Statistics Theory
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
Friday 8 May
This paper deals with the kernel density estimator based on the so-called sinc (or Fourier integral) kernel $K(x)=(\pi x)^{-1}\sin x$. We study in detail both asymptotic and finite sample properties of this estimator. It is shown that, contrary to widespread opinion, the sinc estimator is superior to other estimators in many respects: it is more accurate for quite moderate values of the sample size, has better asymptotics in non-smooth case (the density to be estimated has only first derivative), is more convenient for the bandwidth selection, etc.
Functional Analysis · Cross-listed from cs.LG
Convergent Stochastic Training of Multi-Headed Attention and Understanding LoRA
Transformers have revolutionized machine learning and deploying attention layers in the model is increasingly standard across a myriad of applications. Further, for large models, it is common to implement Low Rank Adaptation (LoRA), whereby a factorized parameterization of them is trained, to achieve a surprisingly beneficial accuracy-size trade-off. In this work, via a unified framework we rigorously establish trainability of such models under stochastic methods. We prove that for a class of mild regularizations, the empirical regression loss on a attention layer and LoRA on a shallow neural net, both induce Poincaré inequality for the corresponding Gibbs' measure. Crucially, we show that the Poincaré constant is free of the data dimension for LoRA and for multi-head attention it is free of the head dimensions, Then it follows, via invoking recent results, that a certain SDE, which mimics the SGD, minimizes the corresponding losses. In both the cases, our first-of-its-kind results of trainability on attention and nets, do not use any assumptions on the data or the model size.
Read the full abstract
Algebraic Geometry
Adjoint test modules along Cohen–Macaulay morphisms
We provide a transformation rule for adjoint test modules along Cohen–Macaulay maps between Cohen–Macaulay varieties that have $F$-rational geometric fibers. This is, in part, an effective version of Enescu's theorem on the ascent of $F$-rationality under local maps with $F$-rational geometric fibers.
Read the full abstract
Analysis of PDEs
Noise-Driven Free Boundaries In The Compressible Navier-Stokes Equations
A stochastic free-boundary problem for the three-dimensional barotropic compressible Navier–Stokes equations is studied. The main feature of the model is that the free boundary is transported by a Stratonovich stochastic flow, so that the noise enters the kinematic boundary condition and hence the evolution of the moving domain. An additional Itô forcing in the momentum equation is also allowed. The problem is transformed by a stochastic Lagrangian map generated by the velocity and the transport vector fields. In these coordinates the density is represented through the Jacobian of the flow, and the remaining system is solved by combining stochastic maximal regularity, deterministic %\rL^p%-%\rL^q$ estimates, and a localized contraction argument. Local pathwise well-posedness is obtained up to an a.s. positive stopping time, with strictly positive density and pathwise uniqueness.
Read the full abstract
Optimization and Control · Cross-listed from cs.CE
Exploring the non-convexity in machine learning using quantum-inspired optimization
The escalating complexity of modern machine learning necessitates solving challenging non-convex optimization problems, particularly in high-dimensional regimes and scenarios contaminated by gross outliers. Traditional approaches, relying on convex relaxations or specialized local search heuristics, frequently succumb to suboptimal local minima and fail to recover the true underlying discrete structures. In this paper, we propose treating these non-convex challenges as a global search problem and introduce a unified framework based on Quantum-Inspired Evolutionary Optimization (QIEO). By leveraging a probabilistic representation inspired by quantum superposition, QIEO maintains a global view of the search space, enabling it to tunnel through local optima that trap conventional gradient-based and greedy solvers. We comprehensively evaluate QIEO across diverse non-convex applications, including sparse signal recovery (gene expression analysis and compressed sensing) and robust linear regression. Extensive benchmarking against state-of-the-art continuous solvers (ADAM, Differential Evolution), classical metaheuristics (Genetic Algorithms), and specialized non-convex algorithms (Iterative Hard Thresholding) demonstrates that QIEO consistently achieves superior structural fidelity, lower mean squared error, and enhanced robustness without support inflation. Our findings suggest that embracing a quantum-inspired global search provides a resilient, unified paradigm for overcoming the inherent intractability of discrete nonconvex machine learning landscapes.
