Optimization and Control
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
Friday 6 March
In this article, we propose an algorithm for the nonlinear conjugate gradient method to find a Pareto critical point of unconstrained multiobjective interval optimization problems. In this algorithm, we use the Wolfe line search procedure to find the step length. After defining the standard Wolfe conditions and the strong Wolfe conditions, we prove that there exists an interval of the step length that satisfies the standard Wolfe conditions and the strong Wolfe conditions. Further, to study the convergence analysis of our proposed algorithm, we derive the result related to the Zoutendijk condition. In the convergence analysis, first, we prove the global convergence property of our proposed algorithm for a general conjugate gradient algorithmic parameter. Further, we consider four variants of the conjugate gradient algorithmic parameter, such as Fletcher-Reeves, conjugate descent, Dai-Yuan, and modified Dai-Yuan. For each variant of the algorithmic parameter, we prove the global convergence results of our proposed algorithm. Finally, we test our algorithm on some test problems and make a performance profile.
Number Theory
On an elementary method for solving $Ax^4-By^2=1$
A new method for solving quartic equations due to Luo and Lin is investigated both computationally and theoretically. As a result, a completely straightforward elementary method is given for solving Bumby's equation $3X^4-2Y^2=1$, along with a conjecture, which if resolved, would enable a similar proof for a possibly infinite family of similar equations.
Read the full abstract
Algebraic Geometry
Compactifications of spaces of symmetric matrices and pointed Kontsevich spaces of isotropic Grassmannians
We study two closely related families of varieties arising from genus $0$ stable maps to the Lagrangian Grassmannian $\operatorname{LG}(n,2n)$. First, we construct the Kausz–type compactification $\mathcal {TL}_n$ of the space of symmetric matrices and give an explicit description of its birational geometry. Second, we realize $\mathcal {TL}_n$ as a general evaluation fiber in a Kontsevich space, and then exploit this modular interpretation to derive consequences for the birational geometry of the space of pointed conics $\overline{M}_{0,1}(\operatorname{LG}(n,2n),2)$. Analogous compactifications related to orthogonal Grassmannians are also presented.
Read the full abstract
Probability
Universality laws for random matrices via exchangeable counterparts
Recently, Brailovskaya & van Handel (GAFA, 2024) established a suite of nonasymptotic universality laws which demonstrate that the spectral statistics of an independent sum of random matrices mirror the spectral statistics of a Gaussian random matrix with the same first- and second-order moments. This paper develops a more elementary proof of their main results by means of a new implementation of the method of exchangeable counterparts.
Read the full abstract
Analysis of PDEs
On the defocusing stationary nonlinear Schrödinger equation on metric graphs
We study the defocusing nonlinear Schrödinger equation on noncompact metric graphs under general self-adjoint vertex conditions ensuring the existence of a negative eigenvalue of the Hamiltonian operator. First, we focus on the existence of energy ground states with prescribed mass. We show that existence and stability always hold for small masses and fail for large masses in the $L^2$-subcritical regime. For $\delta$-type vertex conditions, we provide more precise results: ground states exist for all masses in the $L^2$-critical and supercritical cases, while in the subcritical case, for one vertex graphs, there exists a sharp mass threshold such that ground states exist below it and do not exist above it. Moreover, we show that the ground state bifurcates from the vanishing solution at the bottom of the Hamiltonian spectrum. Finally, we present multiplicity results for stationary solutions, both in the fixed-frequency and fixed-mass settings.
