arXiv subjects

Paul Saxe

Publications and source records attributed to Paul Saxe.

3 recordsLinked to original sources

arXiv2026

BERTology of Molecular Property Prediction

Chemical language models (CLMs) have emerged as promising competitors to popular classical machine learning models for molecular property prediction (MPP) tasks. However, an increasing number of studies have reported inconsistent and contradictory results for the performance of CLMs across various MPP benchmark tasks. In this study, we conduct and analyze hundreds of meticulously controlled experiments to systematically investigate the effects of various factors, such as dataset size, model size, and standardization, on the pre-training and fine-tuning performance of CLMs for MPP. In the absence of well-established scaling laws for encoder-only masked language models, our aim is to provide comprehensive numerical evidence and a deeper understanding of the underlying mechanisms affecting the performance of CLMs for MPP tasks, some of which appear to be entirely overlooked in the literature.

cs.LG↗

arXiv2025

SEAMM: A Simulation Environment for Atomistic and Molecular Modeling

The Simulation Environment for Atomistic and Molecular Modeling (SEAMM) is an open-source software package written in Python that provides a graphical interface for setting up, executing, and analyzing molecular and materials simulations. The graphical interface reduces the entry barrier for the use of new simulation tools, facilitating the interoperability of a wide range of simulation tools available to solve complex scientific and engineering problems in computational molecular science. Workflows are represented graphically by user-friendly flowcharts which are shareable and reproducible. When a flowchart is executed within the SEAMM environment, all results, as well as metadata describing the workflow and codes used, are saved in a datastore that can be viewed using a browser-based dashboard, which allows collaborators to view the results and use the flowcharts to extend the results. SEAMM is a powerful productivity and collaboration tool that enables interoperability between simulation codes and ensures reproducibility and transparency in scientific research.

physics.chem-ph↗

arXiv2020

Quantifying uncertainty in high-throughput density functional theory: a comparison of AFLOW, Materials Project, and OQMD

A central challenge in high throughput density functional theory (HT-DFT) calculations is selecting a combination of input parameters and post-processing techniques that can be used across all materials classes, while also managing accuracy-cost tradeoffs. To investigate the effects of these parameter choices, we consolidate three large HT-DFT databases: Automatic-FLOW (AFLOW), the Materials Project (MP), and the Open Quantum Materials Database (OQMD), and compare reported properties across each pair of databases for materials calculated using the same initial crystal structure. We find that HT-DFT formation energies and volumes are generally more reproducible than band gaps and total magnetizations; for instance, a notable fraction of records disagree on whether a material is metallic (up to 7%) or magnetic (up to 15%). The variance between calculated properties is as high as 0.105 eV/atom (median relative absolute difference, or MRAD, of 6%) for formation energy, 0.65 {\AA}$^3$/atom (MRAD of 4%) for volume, 0.21 eV (MRAD of 9%) for band gap, and 0.15 $\mu_{\rm B}$/formula unit (MRAD of 8%) for total magnetization, comparable to the differences between DFT and experiment. We trace some of the larger discrepancies to choices involving pseudopotentials, the DFT+U formalism, and elemental reference states, and argue that further standardization of HT-DFT would be beneficial to reproducibility.

cond-mat.mtrl-sci↗