Read the full abstract
Algebraic Geometry
Homogeneous pre-foliations of co-degree one and degree four on the projective plane
We classify, up to projective automorphism, all homogeneous pre-foliations of co-degree $1$ and degree $4$ on the complex projective plane $\Ptwo$ whose Legendre transform defines a flat $4$-web. The classification is organized according to the type of the underlying homogeneous foliation $\Hcal$ of degree~$3$, distinguishing the cases $\deg(\Tcal_{\Hcal})=2$, $3$, and~$4$. The case $\deg(\Tcal_{\Hcal})=2$ was treated by Bedrouni, while the cases $\deg(\Tcal_{\Hcal})=3$ and $\deg(\Tcal_{\Hcal})=4$ are completed here. The proof combines Bedrouni's curvature-holomorphy criteria with explicit normal forms and symbolic computation; the result yields a finite list of explicit one-forms, parametrized by the ramification data of the Gauss map of~$\Hcal$, all displayed explicitly in the article (the few cases whose parameters are roots of higher-degree polynomials being written out in full in Appendix A).
Read the full abstract
Statistics Theory
Accelerating Langevin Monte Carlo via Efficient Stochastic Runge–Kutta Methods beyond Log-Concavity
Sampling from a high-dimensional probability distribution is a fundamental algorithmic task arising in wide-ranging applications across multiple disciplines, including scientific computing, computational statistics and machine learning. Langevin Monte Carlo (LMC) algorithms are among the most widely used sampling methods in high-dimensional settings. This paper introduces a novel higher-order and Hessian-free LMC sampling algorithm based on an efficient stochastic Runge–Kutta method of strong order $1.5$ for the overdamped Langevin dynamics. In contrast to the existing Runge–Kutta type LMC (Li et al., 2019) involved with three gradient evaluations, the newly proposed algorithm is computationally cheaper and requires only two gradient evaluations for one iteration. Under certain log-smooth conditions, non-asymptotic error bounds of the proposed algorithms are analyzed in $\mathcal{W}_2$-distance. In particular, a uniform-in-time convergence rate of order $O(d ^{\frac32} h^{\frac32})$ is derived in a non-log-concave setting, matching the convergence rate proved in the aforementioned work but under the log-concavity condition. Numerical experiments are finally presented to demonstrate the effectiveness of the new sampling algorithm.
Read the full abstract
Mathematical Physics · Cross-listed from hep-th
Topological Blocking of the Schwinger Effect in the Salpeter Equation: A Lefschetz Thimble Analysis
We present a comprehensive Lefschetz thimble analysis of the one-dimensional Salpeter equation under a strong electric field. By treating the non-local square-root operator within the framework of algebraic analysis, we construct the full solution space, which includes relativistic generalizations of the Airy Ai and Bi functions and their negative-energy counterparts. Through a direct comparison with the Dirac and Klein-Gordon equations, we provide a geometric explanation for the absence of Klein paradox and the Schwinger effect in the Salpeter equation. Furthermore, our findings establish a unified geometric interpretation of the Schwinger effect across different relativistic wave equations.
Read the full abstract
History and Overview of Mathematics
Notes on Beltrami's Essay
Eugenio Beltrami published his seminal 'Essay on the Interpretation of Non-Euclidean Geometry' in 1868, where he showed that geodesics on a surface of constant negative curvature can be mapped as straight lines on a Euclidean disc. More importantly he showed that figures on the disc would satisfy the identities of hyperbolic geometry characteristic of a surface of negative curvature. However Beltrami did not always give a full explanation of the equations which he used. These notes are an attempt to provide a derivation of some of his principal results, including his formula for hyperbolic distance on the disc, his proof that the sum of the (hyperbolic) angles of a triangle on the disc is less than two right angles and his equations for circles, equidistants and horocycles.