Read the full abstract
Differential Geometry
Twists, Codazzi Tensors, and the $6$-sphere
Let $(M,g,J,\omega)$ be an almost Hermitian manifold. Given an automorphism $\psi\in \mathrm{Aut}(TM)$, the existing structure can be twisted to obtain a new almost Hermitian manifold $(M,g^\psi,J^\psi,\omega^\psi)$. In the current paper, we study these $\psi$-twisted almost Hermitian structures with particular emphasis on questions regarding the integrability of $J^\psi$ and the Riemannian geometry of $g^\psi$. By studying the latter, we identity a certain class of $\mathrm{Aut}(TM)$ with nice transformation properties. We call these automorphisms $g$-Codazzi maps because of their close relationship with Codazzi tensors. The aforementioned results are ultimately applied to the standard nearly Kähler structure on the $6$-sphere where we prove a nonintegrability result for the class of $g$-Codazzi maps.
Read the full abstract
Representation Theory
Unitary and Nonunitary Representations of the Heisenberg-Weyl Lie Algebra
We examine unitary and nonunitary representations of the Heisenberg-Weyl Lie algebra $\mathfrak{hw}_n$, with particular emphasis on tensor products of unitary representations and on indecomposable nonunitary representations. In the unitary setting, the irreducible representations with nontrivial central character are the Schrödinger representations, as classified by the Stone-von Neumann theorem. Although tensor products of these representations are considered in the literature, we give a detailed Lie-algebraic analysis and construct explicit unitary intertwining operators, including the case where the central characters sum to zero. In the nonunitary setting, we consider a natural realization of $\mathfrak{hw}_n$ as a subalgebra of the real symplectic Lie algebra $\mathfrak{sp}_{2n+2}(\mathbb R)$ and prove that every finite-dimensional complex irreducible representation of $\mathfrak{sp}_{2n+2}(\mathbb{R})$ remains indecomposable upon restriction to $\mathfrak{hw}_n$. This yields a large natural family of finite-dimensional, nonunitary indecomposable representations of $\mathfrak{hw}_n$.
Read the full abstract
Analysis of PDEs
Besov space approach to the Navier-Stokes equations with the Neumann boundary condition in bounded domains
Based on the analysis by Iwabuchi-Matsuyama-Taniguchi (2019), we first introduce our framework of Besov spaces $\dot B^s_{p, q}$ on the bounded domain $\Omega \subset {\mathbb R}^d$ with smooth boundary $\partial \Omega$ in terms of the Stokes operator $A=A_2$ with the Neumann boundary condition on $\partial\Omega$ in $L^2_{\sigma}(\Omega)$. Under some geometric assumption on $\Omega$, we establish $L^p-L^q$ type estimates of the semi-group $\{e^{-tA}\}_{t \ge 0}$ in $\dot B^s_{p, q}$ and prove a local well-posedness of the Navier-Stokes equations with the initial data in $\dot B^{-1+\frac dp }_{p, q}$ for $d < p < \infty$ and $1 \le q \le \infty$. Since $d < p$, we have $L^{d, \infty} \subset \dot B^{-1+\frac dp }_{p, \infty}$ so that our space for well-posedness is larger than any other previous one in bounded domains.
Read the full abstract
Combinatorics
Precoloring 3-extension on outerplanar graphs
The precoloring problem of a graph involves assigning colors to some vertices beforehand, and the objective is to determine whether it can be extended to a proper k-coloring of the entire graph. In 1958, Grotzsch proved that every triangle-free planar graph can be properly colored by three colors. One of the further generalizations of it is the recent result by Hoang La et al. in (Discrete Mathematics, 345(6) (2022), 112849 ). They proved that any two non-adjacent vertices and a face with a length at most four are precolored, the precolorings can be extended to a 3-coloring of the graph. In the paper, we consider precoloring extension of connected outerplanar graph with at most one or two triangles. Particularly, we show that precoloring of any two or three non-adjacent vertices can be extend to a 3-coloring of the whole graph.
