Thesis Defenses


  • Julius Baldauf

    Date: Thursday, March 28, 2024 | 2:10pm | Room: 2-449 | Zoom Link

    Committee: Bill Minicozzi (Thesis Advisor and Examination Committee Chair), Tristan Collins, Tristan Ozuch

    The Ricci Flow on Spin Manifolds

    This thesis studies the Ricci flow on manifolds admitting harmonic spinors. It is shown that Perelman's Ricci flow entropy can be expressed in terms of the energy of harmonic spinors in all dimensions, and in four dimensions, in terms of the energy of Seiberg-Witten monopoles. Consequently, Ricci flow is the gradient flow of these energies. The proof relies on a weighted version of the monopole equations, introduced here. Further, a sharp parabolic Hitchin-Thorpe inequality for simply-connected, spin 4-manifolds is proven. From this, it follows that the normalized Ricci flow on any exotic K3 surface must become singular.

  • Deeparaj Bhat

    Date: Tuesday, June 25, 2024 | 3:00pm | Room: 2-136 | Zoom Link

    Committee: Tomasz Mrowka (Advisor and Chair), Daniel Alvarez-Gavela and John Baldwin (Boston College)

    Surgery Exact Triangles in Instanton Theory

    The introduction of instanton Floer theory and Donaldson polynomial invariants in the 1980s revolutionised the study of low dimensional topology. Since then, many Floer theories have been introduced with different structural properties and qualitative features. One of these Floer theories, Heegaard Floer theory, is popular due to its computational ease and rich algebraic structure. One of the computational tools absent in other Floer theories is the integer surgery formula that computes Heegaard Floer homology of 3-manifolds obtained by surgery along knot(s) in them.

    This thesis establishes a new surgery formula in instanton Floer theory. The algebraic language to express this formula is that of the derived category of chain complexes. The first part of the thesis describes this surgery formula whose statement and proof are inspired by the Atiyah-Floer conjectures. The second part then contrasts with the Heegaard Floer analogue by showing that instanton and Heegaard Floer theory cannot agree over integers.

  • Adam Block

    Date: Tuesday, April 30, 2024 | 3:00pm | Room: 4-149 | Zoom Link

    Committee: Alexander Rakhlin (advisor), Yury Polyanskiy, Martin Wainwright, Ankur Moitra (chair)

    Smoothed Online Learning: Theory and Applications

    Many of the algorithms and theoretical results surrounding modern machine learning are predicated on the assumption that data are independent and identically distributed. Motivated by the numerous applications that do not satisfy this assumption, many researchers have been interested in relaxations of this condition, with online learning being the weakest such assumption. In this setting, the learner observes data points one at a time and makes predictions, before incorporating the data into a training set with the goal of predicting new data points as well as possible. Due to the lack of assumptions on the data, this setting is both computationally and statistically challenging. In this thesis, we investigate the statistical rates and efficient algorithms achievable when the data are constrained in a natural way motivated by the smoothed analysis of algorithms. The first part covers the statistical rates achievable by an arbitrary algorithm without regard to efficiency, covering both the fully adversarial setting and the constrained setting in which improved rates are possible. The second part of the thesis focuses on efficient algorithms for this constrained setting, as well as special cases where bounds can be improved under additional structure. Finally, in the third part we investigate applications of these techniques to sequential decicions making, robotics, and differential privacy. We introduce a number of novel techniques, including a Gaussian anti-concentration inequality and a new norm comparison for dependent data.

  • Gonzalo Cao

    Date: Monday, July 1, 2024 | 10:30am | Room: 2-361 | Zoom Link

    Committee: Prof. Gigliola Staffilani (advisor and committee chair), Prof. Semyon Dyatlov and Prof. Larry Guth

    Self-similar singularity formation and wellposedness theory for compressible fluids and dispersive PDE

    This PhD thesis is divided in two parts. The first part consists on the self-similar singularity formation for the compressible Euler equation and its applications. It discusses the existence of smooth and radially symmetric self-similar profiles for all adiabatic constants, and their non-radial linear stability. This is then combined with non-linear stability arguments to show asymptotically self-similar singularity formation for the compressible Navier-Stokes equation and the supercritical defocusing NLS equation, in the euclidean and periodic settings. The second part consists on the improvement of the local wellposedness theory (both deterministic and probabilistic) for the ZK dispersive equation
  • Murilo Corato Zanarella