Read the full abstract
Combinatorics
The number and structure of connected graphs with a fixed degree sequence
We study connected graphs with a fixed degree sequence, in the sparse setting where the number of edges grows linearly in the number of vertices. Using the relation to the configuration model, we identify the number of such connected graphs up to the exponential order. We do this by viewing a connected graph with a given degree distribution as the realization of the giant component in a larger configuration model, and carefully choosing the degree distribution of the larger graph so that it is likely that its giant component has the required degree distribution. To ensure that the connected graph has exactly the correct degrees, we use a switching argument. Additionally, we obtain results on rare event probabilities and describe the local structure of a uniform connected graph with a fixed degree sequence.
Read the full abstract
Probability
Primitive Sequences for Probability Measures on Compact Intervals
We introduce a sequence representation of a random variable $X$ supported on a compact interval $[a,b]$, which we call a primitive sequence. We construct this sequence by repeatedly antidifferentiating the associated cumulative distribution function of $X$ and evaluating the antiderivatives at the endpoint $b$. We show that the primitive sequence of $X$ can be identified as a factorially rescaled moment sequence of the reflected random variable $b-X$. Through this identification, we show that the primitive sequence transparently captures qualitative features of the distribution of $X$. We then connect primitive sequences directly to classical moment theory and exploit this connection to characterize admissible primitive sequences and to show that under natural topologies, the map from probability measures to primitive sequences is a homeomorphism. We end by examining the set of probability measures whose first $m$ primitive sequence terms are fixed, and thereby obtaining sharp upper and lower bounds on two functionals of those measures.
Read the full abstract
Number Theory
A strengthening of Chang's lemma
We prove a strengthening of Chang's lemma for subsets of $\mathbb F_p^n$. The classical conclusion that the large spectrum is contained in a subspace of dimension at most $2\varepsilon^{-2}\log(1/\alpha)$ is refined to show that every character outside this subspace has small correlation with the set not only globally, but also on average over the cosets of the orthogonal complement, in a natural cosetwise $\ell^1$ norm. As a consequence, we obtain a localized counting lemma. We also give an extension of the argument to arbitrary finite abelian groups.
Read the full abstract
Analysis of PDEs
Finite index solutions to the Bernoulli problem in three dimensions are axially symmetric
We show that every entire solution to the Bernoulli (or one-phase) free boundary problem with finite Morse index in $\mathbb{R}^3$ is axially symmetric. In fact, we additionally prove that the same result would follow in any dimension $4 \le n \le 6$ in which stable entire solutions are shown to be flat.
Read the full abstract
Analysis of PDEs
Blow-up of solutions to semilinear parabolic equations driven by mixed local-nonlocal operators with large initial data
We investigate finite-time blow-up for nonnegative solutions to the Cauchy problem associated with semilinear parabolic equations driven by a mixed local–nonlocal operator. The reaction term is assumed to satisfy suitable structural hypotheses, the prototype being $f(u)=u^p$ with $p>1$. By adapting the Kaplan method to the present framework, we prove that solutions blow up in finite time whenever the initial datum is sufficiently large. In the prototype case $f(u)=u^p$, this conclusion holds for every $p>1$. As a particular case of our operator, we also include the fractional Laplacian; to the best of our knowledge, this type of result is new even in that special case.
Read the full abstract
Statistics Theory
Statistical inference with belief functions: A survey
Belief functions are a powerful and popular framework for the mathematical characterisation of uncertainty, in particular in situations in which lack of data renders learning a probability distribution for the problem impractical. The first step in a reasoning chain based on belief functions is inference: how to learn a belief measure from the available data. In this survey we focus, in particular, on making inference from statistical data, and review the most significant contributions in the area.
Read the full abstract
Logic
On Many-logic modal structures and information-based logics
This paper proposes an approach to information-based logics using many-logic modal structures (MLMS). These structures can express accessibility relations between worlds with different underlying logics by anchoring them to a base lattice, which contains the semantics of each logic as a down-complete sublattice. MLMS are suitable for representing connections between information states (i.e., configurations of databases) and the evolution of information states over time. We will illustrate the application of MLMS by means of the six-valued logic of evidence and truth LET+K , related to the lattice L6, and some four-, three-, and two-valued logics related to down-complete sublattices of L6. These logics are capable of representing paracomplete, paraconsistent, and classical contexts with six-, four-, three-, and two-valued scenarios.
