Development of computational multiscale interface for scientific understanding of

interfacial evolutions

Avik Saha

Department of Mechanical and Industrial Engineering

Indian Institute of Technology, Roorkee, India

Email id: aviksaha129@gmail.com

Abstract
Majority of the natural or industrial processes are multiscale in nature. But often the smaller
scale features are computed empirically in practical world, which hinders our basic
understanding of the process or phenomena. In spite of their small existence in spatial and
temporal scale, they often provide utmost influence over the scientific understanding of the
phenomena. Due to high runtime and cost, micro scale simulations over domain of engineering
is not practical with present computational advancement. For that reason, we need to go for
multiscale simulations where microscale simulations will run for limited spatial and temporal
scale and provide required inputs to the macroscale simulations. In present study we have
proposed similar multiscale simulation of thin film for some elemental grey processes like
microlayer dynamics in nucleate boiling, contact line motion during drop spreading and
breaking and making of interfaces during merging of bubbles. For these simulations a coupling
between molecular simulation in LAMMPS and finite volume based solver in OpenFOAM has
been proposed with the help of a indigenously coupler library, which will facilitate the
communication between the solvers and ensure that the solvers run in unison. A flow chart of
the computational steps and the communication has also been provided.

Background and Motivation

In multiscale modeling of any process or phenomena, different scales of the problem are dealt
with different techniques. These types of modeling techniques are having a wide range of
applications in different fields of science and technology. Most of the natural processes are
actually consisting of elemental processes of different spatial as well as time scales. Though
some of these elemental processes are limited to their small spatial and temporal domain but
they are having the utmost contribution to the overall phenomena. For example, in the case of
geologic CO2 sequestration and extraction of shale gas, the gas pipeline extends in the order
of kilometers but the geologic pores and cracks of the coal matrix are in the order of nanometer
to micrometers. The whole process is dependent on the diffusion through these pores and
cracks, which needs to be modeled accurately to capture the overall scenario [1]. Moreover,
modeling such micro-nano scale phenomena requires totally different modeling
methodologies. For these reasons in twenty-first century multiscale modeling is emerging over
a wide range of disciplines starting from Material science, Biomedicine, Energy, Environment,
Mechanical to even in Astrophysics. All these disciplines are having their own range of scales
to be dissolved as shown in Figure 1. Here, one can clearly see in all the disciplines as we
move toward the smaller spatial scale corresponding time scale also reduces. This signifies
that if we want to model any microscale phenomena, the corresponding time scale also needs
to be dissolved in microseconds.

The majority of engineering applications are focused on bulk behavior of the system which
can be grossly predicted by employing macroscopic analysis, obeying homogeneous physics
over a continuous domain. But every physical phenomenon percolates from elemental scale
dynamics involving multiscale transfer processes. Practice of unimodal discretization of
continuum fails in cases, requiring an understanding of physics spanning widely varying scales
and their mutual exchange. Macroscale predictions largely depend on transfer functions which
mimics smaller scale features through empirical knowledge. The majority of the times, these
constitutive equations are not reliable when extrapolated to adjoint domain and makes
researchers handicapped for not going towards elemental discretization in macroscale. On
the other hand, microscale simulations of engineering issues are very costly in terms of
computational cost. Even with the rapid advancement of computational resources, it is
practically impossible to simulate the engineering domain with microscale discretization. For
these reasons, researchers came up with a new style of modeling, which is known as
multiscale modeling.

Figure 1. Scales of multiscale problems for different disciplines [2]

In multiscale simulations most important part is the interconnection between the different
region based parallel simulations. For this purpose, some library is required which will help in
creating a communication bridge between the solvers. Presently two libraries are available in
open domain:

1. Multiscale Universal Interface (MUI) [3]

2. CPL Library [4]

MUI consist of only header files written in C++, but CPL library is written in Fortran. These
organises the massage passing between the solvers using some Message Passing Interface
(MPI) eg. OpenMPI, MPICH.

Problem Statement and Proposed Approach

Three elemental grey processes have been selected to be simulated with the help of
multiscale simulations, which are not only rich in terms of physics but important in different
engineering applications.