    Date: Tuesday, April 23, 2024 | 11:00am | Room: 4-370

    Committee: Wei Zhang, Zhiwei Yun and Spencer Leslie (Boston College)

    First explicit reciprocity law for unitary Friedberg—Jacquet periods

    In the early 2000's, Bertolini and Darmon introduced a new technique to bound Selmer groups of elliptic curves via level raising congruences. This was the first example of what is now termed a "bipartite Euler system", and over the last decade we have seen many breakthroughs on constructing such systems for other Galois representations, including settings such as twisted and cubic triple product, symmetric cube, and Rankin—Selberg, with applications to the Bloch—Kato conjecture and to Iwasawa theory.

    This thesis studies the case of Galois representations attached to automorphic representations on a totally definite unitary group U(2r) over a CM field which are distinguished by the subgroup U(r) x U(r). We prove a new "first explicit reciprocity law" in this setting, which has applications to the rank 0 case of the corresponding Bloch—Kato conjecture.

  • Gefei Dang

    Date: Wednesday, April 24, 2024 | 3:00pm | Room: 2-142

    Committee: Wei Zhang, Julee Kim, Zhiwei Yun

    Local newforms and spherical characters for unitary groups

    We first prove a smooth transfer statement analogous to Jacquet–Rallis’s fundamental lemma and use it to compute the special value of a local spherical character that appears in the Ichino–Ikeda conjecture at a test vector. Then we provide a uniform definition of newforms for representations of both even and odd dimensional unitary groups over p-adic fields. This definition is compatible with the one given by Atobe, Oi, and Yasuda in the odd dimensional case. Using the nonvanishing of the local spherical character at the test vector, we prove the existence of the representation containing newforms in every tempered Vogan L-packet. We also show the uniqueness of such representations in Vogan L-packets and give an explicit description of them using local Langlands correspondence.

  • Patrik Gerber

    Date: Friday, April 26, 2024 | 9:30am | Room: 2-361 | Zoom Link

    Committee: Philippe Rigollet (advisor), Yury Polyanskiy, Martin Wainwright

    Likelihood-Free Hypothesis Testing and Applications of the Energy Distance

    The first part of this thesis studies the problem of likelihood-free hypothesis testing: given three samples X,Y and Z with sample sizes n,n and m respectively, one must decide whether the distribution of Z is closer to that of X or that of Y. We fully characterize the problem's sample complexity for multiple distribution classes and with high probability. We uncover connections to two-sample, goodness of fit and robust testing, and show the existence of a trade-off of the form mn ~ k/ε^4, where k is an appropriate notion of complexity and ε is the total variation separation between the distributions of X and Y. We demonstrate that the family of "classifier accuracy" tests are not only popular in practice but also provably near-optimal, recovering and simplifying a multitude of classical and recent results. We generalize our problem to allow Z to come from a mixture of the distributions of X and Y, and propose a kernel-based test for its solution. Finally, we verify the existence of a trade-off between m and n on experimental data from particle physics.

    In the second part we study applications of the energy distance to minimax statistics. We propose a density estimation routine based on minimizing the generalized energy distance, targeting smooth densities and Gaussian mixtures. We interpret our results in terms of half-plane separability over these classes, and derive analogous results for discrete distributions. As a consequence we deduce that any two discrete distributions are well-separated by a half-plane, provided their support is embedded as a packing of a high-dimensional unit ball. We also scrutinize two recent applications of the energy distance in the two-sample testing literature.

  • Shashi Gowda

    Date: Thursday, April 25, 2024 | 10:00am | Room: 32-G882 | Zoom Link

    Committee: Alan Edelman, Steven Johnson, John Urschel

    Symbolic-numeric programming in scientific computing

    Scientific programming languages should pursue two goals: closeness to mathematical notation, and the ability to express efficient numerical algorithms. To meet these goals simultaneously, languages use imperative surface syntaxes that mimick mathematical notation. However, mimicking does not make them the same—mathematics is declarative, pliable, and caters to exploratory human nature; but algorithms are imperative and must cater to machines. Hence, there is a fundamental limit to this approach and we leave the expressive power of the symbolic representation on the table.