Read the full abstract
Dynamical Systems · Cross-listed from eess.SY
On Koopman Resolvents and Frequency Response of Nonlinear Systems
This paper proposes a novel formulation of frequency response for nonlinear systems in the Koopman operator framework. This framework is a promising direction for the analysis and synthesis of systems with nonlinear dynamics based on (linear) Koopman operators. We show that the frequency response of a nonlinear plant is derived through the Laplace transform of the output of the plant, which is a generalization of the classical approach to LTI plants and is guided by the resolvent theory of Koopman operators. The response is a complex-valued function of the driving angular frequency, allowing one to draw the so-called Bode plots, which display the gain and phase characteristics. Sufficient conditions for the existence of the frequency response are presented for three classes of dynamics.
Read the full abstract
Thursday 5 March
Optimization and Control
A Projected Stochastic Gradient Method for Finite-Sum Problems with Linear Equality Constraints
A stochastic gradient method for finite-sum minimization subject to deterministic linear constraints is proposed and analyzed. The procedure presented adapts the projected gradient method on convex set to the use of both a stochastic gradient and a possibly inexact projection map. Under standard assumptions in the field of stochastic gradient methods, we provide theoretical results in agreement with the theory for unconstrained problems. Numerical results are presented to show the practical behavior of the procedure.
Read the full abstract
Optimization and Control
Compact Lifted Relaxations for Low-Rank Optimization
We develop tractable convex relaxations for rank-constrained quadratic optimization problems over $n \times m$ matrices, a setting for which tractable relaxations are typically only available when the objective or constraints admit spectral structure. We derive lifted semidefinite relaxations that do not require such spectral terms. Although a direct lifting introduces a large semidefinite constraint in dimension $n^2 + nm + 1$, we prove that many blocks of the moment matrix are redundant and derive an equivalent compact relaxation that only involves two semidefinite constraints of dimension $nm + 1$ and $n+m$, respectively. We also derive a new class of valid inequalities for low-rank problems, which we call projection cuts, that exploit the fact that rank constraints are inherited by linear images of a low-rank matrix, to strengthen our low-rank relaxations substantially. For matrix completion and reduced-rank regression problems, among others, we exploit additional structure to obtain even more compact formulations involving semidefinite matrices of dimension at most the sum of the two dimensions of the low-rank decision matrix (i.e., of size at most $n+m$). Overall, we obtain scalable semidefinite bounds for a broad class of low-rank quadratic problems.
Read the full abstract
Optimization and Control
Dynamic Menu-Based Pricing for Electric Vehicle Charging with Vehicle-to-Grid Integration
The number of electric vehicles is rapidly increasing worldwide. This growth brings significant environmental benefits but also introduces new challenges: uncoordinated charging can place stress on the grid, particularly during peak hours. Beyond these challenges lies the opportunity for electric vehicles to feed energy back to the grid (V2G), which helps balance supply and demand and supports renewable energy. However, current pricing schemes such as time-of-use tariffs provide little incentive for discharging. To study incentive design in a realistic context, we focus on a parking lot operator who manages multiple EV chargers. We propose a menu-based pricing mechanism in which each EV declares its energy requirement and parking duration; given the retail real-time electricity prices, the operator offers a menu of options that trade off the allowed level of discharging and the associated price. We formulate this interaction as a bilevel optimization problem and reformulate it into a single-level model. Results show that, relative to a no-V2G baseline, the proposed mechanism increases operator profit by 30% and reduces EV payments by 17%. Compared to widely used tariff baselines, it improves operator profit by 22-29 percent, lowers EV payments by 9-18 percent, and increases V2G contribution by 87-235 percent. Overall, the results show that the proposed dynamic menu-based pricing framework provides a practical, computationally efficient, and economically advantageous approach for real-time EV charging and V2G integration.
Read the full abstract
Optimization and Control
Using Laplace Transform To Optimize the Hallucination of Generation Models
To explore the feasibility of avoiding the confident error (or hallucination) of generation models (GMs), we formalise the system of GMs as a class of stochastic dynamical systems through the lens of control theory. Numerous factors can be attributed to the hallucination of the learning process of GMs, utilising knowledge of control theory allows us to analyse their system functions and system responses. Due to the high complexity of GMs when using various optimization methods, we cannot figure out their solution of Laplace transform, but from a macroscopic perspective, simulating the source response provides a virtual way to address the hallucination of GMs. We also find that the training progress is consistent with the corresponding system response, which offers us a useful way to develop a better optimization component. Finally, the hallucination problem of GMs is fundamentally optimized by using Laplace transform analysis.
