Category Theory · Cross-listed from q-fin.MF
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
Saturday 2 May
We introduce a transport cohomological framework for categorical filtrations. Given a contravariant filtration $F:\mathcal T^{op}\to\mathbf{Prob}$ on a small category \(\mathcal T\), conditional expectation induces transport operators between local probabilistic states. Using the simplicial structure of the nerve \(N_\bullet(\mathcal T)\), we construct simplex-local cochain complexes associated with parametrized simplices and study their transport cohomology. The resulting framework naturally produces loop effects and holonomy structures. In particular, transport around closed simplicial histories may generate nontrivial probabilistic distortions, even when the initial and terminal objects coincide. The associated holonomy operators encode global transport effects between probabilistic states and detect obstructions generated by loop transport. This leads to the notion of homological arbitrage, understood as a global transport phenomenon emerging from probabilistic distortion along loops. From this viewpoint, the essential source of loop effects is the probabilistic distortion generated by transport around closed simplicial histories. The present framework is structurally analogous to parallel transport and holonomy in differential geometry, providing a geometric viewpoint on categorical filtrations and probabilistic transport structures.
Functional Analysis
The proximal point method and its two variants for monotone vector fields in Hadamard spaces
We prove existence and convergence of sequences generated by the proximal point method and its two variants for monotone vector fields in Hadamard spaces. Before obtaining our results, we investigate some fundamental properties of tangent spaces, resolvents, and monotone vector fields in such spaces.
Read the full abstract
Algebraic Geometry
Weierstrass semigroups and the order bound
The aim of this survey is to provide the reader with an essential and accessible introduction to the theory of Weierstrass semigroups, in the context of the theory developed by K.-O. Stöhr and J.F. Voloch. Furthermore, we discuss an application of Stöhr-Voloch theory in coding theory, namely the Feng-Rao bound (also known as the order bound) for the dual minimum distance of one-point algebraic geometry codes from a curve, which relies on the knowledge of certain Weierstrass semigroups of the curve.
Read the full abstract
Optimization and Control
Unified Lyapunov Method for ISS of PDEs: A Tutorial on Constructing Generalized Lyapunov Functionals for Parabolic and Hyperbolic Equations
This tutorial provides an overview of the generalized Lyapunov method (GLM) for analyzing input-to-state stability (ISS) of partial differential equations (PDEs). We begin by revisiting the classical Lyapunov method and the standard ISS-Lyapunov theorem, highlighting their limitations when applied to systems with complex boundary disturbances. In contrast, the GLM, based on the concept of generalized Lyapunov functionals (GLFs) that explicitly depend on the external input, offers greater flexibility and efficiency, particularly for PDEs with Dirichlet-type disturbances. The main objective of this tutorial is to demonstrate how to systematically construct GLFs to establish ISS estimates in $L^q$ spaces with any $q\in[2,\infty]$ for different PDEs. Specifically, we consider three representative classes of PDEs: (i) an $N$-dimensional nonlinear parabolic equation with mixed nonlinear boundary disturbances, (ii) a first order nonlinear hyperbolic equation with boundary disturbances, and (iii) a second order linear hyperbolic equation, i.e., a wave equation, with boundary damping and disturbances. For each case, we provide step-by-step constructions of appropriate GLFs and derive explicit ISS estimates, illustrating the general applicability of the GLM. Finally, we discuss open challenges and future directions, including the systematic construction of GLFs for broader classes of PDEs and their applications in controller design.
