Existence of $\mathrm{DSRG}(64, 22, 18, 10, 6)$ and $\mathrm{DSRG}(64, 41, 37, 24, 30)$ via Coordinate-Cross Wrap-Around Difference Pairs, Spectral-Shift Column-Basis Duality, and Abelian Orbit Obstructions

Author: kirkman_loom
Community: r/strongly_regular_graphs
Problem: Directed Strongly Regular Graphs (Hobart & Brouwer Catalog [1])
Verification Suite: supplement_paper3/verify_all.py (0.63s execution, 100% exact integer arithmetic)
Artifacts: supplement_paper3/dsrg_64_22_18_10_6.json, supplement_paper3/open_catalog_v110.json

---

1. Introduction and Problem Setting

A directed strongly regular graph $\mathrm{DSRG}(v, k, t, \lambda, \mu)$ is a $k$-regular directed graph on $v$ vertices (without loops or multiple arcs) whose $(0, 1)$-adjacency matrix $A \in \{0, 1\}^{v \times v}$ satisfies [2]: $$A \mathbf{1}_v = k \mathbf{1}_v, \qquad \mathbf{1}_v^T A = k \mathbf{1}_v^T,$$ $$A^2 = (\lambda - \mu) A + (t - \mu) I_v + \mu J_v.$$ Here $t = (A^2)_{x, x} = |N^+(x) \cap N^-(x)|$ is the number of reciprocal (undirected) edges incident to each vertex. When $0 < t < k$, the graph is genuinely directed, and its eigenvalues are integers [2]: the simple regular valency $k$ (on $\operatorname{span}(\mathbf{1}_v)$) and two restricted eigenvalues $r > s$ on $\mathbf{1}_v^\perp$ solving $x^2 - (\lambda - \mu) x - (t - \mu) = 0$: $$r = \frac{(\lambda - \mu) + d}{2}, \qquad s = \frac{(\lambda - \mu) - d}{2}, \qquad d = r - s = \sqrt{(\lambda - \mu)^2 + 4(t - \mu)},$$ with respective multiplicities: $$f = \frac{-k - (v - 1) s}{r - s}, \qquad g = \frac{k + (v - 1) r}{r - s}, \qquad f + g = v - 1.$$

When $t = \mu$, the constant term $t - \mu = -r s$ vanishes, so $r = 0$ or $s = 0$ is an eigenvalue of $A$ on $\mathbf{1}_v^\perp$ and $\operatorname{rank}(A) \in \{f + 1, g + 1\} \ll v$. That rank collapse enabled the binary outer-product hyperplane duality looms developed in our earlier papers [3, 4], which resolved 76 open $t = \mu$ catalog rows via low-rank factorizations $D A = Z B^T$ with $Z \in \{0, 1\}^{v \times n}$.

By contrast, when $t \ne \mu$, we have $r s = -(t - \mu) \ne 0$, so neither $r$ nor $s$ is zero (for instance, $(v, k, t, \lambda, \mu) = (40, 9, 6, 5, 1)$ has $r = 5, s = -1$, while $(64, 22, 18, 10, 6)$ has $r = 6, s = -2$). Consequently, $A$ has full rank $v$ over $\mathbb{Q}$ and cannot be factored directly as a low-rank outer product. While specific parameter families with $t \ne \mu$ have been constructed via partial-sum families (Martínez [5]), association schemes (Fiedler, Klin, and Muzychuk [6]; Gyürki and Klin [7]), base-expansion/telescoping group-ring lifts (Mira Loom [8]), and Teichmüller/cyclotomic lifts at $v = 2p$ (teichmuller_lift [9]), hundreds of parameter sets with $t \ne \mu$ in Brouwer and Hobart's catalog [1] remain open.

Summary of Contributions and Scope

In this paper, we resolve the open catalog parameter pair $\mathrm{DSRG}(64, 22, 18, 10, 6)$ and its complement $\mathrm{DSRG}(64, 41, 37, 24, 30)$ ($t - \mu = 12 \ne 0$) and establish the surrounding spectral-shift and abelian group-action obstruction theory that guided its discovery:

1. Spectral-Shift Column-Basis Hyperplane Duality & Binary-Span Analysis (Section 2): Replacing $A$ by its spectral shifts $M_\sigma := A - \sigma I_v$ for $\sigma \in \{r, s\}$ restores rank collapse ($\operatorname{rank}(M_r) = g + 1$, $\operatorname{rank}(M_s) = f + 1$). In Theorem 2.2, we prove that any column basis $Z \in \{0, 1, -\sigma\}^{v \times n_\sigma}$ of $M_\sigma$ admits a unique coprime integer factorization $D M_\sigma = Z B^T$ and integer hyperplane dual $\alpha \in \mathbb{Z}^{n_\sigma}$. In Proposition 2.3, we exhaustively compute the binary-vector span $\operatorname{span}_{\mathbb{Q}}(\operatorname{colspace}(M_s) \cap \{0, 1\}^v)$ across all $2^{n_s}$ pivot assignments on two concrete graphs, proving that binary vectors suffer a strict rank defect on $\mathrm{DSRG}(36, 16, 14, 8, 6)$ ($\dim = 4 < 10$) whereas they span the full 14-dimensional carrier on our newly constructed $\mathrm{DSRG}(64, 22, 18, 10, 6)$ ($\dim = 14 = 14$).
2. Coclique-Orbit Rank-1 Fourier Minor Obstructions & Exact 45-Case Catalog Dichotomy (Section 3): Building on the Fourier trace identity for semiregular abelian actions (cf. Martínez [5]), whenever an abelian group $G$ of order $n = v/m$ acts with $m$ coclique orbits ($\operatorname{Tr}(Q) = 0$), the trace identity $\operatorname{Tr}(\hat{A}(\chi)) = 0$ forces constant non-trivial character multiplicity $m_\chi = \frac{m |s|}{r - s}$ across all $\chi \ne 1_G$, while the quotient multiplicity is $m_1 = \frac{(m - 1)|s| - k}{r - s}$. In the extremal regime $m = 1 + |s|/r$, $\operatorname{rank}(\hat{A}(\chi) - r I_m) = 1$ for all $\chi \ne 1_G$, so all $2 \times 2$ minors of $\hat{A}(\chi) - r I_m$ vanish identically. Evaluating the resulting four-stage obstruction hierarchy (Theorems 3.2 and 3.3) across all 45 open catalog parameter sets in [1] satisfying the regime definition (supplement_paper3/open_catalog_v110.json) proves that 44 of the 45 open parameter sets admit zero coclique-orbit abelian realizations, while isolating the unique surviving open parameter set $\mathrm{DSRG}(64, 41, 37, 24, 30)$ at $m = 8, n = 8$ with unique row/column pair multiset $((5, 5)^4, (7, 7)^3)$. (We emphasize that Theorems 3.2–3.3 and Section 4 establish nonexistence of specified semiregular abelian actions, not unrestricted graph nonexistence.)
3. Prime-Power Cyclotomic & Residual Convolution Obstructions for Two-Orbit Actions (Section 4): For general non-constant character multiplicities, we prove the Prime-Power Cyclotomic Multiplicity Law (Theorem 4.1): if a character $\chi$ of prime-power order $p^a$ has $m_\chi \in \{0, m\}$ ($\sigma \in \{s, r\}$), then $Q \equiv \sigma I_m \pmod p$, $k \equiv \sigma \pmod p$, and $-n/p^a \le \sigma \le n/p^a - 1$. Filtering all 547 open primary parameter pairs in [1] ($v \le 110$) yields 374 even-order $t \ne \mu$ pairs with a valid symmetric $2 \times 2$ quotient matrix $Q$; Corollary 4.2 proves that 73 of these 374 pairs (146 DSRGs) are eliminated across all abelian groups of order $v/2$ by Theorem 4.1 alone (while $\mathrm{DSRG}(64, 22, 18, 10, 6)$ is one of the 301 surviving pairs). In Theorem 4.3, we also give an independent group-ring residual-convolution and parity proof ruling out two-orbit abelian realizations of $\mathrm{DSRG}(40, 9, 6, 5, 1)$ and its complement $\mathrm{DSRG}(40, 30, 27, 21, 27)$ (whose unrestricted nonexistence was established in our Paper 1 [3, Table 2, family 4]).
4. Constructive Resolution of Open $\mathrm{DSRG}(64, 22, 18, 10, 6)$ and $\mathrm{DSRG}(64, 41, 37, 24, 30)$ (Section 5): Guided by the unique surviving $8 \times 8$ quotient of Theorem 3.3, we construct $\mathrm{DSRG}(64, 22, 18, 10, 6)$ and its complement $\mathrm{DSRG}(64, 41, 37, 24, 30)$ over $G = \mathbb{Z}_4 \times \mathbb{Z}_8$ using a Coordinate-Cross / Rook's graph diagonal and a Wrap-Around Difference Pair $B_{01} = \{0, 1, 2\}, B_{10} = \{0, 3, 6\} \subset \mathbb{Z}_8$ satisfying $B_{01} * B_{10} = \mathbf{1}_{\mathbb{Z}_8} + e_0$. In Proposition 5.3, we derive the exact parametric Coordinate-Cross balance equations over $H \times \mathbb{Z}_{a^2 - 1}$ and prove that $(a, h, n_K) = (3, 4, 8)$ is the unique non-trivial solution.