Read the full abstract
Information Theory · Cross-listed from cs.CR
Quantum Key Distribution Secured Federated Learning for Channel Estimation and Radar Spectrum Sensing in 6G Networks
This paper presents a federated learning framework secured by quantum key distribution (QKD) for wireless channel estimation and radar spectrum sensing in the next generation networks (NextG or Beyond 6G). A BB84-style protocol abstraction and pairwise additive masking are utilized to train clients' local models (CNN for channel estimation, U-Net for radar segmentation) and upload only masked model updates. The server aggregates without observing plain parameters; an eavesdropper without QKD keys cannot recover individual updates. Experiments show that secure FL achieves NMSE of 0.216 for channel estimation and 92.1% accuracy with 0.72 mIoU for radar sensing. When an eavesdropper is present, QBER rises to $\sim$25% and all rounds abort as intended; reconstruction error remains below $10^{-5}$, confirming correct aggregation.
Read the full abstract
Classical Analysis and ODEs
Automorphisms of Stokes multipliers in higher-order WKBJ theory
We consider the Stokes phenomenon and higher-order Stokes phenomenon (HOSP) of formal asymptotic transseries arising in the WKBJ analysis of linear differential equations and integral problems. We introduce a framework of automorphisms that act on the Stokes constants of the divergent expansion, explained via late-late-term expansions and parametric Alien calculus, to capture this phenomenon. Our method is applied to a paradigmatic example: we obtain the full Stokes line structure and automorphisms for the Swallowtail problem from catastrophe theory, which contains four WKBJ components. We demonstrate that, in a system with four or more WKBJ components, the automorphism associated with the HOSP can itself change value across another higher-order Stokes line, which occurs when different higher-order Stokes lines intersect. We then argue that no additional special behaviour emerges for transseries with five or more WKBJ components.
Read the full abstract
Mathematical Physics · Cross-listed from nlin.PS
Swinging Waves in the Ablowitz-Ladik Equation
We construct a novel family of exact cnoidal wave and soliton solutions of the focusing and defocusing Ablowitz-Ladik equations. Unlike cnoidal waves that were obtained by earlier authors, the phase variable of the new solutions exhibits a nonlinear dependence on time and site number; the wave “swings". Our approach hinges on the existence of a two-point map governing the absolute value of the complex field; this map gives rise to standing waves centred arbitrarily relative to the lattice sites. Having derived stationary solutions, we use these as a basis for constructing waves with nonzero velocity. The localised members of the new family comprise dark solitons with the nontrivial asymptotic behaviour. We identify periodic and quasiperiodic patterns and establish an explicit quantisation rule for the velocity of the wave circulating in a closed loop of $N$ sites.
Read the full abstract
Logic
Corrigendum & Addendum to "Categoricity-like Properties in the First Order Realm"
This note complements our paper "Categoricity-like properties in the first order realm" (Journal for the Philosophy of Mathematics, 2024).
Mathematical Physics
Self-adjoint realizations of 2d-dimensional canonical systems and applications
This paper studies linear relations and their self-adjoint realizations arising from 2d-dimensional canonical systems, with a focus on how the symplectic structure interacts with boundary conditions. Understanding this interplay allows us to define self-adjoint realizations, which are crucial for analyzing the spectral properties of these systems. We prove that for each pair of Lagrangian boundary matrices {\Theta} and B satisfying appropriate orthonormality conditions, the restricted relation T{\Theta},B is self-adjoint. Our approach relies on the symplectic geometry of boundary spaces and the isotropic structure of Lagrangian subspaces. We also discuss extensions to semi-infinite intervals using asymptotic boundary conditions. In the second part of the paper, we show how this framework applies to spectral problems from partial differential equations, including the stability of traveling waves and the linearization of the focusing nonlinear Schrodinger equation around soliton profiles. In particular, the self-adjoint structure with respect to the H-weighted inner product ensures the spectral properties needed for stability analysis using Evans function and transfer matrix methods. Applications to integrable systems, such as the stability of NLS bright solitons, are also presented.
