Peer Revenue Returns subject to peer review

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

Mathematics

math Watch Mathematics (sign in to use)

Friday 13 March

Numerical Analysis · Cross-listed from cond-mat.dis-nn

A toy model of a protein prototype reveals nontrivial ultrametricity of the energy landscape

A model for studying the ultrametricity of the energy landscape in a disordered heteropolymer is presented. It is treated as a simplified model of a protein molecule in which amino acid residues are modeled as point masses. Pairwise interactions include universal repulsion, the Lennard-Jones potential, the Coulomb potential with screening, and the elastic potential for bonds between adjacent residues. An analogy with spin glass models is used, allowing the application of replica theory methods. Unlike the standard approach to disordered systems, averaging over disorder is not performed. The overlap between replicas is defined as the Pearson correlation coefficient between the vectors of average pairwise energies, which corresponds to a comparison of thermodynamic averages in the spirit of spin glass theory. The results of a computational experiment conducted using the developed algorithm on a graphics processing unit (GPU) are presented. The simulations were performed using a 128-residue-long sequence, with 50 independent disorder realizations and 50 replicas for each sequence at a temperature of T = 1.0. It was found that for 90.0% of the sequences, the distance matrix between replicas contains more than half of the ultrametric triangles, and nontrivial ultrametricity predominates in 97.8% of them, indicating a hierarchical organization of the energy landscape. A repeated computational experiment for selected sequences confirms the reliability of the observations: 95.5% of them again demonstrated ultrametricity, of which 97.7% showed a predominance of the nontrivial type of ultrametricity. The obtained results confirm Frauenfelder's hypothesis of protein ultrametricity and pave the way for a systematic study of ultrametric properties in more realistic protein models.
Read the full abstract

Representation Theory

A characterization of IE-closed subcategories via $\tau$-tilting theory

Enomoto and Sakai classified functorially finite IE-closed subcategories over hereditary algebras in terms of twin rigid modules. Their approach uses the hereditary assumption essentially and therefore does not extend directly to arbitrary finite-dimensional algebras. In this paper, we introduce canonical twin support $\tau$-tilting modules and prove that, for an arbitrary finite-dimensional algebra, they are in bijection with left-and-right finite IE-closed subcategories, namely those whose generated torsion and torsion-free classes are both functorially finite. We further give a characterization of canonicality via the torsion-pair decompositions associated with $\operatorname{Fac} M$ and $\operatorname{Sub} N$, which yields a canonicalization procedure whenever the associated IE-closed subcategory is left-and-right finite. We also introduce canonical Ext-pairs. If the algebra is hereditary or $\tau$-tilting finite, then functorially finite IE-closed subcategories are in bijection with isomorphism classes of canonical Ext-pairs, where the corresponding pair is given by the basic Ext-progenerator and the basic Ext-injective cogenerator. In the hereditary case, this recovers the twin rigid classification of Enomoto and Sakai.
Read the full abstract

Category Theory

Classes of Universal Epi- and Monomorphisms in Quasi-Abelian Categories

This paper provides a solution to the open problen formulated in Glotko and Kuzminov article, as well as examples of non-strict universal epimorphisms and monomorphisms.

Combinatorics

Extending Exact Integrality Gap Computations for the Metric TSP

The subtour relaxation of the traveling salesman problem (TSP) plays a central role in approximation algorithms and polyhedral studies of the TSP. A long-standing conjecture asserts that the integrality gap of the subtour relaxation for the metric TSP is exactly $4/3$. In this paper, we extend the exact verification of this conjecture to a larger number of vertices. Using the framework introduced by Benoit and Boyd in 2008, we confirm their results up to $n=10$. We further show that for $n=11$ and $n=12$, the published lists of extreme points of the subtour polytope are incomplete: one extreme point is missing for $n=11$ and twenty-two extreme points are missing for $n=12$. We extend the enumeration of the extreme points of the subtour polytope to instances with up to $n=16$ vertices in the general case. Restricted to half-integral extreme points, we extend the enumeration to $n=18$. Our results provide additional support for the $4/3$-Conjecture. Our lists of extreme points are available on the public bonndata repository (https://doi.org/10.60507/FK2/JK95PC).
Read the full abstract

