree O(ℓ6N), since the unused III parameter satisfies 2nIII=O(ℓ3) at conductor one. In any coefficient extension the Selmer length and eO scale by the same relative ramification index. Dyadic orders and levels Ceilings Construction and level Finite-character exponent Coefficient extension

At p≥5 prime to M, the split trace is T4=pj+1z1+pjz2+pz3+1, with zik=1. The admissible rank-four systems have integral trace s with ∣s∣≤(3p+1)p(j+1)/2≤(7/2)pj. Since p≥5 and j≥2, the leading term of T4 exceeds ∣s∣+pj+p+1 in modulus. Hence 0<∣s−T4∣≤p2(j+1). Form A4 by multiplying s−T4 over the systems and the conjugates of the phase triple. Galois permutes the integral factors, so A4 is a nonzero integer. Consecutive use of the cotangent estimate, integrality of the unused factors and −log⁡∣A4∣ℓ≤log⁡∣A4∣ gives Lℓ,j,ρ≤2log⁡∣A4∣≪∣Orb⁡(z1,z2,z3)∣ M32(j+1)9log⁡p.(VI.10) Indeed the cotangent estimate charges twice the sum of the negative ℓ-adic log norms over distinct irreducible points. Each point uses a distinct system factor and all other factors have ℓ-adic norm at most one. Split trace Separation and integral product Product and classification Normalized comparison

The source's least good prime p≥5 is 5 or 7 on this host. The Galois orbit contains at most φ(k)≤k phase triples. Equation (VI.10) gives O(ℓ132N65(j+1)9). Alternatively take the least p≡1 mod M. The phases are all one and Linnik gives log⁡p≪log⁡(ℓN). This gives Lℓ,j,ρ≪ℓ128N64(j+1)9log⁡(ℓN)≪(ℓNj)67.(VI.11) For the last bound at an exterior prime, use ℓ≤2j+1, ℓ5<N5/2 and log⁡(ℓN)≪N to absorb ℓ61log⁡(ℓN) in O(j56N3). At the reference primes the first bound is O(N64(j+1)9log⁡N). Good-prime conditions Least good prime Linnik's bound, Xylouris

Count V. Here j=1 and the dyadic conductor threshold in §3.5 gives the source's rank-three local conditions with level N. The sign character cuts out K=Q(i), of discriminant −4. For p≥7 inert in K and prime to ℓN, put Q=p2 and let g be inverse arithmetic Frobenius at the prime of K over p. The diagonal characters are (ρκℓ,1,(ρκℓ)−1), so the trace at g is T3=Q−1z+1+Qz−1, where z=ρ(p−2). In the system count ∣s∣≤Q+2 and Qs is integral. Because Q>Q+3+Q−1, every factor Q(s−T3) is nonzero and has modulus at most 2Q2. The product A3 over the Galois-stable rank-three systems and the conjugates of z is a nonzero integer. The broader source norm includes a rank-two branch. For V, the point classifier sends every tested representation into the rank-three branch. The cotangent bound uses twice the valuation sum and the source allows two embeddings of K. Within each embedding the classification is injective on the points being counted. Since Q is an ℓ-adic unit, this gives Lℓ,1,ρ≤4log⁡∣A3∣≪∣Orb⁡(z)∣N18(4log⁡p+1).(VI.12) Host data and dyadic condition Trace and classification Rank-three branch Two embeddings Trace bound Integer product

At ℓ=3 take p=7. At ℓ=7 take 11 or 19 different from q. Neither can be d≡7 mod 16. The odd quadratic factors disappear at p−2. In the real dyadic group of order 2a−2 the square has order dividing 2a−3, so the phase orbit has size at most 2a−4. Alternatively choose the least prime in the reduced class p≡−1 mod 2a, p≡1 mod qd. It is inert, avoids 3 and 7, and log⁡p≪log⁡N by Linnik. Its phase is one. Thus (VI.12) gives Lℓ,1,ρ≪N18min⁡(2a−4,log⁡N)≪(ℓN)19. The coefficient field can split the host characters by an unramified extension of degree dividing φ(2a−2). Source prime and orbit choices Source phase orbit Linnik's bound

Inputs to the cotangent comparisons. In II, the cotangent comparison uses the weight −1 cotangent length at the split lift, a finite degree-zero algebra, exclusion of the specified block end, automorphy of all irreducible points, and rigidity modulo symplectic conjugation. The source proves rigidity by passing to an auxiliary CM field and returning an adjoint coboundary to sp4. In V the reduced lifting input gives the marked element, finiteness in degree zero and automorphy of the type-I points over K after a Tate twist. The frame comparison uses local tangent conditions, Poitou–Tate, Euler characteristics and vanishing in the relevant weights. Once the type-I adjoint vanishing below applies, the remaining tangents come from the torus orbit. The frame comparison gives the common bound for the finite-coefficient lengths, which then bounds the divisible strict group. The construction of the reduced lift and the classification at the stated local level remain inputs to (VI.12). II cotangent bound · II rigidity · V reduced lift · V cotangent and finite-coefficient bounds · V assembly

A common choice of rational primes for all three counts. For q,d>23 take p=11 unless ℓ=11, in which case take p=23. In I these primes split in Q(−7). In II both avoid every prime in the ramification set dividing 2ℓN when used as indicated. In V, where ℓ=3,7, 11 is inert in Q(i) and its prime has norm 112. The hypotheses on the Galois element used to bound a deformation length in II and V follow from the respective archimedean trace gaps. The uniform field and representation hypotheses in §§3.6–3.7 remain in force.

With this fixed choice, the phase triples of II have at most k≪ℓ4N conjugates and M≪ℓ4N2. Equation (VI.10) and the bound on the logarithm of each factor give Lℓ,j,ρ≪ℓ132N65(j+1)9. In II, ℓ≤max⁡(7,2j+1). Summing over its primes, over j<t and the host characters gives O(DN65t143). For V the phase at 11 has at most 2a−4 conjugates, so the bound on a single length is O(N19) and both reference layers together cost O(DN19). Equation (VI.7) summed over j<t gives O(DNAtA+2) for I, with A the exponent in the count of the systems. Together with §3.5 this proves a uniform bound (Nt)C∗ for the entire odd plus contribution from the stated deformation and automorphic estimates. The fixed choices display how all three comparisons apply on the same host and to the same positive twist range.

Adjoint rigidity of the irreducible rank-three points in I and V

Let σ be the rank-three representation on GK at a type-I point, with K=Q(−7) in I and K=Q(i) in V. Write θK for the quadratic sign and A for the adjoint of the polarized deformation over Q. The normal form at complex conjugation c has matrix part 1 and sign −1, with Ψ(c)=ρ(−1)(−1)j=1 for the host character ρ. If J is the antidiagonal identity matrix, the pairing is σ(g)tJσ(cgc−1)=J. Thus A is the full endomorphism space, with the conjugation action on GK and the action X↦−JXtJ−1 at c. The attachment required in both counts is rπ,ι≃σ(−1) for a regular algebraic conjugate self-dual cuspidal π of GL3(AK). The Tate twist leaves A unchanged and makes the polarized multiplier θKκℓ−2, which has value −1 at c. After killing the finite inertia component of ρ, the ordinary diagonal has cyclotomic exponents (j,0,−j). With HT(κℓ)=−1, the weights of σ(−1) are (1−j,1,1+j) at each embedding, hence (0,1,2) in V. These weights and the symmetric pairing give the normalization of Newton–Thorne’s adjoint. I attachment · V attachment · Polarized action · Newton–Thorne §1

Image at the specified coefficient prime. In the marked deformation frame σ is absolutely irreducible on GK by type I. The pinned open-subgroup lemmas use the local flag and the diagonal residual trace to upgrade this to absolute irreducibility on every open subgroup of GK. To see the test, let U be the normal core of an open subgroup. A power of a local inertia element with cyclotomic value 1+ℓ lies in U, kills the finite character and has three distinct eigenvalues. Any U-stable proper nonzero subspace is a sum of its eigenlines. If it is a plane, its intersection with a distinct GK-translate gives a U-stable line, since a GK-stable plane would contradict type I. The GK-orbit of the line then consists of all three eigenlines, so the global matrices permute them transitively. The local decomposition group preserves the ordinary flag, which is spanned successively by these eigenlines in their flag order. It therefore acts diagonally in this basis. Take a global element g inducing a three-cycle and a local element x with u=Ψ‾(x) and u2≠1. All three matrices of gxn, for n=0,1,2, have trace zero. Reduction in the original integral model with v=Ψ‾(g) would make the three distinct values vun roots of T2+T+1, a contradiction. I open-subgroup lemma · V open-subgroup lemma

In I the square of the finite inertia component is trivial, including at ℓ=q,d, and ℓ−1∤2j supplies x on inertia. In V, both 3 and 7 are inert in K and prime to N. Take a local arithmetic Frobenius lift of norm ℓ2 and adjust it by inertia to have κℓ(x)=1. Cyclotomic inertia is surjective over the unramified local quadratic extension. The odd quadratic factors of ρ are trivial on ℓ2. For dyadic conductor 2b the exact order of the even dyadic factor at ℓ2 is 2b−v2(ℓ2−1), since the character is faithful on the cyclic group 1+4Z/2bZ. Here v2(ℓ2−1) is 3 at 3 and 4 at 7, and b≥14,18, respectively. Reduction preserves these orders. The root-counting argument thus applies even in characteristic 3. The same x also witnesses that the upper triangular ratios of the characteristic-zero diagonal characters differ from κℓ in the V tangent test. Count I conductor and inertia conditions · V local element and inertia subgroup

The identity component of the Zariski closure of σ(GK) acts absolutely irreducibly, since its inverse image is open. Its unipotent radical fixes a nonzero subspace, so it is reductive, and its center acts by scalars. Its connected derived group is therefore irreducible in dimension three. The possibilities for its semisimple Lie algebra inside sl3 are sl3 and sl2 acting by the symmetric square. Two simple factors would require dimension at least four for a faithful irreducible action. Both possibilities have torus elements with three distinct eigenvalues. The Zariski closure of the image of GK(μℓ∞) contains this derived group, since the cyclotomic kernel contains commutators and the commutator map respects Zariski density. The Tate twist is trivial on that kernel. After a finite coefficient enlargement to split the characteristic polynomials, Newton–Thorne Lemma 2.28(2) gives an enormous image in the sense of Definition 2.23. The argument applies over any finite extension of K by the same open-subgroup test. Newton–Thorne Definition 2.23 and Lemmas 2.22, 2.28

The finite tangent conditions. A dual-number deformation (1+ϵy)R of the polarized point R gives a cocycle with values in A. Away from ℓ, the fixed inertia conjugacy class gives an inertia coboundary by differentiation. Where the frame instead kills a finite-index subgroup of inertia, restriction and corestriction in characteristic zero give the same conclusion. At ℓ in I the map of the local deformation ring to the dual numbers kills the specified local ideal. The crystalline comparison for that ideal, applied to the dual-number lift, gives crystallinity over the open subgroup killing the sign and the finite local character. In V, the frame gives a triangular dual-number lift over that subgroup with the fixed cyclotomic powers on inertia. The diagonal blocks are crystalline, since their remaining deformation is unramified. The graded Hom modules for extending the flag have positive cyclotomic exponents j or 2j, with the ratio-κℓ case excluded by x above. The Hom module in each extension step has Fil0=0 and H0=H2=0, by the graded characters and local Tate duality. The local Euler characteristic and Bloch–Kato dimension formula identify Hf1 with H1 for these extensions, so the dual-number lift is crystalline over that subgroup. In both cases its self-extension class becomes a coboundary over Bcris. Restriction and corestriction for the finite local extension give the Bloch–Kato condition over the original field. These are the unramified and crystalline kernels of Newton–Thorne §1. I local ideal and tangent comparison · I inertia and crystalline kernels · V inertia comparison · V crystalline tangent comparison

Vanishing given attachment. Newton–Thorne, Adjoint Selmer groups of automorphic Galois representations of unitary type, Theorem 5.2 (also Theorem A for unitary type), now gives Hf1(Q,A)=0. Finite extension of the coefficient field commutes with the local kernels and detects vanishing. If the polarized attachment is supplied over a finite CM extension L/K, the theorem gives vanishing over L+ and restriction injects the original rational Selmer group there by corestriction. Neither step requires a degree prime to ℓ. The finite tangent cocycle is therefore a global coboundary, corresponding to an infinitesimal change of basis. This gives the type-I rigidity used in (VI.6) and the rigidity test of the V frame used in (VI.12). Newton–Thorne Theorem 5.2 · I cotangent test · V rigidity test

The first construction still needed for this deduction is a polarized cuspidal attachment of σ(−1) at the prescribed coefficient place for each type-I point, including ℓ=3,7 in V. The residual semisimplification is a sum of three characters, so restriction or coefficient enlargement leaves it reducible. In particular BLGGT Theorem 4.5.1, which assumes residual irreducibility on the mod-ℓ cyclotomic kernel, cannot provide the attachment directly. At 3 its numerical lower bound ℓ≥2(d+1) also fails even for residual constituent dimension d=1. The source routes the attachment through an abelian CM field L containing K and then uses solvable automorphic descent. For descent of the attachment, BLGGT Lemma 2.2.2 requires the original polarized representation over K and irreducibility over L, supplied here by the open-subgroup test. The trace count further requires the local level witness over K. An attachment over an arbitrary finite CM extension suffices for Selmer vanishing, while the pinned descent and level comparison use the linked field and its specified local conditions. BLGGT Theorem 4.5.1 · BLGGT Lemma 2.2.2 · Source construction in I and V · Source descent and local conditions

For the trace count the auxiliary field must also carry a uniform collection of systems. To see the distinction, suppose a single finite Galois extension L/K works for all points under consideration, with the probes split completely and the upper levels controlled. At such a probe the upper and lower traces agree. A representation which stays irreducible on GL has at most ∣Gal⁡(L/K)∣ possible extensions up to isomorphism once its restriction is fixed. Indeed Schur's lemma makes the ratio of two extensions a character of the quotient. Thus a single Galois-stable list upstairs, counted with this multiplicity, can bound the required product downstairs whenever its size and the field degree meet the budget. Potential automorphy over fields depending on the point gives no such numerical conclusion by itself. For the linked field in the release the residual outer characters are paired by duality, with a marked element on which they differ. The source uses support on the prescribed ordinary Hecke module for the rank-three lifts and a separate symplectic lifting theorem for Count II. Their precise settings and the solvable descent step are set out in the supplement. Higher-rank attachments and descent

How a rank-three seed reaches the other irreducible points

The Count I seed and its base-level support conditions are given in §3.6. For I and V the rank-three argument begins with a supported cuspidal adjoint point over the chosen linked field L. The projected Hecke quotient is finite and torsion-free over its weight algebra. The same support problem then occurs at characteristic-ℓ curves with the prescribed ordered flags. Seed and weight computation

One required characteristic-ℓ vanishing admits a short proof, also for ℓ=3. Let A be the matrix of the marked decomposition-group element on a residual curve over k′[[t]], and u its cyclotomic value. The characteristic polynomial P of A reduces to (T−c)n, with c≠0 and uˉ≠1. If uAY=YA, then P(uA)Y=YP(A)=0. Modulo t, uˉA−cI is the sum of the nonzero scalar (uˉ−1)c and a nilpotent matrix. Thus P(uA) is invertible modulo t, hence over k′[[t]]. This kills Y, also over the fraction field. Marked curves and weight ideals

The support argument has a delicate step at a reducible curve. At adjoining marked curves the procedure completes the same finite patched module. At each completion the regular sequence of patch and weight parameters and freeness of the augmentation fibre give dimension 3[L+:Q]+1+dpatch for each minimal support component. The Galois deformation estimate at a suitable adjoining prime places its lift above a minimal support prime with irreducible generic determinant. Contract this support prime to the common ring and lift it to the next completion. Flatness after localization supplies a lift. Its residue-space fibre of the common module is nonzero because it is an extension of a nonzero vector space. Passing to a minimal support prime there recovers the dimension needed at the next junction. At the last curve the augmentation-fibre identity puts the desired irreducible prime on the original Hecke support. This uses a common patched module throughout the chain. Dimension, contraction and lifting on support

The local charts used in this argument connect adjacent components in dimension at least 3[L+:Q]+1. The source proves the corresponding global propagation using the connected core of the framed space and the rank-two support on the reducible locus. Given these chart and tower estimates, an integral irreducible lift then has nonzero fibre in the finite Hecke module by Nakayama. Arithmetic classicality identifies its determinant and ordered inertia characters with a cuspidal specialization, giving the required weights (1−j,1,1+j). Section 3.8 gives the integral comparison on the definite-unitary tower and states the spectral hypotheses for its classical evaluations. From local charts to each integral lift · Precise arithmetic inputs

The rank-four seed in Count II

Let F=k0 be the selected real quadratic field, split at 2 and at the odd coefficient prime ℓ. Label the real embeddings by ι and the two places p±∣ℓ. The restricted odd dyadic character ξ has primitive conductor n=2nIIOF, since the dyadic completions are Q2. The tower gives ξˉ2≠1 and ξˉ(Frob⁡p±)≠1 at both places. The refined vector must have level K0(np+p−) with the finite nebentype of ξ. The cohomological parameter is fixed at w=k+1 while the classical weights become (k+1+2t0,k+1), with k,t0≥2 divisible by ℓ−1. The desired representation r has determinant ξκℓk and residual semisimplification 1⊕ξˉ. Its ordered local requirements are the following. Detailed Hilbert seed calculation Seed predicate

Place Inertial cyclotomic exponents, line then quotient Full residual characters, line then quotient
p+ (k+t0,−t0) (ξˉ,1)
p− (k,0) (1,ξˉ)

Parallel cuspidal congruence. For each unramified p, define the finite ideal character by η(p)=ι(ξ(gp−1)), with gp arithmetic Frobenius. The weight-one Hilbert series E1(1,η) has coefficients of positive index ∑d∣mη(d), using zero at ideals not coprime to the conductor. Stabilize at both places using (1−Vp+)(1−η(p−)Vp−), with V shifting ideal coefficients. This gives the ordered U-roots η(p+),1. The first subtraction kills the constant terms at the standard narrow-class expansions and leaves the coefficient of the unit ideal equal to one. To raise the parallel weight, divide the level-one series Ek(1,1) by its constant term 2−2ζF(1−k). The other coefficients are integral. Colmez's residue theorem and Deligne–Ribet interpolation give vℓ(ζF(1−k))<0 for sufficiently divisible positive k on the trivial branch. The regulator is nonzero because the split local logarithm of a generator of the real quadratic unit group is nonzero. The Euler factors at both ℓ-places are units. Thus the raising form is congruent to one at the standard expansions. The Hecke recurrences on the reduced coefficients of positive index agree at the two weights, since (Nq)k≡1 mod ℓ at good primes and the level operators act by shifts. Deligne–Serre lifting on the saturated coefficient lattice retains both Up eigenvalues, the dyadic operators and the finite character. Dasgupta–Darmon–Pollack, Proposition 2.1 Colmez Deligne–Ribet Deligne–Serre, Lemma 6.11 Finite ideal character

A characteristic-zero Eisenstein system at this weight and level would have residual inducing characters {1,ξˉ} by good traces. At either split ℓ-place the local conductor exponents sum to at most one and the central character is unramified, so both inducing characters are unramified. The unit root at both places would be furnished by the same global inducing character, the one whose root lacks the factor ℓk. The two prescribed residual roots and the nontriviality tests at both places exclude this. The lifted packet is cuspidal. Its dyadic U-eigenvalues are units, by reduction to one. This argument and the construction of the parallel packet apply at ℓ=3. Dasgupta–Darmon–Pollack, Proposition 2.1 Colmez Deligne–Ribet Deligne–Serre, Lemma 6.11

Mixed weight at fixed level. Write b=k+1, m=b−2, and tv+=t0, tv−=0 for the two places over ℓ. Every positive multiple t0 of ℓ−1 gives a mixed cuspidal eigenpacket at the original K1(np+p−) with fixed finite-order central character ω and no extra finite character at either ℓ-place. Here K1 means bottom row (0,1) modulo the level. Equivalently, with fixed ω one may use K0 with the prescribed lower-right-entry character at the dyadic places. At v∣ℓ, Iv=K0(v)=Z(Ov)K1(v) and ωv is unramified, so these vectors are Iwahori invariant there. Prescribed level and scalar character Detailed ordinary calculation

Take the quaternion algebra D/F ramified at the two real places and split at every finite place. Classical Jacquet–Langlands preserves the finite local representations and sends holomorphic weight b+2tv to the definite type of symmetric degree m+2tv on the norm-one group. For an integer ring O of a sufficiently large finite coefficient field, let ρt act on

Mt=⨂v∣ℓdet⁡−tvSym⁡m+2tvO2.

Each local scalar acts with power m. This is the Johansson–Newton algebraic coefficient with common central weight m. At infinity the algebraic type is normalized by nrd⁡−m/2, so its real scalar character is sgn⁡b. If f∞(g) is the vector in the real-type model transported by ι−1, passing to f(g)=∣nrd⁡g∣f−m/2ρt(gℓ)−1f∞(g) makes it left D×(F)-invariant. Use the product K of the original tame K1 and the two Iwahori groups. In this model impose

f(guz)=Θ(z)ρt(uℓ)−1f(g),Θ(z)=ωf(z)∣z∣f−m∏v∣ℓzv−m,

for u∈K and finite central ideles z. Here ∣⋅∣f is the product of the finite normalized absolute values and character values are read in the chosen coefficient field. The norm factor cancels the valuation of the ℓ-adic power, so Θ is unit-valued. The product formula and ω∞(−1)=(−1)b at both real places give Θ∣F×=1. On K∩Zf it agrees with the prescribed right action, using the scalar coefficient power and triviality of ω on scalar units in K. Thus the full central character is fixed integrally, including its unramified values. Jacquet–Langlands, Chapter XVI Johansson–Newton, §§2.1–2.4

Evaluation gives fibers over the finite set X=D×(F)\D×(AF,f)/(KZf). For a representative g, the stabilizer is (D×(F)∩gKZfg−1)/F×. It is finite because the rational projective group is discrete in the adelic adjoint representation, with compact local bounds here, including at both real places. At a split dyadic place it injects, after conjugation, into the projective image of K1(2a), with a=nII≥1. The upper Borel over F2 has order two and the principal congruence kernel is pro-2, so that projective image is pro-2. Each finite stabilizer is therefore a 2-group. Averaging its action on the fiber, including the central scalar, is integral at every odd ℓ. The resulting lattice of functions is finite free and its reduction is the full function space for the reduced coefficients and characters. Norm-two stabilizer condition

In the pinned convention the raw operator hb is the double-coset sum with prefactor (Nv)b/2−1=(Nv)m/2, with b fixed. On passing to the algebraic function model, each right translate contributes ∣nrd⁡h∣fm/2, which cancels that prefactor on a determinant-uniformizer coset. At v∣ℓ the remaining ordinary normalization is ℓtvhb. For the upper Iwahori use the pinned representatives

hv(x)=(ℓx01),x mod ℓ.

The K1 and Iwahori sums use the same representatives on our invariant vectors. The term acting on the coefficient is ℓtvdet⁡(hv(x))−tvSym⁡m+2tv(hv(x)). The determinant factor cancels exactly. More generally for h∈Ivdiag⁡(ℓr,1)Iv with r>0, the integral factor is (ℓ−rdet⁡h)−tv. Modulo ℓ each such h has zero first column and lower-right entry dˉ≠0. Its symmetric power factors through the quotient line generated by e2m+2tv, and on that quotient the normalized action reduces to dˉm. The Iwahori action on the quotient also reduces to dˉm, since its extra unit powers have exponents divisible by ℓ−1. This uses Fv=Qℓ. In the pinned coordinates the image of an individual representative is spanned by (xˉe1+e2)m+2tv. Pinned double coset and raw operator Weight normalization Johansson–Newton, Proposition 4.1.1

Let Ct0 be the reduction of the definite function lattice. Project in each symmetric power to the coefficient of e2m+2tv, which defines a map pt0:Ct0→C. For diagonal entries (aˉ,dˉ) of an Iwahori element modulo ℓ, this coefficient quotient has character dˉm(dˉ/aˉ)tv=dˉm. Thus C is the same scalar-valued function space at every weight under consideration, with central law Θˉ and right K-action through ∏v∣ℓ(u22,v‾)−m. In particular dim⁡C≤∣X∣. The coefficient projections are onto, and averaging over the finite 2-group on each fiber commutes with them and preserves surjectivity, including at ℓ=3.

