arXiv subjects

Dawsen Hwang

Publications and source records attributed to Dawsen Hwang.

7 recordsLinked to original sources

arXiv2026

LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks

Large Language Models (LLMs) exhibit strong informal mathematical reasoning but struggle to generate mechanically verifiable proofs in formal languages like Lean. We present LEAP, an agentic framework that enables general-purpose foundation models to achieve state-of-the-art performance on automated formal theorem proving. LEAP leverages foundation model capabilities, such as informal reasoning, instruction following, and iterative self-refinement. By decomposing complex problems into smaller units, the system bridges formal proof construction with informal blueprints through continuous interaction with the Lean compiler. To provide a rigorous evaluation beyond increasingly saturated benchmarks, we introduce Lean-IMO-Bench, a benchmark of IMO-style problems formalized in Lean, with short statements yet highly non-routine and multi-step proofs across a wide range of difficulty levels. Empirically, on the latest 2025 Putnam Competition, an annual mathematics competition for undergraduate students in North America, LEAP solves all 12 problems, matching recent breakthroughs by frontier formal mathematical models. On Lean-IMO-Bench, LEAP boosts the one-shot formal solve rate of general-purpose LLMs from below 10% to 70%, notably surpassing the 48% benchmark set by a specialized, gold-medal-caliber IMO system. Furthermore, we demonstrate LEAP's research-level utility by autonomously formalizing complex proofs for open combinatorial challenges, including a verified proof for a key subproblem in Knuth's Hamiltonian decomposition of even-order Cayley graphs.

cs.AI↗

arXiv2026

Aletheia tackles FirstProof autonomously

We report the performance of Aletheia (Feng et al., 2026b), a mathematics research agent powered by Gemini 3 Deep Think, on the inaugural FirstProof challenge. Within the allowed timeframe of the challenge, Aletheia autonomously solved 6 problems (2, 5, 7, 8, 9, 10) out of 10 according to majority expert assessments; we note that experts were not unanimous on Problem 8 (only). For full transparency, we explain our interpretation of FirstProof and disclose details about our experiments as well as our evaluation. Raw prompts and outputs are available at https://github.com/google-deepmind/superhuman/tree/main/aletheia.

cs.AI↗

arXiv2026

Towards Autonomous Mathematics Research

Recent advances in foundational models have yielded reasoning systems capable of achieving a gold-medal standard at the International Mathematical Olympiad. The transition from competition-level problem-solving to professional research, however, requires navigating vast literature and constructing long-horizon proofs. In this work, we introduce Aletheia, a math research agent that iteratively generates, verifies, and revises solutions end-to-end in natural language. Specifically, Aletheia is powered by an advanced version of Gemini Deep Think for challenging reasoning problems, a novel inference-time scaling law that extends beyond Olympiad-level problems, and intensive tool use to navigate the complexities of mathematical research. We demonstrate the capability of Aletheia from Olympiad problems to PhD-level exercises and most notably, through several distinct milestones in AI-assisted mathematics research: (a) a research paper (Feng26) generated by AI without any human intervention in calculating certain structure constants in arithmetic geometry called eigenweights; (b) a research paper (LeeSeo26) demonstrating human-AI collaboration in proving bounds on systems of interacting particles called independent sets; and (c) an extensive semi-autonomous evaluation (Feng et al., 2026a) of 700 open problems on Bloom's Erdos Conjectures database, including autonomous solutions to four open questions. In order to help the public better understand the developments pertaining to AI and mathematics, we suggest quantifying standard levels of autonomy and novelty of AI-assisted results, as well as propose a novel concept of human-AI interaction cards for transparency. We conclude with reflections on human-AI collaboration in mathematics and share all prompts as well as model outputs at https://github.com/google-deepmind/superhuman/tree/main/aletheia.

cs.LG↗

arXiv2026

Semi-Autonomous Mathematics Discovery with Gemini: A Case Study on the Erdős Problems

We present a case study in semi-autonomous mathematics discovery, using Gemini to systematically evaluate 700 conjectures labeled 'Open' in Bloom's Erdős Problems database. We employ a hybrid methodology: AI-driven natural language verification to narrow the search space, followed by human expert evaluation to gauge correctness and novelty. We address 13 problems that were marked 'Open' in the database: 5 through seemingly novel autonomous solutions, and 8 through identification of previous solutions in the existing literature. Our findings suggest that the 'Open' status of the problems was through obscurity rather than difficulty. We also identify and discuss issues arising in applying AI to math conjectures at scale, highlighting the difficulty of literature identification and the risk of ''subconscious plagiarism'' by AI. We reflect on the takeaways from AI-assisted efforts on the Erdős Problems.