    In this thesis, we ask the question: How can symbolic and numerical modes of computing co-exist, one informing the other? As an answer, we develop a user-extensible system that lifts numerical code into symbolic expressions and can turn symbolic expressions back into high-quality numerical code at staged compilation time, essentially providing the scientific user a way to generate programs and to treat programs as the symbolic artifacts they are. We identified siloing of symbolic software into 3 categories (one can call them “symbolic-only”, “secretly symbolic”, “secretly numerical”) which currently each reproduce similar forms of symbolic capabilities, but cannot share code between each other. Our work demonstrates that this siloing is not essential and an ecosystem of symbolic-numeric libraries can thrive in symbiosis.

    Our system is adaptable to any domain: users can define 1) Symbolic variables of any type 2) the set of primitive (symbolically indivisible) functions in the domain, 3) the propa- gation of partial information, and 4) pattern-based rewrites and simplification rules. There is a tendency in scientific computing to create a “compiler for every problem” starting from scratch every time. Every such effort erects its own towers of symbolic and numerical ca- pabilites. A system like ours eliminates this redundancy and lets every scientific user be a “compiler designer” without any prior knowledge of compiler development.

  • Alasdair Hastewell

    Date: Thursday, April 18, 2024 | 10:30am | Room: 2-449 | Zoom Link

    Committee: Jörn Dunkel (chair), John Bush, Alexander Mietke

    Robust spectral representations and model inference for biological dynamics

    Current developments in automated experimental imaging allow for high-resolution tracking across various scales, from whole animal behavior to tissue scale single-cell trajectories during embryogenesis to spatiotemporal gene expression dynamics or neural dynamics. Transforming these high-dimensional data into effective low-dimensional models is an essential theoretical challenge that enables the characterization, comparison, and prediction of the dynamics within and across biological systems. Spectral mode representations have been used successfully across physics, from quantum mechanics to fluid dynamics, to compress and model dynamical data. However, their use in analyzing biological systems has yet to become prevalent. Here, we present a set of noise-robust, geometry-aware mathematical tools that enable spectral representations to extract quantitative measurements directly from experimental data. We demonstrate the practical utility of these methods by applying them to the extraction defect statistics in signaling fields on membranes of starfish, the inference of partial differential equations directly from videos of active particle dynamics, and the categorization of emergent patterns in spatiotemporal gene expression during bacterial swarming.

    An additional challenge occurs for complex biophysical processes where the underlying dynamics are yet to be entirely determined. Therefore, we would like to use the experimental data to infer effective dynamical models directly that can elucidate the system's underlying biological and physical mechanisms. Building on spectral mode representations, we construct a generic computational framework that can incorporate prior knowledge about biological and physical constraints for inferring the dynamics of living systems through the evolution of their mode representations. We apply this framework first to single-cell imaging data during zebrafish embryogenesis, demonstrating how our framework compactly characterizes developmental symmetry breaking and reveals similarities between pan-embryo cell migration and Brownian particles on curved surfaces. Next, we apply the framework to the undulatory locomotion of worms, centipedes, robots, and snakes to distinguish between locomotion behaviors. Finally, we present an extension of the framework to the case of nonlinear dynamics when all relevant degrees of freedom are only partially observed, with applications to neuronal and chemical dynamics.

  • Arun Kannan

    Date: Tuesday, April 23, 2024 | 1:00pm | Room: 1-273 | Zoom Link

    Committee: Pavel Etingof (advisor), Roman Bezrukavnikov, Victor Kac

    On Lie Theory in the Verlinde Category

    A symmetric tensor category (STC) can be thought of as a “home” to do commutative algebra, algebraic geometry, and Lie theory. They are defined by axiomatizing the basic properties of a representation category of a group (or affine supergroup scheme). Are these the only examples of STCs? In characteristic zero, a famous theorem of Deligne states that, assuming a natural growth condition, representation categories of affine supergroup schemes are the only examples. However, the situation is totally different in positive characteristic, and the Verlinde category Verp is the most fundamental counterexample and appears to play a key role in generalizing the theorem of Deligne to positive characteristic. Moreover, Verp contains the category of supervector spaces. The upshot is that the study of Verp provides new algebraic structures and phenomena beyond that afforded by superalgebra and supergeometry but must also generalize what is already known.