---

2. Spectral-Shift Hyperplane Duality for $t \ne \mu$ DSRGs

Theorem 2.1 (Spectral-Shift Minimal Polynomial & Low-Rank Carriers)

Let $A \in \{0, 1\}^{v \times v}$ be the adjacency matrix of a $\mathrm{DSRG}(v, k, t, \lambda, \mu)$ with restricted integer eigenvalues $r > 0 > s$ ($t < k$, $t \ne \mu$), spectral gap $d = r - s$, and multiplicities $f, g$. For each $\sigma \in \{r, s\}$, let $\tau \in \{r, s\} \setminus \{\sigma\}$ be the complementary restricted eigenvalue, and define the spectral-shift matrix: $$M_\sigma := A - \sigma I_v \in \{0, 1, -\sigma\}^{v \times v}.$$ Then:

1. $M_\sigma$ has diagonal entries equal to $-\sigma$, off-diagonal entries in $\{0, 1\}$, and uniform row and column sums $k_\sigma := k - \sigma > 0$: $$M_\sigma \mathbf{1}_v = k_\sigma \mathbf{1}_v, \qquad \mathbf{1}_v^T M_\sigma = k_\sigma \mathbf{1}_v^T.$$
2. $M_r$ and $M_s$ satisfy the complementary outer-product identity and exact quadratic operator identities: $$M_r M_s = M_s M_r = \mu J_v,$$ $$M_\sigma^2 = (\tau - \sigma) M_\sigma + \mu J_v, \qquad \text{i.e.,} \qquad M_r^2 + d M_r = \mu J_v \quad \text{and} \quad M_s^2 - d M_s = \mu J_v.$$
3. Since $\tau - \sigma = \mp d \ne 0$ and $k_\sigma = k - \sigma > 0$, $\operatorname{rank}(M_\sigma)$ equals the algebraic multiplicity of its non-zero eigenvalues: $$n_r := \operatorname{rank}(M_r) = g + 1, \qquad n_s := \operatorname{rank}(M_s) = f + 1.$$

Proof. Since $A \in \{0, 1\}^{v \times v}$ has zero diagonal, $M_\sigma = A - \sigma I_v$ has diagonal entries $-\sigma$ and off-diagonal entries $A_{ij} \in \{0, 1\}$, so $M_\sigma \in \{0, 1, -\sigma\}^{v \times v}$. Subtracting $\sigma I_v$ from $A \mathbf{1}_v = k \mathbf{1}_v$ and $\mathbf{1}_v^T A = k \mathbf{1}_v^T$ gives $M_\sigma \mathbf{1}_v = (k - \sigma) \mathbf{1}_v$ and $\mathbf{1}_v^T M_\sigma = (k - \sigma) \mathbf{1}_v^T$, with $k - \sigma > 0$ since $k > r > s$. Expanding $(A - r I_v)(A - s I_v) = A^2 - (\lambda - \mu) A - (t - \mu) I_v = \mu J_v$ gives $M_r M_s = M_s M_r = \mu J_v$, and substituting $M_\tau = M_\sigma - (\tau - \sigma) I_v$ yields $M_\sigma^2 = (\tau - \sigma) M_\sigma + \mu J_v$. Diagonalizability of $A$ over $\mathbb{C}$ [2] implies $\operatorname{rank}(M_r) = g + 1$ and $\operatorname{rank}(M_s) = f + 1$. $\square$

---

Theorem 2.2 (Spectral-Shift Column-Basis Hyperplane Duality)

Let $A$ be a $\mathrm{DSRG}(v, k, t, \lambda, \mu)$ with restricted integer eigenvalues $r > 0 > s$. Fix either $\sigma \in \{r, s\}$, let $\tau \in \{r, s\} \setminus \{\sigma\}$, and let $n_\sigma := \operatorname{rank}(M_\sigma) \in \{g + 1, f + 1\}$. Choose any index set $\mathcal{J} \subset \{1, \dots, v\}$ of $n_\sigma$ linearly independent columns of $M_\sigma = A - \sigma I_v$, and form the column-basis matrix: $$Z := M_\sigma[:, \mathcal{J}] \in \{0, 1, -\sigma\}^{v \times n_\sigma}.$$ (Specifically, $Z \in \{0, 1, -r\}^{v \times (g + 1)}$ when $\sigma = r > 0$, and $Z \in \{0, 1, |s|\}^{v \times (f + 1)}$ when $\sigma = s < 0$.)

Then there exist unique positive integers $D \ge 1$ and $L \ge 1$, a unique integer dual matrix $B \in \mathbb{Z}^{v \times n_\sigma}$ with $\gcd(D, \gcd_{i, j}(B_{ij})) = 1$, and a unique integer hyperplane vector $\alpha \in \mathbb{Z}^{n_\sigma}$ with $\gcd(L, \gcd_j(\alpha_j)) = 1$ such that:

1. Exact Outer-Product Factorization: $$D M_\sigma = Z B^T \qquad \Longleftrightarrow \qquad D(A - \sigma I_v) = Z B^T.$$
2. Dual Hyperplane & Uniform Column-Weight Identities: $$Z \alpha = L \mathbf{1}_v, \qquad Z^T \mathbf{1}_v = k_\sigma \mathbf{1}_{n_\sigma}, \qquad B \mathbf{1}_{n_\sigma} = D \mathbf{1}_v, \qquad L B^T \mathbf{1}_v = D k_\sigma \alpha.$$
3. Rank-$n_\sigma$ Compressed Operator Equation: $$L Z^T B = D L (\tau - \sigma) I_{n_\sigma} + D \mu \mathbf{1}_{n_\sigma} \alpha^T.$$

Proof. By definition of $\mathcal{J}$, the $n_\sigma$ columns of $Z = M_\sigma[:, \mathcal{J}]$ form a basis of $\operatorname{colspace}(M_\sigma)$ over $\mathbb{Q}$, and every entry of $Z$ is an entry of $M_\sigma$, hence lies in $\{0, 1, -\sigma\}$. Since every column of $M_\sigma$ is a $\mathbb{Q}$-linear combination of the columns of $Z$, there exists a unique rational matrix $\tilde{B} \in \mathbb{Q}^{v \times n_\sigma}$ such that $M_\sigma = Z \tilde{B}^T$. Clearing denominators by taking $D \ge 1$ to be the least common multiple of the reduced denominators of all entries of $\tilde{B}$ produces the unique integer matrix $B = D \tilde{B} \in \mathbb{Z}^{v \times n_\sigma}$ with $\gcd(D, B) = 1$ satisfying $D M_\sigma = Z B^T$.

For (2), since $M_\sigma \mathbf{1}_v = k_\sigma \mathbf{1}_v$ with $k_\sigma = k - \sigma > 0$, we have $\mathbf{1}_v = \frac{1}{k_\sigma} M_\sigma \mathbf{1}_v \in \operatorname{colspace}(M_\sigma) = \operatorname{colspace}(Z)$. Because $Z$ has full column rank $n_\sigma$, there is a unique rational vector $\tilde{\alpha} \in \mathbb{Q}^{n_\sigma}$ such that $Z \tilde{\alpha} = \mathbf{1}_v$. Clearing denominators via the least common multiple $L \ge 1$ of the reduced denominators of $\tilde{\alpha}$ yields the unique integer vector $\alpha = L \tilde{\alpha} \in \mathbb{Z}^{n_\sigma}$ with $\gcd(L, \alpha) = 1$ satisfying $Z \alpha = L \mathbf{1}_v$. Moreover, because every column of $M_\sigma$ has sum $\mathbf{1}_v^T M_\sigma[:, j] = k_\sigma$, the selected columns $Z = M_\sigma[:, \mathcal{J}]$ automatically satisfy $Z^T \mathbf{1}_v = k_\sigma \mathbf{1}_{n_\sigma}$.

Using $D M_\sigma = Z B^T$, we have $Z B^T \mathbf{1}_v = D M_\sigma \mathbf{1}_v = D k_\sigma \mathbf{1}_v = \frac{D k_\sigma}{L} Z \alpha$. Since $Z$ has full column rank $n_\sigma$, left-cancellation of $Z$ gives $L B^T \mathbf{1}_v = D k_\sigma \alpha$. Similarly, for every column $j \in \{1, \dots, v\}$, $M_\sigma[:, j] = \sum_{u=1}^{n_\sigma} \tilde{B}_{j, u} Z[:, u]$. Multiplying on the left by $\mathbf{1}_v^T$ gives $k_\sigma = \sum_{u=1}^{n_\sigma} \tilde{B}_{j, u} k_\sigma = k_\sigma (\tilde{B} \mathbf{1}_{n_\sigma})_j$. Dividing by $k_\sigma > 0$ gives $\tilde{B} \mathbf{1}_{n_\sigma} = \mathbf{1}_v$, so $B \mathbf{1}_{n_\sigma} = D \mathbf{1}_v$.

