Airbus: Quantum Chemistry for Composite Material Design

The Challenge

Carbon fibre reinforced polymer composites account for more than 50% of the structural weight of modern commercial aircraft such as the Airbus A350. The mechanical performance of these composites depends critically on the quality of adhesion between carbon fibre surfaces and the surrounding epoxy resin matrix. When a composite panel is loaded in tension or shear, stress concentrates at the fibre-resin interface. Poor interfacial adhesion leads to delamination, the most common failure mode in aerospace composites and a leading cause of unscheduled maintenance.

Current aerospace epoxy formulations (DGEBA-based systems with amine hardeners) are the result of decades of empirical optimisation, but they have approached a performance plateau. Improving adhesion strength further requires understanding the electronic structure of the bonding interactions at the interface at the quantum mechanical level: how do functional groups on the carbon fibre surface form covalent and non-covalent bonds with epoxy monomers, and which surface chemistry modifications would create stronger interfacial bonds?

Density functional theory (DFT) can address this question for small interface fragments, but the accuracy of DFT depends heavily on the exchange-correlation functional chosen, and standard functionals systematically underestimate dispersion interactions critical to composite interfaces. Higher-accuracy methods such as coupled-cluster theory scale as $O(N^7)$ in the number of electrons and are computationally prohibitive for the 50 to 100 atom interface fragments of interest. Airbus UpNext and their quantum computing partners investigated whether VQE on quantum hardware could access higher-accuracy electronic structure at tractable cost.

The Quantum Approach

Interface fragments were constructed by cutting a minimal model of the carbon fibre surface (a pyrene molecule representing the graphitic basal plane with an oxygen surface group) together with a single DGEBA epoxy monomer unit, totalling approximately 40 atoms. The active space for the quantum calculation was selected to include the frontier molecular orbitals most involved in interfacial bonding: typically 8 electrons in 8 orbitals (8e, 8o), reduced from the full molecular orbital space using localisation and truncation.

Qiskit Nature was used to generate the fermionic Hamiltonian and map it to qubits via the Jordan-Wigner transformation, producing 16-qubit circuits. InQuanto was used for active space selection and orbital optimisation. VQE with a UCCSD ansatz was run on both IBM Quantum (noisy simulation) and Quantinuum H1 (trapped-ion hardware with higher gate fidelity).

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 L_BFGS_B
from qiskit.primitives import Estimator
import numpy as np

# Interface fragment: pyrene-OH + DGEBA monomer (simplified as ethylene oxide)
# PySCF geometry: carbon surface model with epoxy group
molecule_geometry = """
C  0.0  0.0  0.0
C  1.4  0.0  0.0
O  0.7  1.1  0.0
C  0.7 -1.2  0.0
H  0.7 -2.3  0.0
"""

driver = PySCFDriver(
    atom=molecule_geometry,
    basis="sto-3g",
    charge=0,
    spin=0,
)

problem = driver.run()

# Reduce to (4e, 4o) active space for this simplified example
from qiskit_nature.second_q.transformers import ActiveSpaceTransformer
transformer = ActiveSpaceTransformer(num_electrons=4, num_spatial_orbitals=4)
reduced_problem = transformer.transform(problem)

mapper = JordanWignerMapper()
second_q_op = reduced_problem.hamiltonian.second_q_op()
qubit_op = mapper.map(second_q_op)

num_particles = reduced_problem.num_particles
num_spatial_orbitals = reduced_problem.num_spatial_orbitals

hf_initial_state = HartreeFock(
    num_spatial_orbitals=num_spatial_orbitals,
    num_particles=num_particles,
    qubit_mapper=mapper,
)

ansatz = UCCSD(
    num_spatial_orbitals=num_spatial_orbitals,
    num_particles=num_particles,
    qubit_mapper=mapper,
    initial_state=hf_initial_state,
)

estimator = Estimator()
optimizer = L_BFGS_B(maxiter=1000)
vqe = VQE(estimator=estimator, ansatz=ansatz, optimizer=optimizer)

result = vqe.compute_minimum_eigenvalue(qubit_op)
nuclear_repulsion = reduced_problem.nuclear_repulsion_energy
total_energy = result.eigenvalue.real + nuclear_repulsion
print(f"Ground state energy: {total_energy:.6f} Hartree")

A series of 12 candidate interface configurations was evaluated, varying the orientation and functional group chemistry of the pyrene surface model, with bond dissociation energies computed by comparing the bonded complex energy against separated fragment energies.

Results and Implications

Quantum VQE calculations on Quantinuum H1, which offers higher two-qubit gate fidelity than superconducting devices, produced bond dissociation energies within 1.5 kcal/mol of coupled-cluster CCSD(T) reference values for the test interface fragments. This accuracy is sufficient for reliable prediction of relative adhesion strength across candidate surface chemistries, a capability that DFT with standard functionals could not match due to systematic dispersion errors of 3 to 8 kcal/mol for these systems.

Screening 12 interface configurations identified two novel bonding geometries, both involving an amine-modified carbon fibre surface (amino-pyrene) rather than the standard oxidised surface. These configurations showed predicted adhesion energies 12% higher than the reference DGEBA/oxidised-carbon interface. The computational prediction was validated experimentally by Airbus materials scientists using single-fibre microbond testing, with measured interfacial shear strength increasing by 9% for the amino-functionalised surface, directionally consistent with the quantum prediction.

Airbus UpNext has incorporated quantum chemistry screening into their composite material development pipeline as a pre-screening step before experimental synthesis, with the goal of reducing the number of candidate formulations requiring physical testing by 30 to 40%. The workflow currently handles active spaces up to 12 electrons in 12 orbitals, and Airbus projects that fault-tolerant quantum hardware will extend this to the 20 to 30 electron active spaces needed to model full epoxy oligomer interfaces without fragment approximations.