Battery chemistry on a quantum machine — Li-ion, solid-state, beyond

Why batteries are a natural fit for early quantum hardware

Battery cathodes, electrolytes and interphases are governed by strongly correlated electron behaviour that classical density functional theory handles badly. The 3d transition metals at the heart of every commercial cell — nickel, manganese, cobalt, iron, vanadium — sit in a regime where mean-field approximations under-resolve the magnetic exchange, the Jahn-Teller distortions and the multi-reference character of charged and discharged states. Coupled-cluster methods scale as O(N^7) and choke on transition-metal clusters of meaningful size. Quantum Monte Carlo gives better accuracy but at brutal wall-clock cost.

This is where a 100-physical-qubit transmon machine starts to earn its keep. The Variational Quantum Eigensolver (VQE) and its successors — ADAPT-VQE, contextual subspace VQE, and sample-based quantum diagonalisation — let us encode the active space of a redox-active cluster directly into qubit registers and push the correlation problem onto the hardware. We are not claiming this beats DFT for everything. We are claiming it beats DFT precisely where DFT struggles, and that happens to be the part of the battery problem that matters most: the active site under cycling.

Li-ion: where quantum chemistry adds real signal

For lithium-ion cathode work the high-value targets are well known. Voltage prediction in NMC and NCA stacks is sensitive to the on-site Coulomb U chosen for DFT+U; the answer can swing 200–400 mV depending on parameter choice. Running a small CAS active space (say 12 spatial orbitals, 14 electrons) on a quantum machine using a hardware-efficient ansatz gives a non-empirical reference that pins U against an actual correlated wavefunction rather than a fitted curve.

The other Li-ion target is the cathode-electrolyte interphase. The decomposition of LiPF₆ in carbonate solvents is a multi-step radical process where transition states have open-shell singlet character. These are exactly the cases where single-reference methods fail and where a 60–80 logical-qubit subspace calculation, even noisy, gives a defensible energy ordering of pathways. Anode-side, lithium plating versus intercalation on graphite at fast-charge currents is a kinetics problem with a chemistry root — the Marcus reorganisation energy at the SEI — and that root is a quantum chemistry calculation we can structure for hardware execution.

What we will actually run on first-light hardware

First-light at Co. Tipperary is single-qubit characterisation in Q1 2027. Multi-qubit chemistry workloads come in Q2 2027. The realistic early Li-ion workload is a Hamiltonian in the 20–40 qubit range, mapped via Jordan-Wigner or Bravyi-Kitaev, executed under a heavy-hex topology with active error mitigation — zero-noise extrapolation, probabilistic error cancellation, dynamical decoupling on idle qubits. No surface-code yet; surface-code logical qubits on this device size are a 2028+ conversation. The goal in 2027 is correlated energies on small but real cathode fragments, benchmarked against CCSD(T) where CCSD(T) still fits.

Solid-state batteries: the harder, more valuable problem

Solid-state battery quantum chemistry is the workload I am most interested in personally, because the unsolved problems are interface problems and interface problems are correlated electron problems. Sulfide electrolytes (Li₆PS₅Cl, Li₁₀GeP₂S₁₂) and oxide electrolytes (LLZO, LATP) both fail at the cathode contact for reasons DFT routinely mis-predicts: space-charge layers, polaron formation, and chemical decomposition driven by partial charge-transfer states.

The lithium-metal anode side is similarly painful. Dendrite nucleation is mechanical and electrochemical, and the electrochemical part hinges on the electronic structure of Li adatoms on a defective oxide or sulfide surface. Modelling that needs spin-polarised correlated calculations on clusters that are too big for CCSD(T) and too multi-reference for DFT. A quantum machine does not solve the whole supercell — we are not pretending — but it can solve the embedded active region inside a QM/MM or DFT-embedded quantum scheme. Projection-based embedding plus a quantum solver for the active region is the architecture we expect to actually deploy.

Beyond Li-ion: sodium, magnesium, multivalent and metal-air

Sodium-ion is closest to commercial relevance and shares most of the Li-ion computational stack — Prussian blue analogues and layered oxides slot into the same VQE workflows with minor active-space adjustments. Magnesium and calcium multivalent chemistry is harder and more interesting: the divalent cation gives stronger electronic correlation in the cathode, and the desolvation energetics at the electrode surface are a genuinely open problem. Metal-air chemistry — Li-O₂, Zn-air — involves triplet oxygen, superoxide and peroxide intermediates whose relative energies are notoriously sensitive to method choice.

For battery material science aimed at grid storage rather than vehicles, redox-flow chemistries (vanadium, organic quinones, iron-chromium) are also strong candidates. The organic flow molecules in particular have small enough active spaces that meaningful calculations are tractable on hardware with 50–80 well-characterised qubits.

Engineering reality: what 100 physical qubits actually buys you

Let me be plain about what the machine can and cannot do in its first eighteen months of operation. 100 physical transmon qubits at heavy-hex connectivity, held at sub-15 mK in a dilution refrigerator, with two-qubit gate fidelities in the 99.0–99.7% band, gives roughly 40–60 effectively-usable qubits for variational chemistry once you account for ancilla, readout, and the depth budget your noise floor allows. That is enough for embedded active spaces, not full-cell simulations.

You program it through the standard stack: OpenQASM 3 at the bottom, Qiskit and PennyLane and Cirq above, with Qiskit Nature and PennyLane's quantum chemistry modules handling the fermion-to-qubit mapping. The Hamiltonian construction happens classically — PySCF, OpenFermion — and the quantum machine executes the parameterised circuits and returns expectation values. None of this is exotic. All of it is standard practice in superconducting quantum chemistry today. The contribution