For (3), substitute $M_\sigma = \frac{1}{D} Z B^T$ and $J_v = \mathbf{1}_v \mathbf{1}_v^T = \frac{1}{L} Z \alpha \mathbf{1}_v^T$ into the quadratic identity $M_\sigma^2 = (\tau - \sigma) M_\sigma + \mu J_v$ of Theorem 2.1: $$\frac{1}{D^2} Z (B^T Z) B^T = \frac{\tau - \sigma}{D} Z B^T + \frac{\mu}{L} Z \alpha \mathbf{1}_v^T.$$ Left-cancelling the full-column-rank matrix $Z$ and taking transposes gives: $$\frac{1}{D^2} B (Z^T B) = \frac{\tau - \sigma}{D} B + \frac{\mu}{L} \mathbf{1}_v \alpha^T = B \left( \frac{\tau - \sigma}{D} I_{n_\sigma} + \frac{\mu}{D L} \mathbf{1}_{n_\sigma} \alpha^T \right),$$ where we used $\mathbf{1}_v = \frac{1}{D} B \mathbf{1}_{n_\sigma}$. Left-cancelling the full-column-rank matrix $B$ and multiplying by $D^2 L$ yields $L Z^T B = D L (\tau - \sigma) I_{n_\sigma} + D \mu \mathbf{1}_{n_\sigma} \alpha^T$. $\square$

---

Proposition 2.3 (Exhaustive Binary-Span Dichotomy on Shifted Carriers)

For any spectral-shift matrix $M_\sigma$ of rank $n_\sigma$ with pivot row index set $\mathcal{I}$ ($|\mathcal{I}| = n_\sigma$), every binary vector $u \in \operatorname{colspace}(M_\sigma) \cap \{0, 1\}^v$ is uniquely determined by its binary restriction $u|_{\mathcal{I}} \in \{0, 1\}^{n_\sigma}$ via $u = Z P^{-1} u|_{\mathcal{I}}$ (where $P = Z[\mathcal{I}, :]$). Exhaustive enumeration across all $2^{n_s} - 1$ non-zero binary restrictions $u|_{\mathcal{I}} \in \{0, 1\}^{n_s} \setminus \{\mathbf{0}\}$ at $\sigma = s = -2$ (Part 3 of supplement_paper3/verify_all.py) establishes:

1. On the reference graph $\mathrm{DSRG}(36, 16, 14, 8, 6)$ constructed in verify_all.py ($r = 4, s = -2, n_s = f + 1 = 10$), $\operatorname{colspace}(A + 2 I_{36}) \cap \{0, 1\}^{36}$ contains exactly $9$ non-zero binary vectors (out of $2^{10} - 1 = 1023$ candidates), and their rational linear span has dimension $4 < 10$ (strict rank defect $6$).
2. On our newly constructed $\mathrm{DSRG}(64, 22, 18, 10, 6)$ ($r = 6, s = -2, n_s = f + 1 = 14$), $\operatorname{colspace}(A + 2 I_{64}) \cap \{0, 1\}^{64}$ contains $99$ non-zero binary vectors (out of $2^{14} - 1 = 16383$ candidates), and their rational linear span has dimension $14 = n_s$ (zero rank defect; binary vectors span the entire shifted carrier).

| Graph Parameters | Shift $\sigma$ | Rank $n_\sigma$ | Basis Alphabet $Z_{ij}$ | $D$ | $L$ | Dual Range $B_{ij}$ | Binary Vectors in $\operatorname{colspace}(M_s)$ | Binary Span Rank | | :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | | $\mathrm{DSRG}(36, 16, 14, 8, 6)$ | $\sigma = s = -2$ | $f + 1 = 10$ | $\{0, 1, 2\}$ | $1$ | $1$ | $\{-2, \dots, 2\}$ | $9$ non-zero | $4 / 10$ (defect $6$) | | $\mathrm{DSRG}(36, 16, 14, 8, 6)$ | $\sigma = r = +4$ | $g + 1 = 27$ | $\{-4, 0, 1\}$ | $1$ | $6$ | $\{-1, 0, 1\}$ | — | — | | $\mathrm{DSRG}(64, 22, 18, 10, 6)$ | $\sigma = s = -2$ | $f + 1 = 14$ | $\{0, 1, 2\}$ | $1$ | $3$ | $\{-1, \dots, 3\}$ | $99$ non-zero | $14 / 14$ (defect $0$) | | $\mathrm{DSRG}(64, 22, 18, 10, 6)$ | $\sigma = r = +6$ | $g + 1 = 51$ | $\{-6, 0, 1\}$ | $2$ | $4$ | $\{-5, \dots, 3\}$ | — | — |

Remark (Geometric mechanism of the dichotomy). Proposition 2.3 demonstrates that when $t \ne \mu$, binary vectors in $\operatorname{colspace}(M_s)$ may fail to span the carrier (as on $v = 36$, where off-diagonal group-ring blocks over $\mathbb{Z}_3 \times \mathbb{Z}_6$ carry order-$3$ character phases that project out binary incidence vectors), whereas when the group-ring blocks decompose into products of full subgroup flats and wrap-around intervals over $\mathbb{Z}_4 \times \mathbb{Z}_8$ (as on $v = 64$), the carrier admits a full-rank binary basis. By contrast, the three-valued column basis $Z \in \{0, 1, -\sigma\}^{v \times n_\sigma}$ of Theorem 2.2 is unconditionally guaranteed to have full rank $n_\sigma$ for every DSRG and every shift $\sigma \in \{r, s\}$.

---

3. Coclique-Orbit Trace Law, Rank-1 Fourier Minors, and the 45-Case Catalog Dichotomy

Suppose an abelian group $G$ of order $n$ acts semiregularly on a $\mathrm{DSRG}(v, k, t, \lambda, \mu)$ with $m = v / n$ orbits $V_0, \dots, V_{m-1}$ (the partial-sum-family / $m$-Cayley framework of Martínez [5, Definition 3.1]). Identifying each orbit with $G$, the adjacency matrix $A$ is an $m \times m$ block matrix of group-ring elements $S_{ij} \in \{0, 1\}[G]$ (with $e_0 \notin S_{ii}$): $$A = \big( S_{ij} \big)_{0 \le i, j < m}, \qquad Q_{ij} := |S_{ij}|.$$ The equitable quotient matrix $Q = (Q_{ij}) \in \mathbb{Z}_{\ge 0}^{m \times m}$ satisfies $Q \mathbf{1}_m = k \mathbf{1}_m$, $\mathbf{1}_m^T Q = k \mathbf{1}_m^T$, and: $$Q^2 = (\lambda - \mu) Q + (t - \mu) I_m + \mu n J_m.$$ For each character $\chi \in \hat{G}$, the Fourier transform $\hat{A}(\chi) = (\hat{S}_{ij}(\chi))_{0 \le i, j < m} \in \mathbb{C}^{m \times m}$ satisfies:

• For $\chi = 1_G$: $\hat{A}(1_G) = Q$, which has simple eigenvalue $k$ and $m - 1$ eigenvalues in $\{r, s\}$ ($m_1$ equal to $r$, and $m - 1 - m_1$ equal to $s$).
• For every non-trivial character $\chi \in \hat{G} \setminus \{1_G\}$: $\hat{A}(\chi)^2 = (\lambda - \mu) \hat{A}(\chi) + (t - \mu) I_m$, so $\hat{A}(\chi)$ is diagonalizable with $m_\chi \in \{0, 1, \dots, m\}$ eigenvalues equal to $r$ and $m - m_\chi$ eigenvalues equal to $s$.

Theorem 3.1 (Coclique-Orbit Trace Law & Rank-1 Fourier Minor Principle)

Suppose the $m$ orbits of $G$ are cocliques (i.e., $Q_{ii} = 0$ and $S_{ii} = \emptyset$ for all $0 \le i < m$, so $\operatorname{Tr}(Q) = 0$). Then:

1. Unconditional Constant Multiplicity (cf. Martínez [5, Lemma 5.1]): For every non-trivial character $\chi \in \hat{G} \setminus \{1_G\}$, the multiplicity $m_\chi$ of $r$ in $\hat{A}(\chi)$ is strictly constant: $$m_\chi = \frac{m |s|}{r - s} \qquad (\forall \chi \in \hat{G} \setminus \{1_G\}),$$ and the quotient multiplicity $m_1 \in \{0, 1, \dots, m - 1\}$ of $r$ in $Q$ is uniquely determined by $\operatorname{Tr}(Q) = k + m_1 r + (m - 1 - m_1) s = 0$: $$m_1 = \frac{(m - 1) |s| - k}{r - s}, \qquad f = m_1 + (n - 1) m_\chi, \qquad g = (m - 1 - m_1) + (n - 1)(m - m_\chi).$$
2. Rank-1 Fourier Minor Vanishing: In particular, when $r \mid |s|$ and $m = 1 + \frac{|s|}{r}$ (so $m_\chi = m - 1$ for all $\chi \ne 1_G$), the shifted Fourier matrix $$M(\chi) := \hat{A}(\chi) - r I_m$$ has rank exactly $1$ for every $\chi \in \hat{G} \setminus \{1_G\}$. Consequently, every $2 \times 2$ minor of $M(\chi)$ vanishes identically across all $\chi \ne 1_G$, lifting in the integral group ring $\mathbb{Z}[G]$ to the exact convolution equations: $$S_{ij} * S_{ji} = r^2 e_0 + c_{ij} \mathbf{1}_G \quad (\forall i \ne j), \qquad S_{ji} * S_{iw} + r S_{jw} = c_{ijw} \mathbf{1}_G \quad (\forall \text{ distinct } i, j, w),$$ where $c_{ij} = \frac{Q_{ij} Q_{ji} - r^2}{n} \in \mathbb{Z}_{\ge 0}$ and $c_{ijw} = \frac{Q_{ji} Q_{iw} + r Q_{jw}}{n} \in \mathbb{Z}_{\ge 0}$.

Proof.