Read the full abstract
Dynamical Systems
Constructal Evolution as a Nonsmooth Dynamical System: Stability and Selection of Flow Architectures
Constructal Law states that a finite-size flow system that persists in time evolves its configuration so as to provide progressively easier access to the currents that flow through it. Classical Constructal theory derives hierarchical flow architectures from static resistance minimization under finite-size constraints, but many transport systems operate under irreversible limits that induce regime switching and discontinuous adjustment laws. We formulate Constructal evolution as an autonomous nonsmooth dynamical system. The architectural configuration is modeled as the state of a Filippov differential inclusion defined on a compact forward-invariant admissible set. Irreversible transport constraints generate switching manifolds across which the adjustment field is discontinuous. A resistance dissipation inequality encodes the Constructal principle of progressively improving access as a nonsmooth Lyapunov condition, while a uniform contraction assumption provides spectral bounds on the generalized Jacobians of the regime-dependent dynamics. Under these conditions we prove that the resulting inclusion admits a unique equilibrium architecture and that every admissible trajectory converges to it exponentially. Finite size, irreversibility, and resistance dissipation therefore imply existence, uniqueness, and global stability of persistent flow configurations without invoking static optimization. As an application, the classical area–to–point transport hierarchy of Bejan et. al. is embedded in the dynamical framework. The optimal assembly ratios appear as switching manifolds, while the classical scaling relations arise as sliding invariant sets of the Filippov inclusion. Their intersection defines the uniquely selected globally attracting architecture.
Read the full abstract
General Mathematics
Green-Function and Information-Geometric Correspondences Between Inverse Eigenvalue Loci of Generalized Lucas Sequences and the Mandelbrot Set
We investigate geometric, potential-theoretic, and information-theoretic correspondences between the inverse eigenvalue loci of companion matrices associated with generalized Lucas sequences and the boundary of the Mandelbrot set. Through systematic numerical experiments, we show that these algebraic spectral loci exhibit a striking low-distortion geometric correspondence with the Mandelbrot boundary at macroscopic scales, together with a coherent organization within its external potential field, characterized by concentration along narrow equipotential annuli of the Mandelbrot Green function. This correspondence is quantified using a suite of complementary diagnostics, including optimal transport matching, Procrustes alignment, local distortion measures, fractal and spectral statistics, Green-function-based potential comparisons, and convex simplex update analyses. Taken together, these results indicate that the observed similarity extends beyond visual resemblance, reflecting shared structural organization across geometric, harmonic, and statistical levels. While the present work is entirely numerical in nature, it establishes a robust multi-scale framework for comparing algebraic spectral constructions with nonlinear dynamical fractals, and it highlights several avenues for future analytical investigation.
Read the full abstract
General Mathematics
A Recursion Backbone for Circular and Elliptic Clausen Hierarchies
We introduce an elliptic extension of Clausen-type functions based on a unified recursive framework. Starting from the polylogarithmic master function, we construct a pair of circular functions whose real and imaginary parts correspond to the classical Clausen-type structures. Replacing the trigonometric seed with a Jacobi theta function yields an elliptic deformation that preserves the same recursive backbone. The circular limit recovers the original functions, establishing a structural correspondence between the circular and elliptic settings. Furthermore, we introduce a generating deformation that organizes the recursion into a single analytic object. This viewpoint suggests a unified framework for Clausen-type functions and their elliptic analogues.
