Statistics Theory · Cross-listed from stat.ME
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
Friday 20 February
Multivariate mode estimation arises in many statistical problems such as inverse problems, multimodal sampling, and density-based clustering, but becomes challenging in moderate to high dimensions, especially when the underlying density is not directly evaluable. We introduce GERVE (Gibbs-measure Entropy-Regularized Variational Estimation), a sample-based method for estimating multivariate modes by approximating Gibbs distributions directly from samples, without estimating or evaluating the density. GERVE uses Gaussian-mixture variational annealing and natural-gradient optimization, producing a mixture concentrated in high-density regions whose component responsibilities also provide a clustering of the observations. We prove theoretical guarantees in two regimes: as the Gibbs temperature goes to zero, the optimal variational mixture concentrates around the global modes of the population density; at fixed positive temperature, we prove existence, consistency, and asymptotic normality of empirical maximizers and propose a bootstrap procedure for uncertainty quantification. Simulations and a real-data experiment show that GERVE accurately recovers modes and produces meaningful clusters.
Numerical Analysis
Mathematical and numerical study on the ground states of rotating spin-orbit coupled spin-1 Bose-Einstein condensates
In this article, we study mathematically and numerically the ground states of three-component rotating spin-orbit coupled (SOC) spin-1 Bose-Einstein condensates modeled by the coupled Gross-Pitaevskii equations (CGPEs). Firstly, we rigorously prove existence result of the ground state and derive some analytical properties, including the virial identity and negativity of SOC energy. Secondly, we propose an efficient and accurate preconditioned nonlinear conjugate gradient (PCG) algorithm to compute the ground states. We truncate the whole space into a bounded rectangular domain and readily apply the Fourier spectral method to approximate the wave function. The PCG method is successfully adapted with appropriate modifications to the adaptive step size control strategy for the one-parameter energy minimization problem and to the choice of preconditioners, achieving great performance in terms of accuracy and efficiency. Lastly, we carry out extensive numerical experiments to verify the existence and property results of the ground states, confirm the spatial spectral accuracy by traversing the most commonly-used initial guesses for each component thanks to its great efficiency, which is also attributed to a utilization of cascadic multigrid and discrete Fast Fourier Transform (FFT). Moreover, we investigate the effects of local interaction, rotation and spin-orbit coupling and external trapping potential on the ground state, and unveil some interesting physical phenomena, such as giant vortex and U-shape vortex line.
Read the full abstract
Combinatorics
The Turán number of Berge paths
A Berge path of length $k$ in an $r$-uniform hypergraph is a collection of $k$ hyperedges $h_1,\dots,h_k$ and $k+1$ vertices $v_1,\dots,v_{k+1}$ such that $v_i, v_{i+1}\in h_i$ for each $1\le i\le k$. Győri, Katona and Lemons [European J. Combin. 58 (2016) 238–246] generalized the Erdős-Gallai theorem to Berge paths and established bounds for the Turán number of Berge paths. However, these bounds are sharp only when some divisibility conditions hold. Gy\H ori, Lemons, Salia and Zamora [J. Combin. Theory Ser. B 148 (2021) 239–250] determined the exact value of the Turán number of Berge paths in the case $k\le r$. In this paper, we settle the final open case $k>r$, thereby completing the determination of the Turán number of Berge paths.
Read the full abstract
Analysis of PDEs
Liouville theorems for mixed local and nonlocal indefinite equations
We investigate the qualitative properties of positive solutions to mixed local-nonlocal equations with indefinite nonlinearities, emphasizing the interaction between classical and fractional Laplacians. We first establish maximum principles and prove strict monotonicity along the $x_1$-direction for mixed elliptic operators. By combining a mollified first eigenfunction with a suitable sub-solution, we derive nonexistence results for the mixed operator $ (-\Delta)^s - \Delta$ via a contradiction argument. These results are further extended to the parabolic setting, incorporating both the Marchaud-type fractional time derivative and the classical first-order derivative, revealing new qualitative features under dual nonlocality. A key aspect of our approach is a careful adaptation of the method of moving planes to the mixed local-nonlocal context. By addressing the distinct scaling behaviors of local and nonlocal terms, the method yields monotonicity and Liouville-type results without standard decay assumptions, and provides a framework potentially applicable to a broader class of mixed elliptic and parabolic problems.
