Redefining
Large-scale Atomistic Simulation
GPUMD implements a concise and unified expression for force, virial, and heat current applicable to many-body potentials, effectively breaking through key bottlenecks in parallel efficiency and physical consistency of traditional molecular dynamics methods, providing a solid theoretical and algorithmic foundation for high-performance computing of complex many-body systems on GPUs.
Continuously developed the Neuroevolution Potential (NEP) machine learning potential method, successfully fusing the high precision of first-principles calculations with the high computational efficiency of empirical potentials, achieving fast calculations for complex material systems, making high-precision molecular dynamics simulations at the billion-atom scale possible.
Supported Force Fields
Comprehensive support ranging from empirical potentials to machine learning potentials.
ML Potentials
-
Neuroevolution Potential
Machine learning potential based on neuroevolution algorithms, achieving a perfect balance of accuracy and speed.
-
Deep Potential
Supports Deep Potential molecular dynamics simulations.
Empirical Potentials
Wide Range of Applications
From fundamental physics research to frontier engineering applications, GPUMD provides reliable computational support.
Mechanical Properties
Stress & Strain
Radiation Damage
Defects Evolution
Phase Transition
Structure Change
Shock Simulation
High Velocity
Short-range Order
Local Structure
Ion Transport
Diffusion
Chemical Reactions
Bond Breaking
Tensorial Properties
Anisotropy
Heat Transport
Thermal Cond.
Thermodynamics
Free Energy
Start High-Precision Billion-Atom Simulations
Explore the infinite possibilities of the material world with the extreme performance of GPUMD.