Toyota Research Institute: Quantum Simulation for Solid-State Battery Materials

The Problem

Solid-state batteries are the next major frontier for electric vehicle range and safety. They replace the liquid electrolyte in conventional lithium-ion batteries with a solid ceramic or polymer material, eliminating flammability risk and potentially doubling energy density.

The bottleneck is the interface between the lithium metal anode and the solid electrolyte. At this interface, lithium ions move across a chemically complex boundary layer. The stability and ionic conductivity of this layer depend on quantum mechanical interactions between Li ions and the oxide material that classical Density Functional Theory (DFT) does not capture accurately.

Strong electron correlation at these interfaces means that a single Slater determinant (the approximation underlying DFT) is not sufficient. Multiconfigurational methods are needed, and those scale exponentially with system size classically.

Toyota Research Institute’s quantum team asked whether VQE could eventually handle these strongly correlated interface problems better than classical methods.

Approach: Active Space VQE with PySCF Backend

TRI used Qiskit Nature with PySCF as the classical electronic structure backend. The workflow:

1. Run a classical Hartree-Fock calculation in PySCF to get molecular orbitals
2. Select the active space (the strongly correlated orbitals near the HOMO-LUMO gap)
3. Map the active space Hamiltonian to qubit operators using Jordan-Wigner encoding
4. Optimize the UCCSD ansatz with VQE

Model systems were LiH (a minimal lithium hydride model) and Li2O (a simple lithium oxide, representing electrolyte chemistry), chosen because their exact solutions are known classically.

from qiskit_nature.second_q.drivers import PySCFDriver
from qiskit_nature.second_q.transformers import ActiveSpaceTransformer
from qiskit_nature.second_q.mappers import JordanWignerMapper
from qiskit_nature.second_q.circuit.library import UCCSD, HartreeFock
from qiskit_algorithms.minimum_eigensolvers import VQE
from qiskit_algorithms.optimizers import L_BFGS_B
from qiskit_aer.primitives import Estimator
import numpy as np

# LiH at equilibrium geometry (Li-H bond = 1.595 Angstrom)
driver = PySCFDriver(
    atom="Li 0.0 0.0 0.0; H 0.0 0.0 1.595",
    basis="sto-3g",
    charge=0,
    spin=0,
)
problem = driver.run()

print(f"Full orbital count: {problem.num_spatial_orbitals}")
print(f"Total electrons: {sum(problem.num_particles)}")

# Active space: 2 electrons in 3 orbitals
# Captures the Li-H bonding/antibonding pair plus one lone-pair orbital
transformer = ActiveSpaceTransformer(num_electrons=2, num_spatial_orbitals=3)
active_problem = transformer.transform(problem)

mapper = JordanWignerMapper()
hamiltonian = mapper.map(active_problem.second_q_ops()[0])

print(f"Active space qubit count: {hamiltonian.num_qubits}")

# UCCSD ansatz starting from Hartree-Fock initial state
hf_state = HartreeFock(
    num_spatial_orbitals=active_problem.num_spatial_orbitals,
    num_particles=active_problem.num_particles,
    qubit_mapper=mapper,
)

ansatz = UCCSD(
    num_spatial_orbitals=active_problem.num_spatial_orbitals,
    num_particles=active_problem.num_particles,
    qubit_mapper=mapper,
    initial_state=hf_state,
)

# Run VQE
estimator = Estimator()
optimizer = L_BFGS_B(maxiter=500)

vqe = VQE(estimator=estimator, ansatz=ansatz, optimizer=optimizer)
vqe_result = vqe.compute_minimum_eigenvalue(hamiltonian)

# Compare to reference energies
nuclear_repulsion = active_problem.nuclear_repulsion_energy or 0.0
vqe_energy = vqe_result.eigenvalue.real + nuclear_repulsion

# Classical CCSD(T) reference for LiH/sto-3g: -7.882 Ha
reference_energy = -7.882
delta = abs(vqe_energy - reference_energy) * 627.5  # convert Ha to kcal/mol

print(f"VQE total energy:       {vqe_energy:.6f} Ha")
print(f"CCSD(T) reference:      {reference_energy:.6f} Ha")
print(f"Energy difference:      {delta:.3f} kcal/mol")
print(f"Chemical accuracy met:  {delta < 1.0}")

Comparison with Classical DMRG

Density Matrix Renormalization Group (DMRG) is the state-of-the-art classical method for strongly correlated 1D and quasi-1D systems. It represents the many-body wavefunction as a matrix product state (MPS), allowing controlled approximation with a bond dimension parameter.

TRI ran the same LiH and Li2O active space problems through DMRG using the Block2 library, comparing:

System	Qubits	VQE energy error	DMRG error	DMRG bond dim
LiH	6	0.3 kcal/mol	0.1 kcal/mol	64
Li2O	8	0.8 kcal/mol	0.2 kcal/mol	128

VQE achieved chemical accuracy on both systems. DMRG was more accurate at the same system sizes. The advantage of quantum simulation would emerge when DMRG’s bond dimension requirement scales exponentially with system size, a regime that begins around 50-100 orbitals for strongly correlated 2D materials.

Li4Ti5O12 Fragment Study

As a stretch goal, TRI attempted active space calculations on fragments of Li4Ti5O12, a common battery anode material. The full interface involves hundreds of atoms, but small cluster fragments (8-12 atoms) yielded active spaces of 12-20 orbitals.

At 20 orbitals, the Jordan-Wigner encoding produces 40 qubits, beyond current hardware noise budgets for VQE convergence. The team used these systems as simulator targets, validating that the workflow scaled and identifying the active spaces most likely to exhibit strong correlation at the interface.

Results

Findings from the TRI quantum battery program:

• VQE with UCCSD matched DMRG within chemical accuracy on all 4-8 qubit test systems
• Classical DMRG remains superior through approximately 30-40 qubit equivalents due to lower noise
• Crossover point where quantum simulation overtakes DMRG is estimated at 50-100 logical qubits with full error correction
• LiH and Li2O serve as validated benchmarks for future hardware comparisons

TRI published these results as part of their 10-year quantum computing roadmap, which stages quantum simulation investments alongside their classical computational materials program.

Why This Matters for EV Development

Finding the right solid electrolyte material could cut EV battery development cycles from 10 years to 3-5. Classical simulation narrows the candidate space, but strong correlation at interfaces remains a blind spot. Quantum chemistry that handles these correlations correctly would reduce the experimental screening burden significantly.

Toyota’s quantum team treats current hardware experiments as methodological groundwork. The quantum circuits, active space protocols, and benchmarks developed now will be directly applicable when hardware reaches the 50-100 logical qubit threshold.

Framework and Tools

• Qiskit Nature: Hamiltonian construction, Jordan-Wigner mapping, UCCSD ansatz
• PySCF: Classical Hartree-Fock driver and integral generation
• Block2: Classical DMRG reference calculations
• Qiskit Aer: Statevector simulator for noiseless benchmarks
• IBM Quantum: Real hardware validation of small test cases

Learn more: Qiskit Nature Reference | VQE Algorithm Guide