The map pt0 intertwines all chosen Hecke operators. At either ℓ-place the normalized representative has bottom-right entry 1, so projection after its coefficient action equals projection before it. Away from ℓ, the translates have identity coefficient component at both ℓ-places and the same central law. Write U=Uˉv+Uˉv− for the product of the two normalized residual operators. Each summand of U acts through the tensor of the two coefficient quotients on the function value at its input, so U kills ker⁡pt0. This holds on equivariant functions before choosing representatives of X.

For sufficiently large N, im⁡(UN) has stabilized on each space, and U acts invertibly on this image. The projection maps the stabilized image in Ct0 onto that in C. It is injective there because U kills its kernel. Since the two operators commute, their product is invertible precisely on the summand where both are invertible. The resulting isomorphism intertwines the full Hecke action, including on generalized simultaneous eigenspaces over an algebraic closure of the residue field. A primitive lattice vector for the parallel packet has nonzero reduction in this ordinary summand. Thus the generalized ordinary multiplicity dt0 of its full residual system satisfies

1≤dt0=d0≤∣X∣.

Equivalently, dt0 is the sum of the generic-fiber multiplicities of the systems reducing to this residual system. After extending coefficients to split the generic action, intersect a common invariant flag with the lattice. The intersections are saturated, so the one-dimensional graded pieces reduce with the corresponding residual characters. Their number in the chosen residual system is the dimension of its generalized eigenspace. The isomorphism is at the residual level, and the sum can include several distinct characteristic-zero systems. Independent trace proof with higher congruence precision

Apply Deligne–Serre lifting to the commutative algebra of all these operators on the mixed finite free lattice. The algebra is finite and torsion-free over O as a submodule of its endomorphism algebra, so the simultaneous residual character lifts through a characteristic-zero component after extending coefficients. This gives one eigensystem with ℓtvλv′≡λv at both v∣ℓ and λv′≡λv away from ℓ, where λv′ is the raw hb eigenvalue. Both dyadic unit roots survive. All symmetric degrees are positive, which excludes constituents factoring through reduced norm. Reverse Jacquet–Langlands and select the holomorphic real lines to obtain the mixed cuspidal form with the prescribed finite types and refined level. This proves the fixed-character weight change for the Count II field, including at ℓ=3. Deligne–Serre, Lemma 6.11

Local readout given the mixed eigenpacket. The source first attaches r1 with determinant ξκℓ−k and then sets r=r1⊗κℓk. Put tp+=t0 and tp−=0. Before twisting, the stable line has exponent tp and the quotient has exponent −k−tp. After twisting these exponents become k+tp and −tp. Use HT⁡(κℓ)=−1. The pinned ordinary comparison reads the stable line: after removing its cyclotomic power from r1, the geometric Frobenius value is the normalized operator ι−1(ℓtpλp′), where λp′ is the raw eigenvalue. The normalized operator reduces to ξ(Frob⁡p+−1),1 modulo the coefficient maximal ideal in the geometric convention. Inertia, with trivial extra finite character and ℓ−1∣k,t0, and arithmetic Frobenius then determine the line reductions ξˉ,1 on the full decomposition groups. The determinant gives the opposite quotient reductions in the table. Only the Galois comparison for the classical point enters this readout. Ordinary comparison Tate twist Stable-line identity

At a dyadic place the primitive central conductor and the level give the same lower and upper bound nII on the local conductor exponent. The unit U on that newvector gives a nontrivial local standard L-factor, forcing a principal series with one unramified character and one of conductor exponent nII. Local–global compatibility gives inertia 1⊕ξ and zero monodromy. Away from 2ℓ the representation is unramified. At p±, Iwahori level allows spherical principal series or an unramified twist of Steinberg. The ordinary comparison gives Frobenius slope gap k+2tp≥2, excluding Steinberg, which has gap one. Hence the exponent-one level at ℓ is the level of the refined vector on a spherical representation. These deductions use the classical local attachment and give the prescribed principal-series local conditions used in the induction. Casselman BLGGT, Theorem 2.1.1

Absolute irreducibility of the cuspidal Hilbert attachment upgrades to irreducibility on open subgroups. If an open normal core preserves a line, an inertia element with distinct eigenvalues forces it to preserve at most two lines. Global irreducibility then gives two lines interchanged by the group, and hence a quadratic self-twist δ. Reduction of the twist gives {1,ξˉ}={δˉ,δˉξˉ}. Since ℓ is odd, this would imply ξˉ2=1. Thus the open-subgroup test also works at 3. Saito, §2, citing Taylor, Proposition 3.1

For s∈GQ∖Gk0 representing the marked coset, the induced representation restricts to r⊕rs over k0. In the induced basis s acts by (0r(s2)I0). Since ℓ is odd and the two residual eigenvalues of r(s2) are distinct, the eigenspace projectors are integral in the selected lattice. The change to the four marked lines exhibits the required alternative of residual entries. Before its common twist the four exponents in decreasing order are k+t0,k,0,−t0. The two inducing planes remain inequivalent on the relevant CM field. An isomorphism up to a character on the cyclotomic quotient would, by equality of the determinants, have a quadratic twist. Squaring a suitable inertia element with infinite cyclotomic image would then equate the two distinct pairs of powers above. Given the irreducibility of the two restrictions, Clifford theory gives irreducibility of the induction, and quadratic automorphic induction after base change gives a cuspidal rank-four seed. The target point at twist j has exponents (j+1,j,1,0). The passage from the seed to all of those target points uses a projected rank-four Hecke module, the marked residual support estimate, and classicality at the target weights, using the ordered Hecke comparison of §3.8. Residual entries and symplectic lattice · Conditions on the projected module and descent

3.8. Classical local–global compatibility for the linked ordinary towers

Integral arithmetic specialization. Fix one Count II link on the definite unitary group over L+, with CM field L. Write A for the complete local ordered inertia algebra over the coefficient DVR O. At each ℓ-place use the marked split GL4 coordinate. Fix the away-from-ℓ level U, either at the base or at a prescribed finite patch level, and keep the integral tame coefficient representation τ0 at the split tame places, including the full dyadic type. Its characteristic-zero type test is the prescribed Teichmüller inertia tuple. On forms it is imposed by equivariance. One can work at the kernel of τ0 with that action of the finite quotient, without dividing by its order. The auxiliary split place v0 has principal level one and lies over an odd rational prime p0≠ℓ unramified in L, with residue degree one. These are the hypotheses for the stabilizer and finite-level ordinary module used below. Integral tame type · Finite-level setting and character · Auxiliary prime

The stabilizers in the double-coset description are trivial. For a rational stabilizer, its powers are bounded at every finite place by the compact levels and at infinity by definiteness. Their entries lie in a single fractional lattice with bounded archimedean coordinates, hence the matrix has finite order. At v0 it is conjugate into 1+pw0M4(OLw0), a torsion-free group by the matrix logarithm over the unramified completion at the odd prime p0. This gives arbitrary integral coefficient values at the finite set of representatives. At ℓ=3 the proof uses the same principal level and integral coset sums. No averaging over a potentially noninvertible stabilizer or type quotient is required. Trivial stabilizers

Let Θ be the weight character on the diagonal units. With inverse-Artin coordinate y for an inertia element g at w, the i-th diagonal factor is Θ(ew,i(y))=(ψw,i(g)κℓ(g)i)−1, for 0≤i≤3. At an arithmetic map x:A→O′ into an enlarged coefficient DVR, write xΘ=ϕχλ, with χλ the lowest character of the algebraic coefficient ξλ and ϕ finite. The smooth right translation character is ϕ−1. In particular the scalar-unit law includes κℓ(g)6∏ixψw,i(g) in the denominator of xΘ(yI4). The tame scalar action is included in τ0, and the full determinant step above ℓ is the normalized central uniformizer translation. For the target exponents (j+1,j,1,0) and common linked cyclotomic shift bv, the lowest tuple hv,i−i is (−j−1−bv,−j−1−bv,−3−bv,−3−bv), nondecreasing for j≥2. The seed exponents (k+t0,k,0,−t0) also give a dominant algebraic coefficient in the dual convention after the common shifts. Ordered weight and residual characters · Arithmetic compatibility · Coefficient weights

At w∣ℓ, let N be the integral upper unipotents and Nc− the lower unipotents congruent to 1 modulo pwc+1. Put zw=diag⁡(πw3,πw2,πw,1),Ua=∑u∈N/(aNa−1)R(ua) for a diagonal a with nonincreasing uniformizer exponents. The coset sums compose as UaUb=Uab. In factoring yu for y∈Nc− in upper, diagonal and lower order, the resulting diagonal is congruent to 1 at depth c+1. The permutation of the upper cosets leaves the lower factor conjugated by a−1. For a=zw, conjugation gains πwi−j in each entry i>j. Thus Uzw lowers depth c+1 to c once the smooth character is trivial on the depth-c+1 diagonal units. The three proper step diagonals have product zw, so all their normalized operators are invertible on the ordinary part. The fourth step is central and invertible as well. Contracting and step diagonals · Coset multiplication · Level lowering

Let Cc be the classical lattice at this U,τ0,ϕ and at depth c at least the coefficient-lattice bound, with ϕ trivial on diagonal units congruent to 1 modulo pwc+1 at every w∣ℓ. It is integral in the untwisted coordinates F~(g)=ξλ(gℓ)−1F(g). The control map takes the lowest-weight coordinate in the integral weight basis, keeping the τ0 index. Upper unipotents fix this coordinate functional. In a normalized translate the coefficient matrix is ξλ(u)χλ(a)−1ξλ(a). It induces the identity on the lowest coordinate and, for a=zw, its columns away from the lowest line are divisible by the uniformizer of O′ in the local coefficient factor. Indeed the difference from the lowest weight is a nonzero sum of positive roots, each pairing positively with (3,2,1,0). Use the product over the marked ℓ-places and all labeled embeddings. Each nonlowest tensor column is divisible in at least one factor, and a scalar factor has only its lowest coordinate. Thus the product sends an integral form whose control value is divisible to a form divisible in every coordinate. The map intertwines all the step operators and good spherical operators, and its torus law is (ϕχλ)−1. Integral weight basis · Operator intertwining · Filtration contraction

On the ordinary summand, where the contracting product is invertible, this divisibility gives injectivity modulo the uniformizer and then integrally. For surjectivity modulo the uniformizer, go to a depth c′≥c at which both reduced actions of the lower unipotents are trivial. Prescribe the scalar ordinary values in the lowest coordinate at the finite representatives, with zero in the other coordinates. Equivariance extends these to an integral classical form with the prescribed reduced control values. A sufficiently large power of the contracting product moves it back to depth c. Choose the power to be a multiple of the period of the scalar ordinary vector modulo the uniformizer, so the control values still agree. Eventual factorial powers give the ordinary projection and preserve that agreement. Nakayama gives integral surjectivity. With finite-level ordinary base change ordBC and freeness ordStruct, this gives the Hecke-equivariant isomorphism ezalgCc ≃ O′⊗A,xSAord(U,τ0,Θ). The ordinary lattice is the same under increase of admissible depth by the lowering identity. The comparison applies even when smooth characters coincide. Integral approximation · Ordinary freeness · Integral base change · Arithmetic specialization

Residual projector and characteristic-zero fiber. In the finite commutative image of the chosen operators over A, take the Henselian idempotent ea for the residual profile a, with zero if that profile is absent. The profile includes the good split spherical coefficients and the ordered cumulative unit values at ℓ, including the cyclotomic corrections to ψw,i. Over a finite field of size Q splitting the residual characters, a finite list of operators separates the desired profile from the others. The product ∏i(1−(Hˉi−aˉi)Q−1) has residual values one and zero on the selected and unselected characters. A sufficiently large factorial power gives the idempotent in the residual image algebra, even for nonsemisimple actions. Every residual character of the specialized image pulls back to a residual character of the original algebra with the same values. By uniqueness of idempotent lifting, the integral control map therefore restricts to ea on both sides. The projected module MU=eaSAord is finite free over A. Every characteristic-zero Hecke character of its image algebra above x has a nonzero generalized eigenspace on the classical weight fiber by faithfulness and Nakayama, and hence admits a simultaneous eigenvector. Here the conclusion is a vector on the definite unitary group with the prescribed level, type and normalized ordinary eigenvalues. Integral idempotent · Projected freeness

At a fixed algebraic weight, finite-order variations in the free local unit directions give separating arithmetic evaluations on A. Distinct sufficiently large ℓ-power orders in one direction for each diagonal index at each place give distinct smooth characters and preserve the residual profile, also at 3. Weierstrass preparation applied in each variable gives separation on this grid. Freeness then detects a nonzero vector by one nonzero coordinate, and detects a nonzero operator by its action on a vector. The tame and patch levels are fixed throughout, while the admissible ℓ-depth may depend on the smooth character. Density at a fixed algebraic weight Detailed reader proof and pinned source calls

The cuspidal Galois row. Fix ι:Q‾ℓ≃C. For a polarized regular algebraic cuspidal π of rank a over a CM field E, use πc≃π∨⊗(χ∘NE/E+) with χv(−1)=(−1)a at every real place. The required local normalization is rec⁡w(πw∣det⁡∣w(1−a)/2), with a uniformizer corresponding to geometric Frobenius. Its labeled Hodge–Tate numbers are (a−1)/2−z for infinitesimal-character entries z. Barnet-Lamb–Gee–Geraghty–Taylor, Theorem 2.1.1, gives those lists, spherical crystallinity above ℓ and Frobenius-semisimplified Weil–Deligne compatibility away from ℓ, retaining monodromy. This proves the two away-from-ℓ tests of the pinned cuspidal row. The toral test has zero monodromy. In the large tame test Ew=Qp, with a=1 or p>16a2, a repeated eigenvalue of a tame generator must be a (qw−1)-st root of unity. The full comparison gives its inertia polynomials and zero monodromy. At w∣ℓ, Theorem 1.1 of BLGGT II gives equality of semisimplified Weil representations and a bound on the monodromy partitions in general. If the automorphic monodromy is zero the bound forces Galois monodromy zero, even in even rank without Shin regularity. Rank one follows from reciprocity. Caraiani's full result applies in the conjugate-self-dual case, or after the algebraic character twist of Clozel–Harris–Taylor, Lemma 4.1.4, followed by untwisting. These results include ℓ=3 and concern the characteristic-zero cuspidal attachment. Cuspidal row BLGGT, Theorem 2.1.1 BLGGT II, Theorem 1.1 Caraiani

Constituents of the projected classical space. The linked unitary group is compact at infinity, has hyperspecial level at nonsplit places of the unramified CM extension, and uses the marked general linear coordinate at each split place. Every contributing constituent in its classical evaluations has a standard transfer ⊞bPb⊠[db]. Here Pb is unitary conjugate-self-dual cuspidal of rank ab and ∑babdb=n. Write tb,k=(db−1)/2−k and sb,k=tb,k+(ab−n)/2 for 0≤k<db. The regular infinitesimal-character list is the union of the strings of Pb shifted by tb,k. Shifting by sb,k instead makes Pb∣det⁡∣sb,k regular algebraic in rank ab. Its polarizing character on E+ is ∣⋅∣2sb,kηE/E+ ab, of sign (−1)ab. Apply the cuspidal row to each of these regular algebraic blocks. The direct sum has at each split place the block exponent sb,k+(1−ab)/2=tb,k+(1−n)/2. Its Weil–Deligne monodromy is the sum of the cuspidal block monodromies. In particular a string of shifted trivial characters has zero monodromy. Reversing the strings at conjugate embeddings gives the polarization of the assembled rank-n representation with multiplier κℓ1−n in its conjugate-duality relation. The linked comparison character is a separate common algebraic twist of the ordered pair and multiplier. It shifts the labeled weights and twists the entire local parameter. Its central and finite-character data are the ones fixed by the link. No independent change of twist at a patch place is used. Assembly of the classical row Block calculation and common twist

The release supplies this transfer for every irreducible constituent σ at the base level or a fixed auxiliary level. In RunFin, run the comparison with the empty split test or with a single prescribed split place w. In either run the finite candidate list has the compact real infinitesimal characters of σ and the split match almost everywhere. For every finite set of spherical nonsplit places, some candidate matches throughout that set. If each candidate failed somewhere, one failing place per candidate would give a finite set no candidate could match. Thus one can take a datum D from the empty run and a datum Dw from the run at w, both matching at every spherical nonsplit place. They have the same Satake data almost everywhere by FullSat and are regular by the weight comparison. The pinned LocRigid identifies their blocks at all finite places up to permutation, so PermSub transports the matching split subquotient of Dw back to D. This proves the split comparison for one family at every w, together with the inert spherical comparison and the real weight of D. The pinned baseChangeN supplies BaseChangeIsobaricN to the ordinary construction. In GuWrap, irreducibility of normalized products of unitary local factors upgrades the subquotient to the full product, and the Speh description of the discrete general linear factors gives BcSpeh. Finite candidates and local rigidity Pinned base change Speh assembly Proof for the base and auxiliary levels

Good arithmetic weights and Speh strings. At one marked split w∣ℓ, write k=Lw and let η1,…,η4 be the inverses of the smooth weight characters on Ok×. At a good weight they are distinct on the full unit group, with arbitrarily high conductor allowed. The ordinary Jacquet comparison places each local component of the selected eigenspace in the torus Bernstein class determined by their unordered multiset. Let zs be that class's central idempotent over C. For a compact open J fixing a selected vector, zs∗eJ is a compactly supported Hecke test and fixes the vector. It preserves the selected refinement since zs acts as the identity on its whole smooth block. This projector is defined at the individual weight. The integral residual projector ea continues to specify the projected module. Pinned ordinary Jacquet comparison Roche, §7 and Theorem 7.2 Bezrukavnikov–Kazhdan–Varshavsky, §1.3

At matching split transfer, Pb[db] has the supercuspidal support of Pb,w repeated at each of the unramified shifts ∣det⁡∣(db−1)/2−j, 0≤j<db. Induction takes the union with multiplicity, and subquotients retain this support. A supercuspidal factor of rank greater than one gives a nontoral support Levi. If all factors are characters, db>1 repeats the same unit restrictions. The modulus and rank normalizations are unramified. An auxiliary determinant twist in an endoscopic embedding is common to the copies within one string, so it preserves the repetition. The regular torus projector therefore acts by zero on the whole matched induced representation, including when the trace inserts an intertwiner. The argument also excludes a repeated copy of the same length-one global block. KMSW, §§1.2.2 and 1.3.2 Zelevinsky, §1 and Theorem 4.2

The parameter-refined trace at the marked place. Fix a weak standard parameter Ψ, indexed by its almost-everywhere Satake data and archimedean norm as in KMSW Corollary 3.3.2. Put v=w∣L+ for the marked place and take fv=zs∗eJ. Hold the other factors fv fixed. Take compact coefficient projectors at infinity, the normalized hyperspecial units at nonsplit places and at good unmarked places, and compact tests fixing the chosen vectors at the remaining split places. These include the kernels or isotypic projectors for the tame types, principal level one at v0, any other coefficient places with level, and the actual auxiliary patch levels. The set removed from the unramified Hecke refinement contains all these exceptional places and infinity. Transfer uses the chosen pure inner twist of the definite Hermitian group and hyperspecial normalization at the nonsplit places. KMSW, (3.3.1)–(3.3.7) and Corollary 3.3.2 Mok, Corollary 10.0.2 and Proposition 10.0.3

At a split place the proper elliptic datum H=U(a)×U(4−a) is the block Levi of GL4(Lw), and the identity datum uses the same function in the common coordinate. For H take normalized parabolic descent, with compact conjugation average and half-modulus. If the embedding twists the Levi representation by the determinant character ξH, multiply the descended test by ξH in the convention π(f)=∫f(g)π(g) dg. Its trace on a Levi representation is then the trace of the original test on normalized induction of the twisted factors. The matching orbital-integral identity uses the relative discriminant. At the other finite places the same rule applies to split tests, and hyperspecial transfer applies at the nonsplit places. At infinity use the smooth transfer of the compact coefficient projector. KMSW, Proposition 1.5.1, §1.6 and Lemma 1.6.5 Mok, Theorem 3.2.1(a)

Definiteness leaves only the full group in the rational Levi expansion (3.3.1). In (3.3.2), the stable multiplicity formula (3.3.6)–(3.3.7) gives terms only for endoscopic parameters mapping to the same Ψ. KMSW Proposition 1.5.1 and (1.2.4) identify their stable traces at the split place with general linear characters. After the twists and normalized induction, every surviving term uses the character of the same local representation πΨ,v of the composed parameter. For arbitrary hv in the local Hecke algebra this gives tr⁡Rdisc,Ψ(hvfv)=CΨ(fv)tr⁡πΨ,v(hv). With the other factors projectors, a selected constituent contributes its local character with positive coefficient, and the trace is finite at each compact level. Independence of characters identifies this local constituent with πΨ,v. This is Mok Proposition 10.0.3 for elliptic weak indices. Anisotropy removes the proper rational Levi terms for other indices as well. Varying the split place with Ψ fixed gives the same comparison at the tame and patch places. At the marked place the support calculation makes every term on the right zero for db>1, whereas a selected vector gives positive trace on the left for hv=zs∗eJ. KMSW, (3.3.1)–(3.3.7) and Corollary 3.3.2 Mok, Corollary 10.0.2 and Proposition 10.0.3

The trace derivation uses connected-group stabilization and matching transfer in KMSW §3.1, the split character identity, and the quasi-split stable multiplicity formula of Mok Theorem 5.1.2, imported in KMSW Proposition 3.3.1. Mok assumes twisted stabilization in Hypothesis 4.2.1. Atobe et al. §§0.1 and 0.4 record the general twisted weighted fundamental lemma as a premise of that framework. The argument uses the refined distributions for non-generic indices in the length test. After the test, KMSW Theorem* 1.7.1 and Theorem 5.0.5 apply in their generic pure-inner scope. Their local classification hypothesis (3.6.3) and the intertwining relation of Theorem 2.6.2 are established there in that scope. This gives the real infinitesimal-character comparison and the inert hyperspecial comparison, via restriction of the unramified parameter to WLw. On the pinned route those comparisons are part of BcSpeh. KMSW, §3.1 and Chapter 5 Mok, Hypothesis 4.2.1 and Theorem 5.1.2 Atobe et al., §§0.1 and 0.4

Length-one constituents and their attachments. The surviving generic discrete parameters have standard transfer ⊞bPb[1] with distinct unitary conjugate-self-dual cuspidal blocks of the standard rank-four parity, conjugate-symplectic for the embedding with χ+=1. Proper endoscopic sums can contribute when their combined split support has four distinct unit restrictions after transfer and their packet members satisfy the multiplicity condition and the prescribed levels. Each Pb is individually cuspidal. With several blocks the standard transfer is an isobaric sum. The real, split and inert comparisons above are the inputs to the Galois assembly on this same fixed level. The rank shift Pb∣det⁡∣(ab−4)/2 in the classical row above then gives regular algebraic polarized cusps, with geometric normalization Pb,w∣det⁡∣−3/2 for each block. Apply rank-one reciprocity and BLGGT Theorem 2.1.1 for the higher ranks, and take the direct sum. The split Hecke polynomials come from these same local components, and the unramified factors give the representation of the Galois quotient for that level. At a good coefficient place the matching is toral with zero monodromy, so the monodromy bound in BLGGT II Theorem 1.1 gives equality there, including at ℓ=3. The ordered flag requires the separate ordinary comparison below. If the attached semisimple rank-four Galois representation is irreducible, equality of the determinant forces a single block, giving a cuspidal standard transfer. KMSW, Chapter 5 BLGGT II, Theorem 1.1 Good-weight Speh test and source anchors

The refined local identity. At a split w∤ℓ let the marked geometric Frobenius and pro-ℓ tame inertia generator have inverse-Artin coordinates xw,δw, respectively a uniformizer and a unit. Write qw for the residue size. For a generalized torus eigenspace of character φ on the prescribed Iwahori-level vectors, the geometric lemma gives the normalized inducing characters φ(ei(z))∣z∣w(n−1)/2−i, with coordinates numbered from zero. The common rank normalization therefore yields

αi=qwiφ(ei(xw)),βi=φ(ei(δw)).

