A Concurrent Multiscale Framework Coupling Direct Simulation Monte Carlo and Molecular Dynamics Note: LLNL-JRNL-2006036 Journal: Journal of Computational Physics
Tim Linke, Dane Sterbentz, Niels Grønbech-Jensen, Jean-Pierre Delplanque, Jonathan Belof · 2025
Read A Concurrent Multiscale Framework Coupling Direct Simulation Monte Carlo and Molecular Dynamics Note: LLNL-JRNL-2006036 Journal: Journal of Computational Physics by Tim Linke, Dane Sterbentz, Niels Grønbech-Jensen, Jean-Pierre Delplanque, Jonathan Belof (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-01924.txt.
Introduction
Tim Linke, Dane Sterbentz, and three colleagues couple two simulation methods that see matter at very different scales, in a paper published in two thousand twenty-five. Molecular dynamics tracks individual atoms, while direct simulation Monte Carlo follows representative particles in a gas flow. Many problems, like hypersonic flight, need both views at once.
Earlier couplings decided cell by cell in a limited way. The new framework generalizes that decision and adds operators that translate between the two descriptions, lifting detail up and restricting it down across length and time scales. The authors verify the coupling against basic conservation laws. They then demonstrate it on a hypersonic flow case and weigh its accuracy gain over plain Monte Carlo simulation. Strengths, limits, and further uses are discussed. The goal is one seamless simulation where atoms matter locally but the flow is vast.
Catalog record
| CREATOR | Tim Linke, Dane Sterbentz, Niels Grønbech-Jensen, Jean-Pierre Delplanque, Jonathan Belof |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-01924
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Concurrent Multiscale Framework Coupling Direct Simulation Monte Carlo and Molecular Dynamics Note: LLNL-JRNL-2006036 Journal: Journal of Computational Physics Tim Linke, Dane Sterbentz, Niels Grønbech-Jensen, Jean-Pierre Delplanque, Jonathan Belof Abstract We present a new method to couple the Direct Simulation Monte Carlo (DSMC) algorithm with molecular dynamics (MD). The coupling approach generalizes prior coupling methods using a cell-based decision. The approach is supported by a lifting and restricting operator, which translate length and time scales between DSMC and MD. We verify the framework on basic conservation laws, and demonstrate its usability on a hypersonic flow example. Its accuracy gain is discussed in light of conventional DSMC simulations. Advantages and limitations are discussed, and additional use cases are highlighted. 1 Introduction When modeling a fluid flow field, an essential decision to make is how to treat deviations from equilibrium. Continuum mechanics is widely adopted in computational fluid dynamics, but many cases exist where large gradients occur in very short length and time scales, challenging the continuum assumption. A meaningful measure for the significance of these effects is captured in the Knudsen number; the ratio of the mean free path of a flow’s particle to a characteristic length. High Knudsen number flows are of great interest in fields such as reentry and low-orbit aerodynamics [ 1 ] , microfluidics [ 2 ] , plasmas [ 3 ] , and porous materials [ 4 ] . The validity of continuum mechanics at these conditions is found in the formulation of fluxes, which are assumed to be proportional to gradients of velocity and temperature. However, in high Knudsen number flows (Kn > 0.1), particle collisions occur infrequently over an observed distance and thus restrict the achievement of local equilibrium. Velocity and temperature gradients therefore contain significant fluctuations that are not captured in continuum mechanics. [ 5 ] To fully capture the non-continuum behavior, a microscopic analysis must replace the gradient-based approach. From a molecular perspective, deviations from equilibrium can be captured using velocity distribution functions f(x,v,t) , where x denotes the position and v the velocity of particles at time t . The Boltzmann equation describes the evolution of these distributions f with \frac{\partial f}{\partial t}+v\frac{\partial f}{\partial x}+\frac{F}{m}\frac{\partial f}{\partial v}=(\frac{\delta f}{\delta t})_{c}, (1) where F is an external force, and the right hand side is the collision operator. The differentials of f describe the change in phase space volume due to particle motion that is uncorrelated with collisions. Molecular interactions are described by the collision operator. The Boltzmann equation is established as the main description of moderate to high Knudsen number flows, and stands as a pillar to the kinetic theory of gases and plasmas [ 6 ] [ 7 ] . While its original formulation was designed for simple gases with short-range interactions, adjustments can be added for more complex behavior. These include evaporation and condensation