Mathematical & physical sciences · OECD FoS 1.3 · China MoE 0702 物理学
Physics
物理学
L2Local breakthroughs ↑ Rising Score 1.90 · Last reviewed 2026-10-21 · Next review 2027-01-19
BIGGEST BREAKTHROUGH 2024-10-08 AI leapField · major
Nobel Prize in Physics to Hopfield & Hinton (foundations of neural networks) ↗
First physics Nobel for neural-network foundations
Why this level
Peer-reviewed results in quantum error-correction decoding (AlphaQubit, Nature 2024), autonomous law discovery (SciExplorer, PRX 2026) and particle-model building (Commun. Phys. 2026); not yet the mainstream way physics is done.
S = 0.30·method + 0.35·discovery + 0.25·adoption + 0.10·education (education scores 0 when there is no evidence; if empty, the other weights are renormalised)
Sources
Composite 1.5–2.49: at least one peer-reviewed/officially verified AI-key result, or one dimension at 3; routine workflow largely unchanged. All levels →
Research directions
Quantum error correction & controlAutonomous law discoveryML potentials & simulationCollider data analysis
Papers
- TheoPhysEval: Assessing AI agents at autonomous theoretical physics problem solving
AI: first benchmark of unpublished research-level theory problems for autonomous agents (98% with tools); Field: problems are routine calculations, not discoveries
2026-10-09 · arXiv
- Agentic Exploration of Physics Models
LLM agent infers equations of motion from data (PRX)
2026-07-08 · Physical Review X
- Machine-learning-guided discovery of kagome superconductors
YRu
3
B
2
and
LuRu
3
B
2
AI-predicted superconductors confirmed by three independent measurements
2026-06-17 · Physical Review Research
- Towards AI-assisted neutrino flavor theory design
RL search over flavour models
2026-05-01 · Communications Physics
- Machine-Learned Renormalization-Group-Improved Gauge Actions and Classically Perfect Gradient Flows
ML enters core lattice-QCD computation; <1% discretisation effects on coarse lattices
2026-01-21 · Physical Review Letters
- Learning high-accuracy error decoding for quantum processors
Beats state-of-the-art decoders on real Sycamore data
2024-11-20 · Nature
- AI Feynman: A physics-inspired method for symbolic regression
Rediscovered all 100 Feynman-lecture equations from data
2020-04-17 · Science Advances
- Automated Discovery of Physically Interpretable Subgrid Length Scales for Detached Eddy Simulation via Agentic AI
Four LLM agents propose, review, implement and evaluate candidate subgrid length scales for detached eddy simulation from a single modelling idea.
2026-10-21 · arXiv · Jiawei Chen, Xiang I. A. Yang, Ping He
- Mesoscopic learning dynamics of extensive-width neural networks
Analyses how SGD learns internal representations in two-layer networks when width and input dimension grow proportionally.
2026-10-16 · arXiv · Shu Zhou, Jean Barbier, Sebastian Goldt
- Active Learning for Interatomic Potentials: A Review of Uncertainty-Driven Data Efficiency
Reviews uncertainty-driven active learning that closes the loop for building accurate ML interatomic potentials.
2026-10-16 · arXiv · Ilya V. Chepkasov, Alexander V. Shapeev, Alexander G. Kvashnin
- Properties and challenges of machine learning closures for ten-moment fluid models in the case of Landau damping
Uses ML to discover heat-flux closures for ten-moment plasma fluid models as an alternative to analytic Hammett–Perkins closures.
2026-10-16 · arXiv · Emanuel Jeß, Simon Lautenbach, Sophia Köhne, George Miloshevich
- When do quantum views help? Complementary classical-quantum representations in multi-view fusion
Proposes fusing classical and quantum ML representations when they capture complementary predictive relationships.
2026-10-15 · arXiv · Azadeh Alavi, Fatemeh Kouchmeshki, Hossein Akhoundi, Yongli Re
- DiffGA: Steering Frozen Diffusion Models for Materials Design via Evolutionary Search
Steers a frozen diffusion model with evolutionary search towards black-box materials-design objectives without retraining.
2026-10-10 · arXiv · Shivang Agarwal, Rodrigo Wang, Kevin Ryczko, Aayush R. Singh
- Pressure-transferable universal machine-learning potentials: the WBM-MatPES benchmark and Tuga-FF
Releases WBM-MatPES (182k structures) and reference hulls to benchmark universal ML interatomic potentials across pressures.