If the local representation is spherical, its Frobenius polynomial is ∏i(Z−αi), including repeated roots. When αi≠αj and αi≠qwαj for all i≠j, the principal series is unlinked. Its parameter is toral with zero monodromy and the distinct Frobenius eigenlines identify the inertia scalars βi. Thus for the actual Frobenius and inertia matrices P,T the relation needed before inverting any eigenvalue difference is

∏j≠i(P−αjI)(T−βiI)=0.

This is the split patch-prime calculation in LocRelT and LevPtAlt.lg, over the actual split completion. In particular the large tame clause of the cuspidal row is a separate test. For a split spherical double coset the usual Satake factor qwj(n−j)/2 together with the polynomial coefficient factor qwj(j−1)/2 gives the same rank normalization. Pinned local relation Bernstein–Zelevinsky, §2.12 Zelevinsky, Theorem 4.2

The ordered ℓ-adic readout. With distinct smooth unit characters, the toral parameter at a split w∣ℓ and the ordinary operators give the ordered flag. The coefficient convention is the dual of the prescribed highest weight, with labeled i-th Hodge–Tate number hτ,i=lamτ,n−1−i+i. If the smooth characters are ωi, their restrictions to units are inverse to the prescribed torus eigenspace characters. The unscaled j-step operator has eigenvalue ∏i<jωi(xw)qw−i. In inverse-Artin coordinates the Galois flag character has value ι−1(ωi(z))∏ττ(z)−hτ,i. The unit-normalized ordinary operators at the split places determine the uniformizer factors on the same eigensystem. To see the first line in the potentially semistable module, twist by its prescribed character. Its Weil line is separated by the distinct smooth characters, has Newton number zero, and all labeled Hodge numbers are nonnegative. Weak admissibility forces the induced Hodge numbers on the line to be zero. This gives a weakly admissible subobject, and induction on the quotient gives the ordered flag. This is a local argument at a classical specialization. The small coefficient prime 3 satisfies the same calculation once the prescribed distinct characters and the ordinary operators are supplied. Ordinary operators Ordered flag calculation

The determinant of the projected rank-four module

The Speh test reduces the spectral requirements for constructing the fixed-level determinant to length-one transfers on the separating good weights. Each contributing good eigensystem must still supply a continuous semisimple rank-four representation of the same ΓU, with the prescribed inertia kernels at this level and all four good split Frobenius coefficients. Good-weight separation also requires scalar Hecke action on those classical fibers. Identifying the determinant with a classical representation at an arbitrary arithmetic weight uses its own pointwise attachment and Frobenius comparison. The ordered local and patch relations use their respective pointwise comparisons before separation. Pinned determinant hypothesis Good-weight interpolation

Assume the projected spectral transfer and the characteristic-zero Galois rows above at this level. Write TU for the faithful commutative Hecke image on MU and ΓU for the CM subgroup of the corresponding Galois quotient unramified outside the designated base and patch places. For MU≠0 the construction gives a continuous degree-four determinant DU on the group algebra of ΓU, with values in TU, and four ordered local characters ψw,i. The level's polarized ordinary determinant ring BU maps onto TU, compatibly with the weight map from A. Its kernel JU is Ann⁡BUMU. When MU=0 take JU=BU. Determinant construction · Hecke quotient · Auxiliary levels

For g∈ΓU write Pg(X)=DU(X−g)=∑j=04(−1)jcj(g)X4−j with c0=1. At a good split place w outside the base and patch sets write Fw for geometric Frobenius, the inverse of the source's lift acting on the residue field by the qw-power map. Let Tw,j be the unscaled spherical coset of diag⁡(ϖwIj,I4−j). Then PFw(X)=X4−Tw,1X3+qwTw,2X2−qw3Tw,3X+qw6Tw,4. Indeed if the normalized Satake parameters are bi, the coset has eigenvalue qwj(4−j)/2ej(b1,…,b4). The local shift ∣det⁡∣−3/2 gives geometric eigenvalues qw3/2bi, so their j-th symmetric function is qwj(j−1)/2Tw,j. These factors are units over O. All four coefficients are used integrally and the fourth recovers the central generator. Trace alone can lose the comparison at ℓ=3. For the generator of C2, the representations 1⊕4 and 1⊕sgn⊕3 have the same trace in characteristic three and different polynomials. At arithmetic Frobenius the roots are reciprocal and cj(Fw−1)=c4−j(Fw)/c4(Fw). Spherical normalization · Geometric Frobenius · Integral generators

The linked target is r∣GL⊗λlink for the representation r of the totally real frame. At the seed the character multiplying the roots also includes the parity factor βnpar of the link, with npar specified in the seed descent below. In general a twist by η multiplies cj(g) by η(g)j and ψw,i(g) by η(g). Returning from the linked comparison to the frame divides by the same factors. The formula above uses the linked system, with conjugate duality ryc≃ry∨⊗κℓ−3 at a classical point y. Thus the roots at the conjugate good place are qw3 times the inverse roots at w. Root scaling · Polarization

At good arithmetic weights the projected classical decomposition is scalar for the Hecke image. If an element of TU vanishes on all these characters, its action on each vector has all basis coordinates vanishing at every good weight. The integral control and weight separation above force those coordinates to vanish in A. This proves ⋂y goodker⁡y=0 in TU and hence reducedness of the fixed-level image. To construct DU, embed this compact image algebra as a closed subring of the product of the good evaluation fields. The polynomials of the classical representations in that product lie in the image at every good geometric Frobenius by the displayed formula. Those lifts, with all conjugates, are dense by Chebotarev for degree-one primes outside the excluded set. Continuity and the determinant descent identities then put the product determinant over TU. In particular this uses the polynomial-law identities on group algebras. At any arithmetic y the same good Frobenius identities give yDU=D(ry) for the semisimple representation from its classical constituent. The determinant and flag-character relations impose equations in the modular quotient BU/JU. Separation · Frobenius density · Descent of determinants · Arithmetic comparisons

For w∣ℓ, let gϖ have inverse-Artin coordinate ϖw and put Uw,0=1. For the normalized step operators, indexed by 1≤j≤4, the character of the i-th successive quotient has the values ψw,i(gϖ)=κℓ(gϖ)−iUw,i+1/Uw,i,ψw,i(g)=κℓ(g)−iΘ(ew,i(y))−1when g has unit coordinate y. Here κℓ(gϖ)=qw−1NLw/Qℓ(ϖw) is a unit. The fourth step includes the correction κℓ(gϖ)6 on the product of flag factors. At good points the ordered flag is the one in the preceding local calculation. By separation its polynomial identity Pg(X)=∏i(X−ψw,i(g)) holds over TU for all g∈GLw. At arbitrary arithmetic points it gives the local semisimplification. Realizing the prescribed flag order and making a zero-monodromy comparison there requires the applicable local hypotheses. At each typed split dyadic place the transferred inertia is the diagonal multiset of Teichmüller lifts of the four linked residual inertia characters, with zero monodromy. Its polynomial holds in the family by the same separation. The type fixes the inertia at its actual conductor depth. A dyadic Frobenius polynomial requires the determinant evaluated on a choice of lift. Ordered character formulas · Cyclotomic reciprocity · Dyadic type · Inertia action

The annihilator in an arithmetic fiber. Let x:A→O be an arithmetic O-algebra map and I=ker⁡x. On MU/IMU, the image of the determinant coefficient algebra acts through the image of the good spherical algebra. The spherical image is a finite O-submodule of the endomorphisms of the integral arithmetic fiber, so it is closed in the adic topology. Continuity and Frobenius density approximate each determinant coefficient by good spherical coefficients. Over the fraction field of O the spherical action is scalar on each classical constituent, including at weights with coincident smooth characters. Freeness makes the arithmetic lattice inject into this scalar extension. Thus if μ(t)kMU⊂IMU for t in the determinant coefficient algebra and k≥1, then μ(t)MU⊂IMU. This reducedness concerns the image of that action. An abstract fiber can have extra nilpotents. For example, if a is a nonzero parameter in a domain A, the algebra A[Z]/(Z(Z−a)) acts faithfully by Z↦(0,a) on A2. At a=0 its abstract fiber contains the nonzero class Z of square zero, while its specialized action kills Z. Reducedness modulo ℓ and the full refined action are separate questions. Spherical action · Frobenius approximation · Specialized determinant coefficient action

Detailed determinant comparison

The linked seed at the two definite unitary levels

Let R denote the marked rank-four seed on the totally real tower field F=L+, with symplectic multiplier θ. The starting input is the attached regular algebraic cusp Πseed on GL4(AL) for R∣GL⊗κseed, together with the seed's ordinary flags and irreducibility on the cyclotomic kernel. In the mixed construction this uses cuspidal solvable base change of the Hilbert form, noninvariance of the inducing cusp under the quadratic involution, and normalized quadratic automorphic induction. At the marked element s outside Gk0 the induced matrix sends (u,v) to (r(s2)v,u). If r(s2)=diag⁡(α2,β2), the vectors (βe2,e2),(αe1,−e1),(−βe2,e2),(αe1,e1) have the ordered eigenvalues (β,−α,−β,α). The paired products equal the pre-correction multiplier αβ. The eigenbasis change is integral since 2,α,β are units. At a marked element in the tower field, the scalar multiplier correction scales the four eigenvalues together. Attached seed · Marked eigenbasis

Write wi for the doubled inertia exponents of R and di=(wi−wi+1)/2≥2. The square of κseed has exponent a on open inertia, with a≡w0 mod 2. For each link K∈{K1,K2} take npar=∣w0∣ and ζK=λKβKnpar. The construction uses common λK and βK on both links and distinct auxiliary primes. At each coefficient place write e=2aK+nparbK, where λK and βK2 have respective exponents aK and bK on open inertia and bK is odd. The quotient ζK/κseed has integral exponent (e−a)/2. Reciprocity therefore gives an algebraic Hecke character whose determinant twist of Πseed is the cusp ΠK attached to RK=R∣GL⊗ζK. Its ordered Hodge–Tate weights are hτ,i=−mτ,i with mτ,i=(wi+e)/2 in the selected embedding and consecutive exponent gaps di. Parity calculation · Algebraic twist · Common characters and distinct auxiliaries

The link gives λKλKcθ=κℓ−3 and βKβKc=1. Since complex conjugation acts by an inner conjugation on the restricted frame, the symplectic pairing gives RKc≃RK∨κℓ−3. At a good place geometric Frobenius has cyclotomic value qw−1. Inverting the roots and multiplying by qw3 keeps the factor qw3/2 in the root normalization and inverts the unitary Satake parameters. Strong multiplicity one gives ΠKc≃ΠK∨. The compact real highest weight is νi=mτ,i+i, with infinitesimal entries zi=mτ,i+3/2. Each archimedean parameter character (z/∣z∣)2zi has conjugate-duality sign −1, its value on j2=−1 in the real Weil group. The entries are distinct, so the sign of this sum is uniquely conjugate-symplectic. KMSW Lemma 1.2.5 and Propositions 1.3.1 and 1.3.3 select κ=+1 and χ+=1 for the rank-four embedding. The algebraic highest weight gaps di−1≥1 verify Shin's slightly regular hypothesis. Shin gives finite-place temperedness, and BLGGT II Theorem 1.1 or Caraiani gives the full Frobenius-semisimplified comparison at ℓ, with rec⁡geom(ΠK,w∣det⁡∣−3/2). This comparison permits ramified diagonal characters at the coefficient places. Polarized pairing · KMSW, §§1.2–1.3 · Shin, Theorem 1.2 and Corollary 1.3 · BLGGT II, Theorem 1.1 · Caraiani, Theorem 1.1

The actual unitary group is U(I4) over F. The isometry torsor from the quasi-split Hermitian space gives a global pure inner twist. At an inert finite place Lw/Fv is unramified quadratic. The determinant of two hyperbolic planes agrees with that of I4 modulo local norms, so the local group is quasi-split and the integral self-dual lattice gives the specified hyperspecial subgroup. At a split place the marked coordinate identifies the group with GL4(Fv) and the conjugate coordinate uses inverse transpose. At infinity the group is compact and the regular packet supplies the representation of weight ν. The simple parameter ΠK[1] has trivial component group after quotient by the fixed dual center {±1}. The local characters on that center obey the pure-inner-twist product formula. The global generic character is trivial. KMSW Chapter 5 explicitly verifies Hypothesis 3.6.3 and the intertwining relation of Theorem 2.6.2 for generic parameters on pure inner twists. Its Theorem 5.0.5 gives multiplicity one for this tuple, with the singleton split factors ΠK,w and the unramified spherical members at inert places. The frame and link are unramified at those inert places, as required for these spherical factors. The even degree of F and splitting of the coefficient and ramified places also match the field and ramification conditions in the Labesse application in the proof of BLGGT II, Theorem 3.1. This packet argument uses Mok's quasi-split formula and parameter-refined stabilization as above. The actual unitary group · Unramified setting · Even tower degree · KMSW, §1.7 and Theorem 5.0.5 · Mok, Theorem 5.1.2

Weak descent in the release. The input ΠK is regular algebraic, conjugate-self-dual and cuspidal, with ramification confined to a finite set of split places containing those over ℓ. In DescWeakOfRunTh the RunTh dichotomy gives either a definite-unitary constituent spherical at every nonsplit finite place and matching ΠK at almost every split place, or regular block data D with the same cofinite split Satake data but excluding ΠK. The adapter descWeak_of_runTh rules out the second branch with PreOf and NoOther. In the patching entry point the release supplies SR.descWeak to HypothesesH′. It uses the proved runTh, whose split tests match by the determinant-twisted constant term and whose nonsplit unit tests match at hyperspecial level. In the second branch partNorm normalizes the blocks to unitary data, and splitRigid forces D to contain ΠK from the cofinite split Satake equalities. The generic packet calculation above gives the stronger local statements for the seed at the specified places. Transfer of the contributing constituents of the fixed ordinary modules uses the same-family base-change argument above. RunTh and weak descent Descent adapter Split rigidity Patching entry point

At split tame places the common kernel on inertia of the four linked residual characters acts trivially on RK. The fixed-type deformation relation makes tr⁡(RK(h)(RK(g)−1))=0 for every h and every g in this kernel. Semisimplicity and the trace pairing on the characteristic-zero image algebra give the asserted triviality. The remaining finite abelian image has order prime to ℓ. Reducing and lifting its characteristic roots with multiplicities identifies its four characters as the Teichmüller lifts of the linked reductions. At dyadic places the unlinked list is (ξψ,ψ,ξ,1) on inertia before the frame permutation and the common λˉK twist. This test uses the full restricted conductor and has zero monodromy. The prescribed integral type τ0 supplies the invariant in its tensor with the split constituent via the source's fixed-type test. At each link's own scalar auxiliary place the residual character indexed by the trivial frame character gives λˉK=1 on inertia, hence λK=1 there by its prime-to-ℓ inertia order. The parity factor is unramified there and the original frame kills this inertia. The auxiliary local factor is therefore spherical and supplies a vector at principal level one. At ℓ=3 the auxiliary value test is pK≡2 mod 3 at the chosen degree-one place. Inertia comparison · Prescribed type · Auxiliary inertia · Scalar auxiliary condition

At a marked w∣ℓ let ηi be the four flag characters of RK in order. For the inverse-Artin coordinate u of g, the smooth characters are ωi(u)=ι(ηi(g)∏ττ(u)hτ,i). The flagged representation becomes crystalline after restriction to a finite extension killing the smooth inertia parts. Indeed, for each successive extension, tensor the preceding crystalline subrepresentation by the inverse of the new quotient. The resulting representation has labeled Hodge–Tate weights at most −2, and its Tate dual has weights at least 1, so both H0 groups vanish. The crystalline Bloch–Kato dimension formula, local duality and the Euler characteristic then give Hf1=H1, so the next extension is crystalline there. Thus monodromy is zero. Temperedness makes the normalized inducing characters ωi∣⋅∣3/2 unitary and their ratios avoid ∣⋅∣±1. The local component is the corresponding irreducible principal series. The prescribed smooth torus eigenspace and the refinement in flag order occur at any admissible Iwahori depth. Evaluation at the identity of the induced model detects the simultaneous eigensystem. The unipotent index qwj(4−j) and the half-modulus qw−j(4−j)/2, together with the common shift in the inducing characters, give raw eigenvalue ∏i<jωi(ϖw)qw−i for step j. Local principal series · Ordered Iwahori vector · Bloch–Kato, §3 · Zelevinsky, Theorem 4.2

Put ai=ηi(gϖ) and cϖ=κℓ(gϖ)=qw−1∏ττ(ϖw) for the lift with inverse-Artin coordinate ϖw. The four ordered geometric Weil–Deligne roots are ι(ai∏ττ(ϖw)hτ,i), indexed by i=0,1,2,3. Dividing the raw step by the lowest coefficient value ι(∏i<j,ττ(ϖw)hτ,i−i) gives the four normalized Hecke eigenvalues in ℓ-adic coefficients, (a0, cϖa0a1, cϖ3a0a1a2, cϖ6a0a1a2a3). Each is a unit and the last is the central step. Their reductions give the ordered profile even if local residual characters coincide at ℓ=3. The ratios of adjacent steps and the finite torus character on units recover each marked character by local reciprocity. Cyclotomic reciprocity · Torus and normalized step comparison · Classical ordinary normalization

Tensor these local vectors and contract with the real coefficient representation on each link. The result has that link's prescribed finite type, auxiliary level and simultaneous ordinary eigenvalues. The four good spherical coefficients from the determinant comparison and the four ordered step reductions place it in the same residual projector. With the fixed-level control and determinant interpolation for this projected module, the resulting Hecke character yK satisfies xK=yK∘(BK→TK) for the parity-adjusted arithmetic point xK. The determinant equality follows from the good split polynomials and Chebotarev, and the ordered character equality follows from the local ratios and torus law. Thus xK kills JK. Twist stability for βKnpar and its inverse identifies the ideal extended through the parity-adjusted comparison with that extended through the zero-exponent map. Applying this on both links makes the framed seed kill the sum of the extended ideals. This gives seed support at its evaluation. The prime-containment step uses the separate fixed-weight cotangent comparison. The following patch-level construction uses the transfer and Galois comparisons for all contributing summands on the selected levels. Placement and comparison · Twist stability · Two extended ideals

Detailed seed descent and source comparisons

Fix the two fitted deep Count II links over L/L+ with the same comparison character and distinct scalar auxiliary places. Let Bi be their base-level determinant rings and Ji the annihilators of their projected ordinary modules, for i=1,2. Write X.B for the complete local coefficient ring generated by the determinant coefficients and ordered characters of the symplectic frame, and X.A for the finite-type algebra with its normalized matrices and flags. The two maps ϕi:Bi→X.B at the target normalization scale roots and ordered characters as above. Set JX=ϕ1(J1)X.B+ϕ2(J2)X.B. All support tests use this sum. Two links · Framed space · Prime containment tests

Ordinary modules at the Taylor–Wiles levels. On either link keep the same complete coefficient DVR O, weight algebra A, torus character Θ, integral type τ0 including the dyadic factors, and principal level one at its scalar place v0. This retains the compact scalar action through τ0 and Θ and the full central step in the ordinary Hecke action. Given a Taylor–Wiles realization at depth N, put qj=∣kwj∣ at its new split places, and aj=vℓ(qj−1)≥N. At wj use the hyperspecial Kj=GL4(Owj), the Iwahori Ij, and its normal subgroup Ij,1 cut out by requiring each diagonal residue to have prime-to-ℓ order. Write K=∏jKj, I=∏jIj and I1=∏jIj,1. The quotient GN=I/I1 is ∏j(Z/ℓajZ)4, with all four diagonal coordinates. The fixed level outside the added places and the residual profile on the remaining spherical and ordinary operators are inherited from the base. At K the projected module is the same base module M0 with annihilator Ji. The level-change argument uses equality of the hyperspecial spaces and density of the good Frobenius lifts still indexed after removing the added places to identify their projected parts. The map from the enlarged-level universal determinant ring to Bi pulls back the base determinant along the quotient of Galois groups and preserves the four ordered local characters. The determinant comparison at every level uses the preceding spectral transfer and matched Galois rows on separating classical specializations of these projected spaces. In particular it uses split local–global compatibility at the new places. Selected level and base action · Hyperspecial residual projector · Iwahori subgroups

Let M be the ordinary module at I1 before residual projection. It is already finite free over A by ordinary level control and the trivial-stabilizer argument above, which applies with I and I1 and the fixed principal level at v0. Let e be the A-linear ordinary projector on the space of forms, obtained by taking factorial powers of the contracting ordinary operator in the limit. The level-I1 double cosets with the coefficient-place factor at Iwahori depth zero have free GN-orbits. Choose one in each orbit and let χ be their characteristic function, pulled back to the adelic arguments of the forms. Multiplication by χ preserves the prescribed coefficient equivariance, including the torus character. On the space of forms, the sum of the translates of χ under GN is 1. On M set b(m)=e(χm) and write gm for diamond translation. The factorial powers commute with the diamond action. For a given m and coefficient precision, take one factorial power large enough for m and for all χg−1m. The sum of the translates then gives ∑ggb(g−1m)=m modulo that precision. Separatedness gives the identity over A. Trivial stabilizers · Ordinary limit · Orbit representatives and trace · Trace and ordinary projection

Give A[GN]⊗AM the action on the first factor. The map g⊗m↦gm onto M has the A[GN]-linear section m⟼∑g∈GNg⊗b(g−1m). Indeed, translation by h reindexes the sum by g=hk, and the composite is the trace identity. The source of the split surjection is finite free over A[GN] by the underlying A-freeness of M. The residual Hecke idempotent cuts out the simultaneous topologically nilpotent part of the commuting operators shifted by their specified residual values. It commutes with the diamond action. On that summand the commuting auxiliary operators Uji give another such projector, onto the chosen simultaneous residual Frobenius roots. Its idempotent also commutes with the diamond action, since the 1+Yji act as the diamond generators in the same commutative operator algebra. The resulting MN is thus finite projective over A[GN]. This finite algebra over the local ring A is local, since its reduction over the residue field is the group algebra of a finite abelian ℓ-group in characteristic ℓ, with nilpotent augmentation ideal. Hence MN is finite free over A[GN]≃A[[Yji]]/((1+Yji)ℓaj−1)j,i. This proves freeness even when some labeled Frobenius root residues coincide. Trace criterion · Residual Hecke idempotent · Projection at the auxiliary eigenvalues · Local group ring

Augmentation at finite depth. The diamond norm ∑g∈GNg identifies coinvariants with I-invariants in the projected ordinary representation. For a free group-ring module this is the map taking each basis vector to its group sum, and the kernel is generated by the differences Yjim. Taking a group-ring direct summand preserves the calculation, and the norm intertwines the auxiliary Uji. Thus MN/(Y)MN identifies with the selected residual Frobenius summand of the ordinary module at I. Write Pj for the base determinant polynomial of geometric Frobenius at wj. The elementary symmetric polynomials in Uji act on the hyperspecial vectors by the coefficients of Pj with the usual alternating signs. Sending 1⊗m to the inclusion of m∈M0 and letting the formal roots act by the Uji therefore gives a map from the tensor product of the splitting algebra of the Pj with M0 to the Iwahori module. The map back is obtained by applying the local spherical Hecke sums between the two families of polynomials dual for the divided-difference pairing, and summing with coefficients in the splitting algebra. For each j the Hecke sandwich identity makes the composite on the Iwahori module multiplication by ∏i<i′(Uji′−qjUji). On the splitting extension the corresponding identity is the expansion by the same dual polynomials of the divided difference of that product times an input polynomial, evaluated on the hyperspecial vectors via the symmetric functions. Taking the product over j gives both composites. Multiplying the return map by the complementary polynomial factor then gives the exact factor used by the release. Restricting the maps to the chosen residual summands completes the base splitting algebra at the labeled root residues. Diamond norm · Polynomial pairing · Spherical comparison · Sandwich identities · Composite on the splitting algebra · Multiplying the return map · Passing to completed summands

