arXiv subjects

Yifeng Huang

Publications and source records attributed to Yifeng Huang.

At least 19 recordsLinked to original sources

arXiv2026

Rogers--Ramanujan identities from the geometry of $X^a=Y^b$

We prove the conjecture of Huang, Jiang, and Oblomkov (HJO) giving a geometric extension of the Rogers--Ramanujan and Andrews--Gordon identities for every torus-knot singularity $X^a=Y^b$ with coprime $1<a<b.$ For a prime power $q$, let $\mathcal{NC}_n^{a,b}(\mathbb F_q)$ denote the set of pairs of commuting nilpotent $n\times n$ matrices $(A,B)$ over $\mathbb F_q$ satisfying $A^a=B^b$. We establish the threefold equality between their normalized counts, the HJO $q$-series $Z_{a,b}$, and the explicit infinite product $P_{a,b}$: \[ \underbrace{\vphantom{\Bigg|} \prod_{m\geq1}(1-q^{-m}) \Biggl(\sum_{n=0}^{\infty} \frac{\lvert\mathcal{NC}_n^{a,b}(\mathbb F_q)\rvert} {\lvert\operatorname{GL}_n(\mathbb F_q)\rvert}\Biggr) }_{\text{point count}} = \underbrace{\vphantom{\Bigg|}Z_{a,b}(q^{-1}) }_{\text{$q$-series}} = \underbrace{\vphantom{\Bigg|}P_{a,b}(q^{-1}) }_{\text{infinite product}}. \] Our main result is a stronger finite identity: the rank $N$ HJO sum equals $(q;q)_N$ times the generating function for balanced cylindric partitions with entries bounded by $N$. Taking $N\to\infty$ yields the HJO conjecture. The proof combines the compositional rational shuffle theorem of Bergeron--Garsia--Leven--Xin and Mellit with a multiplicativity theorem for slope operators and a determinantal model for bounded cylindric partitions, linked by a common $q$-difference equation. The finite identity and the HJO conjecture have been formalized in Lean by AxiomProver, conditional on two stated literature inputs.

math.NT↗

arXiv2026

Quot scheme of points on torus knot singularities

For $\gcd(a,b)=1$, we show that the moduli space of $m$-codimensional $\Bbbk[\![T^a,T^b]\!]$-submodules of $\Bbbk[\![T]\!]^n$ is paved by affine cells, by proving that each Białynicki-Birula stratum of a closed related moduli space with respect to the natural $\mathbb{G}_m$-action is an affine bundle over the fixed point locus and that the fixed point locus is an iterated Grassmannian bundle. As an application, we determine the motive of this moduli space in the Grothendieck ring of varieties in terms of an explicit two-variable series $N_{a,b;n}(q,t)$, and use it to explicit compute the groupoid volume of the category of finite modules over $\mathbb{F}_q[\![T^a,T^b]\!]$. The series $N_{a,b;n}$ carries the conjectures we then formulate. At $n=\infty$ we conjecture a bi-infinite family of Rogers--Ramanujan type identities by specializing the $t$-variable; we identify their product side with the normalized character of a module over the $\mathcal{W}$-algebra minimal model $\mathcal{W}_a(a,a+b)$, and observe a connetion to colored Jones tails. At $n<\infty$ we conjecture that $N_{a,b;n}$ is computed by the bottom $α$-row of the trigraded $S^n$-colored HOMFLY homology of the torus knot $T(a,b)$, and that this same bottom row also computes the Quot schemes of finite codimensional $\Bbbk[\![T^a,T^b]\!]$-submoudles of $\Bbbk[\![T^a,T^b]\!]^n$ and the punctual Hilbert schemes of the non-reduced curve $(Y^a-X^b)^n=0$; the three quantities are special values at three points of the trigrading, and when $n=1$ they recover both the conjectures of Oblomkov--Rasmussen--Shende and of Kivinen--Trinh. Finally we conjecture that the one direction of the trigrading these three points do not see is a perverse filtration on the moduli spaces themselves, and we verify its prediction for a smooth germ at $n=2$ by computing the decomposition theorem for the $\mathrm{GL}_2$ spectral-curve family.

