Introducing the BECOME Connector: Get certified energies from your AI chat
You can now order certified ground-state energies from Claude, ChatGPT, Gemini or Grok. Pick a molecule from our catalog of 125,999, set your accuracy, deadline and bid, then approve. You get proven lower and upper bounds on its energy, with proofs you can verify before and after you pay.
Today, we're launching the BECOME Connector, a remote MCP server built on Beloved Ecosystem's BECOME platform. It brings BECOME Quantum, our service for certified molecular energies, into your AI research workflow. When you ask your AI for a molecule's energy, it searches the BECOME Quantum Catalog and retrieves the Hamiltonian file with its Hartree–Fock energy. Then it prepares an order for you to approve: Ground State, proven lower and upper bounds on the energy. The connector is live now at https://mcp.belovedecosystem.com; connecting and browsing the catalog require no account, and ordering requires an early access code.
Why we built it
Researchers increasingly do their thinking with an AI chat open. But while a chat can give you a number, it can't prove it's right. Even full CI, often called exact, returns a floating-point number with no certificate. It can convince you, but it can't prove anything to anyone else.
BECOME Quantum Ground State delivers proof: a lower and an upper bound on the ground-state energy of the exact Hamiltonian you ordered, tied to that file by its SHA-256 fingerprint, with a zero-knowledge proof anyone can verify. The connector puts that result where the work already happens. In October we released the BECOME Quantum Catalog; the connector is the most direct way to order from it.
Why you need lower and upper bounds, even for 8-electron, 8-orbital models
At this size, the question isn't whether you can compute an estimate of the energy. It's whether anyone else can check it without trusting your code. A certified result can be checked by anyone, in one step.
In theory, every variational method, from Hartree–Fock to DMRG, gives an upper bound: the true energy is no higher, but nothing says how much lower it is. In practice, even that bound is only as good as the floating-point arithmetic behind it.
Full CI doesn't solve this, even at this size. It gives you an estimate, with no proof of how far off it is: every operation rounds, the errors pile up across millions of steps, subtracting large energies cancels digits, and the solver stops at a tolerance. So no one else can rely on it without rerunning it or trusting your code.
A lower bound closes the gap, but only if it's rigorous. Lower bounds computed in floating point have already come out above the full CI energy for LiH, HF and four other molecules, by up to 1 mHa (Nakata et al., J. Chem. Phys., 2001). Recomputed with every rounding error accounted for, the bounds came out below full CI (Chaykin et al., J. Chem. Theory Comput., 2020).
With both bounds proven, the true energy is pinned inside an interval, and its width is your error bar. That lets you:
- Trust small differences: if two certified intervals don't overlap, you know which energy is lower.
- Catch wrong answers: any result below the lower bound is wrong, whether it comes from a coupled-cluster code, a quantum device or an AI model.
- Use it as a reference: in benchmarks, published data or AI training, every value carries an error bar anyone can verify.
How it works
Three steps, all from one request
- Ask for an energy. Your AI finds the molecule in the catalog and gets its Hamiltonian file, Hartree–Fock energy and fingerprint.
- Name your terms, then approve. Give your accuracy, deadline and bid. Your AI prepares the order, and you approve it in your browser using your early access code.
- Get your answer. When the result is ready, your AI fetches your certified bounds and verifies the proof.
What you choose
- Accuracy: the widest the interval may be. Standard (1.6×10⁻³ Ha, chemical accuracy), Precise (1.6×10⁻⁶ Ha) or Ultra-precise (1.6×10⁻¹⁰ Ha).
- Deadline: 30, 60 or 90 days. Each order gets a new zero-knowledge proof made for your file and accuracy target, and producing it takes substantial computing power, so we schedule the work.
- Bid: what you want to pay, in US dollars. Your bid covers producing your proof. Nothing is charged when you submit. If we can't reach your accuracy by your deadline, you owe nothing.
How payment works: we accept or decline each bid; a declined bid costs you nothing. You can cancel until we accept; after that, your bid is the fee. When the completion proof shows your target is met, we email an invoice, due on receipt, and you receive your bounds once it's paid.
Designed for checking, not trusting
- Fingerprint: a SHA-256 code that identifies the exact Hamiltonian file. Your order and your result are bound to it.
- Completion proof: before you pay, a zero-knowledge proof shows that the result is bound to your file's fingerprint and meets your accuracy target, while the bounds stay sealed.
- Final proof: after you pay, you get the bounds and can confirm they match the commitment in the completion proof. Your AI can run both checks with the connector's verify tool.
A realistic example
A researcher asks Claude: "Get me a certified ground-state energy for ethanol." Claude finds ethanol (PubChem CID 702, catalog entry 17291), gets its CAS(8e,8o)/6-31G* Hamiltonian and reports its Hartree–Fock energy, −154.0706984695 Ha. The researcher replies with Ultra-precise, a 30-day deadline and their bid, then approves the order in their browser using an early access code. When the job completes, Claude fetches the result and verifies the proof.
How the proof works
For that same ethanol Hamiltonian, we proved the ground-state energy lies between −154.0733873199 and −154.0733873198 Ha, an interval 10⁻¹⁰ Ha wide, within the Ultra-precise target of 1.6×10⁻¹⁰ Ha. Every step needed to establish the bounds is proven. The checker is proven sound in Coq (now Rocq): if it accepts a certificate, the bounds hold for that Hamiltonian. Succinct's SP1, an open-source system for generating and checking zero-knowledge proofs, then proves the checker ran on your exact file, and you can verify that proof yourself with SP1's open-source verifier. The program SP1 runs is built from the Coq-verified checker and pinned by its published verification key, so any proof made by a different program fails to verify.
Good to know
- What a result covers: the Hamiltonian in your file, a model that depends on basis set, active space and geometry (catalog geometries are PM6-optimized structures from PubChemQC; see the Data Descriptor).
- What you can order today: any of the 125,999 catalog entries. BECOME Quantum Custom is coming.
- What's free today: searching the catalog and downloading Hamiltonian files. Only ordering needs an early access code.
- What it runs on: classical computers. No quantum hardware is needed to compute or verify a result.
What's next
We'll add BECOME Quantum Custom, so you can order certified bounds for your own FCIDUMP files.
Get started
Connect your AI app at https://mcp.belovedecosystem.com and start searching the catalog and downloading Hamiltonian files right away. No account or code is needed. Ordering needs an early access code: request one.
Related: Releasing 125,999 ready-to-use molecular Hamiltonians · Data Descriptor: doi:10.5281/zenodo.23601240
Press contact: connect@belovedecosystem.com