arXiv subjects

Eric Li

Publications and source records attributed to Eric Li.

At least 19 recordsLinked to original sources

arXiv2026

The Godsil--McKay Asymptotic for Latin Rectangles in the Sublinear Range of Erd\H{o}s Problem 725

Erd\H{o}s Problem 725 asks for an asymptotic formula for the number $L_{k,n}$ of ordered, labelled $k\times n$ Latin rectangles. Godsil and McKay proved that $L_{k,n}\sim (n!)^k((n)_k/n^k)^n(1-k/n)^{-n/2}e^{-k/2}$ for $k=o(n^{6/7})$. We provide a partial solution to Erd\H{o}s Problem 725 by proving this asymptotic for every $k=o(n)$. More precisely, set $\widetilde A_{k,n}=(n!)^k((n)_k/n^k)^n\exp\{[n(H_n-H_{n-k})-k]/2\}$. For every $K(n)=o(n)$, uniformly for $0\leq k\leq K(n)$, we prove $\log(L_{k,n}/\widetilde A_{k,n})=O(k^2/n^2)$, with an absolute implied constant. The results of this paper have been formally verified in Lean.

math.CO↗

arXiv2026

A Resolution of Erd\H{o}s Problem 731 under Dyadic Regularity

We resolve Erd\H{o}s Problem 731 under the explicit dyadic-regularity formalization of "reasonable." Let $A(n)$ be the least positive integer not dividing $\binom{2n}{n}$. On dyadic intervals $X\le n<2X$, put $L=\log(2X)$ and $F_X=\sqrt{2}(\log 2)^{1/4}L^{1/4}\exp\sqrt{(\log 2)L}$. Uniformly for $1\le z\le Z(X)=o(L^{1/4})$, we prove $\mathbb{P}_X(A(n)\le F_X\exp(-z))\asymp \exp(-2z)$ and $\mathbb{P}_X(A(n)>F_X\exp(z))\ll \exp(-2z)$. Consequently $\log A(n)=\sqrt{(\log 2)\log n}+\frac{1}{4}\log\log n+O_{\mathrm{dens}}(1)$. We also prove dyadic nonconcentration: no scalar center on a large dyadic block, and hence no dyadically regular deterministic scale $f$, can satisfy $A(n)/f(n)\to 1$ in natural density. The proof retains the exact least-common-multiple divisibility condition and replaces heuristic cross-base independence by a moving-base restricted-digit variance theorem. The resolution proved here has been formally verified in Lean.

math.NT↗

arXiv2026

A resolution of Erd\H{o}s Problem 1061 on the sum-of-divisors function

We resolve Erd\H{o}s Problem 1061, the question whether the number \[ S(x)=\#\{(a,b)\in\mathbb{N}^2:a+b\le x, \ \sigma(a)+\sigma(b)=\sigma(a+b)\} \] of ordered solutions has a linear asymptotic $S(x)\sim cx$. In fact the opposite extreme holds at every fixed logarithmic scale: for every $R>0$, \[ \lim_{x\to\infty}\frac{S(x)}{x(\log x)^R}=+\infty. \] The construction begins with three integers having the same abundancy index and reduces the divisor-sum identity to two equations in six primes. After a linear change of variables, these equations lie on a split quadric. A three-parameter rational ruling of the quadric supplies many affine systems of six linear forms. An exact lattice-index calculation, an elementary codimension-two parameter sieve, and Bienvenu's higher-dimensional Siegel--Walfisz theorem give prime points uniformly on these planes. Coprime multiplier amplification then yields the stated resolution.

math.NT↗

arXiv2026

A Resolution of Erd\H{o}s Problem 768: the Sylow Divisor Condition

We resolve Erd\H{o}s Problem 768. Let $A(x)$ count the positive integers $n\le x$ such that, for every prime $p\mid n$, there is a divisor $d>1$ of $n$ with $d\equiv 1 \pmod p$. Erd\H{o}s asked whether $A(x)/x=\exp(-(c+o(1))\sqrt{\log x}\log\log x)$ for some constant $c>0$. We prove that this holds with $c=1/(2\sqrt{\log 2})$; equivalently, $\log(x/A(x))/(\sqrt{\log x}\log\log x)$ tends to $1/(2\sqrt{\log 2})$. The lower bound is obtained from primes in disjoint logarithmic intervals using a fourth-moment argument based on the multiplicative large sieve and a subset-product second moment. The upper bound uses canonical witness divisors, a deterministic compression map, an injective reconstruction theorem for its fibers, and growing divisor moments. Thus the paper determines the exact leading constant in Erd\H{o}s Problem 768. The main theorem and its complete proof have been formally verified in the Lean 4 proof assistant.

