Quantum
A Basis-Dependent Kernel for Learning Quantum Phases of Matter
Publication: quant-ph updates on arXiv.org
Author: Clemens Lindner, Vin\'icius Mohr, Adri\'an P\'erez-Salinas
Source published: 20/11/2026, 05:00:00 UTC
Briefing created: 20/11/2026, 05:44:04 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 20/11/2026, 05:13:24 UTC.
Briefings
A new kernel method called the inverse participation ratio (IPR) kernel was developed to learn quantum phases of matter by analyzing IPRs in physically motivated bases. The study tested the method on ground states of the transverse-field Ising (TFIM) and Heisenberg XXZ models, demonstrating that the IPR kernel retains phase-relevant structure at nonzero temperatures where fidelity-based kernels fail. The research suggests quantum kernels benefit more from physics-motivated feature maps than from maximal expressivity.
Key points
- The IPR kernel identifies quantum states with physically meaningful similarity even when they are orthogonal, unlike fidelity-based kernels.
- The IPR kernel retains phase-relevant structure over a range of nonzero temperatures for TFIM Gibbs states where fidelity-based kernels are degraded.
Context
This work addresses a central problem in condensed-matter physics: identifying quantum phases as parameters vary. The method compares quantum states using IPRs in physically motivated bases, offering a solution to the limitations of existing data-driven approaches.
Limitations
The study was conducted on ground states of specific models (TFIM and Heisenberg XXZ) and did not test the method on excited states or other physical systems. Results are based on ground state data only, and the temperature range tested is limited to nonzero values for the TFIM model.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 20/11/2026, 05:44:04 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeie43pk2d3h6bijyend2lm3hoc7i5l4czyqdupcyzkq37mkdkqxuha