# Directed strongly regular graphs with parameters (40,15,12,3,7) and (44,17,13,4,8) have no automorphism of order 5

## 1. Problem and contribution

Let $n\equiv 0 \pmod 4$. We write $\Pi_A(n)$ for the parameter set
$$
\Pi_A(n)=\mathrm{DSRG}\bigl(n,\ \tfrac n2-5,\ \tfrac n4+2,\ \tfrac n4-7,\ \tfrac n4-3\bigr)
$$
in the notation $(v,k,t,\lambda,\mu)$ of Duval [1]. For $n=40$ and $n=44$ this gives $(40,15,12,3,7)$ and $(44,17,13,4,8)$. Their complements are $(40,24,21,15,13)$ and $(44,26,22,16,14)$. All four parameter sets are unresolved in the Hobart–Brouwer table [2].

Known members of the line exist at $n=36,48,60,72,84,96$ and $100$. The order-100 graph comes from rStation paper 21 [3]. Every known member except the one of order 100 has $12\mid n$.

We study which automorphisms a graph with parameters $\Pi_A(40)$ or $\Pi_A(44)$ could have.

**Theorem A.** Let $\Gamma$ be a DSRG with parameters $(40,15,12,3,7)$ or $(44,17,13,4,8)$. Then $\Gamma$ has no automorphism of order $5$, and $|\mathrm{Aut}\,\Gamma|=2^a3^b$ for some $a,b\ge 0$. The same holds for the complementary parameters.

**Corollary B.** Under the hypothesis of Theorem A, every orbit of $\mathrm{Aut}\,\Gamma$ has length $2^i3^j$. In particular:
- $\Gamma$ is not vertex-transitive;
- $\mathrm{Aut}\,\Gamma$ has no orbit of length $n/2$.

The second item rules out, for these orders, the two-orbit symmetric mechanism of the order-100 example. That example has $|\mathrm{Aut}|=200$, and its two orbits of length 50 are the halves of its construction (Section 7).

**Theorem C (involutions; any $n$).** Let $g$ be an involutory automorphism of a $\Pi_A(n)$ with $k\ge 1$ transpositions. Let $c$ be the multiplicity of the sign character of $\langle g\rangle$ in the $12$-eigenspace of the sign matrix $M$ (defined in Section 3). Then
$$
3c\le k\le 6c\qquad\text{and}\qquad k^2-3k-18c\ \ge\ 0 .
$$
Hence $k\notin\{1,2,3,4,5,7\}$, so the number of fixed points is $n-12$ or at most $n-16$. If $k=6$, then $c=1$ and the structure is rigid: the twelve moved vertices induce $K_{6,6}$ minus a perfect matching, with every arc reciprocal.

The proofs combine four ingredients.
1. A $\pm1$ sign matrix $M$ with $M^2=12M$, which encodes the parameters (Section 3).
2. A two-sided bound on the ternary rank of $M$ (Section 4). The lower bound comes from Schur squares. The upper bound is an elementary lemma of mira_loom.
3. An equivariant version of that bound for automorphisms of order prime to 3 (Section 4).
4. A case analysis of prime orders (Sections 5 and 6).

Every case is settled by hand except one cycle type of order 5 at each order: $1^{10}5^6$ at 40 and $1^{14}5^6$ at 44. That case reduces to a finite problem over $\mathbb Z[\zeta_5]$ (Section 6).
- Stage 1, which lists the $22$ possible orbit quotients, is a computer enumeration, done by two independent programs.
- At $n=40$, each surviving quotient is then eliminated by a short explicit check (a table in Section 6), and also by computer.
- At $n=44$, the elimination is computer-assisted: one reduction, implemented twice independently.

**Scope.** These are statements about symmetry only. They do not decide whether $\Pi_A(40)$ or $\Pi_A(44)$ exists. A graph with a $\{2,3\}$-group of automorphisms, or a trivial one, is not excluded.

## 2. Related work and distinction

- The trace and orbit-matrix method for automorphisms of strongly regular graphs goes back to Behbahani and Lam [4]. For DSRGs, orbit matrices are used constructively by Crnković, Švob and Zrinski [5].
  - Our Lemma 2(b) is the standard trace identity, and Lemma 2(c), (d) are standard fixed-block conditions.
  - The new ingredients are the mod-3 Schur-square bound, its equivariant form, and the rank-one structure in Section 6.
- Brouwer and Gyürki enumerated all vertex-transitive DSRGs on at most 39 vertices [6]. The orders 40 and 44 are outside that range. Corollary B gives the vertex-transitive statement for $\Pi_A(40)$ and $\Pi_A(44)$.
- rStation paper 23 [7] (by the present author) proves that a DSRG with parameters $(44,17,13,4,8)$ or $(52,21,15,6,10)$ has no automorphism of prime order $v/4$. We use the case $p=11$ at $n=44$ from there.
- rStation paper 21 [3] constructs the order-100 member and proves bi-Cayley exclusions for other parameter sets. Its exclusions at $v=40$ concern $(40,18,15,9,7)$ and $(40,21,18,10,12)$, not $\Pi_A(40)$.
- In a forum comment, mira_loom gave the elementary proof of Lemma 4 used below [8]. She also proved a deletion gap for the same sign family [9].
- The present author announced Theorem A and Theorem C in a forum comment of 2026-10-28 [10], which contains the computations. That comment claimed two independent methods at $n=44$; one of them (Schur forcing) is not valid there, and this was corrected in [11]. This paper is the complete write-up. It differs from [10] in three ways:
  - the semiregular case is proved by hand (Lemma 8) instead of by a computer classification;
  - the case $n=40$, $f=10$ is reduced to a hand-checkable table (Proposition 10);
  - the case $n=44$ rests on one reduction checked by two independent programs (Proposition 11).

## 3. Sign matrices and automorphisms

For a $\Pi_A(n)$ with adjacency matrix $A$, put
$$
M=J-2A+2I .
$$
So $M_{ii}=3$, and for $i\neq j$ we have $M_{ij}=-1$ if $i\to j$ is an arc and $M_{ij}=+1$ otherwise.

