aiqbench analysis
This paper establishes a theoretical error floor for polynomial-based graph filters on the Long-Range Ising Model benchmark, proving that standard isotropic message passing architectures cannot capture long-range dependencies without specific architectural modifications. The authors demonstrate that strided filters with sufficient coefficients can bypass this limit, achieving significantly lower test errors than existing spectral networks and dilated convolutions while resolving coupling across the entire lattice. This work provides a rigorous mathematical bound for graph neural network performance on systems with power-law interactions, offering clear guidance for future architecture design in physics-informed machine learning.
Spectral Limits of Polynomial Graph Filters on the Long-Range Ising Model Benchmark
arXiv:2610.22713v1 Announce Type: new Abstract: The Long-Range Ising Model benchmark (LRIM) guarantees long-range dependence by construction: each label is the node's spin times the spin field convolved with a power-law coupling, so any model's distance from the target is measurable. We prove that a readout whose filter is a polynomial in the adjacency, the linear core of isotropic message passing and spectral graph networks, leaves an error floor that no degree lowers, because the target varies along a level set of the spectrum on which every such filter is constant. Computed exactly, the floor and the best filter of each degree predict two spectral networks adapted from published architectures to within a tenth of a decade. A filter escapes the floor only if it weighs separately the offsets the adjacency lumps together, and must still resolve the coupling across the whole lattice. A strided filter of about fifty coefficients does both: fitted to the training labels, its test error is 52 times below the best published network's. A dilated convolutional network beats the best published result at 256 nodes, matches it at 1,024, and gives the first in-domain results at the three larger sizes, while a Fourier neural operator with its standard modes falls further behind as the lattice grows. At its standard budget the dilated network stays two to four orders of magnitude short of a larger strided filter fitted to the labels. The analysis carries over to the benchmark's second coupling.
Abstract:The Long-Range Ising Model benchmark (LRIM) guarantees long-range dependence by construction: each label is the node's spin times the spin field convolved with a power-law coupling, so any model's distance from the target is measurable. We prove that a readout whose filter is a polynomial in the adjacency, the linear core of isotropic message passing and spectral graph networks, leaves an error floor that no degree lowers, because the target varies along a level set of the spectrum on which every such filter is constant. Computed exactly, the floor and the best filter of each degree predict two spectral networks adapted from published architectures to within a tenth of a decade. A filter escapes the floor only if it weighs separately the offsets the adjacency lumps together, and must still resolve the coupling across the whole lattice. A strided filter of about fifty coefficients does both: fitted to the training labels, its test error is 52 times below the best published network's. A dilated convolutional network beats the best published result at 256 nodes, matches it at 1,024, and gives the first in-domain results at the three larger sizes, while a Fourier neural operator with its standard modes falls further behind as the lattice grows. At its standard budget the dilated network stays two to four orders of magnitude short of a larger strided filter fitted to the labels. The analysis carries over to the benchmark's second coupling.
Machine Learning (cs.LG)
arXiv:2610.22713 [cs.LG]
(or arXiv:2610.22713v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.22713 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 332 words
Related coverage
TwinS-GCN: Spectral conjugate for Spectral Graph Convolutional Networks
arXiv · Sep 30, 2026
Spectral Flow Certificates for Depth-Aware Long-Range Propagation in Graph Neural Networks
arXiv · Jul 28, 2026
ButterMamba: Butterworth-Enhanced Spatial-Temporal Mamba for Efficient Traffic Flow Prediction
arXiv · Sep 1, 2026
Utrecht University Cuts Simulation Times with New Symmetry Tool
Quantum Zeitgeist · Sep 10, 2026