In the resulting maps SN⊗BiM0 ⇄ MN/(Y)MN, SN is the completed splitting algebra just described. The maps are linear relative to the ring map that sends Yji to zero, Uji to the corresponding root and the enlarged-level determinant and ordered characters to their base pullbacks. Their composites are multiplication by FN on the right and its image on the left, where the product is over ordered pairs of distinct indices, FN=∏j∏i=03∏i′≠i(Uji−Uji′)(Uji−qjUji′). Inverting FN identifies the augmentation with the base extension. The separate local determinant relations on MN have the form (FNtr⁡(gQ))24MN=0 for each group element g. Here Q is ∏i(Frj−Uji) for geometric Frobenius, or (∏k≠i(Frj−Ujk))(τj−1−Yji) for the chosen tame inertia generator τj. At classical evaluations the factor implements the distinct-root and norm-ratio tests in the local relations. The geometric lemma and Bernstein–Zelevinsky finiteness bound the length of the torus filtration by 4!, so vanishing on its characters gives the stated power on generalized eigenspaces. Separation of the classical evaluations then gives the integral identities. Together with base weight freeness these constructions give HasAut(Bi,Ji) for every datum and realization on each of the two selected links. Exact factor · Augmentation and local relations · Local test · Generalized eigenspaces · Assembly for every datum · Two linked calls

Patch-level ordinary augmentation

The patched test on one link. Use the base setting S′=S, an odd ℓ, local class field data with ranks summing to [L+:Q], and a nontrivial residual cyclotomic value on the linked group. Use the HasAut(Bi,Ji) system constructed above and the paired cohomology input below. Let fi:Bi→k[[u]] be a curve satisfying the source's NiceI conditions. They require the ordered ordinary characters to be distinct with ψw,j≠ψw,kκℓ for j<k, absolute irreducibility on the full cyclotomic kernel, dense separable characteristic polynomials there, and vanishing of the cyclotomic intertwiner at the scalar auxiliary decomposition group. Suppose Ji⊆ker⁡fi. We show Ji⊆pi for every prime pi⊆ker⁡fi. Curve conditions · Patching hypotheses · Single-curve statement

Choose the Taylor–Wiles detection elements with distinct roots on the generic curve and use the corresponding prime families at arbitrarily large depth. Normalize the curve after finite coefficient enlargement. The resulting weight ring is Λ′=O′[[s0,X1,…,XD]], with s0 specializing to the curve uniformizer and D=4[L+:Q]. For qTW auxiliary places the tuple ring R∞ has rTW=4qTW patch variables Y. Write I∞ for the determinant, polarization, ordinary, killed-inertia and auxiliary Frobenius and inertia relations, and set Δ=FN∣qj=1 with the limiting roots in R∞. At each fixed precision the congruences qj≡1 mod ℓN give this limiting factor in the augmentation comparisons. It is ∏j∏i<i′(Uji−Uji′)4, which is nonzero at the generic curve by the distinct-root choice there. The relations (1+Yji)ℓaj−1 vanish at any fixed weight and patch precision for all sufficiently large N, so the finite-level freeness gives freeness over the corresponding truncated power series ring. At Y=0 the finite-precision maps to and from the base augmentation module X0 have composites Δ. The compatible limits after inverting Δ give the finite patched module with quasi-regular Y and augmentation X0, after the coefficient and splitting extensions. The auxiliary relations put I∞ in the radical of its annihilator. Augmentation maps · Patched module · Auxiliary relations · Limiting discriminant factor

Complete at the lifted curve prime q of R∞, which contains (Y)+I∞ and avoids Δ. Write H=(R∞)q^/I∞(R∞)q^ and compute the module before quotienting by I∞. Its localization and completion are nonzero by the hypothesis on Ji and the augmentation fiber test. The completed weight localization is regular of dimension D+1. Indeed Λ′→k′[[u]] is onto by the coefficient residue map and s0↦u, and the kernel has height D+1. The augmentation is a local direct summand of the base-changed finite free augmentation, so it is free over this complete local weight ring. The patch variables followed by a regular system of weight parameters give a module regular sequence of length D+1+rTW. Every minimal support prime is associated, so its quotient dimension is at least this depth. Cutting by the rTW patch parameters leaves the finite augmentation of dimension at most D+1, giving the reverse bound. The same support dimensions hold in H. Weight localization · Dimension of patched support · Augmentation support

The paired cohomology estimates give the upper bound. At precision um the bound B1 for paired first cocycles modulo coboundaries has exponent m(D+rTW)+O(1) in the size of the curve residue field. The B2 estimate gives the even lifting relation with bounded u-power loss among g+1 arrays of g detection classes. Together with the auxiliary determinant relation giving the odd cocycle spanning test, these give the three limit conditions K1–K3. At a Type I point there is one full matrix block, so the off-diagonal term of the relative cotangent bound is zero. Over the complete coefficient DVR at the curve, with residue field k′((u)), the ring H is therefore a quotient of a power series ring in D+rTW variables. A minimal support prime of H has quotient dimension D+1+rTW. Its inverse image in this regular local domain must be zero. Hence the presentation is an isomorphism and the entire localized spectrum is supported. The discriminant test uses the nondegenerate trace pairing on the full matrix algebra, tested on two spanning lists of sixteen group matrices. Its determinant is nonzero generically and stays nonzero at the generizations, also at ℓ=3. Paired cohomology bounds · Cotangent estimate · Dimension comparison · Discriminant test

Lift any pi⊆ker⁡fi below the chosen curve prime through the coefficient and splitting extensions, then through augmentation and the faithfully flat completion. The lifted prime contains Y and lies in the patched support. At such a prime the patched and augmentation residue fibers agree. Nakayama and contraction therefore give Ji⊆pi. This proves the single-curve case of the patched support implication. Support descent · Single-curve application

Application to both links. Consider a curve y:X.A→k′[[u]] specializing to the closed coefficient point, killing JX, with absolutely irreducible generic representation and distinct ordered characters on the designated U1(v) directions. The fitted link converts this into the tests above for each fi=y∣X.B∘ϕi. On U1(v) the cyclotomic character and residual representation are trivial, so distinctness gives the ratio exclusion. The fitted irreducibility and separability argument also uses the absence of the specified residual cyclotomic quadratic self-twist and a residual ratio of order greater than four. At the scalar auxiliary place the reduced matrix at a witness is χ(g)I+N, with N4=0 and κˉℓ(g)≠1. The reduced cyclotomic intertwiner operator has nonzero scalar term (κˉℓ(g)−1)χ(g) plus a nilpotent sum of commuting left and right multiplications by N. The intertwiner operator remains invertible over the curve ring by the unit determinant criterion. Hence there is no nonzero intertwiner over its fraction field. Each prime p⊆ker⁡(y∣X.B) pulls back below both curve kernels. Applying the patched test to the two pullbacks and extending the ideals gives JX⊆p. Rank-four curve conversion · Cyclotomic kernel comparisons · Scalar auxiliary test · Two-link inclusion

From the marked seed to the target. The preceding descent places the linked characteristic-zero seed in fixed-level support. Its fixed-weight tangent injects into the prescribed polarized adjoint Selmer group. With the local tangent-to-finite comparison and adjoint vanishing, the full localized determinant ring has tangent dimension at most the weight dimension, and the supported finite torsion-free quotient gives the matching lower bound. The full ring is regular of this dimension, so the annihilator vanishes there. Apply the fixed-weight test on each link through its parity-adjusted map. The equality of extended ideals established above carries the two conclusions to the common framed ring. The resulting prime-containment statement at the seed gives at least one supported minimal component at its residual endpoint. Mere vanishing on the seed evaluation would only give containment at its evaluation kernel. Seed support · Fixed-weight tangent · Residual endpoint Parity invariance of the extended ideal

At a good residual endpoint, write dF=[F:Q] for the totally real tower degree, sℓ for the number of places over ℓ, and dv=dF/sℓ. The flat complete framed chart has a presentation over O with a variables and b′ equations satisfying a−b′≥2dF+2−2sℓ. Put N∗=2dF+1−2sℓ. Connectedness gives a path of minimal components with adjacent intersection dimensions at least N∗+1. The nine block-support loci have dimensions at most dF+20C+1<N∗ under the tower bound dF≥30(C+1)2, with sℓ≤C and the chart bounds on strict and even cocycles and on the odd global and local terms. On the supported characteristic-ℓ coefficient locus, a collision on U1(v) has dimension at most 2dF+1−dv. This follows by pinning one symplectic pair at every place and the other away from v, then using one comparison at v and weight finiteness up to radical. On the irreducible part, trace products with the fixed diagonal element's projectors recover matrix entries up to two symplectic basis scalings, and each local lower symplectic flag contributes at most four coordinates. The resulting bound 2dF+1−dv+2+4sℓ is below N∗ since dv≥30(C+1)>2+6sℓ. Chart comparison · Reducible dimensions · Collision dimensions · Framed fiber

For an edge P,Q with JX already contained in P, intersect the edge with the special fiber. There is a prime of quotient dimension at least N∗ over P+Q+(ϖ), where ϖ is a coefficient uniformizer. It avoids the finite list of reducibility and irreducible collision loci. To obtain a one-dimensional quotient still avoiding the list, choose a nonzero product in the prime quotient from the defining ideals of those loci and from the maximal ideal, then take a maximal prime after inverting that product. Its contraction has quotient dimension one by the principal ideal theorem and prime avoidance. Normalizing gives a curve over a finite residue extension with the required irreducibility and distinct characters. It kills JX because it lies on P. The two-link curve test applied to the contraction of Q puts Q on support. Induction on the connected graph, followed by flat contraction, shows that one supported component through a good residual point forces all minimal components there to be supported. Edge propagation · Prime avoidance · Curve normalization

The marked residual chain connects the target to either seed pattern, using the additional residual line for the pattern with the two paired entries. Each residual line has a prime kernel below both endpoint kernels, hence a shared minimal component. Starting at the seed, the chart implication gives support on all components through the first endpoint, and the shared component starts the next application. At the target, a minimal prime below the integral kernel also lies below its residual kernel. The all-components result puts JX in that minimal prime, so the integral target kills both extended ideals. To pass from this to automorphy, apply the arithmetic point comparison on each link, use the common twist and distinct auxiliary places to glue the local conclusions, then apply the prescribed descent to the counting fields. At the target gaps (1,j−1,1) the split local test for potential cyclotomic linkage remains necessary. Residual chain · Target support · Gluing and descent · Local level witness

Inputs to the conditional argument. The patched test uses CohP for local and global finiteness, Tate duality, Euler characteristics, Greenberg–Wiles, normalized Poitou–Tate and the tame local descriptions, together with the local ranks, cocycle detection and auxiliary prime realization in the link. The two HasAut systems follow from the level construction above with the same base annihilators. The remaining automorphic premises are the projected spectral transfer at base and auxiliary levels, the matching Galois determinants and ordered characters on the separating arithmetic evaluations, and the local compatibility at the prescribed Taylor–Wiles places. The release packages these through its literature hypotheses and weak definite-unitary descent. Starting the propagation also uses the attached inducing cusp, its fixed-level realization, and the separate fixed-weight adjoint rigidity comparison on each link. The chart step uses the deep tower cohomology bounds and weight finiteness. Finishing uses arithmetic classicality and the stated local and global descent conditions. Thorne's residually Schur module and patching results give a precedent with their own generator count and local hypotheses. Allen–Newton–Thorne's lifting theorem assumes ℓ>3 and individually polarized residual constituents, so the paired calculation at three uses the stated link, spectral and cohomological hypotheses. Cohomology package · Ordinary patching input · First-order count · Obstruction bound

Detailed patching and propagation argument · Thorne, §§4 and 6

4. From one host to the points at a cutoff

Fix ε,J and let S consist of the real cutoffs admitting WinDatum. The scale s records s.x=x∈S, s.J=J, s.ε=ε, s.L=xε and s.b=exp⁡(xε). Release input (admissible ranks and host). Given a tuple (Y,q,d,e) at x, the rank theorem extends its frame to R with RankVal(s,R) and the host admissibility condition. In particular R.t≤⌊x400⌋ eventually. The host theorem uses these ranks, the conductor bounds and the splitting conditions of §2. Its output is H with its arithmetic field, two auxiliary primes g0,g1, two deformation primes, an initial level prime and sets TT,T2 of frame elements. It satisfies the size, local prime class and conductor requirements HostGood. There is a separate simpler conductor-one host theorem with J=1 and a trivial extra requirement. The construction consuming the cohomological windows uses the supplier with the simultaneous conditions below. Rank supply Host data Choice of tuple and common scale Basic host theorem

The extra host requirement Gψ has two parts. First, EAtH(Level.cyc(ψ),H,TT,x,ε,4) asserts the kept and exceptional character conditions, accuracy x−1+ε∣TT∣ in every coset count, and centered-ramp and log-derivative bounds for the bounded characters. Its ideal shares are 4/(3o+1) on the identity coset and 3/(3o+1) on the others, with o=∣⟨ψ⟩∣. For 1 the single share equals one exactly. The nonempty quadratic twists still need the pair estimates. The release invokes its bounded-twist lemma on the same q,d used by the host. Second, Gψ asserts VVle(H.E,R.t,s.b). Proved here. The host field satisfies IsChiField for the chosen e,q,d, so this part follows by restricting the pair's budget to j<R.t. A point with the same s,R,H.E,TT and arithmetic host carries both assertions. Host analytic interface Proof of both host assertions Exact transport into point data

The primes and the level ideal on the selected field

Keep the field and the rank data at this cutoff. Write N=QS qd, K=Q(−d), b=exp⁡(xε). The field construction gives d∣N, d≡3 mod 4, and log⁡N≪xε/5(log⁡x)2. It takes the level ideal in K to be the unit ideal, whose norm is one. Paper II constructs the sets of thick places and tops on this frame. For each shift the thick places keep the same digits, and the weighted exceptional-orbit estimate uses the same tops as the character average. Level ideal and prime set · Thick digits and their translates · Weighted choice of the tops

Primes represented by a norm in a short interval. Suppose d≥100 is prime, d≡3 mod 4, M≥3, and put B=4dM2. There are absolute c,c′>0 such that, whenever 0<h≤1, log⁡B≤clog⁡X, and h≥e−clog⁡X, at least c′hXB3log⁡X rational primes p∈(X,(1+h)X] have a generator of norm p congruent to 1 mod MOK. Writing that generator as (a+b0−d)/2 gives a2+db02=4p, M∣b0, and 2M∣a−b0−2. Indeed 1,(1+−d)/2 is an integral basis. These are the congruences required of a norm prime by the release. Norm and congruence conditions · Short-interval input

To prove the estimate, take the ray class field of K of modulus MOK, and let hM be its degree over K. The ray exact sequence gives hM≤hKM2. The imaginary quadratic Minkowski bound gives an integral ideal of norm O(d) in every class. At norm j there are at most the divisor function τ(j) such ideals. Thus hK≪dlog⁡(2d) and hM≪B. Every Hecke character in this extension has conductor dividing MOK, so its analytic conductor parameter in Thorner–Zaman's Theorem 1.4 is bounded by 4dM2. The base degree is two. For X≤u≤2X, that theorem counts the prime ideals with trivial Artin symbol by F(u)hM(1+O(e−c1log⁡X)). Here F(u)=Li⁡(u) in the absence of an exceptional zero and F(u)=Li⁡(u)−Li⁡(uβ) otherwise. The displayed error follows from the two terms in the theorem, since log⁡X/log⁡B≥log⁡X/c. One can take c1>0 fixed for c≤1, and then decrease c to meet the theorem's size threshold. Thorner–Zaman, Theorems 1.4 and 3.3

In the exceptional case Stark's bound in their Theorem 3.3 gives (1−β)log⁡B≫B−2. Consequently 1−uβ−1≫B−2 on this interval. This follows from 1−e−s≫min⁡(1,s) for s≥0 and log⁡X≥log⁡B. Integration of F′(u)=(1−uβ−1)/log⁡u gives F((1+h)X)−F(X)≫hXB−2/log⁡X. The same bound holds for F=Li⁡. Each endpoint error before division by hM is O(Xe−c1log⁡X/log⁡X), since F(u)≪X/log⁡X. Choose 3c<c1. Then hB−2≥e−3clog⁡X absorbs both errors, uniformly once c is small enough. This gives the lower bound with denominator B3 after division by hM. Prime ideals of residue degree two account for O(X) terms, and primes dividing Md for O(log⁡B). Their total is absorbed because hB−3≥e−4clog⁡X. Shrinking c forces the requisite lower bound on X for all permitted B. Each remaining rational prime contributes at most two ideals. Trivial ray class means the ideal is generated by an element congruent to one at the modulus. Since the ideal is integral, its generator is integral at every finite place. This proves the estimate. Ray class version in the release · Thorner–Zaman

The two primes with prescribed types. Put Y0=(kb)8. For M=N the preceding bound with h=1 gives at least Y09/10 norm primes in (Y0,2Y0] eventually. Indeed log⁡(4dN2)=o(log⁡Y0) and 3log⁡(4dN2)+log⁡log⁡Y0=o(log⁡Y0). For a candidate g, its excluded coefficient primes form the set Br(g)={g} ∪ {ℓ prime: ℓ4vℓ(ga−1)>ga for some 1≤a≤r}. For fixed a at most three primes qualify, since the product of their exact prime powers divides ga−1. Thus ∣Br(g)∣≤3r+1. Fix g0 and ℓ∈Br(g0). For a given a, let w be the least positive integer with ℓw>Y0a/4. Every g for which a witnesses exclusion satisfies ga≡1 mod ℓw. There are at most 2a residue classes, by the cyclic unit group at odd primes and the decomposition of dyadic units into a sign and a cyclic group (the moduli 2 and 4 satisfy the same bound). Each class occurs at most 1+Y03/4 times in this interval. Including g=ℓ, the number of excluded partners for g0 is at most (3r+1)(1+2r2(1+Y03/4))<Y09/10 eventually, since r≤x5 and Y0≥e8xε. A second norm prime g1 therefore has disjoint excluded set. The norm congruence gives gi≡1 mod N. For the later level norm prime at size Ts, use M=N(g0g1)2, h=1/V. Here 1≤V≤x12, log⁡M=O(xε), and log⁡Ts≥csmxε with fixed cs>0 and m≥x1/3. Both short-interval hypotheses hold eventually, and the positive lower bound supplies a prime in the required interval. Excluded prime definition and count · Fixed-coefficient count · Two primes and the later level norm prime Level interval

The deformation primes and the canonical extra prime. Linnik's theorem gives an absolute A≥1 such that a prime congruent to 1 mod D occurs below or at DA for every D≥2, absorbing its fixed multiplicative constant into A. Apply it successively to 15N, 15Npde, and Npdepde′. This gives distinct deformation primes pde,pde′≡1 mod 15N and a provisional level prime in 1 mod N, all bounded by NC8 for an absolute C8. They lie below g0,g1 eventually. For the final set let r0 be the least prime avoiding D=N(pde−1)(pde′−1). A prime divisor of D+1 shows r0≤2D. Since 2≤N and the level ideal has norm one, r0N≤N3+2C8,q0=min⁡{p prime:p≡1 mod r0N}≤N(3+2C8)A. In particular q0 differs from both deformation primes, as r0 divides neither pde−1 nor pde′−1. It lies below the type primes for large x, because log⁡N=o(xε). The five primes split in the abelian host field whose conductor divides N, and the resulting level set is {g0,g1,pde,pde′,q0}. This gives the enlarged host constant at every sufficiently large supported cutoff. Least-prime theorem, Xylouris · Host level selection · Canonical prime bound and releveling

Common weights on the chosen host

Conditional sampling theorem. Fix the host at a supported cutoff, including TT,T2, and fix the constants in the order required by the uniform base-level projection estimate. At all sufficiently late cutoffs, that estimate implies the existence of a single pair a,b supported on TT satisfying both weight windows and WOK. The cutoff threshold is uniform over eligible hosts and indices. The arithmetic proof uses only the local mean loss at large primes. The proof below gives the integral transfers at the serving depth from the release's segment estimates and cleaned paired-column certificate. Local projection and all size and depth checks · Base-level projection statement

Write a∗=xexp⁡(xε), X′=exp⁡(xε/3), b∗=a∗/X′, δ=x−ε/8, and Y0=x13 in this sampling argument. Let N1=Oε(log⁡x) be the bound in Paper II §4 on the number of tops in an orbit window. With the fixed width constant K.La, put L=K.La N12log⁡a∗ and Wb=2nδxεLb∗/k. The two intervals are Ia=[a∗,a∗+a∗/(kY0)]∩Z,Ib=[b∗,b∗+Wb]∩Z. Both widths exceed one eventually. In the first, k≤x6. In the second, 2nδ≥1, L≥1 and b∗/k grows faster than any power of x. Sample uniformly from the finite nonempty product IaTT×IbTT and extend by zero. These intervals impose WinOK. The constants in the projection estimate are chosen before using the boxes, so the sampling distribution is fixed for all primes on this host. Box and width definitions · Weight windows

At a candidate prime p<X′, the cyclotomic coordinate of Frobenius partitions the frame into orbits. The candidates are in the prescribed inert class with orbit order at least xε/4. In the range p≥k2Y02 the candidate condition excludes a power of p in a short relative band about X′. The signal Sp is m times the sum of the rank losses over the good orbits of Paper II §4. For an orbit with NT tops, each assignment of the local weights has exactly (∣Ia∣∣Ib∣)∣TT∣−NT extensions. Its marginal is therefore uniform on the prescribed local boxes. The tops in the good orbits are disjoint, with sum of their cardinalities at most ∣TT∣. Good orbits · Candidate exclusions and marginals

For p<k2Y02 the defined loss is bounded by CNTkx−26≤NTx−12 eventually, with C fixed once the orbit constants are fixed. At large primes the base-level projection estimate gives mean loss at most CprNTN1/Y0. Its certificate is uniform in the bounded interior weight coordinates, the local coefficient ring and the target lattice. It holds at the required depth and bounds the least admissible rank loss from above. In transferring the estimate, the orbit test gives NT≤40n′δ for orbit order n′, and the band exclusion and height bounds apply to the prescribed boxes. The supporting proof checks the signed odd-top term in the serving depth and the dimension-weighted change of boxes in the local estimate. Since CprN1≤x eventually, the large-prime loss has mean at most NTx−12 as well. Local means and quantitative transfer · Serving-depth and numerical bounds

It follows on the single global product that ESp≤m∣TT∣x−12 for each candidate and Pr⁡(Sp>x−ε)≤xεESp. Write u=log⁡X′=xε/3. Summing over all integers up to X′ gives ∑p(log⁡p)/p≤u(2+u) for the candidate sum. Thus E∑p candidatelog⁡pp(21{Sp>x−ε}+4Sp)≤4(xε+1)m∣TT∣x−12 u(2+u)≤48x−19/2+(9+2/3+1/50)ε. For the last inequality use m≤x1/2+ε/100, ∣TT∣≤2x2+8ε+ε/100 and u(2+u)≤3u2. The excess saving γ=1/2−(9+2/3+1/50)ε is positive for ε<1/52. Taking xγ≥48 makes the bound at most x−9. A point in the finite product attains at most the mean and supplies WOK. Linearity of expectation allows the restrictions for different primes to overlap arbitrarily. Independence is used only for the coordinates of the weight box. Finite averaging and uniformity

A determinant estimate on the original interval of weights

The large-prime argument varies one terminal weight while keeping the other weights of its segment fixed. Its Gram matrix M(z) has size s, with entries of degree at most J in the integral binomial basis over Zp. Take W consecutive integers with 1≤W≤ph and J<ph. Assume det⁡M is a nonzero polynomial with leading coefficient l and degree d, and vp(l)+d/(p−1)≤sL. For a positive integer E, let β(z) count the Smith factors divisible by pE, including zero factors. Then 1W∑zβ(z)≤s(JW+L+Jh/WE). This inequality also counts a weight at which the specialized Gram matrix is singular. Terminal Gram window and application

For the proof, expand M(z+phU) in U. Its coefficients are integral by the binomial addition formula and vp(i)<h for 1≤i≤J. In Smith coordinates at z, a determinant coefficient of degree i uses a nonconstant term from at most i rows. Its valuation is at least Emax⁡(0,β(z)−i). Factor det⁡M(Z)=l∏ζ(Z−ζ) over a finite splitting field, counting roots with multiplicity. Let dz be the number of roots in vp(z−ζ)≥h. Divide each factor of det⁡M(z+phU) by a coefficient of least valuation. After reduction the product has degree exactly dz, since each close root contributes degree one and each distant root degree zero. The coefficient of Udz therefore has valuation vp(l)+∑ζmin⁡(h,vp(z−ζ)). Each root belongs to at most one such close disc. For 0<u≤h the number of weights in vp(z−ζ)≥u is at most W/p⌈u⌉+1. Integrating in u, summing over roots, and using d≤sJ proves the bound. Integral change of variable and Smith estimate

Counting the initial segment and pairing its columns

Fix one eligible Frobenius orbit of length n′ with N tops, let q=pn′, and write h=k−1 in this paragraph. A top u contributes symmetric degrees Au,Bu. Put Sk(d)=(d+k−1k−1) for d≥0, zero for d<0, and dbi=∏uSk(Au)Sk(Bu). Cut the cyclic orbit at its even tops. A segment of length Gj has index Qj=pGj, so ∏jQj=q. On an internal letter of degree d and digit modulus P, divide each exponent of a monomial by P. Its remainder vector b∈{0,…,P−1}k satisfies ∣b∣≤d and ∣b∣≡d mod P. Conversely a composition of (d−∣b∣)/P and the vector b recover the monomial. Thus the sum of the reduced symmetric dimensions over all admissible remainders is exactly Sk(d). At a terminal degree B, keeping ties 0≤l≤L=⌊LB/k⌋ gives ∑lSk−1(B−l)=Sk(B)−Sk(B−L−1). Both endpoints enter this count.