Read the full abstract
Statistics Theory · Cross-listed from stat.ML
Mean Testing under Truncation beyond Gaussian
We characterize the fundamental limits of high-dimensional mean testing under arbitrary truncation, where samples are drawn from the conditional distribution $P(\cdot \mid S)$ for an unknown truncation set $S$ that may hide up to an $\varepsilon$-fraction of the probability mass. For distributions with $p$-th directional moments of magnitude at most $\nu_{P,p}$, truncation induces a bias of order $O(\nu_{P,p}\varepsilon^{1-1/p})$. This bias creates a sharp information-theoretic detectability floor: when the signal $\alpha$ falls below this threshold, the null and alternative hypotheses are indistinguishable even with infinite data. Above this floor, we prove that a simple second-order test achieving near-optimal sample complexity $n = O\!\left(\frac{\|\Sigma_P\|}{(\alpha-4\nu_{P,p}\varepsilon^{1-1/p})^2}\sqrt{d}\right)$. We further identify a structural escape from this finite-moment bias barrier. Under a directional median regularity assumption, truncation bias improves to linear order $O(\varepsilon)$. This reveals an intermediate regime in which estimation requires $\Theta(d)$ samples for uniform recovery, while testing recovers the classical $\Theta(\sqrt d)$ rate once truncation bias is eliminated. Together, our results provide a unified framework for mean testing under truncation, connecting finite-moment, sub-Gaussian, and median-regular structural regimes.
Read the full abstract
Analysis of PDEs
Convexity inequalities for eigenvalues and log-concavity of eigenfunctions
We give simple new proofs of two well-known results for the Schrödinger operator: first, the Brunn–Minkowski inequality for Dirichlet eigenvalues and, second, the log-concavity of the first Dirichlet eigenfunction. Our proof of the first applies to a class of domains including $C^{1,1}$ connected domains and convex potentials. In the special case of convex domains, the second result is a simple corollary.
Read the full abstract
Combinatorics
Toric Schubert Varieties in Partial Flag Varieties
In this article, we investigate the toric Schubert varieties in partial flag varieties $G/P$ for a connected semisimple algebraic group $G$. Using Deodhar's decomposition of Richardson varieties and the work of Pasquier, we give an explicit description of the fan of a toric Schubert variety, leading to a combinatorial model for its cones. As an application, we obtain necessary and sufficient conditions for smoothness of toric Schubert varieties in terms of the Cartan integers associated to a reduced expression. Furthermore, we prove that for a Coxeter-type element $w \in W^P$, the interval $[e,w]_{W^P}$ is a supersolvable join-distributive lattice. Finally, we apply these results to the study of spherical and horospherical Schubert varieties, providing a combinatorial method for checking the smoothness via the associated toric Schubert varieties.
Read the full abstract
Mathematical Physics · Cross-listed from hep-th
Exact WKB and Quantum Periods for Extremal Black Hole Quasinormal Modes
We apply exact WKB analysis to the spectral problem arising in black hole perturbation theory. The boundary conditions for quasinormal modes lead to exact quantization conditions for the complex frequencies. To solve these conditions, one needs to evaluate the so-called quantum periods, or Voros symbols. For scalar perturbations of extremal Reissner–Nordström and Kerr black holes, we compute these quantities up to very high orders in the WKB expansion and perform Borel–Padé resummation. The resulting resummed quantization conditions successfully reproduce the correct quasinormal mode frequencies with high precision.
Read the full abstract
Numerical Analysis
Spectral interpolation in semi-implicit semi-Lagrangian methods for shallow water equations on the sphere
Semi-implicit semi-Lagrangian (SISL) methods are commonly used for the shallow water equations (SWE) because they allow for larger time steps than those permitted by the Courant-Friedrichs-Lewy (CFL) stability condition in Eulerian schemes. In these methods, the semi-Lagrangian treatment of advection is typically performed using lower-order interpolation, such as tensor-product Lagrange interpolation with cubic or quintic polynomials. However, operational SISL schemes routinely employ spectrally accurate spatial discretizations, such as spherical harmonics or the double Fourier sphere (DFS) method, for computing horizontal derivatives of the prognostic variables. This creates a mismatch in numerical accuracy, making the use of low-order interpolation less clearly justified. In this work, we present the first numerical investigation of spectrally accurate interpolation in SISL schemes for the SWE. Our approach builds upon the recently developed DFS-based SWE model, incorporating a spectral interpolation scheme that is accelerated using the nonuniform fast Fourier transform (NUFFT) to maintain the same overall computational complexity as the original model. Using several standard SWE test cases, we evaluate the accuracy, conservation, and numerical diffusion of the new model, particularly over long integration times. Compared to an equivalent SISL model with low-order interpolation, the new model achieves higher accuracy, improved mass and energy conservation, and reduced numerical diffusion, demonstrating the potential benefits of incorporating spectrally accurate interpolation into SISL schemes.