math.NT↗

arXiv2026

A Resolution of Erd\H{o}s Problems 593 and 1177: Obligatory Triple Systems and Exact Spectra

We resolve Erd\H{o}s Problems #593 and #1177. Problem #593 asks which finite triple systems occur in every uncountably chromatic triple system; the answer is exactly the class generated from private-vertex expansions of finite bipartite graphs by finite disjoint unions and one-point amalgamations. Equivalently, after isolated vertices are removed, a finite triple system is obligatory precisely when it is linear, every hyperedge-node of its Levi graph has an incident bridge, and every Berge cycle is even. The proof uses an exact bridge-trace theorem for complete-rank one-apex sequence lifts. We also prove that, for every uncountable cardinal kappa, there is a linear triple system of chromatic number exactly kappa, with at most 2^{2^mu} vertices when kappa=mu^+. These two ingredients give a class-valued exact avoidance-spectrum dichotomy for every finite forbidden triple system. As a consequence, Erd\H{o}s Problem #1177 has truth values yes, no, and yes. All results of this paper have been formally verified in Lean-4.

math.CO↗

arXiv2026

A Resolution of Erd\H{o}s Problem 550 on Tree versus Complete Multipartite Ramsey Numbers

We resolve Erd\H{o}s Problem 550, originally asked as question (2) of Erd\H{o}s, Faudree, Rousseau, and Schelp. Precisely, for fixed integers $k\geq 2$ and $1\leq m_1\leq \cdots \leq m_k$, we prove that, for every sufficiently large $n$ and every $n$-vertex tree $T$, $R(T,K_{m_1,\ldots,m_k}) \leq (k-1)(R(T,K_{m_1,m_2})-1)+m_1$. The proof combines an off-Tur\'an tree-embedding theorem, proved by regularity and whole-edge allocation, with a compactness theorem for bounded-rank hypergraph obstructions. The full and unconditional proof has been formally verified in Lean.

math.CO↗

arXiv2026

Prime-Power Rarefaction and a Density-One Lower Bound for Erd\H{o}s Problem 400

For fixed $k\ge 2$, let $g_k(n)$ be the greatest excess $a_1+\cdots+a_k-n$ among positive integers $a_i$ satisfying $a_1!\cdots a_k!\mid n!$. We prove that, for every $\varepsilon>0$, all but $o(x)$ integers $n\le x$ satisfy \[ g_k(n)\ge \left(\frac{3(k-1)}{\log 12}-\varepsilon\right)\log n. \] We also prove, as $n\to\infty$, the pointwise upper bound \[ g_k(n)\le (k-1)\log_2 n+\log_2\log n+O_k(1). \] The central analytic input is uniform phase separation for one or two frequencies on fixed-prime $S$-unit progressions, deduced directly from the finite exceptional-subspace alternative of Drmota and Spiegelhofer, and the resulting uniform digit-sum normal-order theorem. A mixed $2$--$3$ representation, quantitative two-block estimates, and a large-prime Kummer sieve produce the stated coefficient.

