A Hybrid Quantum Neural Network to Analyse Big Experimental Powder X-ray Diffraction Data
Authors: H. Dong, S. D. M. Jacques, M. Q. Hlatshwayo, E. Papoutsellis, K. Georgopoulos, A. M. Beale, A. Vamvakeros
Organizations: Finden ltd, Building R70, Rutherford Appleton Laboratory, Harwell Science and Innovation Campus, OX11 0QX, UK · National Quantum Computing Centre, Harwell Science and Innovation Campus, OX11 0QX, UK · Department of Chemistry, University College London, 20 Gordon Street, WC1H 0AJ, UK · Research Complex at Harwell, Rutherford Appleton Laboratory, Harwell Science and Innovation Campus, OX11 0FA, UK · Dyson School of Design Engineering, Imperial College London, London, SW7 2DB, UK
Quantitative analysis of experimental powder X-ray diffraction data remains challenging when evaluating complex multiphase materials and noisy measurements. We introduce a hybrid quantum neural network framework designed to extract quantitative parameters, such as phase weight fractions and scale factors, directly from one-dimensional powder diffraction patterns without iterative refinement. The model combines noise-aware classical simulator pre-training with fast downstream fine-tuning on quantum processing unit features, ensuring stability against hardware decoherence. We demonstrate the practical utility of this approach by deploying the trained network onto an IBM quantum computer to analyse experimental X-ray diffraction computed tomography datasets from a three-phase solid oxide fuel cell containing ca. 10,000 patterns and a four-phase lithium-ion battery containing ca. 20,000 patterns). The network successfully reconstructs quantitative spatial phase maps in strong agreement with classical Rietveld refinement, paving the way for using quantum computing hardware to analyse real-world materials characterisation data.
Figures & tables
Figure 1: Detailed circuit topology of the 6 qubit parametrised quantum neural network showing sequential Rx , Ry , and Rz single qubit rotations, ring Controlled NOT entangling gates across two repetitions, and final Pauli observable measurements.
Figure 2: End-to-end workflow for simulator pre-training, quantum-hardware calibration, and experimental inference. The hybrid QNN is first pre-trained on 200,000 simulated XRD patterns using stochastic gate-noise injection and local finite-shot measurement sampling. A separate calibration set of 20,000 simulated patterns is subsequently evaluated on an IBM quantum processor using the frozen quantum circuit. Hardware measurements are acquired in four independent parts, each providing 504 effective shots through 18 replicated six-qubit circuits, for 2016 cumulative effective shots per pattern. The hardware-derived simulated features are split into 16,000 training, 2,000 validation, and 2,000 test samples; fixed PMN statistics are estimated from the training subset only. Experimental XRD-CT patterns are processed with the same circuit, shot protocol, fixed PMN transformation, and retrained heads to generate quantitative spatial maps.
Figure 3: Architecture of the hybrid quantum-classical model. A 64-bin XRD pattern is amplitude encoded into a six-qubit parametrised quantum circuit comprising two repetitions of single-qubit rotations and ring entanglement. Measurements of the Pauli X , Y , and Z observables produce 18 quantum features, which are transformed using fixed post-measurement normalisation (PMN) and passed to a classical dual-head MLP. The weight-fraction head produces compositionally constrained outputs using a Softmax function, whereas the scale-factor head combines the quantum features with the pattern maximum and predicts normalised effective scale factors through linear outputs.
System
Phase
a (Å)
c (Å)
Crystallite size (nm)
Li-ion battery
LCO
2.815–2.820
14.160–14.210
100–1000
Li-ion battery
Graphite
2.450–2.460
6.680–6.750
100–1000
Li-ion battery
Al
4.045–4.065
–
100–1000
Li-ion battery
Cu
3.615–3.625
–
100–1000
SOFC
LSM
5.480–5.550
13.350–13.410
100–500
SOFC
YSZ
3.590–3.615
5.150–5.200
100–500
Table 1: Structural and microstructural parameter ranges used to generate the synthetic single-phase diffraction libraries for the Li-ion battery and SOFC systems.
Figure 4: Representative Rietveld refinement plots for the SOFC (left) and Li-ion battery (right) samples. Observed experimental data are shown as black dots, calculated fits as solid red lines, and the difference profiles ( yobs−ycalc ) as gray lines below. Bragg reflection positions for the respective constituent phases are displayed as vertical tick marks beneath the difference curves (SOFC: LSM, YSZ, NiO; Li-ion battery: Al, LCO, Graphite, Cu).
Figure 5: Weight-fraction prediction performance of the hardware-calibrated hybrid quantum neural network (HQNN) on the held-out simulated SOFC test set. Predicted and target weight fractions are shown for LSM, YSZ, and NiO. The dashed diagonal denotes ideal agreement ( y=x ); the coefficient of determination ( R2 ) and root-mean-square error (RMSE) are reported for each phase.
Figure 6: Scale-factor prediction performance of the hardware-calibrated HQNN on the held-out simulated lithium-ion battery test set. Predicted normalised effective scale factors are compared with their target values for LCO, graphite, Al, and Cu. The dashed diagonal represents ideal agreement ( y=x ), with phase-wise R2 and RMSE values reported above each panel.
