PASQAL: Neutral Atom Quantum Computing for Credit Risk

The Challenge

Credit portfolio risk management requires estimating the probability distribution of simultaneous defaults across a portfolio of correlated obligors. The correlations matter enormously: in 2008, underestimated default correlations turned individually acceptable risks into systemic catastrophe. The standard industry tool is Monte Carlo simulation, which samples default scenarios by drawing from a multivariate default model. For portfolios with strong correlation structure and fat-tailed joint default probabilities, the rare-event tail requires enormous sample counts to estimate accurately, making real-time risk assessment computationally prohibitive for large portfolios. PASQAL partnered with a European bank to test whether neutral-atom quantum processors could model correlated defaults more efficiently by encoding the correlation structure directly into physical qubit-qubit interactions.

The Quantum Approach

PASQAL’s Fresnel processor arranges rubidium atoms in configurable two-dimensional arrays. When atoms are brought within the Rydberg blockade radius, they experience a strong repulsive interaction that creates natural entanglement without explicit gate operations. PASQAL’s team encoded a 60-asset credit portfolio by mapping each asset to one atom, then positioned atoms in a 2D layout where inter-atom distances encoded the pairwise default correlations from the bank’s internal credit model. The Pulser framework translated the correlation matrix into a laser pulse sequence that drove the atom array to a quantum state whose measurement statistics approximated the joint default probability distribution.

import pulser
from pulser import Pulse, Sequence, Register
from pulser.devices import Fresnel
import numpy as np

# Load 60-asset correlation matrix from credit model
corr_matrix = np.load("portfolio_correlation_matrix.npy")  # shape (60, 60)

# Map correlation structure to 2D atom positions using MDS
from sklearn.manifold import MDS
mds = MDS(n_components=2, dissimilarity="precomputed", random_state=42)
positions = mds.fit_transform(1 - corr_matrix)

# Scale positions to fit within Fresnel device constraints (max 35 atoms in demo)
# Use top-60 PCA compression for full portfolio
positions = positions[:60] * Fresnel.min_atom_distance * 2.5

reg = Register.from_coordinates(positions, prefix="asset")

seq = Sequence(reg, Fresnel)
seq.declare_channel("global_rydberg", "rydberg_global")

# Drive Rydberg transition with Omega and Delta pulse shaped for default encoding
omega_pulse = Pulse.ConstantDetuning(
    amplitude=pulser.waveforms.BlackmanWaveform(duration=1000, area=np.pi),
    detuning=-2 * np.pi * 3.0,  # MHz, tuned to encode default threshold
    phase=0.0,
)
seq.add(omega_pulse, "global_rydberg")

# Sample default scenarios from measurement outcomes
backend = pulser.backends.QPUBackend(seq, Fresnel)
result = backend.run(num_runs=10000)
samples = result.get_samples()
print(f"Collected {len(samples)} default scenario samples")
joint_default_prob = np.mean([bin(s).count("1") >= 5 for s in samples])
print(f"P(5+ simultaneous defaults): {joint_default_prob:.4f}")

The analog encoding sidesteps the need for discrete two-qubit gate operations, allowing the full 60-atom correlation structure to be embedded in a single coherent evolution rather than decomposed into potentially thousands of two-qubit gates.

Results and Implications

The Fresnel processor generated statistically equivalent estimates of the joint default probability distribution for the 60-asset portfolio 8 times faster than a CPU-bound Monte Carlo simulation running on equivalent compute infrastructure. Crucially, the quantum approach showed no degradation in tail-risk accuracy: Value-at-Risk at the 99.9% confidence level matched the Monte Carlo reference to within 0.3%, within the statistical tolerance of the benchmark. For the bank’s risk desk, this speed advantage directly translates to the ability to run intraday portfolio risk updates rather than overnight batch processes.

The study is notable for demonstrating the business relevance of analog quantum computing, a modality that often receives less attention than gate-based approaches but may reach practical advantage sooner for specific structured problems. PASQAL is developing a dedicated financial risk product built on this methodology, and the bank partner is evaluating integration into its regulatory risk reporting infrastructure. As Fresnel’s atom count scales toward 1,000, the approach could address portfolios an order of magnitude larger than those tractable today.