Numerical Analysis

Augmented Lagrangian preconditioners for fictitious domain formulations of elliptic interface problems

We present a novel augmented Lagrangian (AL) preconditioner for the solution of linear systems arising from finite element discretizations of elliptic interface problems with jump coefficients. The method is based on the Fictitious Domain with Distributed Lagrange Multipliers formulation and it is designed to improve the convergence of the Flexible Generalized Minimal Residual (FGMRES) method in the presence of large coefficient jumps. To reduce the computational cost, we also introduce a cheaper block-triangular variant of the preconditioner. We prove eigenvalue clustering for the ideal AL preconditioner and study the limiting behavior of the spectrum for the modified variant in terms of parameters and the size of the jumps. Numerical experiments on different immersed geometries confirm mesh-independent iteration counts and robustness over large coefficient jumps, with substantial reductions in wall-clock time for the modified approach.
Read the full abstract

All authors Michele Benzi, Marco Feder, Luca Heltai, Federica Mugnaioni

Numerical Analysis

Discretization of the Burgers' equation as a port-Hamiltonian system

The numerical simulation of the inviscid Burgers' equation is often hindered by spurious oscillations near discontinuities. To mitigate this issue, a viscous term can be introduced, leading to the viscous Burgers' equation. In this work, port-Hamiltonian formulations for both the inviscid and the viscous Burgers' equations are proposed, enabling a representation that incorporates both convective and dissipative effects. Boundary control and observation are naturally handled within this framework. Applying a dedicated finite element method, a finite-dimensional port-Hamiltonian system is derived. The relationship between time step, spatial resolution, and viscosity required to achieve numerical stability is analyzed. Numerical experiments validate the effectiveness of the approach.
Read the full abstract

Algebraic Geometry

A Known M-Curve of Degree 8 in OSCAR

We show how the computer algebra system OSCAR can be used to retrace geometric constructions of plane algebraic curves and provide topologically correct and visually pleasing drawings. The approach is illustrated on Shustin's construction from 1987 of a particular M-curve of degree eight. We arrive at an explicit homogeneous trivariate polynomial with rational coefficients, and we provide a computational proof of all claimed properties of Shustin's original article.
Read the full abstract

Numerical Analysis

RUNNs: Ritz-Uzawa Neural Networks for Solving Variational Problems

Solving Partial Differential Equations (PDEs) using neural networks presents different challenges, including integration errors and spectral bias, often leading to poor approximations. In addition, standard neural network-based methods, such as Physics-Informed Neural Networks (PINNs), often lack stability when dealing with PDEs characterized by low-regularity solutions. To address these limitations, we introduce the Ritz–Uzawa Neural Networks (RUNNs) framework, an iterative methodology to solve strong, weak, and ultra-weak variational formulations. Rewriting the PDE as a sequence of Ritz-type minimization problems within a Uzawa loop provides an iterative framework that, in specific cases, reduces both bias and variance during training. We demonstrate that the strong formulation offers a passive variance reduction mechanism, whereas variance remains persistent in weak and ultra-weak regimes. Furthermore, we address the spectral bias of standard architectures through a data-driven frequency tuning strategy. By initializing a Sinusoidal Fourier Feature Mapping based on the Normalized Cumulative Power Spectral Density (NCPSD) of previous residuals or their proxies, the network dynamically adapts its bandwidth to capture high-frequency components and severe singularities. Numerical experiments demonstrate the robustness of RUNNs, accurately resolving highly oscillatory solutions and successfully recovering a discontinuous $L^2$ solution from a distributional $H^{-2}$ source – a scenario where standard energy-based methods fail.
Read the full abstract

All authors Pablo Herrera, Jamie M. Taylor, Carlos Uriarte, Ignacio Muga, David Pardo, Kristoffer G. van der Zee