4. Classical-head prediction for every pattern inside the complete experimental mask
1.3 ms (7,879 patterns)
14.0 ms (18,343 patterns)
14.0 ms (18,343 patterns)
Table 2: Measured processing times for the SOFC and lithium-ion battery workflows. Step 1 reports local wall-clock time for noise-injected pre-training on 200,000 simulated patterns. Step 2 reports IBM quantum-processing time for acquiring QNN features from 20,000 simulated patterns and all masked experimental patterns; queueing, local circuit compilation, and result-download time are excluded. Steps 3 and 4 report local CPU wall-clock times. The battery weight-fraction and scale-factor models share the same QNN pre-training and hardware feature-acquisition stages, whereas their classical heads are retrained and evaluated independently.
Figure 7: Weight fraction maps resolved for the SOFC XRD-CT dataset. Columns correspond to LSM, YSZ, and NiO. The first row contains Rietveld refinement maps, the second row contains calibrated HQNN predictions, and the third row shows the associated relative error maps between Rietveld and HQNN. Red circles and enlarged insets identify representative local structures used for direct visual spatial comparison.
Figure 8: Weight fraction maps for the Li-ion battery XRD-CT dataset. Columns show LCO, graphite, Al, and Cu. The first row contains Rietveld refinement maps, the second row contains calibrated HQNN predictions, and the third row shows the associated relative error maps between Rietveld and HQNN. Red circles and enlarged insets identify representative local structures used for direct visual spatial comparison.
Figure 9: Normalised scale factor maps for the Li-ion battery XRD-CT dataset. Columns correspond to LCO, graphite, Al, and Cu. The first row contains Rietveld refinement maps, the second row contains calibrated HQNN predictions, and the third row shows the associated relative error maps between Rietveld and HQNN. Red circles and enlarged insets identify representative local structures used for direct visual spatial comparison.
Training quantum neural networks (QNNs) on quantum hardware is currently bottlenecked by the cost of gradient estimation: standard parameter-shift methods require a number of circuit evaluations that grows quadratically with the number of trainable parameters, making hardware-based optimisation impractical beyond small system sizes. In this work, we introduce a training framework that reduces this cost to logarithmic in the number of qubits, making gradient-based QNN optimisation feasible on near-term hardware at increasing scales. Our framework combines three co-designed ingredients: (i) a structured, subspace-preserving Butterfly circuit architecture with O(nlogn) parameters and logarithmic depth; (ii) a layer-wise training strategy that confines on-hardware optimisation to one small, well-structured layer at a time; and (iii) a parallelised parameter-shift rule that exploits the commuting structure within each Butterfly layer to extract all gradients in a constant number of circuit executions. Together these reduce the number of distinct circuit evaluations per optimisation step from O(n2) to O(logn). We validate the framework on clinical data imputation using the MIMIC-III electronic health record dataset, a demanding benchmark sensitive to optimisation instability and model variance. Hybrid classical-quantum models are trained directly on IonQ Forte Enterprise trapped-ion hardware at 16 qubits without performance degradation relative to ideal or noisy simulation and via tensor-network simulation at 32 qubits, with 32-qubit inference executed on hardware. The resulting models match or exceed strong classical neural baselines in downstream patient survival prediction while exhibiting reduced variance across runs, demonstrating that the proposed framework enables practical, scalable QNN training under realistic hardware constraints.
Hybrid quantum-classical neural networks have emerged as a promising approach for leveraging quantum computing in machine learning while mitigating current hardware limitations. This paper presents Sim-HVQC, a hybrid Deep Quantum Neural Network that couples an adaptive, parameter-free SimAM weighting module with classical feature extraction to preserve class-discriminative information prior to encoding into a Variational Quantum Circuit (VQC). Previous studies are restricted to binary classification [1] [2] [3] [4] [5]. In contrast, the proposed framework is trained and evaluated on various multi-class datasets(MNIST, KMNIST, Fashion-MNIST, and EMNIST). The framework further demonstrates reproducibility, parameter efficiency, and interpretability through multi-seed evaluation, parameter analysis, and latent/quantum feature inspection. The source code is publicly available at https://github.com/Dilli822/ SimAM-HVQC
Dilli Hang Rai
Institute of Science and Technology Tribhuvan University, Kritipur, Nepal
Artificial intelligence has been transformed by deep neural networks, yet the search for new learning architectures continues. Quantum machine learning offers one such direction, and hybrid quantum neural networks, which combine classical neural-network components with quantum information processing units, have emerged as a practical framework for near-term quantum technologies. However, the rapid development of the field across diverse architectures, benchmarks and hardware assumptions makes it difficult to assess the utility of various proposals, identify where genuine advantages may arise, and determine how practitioners can use these models. While recent benchmarks caution that such gains have not yet been demonstrated at scale, theoretical work has identified tasks on which quantum models hold provable advantages, and hybrid approaches have delivered promising results on practical problems using deliberately compact quantum components and substantially fewer trainable parameters. Here, we review hybrid quantum neural networks for the machine-learning and quantum-machine-learning communities. We summarize their main theoretical and methodological foundations, survey some of the most promising architectures developed so far, and examine their implementation challenges and reported performance. By consolidating these perspectives, this review provides a structured view of the state of the field and helps identify promising paths for future research and application-driven development.
Terra Quantum AG, Kornhausstrasse 25, St. Gallen, 9000, Switzerland · Département de Physique Appliquée, Université de Genève, 24 Quai Ernest-Ansermet, Geneva, 1211, Switzerland