arXiv:2609.33700
On Embedding Design in Quantum Physics-Informed Neural Networks
Ban Q. Tran, Nahid Binandeh Dehaghani, Susan Mengel, Rafal Wisniewski, A. Pedro Aguiar
Abstract
Quantum physics-informed neural networks (QPINNs) solve partial differential equations (PDEs) by training parameterized quantum circuits against physics-based residuals, yet the role of the embedding that maps coordinates into quantum states remains insufficiently understood. In this research, we introduce a unified embedding framework that formulates embedding as a functional transformation shaping the feature representation available to the variational circuit, and we propose two embeddings within it. The Linear Quantum Neural Network Trainable Embedding QPINN (LQNN-TE-QPINN) generates data-dependent features with an auxiliary quantum circuit and adds them to the input coordinates, instead of multiplying them as in our previous formulation. The adaptive-frequency QPINN (AdaFreq-QPINN) promotes the frequency-scaling factors of an arccos-Chebyshev encoding to trainable parameters. We evaluate both against direct and fixed analytical frequency encodings and alternative hybrid architectures on one- and two-dimensional Burgers equations. In one dimension, LQNN-TE-QPINN achieves a relative $L_2$ error of 0.0916 with 720 trainable parameters, compared with 0.4501 for direct-encoding QPINN and 0.1233 for a locally adaptive classical physics-informed neural network with approximately eleven times more parameters, while AdaFreq-QPINN attains the lowest error among the analytical embeddings with only four additional parameters. In two dimensions, LQNN-TE-QPINN attains the lowest training objective, although its solution error is not clearly separated from that of a data re-uploading QPINN. Under simulated hardware noise, LQNN-TE-QPINN retains the lowest absolute derivative error while degrading most relative to its noiseless baseline. These results demonstrate that embedding design substantially influences the approximation and optimization behavior of QPINNs.
AI Summary
The paper examines how coordinate embeddings affect approximation and optimization in quantum physics-informed neural networks (QPINNs). It formulates embeddings as feature transformations and compares direct encoding, fixed arccos–Chebyshev frequency sequences, and two adaptive constructions: AdaFreq, which trains the frequency scales, and LQNN-TE, which adds features measured from an auxiliary quantum circuit to the input coordinates before encoding, replacing a previous multiplicative rule. Evidence comes from classical quantum-circuit simulations of one- and two-dimensional Burgers equations, trained with L-BFGS over three seeds, with circuit-sizing, training-budget, simulated-noise, and finite-shot studies. At the selected one-dimensional configuration, LQNN-TE achieves relative $L_2$ error $0.0916\pm0.0021$ with 720 parameters, versus $0.4501$ for direct encoding and $0.1233\pm0.0739$ for an 8,051-parameter adaptive classical PINN. AdaFreq reaches $0.3262$ with four additional parameters, the lowest error among the analytical embeddings evaluated at that configuration. In two dimensions, LQNN-TE has the lowest training objective, but its unnormalized solution error at 200 epochs, $40.58\pm2.18$, is not clearly separated from data re-uploading’s $39.11\pm1.42$. Training objectives and solution errors frequently rank models differently. LQNN-TE retains the lowest absolute second-derivative error under individual simulated noise channels but degrades most relative to its noiseless baseline. Stated limitations include simulation-memory constraints, evaluation on training rather than held-out grids, configuration-dependent performance, and an unresolved two-dimensional ordering affected by incomplete convergence and run-to-run numerical variation. The evidence is restricted to Burgers benchmarks and does not establish intrinsic noise robustness.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 23 pages