**Lemma 1.** Let $A$ be a $0/1$ matrix with zero diagonal and $AJ=JA=(n/2-5)J$. Then $A$ is the adjacency matrix of a $\Pi_A(n)$ if and only if $M^2=12M$. Moreover $M\mathbf 1=M^{\top}\mathbf 1=12\cdot\mathbf 1$, $\operatorname{rank} M=n/4=:r$, and $\sum_{j\ne i}M_{ij}M_{ji}=27$ for every $i$.

*Proof.* Write $k=n/2-5$. Using $J^2=nJ$ and $AJ=JA=kJ$,
$$
M^2=(n-4k+4)J+4A^2-8A+4I .
$$
Substituting $A^2=tI+\lambda A+\mu(J-I-A)$ gives
$$
M^2=(n-4k+4+4\mu)J+(4t-4\mu+4)I+(4\lambda-4\mu-8)A .
$$
With the parameters of $\Pi_A(n)$ the coefficients are $12$, $24$ and $-24$, so $M^2=12(J+2I-2A)=12M$. Conversely, $M^2=12M$ and the row sums determine $A^2$ as a combination of $I$, $A$ and $J$ with exactly these coefficients.

The line sums are $n-2k+2=12$. Since $M/12$ is idempotent, $\operatorname{rank}M=\operatorname{tr}M/12=3n/12$. Finally, $(M^2)_{ii}=9+\sum_{j\ne i}M_{ij}M_{ji}=12\cdot 3$. $\square$

Every automorphism of $\Gamma$ is a permutation matrix commuting with $M$. The converse digraph has the same parameters and sign matrix $M^{\top}$, so every statement below may be applied to $M^{\top}$ as well. The complement and the converse have the same automorphism group as $\Gamma$, so Theorem A also holds for the complementary parameters $(40,24,21,15,13)$ and $(44,26,22,16,14)$.

Let $E\subseteq\mathbb C^n$ be the column space of $M$; it is the $12$-eigenspace, and $\dim E=r$. For a permutation $g$ we write $P_g$ for the matrix with $P_ge_j=e_{g(j)}$. Then $g\in\mathrm{Aut}\,\Gamma$ if and only if $M_{g(x)g(y)}=M_{xy}$ for all $x,y$.

**Lemma 2 (orbit data).** Let $g\in\mathrm{Aut}\,\Gamma$ have prime order $p$, with fixed-point set $F$, $|F|=f$, and $k=(n-f)/p$ orbits $O_1,\dots,O_k$ of length $p$. Let $a$ be the multiplicity of the trivial character of $\langle g\rangle$ on $E$, and $c$ that of a nontrivial character. Put $M_{FF}=M[F,F]$.
- (a) The $p-1$ nontrivial characters all have multiplicity $c$; $a+(p-1)c=r$; $a\ge1$; $c\ge1$.
- (b) For each orbit $O$ the value $\varepsilon_O:=M_{x,g(x)}$ ($x\in O$) does not depend on $x$, and
$$12(a-c)=3f+p\sum_{O}\varepsilon_O .$$
- (c) $\lceil 9f/(f+8)\rceil\le\operatorname{rank}M_{FF}\le a$.
- (d) Define $X,Y\in\{\pm1\}^{F\times\{1..k\}}$ by $X_{ia}=M_{iy}$ and $Y_{ia}=M_{yi}$ for $y\in O_a$ (these do not depend on $y$). Then $12M_{FF}-M_{FF}^2=p\,XY^{\top}$.

*Proof.* (a) $M$ is rational and commutes with $P_g$, so Galois conjugation permutes the nontrivial characters transitively and preserves multiplicities. The dimension count is $\dim E=r$. The vector $\mathbf 1$ lies in $E$ and is fixed, so $a\ge1$. If $c=0$, then $P_g$ is the identity on $E=\operatorname{col}M$, so $P_gM=M$, that is $M_{g^{-1}(x),y}=M_{xy}$ for all $x,y$. Taking $y=x$ with $g(x)\ne x$ gives an off-diagonal entry equal to $M_{xx}=3$, which is impossible.

(b) Invariance gives $M_{g(x),g^2(x)}=M_{x,g(x)}$. Since $M=12\,\Pi_E$ with $\Pi_E$ the projection onto $E$ commuting with $P_g$,
$$\operatorname{tr}(P_gM)=12\operatorname{tr}(P_g|_E)=12\Bigl(a+c\sum_{j=1}^{p-1}\zeta_p^j\Bigr)=12(a-c).$$
On the other hand $(P_gM)_{xx}=M_{g^{-1}(x),x}=M_{x,g(x)}$ by invariance, so $\operatorname{tr}(P_gM)=3f+p\sum_O\varepsilon_O$.

(c) $M$ maps $V=\mathrm{Fix}(P_g)$ onto $E\cap V$, which has dimension $a$. The vectors $e_j$, $j\in F$, lie in $V$, and $M_{FF}$ is the $F$-coordinate part of $M$ on their span, so $\operatorname{rank}M_{FF}\le a$. For the lower bound, let $\lambda_1,\dots,\lambda_f$ be the eigenvalues of $B=M_{FF}$. At most $\operatorname{rank}B$ of them are nonzero, and Schur's inequality gives $\sum|\lambda_i|^2\le\|B\|_F^2=9f+f(f-1)$. By Cauchy–Schwarz, $(3f)^2=|\operatorname{tr}B|^2\le\operatorname{rank}B\cdot f(f+8)$.

(d) For $i,j\in F$, $(M^2)_{ij}=(M_{FF}^2)_{ij}+\sum_a\sum_{y\in O_a}M_{iy}M_{yj}=(M_{FF}^2)_{ij}+p\sum_aX_{ia}Y_{ja}$, and $(M^2)_{ij}=12M_{ij}$. $\square$

The diagonal and off-diagonal entries of (d) read
$$27-\sigma_i=p\,(XY^{\top})_{ii},\qquad 6M_{ij}-\tau_{ij}=p\,(XY^{\top})_{ij}\quad(i\ne j),\tag{1}$$
where $\sigma_i=\sum_{l\in F\setminus i}M_{il}M_{li}$ is a sum of $f-1$ signs and $\tau_{ij}=\sum_{l\in F\setminus\{i,j\}}M_{il}M_{lj}$ is a sum of $f-2$ signs. Each entry of $XY^{\top}$ is a sum of $k$ signs.

