Higmanian Association Scheme: Structure & Fusion

• Higmanian association schemes are symmetric, imprimitive, and indecomposable rank 5 structures characterized by two nontrivial parabolics.
• Their Bose–Mesner algebra framework and spectral properties enable explicit computation of intersection numbers, eigenvalues, and multiplicities.
• Fusion constructions from these schemes yield symmetric divisible design graphs, offering new infinite families and combinatorial applications.

A Higmanian association scheme is a symmetric, imprimitive, and indecomposable association scheme of rank 5 with a distinctive configuration of equivalence relations (“parabolics”). These schemes play a central role in algebraic combinatorics due to their structure and fusion properties, notably generating infinite families of symmetric divisible design graphs (DDGs). The defining features and classification of Higmanian schemes originate from Higman’s scheme-theoretic generalization of distance-regular graphs, while modern treatments clarify their algebraic and combinatorial context and offer explicit constructions and parameterizations.

1. Definition and Structural Properties

A Higmanian association scheme is an association scheme $\mathcal{X} = (V, S)$, consisting of a finite set $V$ and a partition $S = \{s_0, s_1, s_2, s_3, s_4\}$ of $V^2$ into symmetric basis relations. By definition:

• $s_0$ is the diagonal relation: $s_0 = 1_V = \{(x, x): x \in V\}$.
• Each $s_i$ is symmetric: $s_i^* = s_i$.
• The intersection numbers $$c_{rs}^t = |\{z \in V: (x, z) \in r, (z, y) \in s\}|$$ for $(x, y) \in t$ depend only on the triple $(r,s,t)$, not on $(x, y)$.

A scheme $\mathcal{X}$ is Higmanian if and only if:

1. The rank is 5, i.e., $|S| = 5$.
2. It is symmetric and imprimitive.
3. It is indecomposable (not a wreath product).
4. There are exactly two nontrivial parabolics, explicitly $e_0 = s_0 \cup s_1$ and $e_1 = s_0 \cup s_1 \cup s_2$, while $s_3, s_4$ have trivial radical (no further parabolic structure).

This structural configuration ensures that the quotients $\mathcal{X}_{V/e_i}$ and restrictions to classes $\mathcal{X}_U$ have small rank, matching the classic “Class I–II” cases described by Higman (Ryabov, 26 Jan 2026).

2. Algebraic Framework: Bose–Mesner Algebra

For a Higmanian association scheme, the Bose–Mesner algebra $$\mathcal{A} = \mathrm{Span}_\mathbb{C}\{A_0, \ldots, A_4\}$$, with $A_i$ the adjacency matrix of $s_i$, is a central object:

• The algebra is commutative, semisimple, and closed under matrix multiplication: $A_iA_j = \sum_{k=0}^4 c_{ij}^k A_k$.
• There exists a second distinguished basis of primitive idempotents $E_0, \ldots, E_4$, with $A_i = \sum_{j=0}^4 \theta_j^{(i)} E_j$. Here, $\theta_j^{(i)}$ are the eigenvalues of $A_i$ and the rank of $E_j$ is $m_j$, the multiplicity.
• The standard orthogonality relations among the eigenvalues and multiplicities are:

$$\sum_{i=0}^4 m_j\,\theta_j^{(i)} \theta_{j'}^{(i)} = \delta_{jj'}|V|,\qquad \sum_{j=0}^4 m_j\,\theta_j^{(i)} \theta_j^{(i')} = \delta_{ii'}\, n_i\,|V|.$$

• The valencies $n_i$ are given by $n_i = \deg(A_i)$ and the intersection numbers, eigenvalues, and multiplicities can be calculated or read off from established intersection matrices.

These algebraic invariants are essential for analyzing the combinatorial and spectral properties of Higmanian schemes and guide the process of scheme fusion and construction of related graphs.

3. Fusion Schemes and Divisible Design Graphs

Higmanian association schemes have particularly rich fusion behavior, allowing the construction of symmetric divisible design graphs (DDGs) through the union of basis relations. For a DDG $\Gamma = (V, E)$, the defining properties are:

• $k$–regularity on $v$ vertices.
• An equitable partition of $V$ into $m$ classes of size $n$ such that:
  • Each pair in the same class has $\lambda_1$ common neighbors.
  • Each pair in different classes has $\lambda_2$ common neighbors.

The main result ((Ryabov, 26 Jan 2026), Theorem 1) asserts:

• Given a Higmanian scheme $(V, S)$ with $S = \{s_0, \ldots, s_4\}$, consider $E_i = s_2 \cup s_i$ for $i \in \{3,4\}$. Then $\Gamma_i = (V, E_i)$ is a DDG if and only if:
  1. $c_{33}^3 = c_{33}^4$, and
  2. either

  $\frac{1}{n_3} + \frac{1}{n_4} = \frac{1}{n_1} - \frac{1}{n_1+1}\qquad\text{(canonical partition: classes of $e_1$)}$

  or

  $\frac{n_2}{n_1+1} - \frac{2 n_i}{n_{7-i}} = 1\qquad\text{(canonical partition: classes of $e_0$)}.$

• Exactly one condition must hold for a proper DDG; if both simultaneously hold, the structure reduces to that of a strongly regular graph.

