Quantum simulations, the computer models used to study systems governed by quantum mechanics, often demand specialist numerical methods and fluency with multiple software tools. A new AI system called El Agente Cuántico is being developed to automate parts of that work, delegating tasks to specialist agents tied to different software packages and producing code, numerical results and figures.
The system was demonstrated on a range of problems, including molecular-energy calculations, entangled states, dissipative dynamics, time crystals, quantum error correction and hardware-resource estimation. In repeated tests of five tasks, its average scores in expert assessments ranged from 97 to 100 out of 100. A small number of runs nevertheless departed from the instructions, for example by calculating the wrong observable or preparing an initial state using the wrong sampling procedure.
The approach reflects a broader push to build AI agents that can plan multistep tasks, consult documentation, run programs, examine results and revise their approach. In physics, experimental systems such as SciExplorer have inferred equations and Hamiltonians from simulated observations, and related agent-based methods are being explored in mathematics, including the search for mathematical proofs.
Quantum simulations are used to calculate molecular energies, investigate new materials, model energy transfer and decoherence, test quantum-computing algorithms and predict how quantum systems evolve over time. Such calculations can provide information that is difficult to obtain analytically or experimentally, but they place a heavy burden on researchers who must master both the physics and the software syntax.
The authors present the agents as an aid to scientific judgement rather than a replacement for it. The work suggests that researchers could concentrate more on physical assumptions and interpretation, and less on the mechanics of running code. The findings are published in Reports on Progress in Physics by I. Gustin and colleagues.
15