math.NT↗

arXiv2026

Rank Amplification for Shifted Equal Values of Euler's Totient Function

Let $S_h^\varphi(x)$ denote the number of integers $n\le x$ for which $\varphi(n)=\varphi(n+h)$. For the unit shift, we prove $S_1^\varphi(x)\ll x\exp{-(1/2-o(1))\sqrt{\log x,\log_2 x}}$. More generally, put $A=\log_3 x+\log_4 x-\log 2$, $G=\sqrt{\log x,A}$, and $V=\log x/G$. For every fixed integer $J\ge 1$, uniformly for $1\le h\le \exp{G/\sqrt{J}}$, we obtain $S_h^\varphi(x)=D_{h,>Y_J}^\varphi(x)+O_J(x\exp{-\sqrt{J},G+o_J(V)})$, where $Y_J=\exp{\sqrt{J},G}$. Here $D_{h,>Y_J}^\varphi(x)$ is the above-cutoff part of the classical Graham--Holt--Pomerance same-support family; it is empty for odd $h$. A moving choice $J\asymp \log_2 x/\log_3 x$ gives the unit-shift estimate and an analogous decomposition for a uniform range of shifts. The proof combines the smooth-totient theorem of Banks--Friedlander--Pomerance--Shparlinski with labelled supplier systems, a shifted divisor convolution, and an injective encoding of large supplier products into weighted friable tuples. The results of this paper have been formally verified in Lean.

math.NT↗

arXiv2026

The Erd\H{o}s-Hajnal High-Girth Subgraph Conjecture Holds in the Polynomial Chromatic-Sparsity Regime

For a graph $G$ put $h_r(G)=\max{\chi(H):H\subseteq G,\operatorname{girth}(H)\ge r}.$ Erd\H{o}s and Hajnal asked whether $h_r(G)\to\infty$ as $\chi(G)\to\infty$, for every fixed $r\ge4$. We prove this in every fixed polynomial edge-density regime: for all $r\ge4$, $k\ge2$, $P,C>0$, there is $M=M_{r,k}(P,C)$ such that $\chi(G)\ge M,\ e(G)\le C\chi(G)^P\Longrightarrow h_r(G)\ge k.$ Quantitatively, after replacing $P$ by $P\vee2$ and $C$ by $C\vee2$, $M_{r,k}(P,C)\le \exp!\left(O_{r,k}\bigl((P+2+\log(C\vee2))^2\bigr)\right),$ and consequently the same conclusion holds throughout the quasi-polynomial range $e(G)\le \exp\bigl(C_0(\log\chi(G))^a\bigr),\ 1 < a < 3/2,$ for all sufficiently large $\chi(G)$. In each fixed polynomial-density regime we also obtain $f_{P,C}(k,r)\le k^{O_{r,P,C}(1)}.$ The proof combines a chromatic-defect random extraction lemma, compact and near-quadratic sparse-core bases, and a peeling/thinning bootstrap increasing the admissible edge exponent by $1/(r-1)$. We also prove structural saturation results for possible counterexamples, including Moore-strength exact-cycle packings and quadratic saturation in projected colour-pair space. Finally, writing $h_r^{\mathrm f}(G)=\max{\chi_{\mathrm f}(H):H\subseteq G,\operatorname{girth}(H)\ge r},$ we develop a fractional random-extraction framework based on Mohar-Wu preservation. We prove sufficient cheap-cycle-killing criteria and verify them for several structured families, including clique-organised families, line graphs of incidence graphs of equal-order generalized quadrangles and generalized hexagons, and the Bohman-Keevash tracking-time triangle-free-process graph. We also isolate a density-free obstruction that any proof using this fractional surgery route must overcome.

math.CO↗

arXiv2026

Square-Annular Dynamics and Coalescence Frontiers for $n+\tau(n)$

Let $T(n)=n+\tau(n)$, where $\tau$ is the divisor function. We study the Erdos-Graham coalescence problem by encoding finite-level obstructions in the divisor-successor graph and in square-annular transfer maps. Coalescence is equivalent both to connectedness of this graph and to synchronization along an infinite non-autonomous sequence of finite annular systems. The basic identities are \[ \operatorname{im}(\mathcal A_k)=E_{k+1}, \qquad \mathcal F_{k^2}=k^2+E_k, \] where $E_k$ is the set of square-crossing overshoots from below $k^2$. We prove a transfer parity law, dynamic frontier bounds for the widths $W_{k,s}$, and the criterion that $\liminf_k|\mathcal A_k(E_k)|=1$ would imply connectedness. Unconditionally, \[ R(X)\le \log X+2\gamma+O(X^{-1/4}), \] and the exit sets are residue-universal, satisfy $|E_k|\le k^{o(1)}$, and obey \[ \frac94K+O(1)\le \sum_{k\le K}|E_k|\ll K(\log K)^3. \] Using the shifted-square estimate HST, obtained from the corrected Henriot--Nair--Tenenbaum theorem in the specialized form of Proposition 8.4 and from separate square-shift estimates, we obtain fixed-moment bounds \[ \sum_{k\le K}|E_k|^m\ll_m K(\log K)^{C_m}\quad(m\ge2). \] A further first-moment refinement to $K(\log K)^2$ is conditional on the additional, currently unproved, uniform quadratic Euler-product mean-value hypothesis HQE. We also prove quantitative large-jump and lower-runner race theorems, isolate interval filling, and formulate a square-gated two-branch criterion. No proof of the full Erdos-Graham problem is claimed.