    In this thesis defense, we will first survey the theory of symmetric tensor categories. Then, we will discuss new algebraic structures that arise from the Verlinde category, including new constructions of exceptional Lie superalgebras and a generalization of Jordan algebras unique to characteristic 5. Finally, we will turn to progress made on generalizing useful algebraic techniques and machinery from the super setting to the Verp setting, like the Steinberg tensor product theorem and notions of polynomial functors.

  • Daniil Kliuev

    Date: Tuesday, April 16, 2024 | 2:30pm | Room: 2-131

    Committee: Pavel Etingof, Roman Bezrukavnikov and Ivan Loseu (Yale)

    Positive traces and analytic Langlands correspondence

    I will describe my results with co-authors in two directions.

    The first direction is the problem of classification of positive traces on quantized Coulomb branches. In the most general setting, this problem generalizes the classical problem of describing irreducible unitary representations of real reductive Lie groups. We consider the case of Kleinian singularities of type $A$ and provide a complete classification of positive traces.

    The second direction is analytic Langlands correspondence, which is the following. Let $X$ be a smooth irreducible projective curve over $\mathbb{C}$, $G$ be a complex reductive group. On one side of this conjectural correspondence there are $G^{\vee}$-opers on $X$ satisfying a certain topological condition ({\it real} opers), where $G^{\vee}$ is Langlands dual group. On the other side there is joint spectrum of certain operators on $L^2(Bun_G)$, called Hecke operators, where $Bun_G$ is the variety of stable parabolic $G$-bundles on $X$ and $L^2(Bun_G)$ is a Hilbert space of square-integrable half-densities. We prove most of the main conjectures of analytic Langlands correspondence in the case when $G=\operatorname{PGL}_2(\mathbb{C})$ and $X$ either a genus one curve with points or $X$ is $\mathbb{P}^1$ with higher structures at points.

  • Vasily Krylov

    Date: Monday, April 29, 2024 | 9:30am | Room: 2-143

    Committee: Roman Bezrukaunikov (advisor), Zhiwei Yun, and Ivan Loseu (Yale)

    Geometry and representation theory of symplectic singularities in the context of symplectic duality

    This thesis studies the geometry and representation theory of various symplectic resolutions of singularities from different perspectives. Specifically, we establish a general approach to attack the Hikita-Nakajima conjecture and illustrate this approach in the example of ADHM spaces. We also study minimally supported representations of the quantizations of ADHM spaces and provide explicit formulas for their characters. Lastly, we describe the monodromy of eigenvalues of quantum multiplication operators for type A Nakajima quiver varieties by examining Bethe subalgebras in Yangians and linking their spectrum with Kirillov-Reshetikhin crystals.

  • Jae Hee Lee

    Date: Monday, April 1, 2024 | 3:00pm | Room: 2-361 | Zoom Link

    Committee: Prof. Paul Seidel (thesis advisor), Prof. Pavel Etingof, Prof. Denis Auroux (External, Harvard)

    Equivariant quantum connections in positive characteristic

    In this thesis, we apply techniques from symplectic Gromov--Witten theory to study the equivariant quantum connections in positive characteristic. The main examples of interest arise from symplectic resolutions. We introduce equivariant generalizations of the quantum Steenrod operations of Fukaya, provide nontrivial computations in the example of the cotangent bundle of the projective line, and explore the relationship with Varchenko's construction of mod $p$ solutions to the quantum differential equation. We then prove the compatibility of the equivariant quantum Steenrod operations with the quantum differential and difference connections. As a consequence, we obtain an identification of our operations for divisor classes with the $p$-curvature of the quantum connection in a wide range of examples.
  • Ishan Levy