Read the full abstract
Combinatorics
Hypersphere-Based Restricting Conditions for Colorings of the Euclidean Space
We study colorings of the Euclidean space constrained by hypersphere forcing conditions: if an admissible hypersphere, $S_r(p)$, centered at a point $p$ and of radius $r$ contains a monochromatic set of points satisfying a certain property $\mathcal{P}$, then the center of the hypersphere must have that color. These forcing conditions may be restricted in applicability to a specific set of hyperspheres $S_r(p)$. For cardinality-based forcing conditions we prove a general theorem: for countably many colors and any uncountable set of admissible radii $\mathcal{R}$, such a coloring is locally monochromatic on any admissible center set $\Omega \subseteq \mathbb{R}^n$ (hence constant, for connected $\Omega$). For rigid geometric properties (simplex shape, edge-length, volume constraints) we show that forcing conditions alone are insufficient without regularity assumptions. Our main result shows that for colorings satisfying a certain Baire regularity condition rigid geometric properties enforce local monochromaticity and, in the presence of a certain “uniform cap" condition, global monochromaticity. Applications include dichotomies for edge-length and volume constraints in terms of $\inf(\mathcal{L})$ and $\inf(\mathcal{V})$, and a comeagerness criterion in the “all edges in $\mathcal{L}$” regime.
Read the full abstract
Numerical Analysis
Optimal error estimate of an isoparametric upwind discontinuous Galerkin method for radiation transport equation on curved domains
In recent years, high-order finite element methods on high-order meshes have attracted considerable attention. This work investigates the isoparametric upwind discontinuous Galerkin method for the radiation transport equation on a bounded domain with a piecewise $C^{k+1}$ smooth curved boundary. We use the isoparametric mapping to approximate the curved domain and construct a curved upwind discontinuous Galerkin scheme. The first-order hyperbolic nature and the complexity introduced by non-affine transformation, lead to additional difficulties for geometric approximation, numerical stability and the optimal error estimate. To address these issues, with the help of an isoparametric auxiliary operator, we first prove that the bilinear form is continuous with respect to the DG norm when its first argument is the isoparametric projection error. Then the geometric approximation error of inflow boundary of original domain is precisely estimated. The error order between discrete normal vectors and the continuous ones are also proven. Finally, the rigorous analysis yields an optimal convergence rate in the DG norm. Two- and three-dimensional numerical tests are conducted to support the theoretical results.
Read the full abstract
Representation Theory
Cohomological boundedness of twisted coherent Springer sheaves
We prove that the coherent Springer sheaf and its parabolic analogues are concentrated in cohomological degree $0$, as predicted by Ben-Zvi-Chen-Helm-Nadler, Zhu, Emerton-Gee-Hellmann, Hansen, and others. More generally, we show that the universal trace functor for a mixed partial affine Hecke category is right t-exact with respect to the exotic t-structure given by Bezrukavnikov-Mirković's noncommutative Springer resolution, and left t-exact with respect to the monoidally dual t-structure. To this end, we construct an explicit complex computing the universal trace functor for certain monoidal categories over quotient stacks.
Read the full abstract
Numerical Analysis · Cross-listed from stat.ME
Model Error Embedding with Orthogonal Gaussian Processes
Computational models of complex physical systems often rely on simplifying assumptions which inevitably introduce model error, with consequent predictive errors. Given data on model observables, the estimation of parameterized model-error representations, along with other model parameters, would be ideally done while separating the contributions of each of the two sets of parameters, in order to ensure meaningful stand-alone model predictions. This work builds an embedded model error framework using a weight-space representation of Gaussian processes (GPs) to flexibly capture model-error spatiotemporal correlations and enable inference with GP-embedding in non-linear models. To disambiguate model and model-error/bias parameters, we extend an existing orthogonal GP method to the embedded model-error setting and derive appropriate orthogonality constraints. To address the increased dimensionality introduced by the GP representation, we employ the likelihood-informed subspace method. The construction is demonstrated on linear and non-linear examples, where it effectively corrects model predictions to match data trends. Extrapolation beyond the training data recovers the prior predictive distribution, and the orthogonality constraints lead to meaningful stand-alone model predictions and nearly uncorrelated posteriors between model and model-error parameters.
