[Paper Review] Accelerating Scientific Research with Gemini: Case Studies and Common Techniques

D. P. Woodruff, Vincent Cohen-Addad+34 more|arXiv (Cornell University)|Feb 3, 2026

Machine Learning in Materials Science2 citations

TL;DR

The paper presents case studies of AI-assisted theoretical research using Google Gemini models to solve open problems, refute conjectures, and generate proofs, and distills a playbook of collaboration techniques.

ABSTRACT

Recent advances in large language models (LLMs) have opened new avenues for accelerating scientific research. While models are increasingly capable of assisting with routine tasks, their ability to contribute to novel, expert-level mathematical discovery is less understood. We present a collection of case studies demonstrating how researchers have successfully collaborated with advanced AI models, specifically Google's Gemini-based models (in particular Gemini Deep Think and its advanced variants), to solve open problems, refute conjectures, and generate new proofs across diverse areas in theoretical computer science, as well as other areas such as economics, optimization, and physics. Based on these experiences, we extract common techniques for effective human-AI collaboration in theoretical research, such as iterative refinement, problem decomposition, and cross-disciplinary knowledge transfer. While the majority of our results stem from this interactive, conversational methodology, we also highlight specific instances that push beyond standard chat interfaces. These include deploying the model as a rigorous adversarial reviewer to detect subtle flaws in existing proofs, and embedding it within a "neuro-symbolic" loop that autonomously writes and executes code to verify complex derivations. Together, these examples highlight the potential of AI not just as a tool for automation, but as a versatile, genuine partner in the creative process of scientific discovery.

Motivation & Objective

• Demonstrate how advanced AI models can act as collaborators in tackling open problems across theoretical computer science and related fields.
• Identify and share reusable techniques for effective human-AI collaboration in rigorous mathematical reasoning.
• Showcase concrete flows where AI contributes to conjecture refutation, proof construction, and external validation.

Proposed method

• Documented a series of real-world collaborations using Gemini-based models to tackle open problems.
• Extracted and formalized common techniques such as iterative prompting, cross-domain analogy, and simulation for counterexample search.
• Described neuro-symbolic loops where AI generates and verifies code to corroborate derivations.
• Presented adversarial review and external validation workflows to ensure rigor.
• Outlined an AI-assisted research playbook and dynamics of human-AI collaboration.

Figure 1 : Overview of the reasoning architecture used in many testimonials: an extensive exploration of the solution space combined with deep reasoning and a long tail of automated and human verification and in several cases, guidance and iterative feedback.

Experimental results

Research questions

• RQ1 What AI-driven workflows enable meaningful mathematical and theoretical advances when paired with human expertise?
• RQ2 Which techniques best harness Gemini models for conjecture testing, proof construction, and validation across domains?
• RQ3 Can AI act as adversarial reviewers or autonomous verifiers to improve rigor in proofs and literature?
• RQ4 How do interdisciplinary connections via AI contribute to solving open problems in TCS and related areas?

Key findings

• Case studies show AI-assisted collaboration producing conjecture refutations, novel proofs, and refined insights across TCS, economics, optimization, and physics.
• A set of repeatable techniques emerges: iterative refinement, cross-pollination of ideas, simulation and counterexample search, formalization and rigor checks, interactive proof construction with external validation, and agentic tool-use with automated feedback.
• AI can function as adversarial reviewer and within neuro-symbolic loops to write and execute code for derivation verification.
• Human guidance remains essential; success hinges on structured prompts, problem decomposition, and rigorous external verification.
• The work outlines a practical AI-assisted research playbook and highlights implications for future research workflows.

(a) From Discrete Combinatorics to Continuous Measure Theory: To resolve an open question about bounded-rank SDP solutions for Max-Cut, the AI reframed a discrete combinatorial problem involving unit vectors into an energy minimization problem over continuous probability measures on the unit sphere