Sirui Lu

Sirui Lu he/him

陆思锐

Physics for AI · AI for Physics

I’m a final-year physics PhD candidate at the Max Planck Institute of Quantum Optics and TU Munich, advised by J. Ignacio Cirac↗. I work at the intersection of quantum many-body physics and machine learning: physics for AI, and AI for physics.

My work runs in two directions. I build AI methods for quantum science: quantum simulation and algorithms, neural and tensor-network quantum states, and agent-driven discovery of quantum error-correcting codes. And I use physics and information theory to study AI: tensor networks and stochastic thermodynamics applied to modern generative models.

I also build the tools. I created TeXRA↗, a multi-agent research assistant for theorists that grounds LLM reasoning in Wolfram algebra and Lean 4 proof. I used it to co-design 14,116 certified quantum codes. With Max Welling and Lars Holdijk I co-authored Generative AI and Stochastic Thermodynamics (Cambridge University Press, 2026).

I am looking for research roles at industry AI labs and academic postdoc positions. Get in touch.

Selected publications

all publications →

Algorithms for Quantum Simulation at Finite Energies
Sirui Lu, Mari Carmen Bañuls, J. Ignacio Cirac,
PRX Quantum 2, 020321 • 2021
Abstract
We introduce two kinds of quantum algorithms to explore microcanonical and canonical properties of many-body systems. The first is a hybrid quantum algorithm that, given an efficiently preparable state, computes expectation values in a finite energy interval around its mean energy, using a filtering operator similar to quantum phase estimation and recovering physical values through interferometric measurements. Its computational time scales polynomially with the number of qubits, the inverse variance, and the inverse error, and it does not require long-time evolution. The second is a quantum-assisted Monte Carlo sampling method for microcanonical and canonical ensemble quantities that circumvents the sign problem of classical quantum Monte Carlo. Both can run on small quantum computers and analog quantum simulators.
Cite 
@article{Lu2021Algorithms,
  title = {Algorithms for Quantum Simulation at Finite Energies},
  author = {Lu, Sirui and Ba{\~n}uls, Mari Carmen and Cirac, J. Ignacio},
  year = 2021,
  month = may,
  journal = {PRX Quantum},
  volume = {2},
  number = {2},
  pages = {020321},
  doi = {10.1103/prxquantum.2.020321}
}
🏆 Discovered with TeXRA · 14,116 certified codes

Co-Designing Quantum Codes with Transversal Diagonal Gates via Multi-Agent Systems
Xi He, Sirui Lu, Bei Zeng,
arXiv:2510.20728 • 2025
Abstract
Exact scientific discovery requires more than heuristic search: candidate constructions must be turned into exact objects and checked independently. We extend TeXRA with an independent Lean 4 verification layer, turning it into a human-guided multi-agent platform that couples symbolic synthesis, combinatorial and linear-programming search, exact reconstruction of numerical candidates, and formal verification in Lean. Applying it to nonadditive quantum error-correcting codes with prescribed transversal diagonal gates, we obtain a Lean-certified catalogue of 14,116 distance-2 codes and resolve the transversal-T problem for several distance-3 codes, with all constructions, infinite families, and no-go results formalized and checked in Lean.
Cite 
@misc{He2025CoDesigning,
  title = {Co-Designing Quantum Codes with Transversal Diagonal Gates via Multi-Agent Systems},
  author = {He, Xi and Lu, Sirui and Zeng, Bei},
  year = 2025,
  month = oct,
  eprint = {2510.20728},
  archiveprefix = {arXiv},
  primaryclass = {quant-ph}
}
🏆 Oral at NeurIPS 2025 AI4Science

Can Theoretical Physics Research Benefit from Language Agents?
Sirui Lu, Zhijing Jin, Terry Jingchen Zhang, Pavel Kos, … Bernhard Schölkopf
ICML 2026 • 2026
Abstract
Large Language Models (LLMs) are rapidly advancing across diverse domains, yet their application in theoretical physics remains inadequate. While current models show competence in mathematical reasoning and code generation, we identify critical gaps in physical intuition, constraint satisfaction, and reliable reasoning that cannot be addressed through prompting alone. Physics demands approximation judgment, symmetry exploitation, and physical grounding that require AI agents specifically trained on physics reasoning patterns and equipped with physics-aware verification tools. We envision physics-specialized AI agents that handle multimodal data, propose physically consistent hypotheses, and autonomously verify theoretical results, and we call for collaborative efforts between physics and AI communities to build the specialized infrastructure necessary for AI-driven scientific discovery.
Cite 
@misc{Lu2025Can,
  title = {Can Theoretical Physics Research Benefit from Language Agents?},
  author = {Lu, Sirui and Jin, Zhijing and Zhang, Terry Jingchen and Kos, Pavel and Cirac, J. Ignacio and Sch{\"o}lkopf, Bernhard},
  year = 2026,
  month = jun,
  eprint = {2506.06214},
  archiveprefix = {arXiv},
  primaryclass = {cs.CL}
}
🏆 Editors' Suggestion

Tensor Networks and Efficient Descriptions of Classical Data
Sirui Lu, Márton Kanász-Nagy, Ivan Kukuljan, J. Ignacio Cirac,
Phys. Rev. A 111, 032409 • 2025
Abstract
We investigate the potential of tensor network based machine learning methods to scale to large image and text data sets. For that, we study how the mutual information between a subregion and its complement scales with the subsystem size L, similarly to how it is done in quantum many-body physics. We find that for text, the mutual information scales as a power law with a close to volume law exponent, indicating that text cannot be efficiently described by 1D tensor networks. For images, the scaling is close to an area law, hinting that 2D tensor networks such as PEPS could have adequate expressibility. For the numerical analysis, we introduce a mutual information estimator based on autoregressive networks, and we also use convolutional neural networks in a neural estimator method.
Cite 
@article{Lu2025Tensor,
  title = {Tensor Networks and Efficient Descriptions of Classical Data},
  author = {Lu, Sirui and {Kan{\'a}sz-Nagy}, M{\'a}rton and Kukuljan, Ivan and Cirac, J. Ignacio},
  year = 2025,
  month = mar,
  journal = {Phys. Rev. A},
  volume = {111},
  number = {3},
  pages = {032409},
  doi = {10.1103/PhysRevA.111.032409}
}

Selected projects

all projects →

Generative AI and Stochastic Thermodynamics

- Present

A book with Max Welling and Lars Holdijk on the free-energy unification of generative AI and stochastic thermodynamics. Cambridge University Press, July 2026.

Generative AIStochastic thermodynamicsFree energy
Website

TNLean: tensor networks, formally verified

- Present

Multi-agent autoformalization of tensor-network theory in Lean 4 / Mathlib: the fundamental theorem of matrix-product states, machine-checked.

Lean 4MathlibTensor networksFormal methods
Code Documentation

TeXRA: Multi-agent AI for theorists

- Present

Multi-agent AI research assistant (VS Code / Cursor / CLI) that grounds LLM reasoning in Wolfram algebra and Lean 4 formal proof.

LLM agentsTypeScriptLean 4WolframVS Code
Website

News

Experience

talks, service, full CV →