https://arxiv.org/api/2v/+ueQvBGspRq4NCHL9O3JonEg 2026-07-26T12:39:00Z 30432 660 15 http://arxiv.org/abs/2601.21645v2 2026-05-29T08:30:19Z

We investigate the relation between end-to-end equivariance and layerwise equivariance in deep neural networks. We prove the following: For a network whose end-to-end function is equivariant with respect to group actions on the input and output spaces, there is a parameter choice yielding the same end-to-end function such that its layers are equivariant with respect to some group actions on the latent spaces. Our result assumes that the parameters of the model are identifiable in an appropriate sense. This identifiability property has been established in the literature for a large class of networks, to which our results apply immediately, while it is conjectural for others. The theory we develop is grounded in an abstract formalism, and is therefore architecture-agnostic. Overall, our results provide a mathematical explanation for the emergence of equivariant structures in the weights of neural networks during training -- a phenomenon that is consistently observed in practice.

2026-01-29T12:47:51Z Accepted at ICML 2026 Vahid Shahverdi Giovanni Luca Marchetti Georg Bökman Kathlén Kohn http://arxiv.org/abs/2605.25896v2 2026-05-29T08:09:06Z

We study rational double points over algebraically closed fields in arbitrary characteristics and completely classify the indecomposable objects in their singularity categories, which correspond to the vertices in their Auslander-Reiten quivers. Along the way, we present an alternative proof determining the configuration of these Auslander-Reiten quivers, and provide methods to handle the homotopy categories of matrix factorizations of isolated hypersurface singularities with computer algebra systems.

2026-05-25T14:25:46Z Yuta Takashima http://arxiv.org/abs/2605.19167v2 2026-05-29T04:53:55Z

We prove the conjecture that higher Verlinde categories are geometrically reductive. This is one of the two properties required in order for recent results on algebraic geometry in tensor categories to apply to these categories. We also reduce two further conjectures concerning geometric reductivity for tensor categories to other conjectures appearing in the literature.

2026-05-18T22:42:46Z Kevin Coulembier http://arxiv.org/abs/2605.12680v2 2026-05-29T04:13:04Z

Majorization inequalities for symmetric polynomials have interested mathematicians for centuries, from the AM-GM inequality for two variables going back at least to Baudhāyana's Śulbasūtra and Euclid's Elements in the first millenium BCE, through classical results of Newton, Muirhead and Gantmacher, to more recent extensions to Schur polynomials and zonal spherical functions. These have been established case by case, with no unified approach. Although it is known that majorization inequalities follow from symmetry and convexity in the indexing partition, the difficulty of proving convexity in specific cases has left a number of outstanding conjectures inaccessible until now. The key insight of this paper is that log-convexity provides a more versatile tool and a unifying principle. It implies convexity and hence majorization, and it is preserved under multiplication and weighted averaging, making it well suited to inductive arguments in a wide range of settings. Using this idea, we prove new majorization inequalities for Macdonald polynomials, Jack polynomials and Heckman-Opdam hypergeometric functions, unifying existing results and resolving several open conjectures.

2026-05-12T19:30:23Z 13 pages, 1 figure Colin McSwiggen Siddhartha Sahi http://arxiv.org/abs/2505.04848v3 2026-05-29T02:09:38Z

Let $\mathscr{C}$ be a symmetric tensor category of moderate growth, and let $\mathcal{H}\leq\mathcal{G}$ be algebraic groups in $\mathscr{C}$. We prove that the homogeneous space $\mathcal{G}/\mathcal{H}$ exists as a scheme and is of finite type when $\mathscr{C}$ is geometrically reductive and maximally nilpotent, conditions that are conjecturally equivalent to incompressibility. A key tool is the introduction of a Frobenius kernel of an group scheme. We further show that while $\mathcal{G}_0/\mathcal{H}_0$ and $(\mathcal{G}/\mathcal{H})_0$ need not be the same, they are close enough, so that $\mathcal{G}/\mathcal{H}$ is quasi-affine/affine/proper if and only if $\mathcal{G}_0/\mathcal{H}_0$ is.

