Tim Seppelt

Anuj Dawar, Benedikt Pago, Tim Seppelt

The central open question of algebraic complexity is whether VP is unequal to VNP, which is saying that the permanent cannot be represented by families of polynomial-size algebraic circuits. For symmetric algebraic circuits, this has been confirmed by Dawar and Wilsenach (2020, 2025) who showed exponential lower bounds on the size of symmetric circuits for the permanent. In this work, we set out to develop a more general symmetric algebraic complexity theory. Our main result is that a family of symmetric polynomials admits small symmetric circuits if and only if they can be written as a linear combination of homomorphism counting polynomials of graphs of bounded treewidth. We also establish a relationship between the symmetric complexity of subgraph counting polynomials and the vertex cover number of the pattern graph. As a concrete example, we examine the symmetric complexity of immanant families (a generalisation of the determinant and permanent) and show that a known conditional dichotomy due to Curticapean (2021) holds unconditionally in the symmetric setting.

Snir Hordan, Nadav Dym, Tim Seppelt

Graphs with a simple spectrum admit cubic-time isomorphism testing, yet we prove that for every natural number $k$, the $k$-Weisfeiler-Leman ($k$-WL) test cannot distinguish all non-isomorphic graphs with a simple spectrum. As the WL hierarchy upper-bounds the distinguishing power of widely-used Graph Neural Networks (GNNs), this incompleteness applies to all such GNNs, ruling out completeness for every $k$-WL-aligned GNN family. To close this gap, we introduce PRiSM (Partition, Refine, Solve, Match), the first provably complete canonicalization of simple-spectrum eigendecompositions. PRiSM obtains the completeness guarantee that prior canonicalizations provably lack, and resolves the open problem of achieving complete expressivity on simple-spectrum graphs. When composed with DeepSets or a Transformer, PRiSM achieves universal approximation on simple-spectrum graphs, justifying the use of canonicalized Laplacian positional encodings. Empirically, PRiSM performs comparably to or outperforms existing spectral canonicalizations on graph regression, classification, and expressivity

Isolde Adler, Eva Fluck, Tim Seppelt et al.

We study the expressive power of first-order logic with counting quantifiers, especially the $k$-variable and quantifier-rank-$q$ fragment $\mathsf{C}^k_q$, using homomorphism indistinguishability. Recently, Dawar, Jakl, and Reggio (2021) proved that two graphs satisfy the same $\mathsf{C}^k_q$-sentences iff they are homomorphism indistinguishable over the class $\mathcal{T}^k_q$ of graphs admitting a $k$-pebble forest cover of depth $q$. After reproving this result using elementary means, we provide a graph-theoretic analysis of $\mathcal{T}^k_q$. This allows us to separate $\mathcal{T}^k_q$ from the intersection $\mathcal{TW}_{k-1} \cap \mathcal{TD}_q$, provided that $q$ is sufficiently larger than $k$. Here $\mathcal{TW}_{k-1}$ is the class of all graphs of treewidth at most $k-1$ and $\mathcal{TD}_q$ is the class of all graphs of treedepth at most $q$. We are able to lift this separation to a separation of the respective homomorphism indistinguishability relations $\equiv_{\mathcal{T}^k_q}$ and $\equiv_{\mathcal{TW}_{k-1} \cap \mathcal{TD}_q}$. We do this by showing that the classes $\mathcal{TD}_q$ and $\mathcal{T}^k_q$ are homomorphism distinguishing closed, as conjectured by Roberson (2022). In order to prove Roberson's conjecture for $\mathcal{T}^k_q$, we characterise $\mathcal{T}^k_q$ in terms of a monotone Cops-and-Robber game. The crux is to prove that if Cop has a winning strategy then Cop also has a winning strategy that is monotone. To that end, we transform Cops' winning strategy into a pree-tree-decomposition, which is inspired by decompositions of matroids, and then apply an intricate breadth-first cleaning up procedure along the pree-tree-decomposition (which may temporarily lose the property of representing a strategy). Thereby, we achieve monotonicity while controlling the number of rounds across all branches of the decomposition via a vertex exchange argument.