Read the full abstract
Numerical Analysis
Alikhanov-XfPINNs: Adaptive Physics-Informed Learning for Nonlinear Fractional PDEs on Nonuniform Meshes
To address the initial singularity inherent in solutions to fractional partial differential equations (fPDEs), we propose an accelerated Alikhanov discretization formulation implemented on nonuniform time grids. Based on the physics-informed neural networks (PINNs) framework, we introduce an Alikhanov-extended fractional PINNs (XfPINNs) architecture that combines high-order temporal discretization and deep learning. The nonlocal memory term in fPDEs leads to high computational cost, while the weak singularity near $t\to 0^+$ can deteriorate accuracy on uniform meshes. To separate temporal discretization effects from optimization and sampling errors, we further develop an auxiliary time-marching configuration that enables auditable temporal-convergence studies under controlled training tolerances. This architecture can solve general nonlinear fPDEs. The XfPINNs approach is designed for forward and inverse problems, allowing for data-driven solution reconstruction and parameter estimation. First, the neural network approximates the solution of nonlinear fPDEs; then, an adaptive activation function accelerates convergence and enhances training efficiency. The optimization framework embeds a variational loss function constructed from the Alikhanov scheme, where the initial and boundary conditions are imposed using a combination of hard and soft constraints. Numerical experiments, including cases with known and unknown exact solutions which demonstrate the robustness, computational efficiency, and significant CPU time savings of the Alikhanov-XfPINNs method.
Read the full abstract
Dynamical Systems
Time-periodic carrying simplex for a competitive system of Carathéodory ODEs
We establish the existence of a time-periodic carrying simplex for competitive systems of Kolmogorov-type ordinary differential equations of the form $\dot{x}_i = x_i g_i(t,x)$, $i=1,\dots,d$, on $\mathbb{R}^d_+$, where the vector field $g$ is $T$-periodic in time and satisfies only Carathéodory conditions rather than the classical $C^1$ regularity assumed in earlier work. We show that the time-$T$ Poincaré map $P$ generated by the flow is retrotone, and apply recent results on the carrying simplex for retrotone maps to obtain for $P$ a carrying simplex $\Sigma$ that is an unordered, Lipschitz, invariant manifold homeomorphic to the standard probability simplex, and that attracts all nonzero trajectories of the discrete dynamical system generated by $P$. We then extend this construction to the full nonautonomous flow by introducing an auxiliary autonomous system on the extended phase space $\mathcal{T} \times \mathbb{R}^d_+$, where $\mathcal{T}$ is the circle of length $T$, generated via a skew-product formulation. In the context of this extended phase space we then construct a compact global attractor and identify, as its relative boundary, a time-periodic family of carrying simplices $\{\Sigma_s\}_{s \in \mathcal{T}}$ that depend continuously on the phase $s$, each homeomorphic to the standard simplex and each attracting, uniformly in $s$, all nonzero orbits of the corresponding time-shifted flow. These results provide a rigorous foundation for the qualitative analysis of seasonally forced competitive population models in which growth rates need only be measurable and locally Lipschitz in time, broadening the applicability of carrying simplex theory to more realistic ecological settings.
Read the full abstract
Combinatorics
The deviation from right angles in $k$-subsets of points in the plane
A problem originating with Erdős and Silverman in the 1970s asks for the minimum integer $r(k)$ such that any set of $n \ge r(k)$ points in the plane has some $k$-subset with no right angles. The case $k=4$ has an interesting gap between the known bounds, namely $8 \le r(4) \le 10$. Here, we consider a relaxation that quantifies the deviation from right angles. Specifically, we study $\Gamma_k(n)$, the supremum of angles $\gamma$ such that every $n$-set of points in $\mathbb{R}^2$ has a $k$-subset with all angles outside of the interval $90^\circ \pm \gamma$. We show that $4^\circ \le \Gamma_4(10) \le 9.292^\circ$. For large $n$, the quantity $\Gamma_3(n)$ is closely related to a classical minimax angle problem pioneered by Blumenthal, Erdős and Szekeres. We give bounds on $\Gamma_k(n)$ for a general $k$ and large $n$.
