Reproduce “Scaling Autoregressive Models for Lattice Thermodynamics” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xiaochen Du, Juno Nam, Sulin Liu, Rafael Gómez-Bombarelli
Researchers from MIT developed a framework using Any-Order Autoregressive Models (AO-ARMs) with Marginalization Models (MAMs) to efficiently sample crystal lattice configurations, enabling accurate free energy calculations and phase diagram mapping. The framework successfully captured complex thermodynamic properties in Ising and CuAu alloy systems, achieving up to 160-fold speedups in sampling compared to traditional methods for larger systems.
Key insights
- Joint training of Any-Order Autoregressive Models and Marginalization Models provides both flexible conditional generation and O(L) memory efficiency during backpropagation.
- Transformer architectures with appropriate periodic positional encodings are superior for capturing long-range correlations and critical phenomena in lattice systems compared to MLPs and GNNs.
- The out-painting strategy, utilizing models trained on smaller lattices to generate larger configurations, provides a robust and computationally efficient scaling method, often outperforming direct training for very large systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.