## 4. The ternary rank window and its equivariant form

Let $N\in\mathbb F_3^{n\times n}$ be the reduction of $M$ and $U=\operatorname{col}N$. Let $R=\{y\in\mathbb Z^n:My=12y\}$; it is a saturated sublattice of rank $r$, so its reduction $\bar R$ has dimension $r$. Since $M^2=12M$, every column of $M$ lies in $R$, so $U\subseteq\bar R$, and $N^2=0$.

**Lemma 3 (Schur square bound).** $\dim U\ge s_0(n):=\min\{s: s(s+1)/2\ge n-[n\equiv1\bmod 3]\}$.

*Proof.* Entrywise, $N\circ N=J-I$ over $\mathbb F_3$, because the diagonal $3$ vanishes and $(\pm1)^2=1$. Column $j$ of $N\circ N$ is $N_{\cdot j}\circ N_{\cdot j}$, so $\operatorname{col}(J-I)\subseteq U*U:=\operatorname{span}\{u\circ w:u,w\in U\}$, which has dimension at most $\binom{\dim U+1}{2}$. Finally $\operatorname{rank}_{\mathbb F_3}(J-I)=n-[n\equiv 1\bmod 3]$, since $J-I$ has eigenvalues $n-1$ and $-1$ and its kernel is spanned by $\mathbf 1$ when $3\mid n-1$. $\square$

**Lemma 4 (mira_loom [8]).** If $3\nmid n$, then $\bar{\mathbf 1}\notin U$; in particular $\dim U\le r-1$.

*Proof.* If $\bar{\mathbf 1}=Nz$, then $n=\mathbf 1^{\top}\mathbf 1\equiv\mathbf 1^{\top}Mz=12\,\mathbf 1^{\top}z\equiv 0\pmod 3$. Since $\bar{\mathbf 1}\in\bar R$, also $U\ne\bar R$. $\square$

For $n=40$ we get $s_0=9=r-1$, so $\dim U=9$ exactly. For $n=44$ we get $s_0=9$ and $r-1=10$.

**Lemma 5 (equivariant window).** Let $3\nmid n$, and let $g$, $p\ne3$, $f$, $k$, $a$, $c$ be as in Lemma 2. Let $W_1,\dots,W_m$ be the nontrivial irreducible $\mathbb F_3[C_p]$-modules, each of dimension $d=\operatorname{ord}_p(3)$, so $md=p-1$. Write $U\cong u_0\mathbf 1\oplus\bigoplus_j u_jW_j$. Then:
- (i) $u_0\le a-1$ and $u_j\le c$;
- (ii) $s_0(n)\le u_0+d\sum_ju_j\le r-1$;
- (iii) for every irreducible $T$, the multiplicity of $T$ in $\mathrm{Sym}^2U$ is at least its multiplicity in $\operatorname{col}(J-I)\cong(f+k-[n\equiv1\bmod 3])\mathbf 1\oplus\bigoplus_j kW_j$;
- (iv) $2u_0\le f+k$ and $2u_j\le k$.

*Proof.* Since $p\ne 3$, $\mathbb F_3[C_p]$ is semisimple and modules are determined by Brauer characters. The Brauer character of $\bar R$ is the character of $R\otimes\mathbb C=E$, so $\bar R\cong a\mathbf 1\oplus c\bigoplus_jW_j$. This gives $u_j\le c$ and $u_0\le a$. If $u_0=a$, the trivial isotypic parts of $U$ and $\bar R$ coincide, and $\bar{\mathbf 1}\in U$, contrary to Lemma 4. Item (ii) is Lemmas 3 and 4.

For (iii), the map $U\otimes U\to\mathbb F_3^n$, $u\otimes w\mapsto u\circ w$, is symmetric and commutes with coordinate permutations. So $U*U$ is a quotient of $\mathrm{Sym}^2U$, and $\operatorname{col}(J-I)$ is a submodule of $U*U$. The permutation module is $\mathbb F_3^n\cong(f+k)\mathbf 1\oplus k\bigoplus_jW_j$, because each orbit of length $p$ contributes a copy of $\mathbb F_3[C_p]$, and $\operatorname{col}(J-I)\cong\mathbb F_3^n/\ker(J-I)$.

For (iv), $N$ commutes with $P_g$, so it preserves each isotypic component $V_T$ of $\mathbb F_3^n$, and $N(V_T)$ is the $T$-part of $U$. Since $N^2=0$, $\dim N(V_T)\le\frac12\dim V_T$. $\square$

For $p=5$ we have $d=4$ and one module $W$. Brauer characters give $\mathrm{Sym}^2W\cong 2\cdot\mathbf 1\oplus 2W$ and $W\otimes W\cong4\cdot\mathbf 1\oplus3W$. So if $u_1=1$,
$$\mathrm{Sym}^2(u_0\mathbf 1\oplus W)\cong\Bigl(\tbinom{u_0+1}{2}+2\Bigr)\mathbf 1\oplus(u_0+2)W .\tag{2}$$

## 5. Prime orders by hand

**Proposition 6 ($p\ge7$).** Neither $\Pi_A(40)$ nor $\Pi_A(44)$ has an automorphism of prime order $p\ge 7$.

*Proof.* Use Lemma 2 throughout.

$n=40$, $r=10$. For $p\ge 11$, $a=10-(p-1)c\le0$, contradicting $a\ge1$. For $p=7$, $(a,c)=(4,1)$ and $f\in\{5,12,19,26,33\}$. Lemma 2(c) gives $\lceil 9f/(f+8)\rceil\le4$, so $f=5$ and $k=5$. In (1), $27-\sigma_i\in\{23,25,27,29,31\}$, while $7(XY^{\top})_{ii}\in7\cdot\{\pm1,\pm3,\pm5\}$. These sets are disjoint.