For a fixed digit class, interlacing in successive symmetric and dual Pieri products counts the intermediate paths with multiplicity equal to the dimension of the terminal highest weight. Restricting the initial degree from H0 to H gives the exact factor Sk(H)/Sk(H0). The path exclusions for excessive loss in an internal strip have total dimension at most e−L/8 times the unrestricted dimension at H. These three factors give a lower bound for each class. Summing over digits and multiplying over segments gives dbiq−h(1−κmargin) as a lower bound for the model rank. Here κmargin bounds the total loss from initial, terminal and internal cutoffs. The exponent h counts the free coordinates of a composition with its total prescribed. The edge normalization counts the hyperplanes in the residue space and has denominator 1+q+⋯+qh. Fixed-digit lower bound · Segment product · Digit multiplicities, endpoints and mean

Let κj denote the fraction of model directions in segment j whose integral Gram factor has valuation exceeding its allocated budget Dj, counting zero factors among the discarded directions. Products of good ranks give at least dbiq−h[1−κmargin−∑jκj]+ directions. In passing from the enlarged to the original integer boxes, hold all lower weights fixed and restrict the upper weights first. With N1 the orbit-window bound, sj the number of internal tops and σj the number preceding the excluded prime-power band, the mean of the total loss is at most e0.015N1Y(CsN+CA∑j(sj−σj+1)). This uses the internal Gram estimates, the terminal Gram coefficient bounds and the integral comparison from level lattices to blocks. At a fixed lower weight the enlarged upper box has volume ratio at most (1+Δ/∣Ia∣)N≤e0.015. All losses are nonnegative, so restriction costs at most this ratio. Averaging over the original lower box gives the displayed estimate. Original-box marginals

To identify the pairing on an actual edge e=(S,L), choose a basis b with S=bWk, L=bDp−1Wk, where W is the unramified valuation ring and Dp=diag⁡(p,1,…,1). Put w2=p and Dw=diag⁡(w−1,1,…,1). The reversed dual edge has basis b′=b−TDp. At every Frobenius embedding ιs the adapted matrices satisfy ιs(b′)Dw=(ιs(b)Dw)−T. In the polynomial model the Fischer form pairs equal monomials of multi-index α with value ∏αi!. It is invariant under inverse-transpose substitutions on the two arguments. If dℓ are the individual letter degrees, multiply this form by Z=pk∑ℓ⌊log⁡pdℓ⌋/∏ℓdℓ!, assigning a zero logarithm to degree zero. The resulting form is integral on the coefficient lattices since the multinomial valuation for a degree dℓ is at most k⌊log⁡pdℓ⌋. It maps the cleaned lattice on the reversed edge into the dual test lattice on e. Indeed, duality takes every competing incoming edge to a competing outgoing edge, and multiplication by wc commutes with the map. Lattice and edge definitions · Two-sided lift

In identity coordinates the polynomial lift of each retained block symbol has the prescribed symbol modulo wc on the selected line and vanishes to precision c on every other line. At precision n subtract lifts of each error term wiτi computed at precision n−i. Multiplication by wi restores its denominator allowance. The localizing polynomials preserve the off-line vanishing throughout this induction. Apply these lifts to the source in basis b and to the test in basis b′. Their vertex denominators are at most the stated serving depth ⌊t0⌋+. The two powers of w in the rescaled pairing contribute exactly the required power of p. The comparison on e is therefore Z times the Fischer pairing of the two identity symbols modulo wc. Distinct labels have disjoint monomial supports. For a common label β, the block factorial comparison gives a scalar κβ times the product of the segment forms, with 0≤vp(κβ)≤8Nklog⁡pa. This scalar can have positive valuation. On the tensor of the good Smith frames the matrix is diagonal modulo wc. Its diagonal entries have valuations at most D0=∑jDj+8Nklog⁡pa, and c>4D0. Dividing the rows by those entries gives a matrix congruent to the identity modulo w, hence invertible over the coefficient ring. This realizes the product rank on the two cleaned lattices. The construction uses the identity-coordinate localizing polynomials and their factorial estimates. The internal Gram and level-lattice bounds above control how many directions reach this step. Identity-edge induction · Normalized block forms · Coordinate maps, scalars and depth

The determinant on a capped prefix

Fix an internal tie, the coefficient prime p, and the degrees of the preceding letters. Write K+1 for the length of the prefix ending at its standard letter, r0 for the number of coordinates, and S for the sum of its unsigned degrees dj. On the monomial lattice A over Zp use the form B(xα,xβ)=1α=βα!/∏jdj!. Impose the caps on all prefixes through this letter and take highest vectors of fixed weight for the subgroup fixing its last coordinate x. For order e in x, let Le be the image obtained by taking the coefficient of order e, with its factor xe, on integral vectors divisible by xe. Write Ue=QpLe, Ae=A∩Ue and he=dim⁡Ue.

Two integral estimates govern the answer for K≥1. Put c1=(r02)⌊log⁡p(S+r0−1)⌋,c0=2(r02)⌊log⁡p(4S+r0−1)⌋,b0=c0+c1,D0=(K+1)c1. Then pb0Ae⊆Le⊆Ae. Indeed division by xe shifts the last degree and its cap down by e. Restriction at x=0 in this reduced degree has an integral lift after scaling by pc0. Its proof uses the auxiliary glr0 action on the sum of the restricted degrees. Its highest vectors have last degree zero and hence lift. Type projection and divided-power generation each cost at most c0/2, and the actual restriction image is stable under those divided powers. Returning the lift to the prescribed highest weight costs c1. Next take the Fischer-orthogonal projection Pe to Ue. The nested cap projectors, followed by the highest-weight projector on the chosen weight, give pD0PeA⊆Ae. The length-zero and single-letter caps each act by zero or the identity on this multidegree. This accounts for the number of denominators in D0. Bounded restriction lift · Projection on a positive-length prefix

For a lattice L write L# for its dual under B in its rational span. In the monomial basis the entries of B are reciprocals of multinomial integers, so A#⊂A. Since Ae is saturated in A, every integral functional on Ae extends to A. Orthogonal projection of its representing vector gives Ae#=PeA#. Thus pD0Ae#⊆Ae. Dualizing the inclusion for Le gives pb0Le#⊆Ae#. A second multiplication by pb0 returns from Ae to Le. Every Smith valuation of B on Le is therefore at most D0+2b0. Equivalently, changing from Ae to Le adds 2 length⁡(Ae/Le) to the Gram determinant valuation. Hence vp(det⁡B∣Le)≤he((K+3)c1+2c0)≤he (K+7)r022log⁡p(4S+r0). For K=0 each nonzero graded space has one monomial with Gram entry 1/(oe) for last degree o. The rank-zero determinant is one. Smith argument and integral partner · Graded lattice, cap action and denominator bounds

Internal and terminal averages on the chosen weights

In a segment with s internal ties, let σ be the first late tie. At a late tie i, fixing the digits and all other coordinates makes the compatible upper weights an arithmetic progression of step Pi. The free dual degree ni=(Ai−Ri)/Pi−ji runs through each integer in one interval. Here Ri is the sum of the fixed remainders and ji is the fixed tie level. The caps on the prefix through its standard letter apply throughout this interval. In a basis adapted to order e, permutation terms mixing different orders have smaller degree in the frozen Gram determinant. Its leading coefficient is the product of the graded determinants just bounded and their nonzero leading scalar factors. The width of the interval and the leading coefficient can thus be used together in the determinant-window estimate. Leading coefficient of the internal determinant

More precisely, let bxint count the Smith factors above the internal threshold at level x, and let Yx(A) be the capped polynomial space there. For a fixed compatible digit class C, write A(i) for A with its i-th coordinate at the last compatible value. The continuation multiplicity of a discarded highest vector is bounded by its multiplicity at this endpoint. The integral extension of the highest-weight partner operators places all the retained directions in a saturated sublattice. The internal determinant estimate, with prefix degree at most 7/Y times the progression length, gives ∑A∈Cbxint(A)≤31Y∑i=σs−1∑A∈Cdim⁡Yx(A(i)). At the terminal tie the space is fixed as the scalar exponent varies. Write dx for the free dual letter's dimension, rx=dxdim⁡Yx(A) and W for the number of terminal exponents. The monomial with exponents ν has terminal pairing ν!/n!, where n=∣ν∣. Multiplying its partner by n!/ν! gives a unit. The terminal polynomial after binomial clearing has this internal tensor Gram as its leading matrix up to a specified scalar. Its coefficient bound and the same window estimate give at most W(dxbxint(A)+5rx/Y) discarded directions, summed over the terminal exponents. This gives the segment constant CM=1 on the compatible progressions. Passing to the paired columns uses the integral comparison from the separately normalized level lattices to the monomial block. The highest-weight partner extension, coefficient integrality and this last comparison are the remaining inputs to this deduction. Highest-weight partner extension · Terminal coefficient estimate · Passage from levels to the block · Internal and terminal counts with their thresholds

Integral harmonic splitting on the flat rows

At a large eligible prime p, work over the admitted coefficient DVR O with w2=p and v(p)=1. For one numbered Frobenius orbit let T be the set of tops and NT=∣T∣. The row rank is k≥3, and write mloc=k−1 for the local weight index. At a top s the two degrees (Ps,Qs) are (As,Bs) for even s and (Bs,As) for odd s. A shift subtracts js from both, with 0≤js≤min⁡(Ps,Qs). The flat row then has weight (Ps−js,0,…,0,−Qs+js) and the off-top factors are trivial. Use the rounded radius a with Ps+Qs+1≤8a. Alternating degrees · Bidegree model

For one top write SP,Q for the bidegree space in the standard and dual variables, h=∑ixiyi, and F=∑i∂xi∂yi. On degree (P,Q), Fh−hF=P+Q+k. Thus for Fv0=0 in degree (a′,b′) one has F(hiv0)=i(a′+b′+k+i−1)hi−1v0. The top harmonic projector is HP,Q=∑ℓ′=0min⁡(P,Q)(−1)ℓ′(P+Q+k−2ℓ′) h(ℓ′)F(ℓ′),h(ℓ′)=hℓ′/ℓ′!,F(ℓ′)=Fℓ′/ℓ′!. Applying F cancels consecutive terms, and the projector is the identity modulo h. Induction on the smaller degree gives the direct sum of the strings hjker⁡F∣SP−j,Q−j. The coefficient map for a′=P−j, b′=Q−j is Dj=Ha′,b′F(j)j!(a′+b′+k+j−1j). The string formula shows that hjDj is the projection onto the j-th summand. To bound its denominators in the coefficient lattice, expand the ℓ′-term on a monomial xμyν. For multi-indices ∣α∣=ℓ′ and ∣β∣=j+ℓ′ the factorial ratio from the divided powers and 1/j! is (j+ℓ′j) β!j!α!=(j+ℓ′j)2(ℓ′α)(j+ℓ′β). The remaining monomial coefficients in the numerator are integral binomials. The denominators are one k-part multinomial and two binomials with upper arguments at most P+Q+k. Legendre's formula bounds their loss by (k+1)⌊log⁡p(P+Q+k)⌋. The same bound works for H, and multiplication by hj is integral. Both operations are equivariant for a change of row frame. Harmonic identities · Coefficient map · Gauss lattice estimate

The identification with the actual flat lattices uses the first-column entries and the signed (k−1)-minors of the first k−1 columns. Send x to the first column of Z+ and y to the signed cofactor vector of the independent matrix Z−. Call the resulting map ιpol. It is injective because independent matrices realize arbitrary first columns and any nonzero cofactor vector. Let β be its specialization to Z+=Z−=Z. The two-letter lattice and the flat lattice are the integral images under ιpol and β, respectively, by the integral flag-minor basis theorem for shapes (P,0,…,0), (Q,…,Q,0) and (a′+b′,b′,…,b′,0). The image of the tied matrix is dense in x⋅y=0: when x,y≠0 and y(x)=0, complete x to a basis of ker⁡y and scale another basis vector to obtain the prescribed cofactors. Thus ker⁡β in each bidegree is the corresponding multiple of h. For a lattice frame g, the transported polynomial for weight λ is det⁡(g)−λkf(g−1Z). The cofactor identity C(g−1Z)=det⁡(g)−1gtC(Z) gives exactly this determinant twist on the dual letter and on the flat row. In particular the integral flat image has a coefficient-integral lift in each frame, with its transported two-letter coordinates. Integral Schur modules · Flag minors · Cofactor transformation · Lattice and incidence comparison

On the actual rows use the projection βDjιpol−1 and inclusion β(f)↦ιpol(hjHa′,b′f). The latter is well defined by the incidence kernel. Tensoring gives rj0,ij0 with ∑jij0rj0=1. Each is integral on its row lattices after multiplication by pu, where u=NT⌈(mloc+2)(log⁡p(8a+mloc)+1)⌉. This holds on the Frobenius-twisted source, intermediate edge and target lattices and on the neighboring edges in the intersections. In adapted frames these three lattices use gs, gsdiag⁡(w−1,1,…,1) and gsdiag⁡(w−2,1,…,1). Tensor comparison · Adapted edge bases

Put B=∑sPs, Aj=∑s(Ps−js) and dj=B−Aj. These are the edge exponents. The two pairing conventions are bbi(x,Φ)=Φ(pBx) and bj(y,ψ)=ψ(pAjy). Set rj=wdjrj0 and ij=w−djij0. On the two sides the maps are x↦purjx and Φ↦p2u+dj(Φ∘ij). For an integral row lattice or its integral dual scaled by wE, the corresponding map lands in the flat lattice or flat dual scaled by wE+dj. On the dual side this uses p2u+djw−dj=p2uwdj and integrality of puij0. The exponent −t for the source and target constraints becomes −t+dj≥−t. The exponents −B,c−B become −Aj,c−Aj in the edge and adjacent-edge constraints. The cleaned inclusions therefore hold at the same depth t and conductor c. The pairing identity is bj ⁣(purjx, p2u+dj(Φ∘ij))=p3ubbi(ijrjx,Φ). Consequently a paired bidegree certificate with scalar z and size r yields integral flat pairing matrices whose sum is p3uzIr, after using the original integral change of columns. For each matrix keep the Smith factors dividing p3uz. After division by p3uz, the omitted parts vanish modulo w, so the sum of the retained ranks is at least r. Integral basis changes and rescaling of the retained factors give flat certificates with scalar p3uz. The loss in its valuation is 3u and the release allocates D1=D0+4u when v(z)≤D0. Cleaned lattice inclusions · Smith extraction · Pairing and budgets

The initial paired-column input is admB2 at depth ⌊t0⌋+, conductor c4=c∗+2NTccl+2n′ccl and budget D0=totalDegreeBudget for orbit length n′. It uses the segment and digit-class estimates on both cleaned bidegree lattices. The split itself preserves depth. For the shifts retained in the mean, js≤J0=Lb′/k, each odd-top minimum in t0 decreases by at most J0. The shifted certificate is served at ⌈t0(A−j,B−j)+NTJ0⌉+. The sum of ranks uses all shifts and the edge rank cap at D1 and c4. The serving inequalities give 4D1<c4, also required in the middle-row comparison. Paired-column input · Shift and edge rank cap

Contraction on the bounded middle rows

Put c=ccl and rmid=k−2. At a top with base weight (as,μs,−bs), take (P,Q,η)=(as−c,bs−c,μs) at an even slot and (bs−c,as−c,−rev⁡μs) at an odd slot. The lower, actual and upper operative weights are λ−=(P,0rmid,−Q), λ=(P+c,η,−Q−c) and λ+=(P+2c,0rmid,−Q−2c). Write θ=λ−λ−=(c,η,−c), θ∨=(c,−rev⁡η,−c) and ζ=λ+−λ−. The upper and lower translations here are made in the original sampling coordinates. Operative weights in the two parities · Contraction statement

The middle pairing. Write DA(ν) for the flag-minor module of shape ν over a coefficient ring A. Over the fraction field it consists of polynomials with right upper triangular character ∏jbjjνj under Z↦Zb. The integral flag-minor basis theorem identifies the module over a DVR with the coefficient-integral part of this field space, since the column multidegree and the polynomial identities for column additions descend coefficientwise. Products add shapes. Put β=c+η, α=Q+β, M=Q+2c, and α′=M−rev⁡α=c−rev⁡η. These are the middle shapes of the auxiliary row and the two contraction inputs, with output middle shape (M,…,M). Set e0=αrmid, γ=α−e01, D=M−e0 and δ(Y)=det⁡Y. Deleting full-height minors identifies DA(α) with DA(γ) by division by δe0. The complementary map τf′(Y)=δ(Y)Df′(Y−tJ) uses the reversing permutation matrix J. Jacobi's minor identity replaces each first-column minor by a signed complementary-row first-column minor divided by δ. There are D=α1′ factors, and complementing their heights gives shape γ. Complementing back, with full-height columns for empty minors, gives surjectivity over the integral ring. The two shapes have equal Weyl dimensions by reversal and a determinant shift, so this is a lattice isomorphism. Integral flag minors · Fulton, §§8–9 · Complementary substitution

On DQ(γ) let ⟨ , ⟩ ⁣F be the Fischer form with monomial norms ∏ijuij! on Yu. Normalize by n=⟨h,h⟩ ⁣F, where h=∏jΔjγj−γj+1 is the leading-minor highest polynomial, γrmid+1=0. The normalized form B0 is nondegenerate by positivity over R and contravariant for transpose by the differential formula for the Fischer pairing. Over Z(p) let D=DZ(p)(γ) and let U⊂D be spanned by coefficients of h(gY) for g a product of elementary unipotents in independent parameters. The highest row-weight line is spanned by h, with h(I)=1. Evaluation at I kills the other row weights, which are Fischer orthogonal to h. Thus B0(F,h(gY))=F(gt). A polynomial integral against U therefore takes integral polynomial values on a generic unit lower triangular matrix, by triangular elimination of its transpose into elementary matrices. Gaussian elimination and the right triangular character put F in the localization of the coefficient ring at the product of the leading minors. Intersecting with Q[Y] gives Z(p)[Y], since reduction modulo p shows that multiplying a primitive polynomial by powers of that product preserves primitivity. The converse follows from F(gt) having integral coefficients for integral F. The same evaluation identity shows U spans over Q, since vanishing on the unit lower triangular matrices implies vanishing on the Gaussian open cell by the right character. Hence U#=D and D#=U⊂D, with # taken for B0. Normalized Fischer form

The highest norm is n=∏i=1rmid(γi+rmid−i)!∏i<j(γi−γj+j−i). Indeed the Fischer norm is the complex Gaussian mean of ∣h(Y)∣2, with independent entries of density π−1e−∣z∣2. The polynomial h is the matrix coefficient on the unit highest vector in the tensor product of the exterior powers specified by the column heights of γ. In the factorization Y=UR by Gram–Schmidt, U is Haar unitary and independent of the upper triangular R by left unitary invariance. The squared diagonal entries of R are independent gamma variables of shapes rmid−i+1, by conditioning on the preceding columns. Their γi-th moments give the factors (γi+rmid−i)!/(rmid−i)!. On the irreducible generated by the highest vector, Schur averaging gives 1/dim⁡Vγ for the mean squared unitary coefficient. The Weyl dimension formula gives the displayed quotient. Fulton and Harris, §24

For a monomial of column degrees γi and a power b=pj with j≥1, the contribution of column i to the valuation of the numerator divided by its monomial factorial is ⌊γi+rmid−ib⌋−∑l⌊ulib⌋=⌊∑l(uli mod b)+rmid−ib⌋≤rmid−1. The last column contributes zero and powers above γ1+rmid−1 contribute zero. The denominator in n can only improve this bound. Thus B=pEB0 is integral for E=(rmid−1)2⌊log⁡p(γ1+rmid−1)⌋ when γ1>0. For the constant line use E=0. Since D#⊂D, the image of B contains pED∗, where D∗ is the ordinary linear dual. Smith form over Z(p) gives diagonal powers pei with 0≤ei≤E, and the bases extend to O. In particular γ1≤2c≤8a gives K0=⌈mloc2(log⁡p(8a+mloc)+1)⌉ ≥ ei. The middle pairing is L(f,f′)=B(f/δe0,τf′). For Agf(Y)=f(gY) the identity AgtτAg=det⁡(g)Dτ and transpose contravariance show that simultaneous left substitution multiplies L by det⁡(g)M. The pairing is integral on the two middle lattices and has dim⁡Vη diagonal Smith pairs, by the two lattice isomorphisms. Twisting its target by the inverse determinant character of degree M gives the invariant functional used in the release's QPh. Invariant pairing and its integral bases · Release Smith bound

Row maps. Put D(Y)=diag⁡(1,Y,1). Apply L coefficientwise in Z to v(ZD(Y))s′(ZD(Y′)), with independent Y,Y′, to define C(v,s′) on the shapes ν(λ),ν(θ∨), where ν(ω)=ω−ωk1. The coefficients in Y,Y′ have shapes α,α′ by the right triangular identities of the inputs and are integral on the integral lattices. The output transforms under right middle triangular substitution by det⁡M. Additions from the first column into the middle columns commute through D(Y) as combinations of such additions, and the last column has degree zero. The first column has degree P+Q+4c. Thus the output belongs to DO(ν(λ+)) on integral inputs, and left substitution in Z commutes with C. For the middle Smith vectors ui,uj′, lift ui/δQ and uj′ by adjoining the first row to every middle minor and inserting enough Z11 factors to give first row degree 2c. The lifts si,sj′ have last row degree zero and specialize at D(Y) to the prescribed polynomials. Their shapes are ν(θ),ν(θ∨). Coefficient construction · Output character