Read the full abstract
Optimization and Control · Cross-listed from cs.DC
A Lock-Free Work-Stealing Algorithm for Bulk Operations
Work-stealing is a widely used technique for balancing irregular parallel workloads, and most modern runtime systems adopt lock-free work-stealing deques to reduce contention and improve scalability. However, existing algorithms are designed for general-purpose parallel runtimes and often incur overheads that are unnecessary in specialized settings. In this paper, we present a new lock-free work-stealing queue tailored for a master-worker framework used in the parallelization of a mixed-integer programming optimization solver based on decision diagrams. Our design supports native bulk operations, grows without bounds, and assumes at most one owner and one concurrent stealer, thereby eliminating the need for heavy synchronization. We provide an informal sketch that our queue is linearizable and lock-free under this restricted concurrency model. Benchmarks demonstrate that our implementation achieves constant-latency push performance, remaining stable even as batch size increases, in contrast to existing queues from C++ Taskflow whose latencies grow sharply with batch size. Pop operations perform comparably across all implementations, while our steal operation maintains nearly flat latency across different steal proportions. We also explore an optimized steal variant that reduces latency by up to 3x in practice. Finally, a pseudo workload based on large-graph exploration confirms that all implementations scale linearly. However, we argue that solver workloads with irregular node processing times would further amplify the advantages of our algorithm.
Read the full abstract
Dynamical Systems
Diophantine "Tears of the Heart"
Recent studies of topologically generic unfoldings of vector fields featuring a "tears of the heart" polycycle with one internal and one external winding separatrix have shown that, in a special one-parameter subfamily where the "heart" is preserved and the "tear" loop if broken, at least four invariants of weak topological classification appear. In this paper, we demonstrate that the metrical perspective yields a different result: for Lebesgue almost all values of the coefficients related to the original vector field, the special one-parameter family generates only two such invariants.
Read the full abstract
Group Theory
$E\mathcal{Z}$-boundaries, splittings over finite subgroups, and dense amalgams
The dense amalgam is an operation (introduced in arXiv:1410.4989) which to any finite collection of metrizable compacta associates canonically some new highly disconnected compact metrisable space in which embedded copies of the initial spaces are appropriately uniformly and disjointly distributed. We show that in the very general framework of $E\mathcal{Z}$-boundaries (unifying many frameworks such as Gromov boundaries, CAT(0)-boundaries, systolic boundaries, etc.), any boundary $Z$ of an infinitely ended group $\Gamma$ equipped with an appropriate splitting along finite subgroups has a form of the dense amalgam of the limit sets in $Z$ of the factor subgroups of this splitting.
Read the full abstract
Mathematical Physics
On solutions of the Euler equation for incoherent fluid on a rotating sphere
The motion of compressible, inviscid fluid under the constant pressure on a rotating sphere is studied. The hodograph equations for the corresponding Euler equation are presented. They provide us with the class of solutions of the Euler equation parameterized by two arbitrary functions of two variables. Several particular explicit solutions are given. The blow-up curves, on which the derivatives of velocitiy blows up, are described. The limiting cases of slowly and rapidly rotating sphere are considered. The equation describing the deformations of elliptic functions modulus is presented.
