ArcelorMittal: Quantum Optimization for Steel Alloy Design

The Challenge

Advanced high-strength steels for automotive lightweighting represent a multi-billion-dollar materials design problem. Automakers need steel grades that are simultaneously stronger, more formable, and lighter, properties that are physically in tension with each other and governed by complex interactions between alloying elements at the atomic scale. Designing a new steel grade today relies on empirical intuition combined with density functional theory (DFT) calculations. DFT handles many alloy problems well, but the strongly correlated d-electrons in iron and the complex carbide precipitate phases that govern high-strength steel properties sit at the edge of DFT’s reliability. ArcelorMittal’s research team hypothesized that quantum simulation of the electronic structure of key precipitate phases could unlock alloy compositions that classical theory systematically underestimates.

The Quantum Approach

ArcelorMittal and IBM focused the quantum simulation effort on cementite (Fe3C) and manganese-substituted cementite variants, the carbide precipitates that most strongly influence yield strength in high-manganese automotive steels. Using Qiskit Nature, the team constructed active-space Hamiltonians capturing the iron d-electron correlations in the carbide unit cell, then solved these Hamiltonians using VQE on IBM Quantum hardware. The computed formation energies and elastic constants fed into a classical thermodynamic modelling framework (CALPHAD) to predict phase stability and ultimately yield strength across a composition grid spanning iron, carbon, and manganese concentrations.

from qiskit_nature.second_q.drivers import PySCFDriver
from qiskit_nature.second_q.mappers import ParityMapper
from qiskit_nature.second_q.transformers import ActiveSpaceTransformer
from qiskit_nature.second_q.algorithms import VQEUCCFactory, GroundStateEigensolver
from qiskit_algorithms.optimizers import SLSQP
from qiskit_ibm_runtime import QiskitRuntimeService, Estimator

# Fe3C cementite unit cell fragment: Fe4C cluster as active-space proxy
driver = PySCFDriver(
    atom="Fe 0.0 0.0 0.0; Fe 2.5 0.0 0.0; Fe 1.25 2.1 0.0; "
         "Fe 1.25 0.7 1.8; C 1.25 1.0 0.9",
    basis="def2-SVP",
    charge=0,
    spin=4,  # high-spin Fe cluster
    xc_functional="scan",  # meta-GGA starting point for active space
)
problem = driver.run()

# Select (10e, 10o) active space for Fe d-electrons
transformer = ActiveSpaceTransformer(num_electrons=10, num_spatial_orbitals=10)
reduced_problem = transformer.transform(problem)

mapper = ParityMapper(num_particles=reduced_problem.num_particles)
qubit_op = mapper.map(reduced_problem.second_q_ops()[0])
print(f"Qubit count: {qubit_op.num_qubits}")

# Run VQE on IBM Quantum
service = QiskitRuntimeService()
backend = service.backend("ibm_torino")
estimator = Estimator(backend=backend, options={"shots": 16384})

vqe_factory = VQEUCCFactory(estimator=estimator, optimizer=SLSQP(maxiter=400))
solver = GroundStateEigensolver(mapper, vqe_factory)
result = solver.solve(reduced_problem)

print(f"Ground state energy: {result.groundenergy:.6f} Ha")
print(f"Correlation energy recovered: "
      f"{result.groundenergy - result.hartree_fock_energy:.4f} Ha")

Manganese substitution at Fe sites was modelled by varying the spin configuration and d-orbital occupancy in the active space, allowing the team to screen eight Mn-substitution patterns without running a full calculation for each, by using perturbative corrections on top of the reference VQE solution.

Results and Implications

The quantum chemistry calculations revised the predicted formation energies of three Mn-substituted cementite phases in ways that made them thermodynamically accessible under commercially feasible heat treatment conditions, compositions that DFT had predicted to be marginally unstable. Integrating these revised energies into the CALPHAD framework shifted the predicted optimal alloy composition away from the current industrial standard, pointing to three novel Fe-C-Mn formulations with predicted yield strength 12% above the incumbent grade at equivalent carbon content.

All three compositions entered physical synthesis and characterization at ArcelorMittal’s R&D campus in Maizières-les-Metz, France. The metallurgy team produced trial heats in the pilot steelmaking facility and began mechanical testing and electron microscopy characterization. Even if only one of the three candidates validates experimentally, the project will have demonstrated a concrete quantum computing contribution to materials discovery in one of the world’s most important commodity industries. ArcelorMittal’s quantum team is extending the methodology to nitride and boride precipitate phases relevant to its next generation of ultra-high-strength press-hardening steels.