Every lower-flat F satisfies F(ZD(Y))=δ(Y)QF(Z) by its height-one and height-(k−1) minors. Consequently C(Fsi,sj′)=FRij with Rij independent of F. Use F0=Δk−1Q to compute Rij. The result F0Rij has shape (Q+4c,(Q+2c)rmid,0) and first and last row degrees Q+4c and 0. Every contributing minor product must use row one in every factor and omit the last row, so it is a multiple of Z112cΔk−1Q+2c. Evaluate at I to read off the Smith scalar, and cancel F0. This yields C(Fsi,sj′)=δijpeiFH,H=Z112cΔk−12c. Here H is the highest polynomial of weight ζ, of bidegree (2c,2c) in the two letters. In the frame g the determinant twists add for multiplication by si and H, and for contraction because λk+θk∨=(λ+)k. Thus all three maps are integral on their common row lattices. Diagonal contraction · Frame and determinant factors

Cleaned lattices and primitive inclusion. At a chosen edge e=(S,L) use the adapted source frames described above and tensor the three maps, with trivial factors off T. Call the lower, actual and upper tensor weights H−,X,H+, and let d=NTc. For any tensor weight W put AW=∑sWs,1 and write HW(S),HW(L),HW,e for its integral tensor lattices. The normalized edge lattices RW=w−AWHW,e and RW′=w−AWHW,e∨ are perfect for BW(v,G)=G(pAWv). Source cleaning at (t,c4) imposes membership in w−tHW(S), RW(e) and wc4RW(f) on other outgoing edges at S. For tests use w−tHW(L)∨, RW′(e) and wc4RW′(f) on other incoming edges at L. A row of weight ω and first row degree j scales at the intermediate and target lattices by wj+ωk and pj+ωk, with ωk≤j+ωk≤ω1 on the tableau basis. Thus HW(S)⊆RW(e), and HW(L)⊆RW(e) when ∑sWs,k=−AW. Normalized pairing · Cleaned lattices

Let i multiply by the tensor of transported H's, Ui multiply by the tensor of transported sis's, and Lj contract with the tensor of transported sjs′'s. Thus LjUi=δijpeii with ei=∑seis. The first two maps preserve source cleaning at (t,c4) because the multipliers lie in the auxiliary source lattices, hence in every normalized outgoing lattice, and the leading sums add. In the normalized edge coordinates i is multiplication by a tensor of flag polynomials taking value one at the identity. The auxiliary intermediate scale is w2c per top, which cancels against the normalization. Reduction modulo w shows that a primitive multiplier preserves the least coefficient valuation. Thus each row inclusion is saturated in the coefficient lattice, with free quotient over the DVR. Tensoring the resulting integral row retractions and extending scalars gives a retraction π of i carrying RH+(e) into RH−(e). Primitive inclusion · Tensor comparison

The fixed second input of Lj has adapted exponent c per top. It lies in its auxiliary normalized edge lattice and in p−d times its target lattice, hence in p−d times its normalized lattice at every other incoming edge. Here both auxiliary endpoint sums are d,−d. With AH+−AX=d, the test map is L^jG=pdG∘Lj and satisfies BX(v,L^jG)=BH+(Ljv,G). At L the factor pd cancels the p−d in the input under integral contraction, giving the same depth t. On e normalized pairing gives the test inclusion with no loss. On another incoming edge f, the contraction sends RX(f) into p−dRH+(f). The other-edge condition on the new test is therefore wc4−2dRX′(f). The resulting source and dual conductor is c4−2d. Source and dual inclusion estimates · Normalized pullback identity

Cross pairing and rank. Take the two flat certificates on e at common depth t, conductor c4 and valuation budget D1, with ranks n± and pairings z±I. Form the integral cross matrix of the lower sources after i against the upper tests. Keep the u nonzero Smith entries dividing z−z+, extracting z−z+Iu by integral operations. The other s0=n−−u lower sources pair into wz−z+O against all upper tests. Use the lower tests pulled back through π with multiplier pAH−−AH+, changing them contragrediently to the lower source basis. These are in RH+′(e) by perfectness, and pair with the s0 sources as z−Is0. Combining with the n+ upper vectors and tests gives the block matrix (z−ITbadBz+I),Tbad∈wz−z+Ms0,n+(O). Right multiplication by (z+I0−Bz−I) gives z−z+ times a matrix congruent to identity modulo w, so integral column operations give the scalar identity. All the sources are cleaned and the tests are integral at the chosen normalized edge. Transport to the gq=(qk−1)/(q−1) outgoing edges of the same source, with q=pn′. These edges correspond to lines in p−1S/S and its stabilizer acts transitively. The cross-edge entries lie in wc4O by source cleaning alone. As vw(z−z+)≤4D1<c4, the combined matrix has full rank, whence s0+n+≤d+=dim⁡VH+/gq. In particular u≥n−+n+−d+. Cross-pairing argument · One-sided edge rank cap

Apply Ui and L^j to the retained pairs. The result has rank uDmid for Dmid=∏s∈Tdim⁡Vηs. Its block scalars are peiz−z+, which can be made pNTK0z−z+ by integral test rescaling. Thus the valuation budget is 2D1+NTK0. Write d−=dim⁡VH−/gq, dX=dim⁡VX/gq and let κ± be the flat losses at the lower and upper translations. The rank is at least Dmid((1−κ−)d−−κ+d+). After removing the middle factors in the Weyl product for X, the other 2k−3 factors per row lie between the corresponding flat factors. Their numerators grow by at most 2c between the flat weights, except the first-to-last numerator, which grows by 4c. Since P,Q≥b′/3 in the interior, all ratios are bounded by 1+6c/b′. With ϵ0=12NTkc/b′≤1/9 this gives (1−ϵ0)dX≤Dmidd−≤dX and d+≤2d−. The normalized actual loss is thus bounded by κ−+2κ++ϵ0. The boundary uses the trivial loss when the lower translation leaves the original box. Integral rank budget · Weyl comparison

The serving depth is the upper flat depth t=⌈t0(a+c,b+c)+NTJ0⌉+. Its weight-dependent term is the sum over odd tops s of min⁡(bs+c,(au(s)+c)/pd(s)), with u(s) the preceding even top and d(s) its cyclic distance to s. Both terms use the first operative weight at their respective parity. The minimum increases by at most c from the untranslated weights, and monotonicity serves the lower pair at the same depth. After the top transfer the conductor is c4−2NTc. The off-top argument below uses the remaining 2n′ccl of conductor and at most that much extra depth. The serving inequalities give the full certificate at t≥tI, with conductor c∗ and pairing budget Dp. Odd-top depth term · The two translated pairs · Serving inequalities · Final serving depth

Full bounded middle-row proof

Restoring the bounded weights outside the tops

Fix the orbit of length n′ and its subset T of tops. For s∉T write λs for the dominant weight in its operative order. At an odd position this reverses and negates the original row. Put Ao=∑s∉Tλs,1,Bo=∑s∉Tλs,k,J=Ao−Bo≤2(n′−∣T∣)ccl. Let e=(S,L) be a lattice edge with pL⊆S⊆L such that L/S has residue dimension one. On the top factors take source vectors and dual tests at depth t and conductor c≥J with pairing matrix P. Assume PM=zIh for an integral matrix M and a nonzero scalar z. Restoring all the other factors gives the same scalar z, rank h do, depth t+J and conductor c−J, where do is the dimension of their tensor product. In particular the valuation allowance on z and the proportion of the available dimension both persist. Edge and weight conventions · Intermediate and endpoint lattices · Statement used in the projection estimate

Proof. Work over the coefficient DVR with uniformizer w and w2=p. At a lattice vertex write H(S) and H(L) for the integral tensor lattices of the factors outside T, and let He be their lattice at the intermediate edge. The intermediate lattice over this DVR is S+wL=S+w−1pL. In a basis for S adapted to S⊂L, its first basis vector is w−1 times the source vector, and the first target basis vector is p−1 times that vector. This description holds after every Frobenius twist because the twist fixes p. In a factor of weight λs, the standard integral tableau basis uses shape λs−λs,k1 and the determinant twist. In such a basis let as count the occurrences of the first letter. A column is strictly increasing, so 0≤as≤λs,1−λs,k. Under the intermediate change of lattice the inverse coordinate substitution multiplies a basis polynomial by was, and the factor (det⁡g)−λs,k for a basis matrix g contributes wλs,k. Tensoring gives a basis (vi) of H(S) in which He and H(L) have bases (wjivi) and (pjivi) with Bo≤ji≤Ao. The top and complementary factors use disjoint sets of variables. Products of their bases form an integral basis at every vertex and edge. Integral tableau basis · Adapted generators · Tensor basis at the three lattices

The normalized edge lattices are Re=w−AoHe and Re′=w−AoHe∨. Their respective bases are ri=wji−Aovi and ρi=w−ji−Aovi∨. They pair to δij under (r,ρ)↦ρ(pAor) because pAow−2Ao=1. Relative to the vertex basis p−jivi∨ of H(L)∨, the second multiplier is again wji−Ao. Hence H(S)⊆Re⊆w−JH(S),H(L)∨⊆Re′⊆w−JH(L)∨. The vertex lattices are independent of the adapted frame, since a change in GLk(O) and its inverse preserve the integral polynomial lattice. The following comparisons at neighboring edges use their inclusions of these same vertex lattices. For any other edge f with source S, these inclusions give ri∈w−JRf. For any other edge with target L, they give ρi∈w−JRf′. Normalized edge pairing · Frame independence · Comparison at adjacent edges

Tensor the given source vectors with every ri and the given tests with every ρi. At the chosen edge these tensors are in the normalized lattices because the leading weights add. At the source and dual target vertices their depth is at most t+J. At competing edges the top conditions supply wc and the added basis costs at most w−J, which gives the required divisibility by wc−J. The pairing on tensors is the product of the pairings on the two factors. The new matrix is the original pairing matrix tensored with Ido, and tensoring the integral column operation with Ido puts it in the form zIhdo. Both the certified rank and the ambient dimension multiply by do. The neighboring edges are indexed by the same lattices after restoring the weights. Under the release hypothesis c≥2n′ccl, inclusion gives the stated uniform depth t+2n′ccl and conductor c−2n′ccl. For weights constant along each complementary row, J=0 even if the determinant twists themselves are nontrivial. Depth and competing-edge conditions · Product of pairing matrices and rank normalization

The remaining integral estimates in the initial rank bound concern the internal Gram ties, the terminal Gram coefficients and passage from the untied levels to the block lattices. The pairing step uses admB2 for integral cleaned source and test vectors and admB3 for the margin loss. The determinant estimate, harmonic splitting and the two weight transfers give the mean bound CprNTN1/Y0 from these inputs. Returning from the enlarged intervals to the original uniform measure costs a factor at most two. Segment estimates · Digit-class comparison · Paired-column and margin inputs Recovery of the original mean

Integral harmonic splitting

Realization with prescribed weights and level

Word supply. Write k,m,f for the corresponding sizes in R, and put Γ=exp⁡(CΓm5fxε) and Dword=⌈3Γ2xCF⌉. The rank estimates give m5fxε≤128x9/2+9ε and k≥x9/2+10ε. For fixed CΓ,CF≥0, log⁡Dword=O(x9/2+9ε+log⁡x)=o(k). In particular Dword≤2k−3 eventually. The rank-count argument weakens this integer bound to 2k−2, and Paper III uses its logarithm only to absorb the additive loss into Q0. Thus there is an injection Fin(Dword)→{0,1}k−2 eventually. Choose it once for the kit. The point provenance gives a bijection from its own words to Fin(Dword), with bits equal to those of the common code followed by zeros. Hence identical word indices yield identical bits across points. The additional word tests ChCode_at come from the object and numerical hypotheses on the same kit. Rank estimates and injection Word bijections and agreement Further word tests · Chosen word count · Rank-count application · Peer check of the slope loss

For the geometric construction fix the weights in WinOK, the word code, and a shift u0<n. Use F=E+, f=2n, r=k+m−1, Δ=⌈xcD⌉, and h=⌈xch⌉, with cD,ch fixed and ch≥402. In the ordinary field, finite inertia at q,d can act only on the respective quadratic factor, and inertia at 2 can act only on the real dyadic factor. No finite inertia contains complex conjugation, which changes both imaginary radicals. Hence E/F is unramified at finite places. Arithmetic and realization theorem

At each real place take vi=4Δ(k−2−i), 0≤i≤k−2, and ui+1=vi−12−Δci for the internal coordinates of the word, with ci∈{0,1}. The two endpoints at the top t0 are a(t0)+(k−1)/2 and −b(t0)−(k−1)/2. Away from the tops they are 4Δ(2k−2)+1/2 and −1/2. For 0≤j<m set bj=Bτ+2jh, ej=bj+h, where the base Bτ is b(t0)+⌊b∗/x⌋, ⌈exp⁡(xε)⌉, or ⌈s.w′′(n)⌉ at a top, thick or thin place, respectively, with s.w′′(n)=b∗/(xn). A shift of Bτ at one thick place by less than ∣μ(E)∣ cancels the infinity-type character on CM torsion. Strict interlacing holds since each internal ui+1 lies between vi−4Δ and vi, and the endpoint inequalities use 4kΔ≤a∗, ⌊b∗/x⌋≥k. The b,e gaps are h≥(t+3)/2 since t≤x400 eventually. The compact rank-k dimension is the Weyl product ∏i<j(ui−uj)/(j−i). The row parameter uses vi and −bj, with alternating signs on the latter entries. Its normalized polynomial is ∏i<j∣λi−λj∣/(j−i), with λ=(vi,−bj). Weight formulas and torsion shift · Row formal-degree polynomial · Prescribed family of weights

At either completely split type prime g, choosing one member of a split pair identifies the unitary group with GL on that member. The other component is the inverse adjoint. Reduction gives the column upper unitriangular subgroup in rank k. In the row group of rank r, use the upper unitriangular matrices with the central diagonal block of size k+1 fixing its last basis vector. This imposes k extra zero entries above the diagonal in that block. The central diagonal block is multiplicative on upper block triangular matrices and respects inversion, so these matrices form a subgroup. Counting free entries and dividing ∣GLs(Fg)∣=gs(s−1)/2∏i=1s(gi−1) by the subgroup orders gives Icol(g)=∏i=1k(gi−1),Irow(g)=gk∏i=1r(gi−1). Each gi contributes f such indices. The signed tame list at gi consists of the exponents 0,±2,…,±2m of a primitive residue unit. Outside Br(gi) its prime-to-ℓ order is at least (gi−1)/gi1/4>4m eventually, so the list stays distinct under the prime-to-ℓ projection. The disjoint excluded sets ensure that one of the two choices works at each coefficient prime. Exact subgroup and tame calculation · Local level definitions

The signed row covolume

Let F=E+, f=[F:Q], and let J=Ir−m/2,m/2 be the Gram matrix of the selected row Y0. Give its finite integral stabilizer volume one and use compact-dual Haar at each real place. Then vol⁡(U(J)(F)\U(J)(AF))=∑[L]∈gen⁡(Ir)1∣Aut⁡(L)∣=Mr. The ordinary construction installs the adelic quotient volume for this signature on the datum used by the row count. Choice of quotient measure · Row size

At a finite place the extension is split or unramified quadratic. There is a unit u of norm −1. In the split case take (−1,1). In the unramified case lift a residue-field solution through the principal unit filtration, using surjectivity of the residue trace at each step, including residue characteristic two. The diagonal matrix with entries 1 on the positive coordinates and u elsewhere gives an integral isometry with the identity lattice. To compare rational gauges take an F-rational top form ξ on the skew-Hermitian matrices and set (ωH)1=(det⁡H)−r(X↦HX)∗ξ for Gram matrix H. Under H=A∗H′A the skew-coordinate change Y↦A−∗YA−1 has determinant N(det⁡A)−r, as one sees by splitting the quadratic algebra. The determinant prefactor cancels it. Thus the finite stabilizers have equal gauge volume.

At a real place take imaginary diagonal and complex upper triangular coordinates on the Lie algebra and let dCH be their coordinate density. For the compact form, the last-column map U(s)→S2s−1 gives successive factors 2πs/(s−1)! in its total coordinate Haar. Write Vr=∏s=1r2πs/(s−1)!. The compact-dual convention equates the two real densities in the complexified Lie algebra. The map X↦JX preserves the absolute coordinate density, and ∣det⁡J∣=1. Hence the compact-dual identity density is dCH/Vr for both forms and its ratio to the rational gauge agrees at each real place. The analytic convention multiplies the group measure by Vr and divides its discrete-series formal degree by Vr. The factors cancel over the f real places. Compact-dual convention · Change of Haar

The Tamagawa convergence factors are the same because the two groups have the same character module and dimension. Both unitary groups have Tamagawa number two. The equal local comparisons thus give equal adelic quotient volumes in the specified measures. For the positive form the real group has volume one. The finite double cosets classify the lattices in the genus, and the stabilizer in each class contributes the reciprocal of its finite order. This gives the displayed sum. Spectral occurrence of a row with the required finite type and rational periods is a further assertion about this quotient. Kottwitz, Tamagawa numbers · Definite double cosets · Gauge and compact-dual density calculation

The mass ratio and the word-level prime

Let Ms denote the mass of the genus of the positive identity lattice in rank s. Write Icol and Irow for the products of the corresponding local indices over all 2f type places. For a strictly decreasing list y of length s, put V(y)=∏i<j(yi−yj) and W(y)=V(y)/∏j=0s−1j!. At a real place τ the two lists are u=u(w,τ) and λ=(v0,…,vk−2,−b0,…,−bm−1). The compact Weyl dimension and the normalized discrete-series formal-degree polynomial in the release give the two explicit sizes dw=MkIcol∏τW(u),D∗=MrIrow∏τW(λ). The formal-degree normalization follows by inserting the decreasing parameters in the product of absolute root differences divided by ∏i<j(j−i). The argument also works for any D∗ at least (2/3)f times the displayed row expression. These sizes are defined before the prime for the word and before the selection of the automorphic row. Definitions of both sizes · Row polynomial and the lower comparison

The column size is a positive integer. On a lattice in the genus, the choices of complete residue flags with a basis of each successive line form a set of size Icol. The finite group of lattice automorphisms acts freely on this set. Indeed, at a place above gi a stabilizer embeds into a conjugate of the inverse image of the upper unitriangular group. This inverse image is pro-gi, since each principal congruence quotient has gi-power order and the reduction is a gi-group. A finite subgroup of a pro-gi group has gi-power order. The local representation is faithful by scalar extension, and g0≠g1. Thus the stabilizer has order one. Summing the orbit counts over the genus gives IcolMk. Finally u is strictly decreasing in a single translate of Z with the compact Harish-Chandra shift. Subtracting the half sum of positive roots gives a dominant integral weight, so each W(u) is a positive integer by the Weyl dimension formula. Column level

Put vE=∣disc⁡F∣1/f/2, so π≤vE and log⁡vE≤L=xε on the selected host. Let η be the quadratic character of E/F. Paper V's norm-one sphere calculation gives, for k<s≤r, MsMs−1=((s−1)!)f(vE/π)sf(2vE)−f/2L(s,ηs). At residue norm q, the Euler factor is (1−η(p)sq−s)−1. Its absolute logarithm is at most −log⁡(1−q−s). Grouping over rational primes bounds ∣log⁡L(s,ηs)∣ by flog⁡ζ(s). The factorials in the telescoping product cancel exactly with the Weyl denominators in ∏τW(λ)/W(u). Mass multiplier in the release

For the remaining real factors, compare every interior pair 1≤i<j≤k−2. Interlacing gives ui−uj≤vi−1−vj. The pairs (i−1,j) run once through all nonadjacent pairs of the v's. Each of the other k−2 differences in V(v) is 4Δ. The 2k−3 pairs of u touching an endpoint have difference at most 8a∗ eventually, uniformly in the word and place. At a top this follows from u0≤2a∗ and uk−1≥−a∗, and elsewhere kΔ grows polynomially while a∗=xexp⁡L. Appending the negative entries to v contributes at least ∏jbjk−1, since vi≥0 and the bj have gaps at least one. Hence V(λ)V(u) ≥ (4Δ)k−2(8a∗)2k−3∏j=0m−1bjk−1. The chosen weights satisfy bj<ej≤bj(1+1/x) and 1≤ej≤2exp⁡L. Dividing the logarithm by k therefore gives at least ∑jlog⁡ej−Oε(L+(m/k)L+m/x), using log⁡x=Oε(L). Selected place bounds · Interlacing hypotheses

At a type place the index ratio is gk∏s=k+1r(gs−1). Its power of g is mk+m(m−1)/2. The logarithmic cost from the factors 1−g−s is bounded by an absolute constant, since g≥3. In the mass multiplier, ∑s=k+1rs=(m−1)k+m(m−1)/2. After division by fk, the field power contributes at least mlog⁡(vE/π)−L. The factors (2vE)−f/2 and the Euler products together cost O((m/k)(1+L)), and the optional row factor costs log⁡(3/2)/k. Combining these inequalities, using m≤x eventually, proves uniformly in w log⁡(D∗/dw)fk ≥ m(log⁡g0+log⁡g1)+1f∑τ,jlog⁡(ej(τ)vE/π)−Cε(L+m2L/k). The summands in the average are nonnegative. Both log⁡gi≥8L, and k≥m2, m≥2Cε eventually. The quotient is at least 15mL. Also 2log⁡Γ/(fk)=2CΓm5L/k≤mL eventually. Thus Sw=(D∗/dw)1/(fk)Γ−2/(fk) satisfies log⁡Sw≥14mL. The signed row covolume calculation above identifies the mass factor at the chosen stage. The automorphic interpretation uses the specified level and real weights. Numerical window and penalty · Comparison on realized data

The norm-prime estimate above applies with X=Sw, M=N(g0g1)2, h=1/(fk), since fk≤x12, m≥x1/3 and log⁡M=O(xε). It gives Sw≤T(w)≤(1+1/(fk))Sw,πT(w)πT(w)‾=T(w),πT(w)≡1(modN(g0g1)2OK). The primes split completely in E and are larger than X′ and the five initial level primes eventually. The host level set adds the T(w). The choice is per word, with repetition permitted. Mass normalization, window bounds and their hypotheses · Pinned numerical window proof

At each of the f places above T=T(w) the word level specifies a nonzero covector on the residue space. There are Tk−1 possibilities since GLk(FT) acts transitively on its nonzero covectors. The type flags have already removed all stabilizers. Thus adjoining the word-level data multiplies the class count by (Tk−1)f. A coefficient-valued function on the framed classes can take an arbitrary value of the weight space on each class. Under the identification with that function space its dimension is dw(Tk−1)f. The resulting ratio in the final dimension condition is Γ2dw(Tk−1)fD∗=(TSw)fk(1−T−k)f. The first factor lies in [1,e]. The second lies in [1−fT−k,1], and fT−k≤1/2 eventually. The ratio therefore lies in [1/2,e]. This is where the pre-level mass estimate is used in fitting the sizes after the prime has been chosen. Covector index, class count and space of coefficients · Final dimension condition

Write Nj=k+j(t+1), Rj=r+jt and qj=⌊(m−j)/2⌋. Take E-spaces with diagonal Gram matrix having the indicated number of plus signs first, and give them the self-dual lattices OErank.

