Physics Foundation
Exact structures from quantum field theory and string theory provide the mathematical backbone.
Postdoctoral Associate, Virginia Tech · Scientific Consultant, Anthropic
Xingyang Yu
Theoretical physics × verifier-guided scientific AI.
I am a theoretical physics postdoc working on quantum field theory, string theory, and generalized symmetries. I am currently developing a research direction on verifier-guided AI systems for scientific reasoning, using exact structures in quantum field theory — symmetries, anomalies, line operators, dualities, and consistency conditions — as machine-checkable feedback for AI agents.
Machine-Checkable Feedback
Symmetries, anomalies, line operators, dualities, and consistency conditions become verifiers.
Scientific Reasoning
AI agents can be evaluated on proposing, checking, and repairing structured reasoning steps.
Roles
Appointments
Apr 2026 – Present
Scientific Consultant, Anthropic
Part-time consulting on research-level theoretical physics problems and evaluations for advanced reasoning, with emphasis on high-energy theory, supersymmetric gauge theories, dualities, and generalized symmetries.
Aug 2023 – Present
Postdoctoral Associate, Department of Physics, Virginia Tech
Research in formal quantum field theory and string theory, focusing on generalized and non-invertible symmetries, SymTFTs, anomalies, brane engineering, and dualities.
Jul 2021 – Jul 2023
Visiting Scholar, University of Pennsylvania
Long-term research visit in the Department of Physics and Astronomy, collaborating on generalized symmetries, non-invertible symmetries, SymTFTs, and brane engineering.
Research Program
Research
Quantum Field Theory and String Theory
My work in theoretical physics focuses on generalized global symmetries, non-invertible symmetries, SymTFTs, anomalies, dualities, and string-theoretic constructions of quantum field theories. I am particularly interested in how categorical and topological structures organize quantum field theory data.
Verifier-Guided AI for Scientific Reasoning
I am developing a new direction that treats theoretical physics as a high-precision testbed for scientific AI. The goal is to turn exact structures in quantum field theory — symmetry constraints, anomaly matching, line-operator algebra, duality checks, and consistency conditions — into feedback environments where AI agents can propose, check, and repair scientific reasoning steps.
AI-Facing Research Directions
Projects
QFTCert
A verifier-guided research direction for scientific conjecture generation in quantum field theory, pairing AI-generated physics reasoning with formal, symbolic, or exact-constraint checks. An early milestone is a Seiberg-duality checker for controlled tests of research-level reasoning.
Physics4AI
A broader research question: can concepts and organizing principles from theoretical physics — renormalization group flow, effective theories, phase structure, symmetries, constraints, and coarse-graining — help build conceptual and mathematical frameworks for AI systems?
Grading the Unspoken
An expert-curated study of tacit reasoning in quantum field theory and string theory with large language models, focusing on implicit structural constraints and representation selection.
Exact-Constraint Benchmarks
A planned family of small, verifier-friendly tasks for AI agents, using line-operator algebra, anomaly matching, duality checks, and other exact QFT structures as machine-checkable feedback.
Selected Work
Selected Publications
Contact and Profiles