Universally Adaptable Multiscale Molecular Dynamics (UAMMD). A native-GPU software ecosystem for complex fluids, soft matter, and beyond

Abstract

We introduce UAMMD (Universally Adaptable Multiscale Molecular Dynamics), a novel software infrastructure tailored for mesoscale complex fluid simulations on GPUs. The UAMMD library encompasses a comprehensive range of computational schemes optimized for the GPU, spanning from molecular dynamics to immersed boundary fluctuating-hydrodynamics. Developed in CUDA/C++14, this header-only open-source software serves both as a simulation engine and as a library with a modular architecture, offering a vast array of independent modules, categorized as interactors (neighbor search, bonded, non-bonded and electrostatic interactions, etc.) and integrators (molecular dynamics, dissipative particle dynamics, smooth particle hydrodynamics, Brownian hydrodynamics and a rather complete array of Immersed Boundary -IB- schemes). UAMMD excels in schemes that couple particle-based elastic structures with continuum fields in different regions of the mesoscale. To that end, thermal fluctuations can be added in physically consistent ways, and fast modes can be eliminated to adapt UAMMD to different regimes (compressible or incompressible flow, inertial or Stokesian dynamics, etc.). Thus, UAMMD is extremely useful for coarse-grained simulations of nanoparticles, and soft and biological matter (from proteins to viruses and micro-swimmers). Importantly, all UAMMD developments are hand-to-hand validated against experimental techniques, and it has proven to quantitatively reproduce experimental signals from quartz-crystal microbalance, atomic force microscopy, magnetic sensors, optic-matter interaction and ultrasound.

Program summary

Program Title: UAMMD

CPC Library link to program files: https://doi.org/10.17632/srrt2y5s4m.1

Developer's repository link: https://github.com/RaulPPelaez/UAMMD/

Licensing provisions: GPLv3

Programming language: C++/CUDA

Nature of problem: The key problem addressed in computational physics is simulating the behavior of matter at various scales, encompassing both discrete (particle-based) and continuum (field-based) approaches. The challenge lies in accurately and efficiently modeling interactions at different spatio-temporal scales, ranging from atomic (microscopic) to fluid dynamics (macroscopic). This complexity is further amplified in mesoscale regions, where different physics domains intersect, necessitating advanced computational techniques to capture the nuanced dynamics of systems such as colloids, polymers, and biological structures.

Solution method: The present solution consists in the creation of UAMMD (Universally Adaptable Multiscale Molecular Dynamics), a CUDA/C++14 library designed for GPU-accelerated complex fluid simulations. UAMMD offers a flexible platform that integrates discrete particle dynamics with continuum fluid dynamics. It supports a variety of computational schemes, each tailored for specific spatio-temporal regimes. The library's modular architecture allows for the seamless introduction of new algorithms and easy integration into existing codebases.

Additional comments including restrictions and unusual features: UAMMD's design emphasizes modularity and GPU-native architecture, optimizing computational efficiency and flexibility. However, its focus on GPU acceleration and low level nature means it requires compatible hardware and familiarity with CUDA programming. While UAMMD is versatile in handling various physical regimes, it currently lacks certain standard force field potentials and multi-GPU support. Nonetheless, its ongoing development and open-source nature promise continual enhancements.