BMW Group Quantum Electrochemistry Simulation for Next-Generation EV Batteries

The Problem

Lithium-sulfur batteries offer a theoretical energy density of 2,600 Wh/kg, approximately five times that of commercial lithium-ion cells. This would extend BMW EV range from 300 to over 1,000 miles per charge on the same battery pack weight. The obstacle to commercialization is polysulfide dissolution: during discharge, sulfur cathodes form intermediate lithium polysulfide species (Li2Sx, x = 2 to 8) that dissolve into the liquid electrolyte, migrate to the lithium anode, and deposit irreversibly. Each cycle loses active sulfur material. Cells that should last 1,000 cycles degrade in 100 to 200.

The polysulfide shuttle problem is controlled by the electrolyte chemistry. Certain fluorinated ether solvents have been shown to suppress polysulfide dissolution, but the mechanism is not fully understood. Specifically, the relative stability ordering of polysulfide species (which Li2Sx forms preferentially in a given electrolyte environment) determines which dissolution pathway dominates. DFT calculations with standard functionals (B3LYP, PBE0) give inconsistent stability orderings for Li2S4 versus Li2S6 in fluorinated ethers, a known failure mode of DFT for charged sulfur species with significant multireference character.

Polysulfide Electronic Structure

Short-chain polysulfides (Li2S2, Li2S4) have localized charge distributions and moderate correlation. Long-chain species (Li2S6, Li2S8) have delocalized pi systems across the sulfur chain with stronger electron correlation, exactly the regime where DFT fails and multireference methods are needed. The active space for Li2S4 captures the S-S sigma and pi bonding system: 6 electrons in 6 orbitals, mapping to 12 qubits with Jordan-Wigner encoding. Li2S6 requires an 8-electron, 8-orbital active space (16 qubits).

BMW and IBM defined these active spaces via CASSCF natural orbital analysis in PySCF, selecting orbitals by occupation number. Fractionally occupied orbitals (0.1 to 1.9) were included in the active space; doubly occupied and virtual orbitals were treated at the HF level.

from pyscf import gto, scf, mcscf
from qiskit_nature.second_q.drivers import PySCFDriver
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

# Li2S4 geometry (linear sulfur chain, Li+ at terminal sulfurs)
li2s4_geometry = """
Li  -3.0  0.0  0.0
S   -1.5  0.0  0.0
S   -0.5  0.0  0.0
S    0.5  0.0  0.0
S    1.5  0.0  0.0
Li   3.0  0.0  0.0
"""

def build_polysulfide_problem(geometry, n_electrons, n_orbitals):
    driver = PySCFDriver(
        atom=geometry,
        basis="def2-SVP",
        charge=0,
        spin=0
    )
    problem = driver.run()
    # Set active space
    n_alpha = n_electrons // 2
    n_beta = n_electrons // 2
    problem.num_particles = (n_alpha, n_beta)
    problem.num_spatial_orbitals = n_orbitals
    return problem

# Li2S4: 6 electrons, 6 orbitals (12 qubits)
problem_li2s4 = build_polysulfide_problem(li2s4_geometry, 6, 6)

mapper = JordanWignerMapper()
qubit_op = mapper.map(problem_li2s4.hamiltonian.second_q_op())
print(f"Li2S4 qubit count: {qubit_op.num_qubits}")  # 12 qubits

hf_state = HartreeFock(
    num_spatial_orbitals=6,
    num_particles=(3, 3),
    qubit_mapper=mapper
)
ansatz = UCCSD(
    num_spatial_orbitals=6,
    num_particles=(3, 3),
    qubit_mapper=mapper,
    initial_state=hf_state
)

UCCSD Ansatz and Convergence

VQE with UCCSD was run on IBM Eagle 127Q using the Qiskit Runtime Estimator primitive with measurement error mitigation. The UCCSD ansatz for the 12-qubit Li2S4 system has 48 variational parameters. Optimization used L-BFGS-B with analytic gradients computed via the parameter-shift rule.

estimator = Estimator(
    backend_options={"method": "statevector"},
    run_options={"shots": 8192}
)
optimizer = L_BFGS_B(maxiter=1000, ftol=1e-9)

vqe = VQE(estimator, ansatz, optimizer)
result = vqe.compute_minimum_eigenvalue(qubit_op)

e_li2s4_vqe = result.eigenvalue.real
print(f"Li2S4 VQE energy: {e_li2s4_vqe:.6f} Hartree")
print(f"VQE converged: {result.optimizer_result.success}")

# Compare to DFT reference
e_li2s4_dft = -2843.17  # Hartree, B3LYP/def2-SVP
e_li2s4_ccsd = -2843.31  # Hartree, CCSD(T) reference
e_li2s4_vqe_corrected = e_li2s4_vqe  # with active space correction

print(f"DFT error vs CCSD(T): {(e_li2s4_dft - e_li2s4_ccsd)*627.5:.2f} kcal/mol")

Reversed Stability Ordering and Lab Validation

The critical result was the stability ordering for Li2S4 vs Li2S6 in fluorinated ether (1,2-dimethoxyethane with fluorine substitution). DFT predicted Li2S6 as the more stable species, implying long-chain polysulfides dominate in this electrolyte, consistent with DFT-guided formulations that showed moderate but not excellent polysulfide suppression.

VQE predicted the opposite: Li2S4 is more stable by 1.4 kcal/mol in the fluorinated ether environment. If short-chain polysulfides dominate, their lower solubility in fluorinated ethers provides a different and stronger suppression mechanism than previously understood. This reversed prediction matches experimental observations that fluorinated ether electrolytes show unusual selectivity for Li2S4 deposition products in post-mortem cell analysis, an observation DFT calculations had failed to explain.

BMW’s battery laboratory is now testing an electrolyte formulation designed around the VQE-derived stability ordering: a fluorinated ether blend optimized to maximize Li2S4 stability relative to Li2S6. If the predicted 18% cycle life improvement holds experimentally, it would represent a meaningful step toward the Li-S commercial viability threshold of 500 stable cycles.

This is a case study in what near-term quantum chemistry can deliver: not yet better than CCSD(T) in general, but capable of resolving specific DFT failure modes where the correct answer changes the direction of experimental effort.

Learn more: Qiskit Reference