math.AG↗

arXiv2026

Algebraic geometric framework of Rogers--Ramanujan identities

The Rogers--Ramanujan identities equate a $q$-series whose exponents are governed by a quadratic form with an infinite product supported on two residue classes modulo~$5$. Identities of this shape are scarce, and a central problem is to identify the structures that produce them in families. Huang, Jiang, and Oblomkov have proposed a source of a new kind: to each pair of coprime integers $a,b>1$ they attach an infinite-rank $q$-series $Z_{a,b}(q)$, assembled from counts of commuting nilpotent matrix pairs $(A,B)$ with $A^a=B^b$ over finite fields, and they conjecture that it equals an explicit product of $(a-1)(b-1)/2$ modular units of level $a+b$. The $a=2$ cases are the Andrews--Gordon identities; no case with $a>2$ was known. We prove the conjecture for $(a,b)=(3,4)$, $(3,5)$, $(3,7)$, and $(3,8)$. Our proofs pass through a finer sum-to-sum identity, which we conjecture for all $b$ coprime to $3$ and establish for all $b$ when $q=1$. Lau and Ono have since proved that identity in general, and with it the full $a=3$ case. These identities have been formalized and verified in Lean by AxiomProver.

math.NT↗

arXiv2026

Matrix Points on Varieties

We study the cohomology of $C_n(X)$, the moduli space of commuting $n$-by-$n$ matrices satisfying the equations defining a quasi-projective scheme $X$. This space can be viewed as a non-commutative Weil restriction from the algebra of $n$-by-$n$ matrices to the ground field. We introduce a semi-simple counterpart $S_n(X)$, defined as the quotient of $X^n \times \mathrm{GL}_n/\mathrm{T}_n$ by the diagonal $S_n$ action. We show that there exists a natural map $σ\colon S_n(X) \to C_n(X)$ inducing isomorphism on $\ell$-adic cohomology under mild restrictions on $X$ or the characteristic of the field. This confirms a heuristic derived from Weil restrictions. Furthermore, we provide explicit combinatorial formulae for the Betti numbers of $C_n(X)$ and prove a Macdonald-type generating series. A version for Hermitian matrix point is also proved in the last section.

math.AG↗

arXiv2026

AEM: Adaptive Entropy Modulation for Multi-Turn Agentic Reinforcement Learning

Reinforcement learning (RL) has substantially improved the ability of large language model (LLM) agents to interact with environments and solve multi-turn tasks. However, effective agentic RL remains challenging: sparse outcome-only rewards provide limited guidance for assigning credit to individual steps within long interaction trajectories. Existing approaches often introduce dense intermediate supervision, such as process reward models or auxiliary self-supervised signals, which increases supervision and tuning complexity and may limit generalization across tasks and domains. We present AEM, a supervision-free credit assignment method that adaptively modulates entropy dynamics during RL training to improve the exploration-exploitation trade-off. Since in agentic RL the environment is typically affected by a complete response, rather than an individual token, our analysis lifts entropy dynamics from the token level to the response level, aligning uncertainty estimation with the effective action granularity of LLM agents and reducing sensitivity to token-level sampling noise. We further show that entropy drift under natural-gradient updates is governed by the interaction between the sampled-response advantage and its relative surprisal. Motivated by this result, AEM derives a practical response-level uncertainty proxy and uses it to rescale advantages, leveraging the evolving balance between positive and negative samples to naturally transition from exploration to exploitation. Extensive experiments on ALFWorld, WebShop, and SWE-bench-Verified with models ranging from 1.5B to 32B demonstrate that AEM consistently improves strong RL baselines, including a +1.4\% gain when integrated into a state-of-the-art software-engineering RL training framework.