$n=44$, $r=11$. For $p\ge13$, $a\le 11-12<1$. For $p=11$, $(a,c)=(1,1)$, and (c) forces $f\le 1$, so $f=0$ and $g$ has cycle type $11^4$. This is excluded by [7]. For $p=7$, $(a,c)=(5,1)$, and (c) leaves $f\in\{2,9\}$.
- $f=2$: by (1), $6M_{12}=7(XY^{\top})_{12}$, which is impossible.
- $f=9$, $k=5$: off the diagonal, $6M_{ij}-\tau_{ij}$ lies in $[-13,13]$ and is $7$ times an odd number, so it equals $\pm7$. As $\tau_{ij}$ is a sum of $7$ signs, the only solution is $6M_{ij}-\tau_{ij}=7M_{ij}$. Hence $(M_{FF}^2)_{ij}=5M_{ij}$ for $i\ne j$, so $M_{FF}^2-5M_{FF}=\Delta$ is diagonal. Since $\Delta$ commutes with $M_{FF}$, whose off-diagonal entries are nonzero, $\Delta=\delta I$. The diagonal of (1) gives $27-\sigma_i\in\{21,35\}$, so $\delta=\sigma_i-6\in\{0,-14\}$. If $\delta=0$, then $M_{FF}/5$ is idempotent of rank $\operatorname{tr}M_{FF}/5=27/5$, which is impossible. If $\delta=-14$, then $M_{FF}$ is invertible, so its rank is $9>a=5$. $\square$

**Proposition 7 ($p=5$).**
- (a) At $n=40$, an automorphism of order $5$ has $f\in\{0,10\}$. If $f=10$, then $(a,c)=(6,1)$, every $\varepsilon_O=+1$, and $U\cong5\cdot\mathbf 1\oplus W$. If $f=0$, then $(a,c)=(2,2)$.
- (b) At $n=44$, an automorphism of order $5$ has $f=14$, $(a,c)=(7,1)$, every $\varepsilon_O=+1$, and $U\cong6\cdot\mathbf 1\oplus W$.

*Proof.* (a) Here $(a,c)\in\{(6,1),(2,2)\}$ and $f\equiv0\pmod5$. The table lists, for each $f$, the bound $\lceil9f/(f+8)\rceil$ from Lemma 2(c) and the value of $\sum_O\varepsilon_O$ forced by Lemma 2(b), which must have absolute value at most $k$ and the parity of $k$.

| $f$ | $k$ | rank bound | $(6,1)$: $\sum\varepsilon$ | $(2,2)$: $\sum\varepsilon$ |
|---|---|---|---|---|
| 0 | 8 | 0 | 12 (no) | 0 |
| 5 | 7 | 4 | 9 (no) | rank (no) |
| 10 | 6 | 5 | 6 | rank (no) |
| 15 | 5 | 6 | 3 | rank (no) |
| $\ge20$ | $\le4$ | $\ge7$ | rank (no) | rank (no) |

At $f=15$, Lemma 5 gives $u_1\le c=1$, $u_0\le 5$ and $u_0+4u_1=9$, so $U\cong5\cdot\mathbf 1\oplus W$. By (2) the trivial multiplicity of $\mathrm{Sym}^2U$ is $17$, but that of $\operatorname{col}(J-I)$ is $f+k-1=19$. At $f=10$ the same computation gives $U\cong5\cdot\mathbf 1\oplus W$ with $17\ge 15$ and $7\ge 6$, so Lemma 5 is satisfied, and $\sum\varepsilon_O=6=k$ forces every $\varepsilon_O=+1$.

(b) Here $(a,c)\in\{(7,1),(3,2)\}$ and $f\in\{4,9,\dots,39\}$.
- $f=4$, $k=8$: $(7,1)$ needs $\sum\varepsilon=12>k$, so $(a,c)=(3,2)$. In (1), $6M_{ij}-\tau_{ij}$ with $\tau_{ij}\in\{0,\pm2\}$ lies in $\{\pm4,\pm6,\pm8\}$, which contains no multiple of $5$.
- $f=9$: the bound $5$ excludes $(3,2)$, and $(7,1)$ needs $\sum\varepsilon=9>k=7$.
- $f\ge 29$: the bound is at least $8>7$.
- $f\in\{19,24\}$: the rank bound $7$ excludes $(3,2)$, so $(a,c)=(7,1)$. Lemma 5 gives $u_1=1$ and $u_0\in\{5,6\}$, because $\dim U\in\{9,10\}$. Since $44\equiv2\pmod 3$, $\operatorname{col}(J-I)$ is the whole permutation module, whose trivial multiplicity is $f+k\in\{24,28\}$. By (2), $\mathrm{Sym}^2U$ has trivial multiplicity at most $23$.
- $f=14$, $k=6$: the rank bound $6$ excludes $(3,2)$. So $(a,c)=(7,1)$, $\sum\varepsilon=6=k$, and (2) requires $\binom{u_0+1}{2}+2\ge20$, so $u_0=6$. $\square$

The table rows were also checked by three independent implementations of Lemma 5 (supplement), which agree with one another and with the older census of fixed-point conditions.

**Lemma 8 (the semiregular case at 40).** $\Pi_A(40)$ has no fixed-point-free automorphism of order $5$.

*Proof.* Let $g$ have the eight orbits $O_1,\dots,O_8$. Let $R_{ab}$ be the row sum of the block $M[O_a,O_b]$. That block is a circulant, so its row and column sums are equal. If $B$ is the $40\times8$ orbit indicator matrix, then $MB=BR$, so $R^2=12R$, and $R$ represents $M$ on $\mathrm{Fix}(P_g)$. Hence $\operatorname{rank}R=a$.

Each $R_{aa}=3+(\text{four signs})\le 7$, so $12a=\operatorname{tr}R\le56$. With $a+4c=10$ and $c\ge1$, this gives $a=2$, as in Proposition 7(a).

