QuantumATK - New Release Notes

• On average 10x+ speed-up​ of the most time-consuming parts of bulk and NEGF device calculations, including SCF, bandstructure, PDOS, PLDOS, and transmission spectrum.
  
• Support for multi-node and multi-GPU with linear acceleration with the number of GPUs​.
• Shifting the run-time for 5,000 atoms with DFT or 30,000 atoms with Semi-Empirical NEGF from days to just a few hours.

• ~2x speed-up for all SCF & geometry optimization with MetaGGA, GGA, and Hubbard U for small/medium system sizes (a few hundred atoms).
• Up to ~5x faster PDOS, FatBandstructure and MAE analysis, with speed-up increasing with the number of projections.

• Framework for training new and fine-tuning universal MACE models for a specific system/process with improved accuracy.
• Interface for import, rapid testing and usage of emerging MLP models, based on DeepMD, ORB, SevenNet, MACE, CHGNet, etc.
• Multi-GPU acceleration of MLPs enables large scale MD simulations, e.g. 100,000 atoms with MACE and 1,000,000 atoms with MTPs.  

New methods for simulating room-temperature diffusion and extracting diffusion coefficients:

• Accelerated Collective Variable Hyperdynamics (CVHD) gives up to 300x speed-up over standard MD to capture rare events.
• Adaptive Kinetic Monte Carlo (AKMC) with new Lanczos and ARTn saddle search methods iteratively probes new states, possible transitions, and energy barriers between them.

• Support for complex substrate shapes, such as U-shaped and 2D materials, in deposition/etching processes​.
• ​MD simulations with variable time-step for accurate modeling of impact surface processes without having to use a small time-step for the entire simulation.

• New sequential IDDP method for setting up complex reaction paths​ in NEB.
• New techniques to improve and speed-up NEB reaction path optimization.​
• 2-3x faster refinement of transition state search and reaction barrier calculations when combining NEB with new Dimer and Lanczos methods for saddle point search.