math.NT↗

arXiv2026

An Average-Order Theorem for a Shifted Pairwise-Coprime Extremal Problem

For $n\ge 2$, let $M(n)$ be the supremum of $\sum_{a\in A}1/(n-a)$ over pairwise coprime sets $A\subset[1,n)$. Erd\H{o}s asked whether $M(n)\le \sum_{p d\}$ with a bounded-cost dual certificate and Buchstab--de Bruijn estimates. We also show $M(n)=(e^{-\gamma}+o(1))\log\log n$ for almost all $n$, with a quantitative exceptional-set bound, so Erd\H{o}s's inequality holds for almost all $n$. The argument combines a long-interval two-dimensional beta sieve for two moving forbidden residue classes with an exact finite singular-series cancellation. The remaining obstacles to a full variance estimate are a shifted-prime second moment for $d>\sqrt{2N}$ and finite-$u$ rough correlations in the intermediate tail, yielding a precise conditional criterion. For every $\varepsilon>0$ we also prove the uniform pointwise bound $M(n)\le(2+\varepsilon)\log\log n+O_{\varepsilon}(1)$ and explain the linear-sieve barrier at the constant $2$. The results of this paper have been formally verified in Lean.

math.NT↗

arXiv2026

Erd\H{o}s Problem 684 at Density One: Small-prime Parts of Binomial Coefficients and Gaussian Fluctuations

For $0\leq k\leq n$, let $u(n,k)$ be the largest divisor of $\binom nk$ whose prime factors are at most $k$. Erd\H{o}s Problem #684 concerns the special threshold $u(n,k)>n^2$ and asks how early this small-prime part can be forced to become large. We prove the density-one analogue for every fixed power threshold. If $f_c(n)$ is the least $k$ for which $u(n,k)>n^c$, then, for each fixed $c>0$, \[ f_c(n)=\left(\frac{c}{1-\gamma}+o(1)\right)\log n \] for almost all positive integers $n$. In particular, \[ f_2(n)=\left(\frac{2}{1-\gamma}+o(1)\right)\log n =(4.730544237\ldots+o(1))\log n \] for the Erd\H{o}s #684 threshold. This is a normal-order theorem, not a pointwise resolution of the corresponding worst-case problem. The constant $1-\gamma$ is arithmetic. Kummer's theorem rewrites $\log u(n,k)$ as a sum of carry indicators, and complete-residue averaging gives \[ m(k)=k\sum_{p\leq k}\frac{\log p}{p-1}-\log k!=(1-\gamma)k+o(k). \] The cancellation in this formula moves the typical crossing from the naive scale $c\log n$ to $c(1-\gamma)^{-1}\log n$. We prove the required concentration uniformly for every $k\leq A\log X$ on one dyadic interval, after discarding a zero-density exceptional set caused by large powers of small primes dividing one of the nearby integers $n,n-1,\ldots$. We also prove Gaussian fluctuations in the logarithmic range. If $k=k(X)\to\infty$, $k\leq A\log X$, and $n$ is uniform in $[X,2X)\cap\mathbb Z$, then \[ \frac{\log u(n,k)-m(k)}{\sqrt{V(k)}}\Rightarrow \mathcal N(0,1), \qquad V(k)\sim (2-\log(2\pi))k\log k. \] Higher prime powers are needed for the mean, but after centering their aggregate is $L^2$-negligible on the Gaussian scale; the variance comes only from the prime levels.