2025-05-07T22:54:46Z Updated exposition, fixed typos Kevin Coulembier Alexander Sherman http://arxiv.org/abs/2605.30739v1 2026-05-29T02:07:58Z

This paper studies integral forms for affine vertex superalgebras and their modules. We first obtain integral forms for the universal enveloping superalgebra $U(\hat{\g})$ of an affine Lie superalgebra $\hat{\g}$. For $\g$ a basic classical Lie superalgebra, we give another construction of integral forms for $U(\hat{\g})$ using Chevalley basis, generalizing Kostant-Garland integral form theory. Then we apply the theory to construct integral forms for vertex (operator) superalgebras based on affine Lie superalgebras and their modules, we also investigate when an integral form contains the conformal vector, and integral forms in contragredient modules for vertex operator superalgebras.

2026-05-29T02:07:58Z Hongyan Guo Hongju Zhao http://arxiv.org/abs/1604.08401v3 2026-05-28T19:40:57Z

This paper studies the combinatorics of lattice congruences of the weak order on a finite Weyl group $W$, using representation theory of the corresponding preprojective algebra $Π$. Natural bijections are constructed between important objects including join-irreducible congruences, join-irreducible (respectively, meet-irreducible) elements of $W$, indecomposable $τ$-rigid (respectively, $τ^-$-rigid) modules and layers of $Π$. The lattice-theoretically natural labelling of the Hasse quiver by join-irreducible elements of $W$ is shown to coincide with the algebraically natural labelling by layers of $Π$. We show that layers of $Π$ are nothing but bricks (or equivalently stones, or 2-spherical modules). The forcing order on join-irreducible elements of $W$ (arising from the study of lattice congruences) is described algebraically in terms of the doubleton extension order. We give a combinatorial description of indecomposable $τ^-$-rigid modules for type $A$ and $D$.

2016-04-28T13:11:43Z 32 pages. Version 2: Expository changes only. Version 3: Final pre-publication version Compositio Math. 154 (2018) 1269-1305 Osamu Iyama Nathan Reading Idun Reiten Hugh Thomas 10.1112/S0010437X18007078 http://arxiv.org/abs/2605.30354v1 2026-05-28T17:59:59Z

Global symmetry anomalies of a quantum field theory (QFT) can be packaged as specific couplings of a higher-dimensional symmetry theory (SymTh). In this work we show that for 5D superconformal field theories (SCFTs) engineered from M-theory backgrounds $X$ a Calabi-Yau cone, this data can be extracted from the path algebra of branes probing $X$. This provides a complementary algebraic approach compared with more geometric computations based on the explicit calculation of triple intersection numbers in a resolved geometry and / or $η$-invariants extracted from the boundary geometry $\partial X$. Our method applies in situations where the counterpart geometric computation is either unknown or combinatorially unwieldy. We illustrate with several toric threefold examples, including orbifolds $\mathbb{C}^{3} / Γ$ and more general non-orbifold Calabi-Yau cones of Sasaki-Einstein five-manifolds.

2026-05-28T17:59:59Z 55 pages + appendices, 12 figures Vivek Chakrabhavi Mirjam Cvetič Jonathan J. Heckman Shani Meynet http://arxiv.org/abs/2605.16950v2 2026-05-28T15:13:18Z

We introduce the notion of quasi-Poisson modules over Lie-Rinehart pairs and prove that for the Lie-Rinehart pair $(\dot A,\dot\fk)$ in which $\dot A=\bbbc[t_1^{\pm1},\ldots,t_m^{\pm1}]\ot\Lam_n$ and $\dot\fk={\rm Der}(\dot A)$, there is a one-to-one correspondence between simple cuspidal quasi-Poisson modules over $(\dot A,\dot\fk)$ and simple cuspidal Harish-Chndra $A\fk$-modules for $A:=\bbbc[t_0^{\pm1}]\ot \dot A$ and $\fk:={\rm Der}(A).$ We also classify simple cuspidal quasi-Poisson modules over the Lie-Rinehart pair $(\dot A,\dot\fk)$ and show that each such module is a tensor module $\dot A\ot Ω$ for an admissible $\frak{gl}(m+1,n)$-module $Ω$ via a prescribed action.

2026-05-16T12:02:34Z Malihe Yousofzadeh http://arxiv.org/abs/2508.06657v3 2026-05-28T14:02:21Z

Some of the multiplicity-freeness results in ``Modular Gelfand pairs and multiplicity-free representations'' are stated in overly broad generality. We provide counterexamples and partial corrections.

