arXiv:2610.32001
QuantumMatterExplorer: An Autonomous Agent for the Discovery of Quantum Many-Body Phases of Matter
Jannes Nys, Andrew Jreissaty, Juan Carrasquilla
Abstract
Mapping the phase diagram of a new quantum Hamiltonian demands expert judgement at every step: which observables to measure, where in parameter space to compute, or how to separate genuine order from finite-size artefacts. We introduce QuantumMatterExplorer, an autonomous agent that takes only a Hamiltonian and its parameter ranges as input and returns a phase diagram obtained from its own numerical simulations. The agent is built on a large language model and is designed to reason over a long horizon, maintain a persistent written plan, and drive a set of computational tools, with density-matrix renormalization group (DMRG) calculations cross-checked against independent exact diagonalization. The agent combines a large language model with persistent planning, a library of expert computational-physics skills, and density-matrix renormalization group calculations validated against independent exact diagonalization. It adaptively selects quantities to evaluate, refines parameter scans, and performs finite-size analyses as evidence accumulates. After validating the approach on Hamiltonians with known transitions, we challenge it with a frustrated bilinear-biquadratic spin-1 chain with single-ion anisotropy, whose two-dimensional phase diagram was published only recently and is unlikely to appear in any training corpus. Given no hint about the expected physics, the agent proposes which observables to probe, and identifies candidate Néel, Haldane and dimerized phases, in broad agreement with the reference phase diagram. It also detects incommensurate correlations and infers the frustration-dependent geometry of the phase boundaries, all within an eight-hour compute budget. These results open a practical route to the autonomous computational discovery of new phases of quantum matter.
AI Summary
QuantumMatterExplorer asks whether an LLM-based agent can autonomously explore a quantum many-body phase diagram from a Hamiltonian and parameter ranges, rather than reproduce specified results. It combines persistent planning, curated computational-physics skills, tool execution, and isolated analysis and critic subagents. The agent writes DMRG simulations, validates implementations against independent exact diagonalization and known limits, and adaptively chooses observables, parameter scans, and finite-size analyses. Benchmarks use the anisotropic XY and spin-$1/2$ $J_1$–$J_2$ chains. The tested backends reproduce the XY transition but have greater difficulty locating the BKT dimerization transition at $J_2/J_1\approx0.2411$. An ablation finds that ITensor-specific guidance reduces coding failures for one local model; another backend fails to construct a working DMRG pipeline even with that guidance. The main demonstration is one eight-hour run on a frustrated bilinear-biquadratic spin-1 chain with single-ion anisotropy, at $\theta=0.2\pi$, exploring $J_2\in[0,1]$ and $D\in[-4,4]$. It performs 222 DMRG calculations at 83 parameter coordinates, reaching $N=60$, with ED checks through $N=8$. Magnetic, bond, string-order, entanglement, and gap diagnostics support candidate Néel, dimerized, Haldane, and large-$D$ regions. It also detects varying correlation wavevectors and frustration-dependent shifts of candidate boundaries. The phase assignments remain provisional: finite-size evidence does not establish the full thermodynamic diagram. The agent misses trimerization, cannot establish gaplessness in floating regions, and does not reliably determine transition types. Several points remain unresolved. The plotted regions involve subsequent interpolation, and the recently published reference diagram is only presumed absent from model training data.
The research note generated by \sysname for the spin-1 chain of Eq.~\eqref{eq:blbq_main_body} is reproduced in its entirety at the end of this manuscript, following the references.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 42 pages, 10 figures and 3 tables in the main manuscript. Includes a 13-page agent-generated research note