Algebraic Topology

Power Operations in Morava E-Theory of Flat Ring Spectra

Let $E_n$ be Morava $E$-theory of height $n$. Let $R$ be a $p$-adically flat commutative ring spectrum. Then the Tate-valued Frobenius map endows $\pi_0 R$ with the structure of a $\delta$-ring. On the other hand, we may form the $K(n)$-completed tensor product $L_{K(n)}(R \otimes E_n)$, which is a $K(n)$-local $E_n$-algebra. Then $\pi_0(L_{K(n)}(R \otimes E_n)) = LT_n \widehat{\otimes} \pi_0 R$ admits the structure of an algebra over the monad $\mathbb{T}(n)$ defined by Rezk. The $\mathbb{T}(n)$-algebra structure encodes the power operations of $L_{K(n)}(R \otimes E_n)$. In this paper we describe the $\mathbb{T}(n)$-algebra structure on $\pi_0(L_{K(n)}(R \otimes E_n))$.
Read the full abstract

Numerical Analysis

Reduced rank extrapolation for multi-term Sylvester equations

We investigate the acceleration of stationary iterations for multi-term Sylvester equation by means of reduced rank extrapolation (RRE). Theoretical convergence results and implementations are provided for both small and large-scale problems. For the large-scale problems, an inexact non-stationary iteration is discussed, which makes use of low-rank matrix approximations. Numerical experiments illustrate the potential of the RRE acceleration which often leads to a substantial gain in convergence speed and therefore reducing the consumption of storage and computing time.
Read the full abstract

All authors Peter Benner, Pascal den Boef, Patrick Kürschner, Xiaobo Liu, Jens Saak

Analysis of PDEs

Existence and uniqueness of the global conservative solutions for the generalized Camassa-Holm equation with dual-power nonlinearities

In this paper, we investigate the global conservative solutions to the generalized Camassa-Holm equation with dual-power nonlinearities. By introducing a new set of variables, we transform the original equation into an equivalent semi-linear system, which allows us to establish the global existence of conservative solutions. Furthermore, for a given global conservative solution, we construct some auxiliary variables tailored to its specific structure and demonstrate that they satisfy a semi-linear system with a unique solution, thereby deriving the uniqueness of conservative solutions to the original equation.
Read the full abstract

Representation Theory

Derived equivalences and delooping levels

We construct a finite-dimensional algebra derived equivalent to the example of Kershaw–Rickard. For the Kershaw–Rickard example the delooping level and the sub-derived delooping level are both infinite, while for our algebra both invariants are $0$. Thus the finiteness of these invariants is not preserved under derived equivalences.
Read the full abstract

Numerical Analysis

Adaptive FEM with optimal convergence rate for non-self-adjoint eigenvalue problems

In this paper, we first discuss the optimal convergence of the adaptive finite element methods for non-self-adjoint eigenvalue problems. We present new theoretical error estimators and computable error estimators for multiple and clustered eigenvalues with the help of the error estimators of finite element solutions for the corresponding source problems, and prove the equivalence between these two estimators. We propose an adaptive algorithm for the eigenvalue cluster and demonstrate that it achieves the optimal convergence rate.We also provide numerical experiments to support our theoretical findings.
Read the full abstract

Numerical Analysis

Rapid Identification of Moving Contaminant Sources Through Physics-Based Modelling

In an act of sabotage or terrorism, hazardous material might be released deliberately into the atmosphere to threaten individuals, e.g., those operating critical infrastructure. Hazardous materials in such a scenario include toxic industrial chemicals (TICs), which are often invisible to the human eye, making it difficult to detect and respond to releases in a timely manner. This contribution considers the scenario of an airborne hazardous release requiring rapid and reliable assessment, with a chemical, biological, radiological, and nuclear (CBRN) sensor system providing scarce and local measurements. We present a novel algorithm that couples these data with an advection-diffusion model to detect, localize, and quantify a moving and time-varying contaminant source. Unlike many existing methods, the approach identifies sources with unknown occurrence time and trajectory by incorporating spatial sparsity as prior information. The feasibility of the approach is demonstrated in a two-dimensional computational domain. To further increase the technology readiness level, we additionally propose a calibration methodology for the required three-dimensional flow models based on wind tunnel experiments. Finally, a strategy for coupling the framework with real-time sensor data within a digital twin environment is outlined to enable predictive decision support in emergency scenarios.
Read the full abstract

