Searcharxiv⌕ Search

arXiv subjects

Guoqing Wang

Publications and source records attributed to Guoqing Wang.

At least 19 recordsLinked to original sources

Learning Agent-Centric Vision-Language-Action World Models for Autonomous Driving

Recent progress in integrating vision-language-action (VLA) models with world modeling has significantly advanced end-to-end autonomous driving by allowing systems to reason about imagined futures instead of merely reacting to current observations. However, most world VLA models generate future frames at the pixel level with plenty of dense visual tokens. Although such generated futures can be visually plausible, they preserve many driving-irrelevant appearance details, i.e., only a small portion of the tokens directly encode critical planning-relevant factors, such as agent geometry, motion dynamics, and interactions. This introduces a significant performance bottleneck. An important question thus arises: what world state should a world VLA model reason over to support planning beyond pixel-level prediction? We argue that planning-oriented world modeling should focus on a compact and structured agent-centric latent state that captures the interactions between objects in the driving scene over time. As such, we introduce AgentWorld, an agent-centric VLA world model that uses discrete agent tokens jointly for reasoning and trajectory planning. Each token is dynamically grounded to a surrounding agent associated with its object's state, such as semantic identity, 3D geometry, motion, and interaction context. AgentWorld learns these tokens through three key processes: explicit geometric grounding, reasoning-oriented supervised fine-tuning, and agent-oriented reinforcement learning. Extensive open-loop and closed-loop experiments under diverse driving scenarios show that AgentWorld enhances planning capabilities, fully exploits visual evidence and agent dynamics, and yields interpretable agent-centric world states.

cs.CV↗

Zero-sum Ramsey numbers of forests, cycles and complete bipartite graphs

For a graph $G$ and $m\ge2$ dividing $e(G)$, the zero-sum Ramsey number $R(G,\mathbb Z_m)$ is the least $N$ such that every $\mathbb Z_m$-labeling of $E(K_N)$ contains a copy of $G$ whose edge labels sum to zero. We disprove the Bialostocki--Dierker asymptotic conjecture for forests by constructing trees for every sufficiently large edge count whose ratio of zero-sum to two-color Ramsey numbers has lower limit at least $554/553>1$. For powers of two $q\ge2$, we characterize all $s,k\ge1$ satisfying $R(K_{s,kq},\mathbb Z_q)=s+kq$ and prove $R(mK_{q,q},\mathbb Z_m)=2qm+m/\gcd(m,q)-1$ for every $m\ge2$. The packing upper bound combines Caro's transfer inequality with our spanning result and sharpens earlier bounds when $\gcd(m,q)\ge4$. The formula also gives exact values for counterexamples to Caro's defect conjecture, whose failure was already implicit in earlier work of Caro and Caro--Provstgaard. For every integer $m\ge2$ and every multiple $\ell$ of $m$ with $\ell\ge100m^3$, we also prove $R(C_\ell,\mathbb Z_m)=\ell+m/\gcd(m,2)-1$ for the cycle $C_\ell$ on $\ell$ vertices, determining the eventual value left by the bounds of Chi and He for even moduli $m\ge4$.

math.CO↗

An asymptotic resolution of the Gao--Thangadurai conjecture on the Harborth constant

The Harborth constant $\mathsf g(G)$ of a finite abelian group $G$ of exponent $n$ is the least integer $k$ such that every subset of $G$ of size at least $k$ contains $n$ elements whose sum is zero. Let $C_n$ be a cyclic group of order $n$. Gao and Thangadurai \cite{gt} in 2004 conjectured that $\mathsf g(C_n\oplus C_n)=2n+(-1)^n$ for all integers $n\ge3$. We prove this conjecture for all sufficiently large integers $n$.

math.CO↗

Spin Interferometry Generated by Cavity Memory

We report the control and observation of coherent spin interferometry generated by cavity memory using an ensemble of nitrogen-vacancy (NV) centers coupled to a microwave cavity. The abrupt switch-on and switch-off of a detuned drive produce transient fields at the cavity frequency that supply the two spin rotations for Ramsey interferometry, observed using both microwave and fluorescence readout. Spectrally scanning the spin profile across the cavity and drive frequencies distinguishes cavity-generated Ramsey interference from conventional Rabi dynamics. Temporal weighting of the reflected microwave signal enables spectrally selective population readout at a single probe frequency. By programming the phase and amplitude of the drive envelope, we extend the same mechanism to more quantum control sequences, including spin-echo and Rabi control.

