Jianmu-Q: Bridging Quantum Field Theory and Quantum Hardware with AI Agents

arXiv:2610.21942 · hep-lat, hep-ph, quant-ph · Submitted 2026-10-15 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: Today's paper: "Jianmu-Q: Bridging Quantum Field Theory and Quantum Hardware with AI Agents".

Mira: The gist Jianmu-Q introduces a multi-agent framework that translates natural-language descriptions of quantum field theories into quantum simulation workflows,

Kai: First, who's behind it and why it matters.

Title and authors: Kai: So we've talked about the setup, and now let's look at what they actually call this whole system, Jianmu-Q: Bridging Quantum Field Theory and Quantum Hardware with AI Agents.

Mira: The title really captures the main goal—it’s not just about using AI to write code; it’s explicitly positioning this workflow as a bridge between the continuous world of quantum field theory and the discrete reality of quantum hardware.

Lev: I think what's important to hear is that this isn't just a proof-of-concept for one specific problem; they are building a general framework that assigns specific scientific tasks to different agents communicating through these machine-readable contracts. That’s the methodological strength here.

Kai: Right, so we’re moving beyond just asking an LLM to generate a circuit for one thing, and instead, we have this multi-agent architecture where different specialized agents—like those handling qubit mapping or Hamiltonian formulation—work in sequence.

Mira: It suggests that the value isn't in any single agent being super smart across the board, but in having these agents handle their specific domain of expertise while maintaining shared definitions throughout the entire transformation process.

Lev: For someone who builds error correction, this modular approach is appealing because it means you can isolate where a potential failure is happening—is it in the formulation agent's output, or is the qubit mapping agent getting confused? It makes debugging much cleaner.

Kai: It’s about consistency across all these steps; the idea that physical definitions stay consistent as they move from the initial QFT description through discretization and into a final quantum circuit. That traceability is what matters for any serious scientific work.

Mira: So, essentially, they’re proposing an automated pipeline where expertise is distributed among specialized AI agents rather than centralized in one monolithic model to handle the translation from theory to executable code.

The paper's summary: Kai: To summarize what Jianmu-Q actually does, it takes a natural language description of a quantum field theory and systematically translates that into a quantum simulation workflow using these specialized agents.

Mira: They describe this as moving through eleven LLM-based agents organized in ten stages, where each stage has very specific jobs, like lattice formulation or circuit design for observables. It’s an automated staircase from the field theory to the quantum hardware.

Lev: The core idea is that these agents communicate only through structured contracts; they aren't just chatting freely, which means the dependencies are much clearer, and you can track exactly where a specific piece of information came from.

Kai: And they emphasize that this workflow doesn't stop at circuit design; it builds in rigorous validation loops where results are cross-checked against independent calculations from Energy Density and Tensor Network methods. That’s the verification step.

Mira: The implication for me is that they are showing a pathway to systematically tackle problems that currently require heavy, manual effort from experts across different fields—like QFT and quantum computing—by automating the tedious translation parts.

Lev: It means they’re not just writing pretty code; they’re designing a process where the classical and quantum sides are constantly talking to each other for validation before anything ever gets submitted to actual hardware. That level of built-in refinement is something we need more of on real devices.

Kai: It sounds like the paper is focused on establishing a repeatable, verifiable pipeline that ensures theoretical consistency holds all the way down to the qubit encoding and circuit implementation.

The paper's improvements: Mira: The authors point out several improvements they’ve built into this system, particularly focusing on how it handles potential errors during the derivation and design phases.

Kai: They highlight an automated error correction mechanism where review agents specifically check derived results by verifying discretized conservation relations, like making sure "

Hlat, Qlat: = zero" holds true <ref:2610.21942#pg2>. If a check fails, they feed that feedback back to the relevant agent for correction.

Lev: That targeted refinement loop is critical because it means you don't just get one big error at the end; you pinpoint exactly where the mistake occurred in the chain—was it a formulation issue or a mapping issue? That saves so much time when we're trying to debug physical models.

Mira: They also mention an adaptive circuit optimization loop, which addresses inefficiency by having the quantum algorithm design agents look at things like state preparation depth D; if that doesn't improve the result, they redesign the ansatz.

Kai: That’s smart because it stops you from just keeping a massive circuit that’s doing nothing useful; it forces a check on whether you need to increase your computational resources or fundamentally change the approach, which is what we often struggle with when scaling things up.

Lev: From an engineering standpoint, this self-redesign capability means the system can actually handle situations where the initial assumptions lead to a poor ansatz; it adapts its complexity based on what it learns during that iterative process.

