Peer Revenue Past citations are no guarantee of future returns

Papers

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

Mathematics

math Watch Mathematics (sign in to use)

Sunday 10 May

Mathematical Physics

Two-parameter classes of exactly solvable quantum systems

We introduce two-parameter classes of exactly-solvable novel systems whose Hamiltonian operators could be represented by tridiagonal symmetric matrices in some orthogonal bases. The associated wavefunction is written as point-wise convergent series in the basis elements. The expansion coefficients of the series are orthogonal polynomials in the energy that satisfy the resulting three-term recursion relation starting with two-parameter initial values. These polynomials contain all physical information about the system and they depend on the values of the two parameters. We obtain the associated two-parameter potential function induced by the change in the initial values that causes the system's wavefunction to change. We give several illustrative examples of these systems with continuous and/or discrete energy spectra. Moreover, a curious phenomenon is observed where bound states and/or resonances are induced in a system with pure continuous spectrum (e.g., a free particle) if the two parameters in the initial values exceed certain critical limits.
Read the full abstract

Classical Analysis and ODEs

Sharpness of convolution bounds for measures

In this paper, we determine the optimal universal \(L^p\)-\(L^q\) type sets for convolution operators \(f\mapsto \mu*f\) associated with fractal measures $\mu\in \mathcal P_{\alpha,\beta}(\mathbb R^d)$, which denotes the class of compactly supported Borel probability measures satisfying the \(\alpha\)-Frostman condition \[ \mu(B(x,\rho)) \lesssim \rho^\alpha, \qquad x\in\mathbb R^d,\quad 0<\rho<1, \] and the \(\beta/2\)-Fourier decay condition \[ |\widehat{\mu}(\xi)| \lesssim |\xi|^{-\beta/2}, \qquad \xi\in\mathbb R^d. \] More precisely, we characterize the largest \(L^p\)-\(L^q\) region that is forced solely by the Frostman and Fourier decay assumptions throughout the full admissible range of \((\alpha,\beta)\), with distinct optimal regions in the geometric and nongeometric regimes. We prove optimality in the worst-case sense over \(\mathcal P_{\alpha,\beta}(\mathbb R^d)\) by constructing, for each admissible pair \((\alpha,\beta)\), a single extremal measure whose support has the smallest Hausdorff dimension allowed by the hypotheses. Moreover, variants of the same constructions also yield a single-measure sharpness theorem for the \(L^2\) Fourier restriction theorem of Mockenhaupt–Mitsis–Bak–Seeger: in every dimension and in both the geometric and nongeometric regimes, we construct a measure in \(\mathcal P_{\alpha,\beta}(\mathbb R^d)\) for which the Mockenhaupt–Mitsis–Bak–Seeger threshold exponent is sharp.
Read the full abstract

Number Theory

On Ramanujan Primes for Hecke-Maass Cusp Forms

For a primitive Hecke-Maass cusp form $\phi$ of level $N$ with the $n$-th Hecke eigenvalue $\lambda_{\phi}(n)$ and a prime number $p\nmid N$, the celebrated Ramanujan conjecture at $p$ asserts the following sharp upper bound: \[ |\lambda_{\phi}(p)| \leq 2. \] In this work, we determine an upper bound for the least prime $p$ at which the Ramanujan conjecture holds for two or three distinct primitive Hecke-Maass cusp forms simultaneously. Moreover, given a set of distinct primitive Hecke-Maass cusp forms $\{\phi_i\}$, we also provide a lower bound for the lower natural density of the set of primes at which the Ramanujan conjecture holds for at least one of the $\phi_i$'s.
Read the full abstract

Probability

Stochastic Wright's Equation: Existence of Invariant Measures

Wright's delay differential equation is one of the prime examples of a fully nonlinear equation without an explicit solution and whose dynamics can be understood by analytic means. In this paper, we introduce stochastic perturbations by adding Brownian noise with a bounded Lipschitz noise coefficient to a transformed version of Wright's equation. The transformation considered plays an important role in the deterministic theory as well. We demonstrate that this stochastically perturbed equation has (at least) two invariant measures: a trivial measure concentrated at $-1$ and a nontrivial measure on $(-1,\infty)$. The crucial and most challenging step of the proof is showing that every solution is bounded away from $-1$ in probability. In addition, a major part of our analysis is devoted to deriving detailed estimates for Itô processes with a negative drift.
Read the full abstract

