Variance bound for the degree deviation of connected graphs
Let be a connected graph, and let denote its spectral radius, its degree vector, and its degree variance. Define the degree deviation by , where and are the numbers of vertices and edges of . Variance bound. If is connected, then
The bound is motivated by numerical evidence and would improve an earlier estimate, but it is false for disconnected graphs. The source gives no resolution for connected graphs; it is proved there only under the additional hypothesis that the clique number is at least .
References
Primary source
Felix Goldberg, “New results on eigenvalues and degree deviation”, arXiv:1403.2629 (2014).
Progress summary
A reader-submitted construction claims the bound is false, but nobody has independently checked the claimed counterexample.
Felix Goldberg proposed the conjecture in 2014: every connected graph should satisfy . The source motivated it numerically and noted failure for disconnected graphs.
Known results
- Goldberg, 2014: proved the bound when the clique number satisfies .
Community submission (unverified), August 22, 2026
A submitted argument claims an explicit connected -vertex lollipop graph violates the inequality, using a Rayleigh-quotient lower bound, and also claims infinite counterexample families including bounded-degree trees. These claims have no independent verification in the retrieved sources.
Current status (as of September 2026): The bound is established under , while the general connected-graph conjecture has only an unverified submitted counterexample claim.
Sources
- ar5iv.labs.arxiv.org
- arxiv.org
- quantamagazine.org
- quantamagazine.org
- quantamagazine.org
- quantamagazine.org
- quantamagazine.org
- quantamagazine.org
- quantamagazine.org
- cdn.openai.com
- export.arxiv.org
- export.arxiv.org
- arxiv.org
- arxiv.org
- mathstodon.xyz
- mathstodon.xyz
- mathstodon.xyz
- mathstodon.xyz
- mathstodon.xyz
Solutions 1
CounterexampleThis solution needs a summarySee full solution
Counterexample: the degree-variance spectral bound fails even for bounded-degree trees
Conjecture 1.5 of Felix Goldberg, New results on eigenvalues and degree deviation, arXiv:1403.2629, asserts that every connected simple graph satisfies
where and is the spectral radius of its adjacency matrix. We disprove (1) by an explicit 12-vertex graph, an infinite family of lollipop graphs, and an infinite family of connected bipartite trees of maximum degree three. In fact, no universal constant multiplying the right-hand side can repair the inequality.
A 12-vertex connected counterexample
For , let consist of a complete graph on four vertices together with a path on additional vertices, attached by a single edge to one clique vertex. Its degrees are
Consequently,
and hence
Write for the adjacency matrix of , and choose a vector that equals on the four clique vertices, equals on the first path vertex, and vanishes elsewhere. The Rayleigh principle gives
At , formulas (4) and (5) yield
The final inequality is exact: . Thus is a concrete connected counterexample to (1).
In fact the entire family violates (1) for every . Indeed,
The numerator equals at , and its successive difference is for every . Since the quantity being squared on the left is positive, (7) proves the strict violation for all such .
For an even simpler instance requiring only the clique bound , take . Then
because .
Counterexamples among subcubic bipartite trees
Begin with a three-vertex path . Attach two leaves to , one leaf to , and two leaves to , obtaining an eight-vertex tree with three vertices of degree three and five leaves. For , extend any one of these leaves by a path consisting of new vertices. Denote the resulting tree by . Its degree multiset is
In particular, is connected, bipartite, and has maximum degree three. Since has edges,
so that
Choose a vector that equals on , equals on the five original leaves, and vanishes on every newly added path vertex. The two central edges contribute to , and the five center-leaf edges contribute . Therefore
Combining (10)--(12),
The numerator in (13) equals at , and its successive difference is for every . Hence every with violates the conjecture.
Moreover, (11) and (12) show that
Thus even restricting to connected bipartite trees of maximum degree three, there exists no absolute constant for which
holds universally. Goldberg's positive result for graphs with clique number at least is unaffected: both counterexample families lie outside that additional hypothesis.