math.NT↗

arXiv2026

On the Duke--Erd\H{o}s--R\"odl Problem at the One-Third Threshold

Let $G$ be an $n$-vertex graph with $e(G)\ge n^2/k$. We prove a self-contained internal short-cycle core theorem at the threshold $k\le n^{1/3}$: the graph $G$ contains a subgraph $H_6$ with $\Omega(n^2/k^3)$ edges in which every two distinct edges lie together on a cycle of length at most $6$ contained in $H_6$, and a subgraph $H_8$ with $\Omega(n^2/k^2)$ edges in which every two distinct edges lie together on a cycle of length at most $8$ contained in $H_8$. In density notation $\rho=e(G)/n^2$, this gives internal cores of sizes $\Omega(\rho^3n^2)$ and $\Omega(\rho^2n^2)$ throughout the range $\rho\ge n^{-1/3}$. The $C_{\le6}$ conclusion above is an edge-connected statement and does not impose the adjacent-edge $C_4$ condition appearing in the strongest Duke--Erd\H{o}s--R\"odl formulation. We also include two complementary results clarifying this distinction. First, under the ambient-witness convention, every graph with at least $n^2/k$ edges and $k=o(n^{1/2})$ contains $\Omega(n^2/k^3)$ selected edges whose pairs are witnessed by ambient cycles of length at most $6$, with adjacent pairs witnessed by ambient $C_4$'s. Second, under the standard internal strong $C_6$ convention, for every fixed $\beta\in[1/3,1/2)$ there is an infinite sequence of bipartite graphs $G$ with $n\to\infty$ and $e(G)=\Theta_\beta(n^{2-\beta})$ such that every internally strongly $C_6$-connected subgraph has only $O_\beta(\rho(G)^3n^2/(\log n)^2)$ edges. The obstruction is a random cyclic shift-lift of $K_{q,q}$, together with an occupancy estimate excluding large aligned two-covers.

math.CO↗

arXiv2026

GenMatter: Perceiving Physical Objects with Generative Matter Models

Human visual perception offers valuable insights for understanding computational principles of motion-based scene interpretation. Humans robustly detect and segment moving entities that constitute independently moveable chunks of matter, whether observing sparse moving dots, textured surfaces, or naturalistic scenes. In contrast, existing computer vision systems lack a unified approach that works across these diverse settings. Inspired by principles of human perception, we propose a generative model that hierarchically groups low-level motion cues and high-level appearance features into particles (small Gaussians representing local matter), and groups particles into clusters capturing coherently and independently moveable physical entities. We develop a hardware-accelerated inference algorithm based on parallelized block Gibbs sampling to recover stable particle motion and groupings. Our model operates on different kinds of inputs (random dots, stylized textures, or naturalistic RGB video), enabling it to work across settings where biological vision succeeds but existing computer vision approaches do not. We validate this unified framework across three domains: on 2D random dot kinematograms, our approach captures human object perception including graded uncertainty across ambiguous conditions; on a Gestalt-inspired dataset of camouflaged rotating objects, our approach recovers correct 3D structure from motion and thereby accurate 2D object segmentation; and on naturalistic RGB videos, our model tracks the moving 3D matter that makes up deforming objects, enabling robust object-level scene understanding. This work thus establishes a general framework for motion-based perception grounded in principles of human vision.

cs.CV↗

arXiv2026

SceneTransporter: Optimal Transport-Guided Compositional Latent Diffusion for Single-Image Structured 3D Scene Generation

We introduce SceneTransporter, an end-to-end framework for structured 3D scene generation from a single image. While existing methods generate part-level 3D objects, they often fail to organize these parts into distinct instances in open-world scenes. Through a debiased clustering probe, we reveal a critical insight: this failure stems from the lack of structural constraints within the model's internal assignment mechanism. Based on this finding, we reframe the task of structured 3D scene generation as a global correlation assignment problem. To solve this, SceneTransporter formulates and solves an entropic Optimal Transport (OT) objective within the denoising loop of the compositional DiT model. This formulation imposes two powerful structural constraints. First, the resulting transport plan gates cross-attention to enforce an exclusive, one-to-one routing of image patches to part-level 3D latents, preventing entanglement. Second, the competitive nature of the transport encourages the grouping of similar patches, a process that is further regularized by an edge-based cost, to form coherent objects and prevent fragmentation. Extensive experiments show that SceneTransporter outperforms existing methods on open-world scene generation, significantly improving instance-level coherence and geometric fidelity. Code and models will be publicly available at https://2019epwl.github.io/SceneTransporter/.

