Quantum Computing-Based Wasserstein Generative Adversarial Networks for Airfoil Parameterization

Abdur Rahman Ovy, Xiaosong DuarXiv 2026 · arXiv

TL;DR

Researchers have developed a hybrid Quantum-Classical Wasserstein Generative Adversarial Network (Q-WGAN) to parameterize airfoil geometries. By using a parameterized quantum circuit as a generator, the model achieves comparable accuracy to classical GANs while using significantly fewer parameters and better capturing the underlying statistical distribution of optimal airfoil shapes.

What Problem Does This Paper Address?

In aerodynamic design, "parameterization" is the process of defining an airfoil's shape using a set of variables. Traditional methods like B-splines or Free-Form Deformation (FFD) often require a large number of variables to ensure geometric flexibility. However, this creates a massive "design space" where many configurations are physically unrealistic or aerodynamically poor, leading to inefficient optimization and convergence failures.

Generative Adversarial Networks (GANs) have recently emerged as a solution by learning to generate only "realistic" airfoils, effectively shrinking the design space. However, classical GANs can be parameter-heavy and prone to training instabilities. This paper explores whether quantum computing—specifically parameterized quantum circuits (PQCs)—can offer a more efficient and effective way to represent these complex geometric distributions.

Methods

The authors propose a hybrid Q-WGAN architecture:

  • The Generator (Quantum): A 3-qubit parameterized quantum circuit. It encodes a 3D latent vector using $R_Y$ and $R_Z$ rotations. The circuit consists of 12 variational layers, each using $R_X, R_Y, R_Z$ rotations and a ring of CNOT gates to create entanglement. The quantum output is passed through a small classical "head" (a single fully connected layer) to map the results to 18 FFD control points.
  • The Discriminator (Classical): A standard multi-layer perceptron acting as a Wasserstein "critic." It estimates the Earth-Mover distance between the real airfoil distributions and the generated ones.
  • Training: The model uses the Wasserstein loss function with weight clipping to maintain stability, a standard technique to prevent the "mode collapse" common in traditional GANs.

The most striking aspect of this method is its parameter efficiency. The Q-WGAN generator uses only 180 trainable parameters, whereas the classical WGAN generator used for comparison requires 679,442 parameters.

Datasets

The study utilizes a dataset of 992 optimal airfoil designs generated through high-fidelity aerodynamic shape optimization (minimizing drag under lift and volume constraints).

  • Each airfoil is represented by 18 independent FFD control points.
  • For training, the authors selected a subset of 20 representative airfoils using conditioned Latin Hypercube Sampling (cLHS) to ensure a fair and challenging comparison between the quantum and classical models.

Key Insights and Ideas

  • Quantum Compactness: The PQC generator manages to represent the complex manifold of "optimal airfoils" with a fraction of the parameters required by classical neural networks. This suggests that quantum circuits may have a higher "expressive capacity" for certain geometric tasks.
  • Distribution Fidelity over Raw Accuracy: While both models achieved similar fitting accuracy, the Q-WGAN was superior at capturing the distribution of the data. It achieved lower Jensen-Shannon Divergence and Fréchet Distance, meaning the airfoils it generates are statistically more similar to the "real" optimal airfoils than those from the classical GAN.
  • Representation Learning: t-SNE visualizations revealed that the two models learn different internal representations. The classical WGAN tends to focus on the boundaries of the design space, while the Q-WGAN discovers a more compact, central distribution that overlaps more strongly with the baseline data.
  • Geometric Robustness: The authors noted that a ~92% accuracy in FFD control points translated to a >99% accuracy in actual airfoil coordinates. This implies that the physical shape is somewhat resilient to small fluctuations in the underlying parameters.

Key Results

  • Fitting Accuracy: The Q-WGAN achieved 92.1% accuracy on FFD control points and 99.2% on airfoil coordinates, nearly identical to the classical WGAN (92.29% and 99.26% respectively).
  • Distribution Metrics: The Q-WGAN outperformed the classical model across several key metrics:
  • JSD: 0.496 (Quantum) vs. 0.505 (Classical)
    • Fréchet Distance: $1.67 \times 10^{-5}$ (Quantum) vs. $3.16 \times 10^{-5}$ (Classical) — a 47.2% improvement.
  • Wasserstein Distance: $5.09 \times 10^{-4}$ (Quantum) vs. $1.23 \times 10^{-3}$ (Classical) — a 52.6% reduction.

Limitations and Future Work

  • Small Training Set: The models were trained on only 20 airfoils. While this demonstrates the models' ability to learn from limited data, future work should test scalability with much larger datasets.
  • Simulation vs. Hardware: The study was conducted using the default.qubit simulator in PennyLane. Real-world quantum hardware is noisy (NISQ era), and the performance of Q-WGAN under actual quantum noise remains to be seen.
  • Integration: The current work focuses on parameterization. The next logical step is integrating this Q-WGAN directly into an active aerodynamic optimization loop to see if the "quantum-compact" design space leads to faster or better design discoveries.

Summary updated

Abstract

Airfoil geometry parameterization is essential in aerodynamic design optimization. Optimization search in the large space suffers from low efficiency and modest feasible region, which could further incur failures to convergence. Generative adversarial networks (GAN), aiming to learn the statistical distribution of existing data, pave a new avenue for shape generation and airfoil parameterization. GAN manages to implicitly reduce design space by generating only smooth, realistic airfoils. Meanwhile, quantum computing has been emergingly exhibiting great potential for achieving superior performance in specific types of engineering problems but has not been introduced to airfoil parameterization. To fill this gap, this work develops quantum computing-based Wasserstein GAN (Q-WGAN) and investigates the performance for airfoil parameterization. The Q-WGAN adopts parameterized quantum circuits as the generator to promote the generative performance, while the discriminator remains classical. The developed Q-WGAN is demonstrated on a dataset of optimal airfoils parameterized by free-form deformation (FFD) control points, resulting from 992 airfoil aerodynamic design optimizations. Results show that Q-WGAN achieves 92.1% mean fitting accuracy on the optimal FFD control points and 99.2% accuracy on airfoil coordinates, comparable with the classical WGAN performance of 92.28% on FFD control points and 99.26% on airfoil coordinates. Q-WGAN achieves lower Jensen-Shannon divergence, Fréchet distance, and Wasserstein distance than the classical WGAN, indicating an improved agreement with the reference airfoil distribution. Therefore, this work presents a convincing evidence that Q-WGAN makes a viable option for parameterizing airfoils and possesses a great potential for other relevant engineering applications.
BibTeX
@misc{ovy2026quantum,
  author = {Abdur Rahman Ovy and Xiaosong Du},
  title = {{Quantum Computing-Based Wasserstein Generative Adversarial Networks for Airfoil Parameterization}},
  year = {2026},
  eprint = {2611.20602},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.20602},
}
Figure