PhysicsNeMo Workshop

This workshop provides hands-on approaches on how to use NVIDIA PhysicsNeMo, a framework that combines physics and partial differential equations (PDEs) with artificial intelligence (AI) to build robust models. Participants will learn about the PhysicsNeMo sym utilities to infuse physics during training and inference.

Quick Start with Docker

Prerequisites

• Docker with NVIDIA Container Toolkit
• NVIDIA GPU with CUDA support

Option 1: Docker Compose (Recommended)

# Clone the repository
git clone https://github.com/hasethinvd/physicsnemo_workshop.git
cd physicsnemo_workshop

# Build and start the container
docker compose up --build

# Access Jupyter Lab at http://localhost:8888

Option 2: Docker CLI

# Build the image
docker build -t physicsnemo-workshop .

# Run with GPU support
docker run --gpus all -it \
    -p 8888:8888 \
    -v $(pwd):/workspace/physicsnemo_workshop \
    --shm-size=16gb \
    physicsnemo-workshop

# Access Jupyter Lab at http://localhost:8888

---

Workshop Contents

• Lab 1: Physics Informed FNO for Nonlinear Shallow Water Equations

  Physics informing of a data-driven model using numerical derivatives (PINO) during training.

• Lab 2: Transolver - Physics-Aware Transformers for PDEs

  Introduction to Transolver and Physics-Attention for efficient transformer-based PDE solvers.

• Lab 3: Reservoir Simulation with xMGN

  Multi-scale graph networks for reservoir simulation.

• Lab 4: Learning the flow field of Stokes flow

  Train MeshGraphNet to learn Stokes flow and improve accuracy via physics-informed inference.

• Lab 5: Diffusion Models for FWI

  Conditional diffusion models for full waveform inversion.