Read the full abstract
Geometric Topology
Characterization of non-self OU sequences of two-component link diagrams
A non-self OU sequence is a cyclic sequence of crossing information of non-self crossings that is obtained by traversing a knot component of an oriented link diagram. In this paper, we investigate what information can be derived from non-self OU sequences, and we completely characterize pairs of non-self OU sequences of diagrams of two-component links. We also characterize the pairs for specific prime links with crossing number up to five.
Read the full abstract
Numerical Analysis
A class of low-rank short recurrences for nonsymmetric linear matrix equations
We propose a new class of short matrix recurrences for the solution of nonsymmetric linear equations of the type $\mathbf{A}_1\mathbf{X}\mathbf{B}_1+\ldots+\mathbf{A}_p\mathbf{X}\mathbf{B}_p=CD^T$. Building on ideas underpinning the recently introduced subspace conjugate gradient algorithm, we derive low-rank short recurrences that generalize one-dimensional subspace projection methods to the nonsymmetric matrix equation setting. To limit memory consumption and maximize computational efficiency, rank truncation strategies and modern randomization procedures are incorporated into the proposed algorithms. Computational experiments on a benchmark problem as well as a challenging discretized mixed formulation of a diffusion equation with random inputs illustrate the potential of the proposed methodology.
Read the full abstract
Symplectic Geometry
Symplectic small covers in dimension four
We study symplectic structures on four-dimensional small covers. Our main result shows that every symplectic four-dimensional small cover is aspherical. We then classify symplectic small covers over products of two polygons, proving that symplecticity is equivalent to factor-compatibility. We also classify them up to diffeomorphism. Finally, we construct a symplectic four-dimensional small cover whose orbit polytope is not combinatorially equivalent to a product of two polygons.
Read the full abstract
Analysis of PDEs
Resolving an interface problem for the Dirac equation by using the unified transform method
We use the Unified Transform Method (UTM) for the vector case to resolve an interface problem for the Dirac equation on two semi-infinite domains and two finite domains in the massless and massive cases, respectively. The UTM for the vector case is a variation of the UTM for the scalar case. The solutions obtained for an interface problem on two semi-infinite domains and two finite domains, respectively, in the massive case are convergent explicit integral representations.
Read the full abstract
Combinatorics
Extremal Problems for the Family of $k$-Strongly Connected Digraphs
Let $\mathcal{D}$ be a family of digraphs. A digraph $D$ is \emph{$\mathcal{D}$-saturated} if it contains no member of $\mathcal{D}$ as a subdigraph, but for any arc $e$ in the complement of $D$, the digraph $D + e$ contains some member of $\mathcal{D}$ as a subdigraph. The saturation number $\mathrm{sat}(n,\mathcal{D})$ and the extremal number $\mathrm{ex}(n,\mathcal{D})$ are the minimum number and the maximum number of arcs among all $n$-vertex $\mathcal{D}$-saturated digraphs. For a positive integer $k$, let $\mathcal{D}_k$ denote the family of \emph{$k$-strongly connected digraphs}. In this paper, firstly, we prove that $$\mathrm{sat}(n,\mathcal{D}_k)=(k-1)(2n-k)+\binom{n-k+1}{2}.$$ Then for $n\geq 3(k-1)$, we prove that $$\mathrm{ex}(n,\mathcal{D}_k)\leq \binom{n-k+1}{2}+\frac{17}{6}(k-1)(n-k+1).$$ In addition, we conjecture that for sufficiently large $n$, $$\mathrm{ex}(n,\mathcal{D}_k)=\binom{n}{2}+\frac{3}{2}(k-\frac{4}{3})(n-k+1).$$
Read the full abstract
Combinatorics
The Normal Domination Partizan Game in Stars
The Domination game is an impartial game on graphs, introduced in 2010, and proved PSPACE-complete in the normal variant in 2026. In this game, Alice and Bob alternately select playable vertices, where a vertex is playable if it dominates at least one vertex not dominated by the vertices selected before in the game. The game ends when the selected vertices form a dominating set. In the normal variant, the player unable to move loses. In contrast to the impartial game, the partizan game has the vertices already colored with $A$, $B$, or $C$, in such a way that Alice (resp. Bob) can only select vertices colored with $A$ (resp. $B$) or $C$. The partizan game was proved PSPACE-hard in 2026. In this paper, we determine the winner of the Normal Domination Partizan game in graphs whose connected components are complete bipartite graphs or complete split graphs, including star forests, for any initial coloring of its vertices.