quant-ph↗

Infinite Generalized Sidon Sets with a Negative Answer to a Problem of Szemerédi

Let $A(x)=|A\cap[1,x]|$. For every integer $h\ge3$, every finite $B_h$ set $A_0\subseteq\mathbb N$, and every $0<c\le h^{-2/h}$, we construct an infinite $B_h$ set $A\subseteq\mathbb N$ containing $A_0$ such that $A(x)/x^{1/h}$ has upper limit $c$ and lower limit zero. This extends the range $0<c\le1/h$ obtainable from our first construction, which answered a question of Cilleruelo and Nathanson, later raised by O'Bryant, on the existence of a positive upper limit. For $h=3$, a positive upper limit disproves the assertion $A(x)=o(x^{1/3})$ in Szemerédi's Problem~186(d). The improved range follows from an integer lift of a modular construction of Gómez Ruiz and Trujillo Solarte, with large gaps between sums of distinct $r$-element multisets for each fixed $1\le r<h$. These gaps ensure that successive blocks can be adjoined by translation while preserving the $B_h$ property. The same prescribed limits can be attained inside every set of asymptotic density one containing $A_0$. Separately, we construct $2^{\aleph_0}$ extensions of $A_0$ that are maximal among all $B_h$ subsets of $\mathbb N$, have the prescribed limits, and avoid any prescribed finite subset of $\mathbb N\setminus A_0$. These maximal extensions can be chosen so that $\sum_{a\in A}a^{-1/h}$ diverges arbitrarily slowly.

math.CO↗

Sharp enumeration and growth of admissible sets of integers

A set of positive integers is admissible if equal sums of finite subsets always have equal numbers of summands. Let $a(N)$ count the nonempty admissible subsets of $\{1,\ldots,N\}$, and let $b(N)$ count those with maximum $N$. We prove $\log a(N)\sim\log b(N)\simπ\sqrt{2N/3}$, establishing the sharp logarithmic asymptotic proposed by Nicolas. More generally, let $Q_h(N)$ count the subsets for which each integer has subset-sum representations of at most $h$ distinct cardinalities. Uniformly for all positive integers $h,N$, we prove $\log Q_h(N)\asymp\min\{N,\sqrt{hN}\}$ with absolute implied constants. Every infinite set satisfying the same condition for some fixed $h$ has counting function $o(\sqrt x)$. At $h=1$, the latter two results settle conjectures of Erdős, Nicolas and Sárközy, strengthening the proposed vanishing lower limit to a full limit. The proofs record each subset sum together with its cardinality and use the counting theorem of Campos and the structure theorem of Carpenter, Defant and Kravitz. The sharp count also uses Nicolas's partition enumeration and an elementary construction of a long restricted-sum progression. Finally, we give an exact partition formula for the odd sets with separated alternate cardinality layers suggested by Nicolas.

math.CO↗

Extremal graphs for Alon's graph coupon-collector conjecture

Let \(G\) be a finite simple graph without isolated vertices. Starting with every vertex uncovered, at each step sample a vertex uniformly and independently and cover one of its uncovered neighbors, chosen uniformly, if any. Let \(\cc(G)\) denote the expected number of steps required to cover all vertices. Alon conjectured that disjoint unions of \(K_{d+1}\) maximize \(\cc(G)\) among \(d\)-regular graphs of a prescribed order whenever such unions exist. In this paper, we disprove this conjecture for every fixed \(d\ge2\) and give a complete characterization of the extremal graphs for \(d=2\) at every order \(n\ge3\). In particular, all but at most six vertices of any such extremal graph lie in four-cycle components.

