Date/Time : 12 Aug 2026 15:00
Venue : S17 #04-06 (Seminar Room 1)
Speaker : Boaz Klartag, Tel Aviv University and Weizmann Institute of Science

We prove there exists a lattice sphere packing in R^n whose density is at least c n^2 / 2^n, where c > 0 is a universal constant. Previously known constructions of sphere packings in R^n had densities of the order of magnitude of n / 2^n, up to logarithmic factors. Our proof utilizes a stochastically evolving ellipsoid that accumulates at least c n^2 lattice points on its boundary, while containing no lattice points in its interior except for the origin.

Add to calendar :

Date/Time : 29 Jul 2026 15:00
Venue : S17 #04-06 (Seminar Room 1)
Speaker : Zhongkai Tao, Institut des Hautes Études Scientifiques

The fractal uncertainty principle, introduced by Dyatlov and Zahl, has become an important tool in the study of spectral gaps for systems with trapping. It quantifies how a function and its Fourier transform can both concentrate near fractal sets. I will talk about some recent progress in the fractal uncertainty principle and its applications to spectral problems. This talk is partly based on joint works with Aidan Backus, James Leng, Xiaoqi Huang, Christopher Sogge, Zhexing Zhang, Alex Cohen and Zhenhao Li.

Add to calendar :

Date/Time : 29 Apr 2026 15:00
Venue : S17 #04-06 (Seminar Room 1)
Speaker : Juncheng Wei, The Chinese University of Hong Kong

In 2023, one year before his death, H. Brezis posted 28 open questions which he described as “challenging and difficult”. In this talk I will give complete answer to the first open question—Problem 1.1 on Brezis-Nirenberg problem on three dimensional unit ball, and the seventh open question—Problem 3.1 on harmonic map equation on disk.

Add to calendar :

Date/Time : 25 Mar 2026 15:00
Venue : S17 #04-06 (Seminar Room 1)
Speaker : Jonathan Eskstein, Rutgers University

Parallel computing is a common tool in science and engineering. It is now common for large universities and government labs to own distributed HPC computing system with tens of thousands of processor cores, simulating phenomena such as protein folding, fluid flow, and structure deformation. However, other kinds of numerical computing have lagged behind in their use of large-scale parallel computing. In numerical optimization, the potential of parallel computing has not been fully exploited, except within specialized applications for neural network training. The most popular and robust optimization solver codes do not scale well beyond dozens of cores for some applications, and exhibit poor scaling after just a handful of cores in other cases. There are several reasons why, including the difficulty of parallelizing the central numerical operations in standard linear and nonlinear optimization algorithms, commitment to deterministic execution patterns, and programming difficulties. Languages and compilers for developing parallel algorithms have displayed little meaningful evolution in the last few decades even as hardware capabilities have soared; in fact, they have arguably devolved since the 1990’s.

However, the potential for using parallel computing in optimization remains. This talk will examine how numerical optimization solvers could more effectively take advantage of parallel computing, focusing on first-order methods for continuous optimization and search methods for discrete and nonconvex optimization. For truly challenging applications, it may be possible to combine such approaches. The talk will also discuss the potential obstacles and how they might be overcome. Time permitting, it will also overview some ongoing work on creating a portable parallel-capable version of the PDLP first-order linear programming solver.

Add to calendar :

Date/Time : 11 Mar 2026 15:00
Venue : S17 #04-06 (Seminar Room 1)
Speaker : Sun Song, Zhejiang University

Einstein metrics serve as the counterpart in Riemannian geometry to Einstein spacetimes in general relativity. As one of the most important geometric structures, the study of Einstein metrics is deeply connected to various branches of mathematics, including complex algebraic geometry, low-dimensional topology, partial differential equations, and mathematical physics. In this talk we will discuss recent advances in understanding the construction and classification of Einstein metrics.

Add to calendar :

Date/Time : 06 Mar 2026 14:30
Venue : IMS Executive Seminar Room (S17 Level 3)
Speaker : Max Welling, CuspAI & University of Amsterdam

Prof. Dr. Max Welling is a full professor and research chair in machine learning at the University of Amsterdam and a Merkin distinguished visiting professor at Caltech. He is co-founder and CTO of the startup CuspAI – a frontier AI company harnessing AI to unlock materials breakthroughs. He is also a member of the Royal Dutch Academy of Sciences, a fellow at the Canadian Institute for Advanced Research (CIFAR) and the European Lab for Learning and Intelligent Systems (ELLIS) where he served on the founding board. His previous appointments include Partner and VP at Microsoft Research, VP at Qualcomm Technologies, professor at UC Irvine.

