← News

Agents & MCPSources: arXiv · 18 Nov 2026

BORN framework predicts quantum chemistry code performance before running large calculations

An arXiv paper introduces BORN, an agentic computing framework that estimates a computational strategy's accuracy and runtime in advance, then generates task-specific quantum chemistry programs.

In plain words

Quantum chemistry calculations are usually run with general-purpose software, which can be too slow or costly for very large molecules. BORN is a framework that uses coding agents to write programs tailored to a single task. Before running anything, it predicts how accurate and how fast each candidate strategy will be, then picks one that fits the required accuracy and time budget.

What was announced

The paper argues that general-purpose quantum chemistry software leaves many large-scale calculations outside practical resource budgets, while coding agents make task-specific programs cheap to write but not necessarily efficient. BORN addresses this by anticipating each strategy's behavior before execution: it extrapolates error from representative small systems and predicts runtime from the target-scale workload and measured hardware throughputs. It then decomposes the preset schemes of existing packages into independently configurable algorithmic components, selects the fastest strategy within the error tolerance from a predicted accuracy-runtime Pareto frontier, and guides coding agents to compile it with component-level performance predictions.

The authors report several results. For a Hartree-Fock calculation with 50,024 basis functions, BORN finished in 1.41 hours on a single CPU node, which the authors describe as 77 times faster than the estimated runtime of a general-purpose baseline, or 9.8 times faster when development time is included. On a single GPU, BORN completed a ubiquitin calculation in 2.1 hours, the same order of runtime as the GPU-accelerated package Mako on 64 GPUs, and shortened a B-DNA duplex calculation from an estimated two months to 8.1 hours, a 173-fold reduction. These figures are the authors' own estimates and comparisons and have not been independently verified.

What it does not show

  • The source is the paper's abstract only; full method details, experimental setup and peer review status are not covered.
  • Reported speedups rest on the authors' own baseline estimates and hardware measurements, with no independent confirmation provided.

For security and IT teams

Scope
The available evidence is limited to the paper's abstract, covering the framework's stated approach and headline results only.

Sources

Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 19 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.