Breakeven Complexity: Simulated PDE Trajectories for Neural Solver Benchmarking

20,000 training and 1,000 test trajectories for each of three partial differential equations (Navier-Stokes, Kuramoto-Sivashinsky, Gray-Scott) simulated via Exponax. The dataset was created by author 'yijingz' for the paper 'Breakeven complexity: A new perspective on neural partial differential equation solvers' and was last updated on Hugging Face in May 2026.

Use Cases

• Benchmarking neural PDE solvers based on simulated Navier-Stokes equation trajectories.
• Evaluating model performance on chaotic systems using Kuramoto-Sivashinsky equation data.
• Training physics-informed neural networks on reaction-diffusion systems based on Gray-Scott equation fields.
• Studying the 'breakeven complexity' of neural solvers across different PDE types and scales.

Strengths

• Contains 20,000 training trajectories per PDE, providing substantial data for model training.
• Includes a separate test set of 1,000 trajectories per PDE for reliable evaluation.
• Covers three distinct and challenging PDE systems (Navier-Stokes, Kuramoto-Sivashinsky, Gray-Scott).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific model architectures.
• Description metadata is limited; actual data quality and simulation parameters require manual inspection.

Provenance

Source

Hugging Face dataset by author yijingz.

Collection Method

Simulated via the Exponax method.

Time Range

null

Freshness

Last updated 2026-05-19 05:55:28; freshness should be verified.

Geography

null