The continued development of large-scale scientific computing has enabled unprecedented simulations of complex physics. These simulations rely on advanced computational techniques, particularly developed for accuracy, efficiency and scalability on thousands of computational cores. In addition, the mathematical details, computational algorithm, software design, implementation and optimisation of these codes are often subject specific. Therefore, interfacing these algorithms with the objective of exploring multi-physics problems can be a significant algorithmic and computational challenge. In this project, a coupler library is developed which is used to interface massively-parallel algorithms. The development vehicle for this library was multiscale simulations of continuum-molecular coupled flows.

Many current and emerging technologies rely on flow phenomena which span a wide range of length scales - from the atomistic to the macroscopic level. In such multi-scale flows, the motion of individual atoms near the surface of a wall can determine the bulk motion of the fluid, and involve dynamics which cover more than than six orders of magnitude. A variety of applications, such as electro-kinetic flow in microfluidic channels or the wetting of superhydrophobic surfaces, depend on this complex interaction between the nanoscale and macroscale physics. Thus, the ability to model and understand such multiscale phenomena can lead to significant advances in microfluidic control, and drag reducing surface coatings, among a host of other multiscale technologies.

Current computational approaches have focused on either simulating the dynamics of large groups of atoms, or the macroscopic flow behaviour independently. At the small end of the spectrum, Molecular Dynamics (MD) simulations track the motion of individual atoms based on their mutual interactions inside a confined domain. This approach has been successful in determining the transport and shear properties of liquids [5,6]. On the other hand, Direct Numerical Simulations (DNS) employ continuum-level assumptions to solve the non-linear, partial differential equations for the fluid velocity at all points within a large computational domain. For instance, the DNS code $ \mathcal {T}rans$ $ \mathcal {F}low$ by Zaki and co-workers [13,14,15] has been used in investigating purely hydrodynamic phenomena such as the interaction of vortical modes with turbulence. However, to date, few attempts have been made to combine both methodologies within a single framework, in particular in the context of large-scale, massively-parallel scientific computing.

The goal of the current project is to develop, validate, and optimise a state-of-the-art computational framework to efficiently couple simulations of various physical phenomena. The particular development within the project was focused on coupling MD simulations with high-fidelity DNS for realistic simulations. The combined MD-DNS code provides new capabilities and insight into how phenomena at the molecular scale, such as microscopic surface textures and coatings, can influence macroscopic hydrodynamic motions. The ability to resolve the entire range of scales will also obviate the need for simplified, and often inaccurate, constitutive models for molecular scale effects.