Mathematical Physics · Cross-listed from hep-th

Joint distributions of eigenvectors of symmetric random tensors

We compute the joint distributions of arbitrary numbers of eigenvectors of real and complex symmetric random tensors by the quantum field theoretical methods which were previously used to compute the mean distributions. We obtain the random matrix representations and the large-dimension asymptotics of the joint distributions. The latter can be expressed by a common function of tensor geometries, extending the universality found for the mean distributions to the joint distributions. Several crosschecks of our results are carried out by Monte Carlo computations.
Read the full abstract

Numerical Analysis

Interface Reduction for Elliptic Interface Problems with Conservative Flux Reconstruction

We propose a low-dimensional interface reduction method for elliptic interface problems based on conservative flux reconstruction. The approach combines a fitted $P_1$ finite element discretization with a flux recovery procedure following \cite{ChouTang2000}, yielding locally conservative fluxes that satisfy interface conditions to machine precision. A central result shows that the error of the reduced solution is controlled entirely by the approximation error of the interface data. Numerical experiments for both continuous and discontinuous interface conditions confirm that once the interface data is accurately represented, the full solution is recovered to roundoff accuracy. These results indicate that the essential complexity of elliptic interface problems is concentrated on the interface.
Read the full abstract

Combinatorics

The density of graphs with no $\ell$-path connecting equal-degree vertices: a short proof

Addressing a question posed by Chen and Ma from an asymptotic point of view, we present a short proof for the edge density needed to guarantee that two vertices of the same degree are connected by a path of a fixed length. In particular, we show that for any sufficiently large graph, a density of at least $1/2+o(1)$ enforces the existence of two such vertices. This bound is tight for paths of odd length.
Read the full abstract

All authors Yamaan Attwa, Matías Azócar Carvajal, Simona Boyadzhiyska, Théo Pierron, Anusch Taraz

Analysis of PDEs

Endpoint Architectures for Navier–Stokes

We compare two endpoint analyses for three-dimensional Navier–Stokes flow: a general suitable-weak framework and the axisymmetric equations with swirl. The common backend is selection of physical space–time records, nonoverlapping allocation of positive quantities, retention of the full localized state, and passage either to a compact transition law or to an explicitly named loss of compactness. The front-end theorems are different. In the general problem, bounded-critical sequences lead to symmetry/defect alternatives, while amplitude-fast normalization leads conditionally to an ancient Euler profile only after frequency tightness and a nonvanishing witness are retained. In the axisymmetric problem, logarithmic swirl records are routed to meridional energy or scale-critical signed compression, whereas first-use provenance and recharge packing force recurrent fresh service toward boundary, age, frequency, or other recession mechanisms. We state the interfaces and the remaining endpoints without identifying their currencies or reducing the general problem to the axisymmetric class. No unconditional regularity theorem or singularity construction is claimed.
Read the full abstract

Symplectic Geometry

Symplectic log Kodaira dimension $-\infty$, Hirzebruch–Jung strings and weighted projective planes

We study symplectic minimal resolutions of weighted projective planes $\mathbb{CP}(a,b,c)$ from the perspective of disconnected symplectic divisors with symplectic log Kodaira dimension $-\infty$. Building on the techniques developed in our previous work for connected divisors, we introduce the notion of exceptional gaps between distinct connected components of the divisor and use it to establish a Torelli-type theorem for certain configurations of three Hirzebruch–Jung strings. Motivated by Daigle–Russell's study of affine rulings on complete normal rational surfaces in algebraic context, we also establish a weighted version of Gromov–McDuff's characterization of symplectic $\mathbb{CP}^2$ by showing the existence of symplectic affine rulings implies certain divisor configuration to arise from the minimal resolution of $\mathbb{CP}(a,b,c)$.
Read the full abstract

Optimization and Control · Cross-listed from eess.SY

Action Recommendations for Sequentially Rational Strategic Agents

We consider a finite-horizon discrete-time dynamic system that is jointly controlled by two strategic agents. There is a system designer that has its own reward function but does not have direct control over the agents' actions. We consider an information structure where the current state and all past history are equally accessible by the designer and the agents. The designer sends action recommendations to the agents at each time step. Each agent can use the received recommendation and the available information to choose its action. We are interested in the setting where the designer would like to send recommendations in a way that incentivizes the agents to adopt obedient strategies, i.e., to take the action recommended by the designer. Our goal is to find an optimal action recommendation strategy for the designer that maximizes the designer's objective while ensuring that obedient strategies are sequentially rational for the agents. We provide an algorithm for the designer's problem that involves solving a family of linear programs in a backward inductive manner.
Read the full abstract