cs.CV↗

arXiv2025

Skywork-R1V4: Toward Agentic Multimodal Intelligence through Interleaved Thinking with Images and DeepResearch

Despite recent progress in multimodal agentic systems, existing approaches often treat image manipulation and web search as disjoint capabilities, rely heavily on costly reinforcement learning, and lack planning grounded in real tool-execution traces. To address these limitations, we present Skywork-R1V4, a 30B (A3B) parameter multimodal agentic model that unifies multimodal planning, active image manipulation ("thinking with images"), deep multimodal search, and, most critically, interleaved reasoning that dynamically alternates between visual operations and external knowledge retrieval. Trained solely via supervised fine-tuning on fewer than 30,000 high-quality, planning-execution-consistent trajectories and validated through stepwise consistency filtering, Skywork-R1V4 achieves state-of-the-art results across perception and multimodal search benchmarks: it scores 66.1 on MMSearch and 67.2 on FVQA, surpassing Gemini 2.5 Flash on all 11 metrics. Skywork-R1V4 exhibits emergent long-horizon reasoning at inference time, successfully orchestrating more than 10 tool calls to solve complex, multi-step tasks. Our results demonstrate that sophisticated agentic multimodal intelligence can be achieved through carefully curated supervised learning alone, without any reliance on reinforcement learning.

cs.CV↗

arXiv2025

PanoWorld-X: Generating Explorable Panoramic Worlds via Sphere-Aware Video Diffusion

Generating a complete and explorable 360-degree visual world enables a wide range of downstream applications. While prior works have advanced the field, they remain constrained by either narrow field-of-view limitations, which hinder the synthesis of continuous and holistic scenes, or insufficient camera controllability that restricts free exploration by users or autonomous agents. To address this, we propose PanoWorld-X, a novel framework for high-fidelity and controllable panoramic video generation with diverse camera trajectories. Specifically, we first construct a large-scale dataset of panoramic video-exploration route pairs by simulating camera trajectories in virtual 3D environments via Unreal Engine. As the spherical geometry of panoramic data misaligns with the inductive priors from conventional video diffusion, we then introduce a Sphere-Aware Diffusion Transformer architecture that reprojects equirectangular features onto the spherical surface to model geometric adjacency in latent space, significantly enhancing visual fidelity and spatiotemporal continuity. Extensive experiments demonstrate that our PanoWorld-X achieves superior performance in various aspects, including motion range, control precision, and visual quality, underscoring its potential for real-world applications.

cs.CV↗

arXiv2025

Comparing RAG and GraphRAG for Page-Level Retrieval Question Answering on a Math Textbook

Large language models (LLMs) show promise as educational aids but often lack alignment with specific course materials. We investigate Retrieval-Augmented Generation (RAG) and GraphRAG for page-level question answering on an undergraduate mathematics textbook. Using a curated dataset of 477 question-answer pairs, each tied to a specific textbook page, we compare five embedding-based RAG models, a BM25 baseline, and GraphRAG across two metrics: retrieval accuracy (whether the correct page is retrieved) and answer quality (F1 score). Our results show that embedding-based RAG outperforms GraphRAG for page-level retrieval, with voyage-3-large achieving 99.4% accuracy at top-10 (bootstrap 95% CI for top-1: [.644, .728]). BM25 proves a strong baseline, outperforming several embedding models. Error analysis reveals that 63.3% of top-1 failures retrieve same-chapter content, suggesting pedagogical relevance even in failure cases. GraphRAG retrieves excessive context (~47K tokens vs. ~3.7K for RAG), reducing generation quality. We further replicate key experiments using an open-source local LLM (Qwen3.5-35B-A3B), finding that RAG benefits are proportionally larger for weaker models (+39% vs. +16% relative F1 improvement), an important result for cost-sensitive educational deployments. These findings inform the design of AI tutoring systems that reference specific textbook pages.

cs.IR↗