Read the full abstract
Algebraic Topology
Spectra of bi-incomplete Tambara functors
Bi-incomplete Tambara functors are equivariant generalizations of commutative rings. The most common forms of bi-incomplete Tambara functors are coefficient systems of commutative rings, Green functors, and Tambara functors. In the 1980s, Lewis introduced prime ideals in Green functors, and in the 2010s, Nakaoka introduced prime ideals in Tambara functors. In this work, we define the spectrum of prime ideals for an arbitrary bi-incomplete Tambara functor, simultaneously generalizing Lewis and Nakaoka's notions. We then produce many computational tools which we apply to several examples of interest.
Read the full abstract
Logic
Examples of non-tame abstract elementary classes of abelian groups
We construct an abstract elementary class $K_1$ of torsion-free abelian groups such that $K_1$ is not $(<\aleph_0)$-tame but is $\aleph_0$-tame. This answers a question of [BoVa17]. Furthermore, for every regular uncountable cardinal $\mu$ less than the first measurable cardinal, we construct an abstract elementary class $K_2(2^\mu)$ of torsion-free abelian groups such that $K_2(2^\mu)$ is not $(<\mu)$-tame. $K_1$ and $K_2(2^\mu)$ are non-tame for algebraic reasons. Furthermore, they constitute the first examples of non-tame abstract elementary classes in a natural language.
Read the full abstract
Information Theory
Physics-Inspired Probabilistic Computing for Extremely Large-Scale MIMO Detection in Future 6G Wireless Systems
Extremely large-scale multiple-input multiple-output (XL-MIMO) architectures are a key enabler of forthcoming 6G wireless communication networks by allowing high data rates through massive spatial multiplexing. Here, we approach these problems with physics-inspired unconventional computing based on Ising machines (IMs). For binary modulation, probabilistic IMs (PIMs) and oscillator-based IMs achieve optimal ML detection with systems up to 2048x2048 antennas with only 100 iterations, matching optimal sphere decoder performance for computationally treatable sizes and outperforming the minimum mean-square error (MMSE) industrial standard. For M-QAM up to 256, a generalized PIM-inspired framework, based on d-dimensional probabilistic variables (p-dits) that directly encode QAM symbols, shows low bit-error-rate across sizes up to 256x256 antennas, outperforming or matching MMSE with reduced algorithmic complexity. Unlike the binary mapping, the p-dit interaction matrix is independent of the QAM order, enabling adaptive MIMO modulation. These results show a promising scalable paradigm for XL MIMO detection in future 6G networks.
Read the full abstract
Optimization and Control
Robust Capacity Expansion under Wildfire Ignition Risk and High Renewable Penetration
In power systems, the risk of wildfire ignition has increased significantly in recent years. The impact and severity of these events on energy dispatch, as well as their societal ramifications, make wildfire prevention critical for power system planning and operation. A common intervention by system operators is to de-energize transmission lines to mitigate the risk of fire caused by equipment failures. With the growing integration of variable renewable generation, managing and preparing the system to de-energization under wildfire risk has become even more challenging. In this context, mitigation decisions such as installing battery energy storage systems and undergrounding transmission lines can reduce the risk and adverse effects associated with de-energization and renewable generation variability. This paper presents a robust optimization model to determine the optimal location of battery storage and undergrounding of transmission line investment, utilizing representative weeks and uncertainty sets to capture the temporal relationship of uncertain variables. Specifically, this paper addresses: (i) the worst-case realization of ignition risk leading to the de-energization of transmission lines, combined with the worst-case realization of renewable energy availability, and (ii) the optimal investment decisions for energy storage capacity and undergrounding of transmission lines that are exposed to ignition risk. The proposed model is formulated as a mixed-integer linear programming (MILP) problem, employing duality theory and binary decomposition to address nonlinearities, and is solved using a column-and-constraint generation algorithm. The proposed framework is evaluated on a model of the San Diego power system, demonstrating its practical effectiveness in improving the resilience to wildfire risk.