Metric Geometry

Flotation Rigidity for Convex Polytopes

We prove that the oriented liquid-level function determines a full-dimensional convex polytope $P\subset\mathbb R^d$, $d\ge2$, and its relative density $\delta\in(0,1)\setminus\{\tfrac12\}$. Uniqueness also follows from buoyancy surfaces at distinct densities accumulating at a non-half interior value. For polygons, fixed-density buoyancy data determine the normalized difference body, yielding uniqueness among centrally symmetric polygons at every density.
Read the full abstract

Optimization and Control · Cross-listed from cs.LG

Bayesian Optimization with Structured Measurements: A Vector-Valued RKHS Framework

Bayesian optimization (BO) is an efficient framework for optimizing expensive black-box functions. However, it is typically formulated as learning an end-to-end mapping from inputs to scalar objectives, thereby discarding the potentially rich information whenever a structured system output is available. In this work, we study Bayesian optimization over a vector-valued operator with structured measurements, where each measurement observes multidimensional or functional outputs, e.g., trajectories or spatial fields, rather than a single scalar value. The objective is then defined as a linear functional of these measurements. This allows each observation to reveal substantially richer information about the underlying system compared to scalar observations. Assuming the unknown operator lies in a vector-valued reproducing kernel Hilbert space (RKHS), we derive high-probability concentration bounds for the kernel ridge regression (KRR) estimator directly in the measurement space, characterizing uncertainty in a general Hilbert space. Building on these results, we propose an algorithm based on the upper confidence bound (UCB) acquisition function with regret guarantees under mild assumptions, recovering sublinear rates for common kernels. Empirically, we demonstrate that leveraging structured measurements leads to improved sample efficiency by enabling efficient transfer of information across objectives and adaptation to time-varying settings.
Read the full abstract

Optimization and Control · Cross-listed from eess.SY

Safe Exploration for Nonlinear Processes Using Online Gaussian Process Learning

This paper proposes a safe data-driven control framework for nonlinear systems with partially known dynamics. The method ensures stability and constraint satisfaction during online learning, assuming only a stabilizable linear approximation of the process is available. Unmodeled nonlinear dynamics are captured by a Gaussian process residual learned in real time. Safety is enforced through a probabilistic control-invariant set derived from Lyapunov theory, guaranteeing high-probability stability. A convex quadratic program computes control inputs that maximize information gain while respecting probabilistic safety constraints. The framework provides finite-sample safety guarantees and allows adaptive expansion of the invariant set as uncertainty decreases. Numerical results validate the approach, demonstrating safe and informative exploration under model uncertainty: the safe set expands by about 30% while the Gaussian process root-mean-square error drops from 1.11 to 0.03.
Read the full abstract

All authors Stefano Tonini, Soroush Rastegarpour, Hamid Reza Feyzmahdavian, Nicola Bastianello, Karl Henrik Johansson

Algebraic Geometry

Centralizers of the complex orthogonal and symplectic group

We find a recursive algorithm for computing the precise centralizers of the complex orthogonal and symplectic groups, and hence the isotropy groups, with respect to the similarity transformation on the spaces of skew-symmetric and Hamiltonian matrices, respectively. These groups are conjugate to groups of certain nonsingular block matrices whose blocks are rectangular block Toeplitz.
Read the full abstract

Group Theory

Cyclic Subgroups of Belk-Hyde-Matucci Group $V\!\mathcal{A}$

In this paper it is proved that the Belk-Hyde-Matucci group $V\!\mathcal{A}$, a group containing every countable abelian group, does not contain subgroups with distorted cyclic subgroups.

Algebraic Geometry

Grothendieck Weights on Permutohedral Varieties and Matroids

Grothendieck weights, introduced by Shah, are $K$-theoretic analogues of Minkowski weights on smooth toric varieties. We study Grothendieck weights on the permutohedral fan and prove two main results: a $K$-balancing condition that characterizes Grothendieck weights by a finite system of linear equations, and an explicit product rule for the ring structure. We apply this framework to matroids, giving a combinatorial characterization of Grothendieck weights on matroidal fans. As the main application, we compute the motivic Chern class of the hyperplane arrangement complement in its wonderful compactification and show that the result depends only on the matroid, not on the realization. This allows us to extend the definition of the motivic Chern class to all loopless matroids.
Read the full abstract

