Agentic AI in Mathematics
This project aims to develop agentic AI systems for mathematical research: autonomous discovery of patterns and conjectures, design and execution of computational experiments, and integration with formal verification tools.
IOL Project Members
Funding
This project is being funded by the Berlin Mathematics Research Center MATH+ (project ID EF-LiOpt-3), itself funded by the German Research Foundation (DFG) under Germany's Excellence Strategy (EXC-2046/1, project ID 390685689) from September 2025 to August 2028.
Project Description
Recent AI breakthroughs – discovering knot invariant relationships, generating Hirsch conjecture counterexamples, finding new plane colorings – demonstrate the AI4Math paradigm’s potential.
This project pursues three directions: (1) multi-agent systems for pattern recognition and conjecture generation in discrete mathematics, (2) frameworks that design computational experiments with adaptive parameter tuning, and (3) integration with LEAN for automated proof formalization. A human-in-the-loop approach combines AI’s systematic exploration with mathematical intuition.
Talks and posters
Conference and workshop talks
- Jul 2026
- The Agentic Researcher: A Practical Guide to AI-Assisted Research in Mathematics and Machine Learning by Nico Pelleriti
AI4Research Workshop, Seoul
Poster presentations
- Jul 2026
- The Agentic Researcher: A Practical Guide to AI-Assisted Research in Mathematics and Machine Learning by Nico Pelleriti
AI4Math Workshop, Seoul - Jul 2026
- The Agentic Researcher: A Practical Guide to AI-Assisted Research in Mathematics and Machine Learning by Nico Pelleriti
AI4Research Workshop, Seoul
Publications and preprints
Preprints and reports
- Pokutta, S. (2025). Scalable DC Optimization via Adaptive Frank-Wolfe Algorithms.
[arXiv]
[BibTeX]
@misc{2025_Pokutta_DcoptimizationFrankwolfe_2507-17545, archiveprefix = {arXiv}, eprint = {2507.17545}, arxiv = {arXiv:2507.17545}, primaryclass = {math.OC}, author = {Pokutta, Sebastian}, title = {Scalable DC Optimization via Adaptive Frank-Wolfe Algorithms}, date = {2025-07-23} } - Kaibel, V., and Pokutta, S. (2026). A Counterexample to Ziegler’s Cross-Polytope Conjecture for Simplicial 0/1-Polytopes.
[arXiv]
[BibTeX]
@misc{2026_KaibelPokutta_Counterexampleziegler_2606-31640, archiveprefix = {arXiv}, eprint = {2606.31640}, arxiv = {arXiv:2606.31640}, primaryclass = {math.CO}, author = {Kaibel, Volker and Pokutta, Sebastian}, title = {A Counterexample to Ziegler's Cross-Polytope Conjecture for Simplicial 0/1-Polytopes}, date = {2026-06-30} } - Pelleriti, N., Nelaturu, S. H., Zhou, Z., Li, Z., Zimmer, M., Han, B., and Pokutta, S. (2026). What Do Evolutionary Coding Agents Evolve?
[arXiv]
[BibTeX]
@misc{2026_PelleritiEtAl_Evotrace_2605-20086, archiveprefix = {arXiv}, eprint = {2605.20086}, arxiv = {arXiv:2605.20086}, primaryclass = {cs.NE}, author = {Pelleriti, Nico and Nelaturu, Sree Harsha and Zhou, Zhanke and Li, Zongze and Zimmer, Max and Han, Bo and Pokutta, Sebastian}, title = {What Do Evolutionary Coding Agents Evolve?}, date = {2026-05-19} } - Pokutta, S., and Stellato, B. (2026). The exact worst-case asymptotic rate of Nesterov acceleration for strongly convex optimization.
[arXiv]
[BibTeX]
@misc{2026_PokuttaStellato_NesterovWorstCaseRate_2610-15125, archiveprefix = {arXiv}, eprint = {2610.15125}, arxiv = {arXiv:2610.15125}, primaryclass = {math.OC}, author = {Pokutta, Sebastian and Stellato, Bartolomeo}, title = {The exact worst-case asymptotic rate of Nesterov acceleration for strongly convex optimization}, date = {2026-10-10} } - Pokutta, S. (2026). Lower Bounds for Randomized Linear-Oracle Optimization over Strongly Convex Sets.