math.CO↗

Disproving Conjectures on the Davenport and Narkiewicz Constants

The Davenport constant $\DD(G)$ of a finite abelian group $G$ is a classical invariant arising from the arithmetic of ideal class groups of number fields. Closely related to the Davenport constant is the first Narkiewicz constant $\Nark_1(G)$, which arises in the quantitative study of algebraic integers and their factorizations. Gao and Geroldinger conjectured in 2006 that $\DD(G)\leq \sum_{i=1}^r n_i,$ while Narkiewicz and Śliwa conjectured in 1982 that $\Nark_1(G)=\sum_{i=1}^r n_i,$ where $n_1,\ldots,n_r$ are the invariant factors of $G$. Both are central conjectures concerning arithmetic invariants arising from the theory of number fields. In this paper, we disprove both conjectures. Moreover, we show that, for every fixed $r\geq5$, both $\DD(G)-\sum_{i=1}^r n_i$ and $\Nark_1(G)-\sum_{i=1}^r n_i$ are unbounded over finite abelian groups of rank $r$. We develop a probabilistic representation technique to establish a uniform upper bound for the Davenport constant: $$ \DD(G)\leq\frac{16}{5}\,\rr(G)\exp(G). $$ Our bound gives the first improvement of the classical van Emde Boas--Kruyswijk estimate: $\DD(G)\leq\exp(G) \left(1+\log\frac{|G|}{\exp(G)}\right) \leq\exp(G)\bigl(1+(\rr(G)-1)\log\exp(G)\bigr)$, by removing the logarithmic dependence on the exponent from this classical estimate, replacing it with the absolute constant $16/5$. Their bound stood unimproved since its establishment in 1969. The existence of $C=16/5$ also brings us closer to the most challenging conjecture on the Davenport constant, namely the Emde Boas--Kruyswijk Conjecture in 1969, which predicts that $\DD(C_n^r)=r(n-1)+1.$ As a further application of our results, we disprove a conjecture of Girard in 2008 concerning cross numbers, an invariant with deep roots in the arithmetic of number fields and Krull monoids.

math.CO↗

Submodular decompositions: a bounded-variation counterexample and an exact planar cut formula

Bérczi, Gehér, Imolay, Lovász, and Schwarcz (SIAM J. Discrete Math. 40(2026), 308--335) asked whether every set function of bounded variation is a difference of two bounded submodular functions, and how to determine the optimum in a monotonic sum-decomposition of a planar graph cut function. In this paper, we answer both questions and obtain quantitative extensions. For the first question, we characterize the continuous profiles $f:[0,1]\to\mathbb R$ for which $A\mapsto f(m(A))$ is a bounded submodular difference on an atomless probability space. More precisely, we express the minimum oscillation of the subtracted component through the positive part of the distributional second derivative. We also characterize admissible discontinuous profiles and give an exact summability criterion for countable block functions. In particular, the first question has a negative answer even for increasing functions of variation one whose scalar profiles have log-Lipschitz derivatives and belong to every $C^{1,α}$, $0<α<1$. Turning to graph cuts, we prove that the optimum equals the total edge weight minus the maximum fractional triangle-packing weight for every nonnegative weighting if and only if the graph is $K_5$-free. Moreover, maximum packings parametrize the symmetrized optimal components, while on $K_4$-free graphs we describe all optimal components and characterize uniqueness. Consequently, the formula applies to weighted planar graphs and implies edge-weight monotonicity on $K_5$-free graphs. For rational weights, it also yields polynomial-time constructions.

math.CO↗

Quadratic bounds for intersecting hypergraphs: An affirmative answer to a question of Duke and Erdős

Let $m(k,b)$ denote the minimum number of edges in an intersecting $k$-uniform hypergraph in which every $(k-1)$-set of vertices is disjoint from at least $b$ edges, with $m(k,b)=+\infty$ if no such hypergraph exists. Duke and Erdős [Congressus Numerantium (1980)] asked whether $m(k,k)=o(k^4)$ is possible. In this paper, we answer this question affirmatively and determine the order of growth of $m(k,k)$ by proving that $k^2+1\leq m(k,k)\leq Ck^2$ for all sufficiently large $k$, where $C$ is an absolute constant. More generally, we show that $m(k,b)=Θ(kb)$ for all sufficiently large odd $k$ and every integer $b$ with $k\leq b\leq (k-1)(k-3)/2$, where the implied constants are independent of $k$ and $b$.