Space Rank Signature at each real place
Xj, 0≤j≤m Nj (Nj,0)
X′ Rm (Rm,0)
Yj, 0≤j≤m Rj (Rj−qj,qj)
Vj, 1≤j≤m r+t (r+t,0)
Wj, 1≤j≤m r (k+j−1,m−j)

Each form J defines its group by AˉtJA=J. The positive forms give compact real factors. Successive Xj add t+1 positive lines. For j<m, embed Xj into the first Nj coordinates of Yj. The remaining m−1−j coordinates admit ⌊(m−1−j)/2⌋ hyperbolic pairs e++e−,(e+−e−)/2. A spare negative coordinate occurs exactly when j is even, and the vector πTe− there has norm −T. Here T is the norm prime of the distinguished word. At the top Ym=X′, and the last line of Xm contains the perpendicular vector πTe of norm T. The coordinate lattice is self-dual since J is unimodular. Forms, embeddings and hyperbolic frames · Pinned signed model

At T the auxiliary column lattice Mb=OEk−1⊕πTOE satisfies Mb+πˉTMb#=OEk. The self-dual split move is given on the two factors of the selected coordinate by (T,T−1). On writing O=OF,v at v∣T, O⊕O and TO⊕T−1O meet in TO⊕O, of index T on each side. At a such places the index on each side is Ta, in particular Tf if every place over T is used. Negative or positive orthogonal lines and their fractional ideals extend the lattice models, with the dual ideal on a unit-norm line equal to the inverse conjugate ideal. For a finite adelic isometry af the transported lattice is Erank∩afM^. Clear denominators of af and af−1 by a single integer D. Chinese remaindering in D−1M/DM realizes the prescribed completions and bounds the index on each side of the intersection by D[E:Q]rank⁡M. Left rational isometries transport this intersection and the decorated right stabilizer preserves its reductions. For a sequence of finite adelic moves the product of the successive denominator bounds suffices. The shift u0 resets the distinguished host index, with TT,T2, weights and code carried by provenance. Local moves and integral gluing

The two automorphic passes and the point ledger

Release inputs (the automorphic realization). The first global step after the signed and decorated lattice models is the saturated automorphic seed. It is a row reading at the prescribed real weights and finite types, compatible with the column models and their intertwining maps. In the untwisted pass the active coefficient places, selected by the local required content of the lifts, must be a union of fibres over the fixed row coefficient field L0. A whole fibre is selected whenever one of its places is. The first state has empty localization and zero local depth. Its pinned construction invokes the fine adelic reading and its extension in the ambient coefficient field, the lower-unitary multiplicity equality and the weight-specific rational row test. That test uses the specified local Weil-vector matching, the theta integral comparison and its recurrences for adapted plane changes. The finite-type trace argument and the construction of the separating periods below give the two spectral requirements under their respective hypotheses. The rationality step uses the normalized matched theta comparison on the same grid, including both adapted-plane recurrences. Saturated seed and precise spectral hypotheses

In the first pass choose the fixed fields, split-place orientations and true columns. The twist characters initially equal 1 and the central tuples include the two distinguished characters as well as 1. Keeping the prescribed Galois level, install the rational row lattice, cusp data, defining forms, induced series and moduli data. The strict-henselian and finite spectral steps give a finite-level state with the same support. The discrete-series replacement carries the actual row evaluation and its norm identity, the real volume normalization, the local types and the column models. Join the row, column, division and theta data using the matched characters and intertwiners. This gives the first core and its export. Construction through the spectral replacements · First and second pass

In the second pass use that export to select finite-order twists at the active places with conjugation, agreement over the fixed coefficient field and the local avoidance conditions. Enlarge the ambient coefficient field for their values while preserving the fixed embedded fields and the rational prime list. Run the row, spectral and kernel constructions again with these twists. The resulting core supplies good local twists for every required exponent prime to the relevant residue characteristic. The normalized volume and column-count comparisons identify its actual pre-level sizes with D∗,dw. These comparisons, and the column denominator choice MinkDiv, enter the recursive object theorem. Second pass and the required count and volume equalities · Actual two-pass assembly

The resulting RecOK records Δ, the error parameter 6x−CF, finite active support, vanishing lift corrections off that support, fullness of the ramified division module, the column denominator and tame conditions, and local comparison at T(w). For any valid size record at the same scale and weights, LedOK supplies the local containment and denominator bounds for the canonical parameters. After refilling the parameters, glue the pulled-back kernel lattice at served places, the prescribed comparison lattice above T(w), and the ramified division lattice at all other places. The exceptional set is finite since the served primes are at most X′. The local saturation and comparison bounds give the stated localizations and allow the ledger to record their lengths divided by ramification indices. The division-lattice replacement preserves QD, which requires used tuples, good twists with conjugation and the specified prime-to-residue-characteristic powers, and stability of the row lattice under the full coefficient integer ring. This gives the object at the original scale and weights, with provenance Pt.From and a canonical instance. Exact ledger gluing and QD · Extraction of the point

Saturation of the reference support

At the first pass fix the arithmetic host, the frames, and the prime norms. In the CM field E, there are only finitely many integral generators of a prescribed norm ππˉ=T. In every complex embedding the generator has modulus T, and the Minkowski lattice meets this compact set in finitely many points. Choose a fine automorphic realization for each required generator tuple, assuming these realizations exist. Their reference fields lie in one finite coefficient field L0. Each pair of full lattices in a common vector space is commensurable, so the selected comparisons have finite prime support. Reference choices and the two index pairs

Let A⊆L0⊆L be the chosen reference field, fixed coefficient field and enlarged coefficient field in their distinguished embeddings. Write UA for the support of the two signed reference indices, including their conjugates. For either pair M,N, scalar extension carries M∩N to the intersection after extension by flatness. At w∣v each finite quotient length is multiplied by e(w/v), as one sees on a residue-field composition factor. Thus the signed difference also scales by e(w/v). The Hermitian dual commutes with this extension in a basis by the inverse Gram matrix and its conjugate scalar action. Given the source's simultaneous base change of the two reference pairs, the active set in L is the inverse image of UA together with the primes over rational residue characteristics where the prescribed content is nonzero. Both sets are unions of fibres over primes of L0. Taking this restriction and setting all initial division depths to zero proves the required saturation. This gives finite support for each fixed host. A conductor exponent at a supported prime requires its own estimate. Restriction over the fixed field

How the finite types exclude proper blocks

Here is the cuspidal multiplicity argument for a row with the prescribed good Green types. Write r=k+m−1 and s=m/2. The two unitary groups have signatures (k+s−1,s) and (k+i,m−i−1) at every real place for 0≤i<m. Their finite adelic isomorphism carries the chosen frames and levels. In the image of the finite projector on a passing irreducible constituent choose a single tensor of finite types ⨂vτv appearing there. The polynomial condition on that tensor gives, for each 1≤a<r, a split place v where the degree-r polynomial labeling τv has no divisor of degree a. For 2a>r use complementary divisors. The place can depend on a. Polynomial condition on the selected tuple · Extraction from the actual projector

The projector in this argument can be identified coefficient by coefficient on the fitted finite quotient. Write Ev,f for the character projector of the Green series labelled by f at v. Then E0=∑f good⨂vEv,fv. The embedding of the prescribed coefficient field sends the central idempotent of the reference class lattice to this very sum. For 0<a<r, average over the product Ua of the finite (a,r−a) radicals at the selected places above g0, and denote its idempotent by ua. In each tuple there is a place whose polynomial has no degree-a divisor. The finite parabolic restriction there vanishes, so every tensor summand kills this average. Thus E0ua=0. Equivalently, ∣G∣[1](E0ua)=∑τ: E0∣τ=1dim⁡(τ)dim⁡(τUa)=0, by taking the regular trace. Every passing irreducible tuple therefore supplies a place for each cut. The level-character projection restricts the image of E0 further. This proves the finite support assertion on a row carrying the specified type data. Character descent and the trace test

At this place the radical of the finite parabolic with block sizes a,r−a has zero invariants in τv. Indeed the inducing finite cuspidal blocks have the degrees of the irreducible factors of the Green polynomial with multiplicity. In the double-coset formula a block crossing the cut has invariants under a proper radical and contributes zero. The surviving blocks on the first side would give a divisor of degree a. In the local group, taking principal-congruence invariants of a representation induced from this parabolic gives finite parabolic induction. This follows by restricting its functions to GLr(Ov) and then reducing modulo the maximal ideal. Compact invariants are exact over C, so every subquotient of the local induction also fails the type test. In particular a representation induced from a proper rational parabolic has zero contribution at the split witness for its first isotropic step. This excludes the residual spectrum for the chosen tuple. Finite and local parabolic argument

For the trace comparison assume stabilization of the discrete trace for these unitary inner forms and the archimedean pseudo-coefficient identities, with matching measures. Compress each finite Hecke function on both sides by the idempotent of the chosen tuple. At the split witness for a, averaging the idempotent over the integral radical gives zero. The parabolic descent of every compressed function there consequently vanishes. The proper elliptic endoscopic datum of sizes a,r−a at that split place has exactly this transfer, up to the matching character factors. Every proper term of the stabilized comparison thus vanishes. At infinity the two prescribed discrete series have the same decreasing regular infinitesimal character. The stable elliptic orbital integral of the trace-one pseudo-coefficient is the signed Weyl character expression for that list. At signature (p,q) the sign is (−1)pq=1 since r is odd, and the orbital integral vanishes on nonelliptic regular tori. The strong gap assumption gives the trace selector on the unitary spectrum. The identity datum therefore has equal stable traces on the two forms. Real orbital signs and spectral hypotheses · Stabilization in Kaletha–Mínguez–Shin–White, §3.1

At fixed level and infinity type the contributing cuspidal list is finite. In the common finite Hecke algebra, the nonzero spaces of level invariants of its distinct irreducibles are inequivalent simple modules. The density theorem on their direct sum provides an element acting as the identity on the specified constituent and as zero on all the others. The equality of the two traces gives the difference of the two multiplicities times the rank of the type projector on this constituent. Its rank is positive. This proves equality for the individual passing constituent, including the case of multiplicity zero. The argument uses the polynomial support condition on the same row chosen for the starting datum. Multiplicity separation and scope of the trace argument

A common normalization of the row periods

Work on the fine datum restricted to the active coefficient places, given by the support of the two reference indices and their conjugates, together with the places forced by the lift content. The restriction keeps the framed group, core and good-type idempotent and uses its restricted tuple dictionary, including the reference tuple μ0. In the source, the period calculation takes place on the copy with zero division depths and the same group. The same-group comparison carries the tests back to this datum. Let V be the whole complex row, and write Fμ:Aμ→V for the map of the rational tuple source into that row. The Hilbert model Hrow of V identifies with the good cuspidal multiplicity space EEat(1) via an injective map onto it. In the Hilbert calculation we use Fμ(x)=Thμ(jX,μx) and KH=embLL(L) for the image of the ambient coefficient field. The scalar twist at the end gives the corresponding maps and field in V. Restriction and weight step · Same-group row construction · Good Hilbert model · Complex row

Cuspidal transfer on the restricted datum. For each i<m and each irreducible admissible finite-adelic σ with a nonzero flag-level vector fixed by the good projector, the required equality is mcusp,Y0(λ,σ)=mcusp,Wi(λi,σ∘Φi−1). Here Φi carries the finite frames, and λ,λi are the prescribed real models at all real places. The preceding stable trace argument gives this equality under the stated transfer and pseudo-coefficient hypotheses. The reference reading uses the equality in the lower-unitary comparison. The weight comparison uses it on the restriction alongside rationality of the period maps. Good constituent and transported multiplicities · Separate spectral inputs

The separating periods. Fix the grid and its Weil data and Haar measures on the actual row. The constructor takes a nonzero smooth tensor w by descending a top test vector along the chain of intertwiners. For u∈Hrow let b be its cuspidal row map from the smooth real tensor model, and set f=b(w). With Φ a finite grid datum, ρ real Schwartz data, Ψ a definite column, and ξ a complex-linear functional on its coefficient space, use Pρ,ξ;Ψ,Φ(u)=B+(pairVec⁡(ξ,Ψ), θχ(ρ,Φ)f). Here θχ is the upward row lift on the grid. The period B+ first evaluates the lift at nι(h) and integrates against the conjugate of the prescribed Bessel character over the unipotent flag. It then integrates against the scalarized column pairVec⁡(ξ,Ψ) on the definite stabilizer. The conjugation acts on the fixed character. Evaluation at w and the convergent integrals are complex-linear in u. The row-tower and Bessel convergence results give this linearity on the entire good target. Test vector and period construction · Row lift and Bessel integral · Linearity · Convergence

Take any u≠0. Its map b is nonzero and belongs to the cuspidal row at the principal level above g0g1, since that subgroup lies in the framed row level with trivial character there. The bottom-period argument takes a real test at which the cusp form is nonzero and an adelic point where its modulus exceeds half its supremum. The real period estimate and the grid depth make the distinguished term of a translated coefficient Poincare series dominate the other terms. The series gives a form of the bottom core group whose pairing with the row on the embedded U(Y0) is nonzero. The real translates of the given nonzero w detect this same pairing by irreducibility of the real model and continuity. Changing variables moves the translate to the core form, and density gives a compact-finite core vector. Thus some bottom core form evaluated at a compact-finite vector pairs nontrivially with b(w) in the pinned quotient measure. Principal level · Bottom evaluation and principal depth · Poincare series and its period · Transfer to the fixed tensor

At the specified even-rank core levels this core evaluation is in the span of descending theta chains from definite columns. The release proves surjectivity on these levels by checking the upward adjoints on the spanning Hecke components. It uses transferred parameters, nonvanishing of the real and semilocal zeta factors, and the normalized doubling formula with its nonzero L-value. Pairing with b(w) is linear on the span, so some single descending chain gives a nonzero bottom period. The depth conditions and spanning by the admissible conjugated Fock vectors transfer nonvanishing first to a Fourier–Jacobi period, then alternately to Bessel and Fourier–Jacobi periods. Since m≥2 is even, the final Bessel period is Pρ,ξ;Ψ,Φ(u)≠0 for some choice of its data. These steps use the prescribed real branching and gap estimates and the split and unramified Weil comparisons, with the convergent Siegel–Weil and doubling identities on the auxiliary core tower. Separation inputs · Surjectivity at the even ranks · Spanning by descending columns · Passage to one chain · Depth conditions and Fock tests · Last Bessel period

The terminal Bessel forms have dimension at most one at each real place in the prescribed gap. Taking their tensor product gives a factorization Pρ,ξ;Ψ,Φ(u)=a(ρ,ξ)λ(u,Ψ,Φ) for the fixed w. If the row is nonzero, take one nonzero period and fix its real pair (ρ∗,ξ∗), so a(ρ∗,ξ∗)≠0. For each nonzero u, a period found by the preceding argument has λ(u,Ψ,Φ)≠0 and remains nonzero at this fixed pair. For each Φ, the columns indexed by a definite double coset and a weight basis vector span all definite columns. At the neat level their values at the representatives are the basis vector on the indicated class and zero elsewhere, and equivariance recovers the full column from those values. A period linear in the column and nonzero on this span must be nonzero on a basis column. The finite tests ψa are precisely these periods at (ρ∗,ξ∗), indexed by Φ, class and weight coordinate. Their common kernel on Hrow is zero. The argument applies to arbitrary linear combinations of cusp forms and multiplicity copies, including row vectors outside the span of the tuple images. The good idempotent alone acts as the identity on this whole target. Terminal real line · Fixed pair and whole-row separation · Class and weight values · Span at fixed datum · Choice of the actual tests

The matched integral and relative scalar. The good projector e acts over the coefficient field on Aμ. It fixes the target, so equivariance gives Fμ(ex)=eFμ(x)=Fμ(x). For the projected sources and the same tests just constructed the release uses ζμCμ ψa(Fμ(ex))=κ vμ,a(ex),Cμ=∏i,jcij,μ. Each cij,μ≠0 compares the two orders of the theta operations around a square. Flipping the squares exchanges the downward tuple path and the upward testing path and contributes Cμ and a finite phase. Unfolding the resulting path into finite cosets gives a second phase. Their product ζμ∈KH× is a root of unity. The orthogonal-sum jet comparison gives a single κ for all terms, tuples and tests of this choice. The finite sum vμ,a is over normalized branch pairings on the definite classes. With the prescribed character values, local phases, jets, and coefficient pairings in KH, its rational volume factors and reciprocals of stabilizer orders show vμ,a(ex)∈KH. Projection of the sources · Simultaneous identity · Square flips · Unfolding and jet comparison · Rational finite sum

Matching in the local quotients. Write I(Φ;h) and I′(Φ′;h) for the theta integrals over the two source quotients, of ranks a and a′, with h∈U(W) and dim⁡EW=a+a′. Use the same trace character, the prescribed Weil splittings and ν=α/β. Right translation by the integrated group kills the finite augmentation differences. At a good split place these differences are the entire kernel of the local quotient map, and data of full source rank surject onto the primed quotient. Use such a representative at the distinguished good split place v1. Match the sums of tensor images in the polarized real quotient and the finite quotients, where the spherical images agree off T=S0⊔Q and the two split matrix zeta maps give the images at T. The pinned factorization uses vanishing against cusp forms to descend the polarized comparison to this common quotient. Its spherical, ramified and real Hom calculations determine the local comparisons used in descent. Under the stated rank and real weight gaps, absolute convergence and integrable Schwartz bounds, using full source rank on the primed side, extend the polarized comparison to the allowed Schwartz limits. Source note · Local maps and support · Full-rank partners · Quotient comparison and limits

Definite endpoint. Use the classical anisotropic Siegel–Weil value for the standard evaluation section at s=(M−N)/2, where M is the integrated dimension and N is the rank being doubled. It gives the theta average multiplied by 1 for M>N and by 2 for M≤N. At the definite endpoint, double each source in turn and use M=a+a′ and N=a or a′. For the box tensor of a test and its conjugate, vanishing of its theta lift forces the doubled zeta integral to vanish at this point. Unfolding with the pinned measures and clearing the good denominators before evaluation gives a product of nonzero real and exceptional zeta factors and continued Hecke values at M/2−i, 0≤i<N, for ν±1. The real exponent gap is used for the values outside absolute convergence. Thus neither lift map vanishes identically. Gan–Qiu–Takeda, Theorem 7.1 · Yamana, Theorem 2 · Normalized unfolding · Endpoint nonvanishing

On the compact quotient of U(W) the invariant mixed pairing on the common domain determines a scalar c for which D=I−cI′ is orthogonal to the primed image. Its Gram form is a scalar multiple of the primed Gram form. For a matched test with nonzero primed lift, the reverse theta integrals of D and I′ are constant, with respective averages zero and a nonzero multiple of the primed norm squared. Double W with integrated rank a′ and continue the unfolded Gram comparison with its moderate-growth bound. At s=−a/2 the factor-2 Siegel–Weil value computes both pairings from those reverse integrals. The Gram scalar of D is therefore zero. This gives I(Φ;h)=cI′(Φ′;h) for all matched pairs and every h, and the nonvanishing witnesses give uniqueness and c≠0. Mixed pairing and reverse lift · Endpoint continuation · Conditional comparison constructor

Adapted-plane recurrences. Remove a hyperbolic plane from W and from the changing source. With d=(a−a′)/2 at the upper stage, the β-move lowers a′ and the α-move lowers a after the primed side is definite. On the adapted line parabolic the two Levi powers separate the constant-term components. In the β-component the unprimed term is evaluation and the primed term is a cone integral. In the α-component their roles are reversed. Unramified cone shells give the partial LS0 factor, and ramified Fourier transforms give the epsilon factors with their conductor powers and uniformizer values. In the pinned witnesses the two line-ideal character values have exponents (−1,−1) for β and (1,1) for α. Write Dβ(ν) and Dα(ν) for the products of these ideal values and the epsilon factors at −d and 1−d, respectively. The local maps match evaluation to cone integration on the lower quotient. Lifting a nonzero lower comparison by surjective evaluation makes cancellation in the chosen component legitimate. With c1 the lower scalar this gives c GβDβ(ν)LS0(d+1,ν)=c1,c=GαDα(ν)LS0(d,ν)c1. The geometric factors Gβ,Gα are nonzero and fixed before the characters vary when the geometry and real maps are fixed. Starting at the definite endpoint, these recurrences determine the upper scalar. The pinned induction handles the remaining constant-term components and uses its cuspidality and orthogonality arguments to force the residual to vanish. Adapted planes · Line ideals · Cone factors · Constant terms · Induction

Compatibility on the selected grid. The square constructor transports both orders into the comparison triangle, including their Haar measures. Its nonzero matching scalar multiplies c to give the square scalar. The argument takes the classical theta and doubling facts and the pinned constructors as inputs. It still assumes that the selected grid identifies the unfolded quotient, compact and additive measures, the normalized real and finite maps on the actual tensors, and the required data in its chosen coefficient field. In particular, the stage transports must put the tuple and reference on common real maps before comparing their geometric factors. Rescaling the two quotient measures by tuple-dependent factors would change the square ratio by the corresponding relative Haar factor. Square comparison · Haar transport · Common real maps · Grid assembly

The definite comparison and the local and special-value ratios below give Cμ/Cμ0∈KH× when the indicated field conditions hold. The release chooses the ambient field to contain the required roots, square roots, jets, ratios and definite comparison data. If κ≠0, put ψ~a=(Cμ0/κ)ψa. For every tuple and source vector, ψ~a(Fμ(x))=vμ,a(ex)ζμ Cμ/Cμ0∈KH. The one nonzero rescaling preserves separation, and the reference ratio is 1. Dividing each phase and κ by ζμ0 sets the reference phase to 1. When a period image is nonzero, some ψa(Fμ(ex)) is nonzero, so the simultaneous identity forces κ≠0 and fixes it from that same test once the phases of the matched comparison are fixed. This works even if the reference map vanishes. When κ=0 all period images vanish by separation, and the unscaled tests work. A zero target admits the empty family. A zero good part of any source contributes zero by equivariance. Field data on this family · Rationality of the square ratios · Rescaling and zero case · Reference phase convention

In the complex row V, the source twists the scalar action on the Hilbert model by the identity or complex conjugation, according to its relation to the distinguished embedding at v0. Apply the same conjugation convention to the output of each test. In the conjugate case a complex scalar is conjugated once on the input and once on the output, so the transported test is complex-linear. Its values on the tuple sources are in K=v0(L), with L0 used separately for support saturation. Equivalently, the kernel of the joint map on (⨁μAμ)⊗KC descends to K. Composing with the separating tests gives a matrix over K with the same kernel, and field extension is exact. Conversely, a descended kernel gives a K-basis of the quotient whose images form a complex basis of the joint image. Extending this to a basis of V and taking dual coordinates also separates the complement of the joint image. Scalar twist of tests · Transfer to the complex row of the datum

The special values in the common row normalization

Consider a square (i,j) of the theta construction, with 0≤i,j<m and zero-based indices. Write cij,μ for its nonzero comparison scalar at tuple μ and μ0 for the reference tuple. The associated unitary Hecke character of E is νij,μ=μj0ξμ/χm−i, where ξμ is a finite-order twist and ξμ0=1. Use the finite Hecke L-function with its primitive finite Euler factors, and set Rij,μ(s)=LE(s,νij,μ)LE(s,νij,μ0),E0=trk+r+62,s∗=E0+1. Here trk is the odd rank increment denoted by t in the Hermitian construction, and r=k+m−1 is odd. Indeed the rank at (i,j) is r+jtrk+i(r+6), so the half increment to (i+1,j+1) is the same E0 for every square. Character and tuple convention · Grid ranks

Nonvanishing at the integers in the comparison. At each place τ of the CM type the angular exponent of ν is 2hτ with hτ>Hij, where Hij is half the sum of the two diagonal ranks. To see the strict inequality, the weight construction writes hτ=s+κij(s,τ) with κij(s,τ)>0 up to s=Hij. This exponent is common to every finite-order twist. Let eij be half the difference of the two bottom ranks. The rank formulas give 0≤eij≤Hij. Then LE(n,ν)≠0 for every integer 1−eij≤n≤Hij, and also for every positive integer for either auxiliary finite-order twist of the reference character. Angular weights and ranks · Hecke characters on the selected squares