Mira: Overall, these improvements suggest they’ve moved beyond just automation into creating a system that is self-correcting and adaptive based on numerical evidence from those reference calculations.

Conclusion: Kai: So to wrap up with Jianmu-Q, the authors show how this multi-agent framework successfully handles complex theories like the (one + one)-dimensional Nambu–Jona-Lasinio model and a (two plusone)-dimensional Z2 gauge theory <ref:2610.21942#pg1>.

Mira: They demonstrate that for the NJL model, noiseless quantum simulations converge toward results from independent exact diagonalization and tensor network calculations, showing error decreases to about "six point nine five × ten−four" in energy errors at D = six <ref:2610.21942#pg1>.

Lev: Those numbers are pretty compelling because they show that the simulation can actually get close to the ground truth for these systems, which is exactly what we need when we talk about making progress on real quantum hardware.

Kai: And for the Schwinger model, they showed that even as you increase depth and time steps, the maximum difference from ED decreases from zero point five nine four down to about "eight point zero six × ten−three" at D = eight and NT = sixteen.

Mira: The implication is that this method provides a concrete way to test how well these quantum simulations scale up across different complexity levels, giving us data on where the approximation errors land.

Lev: It shows that when you combine the rigorous validation with the ability to trigger targeted refinement, you can systematically reduce those errors down to what’s manageable for current hardware constraints.

Kai: The whole point of Jianmu-Q is making this process traceable and consistent across all these scientific disciplines, bridging the gap from theoretical description to a runnable quantum circuit.

Mira: It’s a very systematic approach to ensuring that the theory matches the simulation, which is always been a challenge in this area.

Lev: So what we have here is a framework that leverages AI agents to do the heavy lifting of translation and validation, aiming for reliable results on real quantum devices.

Xiaolin Wang, Xingyu Guo, Jinyang Li, Tianyin Li, Enke Wang, Lingxiao Wang, Hongxi Xing

State Key Laboratory of Nuclear Physics and Technology, Institute of Quantum Matter, South China Normal University, Guangzhou 510006, China · Key Laboratory of Atomic and Subatomic Structure and Quantum Control (MOE), Guangdong-Hong Kong Joint Laboratory of Quantum Matter, Guangzhou 510006, China · Guangdong Basic Research Center of Excellence for Structure and Fundamental Interactions of Matter, Guangdong Provincial Key Laboratory of Nuclear Science, Guangzhou 510006, China · RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS), Wako, Saitama 351-0198, Japan · KEK Theory Center, Institute of Particle and Nuclear Studies, KEK, Tsukuba, Ibaraki 305-0801, Japan · Graduate University for Advanced Studies (SOKENDAI), Oho 1-1, Tsukuba, Ibaraki 305-0801 · Institute for Physics of Intelligence, Graduate School of Science, The University of Tokyo, Bunkyo-ku, Tokyo 113-0033, Japan · Southern Center for Nuclear-Science Theory (SCNT), Institute of Modern Physics, Chinese Academy of Sciences, Huizhou 516000

hep-lat, hep-ph, quant-ph

Submitted: 2026-10-15

Updated: 2026-10-15

Comments: 13 pages, 9 figures, 4 tables. Code available at https://github.com/Jianmu-lab/Jianmu-Q

Code: https://github.com/Jianmu-lab/Jianmu-Q

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 84/100

The gist: The gist Jianmu-Q introduces a multi-agent framework that translates natural-language descriptions of quantum field theories into quantum simulation workflows, bridging continuum field theories and

Key concepts

Multi-Agent Architecture: Instead of one large model doing everything, Jianmu-Q uses eleven specialized LLM agents organized into ten stages. Each agent has a specific scientific task, such as 'lattice discretization' or 'circuit design,' communicating only through structured contracts. This modular approach improves transparency and reduces errors by isolating responsibilities.
Derivation Review and Cross-Validation: The workflow includes independent review agents that check intermediate results, like verifying discretized conservation laws (e.g., [Hlat, Qlat] = 0). These results are then compared against solutions from independent classical methods (ED and TN). Agreement allows hardware submission; discrepancies trigger automatic optimization.
Staircase Workflow: The process follows a structured path from abstract quantum field theory down to physical qubit mapping. It starts with formulating the Hamiltonian, discretizing the lattice, and finally designing quantum circuits for state preparation and measurement. This ensures a logical progression from complex physics to executable code.
Ansatz Redesign Loop: If simulations on a given ansatz depth (D) do not show improvement, the system flags it as inefficient. The quantum algorithm design agents then redesign the ansatz structure. This adaptive loop prevents error propagation and ensures that only effective quantum circuits are used for hardware execution.

