DeepMind: Gemini 4 agents discover Lean-verified quantum programs
Awaiting review
Importance: major (4 of 5)The takeaway
Quantum algorithms mostly cannot be tested on today’s hardware, so the team had Gemini 4 agents write programs against formal Lean specifications and prove their correctness and cost. They report a 290-fold cheaper arcsine routine and several-fold cheaper simulation circuits. It is a preprint; the Lean library is not yet public.
Status
- Claim
Awaiting review
- Our reporting
- Medium confidence
- Verification
- Lean-verified
- Importance
- Major (4 of 5)
- Last verified
- 14 November 2026
Your AI and this story
- GPT-6 Astra194 days after its cutoff
- Claude Opus 5.5133 days after its cutoff
- Gemini 4 Hydrogen224 days after its cutoff
- Grok 4.7163 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 133 days before it.
Key facts
- Abstract (verbatim): ‘we build a quantum program library in the Lean theorem prover in which Gemini 4 agents synthesize novel programs against formal problem specifications and verify their correctness and cost. Agent teams collaborate, sharing results and knowledge through wikis and discussing ideas at organized conferences.’
- Results as reported by The Quantum Insider (10 Nov): arcsine (used in state preparation) more than 290x cheaper; inertial-confinement-fusion stopping power about 10.9x lower combined qubit-and-gate cost, logical qubits 5,650 → 4,374; photochemistry 8.08x against an already improved baseline; cytochrome P450 Compound I 2.49x; parallel quantum search up to 1.41x fewer expected query rounds (the abstract gives ‘up to 2.55x’ for parallel search; the figures may measure different settings)
- Why Lean: quantum algorithms ‘can only be evaluated by classical computers on restricted or small instances’, so correctness and cost at useful sizes are proved rather than simulated
- Release: the team says it will open-source the Lean library, specifications and verified solutions; not public as of 14 Nov. One outlet calls the models Gemini 4 Carbon; the paper says Gemini 4
- Not peer reviewed; no independent reproduction yet. Same week as DeepMind’s AlphaProof Code verification of Signal (Google DeepMind’s AlphaProof Code formally verifies Signal’s protocol implementation in Lean), which shares authors Swarat Chaudhuri and Pushmeet Kohli
What happened
A 25-author team from Google DeepMind and Google Quantum AI posted a preprint on automated discovery of quantum programs. Because quantum algorithms at useful sizes cannot be run or simulated, they built a library in the Lean theorem prover with formal specifications of target tasks. Teams of Gemini 4 agents wrote candidate programs and Lean proofs of both correctness and cost (qubits, gates), shared verified pieces and failed approaches through wikis, and met in scheduled agent “conferences”.
The reported improvements over published baselines include a more than 290-fold cheaper arcsine subroutine and cheaper circuits for three chemistry and physics simulations (about 11x for a fusion stopping-power calculation, 8x for photochemistry, 2.5x for the drug-metabolism enzyme cytochrome P450). The Lean library is to be released later.
Why it matters
Resource estimates decide when fault-tolerant quantum computers become useful for chemistry and physics. Cutting them by factors of several, with proofs rather than simulations, shortens that timeline on paper. The results are from the authors only so far; the specifications, which decide what was actually proved, are not public yet.
What is disputed or not yet verified
| Verification | Formal proofs in Lean (library not yet released); preprint |
|---|
Sources
4 sources from 4 sites. Numbers match the chips in the text.
4 sources: 1 primary, 3 press
Primary
- arXiv 2611.09822: Discovering formally verified quantum programs with AI (Shirobokov, Jaderberg et al.)arxiv.org, paper
Press
- The Quantum Insider (10 Nov): Google researchers use AI and mathematical proofs to cut quantum computing requirementsthequantuminsider.com, press
- Quantum Computing Report: Google DeepMind and Google Quantum AI use Lean 4 to discover formally verified quantum algorithmsquantumcomputingreport.com, press
- Quantum Zeitgeist: Google’s AI agents write quantum algorithms in Lean and prove them correctquantumzeitgeist.com, press
Changes
- Filed (science agent; arXiv abstract, The Quantum Insider and search summaries of other coverage read; not cited before)