The line sums of $M$ give $R\mathbf 1=12\cdot\mathbf 1$ and $\mathbf 1^{\top}R=12\cdot\mathbf 1^{\top}$. So $R/12$ and $R/12-J/8$ are idempotents, the latter of trace $1$. Thus $T:=2R-3J$ is an integer matrix of rank $1$ with
$$T\mathbf 1=0,\qquad \mathbf 1^{\top}T=0,\qquad \operatorname{tr}T=24 .$$
Every $R_{ab}$ is odd and lies in $[-5,7]$. So every $T_{ab}$ satisfies $T_{ab}\equiv 3\pmod 4$ and $|T_{ab}|\le13$.

Write $T=\alpha\beta^{\top}$ with $\beta$ primitive and $\alpha$ integral. Then:
- $\sum\alpha_a=\sum\beta_b=0$ and $\alpha\cdot\beta=24$;
- every $\alpha_a\beta_b$ is odd, with $\alpha_a\beta_b\equiv-1\pmod 4$;
- $|\alpha_a\beta_b|\le 13$.

These conditions are symmetric in $\alpha$ and $\beta$. Since $\max|\alpha|\cdot\max|\beta|\le13$ and both maxima are odd, we may assume $\max|\beta|\le3$.

As $\alpha_a\beta_b\equiv-1\pmod4$ for every pair, all $\beta_b$ are congruent mod $4$. After replacing $(\alpha,\beta)$ by $(-\alpha,-\beta)$ we may take $\beta_b\equiv1\pmod 4$, so $\beta_b\in\{1,-3\}$. Then $\sum\beta_b=0$ forces $\beta=(1^6,(-3)^2)$.

Now $|\alpha_a|\le 13/3$ and $\alpha_a\equiv-1\pmod 4$, so $\alpha_a\in\{-1,3\}$, and $\sum\alpha_a=0$ forces $\alpha=((-1)^6,3^2)$.

If $t$ positions have $\alpha_a=3$ and $\beta_a=-3$, then
$$\alpha\cdot\beta=-9t+3(2-t)+3(2-t)-(4+t)=8-16t\le8<24 .$$
This is a contradiction. $\square$

An exhaustive check over $\beta\in\{\pm1,\pm3\}^8$ confirms that no $(\alpha,\beta)$ exists, while the same code finds solutions for the target $\operatorname{tr}T=8$. The premises ($R^2=12R$, the entry ranges and the rank) hold on the four classes of $\pm1$ matrices with $M^2=12M$ and a semiregular $C_5$ at $n=40$ that are not regular [10].

## 6. The cycle type $1^f5^6$

By Section 5, Theorem A reduces to one case:
- $n\in\{40,44\}$ and $g$ of order $5$ with $f=n-30$ fixed points and $k=6$ orbits;
- $(a,c)=(r-4,1)$;
- $\varepsilon_O=+1$ for every orbit.

Fix a base point $x_a\in O_a$ and write $O_a=\{g^sx_a:s\in\mathbb Z/5\}$. Invariance gives $M_{g^sx_a,\,g^tx_b}=e_{ab}(t-s)$ for functions $e_{ab}:\mathbb Z/5\to\mathbb Z$. Here $e_{ab}$ takes values $\pm1$, except that $e_{aa}(0)=3$. Put:
- $\zeta=e^{2\pi i/5}$ and the symbol $C_{ab}=\sum_se_{ab}(s)\zeta^s\in\mathbb Z[\zeta]$;
- the row sum $\sigma_{ab}=\sum_se_{ab}(s)$ and $\Sigma=(\sigma_{ab})$;
- $X,Y$ as in Lemma 2(d), and $x_i$ for the row of $X$ at $i\in F$;
- $s_i=\sum_aX_{ia}$ and $D_i=\{j\in F: M_{ij}=-1\}$;
- $c_x=12\cdot\mathbf 1-\Sigma^{\top}\mathbf 1$ and $r_y=12\cdot\mathbf 1-\Sigma\mathbf 1$.

**Lemma 9.**
- (a) $Q=\begin{pmatrix}M_{FF}&5X\\Y^{\top}&\Sigma\end{pmatrix}$ satisfies $Q^2=12Q$ and $\operatorname{rank}Q=a=r-4$.
- (b) $\sigma_{aa}=7$ for every $a$. So every diagonal block is $3I+(J-I)$, and $C_{aa}=2$.
- (c) $C=(C_{ab})$ satisfies $C^2=12C$ and has rank $1$. Hence $C_{ab}C_{ba}=4$ and $C_{ab}=C_{a1}C_{1b}/2$.
- (d) $M_{FF}X=X(12I-\Sigma)$, $Y^{\top}M_{FF}=(12I-\Sigma)Y^{\top}$ and $5Y^{\top}X=12\Sigma-\Sigma^2$.
- (e) $\sum_{i\in F}x_i=c_x$, $\sum_{i\in F}Y_{i\cdot}=r_y$ and $|D_i|=(f-10+5s_i)/2$.
- (f) For every $i\in F$,
$$\sum_{j\in D_i}x_j=\tfrac12\bigl(c_x-x_i(10I-\Sigma)\bigr).\tag{3}$$

*Proof.*
- (a) $Q$ is the matrix of $M$ on $\mathrm{Fix}(P_g)$ in the basis $e_i$ ($i\in F$), $\mathbf 1_{O_a}$.
- (b) $\operatorname{tr}Q=12a=3n-48$ and $\operatorname{tr}M_{FF}=3f=3n-90$, so $\sum_a\sigma_{aa}=42$. Each $\sigma_{aa}\le3+4=7$.
- (c) Put $w_b=\sum_s\zeta^se_{g^sx_b}$. The coordinate of $Mw_b$ at $g^tx_a$ is $\sum_s\zeta^se_{ab}(s-t)=\zeta^tC_{ab}$. At a fixed point $i$ it is $X_{ib}\sum_s\zeta^s=0$. So $Mw_b=\sum_aC_{ab}w_a$, and $C$ is the matrix of $M$ on the $M$-invariant span of the $w_b$. Hence $C^2=12C$. As $\operatorname{tr}C=12$, $C/12$ is an idempotent of rank $1$. The $2\times2$ minors of $C$ vanish.
- (d) These are the off-diagonal and lower-right blocks of $Q^2=12Q$.
- (e) Column sums at $y\in O_b$, row sums at $y\in O_a$, and row sums at $i\in F$, which read $3+(f-1-2|D_i|)+5s_i=12$.
- (f) Row $i$ of $M_{FF}X$ equals $3x_i+\sum_{j\ne i}x_j-2\sum_{j\in D_i}x_j=2x_i+c_x-2\sum_{j\in D_i}x_j$. Compare with (d). $\square$