    Date: Tuesday, April 23, 2024 | 1:30pm | Room: 13-1143

    Committee: Davesh Maulik, Michael Hopkins, Haynes Miller, and Jeremy Hahn

    The algebraic K-theory of the chromatic filtration and the telescope conjecture

    Chromatic homotopy theory gives a conceptual framework with which to understand the stable homotopy theory, by decomposing the stable homotopy category into monochromatic pieces. There are two variants of these monochromatic pieces, the K(n) and T(n)-local categories, the former of which is often quite understandable in terms of formal groups of height n, and the latter of which detects the so-called v_n-periodic part of the stable homotopy groups of spheres. I will explain how algebraic K-theory has refined our understanding of these monochromatic pieces. On one hand, algebraic K-theory is an important structural invariant of these categories that 'stably' classifies objects and their automorphisms, and I will explain some tools we have to computationally access the K-theory of these categories. On the other hand, the algebraic K-theory of such categories are interesting as spectra: they detect a lot of information about the stable homotopy groups of spheres and have helped us understand the difference between the T(n) and K(n)-local categories.

  • Nicholas Liu

    Date: Tuesday, July 2, 2024 | 11:00am | Room: 2-142

    Committee: John Bush, Matthew Durey, and Rodolfo Ruben Rosales

    Orbital stability in a classical pilot-wave system

    The hydrodynamic bouncing droplet system, consisting of millimetric droplets bouncing on a vibrating fluid bath, displays many quantum mechanical phenomena on a macroscopic scale. These phenomena include tunnelling, diffraction and wave-like statistics. This thesis focuses on the features responsible for the quantisation of orbital radii, and rationalises this quantisation in terms of the stability of circular orbits arising in the presence of a rotating frame and a central force, and are most pronounced when the waves generated by each bounce decay slowly. The faster the waves decay, the less past impacts affect the droplet’s future dynamics, corresponding to lower path memory. We conduct an analytical investigation into the stability of circular orbits using a generalized theoretical framework that allows for an exploration of classical pilot-wave dynamics both inside and outside the experimentally accessible parameter regime. The exploration of parameter regimes beyond those accessible with the hydrodynamic system reveals much richer orbital dynamics. Our novel mathematical approach allows for evaluation of the integrals appearing in the stability problem in terms of Bessel functions of complex order, and thus facilitates asymptotic expansions of the stability problem in various limits. Within the experimental parameter regime, we demonstrate that in a rotating frame, circular orbits destabilise only via resonant instabilities, for which the growing perturbations oscillate at a frequency that is an integer multiple of the orbital frequency. Conversely, in a central force, non-resonant instabilities arise, for reasons discussed herein. Outside the experimental parameter regime, we show how the non-resonant instability leads to counter-intuitive scenarios; for example, where circular orbits are stabilised by increasing memory. In the limit of vanishing particle inertia, infinite path memory and a linear spring force, we demonstrate the intriguing possibility of infinitely many sharply quantised orbital states, where the allowed orbital radii exist in vanishingly thin intervals, and are stabilised by the combined influence of the time-averaged wave field and spring force. We demonstrate that these sharply quantised orbital states are only stable for higher memory. We then consider the effect of weak external forces on spin states, circular orbits arising in the absence of external forces, and show that the destabilisation of spin states depends in a complex manner on the type of external force applied. Finally, we show that the instability of large circular orbits is related to the in-line speed oscillations of free walking droplets in a manner that is independent of the external force.

  • Calder Morton-Ferguson

    Date: Friday, April 26, 2024 | 1:30pm | Room: 2-449 | Zoom Link

    Committee: Roman Bezrukavnikov (advisor), Zhiwei Yun, Ivan Loseu

    Kazhdan-Laumon categories and representations

    In 1988, D. Kazhdan and G. Laumon constructed the \emph{Kazhdan-Laumon category}, an abelian category $\mathcal{A}$ associated to a reductive group $G$ over a finite field, with the aim of using it to construct discrete series representations of the finite Chevalley group $G(\mathbb{F}_q)$. The well-definedness of their construction depended on their conjecture that this category has finite cohomological dimension. This was disproven by R. Bezrukavnikov and A. Polishchuk in 2001, who found a counterexample for $G = SL_3$.