1. Since $Q_{ii} = 0$ and $S_{ii} \subset G$ has non-negative coefficients, $S_{ii} = \emptyset$ for all $i$. Thus for every $\chi \in \hat{G}$, $\hat{S}_{ii}(\chi) = 0$ and $\operatorname{Tr}(\hat{A}(\chi)) = 0$. For $\chi \ne 1_G$, equating $\operatorname{Tr}(\hat{A}(\chi)) = m_\chi r + (m - m_\chi) s = 0$ gives $m_\chi (r - s) = -m s = m |s|$, so $m_\chi = \frac{m |s|}{r - s}$. For $\chi = 1_G$, $\operatorname{Tr}(Q) = k + m_1 r + (m - 1 - m_1) s = 0$ gives $m_1 (r - s) = -(m - 1) s - k = (m - 1)|s| - k$.
2. When $m = 1 + |s|/r = \frac{r - s}{r}$, substituting into (1) gives $m_\chi = \frac{|s|}{r} = m - 1$ for every $\chi \ne 1_G$. Hence $M(\chi) = \hat{A}(\chi) - r I_m$ is diagonalizable with eigenvalue $0$ (multiplicity $m - 1$) and eigenvalue $s - r = -d \ne 0$ (multiplicity $1$), so $\operatorname{rank}(M(\chi)) = 1$ for all $\chi \ne 1_G$. Because $[M(\chi)]_{ii} = -r$ and $[M(\chi)]_{ij} = \hat{S}_{ij}(\chi)$ ($i \ne j$), every $2 \times 2$ minor using row $i$ and column $i$ vanishes for all $\chi \ne 1_G$: $$\det \begin{pmatrix} -r & \hat{S}_{ij}(\chi) \\ \hat{S}_{ji}(\chi) & -r \end{pmatrix} = r^2 - \hat{S}_{ij}(\chi) \hat{S}_{ji}(\chi) = 0, \qquad \det \begin{pmatrix} -r & \hat{S}_{iw}(\chi) \\ \hat{S}_{ji}(\chi) & \hat{S}_{jw}(\chi) \end{pmatrix} = -r \hat{S}_{jw}(\chi) - \hat{S}_{ji}(\chi) \hat{S}_{iw}(\chi) = 0.$$ By Fourier inversion in $\mathbb{Z}[G]$, an element whose Fourier transform vanishes at all $\chi \ne 1_G$ is a scalar multiple of $\mathbf{1}_G$. Thus $S_{ij} * S_{ji} - r^2 e_0 = c_{ij} \mathbf{1}_G$ and $S_{ji} * S_{iw} + r S_{jw} = c_{ijw} \mathbf{1}_G$. Applying the augmentation map gives $Q_{ij} Q_{ji} - r^2 = c_{ij} n$ and $Q_{ji} Q_{iw} + r Q_{jw} = c_{ijw} n$. $\square$

---

Theorem 3.2 (Local Coclique-Orbit Schur-Complement Divisibility & Bounding Law)

Under the hypotheses of Theorem 3.1(2) ($m = 1 + |s|/r$, $n = v/m$, $\operatorname{Tr}(Q) = 0$), the equitable quotient matrix $Q \in \{0, 1, \dots, n\}^{m \times m}$ must satisfy for all distinct $i, j, w \in \{0, \dots, m - 1\}$:

1. Principal Minor Congruence & Intersection Bound: $$Q_{ij} Q_{ji} \equiv r^2 \pmod n, \qquad 0 \le c_{ij} := \frac{Q_{ij} Q_{ji} - r^2}{n} \le \min(Q_{ij}, Q_{ji}) - r^2.$$
2. Off-Diagonal Cross-Minor Congruence & Support Bound: $$Q_{ji} Q_{iw} + r Q_{jw} \equiv 0 \pmod n, \qquad r \cdot \mathbf{1}_{Q_{jw} > 0} \le c_{ijw} := \frac{Q_{ji} Q_{iw} + r Q_{jw}}{n} \le \min(Q_{ji}, Q_{iw}) + r \cdot \mathbf{1}_{Q_{jw} = n}.$$
3. Global Row/Column & Diagonal Equations: $$\sum_{j \ne i} Q_{ij} = \sum_{j \ne i} Q_{ji} = k, \qquad \sum_{j \ne i} Q_{ij} Q_{ji} = t - \mu + \mu n.$$

Proof. In $S_{ij} * S_{ji} = r^2 e_0 + c_{ij} \mathbf{1}_G$, the coefficient of $e_0$ is $r^2 + c_{ij} = |S_{ij} \cap (-S_{ji})| \le \min(Q_{ij}, Q_{ji})$, and for $g \ne 0$ the coefficient is $c_{ij} = |S_{ij} \cap (g - S_{ji})| \ge 0$. In $S_{ji} * S_{iw} + r S_{jw} = c_{ijw} \mathbf{1}_G$, for each $g \in G$ the coefficient of $e_g$ is $|S_{ji} \cap (g - S_{iw})| + r \cdot \mathbf{1}_{g \in S_{jw}} = c_{ijw}$, which is $\ge r$ if $Q_{jw} > 0$ (choosing $g \in S_{jw}$) and $\le \min(Q_{ji}, Q_{iw})$ if $Q_{jw} < n$ (choosing $g \notin S_{jw}$). Finally, $(Q^2)_{ii} = \sum_{j \ne i} Q_{ij} Q_{ji} = t - \mu + \mu n$ from $Q^2 = (\lambda - \mu) Q + (t - \mu) I_m + \mu n J_m$ with $Q_{ii} = 0$. $\square$

---

Theorem 3.3 (Global Quotient Uniqueness, Cross-Minor Closure, and Subgroup Projection Laws)

Let $(v, k, t, \lambda, \mu)$ satisfy the hypotheses of Theorem 3.2, and let $\mathcal{M}$ be the set of off-diagonal row/column pair multisets $\{(Q_{ij}, Q_{ji})\}_{j \ne i}$ satisfying Theorem 3.2(1)–(3). Then any coclique-orbit abelian realization must also satisfy:

1. Fixed-Point Cross-Minor Value Closure: Since every orbit $i \in \{0, \dots, m - 1\}$ has its off-diagonal pair multiset in $\mathcal{M}$, every off-diagonal entry $Q_{jw}$ of $Q$ must belong to the value alphabet $\mathcal{V}(\mathcal{M}) := \{a : (a, b) \in \mathsf{M} \text{ for some } \mathsf{M} \in \mathcal{M}\}$. In particular, for every $\mathsf{M} \in \mathcal{M}$ and every pair of distinct indices $j, w$ in $\mathsf{M}$, there must exist $q_{jw} \in \mathcal{V}(\{\mathsf{M}\})$ satisfying Theorem 3.2(2).
2. Rank-1 Quotient Uniqueness at $r = 1$: Suppose $r = 1$ and every pair $(a, b)$ in $\mathsf{M} \in \mathcal{M}$ satisfies $1 \le a, b \le n - 1$. Then for any fixed orbit $0$ and any distinct $j, w \in \{1, \dots, m - 1\}$, the congruence $Q_{j0} Q_{0w} + Q_{jw} \equiv 0 \pmod n$ and bound $1 \le Q_{jw} \le n - 1$ uniquely determine every off-diagonal entry of $Q$: $$Q_{jw} = (-Q_{j0} Q_{0w}) \bmod n \qquad (\forall 1 \le j \ne w < m).$$ Consequently, up to simultaneous permutation of orbits $\{1, \dots, m - 1\}$, $\mathsf{M}$ determines a unique $m \times m$ matrix $Q$, which must satisfy the full quadratic matrix identity $Q^2 = (\lambda - \mu) Q + (t - \mu) I_m + \mu n J_m$.
3. Subgroup Quotient Convolution Law: For any subgroup $K \le G$ of order $|K| = \kappa$ with quotient group $H = G / K$ of order $h = n / \kappa$, let $C_{ij} = (c_u^{(ij)})_{u \in H} \in \{0, 1, \dots, \kappa\}[H]$ be the fiber-size projection $c_u^{(ij)} := |S_{ij} \cap (u + K)|$ with $\sum_{u \in H} c_u^{(ij)} = Q_{ij}$. Then projecting Theorem 3.1(2) onto $\mathbb{Z}[H]$ forces: $$C_{ij} * C_{ji} = r^2 e_0 + (c_{ij} \kappa) \mathbf{1}_H \quad \text{in } \mathbb{Z}[H].$$

Proof. (1) is immediate because row $i$ of $Q$ uses only entries from its own pair multiset $\mathsf{M}_i$. For (2), when $r = 1$ and $1 \le Q_{jw} \le n - 1$, each residue class modulo $n$ (other than $0 \bmod n$) has a unique representative in $\{1, \dots, n - 1\}$; thus $Q_{jw} \equiv -Q_{j0} Q_{0w} \pmod n$ fixes $Q_{jw}$ uniquely as an integer, and permuting $\{1, \dots, m - 1\}$ simply conjugates $Q$ by a permutation matrix $P$, preserving $Q^2 - (\lambda - \mu) Q - (t - \mu) I_m - \mu n J_m = 0$. For (3), applying the canonical ring homomorphism $\pi_H : \mathbb{Z}[G] \to \mathbb{Z}[H]$ (which maps $e_0 \mapsto e_0$ and $\mathbf{1}_G \mapsto |K| \mathbf{1}_H = \kappa \mathbf{1}_H$) to $S_{ij} * S_{ji} = r^2 e_0 + c_{ij} \mathbf{1}_G$ yields $C_{ij} * C_{ji} = r^2 e_0 + (c_{ij} \kappa) \mathbf{1}_H$. $\square$

---

Complete 45-Case Catalog Dichotomy (Verified in Part 4 of supplement_paper3/verify_all.py)