2025-08-08T19:15:00Z Revised based on referee's comments. This is the final version. To appear in IMRN Int. Math. Res. Not. IMRN, 2026(11):rnag112, 2026 Patrick Allen Preston Wake Robin Zhang 10.1093/imrn/rnag112 http://arxiv.org/abs/2509.21426v2 2026-05-28T13:56:59Z

If $f$ is a mod-$3$ eigenform of weight 2 and level $Γ_0(\ell^2)$ for a prime $\ell$ such that $\ell \equiv -1 \pmod{3}$, and $\ell$ is a vexing prime for $f$, we show that there is no obstruction to finding a minimal lift of $f$, but that there is an obstruction to finding a nonminimal lift. The key new ingredient that we prove is a modular analog of the standard character formula for a cuspidal representation of $\mathrm{GL}_2(\mathbb{F}_\ell)$, an enhancement that allows us to easily compute the group cohomology of a $3$-adic lattice in such a representation. In fact, we provide a general framework for proving such modular analogs for a broader class of representations using results of Broué and Puig in modular representation theory. We show that this class includes certain Deligne--Lusztig representations and representations coming from higher-depth supercuspidal representations of $\mathrm{GL}_2$.

2025-09-25T11:21:36Z Parts of this paper are from arXiv:2508.06657v1, which has now been split into two papers. This version is revised based on referee comments Patrick B. Allen Preston Wake http://arxiv.org/abs/2508.01563v5 2026-05-28T13:41:12Z

Covering theory is an important tool in representation theory of algebras, however, the results and the proofs are scattered in the literature. We give an introduction to covering theory at a level as elementary as possible.

2025-08-03T03:23:13Z 44 pages Yuming Liu Nengqun Li Bohan Xing Pengyun Chen http://arxiv.org/abs/2605.29921v1 2026-05-28T13:33:33Z

We study spaces of conformal blocks associated with line bundles over elliptic curves, with coefficients in a vertex algebra. For vertex algebras satisfying suitable finiteness and semisimplicity conditions, which are met by all admissible affine vertex algebras as well as admissible W-algebras associated with nilpotent elements of standard Levi type, we prove the holonomicity of the sheaf of conformal blocks over the moduli space of bundles. Furthermore, we show that the space of flat sections of the associated Jacobi-invariant connection is spanned by trace functions on modules. This result provides a substantial generalization of the celebrated theorem of Yongchang Zhu to quasi-lisse vertex algebras. As a special case, we deduce that for affine vertex algebras at admissible level, the dimension of the space of conformal blocks coincides with the number of admissible weights at that level.

2026-05-28T13:33:33Z 63 pages Tomoyuki Arakawa Jethro van Ekeren Hao Li http://arxiv.org/abs/2605.29885v1 2026-05-28T13:10:04Z

Modern statistical learning theory and deep learning characterize generalization primarily in terms of continuous capacity control (e.g., norm-based regularization, margin maximization, low-rank bias). While highly successful in continuous domains, deep learning consistently fails to extrapolate exact algorithmic or discrete algebraic rules, reflecting a missing inductive bias toward algorithmic complexity minimization. We propose the Cayley-table completion as the canonical testbed for this missing bias, serving as the discrete algebraic counterpart to matrix completion. Just as matrix factorization combined with weight decay yields an implicit geometric bias toward low linear rank, recent results demonstrate that operator-valued tensor factorizations paired with a flatness prior yield an implicit algorithmic bias toward exact discrete associativity. We pose the open problem of establishing formal exact recovery bounds for Cayley-table completion, and challenge the community to generalize continuous flatness priors to autonomously discover broader discrete algorithmic axioms without combinatorial search.

2026-05-28T13:10:04Z 6 pages. Submitted to the Conference on Learning Theory (COLT) 2026 Open Problem track Dongsung Huh http://arxiv.org/abs/2205.09700v3 2026-05-28T12:49:15Z

Let $\mathfrak{u}_ζ^\vee$ denote the small quantum group associated with a simple Lie algebra $\mathfrak{g}^\vee$ and a root of unity $ζ$. In [9], a geometric realization of $Z(\mathfrak{u}_ζ^\vee)^{G^\vee}$, the $G^\vee$-invariant part of the center of $\mathfrak{u}_ζ^\vee$, was proposed. We compute the dimension of the geometric subalgebra of the center and in the case where $G=SL_n$, we study a bigraded refinement of the result.

2022-05-19T17:07:39Z v3, 28 pages, 2 figures. Final version, to appear in Representation Theory Nicolas Hemelsoet Oscar Kivinen Anna Lachowska