The resulting design parameters $(v, k, \lambda_1, \lambda_2, m, n)$ are given by explicit polynomial expressions in the valencies $n_i$ ((Ryabov, 26 Jan 2026), Lemma 3.7).

4. Classification and Known Constructions

Higman’s original tables ((Ryabov, 26 Jan 2026), citing [Linear Alg. Appl. 226–228 (1995)]) and the subsequent work of Klin, Muzychuk, and Ziv-Av (Michigan Math. J. 58 (2009)) provide feasible intersection matrices for rank 5 schemes up to $|V| \leq 40$ and classify possible parameter tuples $(\alpha, \beta, \gamma, \delta, \varepsilon)$, dictating the schemes’ “shape.” However:

• No complete classification is known for arbitrary $|V|$.
• All known schemes admit precisely two nontrivial parabolics $e_0$ and $e_1$ and fit one of two parameter “shapes.”

Established infinite families of Higmanian schemes and their DDG fusions include:

• Kabanov–Shalaginov family $\mathcal{K}_1$: Schemes constructed from the semidirect product $$G = \mathbb{F}_{q^r}^+ \rtimes \langle \tau \rangle$$, with fusions satisfying the $e_1$-fusion equation.
• Weighing-matrix family $\mathcal{K}_2$: Based on symmetric weighing matrices of type $(4t, 4(t-1))$ and 2×2 block inflations, yielding $e_0$-fusion DDGs.
• Families from divisible difference sets in dihedral and abelian groups: New infinite Cayley DDGs are created from suitable difference sets and intersection conditions, with parameters determined by fusion equations.

These constructions subsume previously known examples and demonstrate the versatility of Higmanian schemes in generating combinatorial structures of interest.

5. Equitable Partitions and the Godsil–Higman Criterion

Equitable partitions in association schemes are characterized by invariance of the characteristic matrix’s column space under the Bose–Mesner algebra. The Godsil–Higman necessary condition provides a spectral integrality test for the existence of such partitions:

• For a symmetric scheme, an equitable partition with projection $F$ onto the partition’s column space must satisfy:

$$(F, E_j) = \frac{1}{v} \sum_{i=0}^d \frac{P_{j i}}{v_i} \operatorname{tr}(N_i) \in \mathbb{N}_0$$

for all $j=0,\ldots,d$, where $P$ is the first eigenmatrix and $N_i$ are the quotient matrices ((Gavrilyuk et al., 2013), Theorem 2).

• This “Higman criterion” ensures that the multiplicities $m_j$ of the eigenvalues in the quotient matrix spectra are nonnegative integers but does not impose the upper bound $m_j \leq f_j$ (the dimension of the corresponding eigenspace).

In comparison, Lloyd’s theorem states that the characteristic polynomial of each quotient matrix $N_i$ divides that of the original $A_i$, hence $m_j \leq f_j$ for all multiplicities. Thus, the Godsil–Higman condition is strictly weaker and does not, by itself, preclude spurious cases where $m_j > f_j$ may occur (Gavrilyuk et al., 2013).

For Higmanian schemes, this observation means that spectral (Godsil–Higman) obstructions are insufficient for classification or for ruling out most non-existence of equitable partitions. Further restrictions must be derived from combinatorial data or stronger algebraic criteria.

6. Families of Divisible Design Graphs from Higmanian Schemes

Fusion constructions from Higmanian schemes yield wide classes of symmetric divisible design graphs (DDGs). Parameter formulas for these DDGs depend on the choice of fusion and are determined by the aforementioned equations on valencies and intersection numbers. Well-known constructions include:

• Fusions from Kabanov–Shalaginov type schemes with explicit parameters:

$$v = q^r \frac{q^r-1}{q-1},\quad k_3 = q^{r-1} \left( \frac{q^r-1}{q-1} + q-2 \right ),\quad m = \frac{q^r-1}{q-1},\quad n = q^r$$

and associated $\lambda_1$, $\lambda_2$ found via the intersection matrices ((Ryabov, 26 Jan 2026), Section 5.1).

• DDGs from weighing matrices and relative difference sets in abelian or generalized dihedral groups, e.g., $v=8t$, $k=4t+2$, $m=4t$, $n=2$, $(\lambda_1, \lambda_2) = (6, 2t+2)$ for the weighing-matrix family ((Ryabov, 26 Jan 2026), Section 5.2).

Explicit constructions provide new infinite families as well as unification of existing classes of DDGs within the framework of Higmanian scheme fusions.

7. Summary and Implications

Higmanian association schemes, through their unique intersection structure and fusion properties, provide a systematic route to the construction and analysis of symmetric divisible design graphs. Their algebraic invariants, structural constraints, and fusion theorems unify several streams of combinatorial design theory and group ring constructions.

A plausible implication is that further advances in the classification and construction of Higmanian schemes, as well as a deeper understanding of associated fusion algebras, will continue to influence the development of algebraic combinatorics, particularly by providing new families of fusions and sharper criteria for the existence of combinatorial configurations such as DDGs.

For equitable partitions of Higmanian schemes, only the full strength of Lloyd’s theorem provides rigorous obstructions; spectral integrality alone (via the Godsil–Higman criterion) is not adequate for classification or nonexistence results in this context (Gavrilyuk et al., 2013). Future progress will likely require new algebraic or combinatorial tools beyond existing spectral conditions.