Read the full abstract
Analysis of PDEs
Hidden Boundary Trace Regularity and an Observability Estimate with Interior Remainder for Boundary-Degenerate Hyperbolic Equations
We study hidden boundary trace regularity for two-dimensional hyperbolic equations with boundary degeneracy governed by $\mcA\vp=-\Div(A\nabla \vp)$, where $A=\diag(1,r^\al)$ and $\al\in(0,1)$. We establish well-posedness in weighted Sobolev spaces and prove an $L^2$ trace estimate for the normal derivative on the nondegenerate side $r=1$. Using truncated geometries and Carleman weights adapted to the anisotropic degeneracy, we derive a large-time observability estimate with a lower-order interior remainder. We also identify a framework-level obstruction at the critical threshold $\al=1$: the weighted Dirichlet coercivity underlying the subcritical analysis loses uniformity and exhibits a logarithmic loss on truncated domains.
Read the full abstract
Functional Analysis
$L^r$- Schwartz spaces on split rank one semisimple symmetric spaces
We study the left $K$-invariant $L^r$-Schwartz space and its Fourier transform on split rank one semisimple symmetric spaces $G/H$ for $0<r\leq 2$. We explicitly determine the kernel of the Fourier transform and show that it is spanned by eigenfunctions associated with the discrete spectrum of the Laplace–Beltrami operator on $G/H$.
Read the full abstract
Optimization and Control
A Single-Loop Penalty-based Algorithm for Stochastic Minimax Optimization with Nonlinear Coupled Constraints
We study stochastic nonconvex-concave minimax optimization with nonlinear coupled constraints that are convex in the maximization variable. To address the nonsmoothness arising from such constraints, we develop a penalty-based smooth approximation that combines quadratic penalization of the coupled constraints with quadratic regularization of the inner maximization problem. Based on this approximation, we propose SPACO, a single-loop stochastic gradient algorithm that tracks the inner maximizer by one stochastic ascent step, updates the outer variable using an inexact stochastic descent direction, and adaptively updates the penalty and regularization parameters over the iterations. For the penalty-based smooth approximation, we establish convergence guarantees from both minimizer and stationarity perspectives. In particular, we introduce enhanced KKT conditions and show that stationary points of the smooth approximations can converge to points satisfying these conditions. An example illustrates that the enhanced KKT conditions can help exclude KKT points that are not local minimizers. For SPACO, we prove non-asymptotic complexity bounds for stationarity and feasibility, as well as asymptotic subsequential convergence to enhanced KKT points. Numerical experiments on synthetic examples, fairness-aware classification, and constrained generative adversarial network training demonstrate the effectiveness of the proposed method.