    Since the early 2000s, there has been little activity in the study of Kazhdan-Laumon categories, despite them being beautiful objects with many interesting properties related to the representation theory of $G$ and the geometry of the basic affine space $G/U$. In the first part of this thesis, we conduct an in-depth study of Kazhdan-Laumon categories from a modern perspective. We first define and study an analogue of the Bernstein-Gelfand-Gelfand Category $\mathcal{O}$ for Kazhdan-Laumon categories and study its combinatorics, establishing connections to Braverman-Kazhdan's Schwartz space on the basic affine space and the semi-infinite flag variety. We then study the braid group action on $D^b(G/U)$ (the main ingredient in Kazhdan and Laumon's construction) and show that it categorifies the \emph{algebra of braids and ties}, an algebra previously studied in knot theory; we then use this to provide conceptual and geometric proofs of new results concerning this algebra.

    After Bezrukavnikov and Polishchuk's counterexample to Kazhdan and Laumon's original conjecture, Polishchuk made an alternative conjecture: though this counterexample shows that the Grothendieck group $K_0(\mathcal{A})$ is not spanned by objects of finite projective dimension, he noted that a graded version of $K_0(\mathcal{A})$ can be thought of as a module over Laurent polynomials and conjectured that a certain localization of this module is generated by objects of finite projective dimension. He suggested that this conjecture could lead toward a proof that Kazhdan and Laumon's construction is well-defined, and he proved this conjecture in Types $A_1, A_2, A_3$, and $B_2$. In the final chapter of this thesis, we prove Polishchuk's conjecture for all types, and prove that Kazhdan and Laumon's construction is indeed well-defined, giving a new geometric construction of discrete series representations of $G(\mathbb{F}_q)$.

  • Matthew Nicoletti

    Date: Monday, April 29, 2024 | 2:30pm | Room: 2-361 | Zoom Link

    Committee: Alexei Borodin (Advisor, chair), Scott Sheffield, Lauren Williams (Harvard)

    Title: Stochastic Dynamics on Integrable Lattice Models

    The purpose of this thesis is to present some new results related to the six-vertex and dimer model. One theme is the construction and analysis of Markov processes which are naturally associated to these lattice models. Certain integrability properties of the six-vertex and dimer model, often related to the Yang--Baxter equation, allow for the construction of associated Markov chains. In some cases, these are measure preserving Markov chains on configurations of the lattice model. In other cases, they arise via transfer matrices, after choosing a distinguished time coordinate, as a continuous time degeneration of the "time evolution" of the lattice model itself. It is often the case that the integrability of the underlying lattice model provides powerful tools to study the associated Markov chains or their marginals, which are sometimes Markov chains themselves.

    In particular together with coauthors, we construct and analyze Markov chains on six-vertex configurations and on dimer model configurations, both of which are models for surface growth in the (2+1)-dimensional "Anisotropic KPZ" (or "AKPZ") universality class; we construct a Markov chain generalizing "domino shuffling" which samples exactly from a recently introduced probability measure on tuples of interacting dimer configurations; using a version of the usual domino shuffling algorithm, we construct and analyze deterministic "t-embeddings" of certain dimer graphs, which discretize minimal surfaces carrying the conformal structure of the limiting Gaussian free field; we construct stationary measures for several colored interacting particle systems using the Yang—Baxter equation.

  • Alexander Ortiz

    Date: Wednesday, April 24, 2024 | 1:15pm | Room: 2-449 | Zoom Link

    Committee: Larry Guth (advisor), David Jerison, Gigliola Staffilani

    Sparse Fourier restriction for the cone

    If the Fourier transform of F(x) is supported near a segment of the light-cone in R^3, what is the shape of the level sets U(N) = {x in R^3 : |F(x)| > N} for large values of N? In 2000, Thomas Wolff had a creative idea to study a related question based on the method of point-circle duality, and used it in a pivotal way to prove the first sharp L^p-decoupling estimates for the cone in R^3 for large values of p.

    I will discuss new weighted L^2 estimates of F(x) which give us insight into the shape of level sets. I will explain how we use some of the same key ideas introduced by Wolff, together with a few new ones in the same spirit. By Wolff's method, our main theorem will partly be an application of a recent circular maximal function estimate due to Pramanik—Yang—Zahl in 2022 from their study of Kaufman-type restricted projection problems.