[arXiv]
[BibTeX]
@misc{2026_Pokutta_RandomizedLinearOracle_2610-13590, archiveprefix = {arXiv}, eprint = {2610.13590}, arxiv = {arXiv:2610.13590}, primaryclass = {math.OC}, author = {Pokutta, Sebastian}, title = {Lower Bounds for Randomized Linear-Oracle Optimization over Strongly Convex Sets}, date = {2026-10-09} } - Pokutta, S. (2026). (Approximate) Symmetric Extension Complexity of the Spanning Tree Polytope.
[arXiv]
[BibTeX]
@misc{2026_Pokutta_Symmetricextensioncomplexity_2606-17017, archiveprefix = {arXiv}, eprint = {2606.17017}, arxiv = {arXiv:2606.17017}, primaryclass = {math.CO}, author = {Pokutta, Sebastian}, title = {(Approximate) Symmetric Extension Complexity of the Spanning Tree Polytope}, date = {2026-06-15} } - Xu, L., Zhou, Y., and Pokutta, S. (2026). Agentic MIP Research: Accelerated Constraint Handler Generation.
[arXiv]
[BibTeX]
@misc{2026_XuZhouPokutta_Agenticmip_2605-09186, archiveprefix = {arXiv}, eprint = {2605.09186}, arxiv = {arXiv:2605.09186}, primaryclass = {cs.AI}, author = {Xu, Liding and Zhou, Yugeng and Pokutta, Sebastian}, title = {Agentic MIP Research: Accelerated Constraint Handler Generation}, date = {2026-05-09} } - Zimmer, M., Pelleriti, N., Roux, C., and Pokutta, S. (2026). The Agentic Researcher: A Practical Guide to AI-Assisted Research in Mathematics and Machine Learning.
[arXiv]
[BibTeX]
@misc{2026_ZimmerPelleritiRouxPokutta_Airesearchframework_2603-15914, archiveprefix = {arXiv}, eprint = {2603.15914}, arxiv = {arXiv:2603.15914}, primaryclass = {cs.LG}, author = {Zimmer, Max and Pelleriti, Nico and Roux, Christophe and Pokutta, Sebastian}, title = {The Agentic Researcher: A Practical Guide to AI-Assisted Research in Mathematics and Machine Learning}, date = {2026-03-16} }
Conference contributions
- Pokutta, S. Frank-Wolfe Beyond 1/t Convergence. Advances in Neural Information Processing Systems, 39.
To appear.
Associated with NeurIPS 2026.
[arXiv]
[BibTeX]
@inproceedings{2026_Pokutta_Localdualsharpness_2604-28006, booktitle = {Advances in Neural Information Processing Systems}, series = {Advances in Neural Information Processing Systems}, volume = {39}, publisher = {Curran Associates}, note = {To appear}, archiveprefix = {arXiv}, eprint = {2604.28006}, arxiv = {arXiv:2604.28006}, primaryclass = {math.OC}, author = {Pokutta, Sebastian}, title = {Frank-Wolfe Beyond 1/t Convergence} } - Pelleriti, N., Nelaturu, S. H., Zhou, Z., Li, Z., Zimmer, M., Han, B., and Pokutta, S. (2026). What Do Evolutionary Coding Agents Evolve? Accepted at NeurIPS 2026 Workshop on Self-Evolving Diversity-Driven Search for Robust AI Systems.
[arXiv]
[BibTeX]
@misc{2026_PelleritiEtAl_Evotrace_2605-20086:1, howpublished = {Accepted at NeurIPS 2026 Workshop on Self-Evolving Diversity-Driven Search for Robust AI Systems}, year = {2026}, archiveprefix = {arXiv}, eprint = {2605.20086}, arxiv = {arXiv:2605.20086}, primaryclass = {cs.NE}, author = {Pelleriti, Nico and Nelaturu, Sree Harsha and Zhou, Zhanke and Li, Zongze and Zimmer, Max and Han, Bo and Pokutta, Sebastian}, title = {What Do Evolutionary Coding Agents Evolve?} }