Representation Theory

Hecke monoids, their homomorphisms and parabolicity

We study homomorphisms of Hecke monoids, notably parabolic homomorphisms, which map parabolic elements to parabolic elements, and injective ones. The importance of the first class stems from the fact that parabolic elements form a rather mysterious submonoid of the Hecke monoid, and we found a plethora of parabolic homomorphisms. Concerning injective ones, as a first step towards their classification, we classified all locally injective connected homomorphisms between Hecke monoids of classical types and expect all of them to be injective. As a surprising byproduct of our study of parabolic and injective homomorphisms we described, to some extent, all homomorphisms between Hecke monoids.
Read the full abstract

Analysis of PDEs

Weighted Neumann-to-Steklov limits for nonlinear eigenvalues and trace constants

We study a nonlinear Neumann-to-Steklov limit generated by a family of interior weights concentrating at the boundary. On a class of admissible possibly irregular domains obtained from the unit ball by trace-compatible Sobolev homeomorphisms, we consider the first nontrivial weighted \((p,q)\)-Neumann eigenvalue with respect to a concentrating bulk weight \(\gamma_a\). We prove that, as \(a\to0\), these eigenvalues converge to the corresponding weighted \((p,q)\)-Steklov eigenvalue with boundary weight \(\beta\). Moreover, normalized minimizers converge, up to subsequences, strongly in \(W^{1,p}\) to Steklov minimizers. Equivalently, the best constants in the weighted Poincaré inequalities converge to the best constants in the weighted trace inequalities; in fact, a quantitative convergence estimate is obtained in the subcritical trace range.
Read the full abstract

Numerical Analysis

Accelerating Power Method with Fast Sketching for Stronger Low-Rank Approximation

The power method is one of the most fundamental tools for extracting top principal components from data through low-rank matrix approximation. Yet, when the target rank is large, the cost of matrix multiplication associated with this procedure becomes a major bottleneck. We develop an algorithmic and theoretical framework for accelerating the power method using fast sketching, which is a popular paradigm in randomized linear algebra. Our framework leads to simple and provably efficient methods for singular value decomposition, low-rank factorization, and Nyström approximation, which attain strong numerical performance on benchmark problems. The key novelty in our analysis is the use of regularized spectral approximation, a property of fast sketching methods which proves more flexible in generalizing power method guarantees than traditional arguments.
Read the full abstract

Information Theory

Learning from Acceptance: Cumulative Regret in the Game of Coding

Classical coding-theoretic guarantees often rely on trust assumptions, such as requiring sufficiently many honest nodes compared with adversarial ones. These assumptions are difficult to enforce in open decentralized systems where participants are not centrally certified. At the same time, such environments often contain incentive mechanisms: participants may be rewarded only when their submitted data are accepted and the system remains functional. This changes the role of an adversary. Rather than acting as a pure saboteur, a strategic adversary may submit data that are consistent enough to be accepted while still degrading the quality of the final estimate. The game-of-coding framework models this strategic interaction between a data collector (DC) and an adversary. Existing works on the game of coding mostly consider the complete-information case, where the DC knows how the adversary trades off acceptance and estimation error. In this paper, we study an incomplete-information version of the game of coding in which the DC, acting as a Stackelberg leader, does not know the adversary's utility trade-off and must learn through repeated interaction. Prior work on the unknown-adversary setting considered an explore-then-commit objective, where only the final selected acceptance rule is evaluated. In contrast, we study the full learning trajectory: every acceptance rule used during the algorithm is executed and contributes to performance. We propose an algorithm that refines its search around promising acceptance rules, prove that it achieves sublinear cumulative regret, and evaluate its performance through numerical experiments.
Read the full abstract

Information Theory

Recovery Models for Linear Batch Codes

Various types of recovery algorithms for batch codes have been considered previously, including asynchronous recovery and recovery algorithms arising from Almost Affinely Disjoint (AAD) families. We initiate a systematic investigation of linear batch codes equipped with recovery algorithms. We introduce online and strongly online batch codes, investigate their relations with asynchronous batch codes, and introduce (m,L)-strongly online batch codes, which provide particularly simple recovery algorithms. We then study simple batch codes associated with arbitrary bipartite graphs, obtaining graph-theoretic conditions for such codes and generalizing several known results.
Read the full abstract

All authors Baran Düzgün, Henk D.L. Hollmann, Ago-Erik Riet, Vitaly Skachek, Vladislav Taranchuk