effects [ 8 ] , chemical reactions and the inclusion of thermal radiation [ 9 ] . Unfortunately, the nonlinearity of these adjustments make the analytical solution of the Boltzmann equation infeasible for practical applications [ 10 ] . In response, various numerical methods have been developed to solve the Boltzmann equation [ 7 ] . The moment method involves expanding the distribution function in terms of its moments, such as density, momentum, and energy, which results in a hierarchy of coupled partial differential equations for these macroscopic quantities that can be truncated and solved [ 11 ] . The method excels near equilibrium conditions, but cannot state the accuracy of the approximated solution. In turn, the Knudsen number expansion method readily offers a solution at arbitrary specified Knudsen numbers by expanding the distribution in terms of the Knudsen number [ 7 ] . However, truncation errors are still inherent to the method. These are also encountered when solving the Boltzmann equation using the Chapman-Enskog expansion, which expands the distribution into an equilibrium and a non-equilibrium part [ 12 ] . While this method successfully recovers the Navier-Stokes equations of continuum mechanics, it performs poorly in conditions far from equilibrium. Lastly, the Direct Simulation Monte Carlo (DSMC) method is shown to accurately capture the widest range of Knudsen numbers [ 9 ] . The DSMC method tracks individual molecular motion by advancing particles collisionless, followed by probabilistic collision steps. By simplifying long-range interactions, it is well suited for high Knudsen number flows. Its initial limitation of high computational cost has long been overcome with current implementations fully capable of running on high-performance computing architectures [ 13 ] . Furthermore, its ability to handle complex geometries has established the method as a staple of gas dynamics. DSMC has seen extensive developments in the past, and its range of applicability is continually challenged. While methods such as the Chapman-Enskog expansion successfully recover continuum formulations, DSMC shows restrictions for low Knudsen number flows. Resolution of short-range interactions undermine the statistical approach to collisions and higher densities limit the computational advantage it has over other particle methods when computationally parallelized. It is the focus of this paper to improve upon the performance of DSMC by coupling it to molecular dynamics simulations. To achieve this, we implement a method that improves the accuracy of arbitrary regions in the domain by using a cell-based decision technique between the coupling of DSMC and MD. 2 Numerical Models 2.1 Direct Simulation Monte Carlo The Direct Simulation Monte Carlo method excels in large-scale particle simulations. It works by dividing the simulation into three steps. First, all particles undergo Newtonian motion in a freely flowing field. Then, boundaries are taken into account. Reflections, energy transfers, and adsorption may be considered, and surface interactions can be evaluated. Next, collisions are modeled. Rather than using direct atomic interactions, collisions are determined stochastically. Based on parameters such as relative velocity, a fraction of the particles present in a DSMC cell are chosen to collide using an acceptance-rejection scheme. If selected, the particles undergo the interaction based on a pre-determined collision model. Popular choices include the Variable Hard Sphere and the Variable Soft Sphere models [ 9 ] . The Variable Hard Sphere model treats particles as inelastic spheres with an infinite repulsive force upon collision. The Variable Soft Sphere model treats molecules also applies an infinite repulsive force when colliding, but treats the cross-section of particles variable to better match properties such as viscosity and diffusion [ 14 ] . Lastly, macroscopic properties are obtained by evaluating the moments of the distribution functions. Due to the stochastic collision operator, DSMC possesses the ability to represent many real particles in the flow by a few simulated particles. The underlying assumption is that particles similar in phase space will undergo similar movement and collisions, such that they can be grouped. The parameter f_{\textrm{num}}=\frac{\textrm{physical}}{\textrm{simulated}}\textrm{particles} has therefore established itself in the DSMC community. f_{\textrm{num}} is a major contributor to DSMC’s computational efficiency advantage over other particle methods. The DSMC method was originally designed for rarefied gas simulations by Bird [ 9 ] . It readily captures effects far from equilibrium through its distribution functions. More recently, it has also been applied successfully near continuum conditions such as hydrodynamic instabilities [ 15 ] [ 16 ] and turbulence modeling [ 17 ] [ 18 ]