Adapting Quantum Machine Learning for Energy Dissociation of Bonds

TL;DR

The paper tackles the challenge of accurately predicting bond dissociation energies (BDEs) by performing a reproducible benchmark that pits classical ML models against quantum ML approaches on a chemically curated dataset. A six-qubit Qiskit Aer setup with ZZFeatureMap encoding and RealAmplitudes evaluates five quantum model families (VQR, QSVR, QNN, QCNN, QRF) alongside strong classical baselines, using a comprehensive feature set and standard metrics. Results show QCNN and QRF achieving accuracy competitive with RF and MLP in the mid-range BDEs, revealing complementary strengths among quantum architectures while exposing challenges at distribution tails. The work outlines a concrete pathway toward near-chemical accuracy, emphasizing high-fidelity labels, physics-informed representations, multi-fidelity Delta-learning, active learning, and calibrated ensembles, with quantum components offering useful inductive biases as complementary tools for quantum-enhanced molecular property prediction, potentially reaching MAEs on the order of $1\ \,\mathrm{kcal/mol}$ in favorable regimes.

Abstract

Accurate prediction of bond dissociation energies (BDEs) underpins mechanistic insight and the rational design of molecules and materials. We present a systematic, reproducible benchmark comparing quantum and classical machine learning models for BDE prediction using a chemically curated feature set encompassing atomic properties (atomic numbers, hybridization), bond characteristics (bond order, type), and local environmental descriptors. Our quantum framework, implemented in Qiskit Aer on six qubits, employs ZZFeatureMap encodings with variational ansatz (RealAmplitudes) across multiple architectures Variational Quantum Regressors (VQR), Quantum Support Vector Regressors (QSVR), Quantum Neural Networks (QNN), Quantum Convolutional Neural Networks (QCNN), and Quantum Random Forests (QRF). These are rigorously benchmarked against strong classical baselines, including Support Vector Regression (SVR), Random Forests (RF), and Multi-Layer Perceptrons (MLP). Comprehensive evaluation spanning absolute and relative error metrics, threshold accuracies, and error distributions shows that top-performing quantum models (QCNN, QRF) match the predictive accuracy and robustness of classical ensembles and deep networks, particularly within the chemically prevalent mid-range BDE regime. These findings establish a transparent baseline for quantum-enhanced molecular property prediction and outline a practical foundation for advancing quantum computational chemistry toward near chemical accuracy.

Figures (6)

• Figure 1: (a) BDE-db Dataset, (b) feature designing, engineering, and scaling to capture atomic, bond, and environmental descriptors relevant to dissociation energetics, (c) Traning of Quantum and classical models, (d) Prediction and evaluation of the model
• Figure 2: Comparison of Actual vs. Predicted BDE Values Across five quantum models: VQR, QSVR, QRF, QCNN, and QNN.
• Figure 3: Comparison of Actual vs. Predicted BDE Values Across classical models: SVR, RF and MLP.
• Figure 4: (a) The top panel shows absolute errors (in kcal/mol) versus actual BDE values. All models demonstrate a U-shaped error pattern, with errors minimized around 80-90 kcal/mol and increasing at both lower and higher BDE values. Horizontal reference lines at approximately 5 and 10 kcal/mol indicate error thresholds. (b) The middle panel presents squared errors versus actual BDE, showing a similar U-shaped pattern but with amplified differences at the extremes due to the squared nature of the metric. Values are capped at 1000 (kcal/mol)². (c) The bottom panel displays relative errors (percentage), again exhibiting the U-shaped trend. This visualization highlights that prediction errors are proportionally larger for smaller BDE values, with some errors exceeding 50% at the lower end of the BDE spectrum.
• Figure 5: Illustration of the distribution of (top) absolute errors, (middle) squared errors (restricted to values $<$ 1000), and (bottom) relative errors (restricted to values $<$ 100%) for the VQR, QSVR, QRF, QNN, and QCNN models. Dashed lines represent tolerance thresholds at 5 and 10 kcal/mol (top), as well as at 5% and 10% relative error (bottom).