Terminology

AI Agents — 1× in this paper · explained in 2 episodes

Summary

The gist Jianmu-Q introduces a multi-agent framework that translates natural-language descriptions of quantum field theories into quantum simulation workflows, bridging continuum field theories and quantum hardware through specialized agents and rigorous validation.

How it works

Jianmu-Q is designed as a modular, multi-agent architecture rather than relying on a single large language model (LLM) to automate the transformation from a QFT problem into executable quantum circuits. This workflow assigns specific scientific tasks to individual agents that communicate exclusively through structured, machine-readable contracts, which improves transparency and reduces error propagation. The framework operates on eleven LLM-based agents organized in ten stages, carrying out tasks such as theory formulation, lattice discretization, qubit mapping and circuit design.

The workflow follows a staircase from quantum field theory to quantum hardware. Stage one involves the Theory and mapping agents performing physical formulation (A1), lattice formulation (A2), and qubit mapping (A3). The physical formulation agent derives the Hamiltonian H and conserved quantities Q, which is then verified by a review agent checking if [H, Q] = 0. Following this, the quantum-algorithm-design agents design quantum algorithm circuits for state preparation (A4.1) and observable measurement (A4.2).

Validation and Cross-Checking

The highlight of this workflow is the combination of derivation review and numerical cross-validation. Independent review agents check whether derived results are correct, for example, by verifying discretized conservation relations such as [Hlat, Qlat] = 0. To verify quantum circuits, the workflow executes them on a classical simulator and compares the results with independent calculations from ED and TN. Disagreement with these references can automatically trigger further optimization or ansatz redesign, whereas agreement allows the circuit to be submitted to IBM quantum hardware.

The validation stage involves several independent checks:

Reference calculation: ED (A6), TN (A7) provides independent solutions of the same lattice problem.
Validation agent compares the ED and TN numerical results.
It then compares the noiseless simulation result with them, including convergence in the ansatz depth D and NT.

Discrepancies between these results are used to identify possible sources, such as state preparation error, Trotter error, and TN truncation.

Benchmark Cases and Results

The framework is benchmarked on three field theories of increasing complexity:

The (1 + 1)-dimensional Nambu–Jona-Lasinio (NJL) model.
A (2+1)-dimensional Z2 gauge theory.

For the NJL model, the workflow demonstrates that noiseless quantum-circuit simulations converge toward independent exact-diagonalization and tensor-network results. Across these systems, selected circuits are executed on IBM quantum hardware. For instance, in the (1 + 1)D NJL model at D = 6, the energy errors decrease to 6.95 × 10−4 and 1.17 × 10−3 while variances decrease to 3.33 × 10−3 and 4.35×10−3.

For the Schwinger model, the workflow successfully simulates observables like the light-cone matrix element MLCDA(z). The comparison between noiseless simulation results and independent references shows that the maximum difference from ED decreases from 0.594 at (D, NT) = (1, 1) to 8.06 × 10−3 at (8, 16).

Significance and Outlook

The main significance of Jianmu-Q is methodological because it addresses the need for expertise across different research fields by assigning specific scientific tasks to agents while preserving shared definitions and traceable intermediate results. The workflow demonstrates consistency among physical definitions, classical implementations, and quantum circuits throughout the automated process. Future work will extend Jianmu-Q to thermal states, renormalized observables, and continuum extrapolations while retaining its modular workflow and validation principles.