math.CO↗

The equality $s(G)=η(G)+\exp(G)-1$ fails for $G=C_2^7\oplus C_6$

Let $G$ be a finite abelian group. The Erdős-Ginzburg-Ziv constant $\mathsf{s}(G)$ is the smallest positive integer $\ell$ such that every sequence over $G$ of length at least $\ell$ contains a zero-sum subsequence of length $\exp(G)$. The short zero-sum constant $η(G)$ is defined similarly, with the requirement that the zero-sum subsequence is nonempty and has length at most $\exp(G)$. It is known that $\mathsf{s}(G)\ge η(G)+\exp(G)-1.$ A conjecture proposed by W. Gao in 2003, which has been widely studied in zero-sum theory, predicts that $$ \mathsf{s}(G)=η(G)+\exp(G)-1 $$ for every nontrivial finite abelian group $G$. In this paper, we prove that this conjecture fails for the group $C_2^7\oplus C_6$.

math.CO↗

Extremal zero-sum sequences without elements of maximal order

Let $\mathsf D(G)$ denote the Davenport constant of a finite abelian group $G$, the maximal length of a minimal zero-sum sequence over $G$. Gao and Geroldinger conjectured that every minimal zero-sum sequence of length $\mathsf D(G)$~\cite[Conjecture~6.1]{GG99} and every zero-sumfree sequence of length $\mathsf D(G)-1$~\cite[Conjecture~4.1]{GG06} contains an element of order $\exp(G)$. We disprove both conjectures by proving a density-one theorem: for every fixed finite abelian group $H$ of exponent $M=2^ap^b$, where $a\ge2$, $b\ge1$, and $p$ is an odd prime, the set of integers $r\ge0$ for which $G_r=C_2^r\oplus H$ admits a minimal zero-sum sequence of length $\mathsf D(G_r)$ with no term of order $M$ has natural density one.

math.CO↗

On a classical zero-sum invariant II: Disproof of a long-standing conjecture

For a nontrivial finite abelian group $G$, let $ν(G)$ be the smallest {nonnegative} integer $\ell$ such that every zero-sum free sequence $T$ over $G$ of length at least $\ell$ has the following property: all nonzero elements of $G$ that do not occur as a subsequence sum of $T$ lie in a proper coset of some subgroup of $G$. It is easy to check that $\mathsf d (G)-1 \le ν(G) \le \mathsf d (G)$, where $\mathsf d (G)$ is the small Davenport constant of $G$. A conjecture by Gao from the year 2000 stated that equality should always hold at the lower bound. This conjecture has since been confirmed for many families of groups (including all p-groups and groups of rank at most two). In this paper, we disprove the conjecture.

math.CO↗

Enumeration of zero-sum subsequences: A proof of Bialostocki-Lotspeich conjecture

Let $C_n$ be a cyclic group of order $n\ge 2$, and let $S=g_1\cdot\ldots\cdot g_{\ell}$ be a sequence over $C_n$, with repetitions allowed. We denote by $\mathsf N_0^n(S)$ the number of subsets $I\subseteq[1,\ell]$ with $|I|=n$ and $\sum_{i\in I}g_i=0$. A conjecture proposed by Bialostocki and Lotspeich in 1992 states that $$\mathsf N_0^n(S)\ge \binom{\lfloor |S|/2\rfloor}{n}+ \binom{\lceil |S|/2\rceil}{n}.$$ Several partial results toward this conjecture are known. In this paper, we prove this conjecture completely.