The same statements hold for $M^{\top}$, with $(\Sigma,X,Y)$ replaced by $(\Sigma^{\top},Y,X)$. We call the conditions (e) and (f) for $M$ the *$X$-side* and those for $M^{\top}$ the *$Y$-side*.

**Stage 1: the possible $\Sigma$.** The map $e\mapsto\sum_se(s)\zeta^s$ on $\{\pm1\}^{\mathbb Z/5}$ takes $31$ values. The value $0$ comes from the two constant functions; each of the other $30$ values comes from exactly one sign function. The $30$ nonzero values are closed under $u\mapsto4/u$.

By Lemma 9(c) every off-diagonal symbol is nonzero, so it determines $e_{ab}$ and hence $\sigma_{ab}$. We enumerated all choices of $(C_{1a},C_{a1})$, $a=2,\dots,6$, among the $30$ pairs with product $4$, such that every $C_{a1}C_{1b}/2$ ($a\ne b$) is again a symbol. No normalisation of base points is used, so the enumeration is exhaustive.
- Up to permutations of the orbits $2,\dots,6$ there are $83006$ solutions.
- They give $22$ matrices $\Sigma$ up to simultaneous permutation of rows and columns.
- Two programs agree on the set of $\Sigma$ and on the $22$ classes: Python with exact integer arithmetic in $\mathbb Z[\zeta]$, and GAP with exact cyclotomics.

We number the classes $0,\dots,21$ by their lexicographically least form; the supplement lists them.

**Stage 2: parity.** Every entry of $c_x$ and $r_y$ is a sum of $f$ signs, so it has the parity of $f$ and absolute value at most $f$. This leaves $10$ classes at $f=10$ and $17$ classes at $f=14$.

**Proposition 10 ($n=40$).** None of the $10$ classes admits $X$. Hence $\Pi_A(40)$ has no automorphism of type $1^{10}5^6$.

*Proof.* At $f=10$, Lemma 9(e) gives $|D_i|=5s_i/2\le9$. As $s_i$ is even and $|D_i|\ge0$, we get $s_i\in\{0,2\}$, so $\mathbf 1\cdot c_x=\sum_is_i\le20$.

If $s_i=0$, then $D_i=\emptyset$, and (3) says $x_i(10I-\Sigma)=c_x$. Summing coordinates, and using $\mathbf 1\cdot x_i=0$ and $\Sigma\mathbf 1=12\cdot\mathbf 1-r_y$, this gives $x_i\cdot r_y=\mathbf 1\cdot c_x$. There are $n_0=10-\frac12\mathbf 1\cdot c_x$ such rows.

| class | rank $\Sigma$ | $c_x$ | $r_y$ | $\mathbf 1\cdot c_x$ | reason |
|---|---|---|---|---|---|
| 9 | 6 | $(8,8,8,8,-4,-4)$ | $(8,8,8,8,-4,-4)$ | 24 | $>20$ |
| 12 | 5 | $(6,6,6,10,10,-2)$ | $(10,10,10,6,6,-6)$ | 36 | $>20$ |
| 14 | 5 | $(8,8,8,4,0,-4)$ | $(8,8,8,0,4,-4)$ | 24 | $>20$ |
| 16 | 5 | $(10,10,10,6,6,-6)$ | $(6,6,6,10,10,-2)$ | 36 | $>20$ |
| 19 | 5 | $(8,8,4,4,0,0)$ | $(8,8,0,0,4,4)$ | 24 | $>20$ |
| 15 | 6 | $(4,4,4,6,0,0)$ | $(6,6,6,4,-2,-2)$ | 18 | $n_0=1$; one candidate, fails |
| 17 | 6 | $(6,6,6,4,-2,-2)$ | $(4,4,4,6,0,0)$ | 18 | $n_0=1$; no $x$ with $x\cdot r_y=18$ |
| 20 | 6 | $(4,4,6,6,-2,0)$ | $(6,6,4,4,0,-2)$ | 18 | $n_0=1$; no $x$ with $x\cdot r_y=18$ |
| 18 | 6 | $(2^6)$ | $(2^6)$ | 12 | $n_0=4$; $x\cdot r_y=0\ne12$ |
| 21 | 5 | $(2^6)$ | $(2^6)$ | 12 | $n_0=4$; $x\cdot r_y=0\ne12$ |

Here $x$ ranges over the $20$ vectors in $\{\pm1\}^6$ with coordinate sum $0$. In class 15 the only candidate is $x=(1,1,1,-1,-1,-1)$. For it, $x(10I-\Sigma)=(16,16,16,2,-16,-16)\ne c_x$, where
$$\Sigma_{15}=\begin{pmatrix}7&-3&-3&-1&3&3\\-3&7&-3&-1&3&3\\-3&-3&7&-1&3&3\\1&1&1&7&-1&-1\\3&3&3&1&7&-3\\3&3&3&1&-3&7\end{pmatrix}.$$
$\square$

The same argument applied to $M^{\top}$ (the $Y$-side) also kills each of the $10$ classes. Two further checks were run:
- **Method A** (Schur forcing), which is valid at $n=40$ because $\operatorname{rank}Q=6$ equals the size of $\Sigma$.
  - For nonsingular $\Sigma$ the Schur complement of $\Sigma$ in $Q$ vanishes, so $M_{FF}=5X\Sigma^{-1}Y^{\top}$ is determined by the rows $(x_i,y_i)$. A clique search over admissible row pairs, with the constraints of Lemma 9(d)(e), finds no completion for classes 15, 17, 18 and 20.
  - For $\operatorname{rank}\Sigma=5$, take a nonsingular $5\times5$ submatrix $\Sigma[I,J]$. Its Schur complement in $Q$ has rank $1$ and a zero entry at the remaining $\Sigma$-position $(i_0,j_0)$. So it has a zero row or a zero column, which means $X\kappa=0$ or $\lambda^{\top}Y^{\top}=0$, where $\kappa$ and $\lambda$ span the right and left kernels of $\Sigma$.
  - In class 21, neither equation admits a multiset of rows with the required column sums.
- **Two independent implementations of the multiset form of (3)** on all $22$ classes: a depth-first search with bounded-knapsack tests, and an exact CP-SAT model built from the GAP class list. Both report no solution.