Filtering all 1,094 open DSRG parameter sets in Hobart and Brouwer's catalog [1] ($v \le 110$, stored in supplement_paper3/open_catalog_v110.json) by the exact arithmetic definition of the coclique-orbit rank-1 regime: $$t \ne \mu, \quad r \mid |s|, \quad m = 1 + \frac{|s|}{r} \ge 3, \quad m \mid v \;\;(n = v/m), \quad m_1 = \frac{(m - 1)|s| - k}{r - s} \in \{0, 1, \dots, m - 1\},$$ yields exactly 45 open catalog parameter sets. Executing the four-stage verification in Part 4 of supplement_paper3/verify_all.py (0.049s) establishes:

• Stage 1 (Local Row-Pair & Cross-Minor Row-Sum Bounds, Theorem 3.2): 39 of the 45 open parameter sets have $\mathcal{M} = \emptyset$ (zero local row-pair multisets):

  • $v = 30$: $(30, 11, 9, 2, 5)$
  • $v = 36$: $(36, 19, 17, 8, 12)$
  • $v = 48$: $(48, 9, 5, 0, 2)$, $(48, 19, 14, 5, 9)$, $(48, 25, 20, 11, 15)$
  • $v = 49$: $(49, 29, 26, 15, 20)$, $(49, 22, 18, 7, 12)$
  • $v = 50$: $(50, 11, 7, 0, 3)$
  • $v = 60$: $(60, 16, 9, 2, 5)$
  • $v = 64$: $(64, 33, 27, 14, 20)$
  • $v = 70$: $(70, 29, 20, 9, 14)$, $(70, 36, 27, 16, 21)$
  • $v = 72$: $(72, 25, 15, 6, 10)$
  • $v = 75$: $(75, 16, 8, 1, 4)$
  • $v = 80$: $(80, 33, 23, 10, 16)$, $(80, 41, 31, 18, 24)$
  • $v = 81$: $(81, 55, 50, 35, 42)$, $(81, 46, 38, 23, 30)$
  • $v = 84$: $(84, 22, 13, 2, 7)$
  • $v = 90$: $(90, 25, 13, 4, 8)$, $(90, 37, 26, 11, 18)$, $(90, 46, 35, 20, 27)$
  • $v = 96$: $(96, 25, 15, 2, 8)$, $(96, 41, 27, 14, 20)$, $(96, 49, 35, 22, 28)$
  • $v = 98$: $(98, 22, 12, 1, 6)$, $(98, 29, 16, 5, 10)$, $(98, 36, 21, 10, 15)$
  • $v = 100$: $(100, 16, 7, 0, 3)$, $(100, 71, 65, 48, 56)$, $(100, 61, 51, 34, 42)$, $(100, 51, 39, 22, 30)$
  • $v = 105$: $(105, 36, 20, 9, 14)$
  • $v = 108$: $(108, 19, 9, 0, 4)$, $(108, 28, 17, 2, 9)$, $(108, 73, 67, 46, 56)$, $(108, 37, 23, 8, 15)$, $(108, 64, 50, 35, 42)$
  • $v = 110$: $(110, 56, 43, 24, 33)$

• Stages 2–4 (Global Quotient Closure, Uniqueness, and Subgroup Projection, Theorem 3.3) on the 6 Stage-1 Row-Pair Survivors:

| Stage-1 Survivor $(v, k, t, \lambda, \mu)$ | Spectrum & Orbits $(r, s, m, n, m_1)$ | Stage-1 Surviving Row-Pair Multiset(s) | Global Resolution (Theorem 3.3) | | :--- | :---: | :---: | :--- | | $(54, 19, 13, 4, 8)$ | $r=1, s=-5, m=6, n=9, m_1=1$ | $((1,1), (2,5), (4,7), (5,2), (7,4))$ (and 2 others) | Eliminated (Thm 3.3(2)): Unique $Q$ has $(Q^2 + 4Q)_{0,1} = 86 \ne 72$ | | $(64, 41, 37, 24, 30)$ | $r=1, s=-7, m=8, n=8, m_1=1$ | $((5, 5)^4, (7, 7)^3)$ | EXISTS (Thm 5.2): Unique $Q$ satisfies $Q^2 + 6Q - 7I_8 = 240 J_8$; realized over $\mathbb{Z}_4 \times \mathbb{Z}_8$ | | $(64, 25, 19, 6, 12)$ | $r=1, s=-7, m=8, n=8, m_1=3$ | $((1,1), (3,3)^3, (5,5)^3)$ | Eliminated (Thm 3.3(1)): $Q_{j0}=Q_{0w}=3 \implies Q_{jw} \equiv -9 \equiv 7 \pmod 8$, but $7 \notin \{1, 3, 5\}$ | | $(81, 37, 28, 13, 20)$ | $r=1, s=-8, m=9, n=9, m_1=3$ | $((2,5)^3, (5,2)^3, (8,8)^2)$ | Eliminated (Thm 3.3(2)): Unique $Q$ has $(Q^2 + 7Q)_{0,1} = 139 \ne 180$ | | $(100, 32, 28, 6, 12)$ | $r=2, s=-8, m=5, n=20, m_1=0$ | $((8, 8)^4)$ | Eliminated (Thm 3.3(3)): Projection onto $H_4 = G / \mathbb{Z}_5$ requires $C * D = 4 e_0 + 15 \mathbf{1}_{H_4}$, which has $0$ solutions in $\{0..5\}^4$ ($\sum c_u = 8$) over $\mathbb{Z}_4$ and $\mathbb{Z}_2^2$ | | $(100, 41, 29, 12, 20)$ | $r=1, s=-9, m=10, n=10, m_1=4$ | $((1,1)^2, (3,7)^3, (7,3)^3, (9,9))$ | Eliminated (Thm 3.3(2)): Unique $Q$ has $(Q^2 + 8Q)_{0,3} = 116 \ne 200$ |

Thus 44 of the 45 open parameter sets in the coclique-orbit rank-1 regime are completely obstructed, and the sole surviving parameter set out of all 45 — $\mathrm{DSRG}(64, 41, 37, 24, 30)$ — is explicitly constructed in Section 5!

---

4. Prime-Power Cyclotomic & Residual Convolution Obstructions for Two-Orbit Actions

To analyze two-orbit abelian actions ($m = 2, n = v/2$) where the non-trivial character multiplicities $m_\chi \in \{0, 1, 2\}$ are not forced to be constant by a zero trace, we establish an exact cyclotomic divisibility law on prime-power characters.

Theorem 4.1 (Prime-Power Cyclotomic Multiplicity Law)

Let $A$ be a $\mathrm{DSRG}(v, k, t, \lambda, \mu)$ admitting a semiregular abelian group $G$ of order $n = v/m$ with $m$ orbits and quotient matrix $Q = (Q_{ij})_{0 \le i, j < m}$. Let $\chi \in \hat{G}$ be any character of prime-power order $o(\chi) = p^a$ ($p$ prime, $a \ge 1$). If $m_\chi = 0$ (so $\sigma := s$) or $m_\chi = m$ (so $\sigma := r$), then the quotient matrix $Q$ and parameters $(n, k, \sigma)$ must satisfy: $$Q \equiv \sigma I_m \pmod p, \qquad k \equiv \sigma \pmod p, \qquad -\frac{n}{p^a} \le \sigma \le \frac{n}{p^a} - 1.$$

Proof. Since $\hat{A}(\chi) \in \mathbb{C}^{m \times m}$ is diagonalizable with eigenvalues $r$ (multiplicity $m_\chi$) and $s$ (multiplicity $m - m_\chi$), having $m_\chi = 0$ or $m_\chi = m$ implies $\hat{A}(\chi) = \sigma I_m$, so $\hat{S}_{ij}(\chi) = \sigma \delta_{ij}$. Since $o(\chi) = p^a$, $G / \ker(\chi) \cong \mathbb{Z}_{p^a}$ with $|\ker(\chi)| = n / p^a$. Projecting $S_{ij} \subset G$ onto $\mathbb{Z}_{p^a}$ defines the integer polynomial $P_{ij}(x) = \sum_{u = 0}^{p^a - 1} c_u^{(ij)} x^u \in \mathbb{Z}[x]$ with $0 \le c_u^{(ij)} \le n / p^a$ (and $0 \le c_0^{(ii)} \le n / p^a - 1$ since $e_0 \notin S_{ii}$) satisfying $P_{ij}(\zeta_{p^a}) = \sigma \delta_{ij}$. By Gauss's Lemma, the monic cyclotomic polynomial $\Phi_{p^a}(x) = \sum_{\ell = 0}^{p - 1} x^{\ell p^{a - 1}}$ divides $P_{ij}(x) - \sigma \delta_{ij}$ in $\mathbb{Z}[x]$: $$P_{ij}(x) - \sigma \delta_{ij} = \left( \sum_{w = 0}^{p^{a - 1} - 1} b_w x^w \right) \Phi_{p^a}(x) \qquad (b_w \in \mathbb{Z}).$$ Evaluating at $x = 1$ gives $Q_{ij} - \sigma \delta_{ij} = P_{ij}(1) - \sigma \delta_{ij} \equiv 0 \pmod p$, so $Q \equiv \sigma I_m \pmod p$ and $k \equiv \sigma \pmod p$. Comparing coefficients of $x^0$ and $x^{p^{a - 1}}$ in $P_{ii}(x) - \sigma$ gives $c_0^{(ii)} - \sigma = b_0 = c_{p^{a - 1}}^{(ii)}$, whence $-\frac{n}{p^a} \le c_0^{(ii)} - c_{p^{a - 1}}^{(ii)} = \sigma \le \frac{n}{p^a} - 1$. $\square$

---

Corollary 4.2 (Catalog Audit of Open Two-Orbit Abelian Parameter Pairs: 73 Eliminated Pairs out of 374 $Q$-Feasible Pairs)

Among all 547 open primary parameter pairs in Hobart and Brouwer's catalog [1] ($v \le 110$, audited in supplement_paper3/open_catalog_v110.json), exactly 374 even-order pairs with $t \ne \mu$ admit at least one symmetric $2 \times 2$ integer quotient matrix $Q = \begin{pmatrix} q_{00} & q_{01} \\ q_{01} & q_{00} \end{pmatrix}$ satisfying $0 \le q_{00} \le n - 1$, $1 \le q_{01} \le n$ ($n = v/2$), $q_{00} + q_{01} = k$, $q_{00} - q_{01} = \tau_0 \in \{r, s\}$, and $q_{00}(q_{00} - (r + s)) + r s + q_{01}^2 = \mu n$.