Read the full abstract
Spectral Theory
Stability of critical equipartitions of graphs
We consider partitions of a finite, simple, weighted graph that minimize a spectral energy functional, defined to be the maximum of the first eigenvalues on each component. These partitions are minimized with respect to a parameter that we view as a small perturbation of a fixed combinatorial partition. It has been shown that critical points of the energy functional in this framework correspond to non-degenerate eigenvectors of the graph Laplacian if and only if the partition is bipartite. In this work we generalize this result to partitions that are not necessarily bipartite by constructing a modified graph Laplacian called the partition Laplacian. The main result states that critical points of the spectral energy functional correspond to nodal partitions of non-degenerate eigenvectors of the partition Laplacian. Furthermore, the stability of each critical point is determined entirely by the nodal deficiency of the associated eigenvector. We also consider more general signed Laplacians and prove that Courant-sharp Laplacian eigenvectors induce globally minimal partitions.
Read the full abstract
Category Theory
Interactions that reshape the interfaces of the interacting parties
Polynomial functors model systems with interfaces: each polynomial specifies the outputs a system can produce and, for each output, the inputs it accepts. The bicategory $\mathbb{O}\mathbf{rg}$ of dynamic organizations \cite{spivak2021learners} gives a notion of state-driven interaction patterns that evolves over time, but each system's interface remains fixed throughout the interaction. Yet in many systems, the outputs sent and inputs received can reshape the interface itself: a cell differentiating in response to chemical signals gains or loses receptors; a sensor damaged by its input loses a channel; a neural network may grow its output resolution during training. Here we introduce *polynomial trees*, elements of the terminal $(u\triangleleft u)$-coalgebra where $u$ is the polynomial associated to a universe of sets, to model such systems: a polynomial tree is a coinductive tree whose nodes carry polynomials, and in which each round of interaction – an output chosen and an input received – determines a child tree, hence the next interface. We construct a monoidal closed category $\mathbf{PolyTr}$ of polynomial trees, with coinductively-defined morphisms, tensor product, and internal hom. We then build a bicategory $\mathbb{O}\mathbf{rgTr}$ generalizing $\mathbb{O}\mathbf{rg}$, whose hom-categories parametrize morphisms by state sets with coinductive action-and-update data. We provide a locally fully faithful functor $\mathbb{O}\mathbf{rg}\to\mathbb{O}\mathbf{rgTr}$ via constant trees, those for which the interfaces do not change through time. We illustrate the generalization by suggesting a notion of progressive generative adversarial networks, where gradient feedback determines when the image-generation interface grows to a higher resolution.
Read the full abstract
Metric Geometry
The local logarithmic Brunn-Minkowski inequality for bodies of revolution
We prove the local logarithmic Brunn-Minkowski inequality for bodies of revolution. Furthermore, we give a generalization for one origin symmetric body of revolution and one body of revolution that does not need to be symmetric and restrict possible equality cases. The proof uses an operator theoretic approach together with the decomposition of spherical functions into isotypical components with respect to rotations around a fixed axis.
Read the full abstract
Thursday 19 February
Mathematical Physics
Constructing Barut-Girardello coherent states for the isotonic oscillator in the DOOT approach
In this work, we study the quantum system of the isotonic oscillator from the perspective of the diagonal operator ordering technique (DOOT). Within this framework, we construct the associated Barut-Girardello and Gazeau-Klauder coherent states. We examine their mathematical properties using reproducing kernels and compute the expectation values of observables that characterize the system and its relevant physical features. Further, we perform the quantization of main classical variables in the complex plane. Then, by exploring the thermal behavior of the physical system in the constructed coherent states, we analyze the properties of mixed states described by a canonical density operator. We also obtain the corresponding Glauber-Sudarshan P-representation.
Read the full abstract
Combinatorics
The number of non-isomorphic arithmetic expressions that can be constructed using +,-,x and /
The goal of this paper is to count the number of distinct functions of n variables, up to permutation of the variables, that can be constructed using each variable exactly once, without constants, using only the operations of addition, subtraction, multiplication, and division. We refer to such a function as an arithmetic expression. Under this definition, two expressions are identical if they represent the same rational function; for example, $x_1-x_2-x_3$ and $x_1-(x_2+x_3)$ are identical arithmetic expressions, as are $x_1(x_2+x_3)$ and $(x_2+x_3)x_1$. Two arithmetic expressions are said to be isomorphic if one can be obtained from the other by a permutation of the variables. For example, $(x_1-x_2)/x_3$ and $(x_2-x_3)/x_1$ are isomorphic. The first few values of the number of non-isomorphic arithmetic expressions with n variables are: $$1,4,18,93,500,2844,16621,99674,608448,...$$ In order to accomplish this enumeration, we classify the set of all arithmetic expressions into 12 disjoint categories. Counting all non-isomorphic expressions in each category allows us to obtain the total required quantity.