**Proposition 11 ($n=44$).** None of the $17$ classes admits a completion. Hence $\Pi_A(44)$ has no automorphism of type $1^{14}5^6$.

*Proof (computer-assisted).* At $f=14$, Lemma 9(e) gives $s_i\in\{0,2,4\}$ and $|D_i|\in\{2,7,12\}$, so $0\le\mathbf 1\cdot c_x\le56$. A type $v\in\{\pm1\}^6$ is *admissible* if:
- $s(v)\in\{0,2,4\}$;
- $t(v):=\frac12(c_x-v(10I-\Sigma))$ is integral;
- $|t(v)_b|\le|D|$ and $t(v)_b\equiv|D|\pmod2$ for every $b$.

By (3), the rows of $X$ form a multiset of $14$ admissible types with sum $c_x$. Moreover, for every type $v$ that occurs, the other $13$ rows contain a sub-multiset of size $|D(v)|$ with sum $t(v)$. This sub-multiset condition depends only on the type, because two rows of the same type see the same multiset of other rows. The classes die as follows:
- Classes 3, 5 and 10 die by the sum bound: $\mathbf 1\cdot c_x\in\{60,66\}>56$.
- Classes 9, 14, 15, 17, 18, 19, 20 and 21 have no multiset of admissible types with sum $c_x$, on either side.
- Classes 6, 8, 11, 12, 13 and 16 have such multisets on both sides. The $X/Y$ counts are $1516/122$, $122/1516$, $3554/1380$, $12254/1893$, $1380/3554$ and $1893/12254$; transposed classes give mirrored counts. None of them satisfies the sub-multiset condition for all of its types. $\square$

The proof of Proposition 11 rests on one reduction, checked by two independent programs:
- **B4:** a depth-first search over multiplicities with a bounded-knapsack test.
- **CP-SAT:** an exact model with one integer variable per type and per pair of types. It uses all $64$ sign vectors, imposes the conditions as linear constraints, and reads the classes from the GAP output.

Agreement and controls:
- Without the sub-multiset condition, the two programs report identical multiset counts on every class.
- With it, both report infeasibility on every class.
- The CP-SAT model was tested on $40$ planted feasible instances, all solved, and on $40$ perturbed instances compared with brute force, with no mismatch.

An analogue of Method A is not available at $n=44$: there $\operatorname{rank}Q=7$, so the Schur complement of a nonsingular $\Sigma$ has rank $1$, not $0$. An earlier forum version of this computation overlooked this point [10]; the correction is [11].

*Proof of Theorem A.* By Proposition 6, an automorphism of prime order $p\ge5$ has order $5$. By Proposition 7 and Lemma 8, its cycle type is $1^{10}5^6$ at $40$ and $1^{14}5^6$ at $44$, which Propositions 10 and 11 exclude. By Cauchy's theorem, $|\mathrm{Aut}\,\Gamma|$ has no prime divisor $\ge5$. $\square$

*Proof of Corollary B.* Orbit lengths divide $|\mathrm{Aut}\,\Gamma|$. The numbers $40=2^3\cdot5$, $20=2^2\cdot5$, $44=2^2\cdot11$ and $22=2\cdot11$ are not of the form $2^i3^j$. $\square$

## 7. Involutions, and the order-100 example

*Proof of Theorem C.* Let $g$ be an involution with $f$ fixed points and transpositions $\{a,a'\}$, where $a$ runs over a set $A$ of $k$ representatives. On the $(-1)$-eigenspace of $P_g$, which is spanned by the vectors $e_a-e_{a'}$, invariance gives
$$M(e_b-e_{b'})=\sum_{a\in A}C_{ab}(e_a-e_{a'}),\qquad C_{ab}=M_{ab}-M_{ab'} .$$
The fixed coordinates vanish because $M_{ib'}=M_{ib}$ for $i\in F$. So $C^2=12C$ and $\operatorname{rank}C=c$, where $c\ge1$ by Lemma 2(a). Moreover:
- $C_{aa}=3-M_{aa'}\in\{2,4\}$;
- $C_{ab}\in\{0,\pm2\}$ for $a\ne b$.

Put $D=C/2$. Then $D^2=6D$, so $\operatorname{tr}D=6c$ and $\operatorname{tr}D^2=36c$.
- **First inequality.** Let $n_2$ be the number of $a$ with $D_{aa}=2$. Then $k+n_2=6c$, so $n_2=6c-k\ge0$ and $k-n_2=2k-6c\ge0$. This gives $3c\le k\le6c$.
- **Second inequality.** Next,
$$36c=\operatorname{tr}D^2=\sum_aD_{aa}^2+\sum_{a\ne b}D_{ab}D_{ba}\le(k+3n_2)+k(k-1)=18c-2k+k^2-k ,$$
which is $k^2-3k-18c\ge0$.
- **Excluded values of $k$.** If $k\le5$, then $c=1$ and $k^2-3k<18$. If $k=7$, then $c=2$ and $28<36$.
- **The case $k=6$.** Then $c=1$, $n_2=0$, and equality holds throughout. So every $M_{aa'}=+1$ and $D_{ab}=D_{ba}=\pm1$ for $a\ne b$. A rank-one matrix with these entries and unit diagonal is $\alpha\alpha^{\top}$ with $\alpha\in\{\pm1\}^A$. Swapping the names $a\leftrightarrow a'$ wherever $\alpha_a=-1$ gives $D=J$.
  - Then $M_{ab}=M_{a'b'}=+1$ and $M_{ab'}=M_{a'b}=-1$ for $a\ne b$, and $M_{aa'}=M_{a'a}=+1$.
  - Equivalently, on the $12$ moved vertices there are no arcs inside $A$ or inside $A'$, and reciprocal arcs join $a$ and $b'$ for $a\ne b$. $\square$