Read the full abstract
Number Theory
Remarks on the distribution of Dirichlet $L$-functions along cosets
In a previous work with B. Garcia, the author considered the asymptotic for the second moment of Dirichlet $L$-functions along cosets, and exhibited a surprising secondary main term that is not predicted by the recipe of Conrey, Farmer, Keating, Rubinstein, and Snaith. In this paper, we re-examine this problem and propose a modified recipe that correctly predicts this secondary main term. The original recipe gives the incorrect answer for this family because the root number is not always independent of the Dirichlet series coefficients along certain cosets, and our proposed fix simply takes this feature into account. In addition, we consider a handful of other problems related to Dirichlet $L$-functions along cosets. One goal is to reformulate Heath-Brown's $q$-analog of van der Corput's shifting method in terms of cosets, which leads to an upper bound on a hybrid second moment. We also revisit the classical van der Corput bound and view it (in more modern terms) as an amplified second moment of a trigonometric polynomial.
Read the full abstract
Group Theory
Universal Structure of Graph Product Kernels
Let $G_\Gamma$ be a graph product over a finite simplicial graph $\Gamma$, and let $K_\Gamma$ denote the kernel of the canonical homomorphism from $G_\Gamma$ to the direct product of its vertex groups. It is known that, up to isomorphism, $K_\Gamma$ depends only on the underlying graph $\Gamma$ and the cardinalities of the vertex groups. In this paper we establish a functorial refinement of this fact. We show that any collection of set maps between the vertex groups naturally induces a homomorphism between the corresponding kernels, and that this construction is functorial. Several applications are discussed.
Read the full abstract
Algebraic Topology
Shape theory for condensed anima
We give different perspectives on the notion of shape for condensed anima. We prove that it recovers more classical notions of shape for topological spaces in the cases of all paracompact compactly generated spaces and all locally contractible spaces. These recovering statements imply and extend comparison results on sheaf and condensed cohomology by Clausen-Scholze and Haine. Another homotopy-theoretical direction for condensed anima are their condensed homotopy groups. Connected to this, we give a description of the underlying topological group functor on condensed groups via quasi-topological groups.
Read the full abstract
Functional Analysis
Stable phase retrieval from short-time linear canonical transforms of signals in Gaussian shift-invariant spaces
Gabor phase retrieval for signals has attracted considerable attention in recent years. For the more general short-time linear canonical transform (STLCT), which arises naturally in optical systems and canonical time–frequency analysis, existing work has so far focused mainly on uniqueness and sampling conditions. Explicit reconstruction formulas, quantitative stability estimates, and robust reconstruction algorithms, however, are still missing. In this paper, we study uniqueness, stability, and robust reconstruction for phase retrieval from phaseless STLCT measurements in the complex Gaussian shift-invariant space $V_\beta^\infty(\varphi)$. We first prove that every signal in $V_\beta^\infty(\varphi)$ is uniquely determined, up to a global unimodular constant, by its phaseless STLCT measurements on the semi-discrete set $\frac{\beta}{2}\mathbb Z\times\mathbb R$, and we derive an explicit reconstruction formula. We then establish stability on intervals under an anchor-point condition, showing that the stability constant is governed by the maximal spacing between adjacent anchor points rather than by the radius of the whole interval. This prevents exponential deterioration with respect to the interval size. Motivated by the practical setting in which only finitely many discrete noisy magnitude samples are available, we further develop an explicit reconstruction algorithm with quantitative robustness guarantees, where the reconstruction error is controlled by the discretization parameters, the noise level, and the conditioning induced by the anchor points. In the Fourier case, our results recover the corresponding Gabor phase retrieval results of Grohs and Liehr and provide improved stability constants.
Read the full abstract
Probability
A Bismut-Elworthy formula for BSDEs with degenerate noise
In this paper we derive a Bismut-Elworthy formula under assumptions weaker than the non degeneracy of the noise. By Bismut-Elworthy formula we mean a gradient type estimate on the transition semigroup of a stochastic differential equation in a possibly infinite dimensional Hilbert space. We also consider a nonlinear version of the Bismut formula for a backward stochastic differential equation, in analogy to what is done in \cite{futeBismut}, where a non-degenerate noise is considered. Our study is motivated by applications to stochastic wave equations and to stochastic damped wave equation.
