Searcharxiv⌕ Search

arXiv subjects

QuocViet Pham

Publications and source records attributed to QuocViet Pham.

2 recordsLinked to original sources

LongTails: Automated Research in Optimization

We introduce LongTails, a framework for discovering and improving oracle-complexity bounds with formal verification in Lean 4. Using LongTails, we have already obtained more than 100 new results across a broad range of optimization problems, including generalized smoothness, heavy-tailed noise, finite-sum and composite optimization, and higher-order and zeroth-order methods. Our goal is to scale discovery and continually develop new methods with non-trivial and valuable theoretical guarantees using modern agentic systems. The framework is specialized for optimization, where we separate a Setup, specifying the problem class, assumptions, available information, and success criterion, from a Complexity, specifying a sufficient oracle-call budget. Once a Setup is fixed (e.g., $L$-smooth convex optimization with the goal of finding $x_{\mathrm{out}}$ such that $f(x_{\mathrm{out}}) - f(x^\star) \leq \varepsilon$), agents search for better Complexities (e.g., $O(\sqrt{L R^2 / \varepsilon})$, where $R \geq \|x_0 - x^\star\|$) by designing algorithms and proving their convergence. The framework then checks the proofs and guarantees that every accepted Complexity can indeed be attained under the given Setup. This turns the search for new methods into an ongoing effort to improve their theoretical guarantees. To the best of our knowledge, LongTails is the first framework to scale automated research in optimization across such a broad range of problems with formally verified theoretical guarantees.

math.OC↗

TheoremBench: Evaluating LLMs on Theorem Proving in Formal Mathematics

LLMs have recently achieved strong results on formal proving benchmarks. However, existing evaluations remain heavily concentrated on competition-style problems and often fail to capture how models behave on longer, more dependency-rich mathematical developments. We introduce TheoremBench, a Lean4 benchmark designed to evaluate theorem provers beyond contest settings. The benchmark is built from nearly one hundred classical theorems and is released in two complementary forms: a plain main version containing one target theorem per instance, and a premised version that expands each theorem into a structured family of related proving tasks consisting of the main theorem together with automatically extracted supporting subtheorems. This design enables evaluation of not only whether the final theorem was proved from scratch, but also of partial progress through the internal proof structure of a theorem. Our experiments show that explicit premises substantially improve performance for Lean4-capable prover models. To provide a comprehensive evaluation, we introduce theorem-level coverage and token-efficiency metrics that expose qualitative differences in proof behavior. The results show that current provers remain strongly biased toward easy subtheorems and often solve theorems through long and inefficient tactic traces rather than compact proof plans. TheoremBench therefore provides a more fine-grained view of formal reasoning ability and highlights the importance of structural benchmark design for evaluating Lean4 theorem provers.

cs.AI↗