All authors Marco Mattuschka, Jacopo Bonari, Max von Danwitz, Alexander Popp

Operator Algebras

Toward the classification of strongly self-absorbing $\mathrm{C}^*$-dynamical systems of compact groups

Strongly self-absorbing $\mathrm{C}^*$-algebras play a distinguished role in the classification of nuclear $\mathrm{C}^*$-algebras. Their dynamical analogues were introduced and extensively studied by Szabó. In this paper, we propose a conjecture regarding the equivariant $KK$-theory of strongly self-absorbing $\mathrm{C}^*$-dynamical systems of compact groups in the equivariant bootstrap category; an affirmative answer to this conjecture would lead to classification results. We settle this conjecture for all finite EPPO (every element has a prime-power order) groups. In the course of our proof, we establish a Künneth-type formula for the equivariant $K$-theory of $\mathrm{C}^*$-algebras equipped with finite cyclic group actions – more precisely, for the cyclotomic part of the equivariant $K$-groups introduced by Meyer and Nadareishvili – under a certain unique divisibility assumption.
Read the full abstract

Optimization and Control

Deep Predictor-Corrector Networks for Robust Parameter Estimation in Non-autonomous System with Discontinuous Inputs

Learning under non-smooth objectives remains a fundamental challenge in machine learning, as abrupt changes in conditioning variables can induce highly non-smooth loss landscapes and destabilize optimization. This difficulty is particularly pronounced in non-autonomous dynamical systems driven by discontinuous inputs, where widely used optimization methods, including recent neural smoothing approaches, exhibit unreliable convergence or strong hyperparameter sensitivity. To address this issue, we propose Deep Predictor-Corrector Networks (DePCoN), a multi-scale learning framework that stabilizes optimization by learning scale-consistent parameter update rules across a hierarchy of smoothed inputs. Rather than treating smoothing as a fixed preprocessing choice, DePCoN integrates smoothing into the learning dynamics itself through a learned predictor-corrector mechanism. Across biological and ecological benchmarks with discontinuous inputs, DePCoN consistently achieves more robust and faster convergence than existing methods while substantially reducing sensitivity to hyperparameter choices. Beyond dynamical systems, our approach provides a general learning principle for stabilizing optimization under non-smooth objectives.
Read the full abstract

All authors Gyeongwan Gu, Jinwoo Hyun, Hyeontae Jo, Jae Kyoung Kim

Numerical Analysis

A regularized method for quadratic optimization problems with finite-dimensional degeneracy

We propose and analyze a perturbative regularization method to approximate quadratic optimization problems with finite-dimensional degeneracy. The original problem is first approximated by a regularized problem depending on a small positive parameter, and then discretized using the finite element method. The resulting families of continuous and discrete functionals $\Gamma$-converge to the functional of the original problem and the corresponding minimizers converge as well. Our method generalizes the approach proposed in Kaleem et al. (2026) for numerically approximating pure Neumann problems, which represents the cornerstone of a sparsity-preserving, numerically efficient alternative to the methods developed in Bochev and Lehoucq (2005), Ivanov et al. (2019) and Roccia et al. (2024). References: A. Kaleem, C. Gebhardt, and I. Romero. On the pure traction problem of linear elasticity: a regularized formulation and its robust approximation. arXiv preprint arXiv:2602.04359, 2026. P. Bochev and R. Lehoucq. On the finite element solution of the pure Neumann problem. SIAM Review, 47:50-66, 2005. M. Ivanov, I. Kremer, and M. Urev. Solving the pure Neumann problem by a finite element method. Numerical Analysis and Applications, 12:359-371, 2019. B. Roccia, C. Alturria, F. Mazzone, and C. Gebhardt. On the homogeneous torsion problem for heterogeneous and orthotropic cross-sections: theoretical and numerical aspects. Applied Numerical Mathematics, 201:579-607, 2024.
Read the full abstract

