Stellantis Quantum Chemistry for Solid-State Battery Cathode Design

The Problem

Solid-state batteries promise higher energy density and improved safety over lithium-ion cells by replacing the liquid electrolyte with a ceramic solid. The cathode material is a critical bottleneck: conventional lithium nickel manganese oxide (LNMO, LiNi0.5Mn1.5O4) and lithium nickel cobalt oxide (LNCO) cathodes suffer from Jahn-Teller distortion in Mn3+ ions during charge/discharge cycling. This distortion causes local lattice strain, crack propagation, and capacity fade, the main reason solid-state batteries have not yet reached production at the energy densities their theoretical voltage allows.

Stellantis (parent of Jeep, RAM, Dodge, Peugeot, and Fiat Chrysler brands) has aggressive electrification commitments requiring energy density improvements beyond what current lithium-ion chemistry can deliver. Their partnership with QuantumScape (specializing in solid electrolyte interfaces) and IBM Quantum aimed to identify cathode compositions where Ni-rich substitution suppresses Mn3+ Jahn-Teller distortion without sacrificing cycle stability.

VQE for Ni-Mn-O Cluster Chemistry

The Jahn-Teller effect in MnO6 octahedra arises from partial occupation of degenerate d-orbitals in Mn3+ (3d4 configuration), causing spontaneous symmetry breaking that elongates the octahedron along one axis. Standard DFT with hybrid functionals overestimates the distortion magnitude because it does not correctly treat the near-degenerate d-orbital occupation. VQE with a UCCSD ansatz was applied to an Ni-Mn-O cluster to compute the distortion energy as a function of Ni:Mn ratio.

from pyscf import gto, scf
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 import VQE
from qiskit_algorithms.optimizers import SLSQP
from qiskit.primitives import Estimator
import numpy as np

def compute_mn_o_energy(ni_fraction, geometry_distortion_angstrom):
    """
    Compute Mn-O cluster energy as function of Ni substitution and
    axial distortion. ni_fraction in [0, 1] replaces Mn with Ni.
    """
    # Simplified MnO4 cluster; real calc uses MnO6 octahedral geometry
    # Bond lengths: equatorial 1.91 A, axial = 1.91 + distortion
    axial_z = 1.91 + geometry_distortion_angstrom

    atom_str = f"""
    Mn  0.000  0.000  0.000;
    O   1.910  0.000  0.000;
    O  -1.910  0.000  0.000;
    O   0.000  1.910  0.000;
    O   0.000  0.000  {axial_z:.3f}
    """
    # Effective charge approximates Ni substitution effect
    effective_charge = int(round(ni_fraction * 1))

    driver = PySCFDriver(
        atom=atom_str,
        basis="cc-pVDZ",
        charge=effective_charge,
        spin=2,
        unit="Angstrom"
    )
    problem = driver.run()
    problem.active_space = (6, 6)  # Mn 3d + O 2p bonding orbitals

    mapper = JordanWignerMapper()
    ansatz = UCCSD(
        num_spatial_orbitals=problem.num_spatial_orbitals,
        num_particles=problem.num_particles,
        mapper=mapper
    )
    vqe = VQE(
        estimator=Estimator(),
        ansatz=ansatz,
        optimizer=SLSQP(maxiter=200)
    )

    from qiskit_nature.second_q.algorithms import GroundStateEigensolver
    result = GroundStateEigensolver(mapper, vqe).solve(problem)
    return result.total_energies[0]

# Scan distortion energy for pure Mn vs Ni-rich (ni_fraction=0.5)
distortions = [0.0, 0.05, 0.10, 0.15, 0.20]
print("Distortion (A) | E(Mn-only) Ha | E(Ni-rich) Ha | JT reduction")
for d in distortions:
    e_mn = compute_mn_o_energy(ni_fraction=0.0, geometry_distortion_angstrom=d)
    e_ni = compute_mn_o_energy(ni_fraction=0.5, geometry_distortion_angstrom=d)
    print(f"  {d:.2f}         | {e_mn:.4f}       | {e_ni:.4f}     | {e_mn - e_ni:.4f} Ha")

UCCSD vs Hardware-Efficient Ansatz

The team compared two ansatz choices on IBM Heron 133Q. UCCSD is chemically motivated (captures single and double excitations from the reference Hartree-Fock state) but has circuit depth that scales as O(N^4) in the number of molecular orbitals, leading to significant noise accumulation on current hardware. A hardware-efficient ansatz (HEA) with alternating Ry rotation and CNOT layers was also tested: it ran with shallower circuits but converged to higher (less accurate) energies. For the 6-orbital active space, UCCSD with error mitigation via probabilistic error cancellation gave results within 5 mHa of CCSD(T) benchmark calculations.

The key finding was that at Ni:Mn ratio of 0.5 (LiNi0.5Mn1.5O4, the standard LNMO composition), the VQE Jahn-Teller distortion energy was 0.12 eV, substantially lower than the 0.18 eV predicted by DFT+U with standard U parameters. This 33% lower distortion energy means the lattice resists symmetry breaking more strongly, correlating with experimental observations that Ni-rich LNMO shows improved cycle retention over Mn-rich variants.

Projected Cell Performance

The 0.12 eV Jahn-Teller distortion energy implies reduced lattice strain accumulation per cycle. Using a classical crack mechanics model calibrated to QuantumScape’s solid electrolyte interface characterization data, this lower distortion projects to a 40% reduction in capacity fade rate over 500 cycles. Combined with the higher operating voltage of LNMO (4.7 V vs 4.2 V for NMC), the net effect is a projected 8% increase in energy density relative to the current QuantumScape cell design. Stellantis is targeting this cathode chemistry for integration into the next-generation Dodge electric muscle car platform with a production-intent cell validation program beginning in 2025.