Information Theory · Cross-listed from cs.LG

Entropy-informed Decoding: Adaptive Information-Driven Branching

Large language models (LLMs) achieve remarkable generative performance, yet their output quality is dependent on the decoding strategy. While sampling-based methods (e.g., top-k, nucleus) and search-and-select based methods (e.g., beam search, best-of-n, majority voting) can improve upon greedy decoding, both approaches suffer from limitations: sampling generally commits to a single path, while search often expends excessive computation regardless of task complexity. To address these, we introduce Entropy-informed decoding (EDEN), a plug-and-play, model-agnostic decoding framework that adaptively allocates computation based on the model's own uncertainty, approximating higher-width beam search with fewer expansions. At each generation step, EDEN estimates the entropy of the output token distribution and adjusts the branching factor monotonically with the entropy, expanding more candidates in high-entropy regions and following a greedier path in low-entropy regions, improving token efficiency. Experiments across complex tasks, including mathematical reasoning, code generation, and scientific questions, demonstrate that EDEN consistently improves output quality over existing decoding strategies, achieving better accuracy-expansion trade-offs than fixed-width beam search. By treating next-token selection as a noisy maximisation problem, we prove that branching factors monotone in entropy are guaranteed to find better (i.e. more probable) continuations than any fixed branching factor within the same total expansion budget, and derive explicit regret rates characterising the benefit of the adaptive allocation.
Read the full abstract

Analysis of PDEs

Solutions of the Navier-Stokes equations with forced rapid space-time decay

We study the pointwise decay properties of solutions to the incompressible Navier-Stokes equations, both in the space and time variables. It is well known that generic global solutions on $\mathbb{R}^n$ do not decay faster at infinity than $|x|^{-(n+1)}$ and $t^{-(n+1)/2}$ in the pointwise sense. In this paper, we address the control problem of constructing an external forcing and a solution to the Navier-Stokes equations whose space-time decay properties go beyond these limiting rates. A distinctive feature of the forcing term is that its spatial profile can be fixed once and for all, independently of the initial data of the problem, and localized in an arbitrarily small region of $\mathbb{R}^n$. Only the temporal profile of the external force displays a dependency on the initial datum.
Read the full abstract

Number Theory

Lacunary recurrences and 2-adic properties of Eisenstein series

We study the rational coefficients that arise when the Eisenstein series $G_k$ is expressed as a polynomial in $G_4$ and $G_6$, proving a conjecture that gives an exact formula for their minimal 2-adic valuation in terms of the binary expansion of the weight. The proof uses lacunary recurrences for Eisenstein series and yields refined information about the first valuation levels. As an application, we prove irreducibility results for Faber polynomials associated to dyadic linear combinations of powers of Eisenstein series.
Read the full abstract

Combinatorics · Cross-listed from cs.DS

TreeWidzard: An Engine for Width-Based Dynamic Programming and Automated Theorem Proving

In this work, we introduce TreeWidzard, an engine for developing dynamic programming algorithms that decide graph-theoretic properties parameterized by treewidth and pathwidth. Besides providing a unified framework for algorithms deciding atomic graph-theoretic properties, our engine allows one to combine such algorithms for two purposes: to obtain dynamic programming algorithms for more complex graph properties, and to support treewidth-based automated theorem proving. Within this context, given the specification of a Boolean combination \(P\) of graph properties \(P_1, P_2, \ldots, P_r\), and a positive integer \(k\), our engine can be used to determine whether all graphs of treewidth at most \(k\) satisfy \(P\). The main goal of the present work is to provide a system description of TreeWidzard. In particular, we provide a step-by-step account of how to implement dynamic programming algorithms in our framework and how to combine these algorithms for model checking and automated theorem proving.
Read the full abstract

Number Theory

The variation of zeros of the Miller basis

We exhibit a connection between the variation of zeros in the Miller basis of modular forms $q^m+O(q^{\ell+1})$ and a logarithmic version $\mathcal{S}_\delta$ of the Szegő curve, where $\delta=m/\ell$. When $\delta<0.6194$ we show that all the zeros are on the unit arc for $k\gg 0$, while if $\delta$ is asymptotically close to 1, we show that all the zeros lie on $\mathcal{S}_{\delta}$. In general, we posit that for all $\delta$, the zeros are located on the union of the unit arc and the log Szegő curve, obtaining a partial result, and find conjectural thresholds for $m/\ell$ with all zeros on the unit arc, and no zeros on the arc. Finally, we enumerate all algebraic zeros of Miller forms up to $\ell-m\leq 25$.
Read the full abstract