Numerical Analysis

A Priori Adaptive Numerical Methods for Estimating Blow-up Times of Autonomous ODEs

We present effective a priori adaptive numerical methods for estimating the blow-up time for solutions of autonomous ODEs. The novelty of our approach is to base our adaptive steps on the sensitivity of an auxiliary hitting time. We provide results on the theoretical error rates, and show that there is a benefit in terms of computational effort in choosing our adaptive algorithm over alternative approaches. Numerical experiments support our theoretical results and show how the methods perform in practice.
Read the full abstract

Optimization and Control

Numerically stable variants of overrelaxation for operator Sinkhorn iteration

We consider accelerated versions of the operator Sinkhorn iteration (OSI) for solving scaling problems for completely positive maps. Based on the interpretation of OSI as alternating fixed point iteration, it has been recently proposed to achieve acceleration by means of nonlinear successive overrelaxation (SOR), e.g.~with respect to geodesics in Hilbert metric. The direct implementation of the proposed SOR algorithms, however, can be numerically unstable for ill-conditioned instances, limiting the achievable accuracy. Here we derive equivalent versions of OSI with SOR where, similar to the original OSI formulation, scalings are applied on the fly in order to take advantage of preconditioning effects. Numerical experiments confirm that this modification allows for numerically stable SOR-acceleration of OSI even in ill-conditioned cases.
Read the full abstract

All authors Henrik Eisenmann, Tasuku Soma, Xun Tang, André Uschmajew

Functional Analysis

Fine properties of Besov functions $B^r_{q,\infty}$ in metric spaces

Let $X$ be a metric space and $\mu$ an $s$-regular Ahlfors measure. Let $Y$ be a metric space. We prove that for Besov functions $u \in B^r_{q,\infty}(X,\mu;Y)$, every point is a general average Lebesgue point of $u$ outside a $\sigma$-finite set with respect to the Hausdorff measure $\mathcal{H}^{s - rq}$. The proof is based on density-type estimates involving Hausdorff measure. In addition, we prove that for functions $u$ in the fractional Sobolev space $W^{r,q}(X,\mu;Y)$, almost every point with respect to $\mathcal{H}^{s - rq}$ is an average Lebesgue point of $u$. Finally, if $Y$ is also complete, we prove that for $u \in B^r_{q,\infty}(X,\mu;Y)$, almost every point is a Lebesgue point outside a set of Hausdorff dimension at most $s - rq$.
Read the full abstract

Analysis of PDEs

Bifurcation of radial solutions for prescribed mean curvature equations

We prove global bifurcation results for prescribed mean curvature equations. These equations are defined on R3 and the radial solutions belonging in these branches are smooth and positive.

Functional Analysis

A Banach space with an unconditional basis which is not slicely countably determined

In this note, we study the geometry of the unit ball of the Banach space generated by the adequate family of all subsets of branches of the infinite binary tree, and answer several open questions related to slicely countably determined Banach spaces. Our main result is that the binary tree space is an example of a Banach space with an unconditional basis which fails to be slicely countably determined. In particular, it provides an example of a non slicely countably determined separable Banach space which contains no isomorphic copy of a space with the Daugavet property. We also exhibit some other geometric features of this space: we prove that its unit ball is dentable, that it has numerical index~1, and that the points of continuity of its unit ball form a weakly dense set. Finally, we show that the binary tree space contains a non-convex subset which is slicely countably determined, but does not admit a countable $\pi$-base for its relative weak topology, and that there is a 2-equivalent renorming of this space whose unit ball fails to be slicely countably determined.
Read the full abstract