Read the full abstract
Combinatorics
On the Generalized Honeymoon Oberwolfach Problem
The generalized Honeymoon Oberwolfach Problem (HOP) asks whether it is possible to seat $2n$ participants consisting of $n$ newlywed couples at a conference with $s$ tables of size $2$ and $t$ ”round” tables of sizes $2m_1, 2m_2, \ldots, 2m_t$, where $n = s + \sum_{i=1}^{t} m_i $ with all $m_i \geq 2$, over several nights so that each participant sits next to their spouse every time and next to each other participant exactly once. We denote this problem by $\mathrm{HOP}(2^{\langle s \rangle}, 2m_1, \ldots, 2m_t)$. This paper is the first of two papers investigating the generalized HOP. While the second paper will deal with the generalized HOP with a single round table (i.e. table of size at least $4$), the present work develops solutions for the generalized HOP with multiple round tables. In particular, we present solutions to certain cases with two round tables, showing that a solution to $\mathrm{HOP}(2^{\langle s \rangle}, 2m_1, 2m_2)$ exists when $n \equiv 1 \pmod{(2m_1 + 2m_2)}$ or $n \equiv m_1 + m_2 \pmod{(2m_1 + 2m_2)}$. We also develop solutions for cases with small round tables, showing that $\mathrm{HOP}(2^{\langle s \rangle}, 2m_1, \dots, 2m_t)$ has a solution whenever $m = m_1 + \dots + m_t \leq 10$, $n = s + m$ is odd, and $n(n - 1) \equiv 0 \pmod{2m}$.
Read the full abstract
Functional Analysis
Generalized b-weakly compact operators and their factorization through KR-spaces
We investigate more closely the class of generalized b-weakly compact operators on locally convex-solid Riesz spaces and we provide new sequential and operator characterizations in relation with the subject. We introduce explicitly the so-called KR-spaces in order to study the factorization problem for generalized b-weakly compact operators by analogy with the well-known factorization of b-weakly compact operators through KB-spaces.
Read the full abstract
Numerical Analysis · Cross-listed from cs.CL
Structured Multidimensional Representation Learning for Large Language Models
Transformer architectures achieve state-of-the-art performance across a wide range of pattern recognition and natural language processing tasks, but their scaling is accompanied by substantial parameter growth and redundancy in the embedding dimension. In this work, we introduce a structured spectral factorization of the embedding space based on the L-product for third-order tensors. By reshaping token representations into spectral tensor slices and performing attention and feed-forward operations in the transform domain, we obtain a Tensor Transformer architecture that decomposes the encoder into p independent spectral sub-transformers while preserving standard Transformer semantics. We prove that the proposed L-Transformer is spectrally equivalent to p parallel Transformers operating on reduceddimensional embeddings, which yields approximately 1/p reduction (up to lower-order terms such as biases and normalization parameters) in encoder parameters under fixed total embedding size. When instantiated with a real-valued Discrete Cosine Transform (DCT), the method remains fully differentiable and compatible with existing training pipelines. Beyond compression, the spectral decomposition introduces an inductive bias over embedding frequencies, enabling slice-dependent frequency scaling that improves generalization. Experiments on IMDB and AG~News show that the proposed model can substantially reduce encoder parameters (up to 75% for p=4) while maintaining competitive accuracy. On IMDB, the tensorized encoder matches or improves upon the standard baseline under compression, whereas on AG~News at moderate width we observe a small accuracy decrease in exchange for a 4 times encoder reduction; at BERT-base width (d=768), performance returns to parity.
Read the full abstract
Numerical Analysis
FlexTrace: Exchangeable Randomized Trace Estimation for Matrix Functions
We consider the task of estimating the trace of a matrix function, ${\rm tr}(f({\bf A}))$, of a large symmetric positive semi-definite matrix ${\bf A}$. This problem arises in multiple applications, including kernel methods and inverse problems. A key challenge across existing trace estimation methods is the need for matrix-vector products (matvecs) with $f({\bf A})$, which can be very expensive. In this article, we introduce a novel trace estimator, FlexTrace, an exchangeable, single-pass method that estimates ${\rm tr}(f({\bf A}))$ solely using matvecs with ${\bf A}$. We consider the case where $f$ is an operator monotone matrix function with $f(0)=0$, which includes functions such as $\log(1+x)$ and $x^{1/2}$, and derive probabilistic bounds showcasing the theoretical advantages of FlexTrace. Numerical experiments across synthetic examples and application domains demonstrate that FlexTrace provides substantially more accurate estimates of the trace of $f({\bf A})$ compared to existing methods.