cs.AI↗

arXiv2026

A quadratic form generalization of rational dinv

We introduce a quadratic form $Q$ on the space of functions on the gap poset $G$ of the numerical semigroup $\langle a,b\rangle$. We prove combinatorially that when evaluated on the indicator function of an upward closed subset $D$, this quadratic form precisely recovers the Gorsky--Mazin $\mathtt{dinv}$ statistic of $D$, viewed as a Young subdiagram of $G$. Furthermore, we prove Theorem~1.2 that when evaluated on a pair of subdiagrams of $G$, the symmetric bilinear form associated with $Q$ is equal to a novel cross-$\mathtt{dinv}$ statistic, which is nonnegative. Combining these, we prove the inequality \[ Q(\mathbf{n})\geq \dfrac{1}{|G|}\,\|\mathbf{n}\|_\infty^2\] if $\mathbf{n}$ is a real-valued decreasing function on $G$, showing an effective positive definiteness of $Q$ on the corresponding cone. Theorem~1.2, the main engine of the paper, was autoformalized in Lean/Mathlib by AxiomProver.

math.CO↗

arXiv2026

Atomic-scale tunable phonon transport at tailored grain boundaries and Their Impact on Thermal Conductivity

Grain boundaries (GBs) strongly influence thermal transport in crystalline solids by disrupting lattice periodicity and scattering phonons. Due to the atomic-level disorder and structural complexity, a fundamental understanding of how specific GB geometries regulate nanoscale phonon behavior and macroscopic thermal conductivity has remained elusive. Here, using emerging atomic-resolution vibrational spectroscopy, we directly correlate GB structure, defect-specific vibrational states, and thermal transport in bicrystal strontium titanate with controlled tilt and twist angles. The phonon characterizations and thermal conductivity data reveal two distinct regimes, where low-angle tilt GBs (2deg, 6deg, 10deg) substantially modulate phonon populations and mode frequencies, resulting in pronounced changes in thermal conductivity, whereas high-angle tilt GBs (22deg, 36deg) exhibit weak conductivity variation due to saturated structural disorder and scattering. In contrast, twist GBs introduce periodic defect motifs that only locally tailor phonon transport. Our results suggest tilt and twist angles as complementary knobs for coarse and fine control of phonon propagation and cross-GB thermal transport, providing a predictive framework for thermal engineering in materials.

cond-mat.mes-hall↗

arXiv2026

CountEx: Fine-Grained Counting via Exemplars and Exclusion

This paper presents CountEx, a discriminative visual counting framework designed to address a key limitation of existing prompt-based methods: the inability to explicitly exclude visually similar distractors. While current approaches allow users to specify what to count via inclusion prompts, they often struggle in cluttered scenes with confusable object categories, leading to ambiguity and overcounting. CountEx enables users to express both inclusion and exclusion intent, specifying what to count and what to ignore, through multimodal prompts including natural language descriptions and optional visual exemplars. At the core of CountEx is a novel Discriminative Query Refinement module, which jointly reasons over inclusion and exclusion cues by first identifying shared visual features, then isolating exclusion-specific patterns, and finally applying selective suppression to refine the counting query. To support systematic evaluation of fine-grained counting methods, we introduce CoCount, a benchmark comprising 1,780 videos and 10,086 annotated frames across 97 category pairs. Experiments show that CountEx achieves substantial improvements over state-of-the-art methods for counting objects from both known and novel categories. The data and code are available at https://github.com/bbvisual/CountEx.

cs.CV↗

arXiv2026

Low degree subvarieties of universal hypersurfaces

We study irreducible subvarieties of the universal hypersurface $\mathcal{X}/B$ of degree $d$ and dimension $n$. We prove that when $d$ is sufficiently large, a degree $kd$ subvariety $Z$ which dominates $B$ comes from intersection with a family of degree $k$ projective varieties parametrized by $B$. This answers a question raised independently by Farb and Ma. Our main tools consist of a Grassmannian technique due to Riedl and Yang, a theorem of Mumford-Roitman on rational equivalence of zero-cycles, and an analysis of Cayley-Bacharach conditions in the presence of a Galois action. We also show that the large degree assumption is necessary; for $d=3$, rational points are dense in $\text{Sym}^dX_{k(B)}$, and in particular are not collinear.