Analysis of PDEs

Regularity of solution maps of the generalized surface quasi-geostrophic equations

We study regularity properties of the data-to-solution maps of the family of generalized surface quasi-geostrophic equations which includes both the 2D incompressible Euler and the standard surface quasi-geostrophic equations. We prove that the Lagrangian solution maps, interpreted as Riemannian exponential maps on the group of exact Sobolev class diffeomorphisms, are real analytic and, consequently, the Cauchy problems are locally well-posed in the sense of Hadamard. On the other hand, we also show that the corresponding Eulerian solution maps are nowhere locally uniformly continuous on bounded subsets in the Sobolev topology and fail to be continuous in the standard (large-) Hölder topologies. These results sharpen earlier theorems and further highlight the striking dichotomy between regularity properties of the solution maps in the Lagrangian and Eulerian formulations.
Read the full abstract

Analysis of PDEs

Topological Degree Methods for Age-Structured Epidemic Models

This paper is devoted to the study of an age-structured SIRS epidemic model, in which a population affected by a disease is divided into susceptible, infected, and removed individuals. We assume that the force of infection may be nonlinear and time-dependent. The model, originally introduced and studied by Iannelli and his co-authors, can be naturally formulated in an abstract setting and has traditionally been analyzed using fixed point techniques, most often the Banach contraction principle. Following the approaches of Inaba and Banasiak, our investigation is based on the semigroup theory, through which we study the existence of mild (integral) solutions. The main novelty of our work lies in the use of the topological degree for condensing maps instead of classical fixed-point arguments. We prove the existence of a unique, global, nonnegative solution to the model that satisfies the prescribed initial and nonlocal conditions and takes values in the space $L^1$ with respect to the age variable. Moreover, this solution depends continuously on the initial data.
Read the full abstract

Analysis of PDEs

Some minimum principles for a class of nonlinear elliptic problems in divergence form

In this paper we study a general class of nonlinear elliptic problems in divergence form. First, we prove that the solutions to these problems satisfy a convexity property when the given domain is strictly convex. Then, making use of this convexity property, we develop some minimum principles for an appropriate $P$-function, in the sense of L.~E.~Payne. Finally, this new minimum principle is applied to find a priori estimates for the solutions, in terms of the mean curvature of the boundary of the underlying domain.
Read the full abstract

Functional Analysis

Complementability of separable spaces $\mathcal{C}(K)$ in Banach spaces

For a metric compact space $L$ and a Banach space $E$, we provide a characterization of the complementability of the Banach space $\mathcal{C}(L)$ of continuous functions on $L$ inside $E$ in terms of the existence of a certain tree in the product $E \times E^*$, based on new descriptions of the Banach spaces $\mathcal{C}([1, \omega^{\alpha}])$ for countable ordinal numbers $\alpha$ and $\mathcal{C}(2^{\omega})$. Applying this general result in the case where $E=\mathcal{C}(K)$ for some compact space $K$, we further obtain a characterization of the existence of a positively $1$-complemented positively isometric copy of $\mathcal{C}(L)$ inside $\mathcal{C}(K)$ in terms of the topology of $K$ and the space of probability Radon measures on $K$. In the process, we also prove a variant of the classical Holsztyński theorem for isometric embeddings onto complemented subspaces.
Read the full abstract