Read the full abstract
Statistics Theory
An Exact Pointwise Characterization for Total Variation Denoising in Quantile Regression
Total variation denoising (TVD) is a classical method for denoising and curve fitting, yet an explicit pointwise description of its fitted values has only recently been established in the mean regression setting by arXiv:2410.03041v4. This raises the question of whether a similar representation holds for quantile regression. We answer this question affirmatively by deriving an exact minmax/maxmin representation for the quantile TVD estimator, providing a complete pointwise characterization of its solution set. Given that the quantile TVD estimator is generally non-unique, the existence of such a representation is perhaps surprising. We show that the set of admissible fitted values at any location forms a compact interval, whose endpoints are characterized exactly by minmax/maxmin functionals of local order statistics over nested intervals. We next develop several structural properties of the quantile TVD solution set. First, the solution set is closed under coordinatewise maximum and minimum, guaranteeing the existence of extremal elements – upper and lower envelope solutions. Second, this reveals that quantile TVD is intrinsically non-crossing across quantile levels when a common tuning parameter is used. We prove this is driven by submodularity of the total variation penalty, and show that any penalized quantile regression estimator with a submodular penalty enjoys this property. From an estimation error perspective, our representation enables a refined pointwise analysis via a transparent local bias-variance decomposition, facilitating new pointwise risk bounds and near-optimal rates for locally Holder smooth functions. Our results hold under heavy-tailed noise (e.g., Cauchy) and substantially extend existing guarantees beyond locally constant signals. Altogether, these results advance the theory of quantile TV regression via exact pointwise min-max representations.
Read the full abstract
Numerical Analysis
Inverse Scattering for Dirac Equations Arising in Waveguide Arrays
We investigate inverse scattering problems for Dirac equations that arise as continuum models of waveguide arrays. We first establish the well-posedness of the forward models. For the associated inverse problems, we develop the inverse Born series and the reduced inverse Born series, providing analysis of convergence and rigorous error estimates. Numerical experiments are presented to validate the proposed algorithms and demonstrate their effectiveness.
Read the full abstract
Combinatorics
Tree-alpha and excluding finitely many graphs
We prove that a hereditary graph class $\mathcal{G}$ defined by finitely many excluded induced subgraphs has bounded tree-$\alpha$ if and only if it is "$(\mathrm{tw},\omega)$-bounded" (that is, for all $t\in \mathbb N$, the class of all $K_t$-free graphs in $\mathcal{G}$ has bounded treewidth). Equivalently, $\mathcal{G}$ has bounded tree-$\alpha$ if and only if it excludes a complete bipartite graph, a forest whose components each have at most three leaves, and the line graph of such a forest. This resolves two conjectures of Dallard, Krnc, Kwon, Milanič, Munaro, Štorgel, and Wiederrecht: the above, and a weaker one that for all $a,b\in \mathbb N$, every hereditary class that excludes $K_{a,a}$ and the $b$-vertex path has bounded tree-$\alpha$. The latter was already open even for $(a,b)\in \{(2,7),(3,5)\}$, and only recently proved for $(a,b)=(2,6)$.
Read the full abstract
Analysis of PDEs
A Projected Tug-of-War Game for the Regularized $p$-Laplacian
We give a tug-of-war interpretation of the regularized $p$-Laplacian $\operatorname{div}\big((1+|Dv|^2)^{p/2-1}Dv\big)=0$ in a bounded domain $\Omega \subset \mathbb{R}^n$, $p\ge 2$. The key is the linear lift $w(x,x_{n+1})=v(x)+x_{n+1}$, which identifies this equation with $\Delta_p w=0$ in $\mathbb{R}^n+1$. Projecting the standard $(n+1)$-dimensional $p$-harmonious scheme onto $\mathbb{R}^n$ yields a discrete dynamic programming principle for which we prove existence, uniqueness, and Borel measurability of solutions with strip boundary data, identify the unique fixed point with the value of the projected game, and establish convergence to the viscosity solution as $\varepsilon\to 0$.
Read the full abstract
Representation Theory
Cohomological Maschke's Theorem for Generalized Digroups
We study Maschke-type phenomena in the representation theory of generalized digroups. For a generalized digroup $D$, we construct an associative enveloping algebra $A_D$ and prove that $Rep(D)$ is equivalent to the category of left $A_D$-modules. Under a Maschke-type condition on the group component, we show that short exact sequences split on the $\rho$-side, while the obstruction to full splitting is described by cocycles and identified with $Ext^1_{Rep(D)}(Q,W)$. We also derive a spectral sequence with consequences for splitting and non-semisimplicity.