math.CO↗

Quasirandomness from balanced and repeated subgraph counts

Let $F$ be a graph on $m$ vertices with at least one edge, and let $0 1$ or $\balpha\ne\bm/m$. We also obtain a complete classification of labelled transversal counts: for \rev{$0<p<1$, the only nonforcing profile is $F=K_2$ with class proportions $(1/2,1/2)$.} Together, these results settle conjectures of Janson and Sós [European J. Combin. 46 (2015)], Hatami, Hatami and Li [European J. Combin. 56 (2016)], and Janson [arXiv:2606.20186v1 (2026)]. The common ingredient is a balanced-count theorem for every graph with more than one edge, proved through a degree-sequence inequality, and its consequence of additive rigidity. Our arguments also determine the bad indices of the associated functional nullspaces and isolate the exceptional single-edge profiles. In particular, a non-divisibility theorem for grouped elementary symmetric polynomials excludes the higher-order obstructions in the repeated-class setting. In contrast, for two repeated classes and every $p\in(0,1)$, we construct nonquasirandom graph sequences satisfying the crossing-triangle counts at proportions $(1/3,2/3)$, with limiting edge density strictly below $p$. These counterexamples answer the crossing-triangle question negatively and show why the three-class assumption cannot be omitted.

math.CO↗

Inverse Zero-Sum Problems of Erdős--Ginzburg--Ziv constant: Property C and Property D conjectures

Let $C_n^r$ denote the direct sum of $r$ cyclic groups of order $n$, where $n\ge2$ and $r\ge1$. Properties C and D predict a uniform multiplicity structure for extremal zero-sum problems over $C_n^r$: every sequence of maximum length avoiding nonempty zero-sum subsequences of length at most $n$, or of length exactly $n$, respectively, should have the form $T^{n-1}$. We prove that, for every $r\ge4$ and every even $n\ge2$, each of these properties holds if and only if $n$ is a power of two. This gives negative answers to the general Property C problem posed by Gao, Hou, Schmid and Thangadurai Open Problems 1.2{GHST} in 2007, and disproves the Property D conjecture proposed by Gao and Geroldinger [Conjecture 7.2]{GG} in 2006.

math.CO↗

MENTOR: A Metacognition-Driven Self-Evolution Framework for Uncovering and Mitigating Implicit Domain Risks in LLMs

Ensuring the safety of Large Language Models (LLMs) is critical for real-world deployment. However, current safety measures often fail to address implicit, domain-specific risks. To investigate this gap, we introduce a dataset of 3,000 annotated queries spanning education, finance, and management. Evaluations across 14 leading LLMs reveal a concerning vulnerability: an average jailbreak success rate of 57.8\%. In response, we propose MENTOR, a metacognition-driven self-evolution framework. MENTOR performs metacognitive self-assessment, using strategies such as perspective-taking and consequential reasoning to uncover latent model misalignments. MENTOR couples single-pass rule-guided inference for routine requests with a selectively invoked metacognitive evolution cycle that revises residual unsafe responses, distills successful corrections into a dynamic rule graph, and compiles validated rules into activation-level steering signals for future inference. Experiments demonstrate that MENTOR substantially reduces attack success rates across all tested domains and outperforms existing safety alignment methods. The code and dataset for MENTOR are available at: https://anonymous.4open.science/r/MENTOR-Evo.

cs.AI↗

BridgeAfford: Bridging Query-Target Gap for Generalizable 3D Affordance Grounding

Language-guided 3D affordance grounding aims to localize functional regions on 3D objects according to interaction intents. However, existing methods exhibit poor generalization to unseen objects. One challenge is the query-target gap: a semantic discrepancy between query features encoding interaction intents and target-region features, leading to inaccurate grounding. End-to-end training favors query representations tailored to seen objects, limiting their compatibility with shifted target-region features on unseen objects. To address this issue, we propose BridgeAfford, a framework that reduces the query-target semantic gap from two complementary perspectives. First, Interaction Region Alignment (IRA) promotes cross-object alignment among geometrically similar interaction regions during training. This enables structures on unseen objects that resemble previously observed ones to produce representations similar to those learned during training, facilitating target-region recognition. Second, Visual Evidence Aggregation (VEA) leverages a 2D affordance model to refine the attention distribution and guide the query toward the correct region, yielding more target-aware queries. Extensive experiments on the LASO and PIAD datasets demonstrate that BridgeAfford outperforms existing methods, particularly on unseen objects, highlighting its robust generalization ability.

cs.CV↗