A generalized quadrangle is an incidence structure of points and lines such that

These axioms are self-dual, and therefore the incidence structure dual to a $GQ$ is again a $GQ$.

The smallest interesting example has as points the 15 edges of $K_6$ and as lines the 15 1-factors of $K_6$; each edge is incident with the three 1-factors that contain it. A $GQ$ is thick if each point is on at least three lines and each line contains at least three points. If a $GQ$ is thick then it is point and line regular, this means there are integers $s$ and $t$ such that each point lies on exactly $t+1$ lines and each line contains exactly $s+1$ points. Our example above is a $GQ(2,2)$, traditionally denoted by $W(2)$.

Associated to an incidence structure we have a point graph, a line graph and an incidence graph. For $W(2)$ the point and line graphs are isomorphic to $L(K_6)$ (which is very easy to check); the incidence graph is a bipartite cubic graph on 30 vertices with diameter four and girth eight. It is known as Tutte's 8-cage. For a thick $GQ$, both the point and line graphs are strongly regular.

We describe how to construct a family of $GQ$'s denoted by $W(q)$, where $q$ is a primes power. We will accompany the description with the code for $W(3)$.

Let $V$ be the vector space of dimension four over $GF(q)$. Its 1-dimensional subspaces will be the points of our generalized quadrangle. To define the lines we want a non-degenerate alternating form on $V$; this is given by an invertible matrix $S$ with diagonal entries zero such that $S+S^T=0$. (So if $q$ is odd, then $S$ is skew symmetric; if $q$ is even it's symmetric with zero diagonal.) A subspace $U$ of $V$ is isotropic if \[ u^THv = 0 \] for all $u$, $v$ in $U$. All 1-dimensional subspaces of $V$ are isotropic and the lines of $W(q)$ will be the 2-dimensional isotropic subspaces.

Time for some actual work. We define our form:

def beta(u,v):
    return u[0]*v[2]+u[1]*v[3] -u[2]*v[0] -u[3]*v[1]

and create our points and lines:

V = VectorSpace( GF(3), 4)
pnts = [u[1] for u in V.subspaces(1)]
lines = [sub for sub in V.subspaces(2)\
    if beta(sub.matrix()[0],sub.matrix()[1])==0]

Two points $u$ and $v$ are collinear if $\beta(u,v)=0$. Two lines $L$ and $M$ are incident if

(L != M) and (L.intersection(M) != V.zero_subspace()) # not tested

or if they are not equal and

det( L.matrix().stack(M.matrix)) == 0 # not tested

Elements of our vector space $V$ are "mutable", so not hashable, and therefore cannot be used as vertices of a graph. This is easily circumvented:

adj = lambda i,j: beta(pnts[i],pnts[j])==0
W3 = Graph([range(len(pnts)),adj], loops=False)

We can check that $W3$ is connected and regular, and that it has exactly three eigenvalues (obvious in the factored characteristic polynomial):

W3.is_connected(), W3.is_regular(), W3.am().fcp()

(True, True, (x - 12) * (x + 4)^15 * (x - 2)^24)

The lines of the $GQ$ correspond to the cliques of maximal size, which we can find by

cliques = W3.cliques_maximum()

We get the point graph of the dual $GQ$ by

adj = lambda i,j: i != j and det( lines[i].matrix().stack(lines[j].matrix())) == 0
W3d = Graph( [range(len(lines)), adj])

As expected $W3d$ is not isomorphic to $W3$, but it is strongly regular with the same parameters.

W3.is_isomorphic(W3d)

False

dp = distance_partition( W3, W3.vertices()[0])
W3.is_equitable( dp, quotient_matrix=True) 

[ 0 12  0]
[ 1  2  9]
[ 0  4  8]

dp = distance_partition( W3d, W3d.vertices()[0])
W3d.is_equitable( dp, quotient_matrix=True) 

[ 0 12  0]
[ 1  2  9]
[ 0  4  8]