arXiv subjects

Etinosa Osaro

Publications and source records attributed to Etinosa Osaro.

4 recordsLinked to original sources

arXiv2026

Agentic AI for Density-Functional Development: Revisiting r2SCAN

We demonstrate physics-constrained agentic development of a meta-generalized gradient approximation (meta-GGA) functional using a large language model (LLM) to assist the search and optimization of a band-gap-oriented revision of r2SCAN. No real bonded systems were fitted, preserving r2SCAN's nonempirical philosophy. Across the finalist set, band gaps and several molecular subsets improve relative to r2SCAN; the top finalist, r2SCAN+, reduces the band-gap MAE on a benchmark comprising 24 solids from 1.26 to 0.96 eV and the aggregate MAE on 329 molecular properties from 4.79 to 4.42 kcal/mol. We first curated 78 exchange and 90 correlation candidate correction terms from r2SCAN's dimensionless ingredients, spanning polynomial terms through third degree, exponentials, exponentially damped products, and ratios. Allowing each candidate to combine one to three correction terms from the exchange catalog, the correlation catalog, or both yields about 8 x 10^5 distinct forms, making exhaustive high-throughput screening impractical. We defined the search criteria for the LLM agent using r2SCAN's exact constraints, physical norms, and the targeted iso-orbital derivative response. The agent then combined these criteria with its pretrained knowledge and accumulated search feedback to propose and refine sparse forms, prioritizing terms tied to the iso-orbital response; a second LLM critic screened proposals before deterministic verification. Compared with uniform random search, the workflow learned from prior evaluations, incurred far fewer downstream rejections (0.6% versus 24.6%), and located stronger high-response candidates: 51 agentic candidates exceeded the best random-search response of 1.263, with the overall best reaching 1.331. These results show that agentic search can support density-functional development when flexible hypothesis generation is coupled to automated physical verification.

physics.chem-ph↗

arXiv2026

RLEASE: Reinforcement Learning Efficient Active Space Engine

Selecting the active space for multireference electronic-structure calculations is a long-standing bottleneck that often requires expert chemical intuition and costly trial-and-error. We introduce RLEASE (Reinforcement Learning Efficient Active Space Engine), a low-cost method for automatic, geometry-dependent active-space selection. A neural network predicts per-orbital diagnostic scores ($\hat{s}_{1}$) from inexpensive Hartree-Fock orbital descriptors, and a learned threshold partitions orbitals into active and inactive sets. The threshold policy is optimized with proximal policy optimization, using the discrepancy between sc-NEVPT2 energies computed with the selected active space and DMRG reference energies as the reward. After training, the same RLEASE-selected active spaces can be used with multireference perturbation theory or composite coupled-cluster energy estimators. Despite being trained on a small set of molecules and geometries, RLEASE transfers to chemically diverse test systems, producing compact active spaces and competitive potential-energy surfaces relative to established entropy-based selectors. Because deployment requires only inexpensive orbital descriptors and neural-network inference, RLEASE enables high-throughput multireference workflows without molecule-specific retraining or target-system pilot DMRG calculations.

physics.chem-ph↗

arXiv2026

MLIPilot: LLM-Driven Auto-Research for Machine-Learned Interatomic Potentials

Constructing production-quality machine-learned interatomic potentials (MLIPs) requires balancing accuracy, dynamical stability, and computational throughput under constraints that are not captured by a single training loss. We introduce MLIPilot, an auto-research framework in which tool-calling large language models propose hypotheses, edit MLIP training code, launch HPC jobs, and accept or revert changes using a fixed, physically constrained scorecard. We evaluate MLIPilot on MACE potential optimization using both commercial and open-weight LLM agents, including GPT-5.5, GPT-4.1, Mistral-24B, and Qwen3-32B. The benchmarks span molecular and periodic settings: a QM7-derived dataset for which we generated B3LYP/6-31G(d) energies and forces, and a Cu EMT dataset with periodic copper supercells labeled by ASE's Effective Medium Theory calculator. Across these benchmarks, the strongest agents move initially constraint-violating baselines to accepted models by discovering useful training strategies, including output normalization, loss-function changes, progressive training schedules, and model-capacity adjustments. These results suggest that LLM agents can serve as autonomous operators for scientific machine-learning workflows when their search is constrained by domain-specific validation criteria, shifting part of MLIP development from manual trial-and-error toward auditable, automated experimentation.

physics.chem-ph↗

arXiv2026

A Matched Spectral Benchmark of Quantum Inspired Feature Maps

Quantum machine learning is often motivated by the idea that quantum systems can expose useful high-dimensional structure that is difficult to access with classical models. We isolate one central component of this claim: the fixed data-encoding map. Amplitude, angle, and basis encoding are evaluated as deterministic feature maps for classical supervised learning under matched output dimensionality and strong classical controls. The benchmark compares these encodings against raw linear models, random Fourier features, polynomial features, PCA, RBF SVMs, and shallow neural networks across diverse classical datasets. Rather than treating performance as a single endpoint, we analyze the geometry of each representation through effective rank, condition number, centered kernel alignment, predictive performance, and practical overhead. The resulting picture is mechanistic: amplitude encoding can remove magnitude information through unit-sphere normalization, angle encoding can become geometrically redundant with raw linear features, and basis encoding can impose a binary Hamming geometry that is poorly aligned with smooth decision structure. These findings do not argue against quantum computation, however, they show that fixed quantum-inspired encoding geometry alone is not a reliable source of machine-learning advantage on classical data.

quant-ph↗