Read the full abstract
Probability · Cross-listed from stat.ME
Parametric ROC Analysis and Optimal Cutoff Selection under Scale Mixtures of Skew-Normal Distributions: A Decision-Theoretic Framework with Asymptotic Inference
We study an optimal threshold functional arising in binary classification for continuous biomarkers. While the ROC curve summarizes discriminatory performance across all thresholds, practical threshold selection must also account for disease prevalence and asymmetric misclassification costs. The classical Youden index corresponds to a symmetric special case and may therefore be suboptimal in realistic decision settings. In addition, biomarker distributions in serological and immunological studies often display skewness and heavy tails, making Gaussian ROC models inadequate. We develop a parametric framework for ROC analysis and optimal cutoff selection under the family of scale mixtures of skew-normal (SMSN) distributions, including the skew-normal and skew-t models. The ROC curve and AUC are estimated by plug-in maximum likelihood from separate group fits. The optimal cutoff is defined as the minimiser of a weighted misclassification risk, which yields a likelihood ratio equation extending the Youden criterion. Under a monotone likelihood ratio condition, we establish existence, uniqueness, and global optimality of the cutoff. We further study its local regularity as an implicitly defined functional of the model parameter and derive consistency, asymptotic normality, and a closed-form plug-in variance estimator. A central term in this variance is the local slope of the estimating equation at the optimal threshold, which acts as a local identifiability diagnostic. Monte Carlo experiments across six scenarios show that the asymptotic approximation is accurate and that Wald confidence intervals attain near nominal coverage. An application to SARS-CoV-2 serological data illustrates that the proposed cutoff can differ substantially from the Youden threshold and may reduce estimated misclassification risk by up to 63% under asymmetric decision settings.
Read the full abstract
Numerical Analysis
NSPOD: Accelerating Krylov solvers via DeepONet-learned POD subspaces
The convergence of Krylov-based linear iterative solvers applied to parametric partial differential equations (PDEs) is often highly sensitive to the domain, its discretization, the location/values of the applied Dirichlet/Neumann boundary conditions, body forces and material properties, among others. We have previously introduced hybridization of classical linear iterative solvers with neural operators for specific geometries, but they tend to not perform well on geometries not previously seen during training. We partially addressed this challenge by introducing the deep operator network Geo-DeepONet and hybridizing it with Krylov-based iterative linear solvers, which, despite learning effectively across arbitrary unstructured meshes without requiring retraining, led to only modest reductions in iterations compared to state-of-the-art preconditioners. In this study we introduce Neural Subspace Proper Orthogonal Decomposition (NSPOD), a multigrid-like deep operator network-based preconditioner which can dramatically reduce the number of iterations needed for convergence in Krylov-based linear iterative solvers, even when compared to state-of-the-art methods such as algebraic multigrid preconditioners. We demonstrate its efficiency via numerical experiments on a linearized version of solid mechanics PDEs applied to unstructured domains obtained from complex CAD geometries. We expect that the findings in this study lead to more efficient hybrid preconditioners that can match, or possibly even surpass, the convergence properties of the current gold standard preconditioning methods for solid mechanics PDEs.