2026-10-10 · arXiv · Pierre-Paul De Breuck, Thalis H. B. da Silva, Antoine Loew, Silvana Botti
- From arXiv, bioRxiv, medRxiv and OpenAlex (last ~30 days); summaries written by AIMap from the abstract.
- From Research Goals to Scientific Closure: Evaluating Agents Through Evidence-Supported Physics Research 2026-10-15
- Into the Unknown: Benchmarking AI agents on open scientific research problems 2026-10-13
- Tetris3D: 3D Scene Generation With Objects That Fit Together 2026-10-07
- Optimizing the Optimizer: Language Models Discover Faster Molecular Relaxation Algorithms 2026-10-05
- All related papers in the archive →
Milestones 11· 1 leaps
Big glowing nodes are leaps: AI leap cool · Field leap warm · double ring = both
-
Discovery
Anthropic: Claude designs more efficient quantum circuits, ~17× fewer gates for elliptic-curve point additionAI: from plain-language specs the model beat prior circuits on 8 tasks (arXiv 2610.22367); Field: cuts gate/qubit costs for fault-tolerant algorithms (vendor, not peer-reviewed)
AnthropicVendor claimPreprint -
Industry
CERN and Mistral sign a statement of intent to build the first foundation model for collider physicsAI: one model pretrained on billions of simulated LHC collisions; Field: CERN officially pursues replacing the staged reconstruction chain (statement of intent only)
CERN -
Benchmark
TheoPhysEval: 50 unpublished research-level theory problems; GPT-6 Astra agent solves 49AI: first benchmark of unpublished research-level theory problems for autonomous agents (98% with tools); Field: problems are routine calculations, not discoveries
arXivPreprint -
Paper
AI Theorist proposes new exciton model for α-RuCl3 data (preprint)AI theorist proposes a new model; not yet peer-reviewed
arXivPreprint -
Paper
SciExplorer: LLM agent infers equations of motion and Hamiltonians (Physical Review X)LLM agent infers equations of motion from data (PRX)
Physical Review X -
Paper
ML-guided discovery of kagome superconductors YRu3B2 and LuRu3B2 (PR Research)AI-predicted superconductors confirmed by three independent measurements
Physical Review Research -
Paper
AMBer: RL agent builds neutrino-flavour models (Communications Physics)RL search over flavour models
Communications Physics -
Paper
Machine-learned RG-improved gauge action with classically perfect gradient flow (PRL)ML enters core lattice-QCD computation; <1% discretisation effects on coarse lattices
Physical Review Letters -
Paper
AlphaQubit decodes quantum error-correcting codes better than prior decoders (Nature)Beats state-of-the-art decoders on real Sycamore data
Nature -
PrizeAI leap
Nobel Prize in Physics to Hopfield & Hinton (foundations of neural networks)First physics Nobel for neural-network foundations
Nobel Prize -
Paper
AI Feynman: physics-inspired symbolic regression (Science Advances)Rediscovered all 100 Feynman-lecture equations from data
Science Advances
No milestones match this filter.
More papers & repos 8
| Type | Title | Date |
|---|---|---|
| Paper | TheoPhysEval: Assessing AI agents at autonomous theoretical physics problem solving Preprint | 2026-10-09 |
| Paper | The AI Theorist reveals excitonic structure in α-RuCl3 (arXiv 2610.02417) Preprint | 2026-10-01 |
| Paper | Agentic Exploration of Physics Models | 2026-07-08 |
| Paper | Machine-learning-guided discovery of kagome superconductors YRu 3 B 2 and LuRu 3 B 2 | 2026-06-17 |
| Paper | Towards AI-assisted neutrino flavor theory design | 2026-05-01 |
| Paper | Machine-Learned Renormalization-Group-Improved Gauge Actions and Classically Perfect Gradient Flows | 2026-01-21 |
| Paper | Learning high-accuracy error decoding for quantum processors | 2024-11-20 |
| Paper | AI Feynman: A physics-inspired method for symbolic regression | 2020-04-17 |
Latest news 9
- Measurement-efficient energy estimator for quantum chemistry of photodynamic materialsarXiv physics.chem-phenPaperPreprint
- 光量子计算，距离真正的产业化还有多远？雷峰网zh
- Towards a quantum computer that learns from its errorsGoogle Research Blogen