Analysis of PDEs

Comparison results for the $p$-torsional rigidity on convex domains

For each open, bounded and convex domain $\Omega \subset \mathbb{R}^{D},$ $D\geq 2$, and each real number $p>1,$ we denote by $u_{p}$ the $p$-torsion function on $\Omega $, i.e. the solution of the torsional creep problem $\Delta_{p}u=-1$ in $\Omega $, $u=0$ on $\partial \Omega $, where $\Delta _{p}u:=\operatorname{div}( \left\vert \nabla u\right\vert ^{p-2}\nabla u) $ is the $p$-Laplacian. Let $T_p(\Omega)$ be the $p$-torsional rigidity on $\Omega $, defined as $T_{p}\left( \Omega \right) :=\int_{\Omega }u_{p}dx$. Define $T\left( p;\Omega \right) :=\left\vert \Omega \right\vert ^{p-1}T_{p}\left( \Omega \right) ^{1-p}$, where $|\Omega|$ stands for the Lebesgue measure of $\Omega$. The main purpose of this paper is to compare the values of $T(p;\Omega)$ for bounded convex domains having different inradii. We prove that for any $0<a<b$ there exists a constant $\gamma_{D,p}\in[1/D,1)$, depending only on the dimension $D$ and the parameter $p$, such that $T(p;\Omega_b)\leq T(p;\Omega_a)$, for all $ \Omega_a\in\PP^D(a)$, and $\Omega_b\in\PP^D(b)$, if and only if $\gamma_{D,p}b\geq a$, where $\PP^D(r)$ denotes the family of convex bounded domains in $\mathbb{R}^D$ of inradius $r$. In addition, we discuss the asymptotic equality case, the limiting regimes $p\rightarrow 1^+$ and $p\rightarrow\infty$, and the sharpness of our bounds on model families such as rectangles, orthotopes, ellipses, and triangles}. We also derive a Saint-Venant type comparison result under additional geometric constraints, as a direct consequence of our main theorem.
Read the full abstract

Dynamical Systems

$C^1$-generic continuum-wise expansive surface diffeomorphisms

We exhibit a local residual set of surface $C^1$ diffeomorphisms that are continuum-wise expansive but are not expansive. We also exhibit an open and dense set of surface $C^1$ diffeomorphisms where expansiveness implies being Anosov.

Information Theory · Cross-listed from stat.ML

A theory of learning data statistics in diffusion models, from easy to hard

While diffusion models have emerged as a powerful class of generative models, their learning dynamics remain poorly understood. We address this issue first by empirically showing that standard diffusion models trained on natural images exhibit a distributional simplicity bias, learning simple, pair-wise input statistics before specializing to higher-order correlations. We reproduce this behaviour in simple denoisers trained on a minimal data model, the mixed cumulant model, where we precisely control both pair-wise and higher-order correlations of the inputs. We identify a scalar invariant of the model that governs the sample complexity of learning pair-wise and higher-order correlations that we call the diffusion information exponent, in analogy to related invariants in different learning paradigms. Using this invariant, we prove that the denoiser learns simple, pair-wise statistics of the inputs at linear sample complexity, while more complex higher-order statistics, such as the fourth cumulant, require at least cubic sample complexity. We also prove that the sample complexity of learning the fourth cumulant is linear if pair-wise and higher-order statistics share a correlated latent structure. Our work describes a key mechanism for how diffusion models can learn distributions of increasing complexity.
Read the full abstract

Numerical Analysis · Cross-listed from cs.CE

Development of a Methodology for the Automated Spatial Mapping of Heterogeneous Elastoplastic Properties of Welded Joints

Knowledge of the mechanical properties of materials is required for the design and analysis of engineering products, however, the characterisation of heterogeneous properties using traditional techniques is limited by spatial resolution or insufficient reliability. This paper presents a novel methodology for the characterisation of heterogeneous mechanical properties by extending the virtual fields method through the automated spatial parameterisation of constitutive parameters. Collaboration with the United Kingdom Atomic Energy Authority provided this project with an application focus on the characterisation of the spatially-varying, elastoplastic mechanical properties of welded joints. The developed methodology enables the novel characterisation of welds with assorted geometries, varied loading configurations and dissimilar materials. Numerical verification of the developed method was performed using synthetic data equivalent to that obtained experimentally using optical measurements, where the kinematic fields are known and controlled. The results confirm that the proposed approach converges towards the target parameter maps without any a priori information on the distribution of the properties, successfully demonstrating the established methodology as a proof of concept.
Read the full abstract

All authors Robert Hamill, Allan Harte, Aleksander Marek, Fabrice Pierron