The path-matrix conjecture for 2-connected graphs

At least 3 years old · documented by

Let MM be a graph, and let its path matrix be the matrix whose eigenvalues are called path eigenvalues. The graph MM is 2-connected if it remains connected after the deletion of any one vertex.

The path-matrix conjecture. If MM is 2-connected, then its path matrix has exactly one positive eigenvalue.

The claim extends the observed class of graphs whose path matrices have one positive eigenvalue. The supplied text gives no resolution or further general context, so its status remains open.

References

Primary source

Amol P. Narke, Prashant P. Malavadkar and Maruti M. Shikare, “Bounds on Path Energy of Graphs”, arXiv:2205.05100 (2024).

Progress summary

Refreshed
Open

A reader-submitted calculation claims an infinite family of 2-connected graphs disproves the conjecture, but the calculation has not been independently verified.

The conjecture asserts that every 2-connected graph has a path matrix with exactly one positive eigenvalue. No proposer or date is identified in the supplied material.

Community submission (unverified)

A submitted calculation argues that Gm=K2∨Km‾G_m=K_2\vee\overline{K_m} is 2-connected and that its path matrix has two positive eigenvalues for every integer m≥5m\ge 5. It derives a two-dimensional quotient matrix with determinant 2(m2−4m−1)>02(m^2-4m-1)>0 and positive trace, while the remaining eigenvalues are negative; if correct, this gives an infinite family of counterexamples, including planar chordal graphs.

Current status (as of August 2026): The conjecture remains unresolved publicly; an unverified community calculation claims it is false for the family K2∨Km‾K_2\vee\overline{K_m} when m≥5m\ge 5.

Solutions 1

CounterexampleThis solution needs a summarySee full solutionHide full solution

An infinite family of 2-connected counterexamples

The conjecture is false even for planar chordal graphs. For each integer m≥1m\geq1, let

Gm=K2∨Km‾.G_m=K_2\vee\overline{K_m}.

More explicitly, its vertices are

V(Gm)={u,v,x1,…,xm},V(G_m)=\{u,v,x_1,\ldots,x_m\},

and its edges are

E(Gm)={uv}∪{uxi,vxi:1≤i≤m}.E(G_m)=\{uv\}\cup\{ux_i,vx_i:1\leq i\leq m\}.

Thus GmG_m consists of mm triangles sharing the edge uvuv. Removing any xix_i leaves a connected graph, while removing either uu or vv leaves a star. Consequently GmG_m is 2-connected.

Write p(a,b)p(a,b) for the maximum number of internally vertex-disjoint paths between distinct vertices a,ba,b. If one endpoint is xix_i, its degree is two, so there are at most two such paths. There are also two explicitly:

xiuxj,xivxj(i≠j),x_iux_j,\qquad x_ivx_j \qquad (i\neq j),

and

xiu,xivu.x_iu,\qquad x_iv u.

The analogous paths connect xix_i to vv. Hence

p(xi,xj)=2(i≠j),p(xi,u)=p(xi,v)=2.p(x_i,x_j)=2\quad(i\neq j), \qquad p(x_i,u)=p(x_i,v)=2.

For the two vertices on the common edge, the paths

uv,ux1v,ux2v,…,uxmvuv,\qquad ux_1v,\qquad ux_2v,\qquad\ldots,\qquad ux_mv

are internally vertex-disjoint. Since both uu and vv have degree m+1m+1, this is optimal, and therefore

p(u,v)=m+1.p(u,v)=m+1.

In the vertex order x1,…,xm,u,vx_1,\ldots,x_m,u,v, the path matrix is

P(Gm)=(2(Jm−Im)21m×2212×m(m+1)(J2−I2)).P(G_m)= \begin{pmatrix} 2(J_m-I_m)&2\mathbf 1_{m\times2}\\ 2\mathbf 1_{2\times m}&(m+1)(J_2-I_2) \end{pmatrix}.