Read the full abstract
Number Theory
Product of powers of distinct primes as sums of Fibonacci numbers
Let $F_n$ be the $n$-th Fibonacci number. In this paper, we study the Diophantine equation $F_n+F_m=p^xq^y$ in nonnegative integers $n\ge m$, $x$ and $y$, where $p$ and $q$ are fixed distinct prime numbers. We determine all pairs of primes $(q,p)$ with $q\le \min\{1000,p\}$ such that the above equation has at least two solutions $(x,y)$ (and corresponding $m,n$) in positive integers.
Read the full abstract
Statistics Theory
Central limit theorem for the global clustering coefficient of random geometric graphs
The global clustering coefficient serves as a powerful metric for the structural analysis and comparison of complex networks. Random geometric graphs offer a realistic framework for representing the spatial constraints and geometry often found in real-world network datasets. In this paper, we establish a central limit theorem for the global clustering coefficient of random geometric graphs. Our main result identifies the centering and scaling sequences required for convergence in law to the standard normal distribution. Our approach varies by regime: in the dense case, we employ the Lyapunov CLT; in the intermediate case, we utilize the asymptotic theory of $U$-statistics with sample-size-dependent kernels; and in the sparse regime, we use the method of moments to derive the asymptotic distribution. Notably, the convergence rates for non-uniform and uniform random geometric graphs diverge in the dense regime, yet they coincide in the sparse regime. In addition, we find that the global clustering coefficient for both uniform and non-uniform RGGs is asymptotically equal to $3/4$
Read the full abstract
Statistics Theory · Cross-listed from stat.ML
Learning from Biased and Costly Data Sources: Minimax-optimal Data Collection under a Budget
Data collection is a critical component of modern statistical and machine learning pipelines, particularly when data must be gathered from multiple heterogeneous sources to study a target population of interest. In many use cases, such as medical studies or political polling, different sources incur different sampling costs. Observations often have associated group identities - for example, health markers, demographics, or political affiliations - and the relative composition of these groups may differ substantially, both among the source populations and between sources and target population. In this work, we study multi-source data collection under a fixed budget, focusing on the estimation of population means and group-conditional means. We show that naive data collection strategies (e.g. attempting to "match" the target distribution) or relying on standard estimators (e.g. sample mean) can be highly suboptimal. Instead, we develop a sampling plan which maximizes the effective sample size - the total sample size divided by $D_{\chi^2}(q\mid\mid\overline{p}) + 1$, where $q$ is the target distribution, $\overline{p}$ is the aggregated source distribution, and $D_{\chi^2}$ is the $\chi^2$-divergence. We pair this sampling plan with a classical post-stratification estimator and upper bound its risk. We provide matching lower bounds, establishing that our approach achieves the budgeted minimax optimal risk. Our techniques also extend to prediction problems when minimizing the excess risk, providing a principled approach to multi-source learning with costly and heterogeneous data sources.
Read the full abstract
Numerical Analysis
Hybrid ABBA-GMRES for Unmatched Backprojectors in Large Scale X-Ray Computerized Tomography
In large-scale X-ray computed tomography (CT), matrix-free iterative methods are essential due to the prohibitive cost of explicitly forming the system matrix. In practice, forward projectors and backprojectors are often implemented with different discretizations or accelerations, leading to unmatched projector pairs. This mismatch violates the adjointness assumptions underlying classical least-squares solvers, so the resulting iterations no longer correspond to a true least-squares problem and can exhibit non-symmetric or inconsistent behavior. Prior work has explored Krylov subspace solvers such as AB-GMRES and BA-GMRES to handle unmatched projector pairs, where these methods exhibit semi-convergent regularizing behavior. Under matched conditions, AB-GMRES and BA-GMRES reduce to LSQR and LSMR, respectively. However, in the presence of unmatched projectors, AB- and BA-GMRES have been observed to yield improved reconstruction quality compared to classical least-squares solvers. In this paper, we develop hybrid AB- and BA-GMRES methods that incorporate Tikhonov regularization directly into the Krylov subspace iterations. We also examine the relationship between the proposed methods and hybrid variants of LSQR and LSMR, considering both matched and unmatched backprojectors. We propose automatic strategies for selecting regularization parameters, including approaches based on the L-curve and generalized cross validation (GCV), and analyze their effect on convergence behavior and image quality. Numerical experiments on two-dimensional CT problems using GPU-accelerated projectors demonstrate that the proposed hybrid AB- and BA-GMRES methods mitigate semi-convergence, produce higher-quality reconstructions, and exhibit more stable stopping behavior than their non-hybrid counterparts.