  • Ashwin Sah

    Date: Wednesday, April 3, 2024 | 3:30pm | Room: 2-449

    Committee: Prof. Yufei Zhao (advisor and chair), Prof. Dor Minzer, and Prof. Philippe Rigollet

    Random and exact structures in combinatorics

    We aim to show various developments related to notions of randomness and structure in combinatorics and probability. One central notion, that of the pseudorandomness-structure dichotomy, has played a key role in additive combinatorics and extremal graph theory. In a broader view, randomness (and the pseudorandomness notions which resemble it along various axes) can be viewed as a type of structure in and of itself which has certain typical and global properties that may be exploited to exhibit or constrain combinatorial and probabilistic behavior.

    These broader ideas often come in concert to allow the construction or extraction of exact behavior. We look at three particular directions: the singularity of discrete random matrices, thresholds for Steiner triple systems, and improved bounds for Szemerédi's theorem. These concern central questions of the areas of random matrices, combinatorial designs, and additive combinatorics.

  • Mehtaab Sawhney

    Date: Wednesday, April 17, 2024 | 2:00pm | Room: 2-449

    Committee: Yufei Zhao, Dor Minzer, and Philippe Rigollet

    Probabilistic and Analytic Methods in Combinatorics

    The defense will center on fast algorithms for discrepancy theory. Discrepancy theory is broadly concerned with the following problem; given a set of objects, we aim to partition them into pieces which are “roughly equal”. We will focus specifically on vector balancing: given a set of vectors, one seeks to divide them into two parts with approximately equal sum.

    Important results in this area, including Spencer’s six standard deviations suffice and Banaszczyk's results towards the Komlós conjecture, were originally purely existential. However, since work of Bansal from 2010, it has become clear that such existential results can often be made algorithmic. We will explain a pair of such results. The first concerns bounds for online vector balancing obtained via a certain Gaussian fixed point random walk. The second gives an algorithmic form of Spencer's theorem that runs in near input sparsity time.

  • George Stepaniants

    Date: Thursday, April 25, 2024 | 2:30pm | Room: 4-149 | Zoom Link

    Committee: Philippe Rigollet, Jörn Dunkel, Sasha Rakhlin

    Inference from Limited Observations in Statistical, Dynamical, and Functional Problems

    Observational data in physics and the life sciences comes in many varieties. Broadly, we can divide datasets into cross-sectional data which record a set of observations at a given time, dynamical data which follow how observations change in time, and functional data which observe data points over a space (and possibly time) domain. In each setting, prior knowledge of statistical, dynamical systems, and physical theory allow us to constrain the inferences and predictions we make from observational data. This domain knowledge becomes of paramount importance when the data we observe is limited: due to missing labels, small sample sizes, unobserved variables, and noise corruption.

    This thesis explores several problems in physics and the life sciences, where the interplay of domain knowledge with statistical theory and machine learning allows us to make inferences from such limited data. We begin in Part I by studying the problem of feature matching or dataset alignment which arises frequently when combining untargeted (unlabeled) biological datasets with low sample sizes. Leveraging the fast numerical methods of optimal transport, we develop an algorithm that gives a state-of-the-art solution to this alignment problem with optimal statistical guarantees. In Part II we study the problem of interpolating the dynamics of points clouds (e.g., cells, particles) given only a few sparse snapshot recordings. We show how tools from spline interpolation coupled with optimal transport give efficient algorithms returning smooth dynamically plausible interpolations. Part III of our thesis studies how dynamical equations of motion can be learned from time series recordings of dynamical systems when only partial observations of these systems are captured in time. Here we develop fast routines for gradient optimization and novel tools for model comparison to learn such physically interpretable models from incomplete time series data. Finally, in Part IV we address the problem of surrogate modeling, translating expensive solvers of partial differential equations for physics simulations into fast and easily-trainable machine learning algorithms. For linear PDEs, our prior knowledge of PDE theory and the statistical theory of kernel methods allows us to learn the Green's functions of various linear PDEs, offering more efficient ways to simulate physical systems.