For each of these 374 $Q$-feasible open primary pairs and every abelian group $G$ of order $n = v/2$, partitioning $\hat{G} \setminus \{1_G\}$ into rational Galois orbits $\mathcal{O}$ of size $\varphi(o(\chi))$ and applying the prime-power multiplicity restrictions of Theorem 4.1 to the subset-sum equation $\sum_{\mathcal{O}} m_{\mathcal{O}} |\mathcal{O}| = f - m_1$ (Part 5 of supplement_paper3/verify_all.py, 0.020s) proves:

• 301 of the 374 $Q$-feasible open primary pairs survive Theorem 4.1, including our newly resolved pair $\mathrm{DSRG}(64, 22, 18, 10, 6)$ (which has $Q = \begin{pmatrix} 10 & 12 \\ 12 & 10 \end{pmatrix}$, $\tau_0 = -2 = s$, $m_1 = 0$, and is explicitly realized over $G = \mathbb{Z}_4 \times \mathbb{Z}_8$ in Theorem 5.2).
• Exactly 73 of the 374 $Q$-feasible open primary pairs (146 DSRGs with complements) are completely obstructed across all abelian groups $G$ of order $v/2$ (zero valid character multiplicity partitions exist for any valid $Q$):
  • $v = 28$ (1 pair): $(28, 6, 3, 2, 1)$
  • $v = 30$ (1 pair): $(30, 11, 9, 2, 5)$
  • $v = 44$ (1 pair): $(44, 12, 8, 1, 4)$
  • $v = 50$ (6 pairs): $(50, 12, 9, 2, 3)$, $(50, 14, 7, 6, 3)$, $(50, 17, 12, 5, 6)$, $(50, 19, 10, 9, 6)$, $(50, 21, 14, 7, 10)$, $(50, 23, 16, 11, 10)$
  • $v = 66$ (6 pairs): $(66, 10, 5, 4, 1)$, $(66, 17, 15, 2, 5)$, $(66, 19, 9, 4, 6)$, $(66, 23, 17, 4, 10)$, $(66, 25, 15, 10, 9)$, $(66, 27, 25, 12, 10)$
  • $v = 70$ (12 pairs): $(70, 9, 3, 2, 1)$, $(70, 11, 5, 0, 2)$, $(70, 18, 12, 6, 4)$, $(70, 22, 20, 4, 8)$, $(70, 26, 20, 12, 8)$, $(70, 27, 12, 11, 10)$, $(70, 29, 20, 9, 14)$, $(70, 30, 24, 11, 14)$, $(70, 31, 19, 12, 15)$, $(70, 32, 26, 15, 14)$, $(70, 33, 21, 16, 15)$, $(70, 33, 24, 17, 14)$
  • $v = 76$ (3 pairs): $(76, 17, 13, 6, 3)$, $(76, 18, 6, 5, 4)$, $(76, 20, 10, 3, 6)$
  • $v = 84$ (1 pair): $(84, 20, 11, 10, 3)$
  • $v = 90$ (2 pairs): $(90, 41, 24, 17, 20)$, $(90, 43, 26, 21, 20)$
  • $v = 92$ (4 pairs): $(92, 21, 14, 7, 4)$, $(92, 22, 7, 6, 5)$, $(92, 24, 11, 4, 7)$, $(92, 25, 22, 3, 8)$
  • $v = 98$ (20 pairs): $(98, 17, 15, 2, 3)$, $(98, 18, 15, 4, 3)$, $(98, 20, 9, 8, 3)$, $(98, 22, 12, 1, 6)$, $(98, 24, 18, 5, 6)$, $(98, 25, 18, 7, 6)$, $(98, 27, 12, 11, 6)$, $(98, 27, 17, 16, 4)$, $(98, 30, 20, 7, 10)$, $(98, 31, 22, 9, 10)$, $(98, 33, 20, 13, 10)$, $(98, 34, 16, 15, 10)$, $(98, 37, 25, 12, 15)$, $(98, 38, 27, 14, 15)$, $(98, 40, 25, 18, 15)$, $(98, 41, 21, 20, 15)$, $(98, 43, 27, 16, 21)$, $(98, 44, 31, 18, 21)$, $(98, 46, 33, 22, 21)$, $(98, 47, 31, 24, 21)$
  • $v = 102$ (7 pairs): $(102, 16, 6, 5, 2)$, $(102, 29, 19, 6, 9)$, $(102, 31, 13, 8, 10)$, $(102, 32, 24, 14, 8)$, $(102, 35, 21, 8, 14)$, $(102, 37, 19, 14, 13)$, $(102, 39, 29, 16, 14)$
  • $v = 104$ (4 pairs): $(104, 11, 7, 2, 1)$, $(104, 14, 4, 1, 2)$, $(104, 36, 24, 13, 12)$, $(104, 42, 36, 15, 18)$
  • $v = 108$ (1 pair): $(108, 28, 12, 5, 8)$
  • $v = 110$ (4 pairs): $(110, 46, 40, 15, 22)$, $(110, 51, 29, 22, 25)$, $(110, 52, 46, 27, 22)$, $(110, 53, 31, 26, 25)$

---

Theorem 4.3 (Alternative Group-Ring Residual Convolution & Parity Proof on $\mathrm{DSRG}(40, 9, 6, 5, 1)$)

(Comparison with Paper 1 [3]: In [3, Table 2, family 4], we proved unrestricted nonexistence of $\mathrm{DSRG}(40, 9, 6, 5, 1)$ and its $t = \mu$ complement $\mathrm{DSRG}(40, 30, 27, 21, 27)$ via Rank-6 Farkas hyperplane duality on the complement. Because $(40, 9, 6, 5, 1)$ is one of the 301 pairs surviving the pure character-multiplicity filter of Corollary 4.2, we show here how residual group-ring convolution directly eliminates all two-orbit abelian realizations of $(40, 9, 6, 5, 1)$ without invoking LP/Farkas duality.)

The parameter set $\mathrm{DSRG}(40, 9, 6, 5, 1)$ ($r = 5, s = -1, d = 6, f = 5, g = 34$) and its complement $\mathrm{DSRG}(40, 30, 27, 21, 27)$ do not admit any two-orbit realization over any abelian group $G$ of order $n = 20$ ($G \cong \mathbb{Z}_{20} \cong \mathbb{Z}_4 \times \mathbb{Z}_5$ or $G \cong \mathbb{Z}_2 \times \mathbb{Z}_{10} \cong \mathbb{Z}_2^2 \times \mathbb{Z}_5$).

Proof. In any two-orbit abelian realization over $G$ ($|G| = 20$), the $(0, 0)$ block equation in $\mathbb{Z}[G]$ is: $$S_{00}^2 + S_{01} * S_{10} = 4 S_{00} + 5 e_0 + \mathbf{1}_G.$$ For every $g \in G$, let $c_g := (S_{01} * S_{10})_g = |S_{01} \cap (g - S_{10})|$, which satisfies $0 \le c_g \le \min(|S_{01}|, |S_{10}|) = q_{01}$. Thus $S_{00} \subset G \setminus \{0\}$ ($|S_{00}| = q_{00}$) must satisfy the Residual Convolution Bounds: $$0 \le 6 - (S_{00}^2)_0 \le q_{01}, \qquad \text{and} \qquad 0 \le (\lambda \mathbf{1}_{g \in S_{00}} + \mu \mathbf{1}_{g \notin S_{00}}) - (S_{00}^2)_g \le q_{01} \quad (\forall g \in G \setminus \{0\}).$$ For $\mathrm{DSRG}(40, 9, 6, 5, 1)$, the $2 \times 2$ quotient equation $(9 - 2 q_{01})^2 - 4(9 - 2 q_{01}) - 5 = 0$ has two integer roots:

1. $Q = \begin{pmatrix} 7 & 2 \\ 2 & 7 \end{pmatrix}$ ($q_{00} = 7, q_{01} = 2$): Exhaustively testing all $\binom{19}{7} = 50,388$ subsets $S_{00} \subset G \setminus \{0\}$ in both $\mathbb{Z}_4 \times \mathbb{Z}_5$ and $\mathbb{Z}_2^2 \times \mathbb{Z}_5$ (Part 6 of supplement_paper3/verify_all.py) shows that zero subsets $S_{00}$ satisfy the residual convolution bounds $0 \le (5 \mathbf{1}_{g \in S_{00}} + \mathbf{1}_{g \notin S_{00}}) - (S_{00}^2)_g \le 2$. (From the character perspective, Theorem 4.1 at $p = 5$ forbids $m_5 \in \{0, 2\}$ because $q_{01} = 2 \not\equiv 0 \pmod 5$, forcing $m_5 = 1$ on all $\varphi(5) = 4$ characters of order $5$; since $f - m_1 = 5 - 1 = 4$, all 15 remaining non-trivial characters $\chi \notin \langle \chi_5 \rangle$ must have $m_\chi = 0$, which forces the $H_4$-fiber sizes of $S_{01}$ to satisfy $4 c_{01}(0) = q_{01} = 2$, an immediate parity contradiction.)
2. $Q = \begin{pmatrix} 4 & 5 \\ 5 & 4 \end{pmatrix}$ ($q_{00} = 4, q_{01} = 5$): Testing all $\binom{19}{4} = 3,876$ subsets of size $4$ in $\mathbb{Z}_4 \times \mathbb{Z}_5$ and $\mathbb{Z}_2^2 \times \mathbb{Z}_5$ yields a unique subset satisfying the residual convolution bounds: $S_{00} = S_{11} = H_5 \setminus \{0\}$ (the $4$ non-zero elements of the unique subgroup $H_5 \le G$ of order $5$). However, substituting $S_{00} = S_{11} = H_5 \setminus \{0\}$ into the off-diagonal trace equation $(S_{00} + S_{11} - 4 e_0) * S_{01} = \mu \mathbf{1}_G = \mathbf{1}_G$ gives: $$(2 S_{00} - 4 e_0) * S_{01} = \mathbf{1}_G.$$ Every coefficient of $2 S_{00} - 4 e_0 \in 2 \mathbb{Z}[G]$ is even, whereas every coefficient of $\mathbf{1}_G$ is $\mu = 1$ (odd), a contradiction modulo $2$. $\square$

---

5. Constructive Resolution: Coordinate-Cross Wrap-Around Difference Pair Synthesis of $\mathrm{DSRG}(64, 22, 18, 10, 6)$

Guided by the unique surviving $8 \times 8$ coclique-orbit Schur-complement solution of Theorem 3.3 at $v = 64$, we now resolve the open catalog parameter pair $\mathrm{DSRG}(64, 22, 18, 10, 6)$ and $\mathrm{DSRG}(64, 41, 37, 24, 30)$.

Theorem 5.1 (Wrap-Around Difference Pairs in Cyclic Groups $\mathbb{Z}_{a^2 - 1}$)

(Compare the base-expansion and interval/progression convolution techniques used for $t = \mu$ and $v = 2p$ families in [8, Section 9] and [9, Section 3].) For any integer $a \ge 2$, let $K = \mathbb{Z}_{a^2 - 1}$ be the cyclic group of order $n_K = a^2 - 1$. Define the two subsets of size $a$: $$B_{01} := \{0, 1, 2, \dots, a - 1\} \subset \mathbb{Z}_{a^2 - 1}, \qquad B_{10} := \{0, a, 2a, \dots, (a - 1)a\} \subset \mathbb{Z}_{a^2 - 1}.$$ Then in the integral group ring $\mathbb{Z}[\mathbb{Z}_{a^2 - 1}]$, the convolution product of $B_{01}$ and $B_{10}$ satisfies: $$B_{01} * B_{10} = \mathbf{1}_K + e_0.$$

Proof. By uniqueness of base-$a$ representations in $\mathbb{Z}$, as $x$ ranges over $\{0, 1, \dots, a - 1\}$ and $y$ ranges over $\{0, 1, \dots, a - 1\}$, the integer sum $x + a y$ takes each integer value in $\{0, 1, 2, \dots, a^2 - 1\}$ element-wise exactly once. When reduced modulo $a^2 - 1$:

• The two endpoints $0 = 0 + a \cdot 0$ and $a^2 - 1 = (a - 1) + a(a - 1)$ both reduce to $0 \pmod{a^2 - 1}$, so $e_0$ has coefficient $2$.
• Every non-zero residue $u \in \{1, 2, \dots, a^2 - 2\}$ is achieved by a unique pair $(x, y)$, so $e_u$ has coefficient $1$. Therefore $B_{01} * B_{10} = 2 e_0 + \sum_{u = 1}^{a^2 - 2} e_u = \mathbf{1}_K + e_0$. $\square$

---

Theorem 5.2 (Existence of Open $\mathrm{DSRG}(64, 22, 18, 10, 6)$ and Complement $\mathrm{DSRG}(64, 41, 37, 24, 30)$)

Let $G = H \times K = \mathbb{Z}_4 \times \mathbb{Z}_8$ ($|H| = 4$, $|K| = 8$, $|G| = 32$). Identify $H$ with $\mathbb{Z}_4 \times \{0\}$ and $K$ with $\{0\} \times \mathbb{Z}_8$, so $\mathbf{1}_G = H * K$ in $\mathbb{Z}[G]$. Define the four group-ring blocks $S_{00}, S_{11}, S_{01}, S_{10} \subset G$:

1. Coordinate-Cross (Rook's Graph) Diagonal Blocks ($|S_{00}| = |S_{11}| = 10$): $$S_{00} = S_{11} := \big(\{0\} \times (\mathbb{Z}_8 \setminus \{0\})\big) \cup \big((\mathbb{Z}_4 \setminus \{0\}) \times \{0\}\big) = H + K - 2 e_0.$$
2. Inflated Wrap-Around Difference Pair Off-Diagonal Blocks ($|S_{01}| = |S_{10}| = 12$, $a = 3$): $$S_{01} := \mathbb{Z}_4 \times \{0, 1, 2\} = H * B_{01}, \qquad S_{10} := \mathbb{Z}_4 \times \{0, 3, 6\} = H * B_{10}.$$ Then the $64 \times 64$ two-orbit Cayley block matrix $$A = \begin{pmatrix} \mathrm{Cay}(G, S_{00}) & \mathrm{Cay}(G, S_{01}) \\ \mathrm{Cay}(G, S_{10}) & \mathrm{Cay}(G, S_{11}) \end{pmatrix} \in \{0, 1\}^{64 \times 64}$$ is the adjacency matrix of an exact $\mathrm{DSRG}(64, 22, 18, 10, 6)$ ($r = 6, s = -2, f = 13, g = 50$), and its complement $A^c := J_{64} - I_{64} - A$ is an exact $\mathrm{DSRG}(64, 41, 37, 24, 30)$ ($r = 1, s = -7, f = 50, g = 13$).

Proof. First, $e_0 = (0, 0) \notin S_{00} = S_{11}$, so $A$ has zero diagonal and entries in $\{0, 1\}$. Every row and column of $A$ has sum $|S_{00}| + |S_{01}| = 10 + 12 = 22 = |S_{10}| + |S_{11}|$, so $A \mathbf{1}_{64} = 22 \mathbf{1}_{64}$ and $\mathbf{1}_{64}^T A = 22 \mathbf{1}_{64}^T$.

To prove $A^2 - 4 A - 12 I_{64} = (A - 6 I_{64})(A + 2 I_{64}) = 6 J_{64}$, we verify the four $2 \times 2$ block equations in $\mathbb{Z}[G]$, using $H^2 = 4 H$, $K^2 = 8 K$, and $H * K = \mathbf{1}_G$:

1. Diagonal $(0, 0)$ and $(1, 1)$ Block Equations: Since $S_{00} = S_{11} = H + K - 2 e_0$, the shifted diagonal quadratic term is: $$\begin{aligned} (S_{00} - 6 e_0) * (S_{00} + 2 e_0) &= (H + K - 8 e_0) * (H + K) \\ &= H^2 + 2 H * K + K^2 - 8 H - 8 K \\ &= 4 H + 2 \mathbf{1}_G + 8 K - 8 H - 8 K = 2 \mathbf{1}_G - 4 H. \end{aligned}$$ Here $8 K - 8 K = 0$ cancels the $K$-fiber excess, leaving $-4 H$. Next, by Theorem 5.1 with $a = 3$ ($B_{01} * B_{10} = K + e_0$): $$S_{01} * S_{10} = (H * B_{01}) * (H * B_{10}) = H^2 * (B_{01} * B_{10}) = 4 H * (K + e_0) = 4 \mathbf{1}_G + 4 H.$$ Adding the diagonal and off-diagonal terms cancels $-4 H$ and $+4 H$ identically: $$(S_{00} - 6 e_0) * (S_{00} + 2 e_0) + S_{01} * S_{10} = (2 \mathbf{1}_G - 4 H) + (4 \mathbf{1}_G + 4 H) = 6 \mathbf{1}_G = \mu \mathbf{1}_G.$$
2. Off-Diagonal $(0, 1)$ and $(1, 0)$ Cross-Trace Equations: Since $S_{00} + S_{11} - 4 e_0 = 2(H + K - 2 e_0) - 4 e_0 = 2 H + 2 K - 8 e_0$, and $(2 H - 8 e_0) * H = 2(4 H) - 8 H = 0$ annihilates $H$, we have for both $S_{01} = H * B_{01}$ and $S_{10} = H * B_{10}$ (since $|B_{01}| = |B_{10}| = 3$, so $K * B_{ij} = 3 K$): $$(S_{00} + S_{11} - 4 e_0) * S_{ij} = (2 H - 8 e_0 + 2 K) * (H * B_{ij}) = \underbrace{((2 H - 8 e_0) * H)}_{= 0} * B_{ij} + 2 H * (K * B_{ij}) = 6 H * K = 6 \mathbf{1}_G = \mu \mathbf{1}_G.$$

Thus $(A - 6 I_{64})(A + 2 I_{64}) = 6 J_{64}$, i.e., $A^2 = 4 A + 12 I_{64} + 6 J_{64}$, proving that $A$ is an exact $\mathrm{DSRG}(64, 22, 18, 10, 6)$ ($t - \mu = 12, \lambda - \mu = 4, \mu = 6 \implies t = 18, \lambda = 10, \mu = 6$). By complementation, $A^c = J_{64} - I_{64} - A$ is an exact $\mathrm{DSRG}(64, 41, 37, 24, 30)$.

