# Symmetry-Based Parameter Efficient Low-Rank Fine-Tuning for High-Throughput Calculation of Lattice Thermal Conductivity

Preprint · Rustam Arabov, Konstantin Sozykin, Alexander Shapeev, Nikita Rybin

- Archive id: arxiv:2610.29634
- Published: 2026-10-21
- arXiv: https://arxiv.org/abs/2610.29634
- Citations: 0
- Source URL: https://arxiv.org/abs/2610.29634
- PDF: https://arxiv.org/pdf/2610.29634
- Provenance: arxiv

## Abstract

Accurate and efficient prediction of lattice thermal conductivity (LTC) is essential for developing materials for thermoelectric, microelectronic, and photovoltaic applications. Here, we present a methodology for high-throughput calculation of LTC across large datasets of compounds sharing a common crystal symmetry. Our approach combines fine-tuning of the MatterSim 1M universal machine-learned interatomic potential with LTC calculations based on the solution of the Boltzmann transport equation. We introduce a symmetry-based few-shot fine-tuning strategy (SBFS-FT) based on low-rank adaptation. In this approach, MatterSim 1M is fine-tuned on a small, carefully selected set of compounds that share a common crystal symmetry. The resulting potential is then applicable for accurate LTC calculations for compounds with this crystal symmetry, even if they were not included in the fine-tuning set. We tested our methodology on datasets of half-Heusler compounds (space group 216), materials with space group 225, and spinels (space group 227). We demonstrated that LTC calculations with MatterSim 1M fine-tuned using SBFS-FT exhibited a mean absolute relative difference from the reference results (obtained using moment tensor potentials and polyMLPs) 2-3 times smaller than that achieved with the non-fine-tuned MatterSim 1M potential. Moreover, the accuracy of LTC calculations with MatterSim 1M after SBFS-FT was comparable to that achieved using the MatterSim 1M potentials individually fine-tuned for each compound. Importantly, SBFS-FT enabled generation of the potential suitable for accurate LTC calculations across large compound datasets substantially faster than fine-tuning of MatterSim 1M for each compound individually, reducing the computational time by approximately a factor of 5-7 for a dataset of 100 compounds.

_Abstract reproduced under CC0-1.0 (arXiv descriptive metadata)._

Topics: [AI for Science](https://space-kitty.com/topics/ai-for-science), [Computational & AI-Driven Superconductor Discovery](https://space-kitty.com/topics/superconductor-discovery)

---
Space Kitty Archive · https://space-kitty.com · API: https://space-kitty.com/api/v1 · MCP: https://space-kitty.com/mcp · Premium: $0.20/request via x402 (Base Sepolia (testnet)). Metadata: arXiv (CC0), OpenAlex (CC0), Europe PMC (CC BY/CC0 works), ClinicalTrials.gov.