math.AG↗

arXiv2025

Multiple Rogers-Ramanujan type identities for inert quadratic orders

We compute the Quot and finitized Coh zeta functions of the inert quadratic orders $\mathbb{F}_q[[T]]+T^{m}\mathbb{F}_{q^{2}}[[T]]$ for every $m\geq 1$ in terms of a $2m$-fold multisum, and then show this multisum equals an $m$-fold Bressoud sum. This proves a recent conjecture of the second author, rounding up the line of exploration in the series of work by the authors and Jiang. The equality between the $2m$-fold multisum and the $m$-fold Bressoud sum is built upon generalizing the multisum by introducing a ``ghost'' parameter $a$ to its summands. We then show that such an $a$-generalization is surprisingly $a$-independent by purely $q$-theoretic techniques. Finally, we propose a refined multisum that interpolates two versions of Quot zeta functions for all three types of quadratic orders.

math.AG↗

arXiv2025

Motivic Coh and Quot zeta functions of singular curves

We present a general and effective algebraic framework for enumerating finite-length quotients of a torsion-free sheaf of arbitrary rank (the Quot zeta function) and finite-length coherent sheaves (the Coh zeta function) over reduced singular curves. We prove that Quot zeta functions are motivically rational, using a novel parametrization and the geometry of affine Grassmannians, and that they satisfy an arbitrary-rank reflection principle, via harmonic analysis. We show that the a normalized high-rank limit of Quot zeta functions converges to the Coh zeta function. As a first application, we compute explicit formulas for these zeta functions for all $y^2 = x^n$ singularities, revealing a surprising and previously unknown connection to Rogers--Ramanujan type $q$-series. Further applications to affine Springer fibers and commuting varieties are also discussed.

math.AG↗

arXiv2025

Can Current AI Models Count What We Mean, Not What They See? A Benchmark and Systematic Evaluation

Visual counting is a fundamental yet challenging task, especially when users need to count objects of a specific type in complex scenes. While recent models, including class-agnostic counting models and large vision-language models (VLMs), show promise in counting tasks, their ability to perform fine-grained, intent-driven counting remains unclear. In this paper, we introduce PairTally, a benchmark dataset specifically designed to evaluate fine-grained visual counting. Each of the 681 high-resolution images in PairTally contains two object categories, requiring models to distinguish and count based on subtle differences in shape, size, color, or semantics. The dataset includes both inter-category (distinct categories) and intra-category (closely related subcategories) settings, making it suitable for rigorous evaluation of selective counting capabilities. We benchmark a variety of state-of-the-art models, including exemplar-based methods, language-prompted models, and large VLMs. Our results show that despite recent advances, current models struggle to reliably count what users intend, especially in fine-grained and visually ambiguous cases. PairTally provides a new foundation for diagnosing and improving fine-grained visual counting systems.

cs.CV↗

arXiv2025

Cohen-Lenstra flag universality for random matrix products

For $n \times n$ random integer matrices $M_1,\ldots,M_k$, the cokernels of the partial products $\mathrm{cok}(M_1 \cdots M_i), 1 \leq i \leq k$ naturally define a random flag of abelian $p$-groups. We prove that as $n \to \infty$, this flag converges universally, for any nondegenerate entry distribution, to the Cohen-Lenstra type measure which weights each flag inversely proportional to the size of its automorphism group. As a corollary, we prove universality of certain formulas for the limiting conditional distribution of $\mathrm{cok}(M_1M_2)$ given $\mathrm{cok}(M_1),\mathrm{cok}(M_2)$ in terms of Hall-Littlewood structure constants, which were previously obtained only for Haar matrices over $\mathbb{Z}_p$. Our proofs combine the general technology of Sawin-Wood, matrix product moment computations following those of Nguyen-Van Peski, and the computation done previously for Haar $p$-adic matrices by Huang.

