Post by qBraid

We used Google's AlphaEvolve to find more efficient error-correcting codes for quantum chemistry. It discovered something no one had managed to build by hand: a distance-5 encoding for molecular systems. ⚛️ Here's why that's hard. Getting useful chemistry onto a quantum computer hinges on one choice — how you translate a molecule's electrons into qubits. One such mapping is the Generalized Superfast Encoding (GSE), a fermionic encoding that comes with built-in stabilizers you can use for error correction. For a molecule with just 8 orbitals, there are more than 10^50 ways to do it. Searched by hand, the best structures have always topped out at distance 3. So we seeded AlphaEvolve with a working baseline built on our GSE and let it evolve thousands of variations. What it handed back beat our hand-designed baseline on every axis that mattered: 🔹 Exact distance 5 on dense molecular Hamiltonians — hand design only reached distance 3 🔹 4.2–5.0× fewer data qubits than the standard fault-tolerant route 🔹 3.4–7.9× lower logical error rate under exact decoding in the code-capacity model 🔹 Held distance 5 on BeH₂ and H₂O — molecules the search never saw And because the output was readable code, we could verify exactly what it found and why it worked. Thanks to the AlphaEvolve and AI for Science teams at Google Cloud for the early access and support throughout. 🙏 📄 Paper in the comments. Full write-up on our blog 👇 #QuantumComputing #QuantumErrorCorrection #AlphaEvolve #QuantumChemistry #qBraid