Quantum Extreme Reservoir Computing Demonstrates Classification of Engineering Data Using Self-Consistent Field Theory

The challenge of classifying complex microstructures in polymer alloys has driven Arisa Ikeda, Akitada Sakurai, and Kae Nemoto, from Keio University and the Okinawa Institute of Science and Technology Graduate University, to explore the potential of quantum machine learning. Their research applies quantum extreme reservoir computing (QERC) to images generated by self-consistent field theory, offering a novel approach to analysing materials data. This work distinguishes itself from previous quantum machine learning studies, which largely focused on standard datasets, by tackling a problem with direct relevance to materials science and engineering. By investigating the impact of computational parameters on classification accuracy, the team produces visual diagrams linking model outputs to actual polymer behaviour. This establishes a crucial step towards integrating advanced learning techniques into the field of materials informatics, potentially accelerating the discovery of new and improved materials.

Quantum machine learning is expected to offer new opportunities to process high-dimensional data efficiently by exploiting the exponentially large state space of quantum systems. In this work, researchers apply quantum extreme reservoir computing (QERC) to the classification of microstructure images of polymer alloys generated using self-consistent field theory (SCFT). Previous quantum machine learning efforts have primarily focused on benchmark datasets, but this work demonstrates the applicability of QERC to engineering data with direct materials relevance. Through numerical experiments, the influence of key parameters on classification accuracy is examined, contributing a novel application of QML techniques to a challenging materials science problem.

Polymer Alloy Microstructure Classification via QERC

The study pioneers the application of extreme reservoir computing (QERC) to classify microstructure images of polymer alloys, addressing a materials science challenge beyond traditional quantum machine learning datasets. Researchers generated a comprehensive dataset using self-consistent field theory (SCFT) simulations, creating images representing different polymer alloy microstructures. To account for the inherent variability in SCFT simulations, the team generated one unique microstructure for each random seed, ensuring reproducibility and capturing the diversity of possible morphologies.

This innovative approach yielded 5,780 images, systematically distributed across a parameter space defined by volume fraction (f) and repulsive interaction coefficient (χ). The experimental setup involved systematically varying f (from 0.3 to 0.5 with increments of 0.0125) and χ (ranging from 0.1 to 1.0 with steps of 0.1), keeping the degree of polymerization N fixed at 25. This grid-based approach allowed for the creation of a phase diagram, mapping the relationship between parameters and resulting microstructures. Crucially, the researchers established clear boundaries between four distinct microphase types , hexagonal, gyroid, lamellar, and disordered , utilizing established theoretical frameworks. The classification task framed the problem as predicting the structural type from the microstructure image, treating the images as data points for the QERC model.

The study meticulously prepared labelled data, generating 24 training images and 10 test images for each of the 170 grid points within the parameter space. This careful data preparation formed the foundation for evaluating the feasibility and performance of QERC on realistic engineering datasets. The resulting classifications are visualized as phase diagrams, offering an intuitive connection between model outputs and underlying material behaviour. Beyond assessing classification accuracy, the research team developed a method for visualizing the predicted phase diagrams, enabling assessment of class boundaries and providing insights beyond traditional metrics. This visualization technique transforms the classification results into a two-dimensional representation of the parameter space, revealing the model’s ability to discern subtle differences in microstructure and accurately predict phase behaviour.

Scientists Results

Scientists have successfully applied quantum extreme reservoir computing (QERC) to the classification of microstructure images generated from polymer alloys using self-consistent field theory (SCFT). This work demonstrates the applicability of quantum machine learning to engineering data possessing direct materials relevance. Experiments focused on classifying polymer alloy microstructures into four distinct phases: disordered, hexagonal, gyroid, and lamellar, as depicted in phase diagrams illustrating transitions in polymer morphology. The resulting phase classifications establish a clear connection between quantum model outputs and the underlying material behavior.

The research team systematically examined the influence of key computational parameters on classification performance, varying the number of qubits, the sampling cost measured by the number of measurement shots, and the configuration of the quantum reservoir itself. Through numerical experiments, scientists were able to reconstruct phase diagrams from the quantum model outputs, providing a direct interpretation of model performance and accurate phase boundary prediction. These diagrams visually represent the relationships between interaction parameters and resulting polymer morphologies, offering insight into how the quantum model generalizes across different materials parameter spaces. Measurements confirm that QERC can effectively process high-dimensional data derived from SCFT simulations.

The team’s work establishes a foundation for integrating quantum learning techniques into materials informatics, paving the way for more efficient materials design and discovery. By demonstrating the utility of QERC in this context, the researchers bridge the gap between theoretical quantum learning and practical materials engineering applications. Furthermore, the study highlights the importance of encoder design and its impact on the interpretability of quantum feature representations. The ability to visualize the results as phase diagrams allows for a direct understanding of how the model arrives at its classifications, enhancing trust and facilitating further refinement.

QERC Classifies Polymer Alloy Microstructures Accurately Quantum extreme

This research successfully applied quantum extreme reservoir computing (QERC) to classify microstructure images of polymer alloys generated through self-consistent field theory, demonstrating the potential of quantum machine learning with data directly relevant to materials science. The study achieved high-precision classification using a relatively small number of qubits, approximately seven, and explored how parameters like measurement shots influence performance. Importantly, the QERC model demonstrated effective generalization even when classifying data points between different phases, accurately reproducing phase structures within the parameter space. The work’s visualization of the classification results as a phase diagram not only confirms accuracy but also offers insights into the internal workings of quantum machine learning models, potentially guiding future design and interpretability improvements. The authors acknowledge that performance decreased with a smaller training dataset relative to the test dataset, but consider the results satisfactory given this constraint. Furthermore, the physically-based dataset created for this study provides a flexible benchmark for evaluating the robustness and generalization capabilities of learning models, and future work could build upon this foundation to integrate learning techniques more broadly into materials informatics.