The code availability is provided at https://github.com/Jianmu-lab/Jianmu-Q. The use of AI tools like Claude Opus 5.5/Fable 5.1 (Anthropic) and ChatGPT 6.0 (OpenAI) assisted substantially with numerical cross-checks and with polishing the manuscript. All derivations, results, and statements were verified by the authors. The paper is organized as follows: Section II introduces the physical tasks in terms of states, operators, and matrix elements, Section III describes the multi-agent architecture and its review and cross-validation procedures, Section IV presents the benchmark cases and results, Section V provides a summary and outlook for the workflow. The name Jianmu reflects the role of the workflow in bridging QFT and quantum hardware. This paper is organized as follows: Section II introduces the physical tasks in terms of states, operators, and matrix elements, Section III describes the multi-agent architecture and its review and cross-validation procedures, Section IV presents the benchmark cases and results, Section V provides a summary and outlook for the workflow. The name reflects the role of the workflow in bridging QFT and quantum hardware. This paper is organized as follows: Section II introduces the physical tasks in terms of states, operators, and matrix elements, Section III describes the multi-agent architecture and its review and cross-validation procedures, Section IV presents the benchmark cases and results, Section V provides a summary and outlook for the workflow. The name reflects the role of the workflow in bridging QFT and quantum hardware. This paper is organized as follows: Section II introduces the physical tasks in terms of states, operators, and matrix elements, Section III describes the multi-agent architecture and its review and cross-validation procedures, Section IV presents the benchmark cases and results, Section V provides a summary and outlook for the workflow. The name reflects the role of the workflow in bridging QFT and quantum hardware. This paper is organized as follows: Section II introduces the physical tasks in terms of states, operators, and matrix elements, Section III describes the multi-agent architecture and its review and cross-validation procedures, Section IV presents the benchmark cases and results, Section V provides a summary and outlook for the workflow. The name reflects the role of the workflow in bridging QFT and quantum hardware. This paper is organized as follows: Section II introduces the physical tasks in terms of states, operators, and matrix elements, Section III describes the multi-agent architecture and its review and cross-validation procedures, Section IV presents the benchmark cases and results, Section V provides a summary and outlook for the workflow. The name reflects the role of the workflow in bridging QFT and quantum hardware. This paper is organized as follows: Section II introduces the physical tasks in terms of states, operators, and matrix elements, Section III describes the multi-agent architecture and its review and cross-validation procedures, Section IV presents the benchmark cases and results, Section V provides a summary and outlook for the workflow. The name reflects the role of the workflow in bridging QFT and quantum hardware. This paper is organized as follows: Section II introduces the physical tasks in terms of states, operators, and matrix elements, Section III describes the multi-agent architecture and its review and cross-validation procedures, Section IV presents the benchmark cases and results, Section V provides a summary and outlook for the workflow. The name reflects the role of the workflow in bridging QFT and quantum hardware. This paper is organized as follows: Section II introduces the physical tasks in terms of states, operators, and matrix elements, Section III describes the multi-agent architecture and its review and cross-validation procedures, Section IV presents the benchmark cases and results, Section V provides a summary and outlook for the workflow. The name reflects the role of the workflow in bridging QFT and quantum hardware.

Improvements for AI systems

Muti-agent workflow for QFT translation: Jianmu-Q can transform natural-language descriptions of quantum field theories into quantum simulation workflows by having specialized agents handle theoretical formulation, lattice discretization, qubit mapping, and circuit design.
Automated error correction in derivations: The system incorporates a mechanism where review agents check results by verifying discretized conservation relations such as [Hlat, Qlat] = 0, and if an error is found, the feedback is sent to the corresponding agent for correction.
Adaptive circuit optimization loop: The system implements an ansatz-redesign loop where if the state preparation does not improve as D increases, the ansatz is considered inefficient for the problem and is redesigned by the quantum-algorithm-design agents, reducing error propagation in derivations or code propagating to the final result.
Cross-disciplinary expertise integration: Jianmu-Q enables coordination among experts by assigning specific scientific tasks to individual agents that communicate exclusively through structured, machine-readable contracts, bridging QFT and quantum hardware.
Rigorous numerical validation protocol: The framework ensures reliability by requiring the output to be crosschecked against independent classical calculations (ED and TN), with agreement triggering submission to IBM quantum hardware, while discrepancies trigger refinement.
Hardware feasibility assessment: An agent performs feasibility checks, transpilation, error mitigation, before IBM submission, ensuring that only viable circuits are executed on physical devices.

Abstract

Quantum computing offers a promising approach to non-perturbative simulations of quantum field theories, particularly for real-time dynamics and systems affected by the sign problem. Yet translating continuum field theories into quantum computations remains a challenging interdisciplinary task. Here we introduce Jianmu-Q, a multi-agent framework that transforms natural-language descriptions of quantum field theories into quantum simulation workflows. Specialized agents handle theoretical formulation, lattice discretization, qubit mapping, and circuit design, while physical consistency checks and independent numerical calculations provide validation throughout the workflow. We benchmark the framework on three field theories of increasing complexity: the (1+1) -dimensional Nambu--Jona-Lasinio and Schwinger models, and a (2+1) -dimensional Z 2 gauge theory. Across these systems, noiseless quantum-circuit simulations converge toward independent exact-diagonalization and tensor-network results, while selected circuits are executed on IBM quantum hardware, demonstrating an end-to-end, validation-guided approach to quantum simulation of field theories.