Read the full abstract
Information Theory
Improved Rate-versus-Distance Upper Bounds for LDPC Codes
LDPC codes play a vital role in coding theory and practical error correction. A central problem in this direction is to understand their rate–distance tradeoff. In this paper, we develop a local-growth argument for estimating ball sizes in the coset graphs of LDPC codes. Rather than estimating coset balls directly, we use a local-growth analysis to bound the coset-weight generating function of linear spaces spanned by low-weight vectors. This approach sharpens the previous ball-size estimates of Iceland and Samorodnitsky. Combined with a general method of Friedman and Tillich that relates balls in coset graphs to sizes of error-correcting codes, it further improves the upper bounds on the rate of LDPC codes for a significant range of relative distances.
Read the full abstract
Spectral Theory
Generic Irreducibility of Bloch Varieties for Periodic Graph Operators
We give a complete characterization of generic irreducibility for dispersion polynomials and Bloch varieties of periodic graph operators. More precisely, we prove that for a generic choice of edge weights and potentials, the dispersion polynomial/Bloch variety of a nontrivial periodic graph is irreducible if and only if the quotient graph is connected. Our proof uses a strong dichotomy for parameterized Laurent polynomials: reducibility either occurs for every parameter or fails on a nonempty Zariski-open set. After establishing this dichotomy, we reduce the problem to minimally connected periodic graphs.
Read the full abstract
Numerical Analysis
Sampling Pfaffian point processes and the symplectic Arnoldi method
We present an exact sampling algorithm for Pfaffian point processes based on a skew-symmetric analogue of the Cholesky factorization. This algorithm enables efficient sampling of a wide range of statistics arising in random matrix theory and combinatorics. For instance, we can sample eigenvalues of the orthogonal and symplectic ensembles ($\beta = 1,4$). In addition, we introduce a symplectic Arnoldi method for computing skew-orthogonal polynomials associated with a general weight function. This method can be used to efficiently construct the $2 \times 2$ matrix valued skew-symmetric kernels that arise in $\beta = 1,4$ polynomial ensembles. We illustrate our approach with several numerical examples and experiments, including the symmetric corner growth model, the finite-$N$ Gaussian (Hermite) orthogonal and symplectic ensembles, and the $\beta = 1,4$ Airy point processes and Tracy-Widom distributions.
Read the full abstract
Information Theory · Cross-listed from cs.LG
Near-Floor Geometry Is Generic: Leverage Dispersion in Trained Overcomplete Codes
An overcomplete code packs $F$ features into $d<F$ dimensions, so linear readout of one feature picks up cross-talk from the others, bounded below by the rank-trace floor $W(F,d)=(F-d)/(d(F-1))$. Proximity to this floor is read as evidence of an efficient arrangement. It is not: an i.i.d. code of the same shape attains it at every shape and load we measure (1.0000-1.0005 over 7 matched controls), and the reason is exact. For the gain-calibrated pseudoinverse the attainment ratio is $\frac{d}{F(F-d)}(\sum_i h_i^{-1}-F)$ with $\sum_i h_i=d$, so it equals one precisely when the leverage $h_i$ is equalised across features: distance from the floor IS leverage dispersion. On 40 released decoder matrices - 24 sparse-autoencoder decoders plus 16 MLP down-projections of a hybrid model - all sit above their matched control (1.0055-1.2018), and all reach guaranteed affine failure 4 to 40 sparsity levels earlier: where a random code of the same shape is still safe, a released dictionary already has a feature no affine rule recovers. Autoencoders trained by the same recipe on a randomly initialised copy of the same model return to the floor (1.0038-1.0065) in all 10 paired layers, so the excess reflects what the model learned, not the act of fitting an autoencoder. Under controlled conditions the loss decides whether an interface exists at all: $L^2$ reaches $R_{geom}=18.5635$ to 32.3131 with its frontier collapsed to $1$, against $1.0002$ for $L^4$, and that frontier follows a rate we registered before the widest run (10.79 predicted, 10.510 measured). Two limits are stated rather than implied: frontier statements are worst cases, not typical-case claims; and because the probe class contains the network's own decoder, our decoder comparison bounds supervised fitting rather than affine expressiveness.