Read the full abstract
Optimization and Control
Multi-agent path-planning in a moving medium via Wasserstein Hamiltonian Flow
We present a finite dimensional variational model for multi-agent path-planning in which a group of agents traverses from initial positions to a target distribution in a moving medium. The model is derived using the agent-based formulation of the Wasserstein Hamiltonian flows that transport between probability distributions while optimizing a running cost. The objective is the mismatch between their final positions and the target distribution. The constraints are a system of Hamiltonian equations that provide the trajectories of the agents. The free variables on which the optimization is defined form a finite vector of the initial velocities for the agents. The model is solved numerically by the L-BFGS method in conjunction with a shooting strategy. Several simulation examples, including a time-dependent moving medium, are presented to illustrate the performance of the model.
Read the full abstract
Functional Analysis
Isometries and geometric liftings for Alexiewicz-normed $L^\infty$ spaces
We study spaces of essentially bounded functions on compact subsets of the real line, equipped with the Alexiewicz norm given by the supremum norm of the primitive. Using the associated measure projection, we classify their surjective linear isometries as weighted composition operators determined by a sign and an increasing bi-Lipschitz map between the corresponding measure intervals. We also give geometric criteria characterizing when this interval-level map lifts to a homeomorphism or to a bi-Lipschitz homeomorphism between the underlying compact sets.
Read the full abstract
Optimization and Control
The mean-field control problem for heterogeneous forward-backward systems
We study the problem of mean-field control when the state dynamics are given by general systems of forward-backward stochastic differential equations (FBSDEs) with heterogeneous mean-field interactions. Firstly, we introduce a novel methodology for reducing the well-posedness of such systems to that of a single randomized mean-field FBSDE. As a consequence, we show that, in the fully coupled case, smallness conditions yield existence and uniqueness for both the system itself and the associated variational and adjoint systems. Secondly, we derive a stochastic maximum principle and a verification for the mean-field control problem. This provides necessary and sufficient conditions for optimality.
Read the full abstract
Optimization and Control
Improved Analysis of Restarted Accelerated Gradient and Augmented Lagrangian Methods via Inexact Proximal Point Frameworks
This paper studies a class of double-loop (inner-outer) algorithms for convex composite optimization. For unconstrained problems, we develop a restarted accelerated composite gradient method that attains the optimal first-order complexity in both the convex and strongly convex settings. For linearly constrained problems, we introduce inexact augmented Lagrangian methods, including a basic method and an outer-accelerated variant, and establish near-optimal first-order complexity for both methods. The established complexity bounds follow from a unified analysis based on new inexact proximal point frameworks that accommodate relative and absolute inexactness, acceleration, and strongly convex objectives. Numerical experiments on LASSO and linearly constrained quadratic programs demonstrate the practical efficiency of the proposed methods.
Read the full abstract
Statistics Theory · Cross-listed from stat.ML
Interactive Learning of Single-Index Models via Stochastic Gradient Descent
Stochastic gradient descent (SGD) is a cornerstone algorithm for high-dimensional optimization, renowned for its empirical successes. Recent theoretical advances have provided a deep understanding of how SGD enables feature learning in high-dimensional nonlinear models, most notably the single-index model with i.i.d. data. In this work, we study the sequential learning problem for single-index models, also known as generalized linear bandits or ridge bandits, where SGD is a simple and natural solution, yet its learning dynamics remain largely unexplored. We show that, similar to the optimal interactive learner, SGD undergoes a distinct “burn-in” phase before entering the “learning” phase in this setting. Moreover, with an appropriately chosen learning rate schedule, a single SGD procedure simultaneously achieves near-optimal (or best-known) sample complexity and regret guarantees across both phases, for a broad class of link functions. Our results demonstrate that SGD remains highly competitive for learning single-index models under adaptive data.
