Let $D$ be the incidence matrix of an orientation of the graph $X$. Then \[ DD^T = \De -A,\quad D^TD = 2I +S \] where $\De$ is the diagonal matrix of valencies of $X$ and $S$ is a symmetric matrix with zero diaogonal and off-diagonal entries from $\{0,1,-1\}$. In other words, $S$ is a signed adjacency matrix. The entries of $S$ are indexed by $E(X)$, and the $ab$-entry is non-zero if and only if the edges $a$ and $b$ are adjacent in the line graph $L(X)$ of $X$, so we have a signed adjacency matrix for $L(X)$.

A signed ajacency matrix $T$ of a graph $Y$ determines a two-fold cover $Z$ of $Y$, as follows. The vertex set of the cover is \[ V(Y) \times \{0,1\} \] and $(u,i)$ and $(v,j)$ are adjacent if and only if $u\sim v$ and either \[ i=j,\quad T_{u,v}=1 \] or \[ i\ne j,\quad T_{u,v}=-1. \] It is easy to check that the map that sends $(u,i)$ to $(u,1-i)$ is an automorphism of $Z$. In matrix terms, we get the adjacency matrix of $Z$ by applying the following substitutions to the entries of $A(Y)$: \[ 0\to\pmat{0&0\\0&0},\qquad 1\to\pmat{1&0\\0&1} \qquad -1\to\pmat{0&1\\1&0}. \]

The pairs \[ \{(u,0),(u,1)\} \] are called the fibres of the cover, you might verify that these form an equitable partition of the cover, and that the quotient over the fibres is $Y$. Thus we see that the characteristic polynomial of the cover is divisible by the characteristic polynomial of the graph.

Lemma \[ \phi(Z,t) =\phi(T,t)\phi(Y,t). \] Proof (Outline.) Let $H$ be the matrix \[ \frac{1}{\sqrt{2}} \pmat{1&1\\1&-1} \] and let $K$ be the block-diagonal matrix formed using $|V(Y)|$ copies of $H$. If $M$ is a complex matrix, let $|M$ denote the matrix we get by replacing each entry of $M$ by its absolute value. Then $K$ is orthogonal and symmetric and $KA(Z)K$ is permutation equivalent to \[ \pmat{S&0\\0&|S|}. \] $\blacksquare$

Here $|S|=A(Y)$. We offer another view of this result. If $A$ and $B$ are symmetric $01$-matrices such that $A\circ B=0$, then $A-B$ is a signed adjacency matrix and \[ \pmat{A&B\\B&A} \] is the adjacency matrix of the corresponding cover. We can now use the following: \[ \frac12\pmat{I&I\\I&-I}\pmat{A&B\\B&A}\pmat{I&I\\I&-I} =\pmat{A+B&0\\0&A-B}. \]

This leads us to an easy construction of the adjacency matrix of the cover.

def lgcvr( X):
    D = X.incidence_matrix()
    IDM = identity_matrix( X.num_edges())
    M = D.transpose()*D -2*IDM
    MM = M.elementwise_product(M)
    A = (1/2)*(M+MM)
    B = (1/2)*(M-MM)
    return block_matrix(2, 2, [A,B,B,A])

Now we turn to our line graphs. The cover can be constructed as follows. Choose one arc $(u,v)$ for each edge $\{u,v\}$ of $X$. (This defines an orientation of the graph.) Our matrix $S$ is a signing of $A(L(X))$---its rows and columns are indexed by our chosen arcs, and the entry corresponding to a pair of distinct overlapping arcs is 1 if they have the same head or tail, and $-1$ otherwise. Rather than represent the vertices of the cover by pairs $((u,v),i)$, we proceed thus: if $(u,v)$ is one of our chosen arcs then we use $(u,v)$ to denote $((u,v),0)$ and $(v,u)$ for $((u,v),1)$. The following procedure implements this.

def dbl_lng( X):
    V = X.vertices()
    E = X.edges( labels=False)
    arcs = [(i,j) for i in V for j in V\
        if ((i,j) in E or (j,i) in E)]
    return  Graph( [arcs, lambda a,b: a != b and (a[0]==b[0] or a[1]==b[1])])

We test that our two constructions agree on the Petersen graph, and see that cover is isomorphic to the line graph of the direct product of the Petersen graph with $K_2$. (This product is itself a cover of the Petersen graph.)

P = graphs.PetersenGraph()
Q1 = Graph( lgcvr( P))
Q2 = dbl_lng( P)
Q1.is_isomorphic( Q2)

True

K2 = graphs.CompleteGraph(2)
LP2 = (P.tensor_product(K2)).line_graph()
LP2.is_isomorphic( Q1)

True