🔢 mathematics

Signed graphs with fixed smallest eigenvalue at least $-3$ and their lattices

This paper establishes that connected signed graphs with sufficiently large minimum valency and smallest eigenvalue slightly above $-3$ must have eigenvalues at least $-3$ and generate lattices that are sublattices of direct sums of $\mathbb{Z}^n$ and $E_8$ , while also exploring the connection between such graphs and rootless irreducible unimodular lattices.

Original authors: Meng-Yue Cao, Jack H. Koolen, Jing-Yuan Liu, Qianqian Yang

Published 2026-07-13

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Meng-Yue Cao, Jack H. Koolen, Jing-Yuan Liu, Qianqian Yang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a giant, invisible playground made of dots (vertices) and lines (edges). In this playground, every line has a secret personality: it's either a friendly "plus" (+) or a grumpy "minus" (−). Mathematicians call this a signed graph. Now, imagine these dots and lines are vibrating like guitar strings. Every graph has a specific "lowest note" it can hum, called its smallest eigenvalue.

For a long time, mathematicians have been trying to figure out what happens when these graphs get really, really big and busy (meaning every dot is connected to lots of other dots). Specifically, they wanted to know: If a graph is huge and its lowest note is just a tiny bit higher than a very low pitch (specifically, higher than -3 minus a tiny bit of "epsilon"), what does the graph actually look like?

The Big Discovery: The "Magic Floor"

The authors of this paper, Cao, Koolen, Liu, and Yang, proved a fascinating rule. They showed that if you have a connected signed graph that is busy enough (meaning every dot has a high number of neighbors) and its lowest note is higher than -3.000...1 (just a smidge above -3), then two amazing things happen:

The Pitch Stabilizes: The graph's lowest note actually bumps up to be at least -3. It can't stay in that tiny gap between -3 and -3.000...1 if the graph is big enough. It's like a ball rolling down a hill that suddenly hits a flat, solid floor at -3 and stops.
The Lattice Structure: If you turn this graph into a mathematical "lattice" (a grid-like structure made of vectors, which are like arrows with specific lengths), this lattice turns out to be built from very specific, famous building blocks. It is a piece of a giant structure made by combining:
- Standard grids (called $Z_n$ ).
- Copies of a super-special, 8-dimensional shape called the $E_8$ root lattice.

Think of it like this: If you build a massive, complex castle out of Lego bricks, and you find that the castle is huge and stable, the authors proved that the castle must be built using only standard bricks and a specific, rare "super-brick" called $E_8$ . You can't use just any random brick; the math forces the structure to be made of these specific types.

What They Ruled Out

The paper is very clear about what doesn't happen.

No "In-Between" Chaos: They proved that you cannot have a huge, busy graph with a lowest eigenvalue stuck in that tiny, mysterious gap between -3 and -3 minus a tiny bit. If the graph is big enough, it either snaps to -3 or goes higher.
No Infinite Variety of "Dead Ends": The authors investigated "non-extendable" graphs—graphs that are so complete they can't be made bigger without breaking the rules. They found that while there are some famous, massive examples of these (like one with 2,300 dots and 891 connections per dot), they expect the answer to "Are there infinitely many?" to be no. In fact, based on Theorem 1.7, they suggest that the answer is likely no.

The "Fat" and "Slim" Analogy

To prove this, the authors used a clever trick involving "Hoffman signed graphs." Imagine a graph where some dots are "slim" (regular) and others are "fat" (special, heavy dots).

They showed that if your graph is big enough, it must be the "slim" part of a larger, "fat" graph that has a lowest note of at least -3.
They proved that the list of "forbidden" fat graphs (the ones that would break the rules) is finite. There are only a limited number of ways to build a "bad" fat graph that is just small enough to be a problem. Once you know there are only a finite number of these bad shapes, you can prove that big graphs can't accidentally stumble into the forbidden zone.

The "Leech" and "Conway" Connections

The paper also connects these graphs to some legendary mathematical objects called lattices.

They looked at special, "rootless" lattices (grids where the shortest arrows have a squared length of 3, not 2).
They found that if you take these special lattices (like the shorter Leech lattice in 23 dimensions or the odd Leech lattice in 24 dimensions) and pick specific arrows to build a graph, you get a signed graph with a lowest eigenvalue of exactly -3.
These graphs are "non-extendable," meaning you can't add any more dots to them without changing their lowest note.
The paper lists specific numbers for these famous examples:
- One graph has 2,300 dots and a valency (number of connections) of 891.
- Another has 2,048 dots and 759 connections.
- There are others with 1,560, 1,332, 820, 1,120, 864, 928, and 800 dots.

How Sure Are They?

The authors didn't just guess or run simulations; they proved it with rigorous math.

They proved that for any graph with a minimum valency (connectivity) above a certain number (let's call it $\kappa_3$ ), the lowest eigenvalue must be at least -3.
They proved that the associated lattice is a sublattice of $Z_n$ and copies of $E_8$ .
They proved that there are infinitely many graphs that contain a specific graph as a smaller piece (meaning the graph is "extendable" unless it's one of those special, rare ones).
They expect (based on Theorem 1.7) that there are only finitely many "non-extendable" graphs with a lowest eigenvalue of -3. They point out that the constant $\kappa_4$ (the minimum connectivity needed to guarantee a graph is extendable) must be at least 892, based on that famous 2,300-dot example.

In short, the paper draws a hard line in the sand: if your graph is big and busy, it can't be weirdly stuck between -3 and -3.000...1. It has to settle on -3 or higher, and its underlying structure is built from a very specific, elegant set of mathematical bricks.