Read the full abstract
Geometric Topology
Positive braid closures and taut foliations
We study taut foliations on the complements of non-split positive braid closures in $S^3$. If $L$ is such a link with components $L_1,\ldots,L_n$ and at least one component is not the unknot, then the Dehn surgery along a multislope $(s_1,\ldots,s_n)\in\mathbb{Q}^n$ satisfying $s_i<2g(L_i)-1$ for $i=1,2,\ldots, n$ yields a non-L-space that admits a co-oriented taut foliation.
Read the full abstract
Differential Geometry
A discrete view of Gromov's filling area conjecture
A compact metric surface $M$ isometrically fills a closed metric curve $C$ if $\partial M=C$ and $d_M(x,y)=d_C(x,y)$ for every $x,y\in C=\partial M$; that is, $M$ does not introduce any “shortcuts” between points on its boundary. Gromov's filling area conjecture in differential geometry from 1983 asserts that among all compact, orientable Riemannian surfaces which isometrically fill the Riemannian circle, the one with the smallest surface area is the hemisphere. Gromov demonstrated that this is indeed the case if $M$ is homeomorphic to the disk. While Gromov's conjecture has since been verified in some other cases, the full conjecture remains unresolved. In this paper, we consider a discrete analogue of Gromov's problem, which is likely natural to those who study graph embeddings on arbitrary surfaces. Using standard graph-theoretic tools, such as Menger's theorem, we obtain reasonable asymptotic bounds on this discrete variant. We then demonstrate how these discrete bounds can be translated to the continuous setting, showing that any isometric filling of the Riemannian circle of length $2\pi$ has surface area at least $1.36\pi$ (the hemisphere has surface area $2\pi$). This appears to be the first quantitative lower-bound on Gromov's problem that applies to arbitrary isometric fillings.
Read the full abstract
Mathematical Physics · Cross-listed from hep-th
Chern-Simons deformations of the gauged O(3) Sigma model on compact surfaces
Existence of solutions to the field equations of the gauged Chern-Simons-O(3)-Sigma model on a compact Riemann surface is proved by a topological method. Existence of a minimal deformation constant $\kappa_{*} > 0$ is proved, such that for any prescribed configuration of vortices and antivortices, at least one solution exists for $|\kappa| \leq \kappa_{*}$. For small values of the Chern-Simons deformation parameter $\kappa$, it is proved that the field equations admit multiple solutions, provided the total number of vortices and antivortices are different. The Maxwell limit is computed for solutions of the field equations. In contrast, if the number of vortices equals the number of antivortices, it is proved that the field equations admit at least one solution for any value of $\kappa$ and the limit $\kappa \to \infty$ is proved. dependence of the fields on the deformation parameter is investigated numerically on the sphere.
Read the full abstract
Dynamical Systems
Mean-field dynamics of attractive resource interaction: From uniform to aggregated states
We introduce and study a nonlinear discrete dynamical system describing the evolution of a resource distribution among interacting agents. The model generalizes several classical mean-field and opinion-dynamics frameworks and is defined on the standard simplex, where each coordinate evolves according to an interaction rule depending on preference-based mean-field interactions. We provide a complete analytical description of the long-term behavior of the system. First, we establish monotonicity properties and show that the dynamics always remains in a positively invariant region determined by initial conditions. We prove the existence of a unique fixed point for any admissible parameter set and derive an explicit closed-form formula for the equilibrium in arbitrary dimension. We then analyze the local stability of the fixed point and identify parameter regimes leading to aggregation or uniform distributions. Finally, we characterize all possible asymptotic scenarios and show that, despite the nonlinear structure, the system does not exhibit oscillatory or chaotic behavior: every trajectory converges to the unique equilibrium. The results provide a full qualitative theory for this class of monotone resource-interaction models and offer a mathematical explanation for the transition from uniform to aggregated states.