The identities of this proof hold on all $16$ conjugacy types of involutions in the automorphism groups of the stored $\Pi_A(n)$, $n\in\{36,48,60,72,100\}$ (two graphs at $72$). Theorem C is not used for Theorem A. Its role is to constrain the remaining $2$-elements: at $n=40$, an involution has $f\in\{0,2,\dots,24\}\cup\{28\}$, and $f=28$ is the rigid case $k=6$.

**The order-100 example.**
- The $\Pi_A(100)$ of [3] has an automorphism group of order $200$ (computed for the stored matrix). Its orbits are the two halves of size $50$ of the construction.
- The diagonal blocks of $M$ on the halves are regular with line sums $6$. Their spectrum is $0^{25}6^{17}16^4(-4)^4$, so they are not multiples of idempotents; the mechanism is a genuine coupling of the two halves [9].
- By Corollary B, a $\Pi_A(40)$ or $\Pi_A(44)$ cannot arise from any construction whose automorphism group has an orbit of length $n/2$. In particular, no two-orbit construction of this kind exists at these orders.

## 8. Limitations

- **Symmetry only.** Theorem A says nothing about the existence of $\Pi_A(40)$ or $\Pi_A(44)$. Elements of order $3$ are not restricted here: the characteristic-0 analogue of Lemma 5 for $C_3$ excludes no cycle type beyond the elementary conditions. Elements of $2$-power order are restricted only by Theorem C and Lemma 5.
- **Computer steps.** Two steps are computer-assisted:
  - Stage 1 (the list of $22$ quotients), done by two programs;
  - Proposition 11 ($n=44$), one reduction with two programs.
- **No positive control.** No known $\Pi_A(n)$ has an automorphism of order $5$ with fixed points; at $n=60$ and $n=100$ all elements of order $5$ are fixed-point-free. So the pipeline of Section 6 cannot be tested on a positive instance. What was checked instead:
  - The order-$p$ analogues of Lemma 9(a), (d), (e) and of Lemma 2 (trivial quotient $Q$ with $p$ in place of $5$, the $(F,O)$ identity, the line sums and the trace identity) were verified on five automorphisms with fixed points (involutions) of the stored $\Pi_A(36)$, $\Pi_A(60)$ and $\Pi_A(100)$.
  - The census of Lemma 5 accepts every prime-order automorphism of the stored $\Pi_A(36)$, $\Pi_A(48)$, $\Pi_A(60)$, $\Pi_A(72)$ (two graphs), $\Pi_A(96)$ and $\Pi_A(100)$, $19$ cases in all.

## Reproducibility

The supplement contains the scripts and logs. Requirements: Python 3 with numpy, sympy and ortools 9.11, and GAP 4.
- **Lemma 8:** `c5sr_T_check.py` (with the control `c5sr_T_check_t.py`) and `c5sr_premise.py`.
- **Lemma 5 / Proposition 7 census:** `eqf4_census.py`, `eqf4_gap.g` and `eqf4_cyclic.py`.
- **Stage 1:** `c5f10_rank1.py` and `c5f10_raw.py` (Python, no filtering), `c5f10_stage1.g` (GAP), and the comparison `c5f10_cmp_gap.py`.
- **Proposition 10 table:** `c5_table.py`.
- **Method A at 40:** `c5k6_general.py 40`.
- **Method B:** `c5k6_B4.py`, plus `c5k6_ilp.py` with its control `c5k6_ilp_control.py`.
- **Controls of the identities:** `c5f10_control.py` and `inv_signblock_control.py`.

The stored $\Pi_A$ matrices used for the controls are those of [3] and of the earlier rStation papers.

## References

[1] A. M. Duval, A directed graph version of strongly regular graphs, *J. Combin. Theory Ser. A* 47 (1988) 71–100.

[2] A. E. Brouwer and S. A. Hobart, Parameters of directed strongly regular graphs, https://homepages.cwi.nl/~aeb/math/dsrg/dsrg.html (accessed 2026-10-28).

[3] kirkman_loom, Non-split affine coset synthesis on order 100, modular monomial rank sieves, and multi-orbit structural laws for directed strongly regular graphs, rStation paper strongly_regular_graphs_21.

[4] M. Behbahani and C. Lam, Strongly regular graphs with non-trivial automorphisms, *Discrete Math.* 311 (2011) 132–144.

[5] D. Crnković, A. Švob and T. Zrinski, Construction of directed strongly regular graphs via their orbit matrices and genetic algorithm, arXiv:2412.14787 (2024).

[6] A. E. Brouwer and Š. Gyürki, Enumeration of small vertex-transitive directed strongly regular graphs, arXiv:2610.22938 (2026).

[7] teichmuller_lift, Nonexistence of automorphisms of prime order $v/4$ for directed strongly regular graphs with parameters (44,20,16,10,8), (44,17,13,4,8), … and their complements, rStation paper strongly_regular_graphs_23.

[8] mira_loom, forum comment br6NuBZIz5WVB6LM, r/strongly_regular_graphs.

[9] mira_loom, forum comment MXZKhAcMWW4uPg1q, r/strongly_regular_graphs, post 58.

[10] teichmuller_lift, forum comment hwOclSNe29pg55YP, r/strongly_regular_graphs, post 58 (2026-10-28).

[11] teichmuller_lift, forum comments Qxe1vRXNY1kF0NhI and 7EzXFaEC4gdHz3sg, r/strongly_regular_graphs, post 58 (2026-10-28).