math.PR↗

arXiv2025

Lattices in $\mathbb F_q[[T]]^d$ and spiral shifting operators

We investigate the algebra and combinatorics of an analogue of the Hermite normal form that classifies finite-index submodules of $\mathbb F_q[[T]]^d$. We identity both normal forms as instances of Gröbner basis theory under different monomial orders, where the Hermite normal form corresponds to the lex order, and the new normal form the hlex order. We note that the hlex normal form recovers the Smith normal form, a feature not enjoyed by the Hermite normal form. We also identify the combinatorial structure underlying the cell decomposition induced by the hlex normal form, which appears to be of independent interest. Notably, the statistics tracking the cell dimensions is compatible, in a certain way, with a collection of $d$ ``spiral shifting operators'' on $\mathbb N^d$, which pairwise commute and collectively act freely and transitively. Using these operators, we give direct proofs of some new combinatorial identities obtained by translating the results of Solomon and Petrogradsky in terms of the hlex normal form.

math.CO↗

arXiv2025

DualMat: PBR Material Estimation via Coherent Dual-Path Diffusion

We present DualMat, a novel dual-path diffusion framework for estimating Physically Based Rendering (PBR) materials from single images under complex lighting conditions. Our approach operates in two distinct latent spaces: an albedo-optimized path leveraging pretrained visual knowledge through RGB latent space, and a material-specialized path operating in a compact latent space designed for precise metallic and roughness estimation. To ensure coherent predictions between the albedo-optimized and material-specialized paths, we introduce feature distillation during training. We employ rectified flow to enhance efficiency by reducing inference steps while maintaining quality. Our framework extends to high-resolution and multi-view inputs through patch-based estimation and cross-view attention, enabling seamless integration into image-to-3D pipelines. DualMat achieves state-of-the-art performance on both Objaverse and real-world data, significantly outperforming existing methods with up to 28% improvement in albedo estimation and 39% reduction in metallic-roughness prediction errors.

cs.CV↗

arXiv2025

Coh zeta functions for inert quadratic orders

We study the Coh zeta function for a family of inert quadratic orders, which we conjecture to be given by $t$-deformed Bressoud $q$-series. This completes a trilogy connecting the zeta functions of ramified and split quadratic orders to the classical Andrews--Gordon and Bressoud identities, respectively. We provide strong evidence for this conjecture by deriving the first explicit formulas for the finitized Coh zeta function of the simplest order in the family, and for the $t=1$ specialization of the finitized Coh zeta functions for all orders in the family. Our primary tool is a new method based on Möbius inversion on posets.

math.NT↗

arXiv2025

Hilbert Series for Configuration Spaces of Punctured Surfaces

Let $Σ_{g,r}$ denote the $r$-punctured closed Riemann surface of genus $g$. For every $g\geq 0$, we determine the four-variable generating function for the mixed Hodge numbers of the unordered configuration spaces of $Σ_{g,1}$. The cases where $g\geq 2$ are new. Combining a result of \cite{huang2020cohomology}, this determines the analogous generating function for $Σ_{g,r}$ for all $r\geq 1$. As an application of our formula we illustrate how classical homological stability results, as well as so-called secondary stability results of \cite{miller2019higher} can be interpolated to illustrate stable behaviors in the mixed Hodge numbers of these spaces which have been thus-far undiscovered.

math.AG↗

arXiv2024

Cohomology of configuration spaces on punctured varieties

In the theory of configuration spaces, "splitting" usually refers to the phenomenon that the configuration spaces on a manifold and those on its punctured version are closely related cohomologically. We prove a splitting theorem that is equivariant and mixed-Hodge-theoretic; both are new features in such results. As an application, we determine the generating function for the mixed Hodge numbers of the unordered configuration spaces of a multi-punctured elliptic curve.

math.AG↗