Read the full abstract
Information Theory · Cross-listed from cs.LG
Quad Length Codes for Lossless Compression of e4m3
Training and serving Large Language Models (LLMs) relies heavily on parallelization and collective operations, which are frequently bottlenecked by network bandwidth. Lossless compression using e.g., Huffman codes can alleviate the issue, however, Huffman codes suffer from slow, bit-sequential decoding and high hardware complexity due to deep tree traversals. Universal codes e.g., Exponential-Golomb codes are faster to decode but do not exploit the symbol frequency distributions. To address these limitations, this paper introduces Quad Length Codes, a hybrid approach designed to balance compression efficiency with decoding speed. The coding scheme uses 3 prefix bits to divide the 256 symbols into 8 areas. Each area has a different code length and encodes a different number of symbols. The scheme uses a Look Up Table with 256 entries, significantly simplifying the hardware implementation compared to Huffman trees. The coding scheme can be adapted for different distributions. For the e4m3 data type, the scheme achieves a compressibility of 13.9% in comparison to 15.9% achieved by Huffman codes, but it significantly speeds up the decoding and simplifies the hardware complexity.
Read the full abstract
Optimization and Control
Brockett Openness Profiles and Gain-Limited Feedback Stabilization
Brockett's necessary condition asserts that a continuously stabilizable nonlinear control system must have a vector field that is open at the equilibrium. We show that the quantitative data behind this openness condition constrains the possible growth of stabilizing feedbacks. To a system vector field $f$, we associate its openness profile $\Omega_f(r)=\sup\{\rho:\mathbb{B}_\rho(0)\subset f(\mathbb{B}_r(0,0))\}$, so that Brockett's condition becomes $\Omega_f(r)>0$ for all sufficiently small $r>0$. If a feedback $u$ satisfies $\|u(x)\|\leq d(\|x\|)$, then the openness profile of the closed-loop field $F_u(x)=f(x,u(x))$ satisfies $\Omega_{F_u}(r)\leq \Omega_f\!\left(\sqrt{r^2+d(r)^2}\right)$. Consequently, any prescribed lower openness rate for the closed-loop dynamics yields a necessary lower bound on the feedback growth. For systems with $\Omega_f(r)\lesssim r^q$, linear-rate closed-loop openness forces $d(r)\gtrsim r^{1/q}$, and this exponent is sharp in elementary polynomial examples. Thus Brockett's condition is not merely a binary topological obstruction; its quantitative profile governs gain requirements for stabilizing feedback.
Read the full abstract
Optimization and Control
A Refinement in Čech Cohomology of Coron's Necessary Condition
Coron established a homological obstruction to continuous feedback stabilization of nonlinear control systems $\dot{x}=f(x,u)$ with $f \in C(\Omega,\mathbb{R}^n)$ and $f(0,0)=0$, showing that local asymptotic stabilizability implies the induced homomorphism $f_*$ satisfies $f_*\big(H_{n-1}(\Sigma_\epsilon)\big)=H_{n-1}(S^{n-1})$, where $\Sigma_\epsilon:=\Big(\big(\mathbb{B}_\epsilon^{\mathbb{R}^n}(0)\times\mathbb{B}_\epsilon^{\mathbb{R}^m}(0)\big)\cap \Omega\Big)\setminus f^{-1}(0)$. In this paper, we refine Coron's necessary condition using Čech cohomology and the Vietoris-Begle mapping theorem. Specifically, we prove that the closed version of $\Sigma_\epsilon$ must be a Čech cohomology $(n-1)$-sphere and that the restriction of $f$ to this subset induces an isomorphism on its Čech cohomology groups in all degrees. This strengthens Coron's condition from a constraint on the top class to a full cohomological rigidity statement.
Read the full abstract
Analysis of PDEs
Invariant Manifolds for Capillary Waves and a Class of Quasilinear PDEs
This paper studies the local stable and unstable manifolds of equilibria for quasilinear and fully nonlinear PDEs. These manifolds are fundamental objects in the analysis of local dynamics. While their existence is well understood for ODEs, semilinear PDEs, and certain parabolic-type quasilinear PDEs, invariant manifold theorems are often unavailable for quasilinear PDEs whose nonlinearities involve a loss of regularity and whose linear parts do not provide sufficient smoothing. Our main results establish the existence, uniqueness, and smoothness of local stable and unstable manifolds for nonlinear PDEs that satisfy suitable energy estimates. With the main focus on irrotational water waves with surface tension, this framework applies to a broad class of PDEs, including nonlinear Schrödinger equations, nonlinear wave equations, and the MMT model, as well as to certain gradient-type PDEs.