For n>1 this follows from the absolutely convergent Euler product. At 1, Hecke's continuation theorem makes both LE(s,ν) and LE(s,ν2) holomorphic because the angular type of each is nontrivial. For real σ>1 the Euler logarithm of ζE(σ)3∣LE(σ,ν)∣4∣LE(σ,ν2)∣ is nonnegative. At each unramified prime power its numerator is 3+4cos⁡θ+cos⁡2θ=2(1+cos⁡θ)2, and at a ramified prime the missing factors leave a numerator at least two. If LE(1,ν) vanished, its fourth power would overwhelm the pole of order three, contradicting the lower bound one. The same proof applies to the inverse and the indicated twists. For 1−eij≤n≤0 use Hecke's functional equation. The finite value of the inverse at 1−n is nonzero. The gamma arguments on the two sides are n+hτ and 1−n+hτ, positive integers at every place. All gamma, conductor and root-number factors are finite and nonzero. This proves nonvanishing of the finite value at n. Finite Euler-product convention · Reflection at integers

For the joint rational row test let K⊂C be the image of its coefficient field. On the compatible grid the upper half-rank difference starts at E0. Put q=⌊(m−j)/2⌋, p=⌊(m−j−1)/2⌋ and write R=Rij,μ. At the integers in the recurrences, LS0 equals the primitive finite LE because the omitted factors above S0 are already 1 by ramification. Compare each move at μ with the same move at μ0 after the common real-map transport. The β-moves increase d by one and put R(E0+1),…,R(E0+q) on the left. The α-moves decrease d by one and put the p ratios starting at R(E0+q) in descending order on the right. Division leaves reference-over-tuple local factors for β and tuple-over-reference factors for α. The remaining factor Bij,μ is the product of the step-factor ratios, the stage and endpoint transport units, and the prescribed definite bottom ratio. The latter comes from evaluating a common nonzero test in the polynomial quotient by finite orbit sums with the prescribed coefficients and volumes. Primitive Euler factors · Tower transports · Definite bottom ratio

The resulting identity is cij,μcij,μ0∏a=1qR(E0+a)=Bij,μ∏a=q−p+1qR(E0+a). The empty product is one. Every cancelled factor is nonzero already in the half-plane of absolute convergence. Since m is even, q−p=0 for odd j and q−p=1 for even j. The square ratio is therefore Bij,μ for odd j and Bij,μ/Rij,μ(s∗) for even j. The remaining factor is inverted because it occurs on the left. The prescribed field data must place the local and transport factors, the bottom ratio and the required surviving L-ratios in K× for the individual square conclusion. Multiplication gives the ratio Cμ/Cμ0 used in the common rescaling. Step-factor arithmetic · The two lengths · Square ratio and field implication

In particular, once the definite and local comparisons give Bij,μ∈K×, the single product Rμ=∏i=0m−1∏0≤j<mj evenRij,μ(s∗) belongs to K× exactly when Cμ/Cμ0∈K×, where Cμ=∏i,jcij,μ. This supplies a sufficient field condition for the common rescaling of the separating tests in the row argument. There is a uniform archimedean estimate at the chosen embedding. Taking the logarithm from the convergent Euler series gives ∣log⁡Rij,μ(s)∣≤2[E:Q]log⁡ζ(s). For s≥2 the sum after the term 2−s is at most ∫2∞u−sdu, hence ∣log⁡Rμ∣≤3[E:Q]m22−s∗. The following field calculation determines the extension generated by these values. Later valuation estimates in the release also use reflected values and bottom theta norms. Field comparison on the bottom triangle · Finite-place valuation argument

The field generated by the critical ratios

Write K0=Q(−d)⊂E for the imaginary quadratic subfield, reserving B for an embedded coefficient field containing all embeddings of E and the finite character values on the grid. For a fixed tuple μ, let θμ be the finite character of GK0ab corresponding by class field theory to ξμ∣AK0×. If Hμ(n) is the fixed field of ker⁡θμn in K0ab, then B(Rij,μ(s∗))=BHμ(1),B(Rμ)=BHμ(m2/2). The first equality also holds for every integer 1≤s≤bigN⁡(D)+1 required by the individual-ratio clause. Here bigN⁡(D)+1 is the release's common upper bound on that list of arguments. Field clause · Angular exponents throughout the list

Indeed both Hecke characters are algebraic of weight zero. Their angular exponents 2hτ agree at every complex place and hτ>s on this range. The arguments s+hτ and 1−s+hτ of the archimedean gamma factors are positive. The finite-order twist contributes a rank-one Artin character. Blasius's critical-value theorem for CM Hecke characters and his twist formula over a totally complex field give the ratio, up to B×, as one nonzero Artin period. The common powers of 2πi cancel. The de Rham quotient selects one of the two Hodge entries −hτ,hτ at each place. This half is independent of the square and of s, since there is no middle Hodge component. The embeddings in the chosen CM type all restrict to the same embedding of K0. The transfer defining the Artin period is consequently the usual transfer from GK0 to GEab, or its conjugate for the other choice of half. On ideles it sends a K0-idele diagonally into the E-ideles. Conjugation inverts the restricted twist because the twist is trivial on AF×. Blasius, §§M.11–M.17 and L.6–L.8 · Restriction of the selected CM type

For g∈GB the multiplier g(Rij,μ(s))/Rij,μ(s) is thus θμ(g)±1, with a common sign. All these nonzero ratios for a fixed tuple span the same B-line. Fixing a single ratio amounts to killing the character, whereas fixing the product amounts to killing its m2/2-th power. If dμ=∣θμ(GB)∣, the two extension degrees are respectively dμ and dμ/gcd⁡(dμ,m2/2). In particular the full field condition for the primitive ratios in one tuple can be tested at one square and one integer. The row rescaling uses only the product, given the local and definite bottom comparisons. · Primitive conductors and Blasius periods

In the ordinary existence construction the ambient row field can include all the individual class fields. At fixed preliminary datum and frame the moving seed supplies a finite set Ξ of twist characters before the final ambient field is chosen. Add generators for Hξ(1) for each ξ∈Ξ to the finite list of algebraic numbers prescribed in that ambient field. Generators for the host embeddings and character values are included as well. For the latter one may use the existing character-value lists on the representative grids for the finitely many keys (ξ,Q), where Q records oriented ramified places. Pullback along enlargement is surjective on coefficient places and preserves their residue characteristic. It therefore gives exactly the same twist set and the same avoided norms on E. The construction preserves the embedded small field FO(D), whose degree and root discriminant occur in the size estimate. Consequently this prescription accommodates the ratios at the original cutoff and field-size budget. It supplies a field of definition. The local valuation inequalities for the squares, the definite bottom comparison and the identities for the chosen theta kernels enter separately. The field choices and order of quantifiers · Pullback along enlargement · Finite prescription and ring class fields

Conjugates and the signed finite degree

Let B contain the values of the characters on finite ideles and the images of the host under all its embeddings. For η:B↪C, form Rij,η at s∗ by applying η to the Euler coefficients at each prime of E. Blasius's theorem is an identity in a tensor product. With Ψ the selected CM type and φ the Artin character of the twist, it gives (Rij,η)η=(bij⊗1)c(Ψ,φ),bij∈B×. If g∈GQ and η=g∣B, the η-component of 1⊗g on a tuple (zλ) is g(zid). The reciprocity formula for c(Ψ,φ) therefore gives g(Rij,id)=ω(g)Rij,η with ω(g) a root of unity common to all squares. For g∈GK0 the induced type gives ω(g)=g(θμ(g)±1) with the same sign as in the field comparison. This application uses the full tensor identity, including the conjugates η(bij). Blasius, M.13, M.17 and L.6–L.8

For either Hecke character χ in the ratio, take a ray-principal power of an unramified prime p. Its chosen generator a is 1 modulo the conductor, so a power of χ(p) is a product of integral powers of τ(a)/τˉ(a). The embeddings of a CM field preserve these conjugate pairs. Thus every conjugate Euler coefficient has modulus one. The identity χ(aˉ)=χ(a)−1 follows from triviality on the real ideles and persists at every coefficient embedding. Distinct conjugate primes give positive paired factors in the Euler product. An unramified prime fixed by conjugation has coefficient 1 or −1, again giving a positive factor. Thus Rij,η>0. For s≥2, log⁡ζ(s)≤2−s+∫2∞u−sdu≤3⋅2−s. Applying the Euler logarithm bound to both values in each of the n=m2/2 ratios gives ∣log⁡∣g(Rμ)∣∣≤3[E:Q]m2 2−s∗(g∈GQ). Consequently for any number field L containing Rμ the same upper bound holds for ∣∑Pord⁡P(Rμ)log⁡NP∣/[L:Q], by taking the absolute norm. The finite prime contributions enter this sum with their signs. Angular characters and their restrictions

There is also an exact exponent for total positivity. Put d=∣θμ(GK0)∣. For g fixing K0, the phase of g(Rμ) is ω(g)n, of order dividing d/gcd⁡(d,n). Every other g is hc for h fixing K0 and complex conjugation c, and c fixes the distinguished positive real value. Choose h with θμ(h) of order d. Its image under h, and also its inverse, have order d. The phase of h(Rμ) thus has order exactly d/gcd⁡(d,n). This proves that d/gcd⁡(d,n) is the least power giving a positive real number at every embedding. That power belongs to B by the field calculation. The relative extension degree over B uses ∣θμ(GB)∣, which may be smaller. Transfer and field of the Artin period · All conjugates and their phases

The valuation at the prime selecting the twist

Fix a coefficient place w over ℓ. Off the active set its tuple is the reference and every ratio is one. On the active set its character is Δψw. The deformation Δ is the first deformation character except when ℓ equals its ramification prime, in which case it is the second. It is unramified above ℓ. The moving character ψw is trivial on the full local group EP× for every P∣ℓ, including the uniformizer. Restriction to K0,p× multiplies the local characters over P∣p. Thus at every p∣ℓ the class field of the restricted seed splits completely and the fields Hμ(w)(a) are unramified for every positive power a. These statements persist over any completion Bb at b∣ℓ. For vℓ(ℓ)=1 and eb=e(b/ℓ), every individual ratio has valuation in eb−1Z. Dividing it by the corresponding power of a uniformizer of Bb gives a unit whose residue transforms by the prime-to-ℓ part of the restricted deformation character, up to the same inversion. Selection and switch · Triviality on the local group · The restricted class field at the selected prime

At an even column put ν=νij,μ0, R=LE(s∗,νΔψw)/LE(s∗,ν) and Rd=LE(s∗,νΔ)/LE(s∗,ν). The quotient Z=R/Rd has local multiplier that of ψw, so Z∈Bb× at the compatible embedding. Its primitive numerator and denominator have identical Euler factors above ℓ. In the toric evaluation the denominator uses the deformation at the base order, and the numerator uses the enlarged order at which the moving primes are ramified. This gives the primitive Z in the square identity. Euler factors at the orders used in the comparison · The square at the selected place

The fitted theta estimate in the release has the following numerical consequence. Its local comparison gives Z Sλ=uTλ(a), with u a unit, Sλ the base sum of theta coordinates, and Tλ(a) the twisted sum for an exponent prime to ℓ. The estimate for all these powers gives vℓ(Tλ(a))≤vℓ(Sλ)+1/ℓ2. Taking valuations in the identity and using Z∈Bb× gives vℓ(R)−vℓ(Rd)≤⌊eb/ℓ2⌋eb. In particular the error is at most zero when eb<ℓ2. The bound on the smaller field's root discriminant allows ramification at the deformation primes, so this last criterion is local to the specified base. Fitted theta hypothesis · Passage to the valuation of the ratio

The lower estimate uses the reference value LE(s∗,ν). Let β0 be its twelfth power with the prescribed CM-period normalization, P0 its monic minimal polynomial over the row field, ew=ord⁡w(ℓ), and Dℓ the loss per twelfth power in the toric denominator theorem. That theorem controls the normalized twisted value βt=R12β0 at every place above ℓ. Summing the lower bound vℓ(βt)≥−12Dℓ over conjugates fixing the row field gives vℓ(Z) ≥ −Dℓ−ord⁡w(P0(0))12ewdeg⁡P0−vℓ(Rd). Here R is in the row field. The local degree, conductor and partial-zeta denominators enter Dℓ through integrality of the toric lattice values. Rounding this lower endpoint up in eb−1Z is valid. These two estimates identify the arithmetic needed to bound the scalar at b. Reciprocity fixes its eigenline, and multiplication by an element of Bb× preserves that eigenline. Denominator theorem · Reference normalization and its valuation · Both valuation bounds at the surviving integer

One character for all the theta evaluations

The estimate for Tλ(a) above comes from choosing the order of the twist after the finite set of theta, row and Selmer tests is known. The numerical argument permits every exponent a prime to ℓ at the chosen order. Write π for a uniformizer in the coefficient ring of a test and e=ord⁡π(ℓ). If h(X)=∑hjXj is integral with one known nonzero coefficient hw, put dj=ord⁡π(hj) for nonzero coefficients. At a primitive ℓk-th root ζ, the valuations of the terms of h(ζ−1) are dj/e+j/φ(ℓk). Indeed (1−ζa)/(1−ζ) is a unit for every coprime a, and the product of 1−ζa over the primitive roots is ℓ. At most w+dw positive depths k have two minimizing indices. At such a depth choose minimizers ik<jk. Since 1/φ(ℓk) strictly decreases with k, the two minimizing inequalities give jk≤ik′ when k<k′. The depths with ik<w therefore number at most w. For the others jk>w, hence djk<dw by comparison with the w-th term. These nonnegative orders strictly decrease through the remaining exceptional depths. At every other depth the unique minimum is the exact valuation for every primitive root. Exceptional depths · Valuation for all primitive roots

To apply this count, divide the split auxiliary prime pairs into left and right lists with their prescribed Frobenius coordinates. Clear the toric coefficients by ℓN0. The group ring, written in variables 1+Xq, gives a polynomial whose specialization is the cleared theta sum. Dropping the right list with its character set equal to one gives the polynomial on the left list multiplied by the pair Euler corrections. The first nonzero residual term of each correction has degree zero or two in its own left variable. The constant term of the left-list polynomial has order e(N0+vℓ(Sλ)), since the corrections for deleting its pairs at the trivial character are units. Substitute Xq=(1+T)ℓeq−1 with fixed widely spaced exponents eq. Reduction gives distinct powers of T for the designated lowest terms of the Selmer polynomials. The right exponents can be taken large enough that the group relations contribute zero modulo πb+1 through the required degree, with b the order bound on the left constant. If the reduced Euler product first has a unit in degree d, its product with a polynomial of constant order ≤b has a coefficient of order ≤b by degree (b+1)d. Otherwise reducing the product first modulo π and then successively modulo πb+1 forces the constant to be divisible by πb+1. These observations give one bounded-degree, bounded-order coefficient for each test. The bounds hold at arbitrarily high common auxiliary levels using the toric clearing, pair deletion, row determinant and auxiliary-prime hypotheses on the same data. Low-coefficient implication · Simultaneous substitutions

Sum w+dw over the finite test list, using the uniform bounds for the chosen coefficients. Choose the auxiliary level large enough to contain a consecutive interval of depths longer than this sum, with w/φ(ℓk) below the allocated error for every test. One depth then gives unique minima for them all. Its character has the prescribed nontrivial inertia on each chosen pair and is trivial above ℓ. For a toric label divide the selected value by ℓN0. Its normalized valuation lies between −N0 and vℓ(Sλ)+ℓ−2 and is the same for every coprime power at that label. The second pass identifies the new character with a coprime power of the first, with the same test coefficients and a possible inversion. This is why the Fourier estimate applies to the character in the local ratio Z. Common depth and error · All-powers comparison · Full proof of the Fourier bound for the selected seed

The provisional and final kits on the same host

Choose the constants before the cutoff varies. For fixed sufficiently small ε and the host constants, choose the lower bounds for the standard record C, including the local projection thresholds. Choose cD above its bound and ch≥402 above the bounds that then apply. Take one onset satisfying the host, construction, weight and code requirements. These thresholds are uniform in x, the eligible hosts, and the shifts and words at each cutoff. In particular the word bound and the local projection estimates do not call for a separate onset for each word or prime. Recursive quantifiers · Uniform projection thresholds

Use a pair in the nonempty integer windows and one code to build a provisional kit at each supported cutoff. Each kit consists of a point for every shift 0≤u0<n. Apply the weight result in its family form, WClosed, to a member at that cutoff. Provenance identifies its scale, rank, TT, prime classes and Frobenius coordinates with those of the originating host. The point and host signals agree with the same numerical parameters and orbit representatives. Hence the recovered pair satisfies both weight requirements on that same host. Reapply object existence with this pair, the same host and code, once for every shift. The conditional arithmetic sampling proof above already takes place on the host, and the provisional-kit route is how the release carries it through to the final objects. Both passes of the automorphic construction take place within each application of object existence. Transport from the provisional point · Provisional and final choices · Final kit

Proved here (one family). Store the final kit at every x∈S above one threshold x1, with its objects and sizes as F.D(x,i) and F.L(x,i). For formal indexing the release chooses one fixed supported cutoff y≥x1 as a fallback. It uses the fallback at x≤x1, and the index subtype is empty for x>x1 outside S. Every supported x>x1 uses its own kit. Within a kit, Pt.From and TieSame preserve the ranks, q,d,QS, the Artin data, the torus and auxiliary and deformation primes, TT,T2, the weights and code. Consequently every member has Gψ at its actual scale. The design theorem requests a standard record large enough for the prescribed lower bounds, takes the common onset with the numerical requirements, carries the thick set, tame, QD, TT and word tests into the design, and sets CTu=cT(ε,9) with the lower bounds adjusted for that assignment. Its cutoff set is still S. Indexing and same-cutoff data Actual scale and retained cutoff set Design construction

5. Fitting the comparison and consuming the family

The geometric existence route fixes a class of families and a selection on that class. Release input (representation and geometric rows). The input from local Langlands, the unitary group and theta correspondence includes the two theta-integral identities, matching local factors, and the definite-side covolume bound. The source obtains a Fourier coefficient fit and an evaluation grid. It takes the larger of the two grid thresholds for the row functions and for the evaluation theorem and uses that grid for both. The linked row bundles and unramified exponent condition give witnesses for the definite-side hypotheses. classSel combines the row-existence statements with the lattice-volume identities and the base and retwisting existence statements to obtain one class and one selection satisfying the row requirements together. Its existence proof is recS, using the host-to-design construction of §4 with the recursive input obtained from these rows. In particular the release calls classSel on a conjunction of the required rows. Choosing a separate family for each row would give a weaker existence assertion. Theta, grid and linked bundles Representation and covolume premises in the final call Common class and selection Simultaneous rows and existence on that class

Release inputs for the comparison, with the divisor and column constructions supplied through the interfaces of Papers II–V. On this class the closed row requirements include the common map on the column lattices, the Bessel pairing identification, the rank and fullness assertions, the twisted-place comparison LinkN, and the remaining norm and irreducibility estimates used in LinkGLOBAL. “Closed” here specifies quantifiers. There is a positive cap for ε, then a required bound on the constants for each ε, then a conclusion for every family of the class satisfying that bound. Eventual pointwise conclusions may have an onset depending on that family. The rank input identified as SmallLevelRankEq equates each selected lift rank dZ,0 to D0. The divisibility input LinkDIVt takes LinkEta and DIVinN and supplies LinkDIV. That output includes Bk(u)(νu)⊆CC for every word u, where νu is its column lattice and CC is the common target lattice, the Bessel comparison by unitary maps and a nonzero scalar, full row lattice, and the stated bounds on the heights and local errors. The common-map theorem and the quantitative divisor, column and norm statements are substantial inputs. Closed quantifiers Common-map definition Bessel, rank and divisibility interfaces Actual comparison inputs

Proved here (using the length budget). At each point, VVle and the identification sha_eq of its record lengths with the localization lengths give finiteness and bound the nonnegative finite partial sums of vvTerm by b. Their supremum therefore bounds the infinite sum by b. Let L=xε, B0=fmbL, and Nc≥0 be the size budget of the point. With the selected lift rank identified by SmallLevelRankEq, LinkN bounds the local sum etaSum by D0[C(fmbL2+(log⁡x)2Nc)+∑ℓvvTerm(ℓ)]. At a sufficiently late cutoff, L≥1, fm≥1, (log⁡x)2≤L and b≤b≤B0. The first term in the bracket is at most ∣C∣L(B0+Nc), and its other term is at most L(B0+Nc). This is DIVinN with constant ∣C∣+1. A maximum of the thresholds for the three inputs and the logarithm estimate makes the deduction uniform over the late members. Thus the localization lengths supply the error allowance in the divisibility bridge. The independent place coverage, column and lattice assertions still enter the bridge. Identification and summability Input inequality and desired bound Pointwise and family deductions

Proved here (order of choices). Intersect the cap from design existence with the caps from the closed assertions. After fixing ε below this positive minimum, join the finitely many required lower bounds on the parameters. Request the single design family at this joined bound. Apply its closed assertions, including the preceding cohomological deduction for divisibility, to this witness F. The output LinkGLOBAL takes ChosenE,LinkG,LinkS,LinkIRR,LinkDIV,LinkMu on F, and returns constants and one onset with ChainTied at every later x∈S. Its proof combines the onsets of its input estimates. A tied chain has Dd.x=x, chooses all its points from Ix, and identifies its arithmetic data, top set, weight windows, and word types with those of F. The actual shape call supplies eD=2,eL=0, independent of ε. This proves the assembly theorem of §1 relative to its stated inputs. Closed join and same-family application Actual exponents Global premises and onset Common size threshold Chain ties

Supplied by Papers II–V (scalar comparison). The scalar comparison uses the chain and the arithmetic condition at the same member i and the same TT. It yields, after enlarging the threshold to cover the chain, arithmetic and scalar onsets, Gz(x)=∑n≤xΛvM(n)(x/n−1)−xlog⁡x+(1+γE)x ≤ x1/2+9ε(x∈S, x sufficiently large). Here Level.cyc(1) keeps only the trivial character, so its correction sum over the other kept characters is empty. Given 1/2<θ<1, choose ε below the common cap with 9ε≤θ−1/2 and then take the family. Proved for ordinary zeta in Paper I, Theorem 1. The released window criterion uses the resulting bound by xθ on arbitrarily late [Z,Z3/2]. Its analytic premise, FormB, turns this into zero-freeness to the right of θ for the kept characters in the release's pole convention. Ordinary zeta is the conductor-one case. The classical zeta functional equation and the given nonvanishing statements give the asserted critical-line conclusion. The full theorem takes this step for general kept characters and then passes through the imprimitive extension. Conductor-one kept set Scalar input uses the same member Window estimate and analytic quantifiers Analytic window premise Final pass to all characters Ordinary critical-line reduction

The favorable-prime refinement of §3.3 can also give a real tail of cutoffs, with the other inputs of this construction. Proved here from that refinement. Given a sufficiently large Z, use X=Z2 in the conductor-one selection and take W=Z. Since c=k=1, §2 supplies [Z,Z3/2], including its left endpoint, so every sufficiently large x belongs to S. The releveling, provisional-family and final-family constructions preserve that set. The same-family comparison therefore holds on a full tail and the ordinary all-cutoff Landau consumer applies. The proof of the exported theorem already uses the weaker window assertion. This refinement still needs the pointwise host cohomological estimates and the common scalar, divisor and column inputs. Preserved cutoff set in design All-cutoff analytic interface Prior tail trace

For the favorable ordinary pair, §3.3 supplies the dyadic term on the entire prescribed interval. The general pair-independent route in §3.1 still uses the exceptional-twist H2 estimate. The classical proof in §3.4 supplies the odd minus term for every host in this specialization. In the odd plus term the remaining steps include the ordinary model and special-fibre hypotheses of the Count I cotangent comparison in §3.6, the rank-three attachments in I and V, and the rank-four automorphic construction in II. The weight selection and signed Hermitian spaces in §4 are on the same arithmetic host. The internal Gram and integral block-lattice estimates in §4, the occurrence and multiplicities of the selected representations, the saturated automorphic seed with the prescribed types, and the theta comparisons in §§4–5 still enter the geometric construction. Cohomological assembly Existence and comparison assembly