cs.AI↗

arXiv2025

Towards Robust Mathematical Reasoning

Finding the right north-star metrics is highly critical for advancing the mathematical reasoning capabilities of foundation models, especially given that existing evaluations are either too easy or only focus on getting correct short answers. To address these issues, we present IMO-Bench, a suite of advanced reasoning benchmarks, vetted by a panel of top specialists and that specifically targets the level of the International Mathematical Olympiad (IMO), the most prestigious venue for young mathematicians. IMO-AnswerBench first tests models on 400 diverse Olympiad problems with verifiable short answers. IMO-Proof Bench is the next-level evaluation for proof-writing capabilities, which includes both basic and advanced IMO level problems as well as detailed grading guidelines to facilitate automatic grading. These benchmarks played a crucial role in our historic achievement of the gold-level performance at IMO 2025 with Gemini Deep Think (Luong and Lockhart, 2025). Our model achieved 80.0% on IMO-AnswerBench and 65.7% on the advanced IMO-Proof Bench, surpassing the best non-Gemini models by large margins of 6.9% and 42.4% respectively. We also showed that autograders built with Gemini reasoning correlate well with human evaluations and construct IMO-GradingBench, with 1000 human gradings on proofs, to enable further progress in automatic evaluation of long-form answers. We hope that IMO-Bench will help the community towards advancing robust mathematical reasoning and release it at https://imobench.github.io/.

cs.CL↗

arXiv2018

Online Resource Allocation under Partially Predictable Demand

For online resource allocation problems, we propose a new demand arrival model where the sequence of arrivals contains both an adversarial component and a stochastic one. Our model requires no demand forecasting; however, due to the presence of the stochastic component, we can partially predict future demand as the sequence of arrivals unfolds. Under the proposed model, we study the problem of the online allocation of a single resource to two types of customers, and design online algorithms that outperform existing ones. Our algorithms are adjustable to the relative size of the stochastic component, and our analysis reveals that as the portion of the stochastic component grows, the loss due to making online decisions decreases. This highlights the value of (even partial) predictability in online resource allocation. We impose no conditions on how the resource capacity scales with the maximum number of customers. However, we show that using an adaptive algorithm---which makes online decisions based on observed data---is particularly beneficial when capacity scales linearly with the number of customers. Our work serves as a first step in bridging the long-standing gap between the two well-studied approaches to the design and analysis of online algorithms based on (1) adversarial models and (2) stochastic ones. Using novel algorithm design, we demonstrate that even if the arrival sequence contains an adversarial component, we can take advantage of the limited information that the data reveals to improve allocation decisions. We also study the classical secretary problem under our proposed arrival model, and we show that randomizing over multiple stopping rules may increase the probability of success.

cs.DS↗

arXiv2014

Distributed Multi-Depot Routing without Communications

We consider and formulate a class of distributed multi-depot routing problems, where servers are to visit a set of requests, with the aim of minimizing the total distance travelled by all servers. These problems fall into two categories: distributed offline routing problems where all the requests that need to be visited are known from the start; distributed online routing problems where the requests come to be known incrementally. A critical and novel feature of our formulations is that communications are not allowed among the servers, hence posing an interesting and challenging question: what performance can be achieved in comparison to the best possible solution obtained from an omniscience planner with perfect communication capabilities? The worst-case (over all possible request-set instances) performance metrics are given by the approximation ratio (offline case) and the competitive ratio (online case). Our first result indicates that the online and offline problems are effectively equivalent: for the same request-set instance, the approximation ratio and the competitive ratio differ by at most an additive factor of 2, irrespective of the release dates in the online case. Therefore, we can restrict our attention to the offline problem. For the offline problem, we show that the approximation ratio given by the Voronoi partition is m (the number of servers). For two classes of depot configurations, when the depots form a line and when the ratios between the distances of pairs of depots are upper bounded by a sublinear function f(m) (i.e., f(m) = o(m)), we give partition schemes with sublinear approximation ratios O(log m) and Θ(f(m)) respectively. We also discuss several interesting open problems in our formulations: in particular, how our initial results (on the two deliberately chosen classes of depots) shape our conjecture on the open problems.

cs.DC↗