Every vector supported on the first mm coordinates whose coordinates sum to zero is an eigenvector with eigenvalue −2-2. This gives multiplicity m−1m-1. Likewise, the vector supported on u,vu,v with coordinates 1,−11,-1 is an eigenvector with eigenvalue −(m+1)-(m+1).

The remaining two-dimensional invariant subspace consists of vectors constant on {x1,…,xm}\{x_1,\ldots,x_m\} and constant on {u,v}\{u,v\}. The corresponding quotient matrix is

Qm=(2(m−1)42mm+1).Q_m= \begin{pmatrix} 2(m-1)&4\\ 2m&m+1 \end{pmatrix}.

Its trace and determinant are

tr⁡(Qm)=3m−1,det⁡(Qm)=2(m2−4m−1).\operatorname{tr}(Q_m)=3m-1, \qquad \det(Q_m)=2(m^2-4m-1).

Consequently, the complete characteristic polynomial is

det⁡(λI−P(Gm))=(λ+2)m−1(λ+m+1)(λ2−(3m−1)λ+2(m2−4m−1)),\det(\lambda I-P(G_m)) =(\lambda+2)^{m-1}(\lambda+m+1) \bigl(\lambda^2-(3m-1)\lambda+2(m^2-4m-1)\bigr),

and the two remaining eigenvalues are

λ±(m)=3m−1±m2+26m+92.\lambda_{\pm}(m) =\frac{3m-1\pm\sqrt{m^2+26m+9}}{2}.

For every integer m≥5m\geq5, their sum and product are positive:

λ+(m)+λ−(m)=3m−1>0,λ+(m)λ−(m)=2(m2−4m−1)>0.\lambda_+(m)+\lambda_-(m)=3m-1>0, \qquad \lambda_+(m)\lambda_-(m)=2(m^2-4m-1)>0.

Therefore both are strictly positive. All other path eigenvalues are strictly negative, so

#{λ∈Spec⁡(P(Gm)):λ>0}=2(m≥5).\#\{\lambda\in\operatorname{Spec}(P(G_m)):\lambda>0\}=2 \qquad(m\geq5).

An explicit seven-vertex counterexample

Taking m=5m=5 gives the 2-connected graph with edges

uv,ux1,vx1,ux2,vx2,ux3,vx3,ux4,vx4,ux5,vx5.uv,\quad ux_1,vx_1,\quad ux_2,vx_2,\quad ux_3,vx_3, \quad ux_4,vx_4,\quad ux_5,vx_5.

Its path matrix is

P(G5)=(0222222202222222022222220222222202222222062222260).P(G_5)= \begin{pmatrix} 0&2&2&2&2&2&2\\ 2&0&2&2&2&2&2\\ 2&2&0&2&2&2&2\\ 2&2&2&0&2&2&2\\ 2&2&2&2&0&2&2\\ 2&2&2&2&2&0&6\\ 2&2&2&2&2&6&0 \end{pmatrix}.

The characteristic polynomial factors as

det⁡(λI−P(G5))=(λ+2)4(λ+6)(λ2−14λ+8),\det(\lambda I-P(G_5)) =(\lambda+2)^4(\lambda+6)(\lambda^2-14\lambda+8),

and thus its path spectrum is

Spec⁡(P(G5))={−2,−2,−2,−2,−6,7−41,7+41}.\operatorname{Spec}(P(G_5)) =\{-2,-2,-2,-2,-6,7-\sqrt{41},7+\sqrt{41}\}.

Since 41<7\sqrt{41}<7, both 7−417-\sqrt{41} and 7+417+\sqrt{41} are positive. This directly contradicts the claim that every 2-connected graph has exactly one positive path eigenvalue.

The assertion refuted here is Conjecture 1 of A. P. Narke, P. P. Malavadkar, and M. M. Shikare, Bounds on Path Energy of Graphs, current arXiv version 4 (2024), https://arxiv.org/abs/2205.05100. The path matrix used above is exactly the internally vertex-disjoint-path matrix in Definition 1.1 of that source.