QUASAR: A Universal Autonomous System for Atomistic Simulation and a Benchmark of Its Capabilities

Area of Science:

• Materials Science
• Computational Chemistry
• Artificial Intelligence

Background:

• Current agentic AI systems for materials science are limited by rigid, domain-specific tool-calling paradigms and narrowly scoped agents.
• Streamlining complex computational workflows in materials science remains a significant challenge.

Purpose of the Study:

• To introduce QUASAR, a universal autonomous system for atomistic simulations.
• To facilitate production-grade scientific discovery by autonomously orchestrating complex multiscale workflows.
• To enable real-world research scenarios without human intervention.

Main Methods:

• QUASAR autonomously orchestrates multiscale workflows across Density Functional Theory (DFT), machine learning potentials, Molecular Dynamics (MD), and Monte Carlo (MC) simulations.
• The system employs adaptive planning, context-efficient memory management, and hybrid knowledge retrieval.
• Benchmarking involved a series of tiered tasks, from routine to frontier research challenges like photocatalyst screening.

Main Results:

• QUASAR demonstrated the ability to function as a general atomistic reasoning system, not merely a task-specific automation framework.
• The system successfully navigated complex multiscale workflows autonomously.
• Performance was evaluated across routine tasks and advanced research challenges, including novel material assessment.

Conclusions:

• QUASAR represents a significant advancement in autonomous systems for atomistic simulations.
• The findings suggest potential for agentic AI deployment in computational chemistry research.
• Further development is needed to fully realize the capabilities of such AI systems in scientific discovery.