Logic

Algebraic characterisation of pseudo-elementary and second-order classes

In this paper we provide purely model-theoretic (algebraic) characterisations for classes definable in second-order logic and for pseudo-elementary classes (including PC and PC_{\Delta} classes). Classical results of this flavour include Keisler-Shelah type theorems (characterising first-order definability by closure under ultraproducts and ultraroots) and Birkhoff's HSP theorem; a key starting point for this paper is Sági's work, which provides an algebraic description of classes definable by existential second-order sentences. Here we resolve several open problems from the literature. Our main results are the following. We solve the long-standing problem of giving a purely algebraic characterisation of pseudo-elementary classes: we characterise PC_{\Delta} classes by intrinsic closure properties. We also give a characterisation for the basic pseudo-elementary classes (PC). We provide a structural classification of second-order equivalent structures, and we obtain purely algebraic characterisations of the classes definable by second-order formulas as well as those definable by finitely many second-order sentences.
Read the full abstract

Statistics Theory

On the Impossibility of Specification Testing of Interference Models Based on Exposure Mappings

Researchers use interference models based on exposure mappings to facilitate estimation of causal effects in randomized experiments with interference. To test the veracity of such models, researchers can use specification tests that aim to detect departures from the stipulated model. However, existing tests suffer from poor power and are often unable to detect important model violations. The main result in this paper is to show that the specification testing problem for exposure mapping models is inherently difficult, and the poor power of existing tests is inescapable. In particular, the worst-case Type I and Type II error rates must sum to one for any specification test of such models, ruling out the existence of a uniformly consistent test. This is the worst-case overall error rate achieved by a naive test that discards all data and arbitrarily rejects the null at random; the testing problem is in this sense impossible. This negative result holds true for all exposure mappings, all sample sizes, for uniformly bounded outcomes, and for alternatives that are maximally separated from the null. While some tests can detect some type of departures from the null model, there will always be relevant departures from the null that are undetectable. Informative specification tests must therefore restrict the alternative model against which they seek to attain power for, beyond the restrictions imposed by the exposure mappings alone. We illustrate this by providing a uniformly consistent test for differentiating no-interference from a network-linear-in-means model.
Read the full abstract

All authors Chao Gao, Christopher Harshaw, Fredrik Sävje, Yitan Wang

Probability · Cross-listed from stat.ML

Learning stochastic multiscale models through normalizing flows

Many systems in physics, engineering, and biology exhibit multiscale stochastic dynamics, where low-dimensional slow variables evolve under the influence of high-dimensional fast processes. In practice, observations are often limited to a single trajectory of the slow component, while the fast dynamics remain unobserved, making statistical learning challenging. Approaches based on partial differential equations (PDE), such as Fokker-Planck formulations, aim to characterize the evolution of probability densities, typically requiring dense space-time data or grid-based solvers. In contrast, we adopt a trajectory-based perspective and develop a data-driven framework for learning effective stochastic dynamics from a single observed path. We model the dynamics by coupled multiscale stochastic differential equations (SDEs) and first obtain a principled model reduction through stochastic averaging. Unlike generic model reduction techniques such as PCA, this respects the dynamical structure of the original system and explicitly incorporates the interaction between slow and fast scales. A central challenge, however, is that the reduced model depends on the invariant distribution of the fast process, which is a solution to an intractable and often unknown PDE. We introduce a novel learning framework that parameterizes the invariant distribution using normalizing flows, enabling expressive density modeling in the latent fast-variable space. The flow is trained end-to-end by optimizing a penalized likelihood objective induced by the reduced stochastic dynamics. Furthermore, we develop a Bayesian variational inference procedure for uncertainty quantification, employing a second normalizing flow to approximate the posterior distribution over model parameters. This yields a scalable approach to capturing epistemic uncertainty in multiscale systems.
Read the full abstract

Statistics Theory

The general regularisation scheme applied to conditional density estimation

The general regularisation scheme, a versatile approach for nonparametric estimation, has been successfully applied to regression, density ratio, and score estimation. In this paper, we introduce a unified framework encompassing these settings and extend it to conditional density estimation, deriving a new estimator with rigorously established convergence rates. We implement the Landweber regularisation, which is computationally more tractable than Tikhonov regularisation in this context. Numerical experiments demonstrate that our estimator matches or outperforms the Nadaraya-Watson estimator in various scenarios, including time series models.
Read the full abstract