Prof. Max Welling finished his PhD in theoretical high energy physics under supervision of Nobel laureate prof. Gerard ‘t Hooft. He then switched fields to focus on machine learning, first as a postdoc at Caltech under supervision of prof. Pietro Perona and then as postdoc under supervision of Nobel laureate prof. Geoffrey Hinton at UCL & U. Toronto.

Prof. Max Welling is a preeminent computer scientist and physicist with over 200,000 Google Scholar citations. His landmark contributions include inventing the Variational Autoencoder (VAE), a foundational architecture in generative AI, and Graph Convolutional Networks (GCNs), which revolutionized the processing of structured data such as molecules and social networks. Furthermore, he pioneered Stochastic Gradient Langevin Dynamics (SGLD) for Bayesian learning and has led the advancement of Equivariant Neural Networks, which integrate fundamental geometric symmetries into deep learning models.

Add to calendar :

Date/Time : 04 Mar 2026 15:00
Venue : S17 #04-06 (Seminar Room 1)
Speaker : Sven Leyffer, Argonne National Laboratory

Many design, planning and decision problems arising in engineering, sciences, finance, and statistics can be modeled as mixed-integer nonlinear optimization problems. A challenging class of mixed-integer problems are topology design problem, such as additive manufacturing and the design of cloaking devices. In topology optimization, the physical response of the design is modeled as partial-differential equations (PDEs) and the design is modeled with binary variables defined on each element of the discretization of the PDE. This approach results in mixed integer PDE-constrained optimization (MIPDECO) problem that combine the computational challenges of PDEs with the combinatorial challenges of a massive number of discrete variables. We show that despite their seemingly hopeless complexity, certain MIPDECOs can be solved efficiently (at a cost comparable to a few continuous PDE-constrained optimization solves). We discuss two classes of methods: rounding techniques that are asymptotically optimal, and randomized search techniques. We illustrate these solution techniques with examples from topology optimization.

Add to calendar :

Date/Time : 25 Feb 2026 15:00
Venue : S17 #04-06 (Seminar Room 1)
Speaker : Ralf Spatzier, University of Michigan

Symmetry and extremal properties have played a major role in mathematics for a long time. While I will start to discuss this in the context of some Riemannian geometry, I will emphasize more recent work on dynamical systems. Here symmetry expresses itself in terms of having non-trivially commuting maps or flows, or an action of some group with complicated relations. Case in point are actions of semisimple Lie groups, especially ones of higher rank, e.g. SL(n,R) with n at least 3. This is the so-called Zimmer program. I will discuss some recent highlights.

Add to calendar :

Date/Time : 04 Feb 2026 15:00
Venue : IMS Executive Seminar Room (S17 Level 3)
Speaker : Zhang Pingwen, Wuhan University and Peking University

Real numbers underpin mathematical research and the understanding of natural phenomena. The real number set comprises rational numbers (a measure-zero set) and irrational numbers (a full-measure set), implying that systems governed by irrational numbers are more prevalent. However, current computers cannot exactly store or represent irrational numbers, rendering numerical computations for irrational-dominated systems susceptible to Diophantine errors—incurred when approximating irrationals with rationals. Such errors significantly compromise numerical results. In this talk, we analyze the mechanism of Diophantine error and its impact on numerical computations. We also develop an approximation theory for multi-dimensional quasiperiodic functions with Diophantine frequencies when approximated by periodic functions. To avoid Diophantine errors, we propose two novel algorithms: the projection method and the finite points recovery method, which extend high-precision numerical computation from the rational to the real domain. Leveraging the ergodic and arithmetic properties of irrational numbers, we establish the mathematical foundations of these algorithms. Furthermore, we present their applications to several irrational-dominated scientific problems to discover new physical phenomena, including quasicrystals and their phase transitions, interfacial structures in materials, and quasiperiodic Schrödinger operators.

Add to calendar :