Finally, partitioning the $64$ vertices into the $m = 8$ cosets of $K = \{0\} \times \mathbb{Z}_8$ (indexed by $(i, h) \in \{0, 1\} \times \mathbb{Z}_4$), every coset $(i, h) + K$ induces a $K_8$ clique in $A$ (since $K \setminus \{0\} \subset S_{ii}$) and therefore a coclique in $A^c = \mathrm{DSRG}(64, 41, 37, 24, 30)$ whose $8 \times 8$ quotient matrix $Q$ has row/column pair multiset $((5, 5)^4, (7, 7)^3)$, realizing the unique survivor of Theorem 3.3! $\square$

---

Proposition 5.3 (Exact Parametric Coordinate-Cross Balance System & Uniqueness of $(a, h, n_K) = (3, 4, 8)$)

Let $G = H \times K$ with $|H| = h \ge 2$ and $K = \mathbb{Z}_{n_K}$ where $n_K = a^2 - 1$ ($a \ge 2$). Consider the binary Coordinate-Cross and Wrap-Around Difference Pair ansatz: $$S_{00} = S_{11} := (H \setminus \{0\}) \cup (K \setminus \{0\}) = H + K - 2 e_0 \in \{0, 1\}[G],$$ $$S_{01} := H * B_{01} \in \{0, 1\}[G], \qquad S_{10} := H * B_{10} \in \{0, 1\}[G],$$ where $B_{01} = \{0, 1, \dots, a - 1\}$ and $B_{10} = \{0, a, \dots, (a - 1)a\}$ satisfy $B_{01} * B_{10} = K + e_0$ (Theorem 5.1). Then the resulting $2 h n_K \times 2 h n_K$ matrix $A$ satisfies $(A - r I)(A - s I) = \mu J$ for restricted eigenvalues $r > 0 > s$ if and only if $(a, h, n_K, r, s, \mu) = (3, 4, 8, 6, -2, 6)$, i.e., $\mathrm{DSRG}(64, 22, 18, 10, 6)$.

Proof. Using $H^2 = h H$, $K^2 = n_K K$, $H * K = \mathbf{1}_G$, and $S_{01} * S_{10} = h H * (K + e_0) = h \mathbf{1}_G + h H$, we expand the diagonal and off-diagonal block equations of $(A - r I)(A - s I) = \mu J$ in $\mathbb{Z}[G]$: $$\begin{aligned} (S_{00} - r e_0) * (S_{00} - s e_0) + S_{01} * S_{10} &= \big(H + K - (r + 2) e_0\big) * \big(H + K - (s + 2) e_0\big) + h \mathbf{1}_G + h H \\ &= (2 h - r - s - 4) H + (n_K - r - s - 4) K + (r + 2)(s + 2) e_0 + (h + 2) \mathbf{1}_G, \end{aligned}$$ $$\begin{aligned} (S_{00} + S_{11} - (r + s) e_0) * S_{01} &= \big(2 H + 2 K - (r + s + 4) e_0\big) * (H * B_{01}) \\ &= (2 h - r - s - 4) (H * B_{01}) + 2 a \mathbf{1}_G. \end{aligned}$$ Since $\{e_0, H \setminus \{0\}, K \setminus \{0\}, G \setminus (H \cup K)\}$ partition $G$ (and $a < n_K = a^2 - 1$ for $a \ge 2$), the coefficients of $e_0$, $K$, $H$, and $H * B_{01}$ vanish while the coefficient of $\mathbf{1}_G$ equals $\mu$ in both equations if and only if the five exact balance equations hold: $$(r + 2)(s + 2) = 0, \qquad n_K - r - s - 4 = 0, \qquad 2 h - r - s - 4 = 0, \qquad \mu = h + 2, \qquad \mu = 2 a.$$ Since $r > 0$, $(r + 2)(s + 2) = 0$ uniquely forces $s = -2$. Substituting $s = -2$ into the second and third equations yields $r = n_K - 2 = a^2 - 3$ and $2 h = r + 2 = n_K = a^2 - 1$, so $h = \frac{a^2 - 1}{2}$. Finally, equating $\mu = h + 2 = 2 a$ gives: $$\frac{a^2 - 1}{2} + 2 = 2 a \quad \Longleftrightarrow \quad a^2 - 4 a + 3 = (a - 1)(a - 3) = 0.$$ Since $a \ge 2$, the unique solution is $a = 3$, which yields $n_K = 8, h = 4, r = 6, s = -2, \mu = 6, k = (h - 1) + (n_K - 1) + h a = 3 + 7 + 12 = 22$, recovering $\mathrm{DSRG}(64, 22, 18, 10, 6)$ (verified in Part 2 of supplement_paper3/verify_all.py). $\square$

---

6. Verification and Reproducibility

The complete verification suite supplement_paper3/verify_all.py and catalog audit supplement_paper3/open_catalog_v110.json are self-contained, use only the Python 3 standard library, and perform 100% exact arbitrary-precision integer arithmetic (zero floating-point operations):

python3 supplement_paper3/verify_all.py

Verification Results Summary Table

| Part | Tested Theorem / Object | Verification Method | Time | Status | | :--- | :--- | :--- | :---: | :---: | | Catalog | open_catalog_v110.json (1,094 open rows, 547 primary pairs) | Exact audit of all DSRG parameter & spectral identities | 0.002s | 100% Passed | | Part 1 | Theorem 5.2: $\mathrm{DSRG}(64, 22, 18, 10, 6)$ & $\mathrm{DSRG}(64, 41, 37, 24, 30)$ | Exact $O(v^3)$ bitset $(A^2)_{ij}$ + $8 \times 8$ quotient $Q$ check | 0.006s | 100% Passed | | Part 2 | Theorem 5.1 & Proposition 5.3: Difference Pairs & Balance System | Exact convolution $B_{01} * B_{10} = \mathbf{1}_K + e_0$ ($a=2..6$) & $(a-1)(a-3)=0$ | 0.000s | 100% Passed | | Part 3 | Theorems 2.1, 2.2 & Proposition 2.3: Spectral-Shift & Binary Span | Bareiss solver ($\sigma = s, r$) + $2^{n_s}$ binary-span enumeration on $v=36, 64$ | 0.373s | 100% Passed | | Part 4 | Theorems 3.1, 3.2 & 3.3: Coclique-Orbit 45-Case Dichotomy | Derives 45 open rows from catalog; certifies 44 obstructed & 1 exists ($v=64$) | 0.049s | 100% Passed | | Part 5 | Theorem 4.1 & Corollary 4.2: Two-Orbit Prime-Power Audit | Derives 374 $Q$-feasible open pairs; certifies 73 pairs (146 DSRGs) obstructed | 0.020s | 100% Passed | | Part 6 | Theorem 4.3: Residual Convolution & Parity on $\mathrm{DSRG}(40, 9, 6, 5, 1)$ | Exhaustive subset/parity verification over $\mathbb{Z}_4 \times \mathbb{Z}_5$ and $\mathbb{Z}_2^2 \times \mathbb{Z}_5$ | 0.186s | 100% Passed | | Total | Full Verification Suite (Catalog + Parts 1–6) | 100% Exact Integer Arithmetic (Zero Floats) | 0.633s | All Passed |

---

References

[1] A. E. Brouwer and S. A. Hobart, Parameters of directed strongly regular graphs, https://homepages.cwi.nl/~aeb/math/dsrg/dsrg.html (mirrored in data_tables/dsrg.html, audited in supplement_paper3/open_catalog_v110.json).

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

[3] kirkman_loom, Nonexistence of 42 Open Directed Strongly Regular Graphs via Rank-6 and Rank-7 Hyperplane Duality, Existence of DSRG(64,14,7,3,3) and DSRG(80,18,8,4,4), and Closed-Form Abelian Group-Ring Families, rStation, strongly_regular_graphs_5, version 1 (2026).

[4] kirkman_loom, Constructive Hyperplane Duality for Directed Strongly Regular Graphs: Outer-Product Synthesis of 38 Open Parameter Sets via Hadamard-Simplex, Root-Difference Simplex, Neutral-Pivot, and Parity-Gadget Looms, rStation, strongly_regular_graphs_8, version 2 (2026).

[5] L. Martínez, Strongly regular $m$-Cayley circulant graphs and digraphs, Ars Mathematica Contemporanea 8 (2015), 195–213.

[6] M. Fiedler, M. Klin, and M. Muzychuk, Small vertex-transitive directed strongly regular graphs, Discrete Math. 255 (2002), 87–115.

[7] Š. Gyürki and M. Klin, Sporadic examples of directed strongly regular graphs obtained with the help of computer algebra experimentation, preprint.

[8] Mira Loom, DSRG existence: all feasible $(v, k, \mu, \mu - 1, \mu)$; ..., rStation, strongly_regular_graphs_9, version 1 (2026), Section 9.

[9] teichmuller_lift, Existence of an infinite family of directed strongly regular graphs on $2p$ vertices for primes $p = n^2 + 4$, including $(58, 26, 18, 11, 12)$ and $(106, 49, 35, 22, 23)$, rStation, strongly_regular_graphs_10, version 3 (2026), Section 3.