  • Pu Yu

    Date: Wednesday, April 3, 2024 | 2:00pm | Room: 2-255

    Committee: Scott Sheffield (advisor), Alexei Borodin, Nike Sun

    Conformal welding of random surfaces from Liouville theory

    Liouville quantum gravity (LQG) is a natural model describing random surfaces, which arises as the scaling limit for random planar maps. Liouville conformal field theory (LCFT) is the underlying 2D CFT that governs LQG. Schramm-Loewner evolution (SLE) is a random planar curve, which describes the scaling limits of interfaces in many statistical physics models. As discovered by Sheffield (2010), one of the deepest results in random geometry is that SLE curves arises as the interfaces under conformal welding of LQG surfaces.

    In this thesis, we present some new results on conformal welding of LQG surfaces as well as their applications towards the theory of SLE. We first define a three-parameter family of random surfaces in LQG which can be viewed as the quantum version of triangles. Then we prove the conformal welding result of a quantum triangle and a two-pointed quantum disk, and deduce integrability results for chordal SLE with three force points.

    The second main result is regarding the conformal welding of a multiple number of LQG surfaces, where under several scenarios, we prove that the output surfaces can be described in terms of LCFT, and the random moduli of the surface is encoded in terms of the partition functions for the SLE curves.

    The third part is about the conformal welding of the quantum disks with forested boundary, where we prove that this conformal welding gives a two-pointed quantum disk with an independent SLE$_\kappa$ for $\kappa\in(4,8)$. We further extend to the conformal welding of a multiple number of forested quantum disks, where as an application, for $\kappa\in(4,8)$, we prove the existence of the multiple SLE partition functions, which are smooth functions satisfying a system of PDEs and conformal covariance. This was open for $\kappa \in (6,8)$ and $N\ge 3$ prior to our work.

    The conformal loop ensemble (CLE) is a random collection of planar loops which locally look like SLE. For $\kappa \in (4,8)$, the loops are non-simple and may touch each other and the boundary. As a second application, we derive the probability that the loop surrounding a given point in the non-simple conformal loop ensemble touches the domain boundary.

  • Danielle Wang

    Date: Tuesday, April 23, 2024 | 1:00pm | Room: 4-265

    Committee: Wei Zhang (advisor/chair), Julee Kim, Spencer Leslie (Boston College)

    Twisted Gan-Gross-Prasad conjecture for unramified quadratic extensions

    The global twisted GGP conjecture is a variant of the Gan-Gross-Prasad conjecture for unitary groups, which considers the restriction of an automorphic representation of GL(V) to its subgroup U(V), for a skew-Hermitian space V. It relates the nonvanishing of a certain period integral to the central value of an L-function attached to the representation.

    In this thesis, using a relative trace formula approach, we prove the global twisted GGP conjecture in the unramified case, under some additional local assumptions on the quadratic extension and the automorphic representation. In particular, we reduce the required fundamental lemma to the Jacquet-Rallis fundamental lemma.

  • Catherine Wolfram

    Date: Tuesday, April 30, 2024 | 3:30pm | Room: 2-255 | Zoom Link

    Committee: Scott Sheffield (thesis advisor), Alexei Borodin, Curtis McMullen

    Random geometry in two and three dimensions

    A central theme in random geometry is the interplay between discrete models and continuum ones that appear in scaling limits. Surprising structure and symmetry often arises in these scaling limits, leading to an interplay between combinatorics, probability, complex analysis, and geometry.

    The dimer model is one of the classical lattice models of statistical mechanics and can be defined in any dimension. In the first half of this thesis, we prove a large deviation principle for dimer tilings in three dimensions. This generalizes a two-dimensional result of Cohn, Kenyon, and Propp, and is one of the first results for dimers in any dimension $d>2$. Many ideas and constructions used to study dimers are specific to two dimensions, so our arguments start from a smaller set of tools including Hall's matching theorem, the qualitative description of the Gibbs property, and a double dimer swapping operation.

    In the second half of this thesis, we study discrete, geometrically-motivated coordinates called shears on the space of circle homeomorphisms up to M\"obius transformations. The Weil-Petersson Teichm\"uller space is a subspace of this which has been of long-term interest in geometry and string theory and has recent connections to SLE curves in probability. We introduce and study natural $\ell^2$ spaces in terms of shears, and obtain sharp results comparing them to H\"older classes of circle homeomorphisms and the Weil-Petersson class. We also give some preliminary results about i.i.d. Gaussian random shears.