Technical Summary: Signed graphs with fixed smallest eigenvalue at least −3 and their lattices

Problem Statement
This paper investigates the structural properties of connected signed graphs $(G, \sigma)$ whose smallest eigenvalue, $\lambda_{\min}(G, \sigma)$ , is bounded from below by a value close to $-3$. Specifically, the authors address the behavior of such graphs when they possess a sufficiently large minimum valency (degree). The study focuses on two primary aspects:

Determining whether the smallest eigenvalue is forced to be at least $-3$ (rather than just greater than $-3 - \varepsilon$ ) under high valency conditions.
Characterizing the integral lattice $\Lambda(G, \sigma)$ associated with these graphs, specifically determining if it embeds into a direct sum of standard lattices $\mathbb{Z}^n$ and copies of the root lattice $E_8$ .
Analyzing the "extendability" of these graphs: whether a signed graph with $\lambda_{\min} \ge -3$ can be a proper induced subgraph of a larger connected signed graph with the same eigenvalue bound.

Methodology
The authors employ a combination of spectral graph theory and lattice theory, utilizing the following key tools:

Lattice Association: For a signed graph with smallest eigenvalue $\lambda$ , the matrix $G_r(G, \sigma) = A(G, \sigma) - \lfloor \lambda \rfloor I$ is positive semidefinite. The lattice $\Lambda(G, \sigma)$ is defined as the integral lattice generated by the columns of a matrix $N$ such that $G_r(G, \sigma) = N^T N$ .
Hoffman Signed Graphs: The authors utilize the framework of Hoffman signed graphs, which consist of "slim" and "fat" vertices. Fat vertices are replaced by positive cliques to construct larger graphs. This framework allows for the analysis of eigenvalue limits as clique sizes grow.
Minimal Forbidden Subgraphs: The proof strategy relies on identifying a finite set of "minimal forbidden fat Hoffman signed graphs" for the threshold $\lambda = -3$ . By proving this set is finite, the authors establish that graphs with sufficiently large minimum valency cannot contain these forbidden structures, thereby forcing the eigenvalue to satisfy specific bounds.
Rootless Unimodular Lattices: The paper constructs examples of non-extendable signed graphs using irreducible unimodular lattices generated by vectors of squared norm 3 (without roots), such as the shorter Leech lattice and the Conway-Borcherds lattice.

Key Contributions and Results

Eigenvalue Gap Elimination (Theorem 1.3):
The paper proves that there exist a positive real number $\varepsilon$ and a positive integer $\kappa_3$ such that if a connected signed graph has $\lambda_{\min} > -3 - \varepsilon$ and minimum valency at least $\kappa_3$ , then $\lambda_{\min} \ge -3$ . This strengthens previous results which characterized graphs with $\lambda_{\min} \in (-2, -1]$ and $(-1-\sqrt{2}, -1]$ .
Lattice Structure (Theorem 1.3):
Under the conditions above, the associated lattice $\Lambda(G, \sigma)$ is proven to be a sublattice of a direct sum of the standard lattice $\mathbb{Z}^n$ and copies of the root lattice $E_8$ . This generalizes earlier findings where lattices for $\lambda_{\min} \ge -2$ were restricted to root lattices $A_n, D_n, E_6, E_7, E_8$ .
Extendability of Graphs (Theorem 1.7):
The authors show that if a connected signed graph has $\lambda_{\min} \ge -3$ and sufficiently large minimum valency, it is extendable. That is, it can be embedded as a proper induced subgraph into infinitely many larger connected signed graphs that also satisfy $\lambda_{\min} \ge -3$ . This implies that such graphs are not maximal in the context of eigenvalue preservation.
Construction of Non-Extendable Graphs:
The paper constructs specific examples of non-extendable regular signed graphs with $\lambda_{\min} = -3$ . These are derived from irreducible unimodular lattices generated by squared norm 3 vectors without roots (e.g., the shorter Leech lattice, odd Leech lattice, and Conway-Borcherds lattice).
- An example is provided with order 2300 and valency 891.
- This construction demonstrates that the constant $\kappa_4$ in Theorem 1.7 (the threshold for extendability) must be at least 892.
Finiteness of Forbidden Structures (Theorem 3.2):
A critical technical result is the proof that the set of minimal forbidden fat Hoffman signed graphs for $\lambda = -3$ is finite. This finiteness is essential for establishing the existence of the valency threshold $\kappa_3$ .

Significance and Claims
The paper positions itself as a substantial strengthening of prior characterizations of signed graphs with fixed smallest eigenvalues.

Motivation: The work is motivated by the classification of signed graphs with $\lambda_{\min} \ge -2$ (which correspond to root lattices) and the broader problem of determining the maximum number of biangular lines in Euclidean space with specific angles.
Classification: The authors suggest that while there are infinitely many extendable graphs with $\lambda_{\min} \ge -3$ , the existence of non-extendable graphs is likely finite (or at least restricted to specific lattice constructions). They pose Problem 1.9: "Are there infinitely many switching classes of non-extendable signed graphs with smallest eigenvalue at least $-3$?" Based on their results, they expect the answer to be negative.
Limitations: The paper does not claim to fully classify all non-extendable signed graphs with $\lambda_{\min} = -3$ , but rather provides a framework for understanding their lattice properties and constructing specific examples. It highlights that the constant $\kappa_4$ required for the extendability theorem is non-trivial (at least 892), indicating that small graphs may behave differently than those with large valency.

In summary, the paper establishes that high valency forces signed graphs with eigenvalues near $-3$ to conform to a specific lattice structure ( $\mathbb{Z}^n \oplus E_8^k$ ) and become extendable, while identifying specific lattice-derived constructions as the likely candidates for non-extendable, maximal signed graphs in this regime.

The Big Discovery: The "Magic Floor"

What They Ruled Out

The "Fat" and "Slim" Analogy

The "Leech" and "Conway" Connections

How Sure Are They?

More like this