Read the full abstract
Mathematical Physics · Cross-listed from cond-mat.stat-mech
Tamed Feynman-Kac diffusion processes: Killing-branching intertwine
Relaxation to equilibrium of a drifted Brownian motion is quantified by a transition probability density function, whose main (multiplicative) entry is an inferred Feynman-Kac kernel of the Schrödinger semigroup operator. Although seemingly devoid of a natural probabilistic significance (except for its explicit path integral definition), the pertinent kernel relaxes to equilibrium as well. The implicit Feynman-Kac potential ${\cal{V}}(x)$, continuous, confining and bounded from below, may take negative values. If positive, ${\cal{V}}(x)$ can be interpreted as the killing rate of the decaying diffusion process. In case of relaxing F-K kernels the killing effects are tamed (often overcompensated). The taming inavoidably appears in conjunction with the existence of the negativity subdomains of ${\cal{V}}(x)$ in $R$. If locally ${\cal{V}}(x) < 0$, its sign inversion $- {\cal{V}}(x)$ can be interpreted as the branching (cloning, alternatively bifurcation) rate in the course of the other wise free random motion. The arising killed diffusion processes with branching, we interpret as the possible path-wise background of tamed (relaxing) Feynman-Kac diffusions. We present acomputer-assisted path-wise arguments, towards a consistency of the killing/branching taming scenario, for a number of nonlinear model systems in one space dimension. Special attention is paid to Feynman-Kac potential shapes in the double well form, where an analytic access to eigenvalues and eigenfunctions is scarce. Throughout the paper the dynamics refers to the positive real time. Since the Newton-type equations of motion for admissible classical trajectories have a Euclidean form (due to the sign inverted force term), we give a brief resume of a couple of their explicit solutions, without recourse to the Euclidean time intuitions, and the instanton lore of related quantum model systems.
Read the full abstract
Statistics Theory · Cross-listed from stat.ML
Expectation-Maximization as a Spectrally Governed Relaxation Flow
The expectation–maximization (EM) algorithm combines global monotonicity, local linear convergence, and strong practical robustness, but these features are usually analyzed separately. Global descent is nonlinear, whereas local convergence is governed by the spectrum of the linearized EM map. How these two levels fit into a single dynamical picture has remained less transparent. We make explicit the latent-variable operator that connects them. Along the EM trajectory, the likelihood increment admits a global energy decomposition in terms of posterior-relative entropy. Linearization at a nondegenerate maximizer $\theta^\ast$ reveals the local operator \[ \mathcal G_{\theta^\ast}=I-DT(\theta^\ast), \] which coincides with both the missing-information ratio and the information-geometric Hessian of the observed likelihood. From this operator we derive two acceleration strategies. The G-Accelerator uses the spectral gap to obtain an optimal Nesterov-type momentum $\beta^* = (1-\sqrt{\lambda_*})/(1+\sqrt{\lambda_*})$. The Geo-Adaptive accelerator extends the geometric EM framework of Zhou, Alexander & Lange by replacing their fixed correction strength $\gamma=8$ with the adaptive rule $\gamma_k = 1/\hat\lambda_k$, where $\hat\lambda_k$ is estimated online from the parameter trajectory. Both methods are parameter-free; Geo-Adaptive achieves dramatic acceleration precisely when the spectral gap is smallest. Numerical experiments on Gaussian mixtures demonstrate that both accelerators consistently outperform standard EM and fixed-$\gamma$ DCC-EM, with Geo-Adaptive attaining speedups exceeding $8\times$ in the most challenging regimes.
Read the full abstract
Differential Geometry
On generalized imaginary $\mathrm{Spin}^c$-Killing spinors
A non-trivial spinor field $\psi$ is called a generalized imaginary $\mathrm{Spin}^c$-Killing spinor if $\nabla^{g,A} _X \psi = i\mu X \cdot \psi$ for all vector fields $X$, where $\mu$ is a real function that is not identically zero and $\nabla^{g,A}$ is the $\mathrm{Spin}^c$ Levi-Civita connection with $\mathrm{U}(1)$-connection $A$. Associated with $\psi$ is a vector field $V$, the Dirac current, defined by $g(V,X) = i \langle X\cdot \psi, \psi \rangle$. We prove that if $V$ vanishes somewhere and $\operatorname{dim} M \geq 3$, the manifold is locally isometric to real hyperbolic space. When $V$ never vanishes and $\operatorname{dim} M \geq 3$, we obtain a global geometric description of all $\mathrm{Spin}^c$-Riemannian manifolds carrying such spinors, under the assumption that either the normalized Dirac current $\xi = \frac{V}{|V|}$ is complete or the leaves of $\mathcal{D} = \ker(\xi^\flat)$ are complete. Finally, we reinterpret the case of type~I generalized imaginary $\mathrm{Spin}^c$-Killing spinors in terms of parallel spinors for a suitable connection with vectorial torsion.