Prediction of microlayer dynamics in nucleate boiling

Boiling of micro and macro layer is assessed as the prime reason of growth of nucleation.
Accurate prediction of layers is important to obtain bulk growth rate of nucleating bubble, in
which microlayer demands elemental scale understanding which feeds macroscale bulk
growth of vapour mass. Due to ultra-short existence of such microfilm both in space and time,
measurement of this thickness is very difficult. Due to ultra-short existence of such microfilm
both in space and time, measurement of this thickness is very challenging. Researchers tried
to propose correlations based on temperature information [5] or in-situ sophistications like
laser interferometry [6] for the same. These correlations are reliable but difficult to extrapolate
in situations demanding complex physics like deformable triple point in liquid-liquid boiling.
Calculation of this film thickness from the fundamentals is needed to be done from multiscale
point of view to tackle such problems. In proposed multiscale simulation, microlayer will be
simulated with the help of molecular dynamics (MD) as shown in figure 2 (a). MD will be
calculating microfilm thickness, the heat flux and the rate of phase change at different location.
The same information will be passed to the volume of fluid (VOF) based finite volume solver
with the help of coupling library. Also, there will a thin region of overlap in between two solver
domains. VOF solver will solve for the rest of the domain and determining the overall bubble
shape and dynamics. VOF solver will also calculate the forces and stresses at different
locations and will pass the same to the MD solver as boundary conditions.
 Liquid Pool
Vapor
Bubble

Wall molecules
Water molecules Liquid
Microfilm
Finite volume domain
Nucleation
Molecular dynamics domain Site
Contact line
Two phase interface
Overlapping region

Liquid Pool Thin liquid Wall

film

Vapor Vapor Heat Supply

Bubble Bubble (a)

Liquid
Drop

Contact line

(c) (b)

Figure 2: Schematic of multiscale modelling of (a) microlayer dynamics in nucleate boiling,

(b) drop spreading and (c) merging of bubbles

Contact line motion during drop spreading

Most critical part in droplet spreading over solid surface is the assessment of motion of contact
line and evaluation of dynamic contact angle. In the era of miniaturisation with advancement
of microfluidics and micro structured surfaces, study of droplet spreading for fluid dispensing
and heat transfer demands more attention in the motion of contact line. Researchers have
proposed different models of dynamic contact angle [7] based on the experimental
observations, but understanding the physical phenomena at triple point from the fundamental
direction is still lacking. Microscale apprehension of molecular behaviour in the wedge-shaped
liquid film surrounding contact line can be studied using molecular dynamics and overall drop
spreading needs to be modelled in macroscopic frame work. Hence, motion of contact line
can be a perfect example of multi scale simulation. In present study, VOF based finite volume
and molecular dynamics will be coupled with a coupling library, keeping an overlap zone for
information exchange shown in figure 2 (b). Dynamic contact angle will be calculated in
microscopic frame work and contact line forces will be exchanged with finite volume solver for
evaluation of drop shape and location during spreading.

Breaking and making of interfaces during merging of bubbles

Bubble-bubble merging is a very important phenomenon, which takes crucial role in deciding
the bubble size and population inside any liquid column. Bubble merging consists of different
stages like approach, liquid drainage and film rupture. At the onset of rupture, the film becomes
very thin (~ 100Å) at which molecular forces cannot be neglected. In such scenario, van der
Waal force becomes dominant in stabilising the converging interfaces. Due to smaller length
scale, last stage of film rupture cannot be captured efficiently with macroscopic methods
directly. To take care of such intermolecular forces in macroscopic simulations, generally,
additional pressure term is added in the momentum equation as adjoining pressure which is
calculated based on constitutive formulations, mentioned in literature. To avoid this adhocism,
in the present proposal, a multiscale simulation is planned where the thinnest portion of the
liquid film between the bubbles will be simulated via molecular dynamics and rest interface
will be captured through finite volume solvers. Information like thin film stress and curvature
of converging interfaces will be shared through an exchange zone between the solvers.
Schematic diagram of the domain decomposition and subsequent calculation plan is shown
in figure 2 (c).

Figure 3. Schematic of molecular dynamics continuum interface proposed by O'Connell and

Thompson [7]
Developed multiscale model through coupling library will help in prediction of many more
critical physical issues involving mass, phase, momentum and energy barrier.

Brief Literature Review

In 1995 O'Connell and Thompson [8] first proposed the algorithm of the multiscale modeling
for coupling molecular dynamics and continuum. For a demonstration of the algorithm, they
have considered the startup flow of Newtonian fluid in Couette cell. They have been divided
the domain into two parts of molecular dynamics and continuum domain with some
overlapping region as shown in figure 2. In between two domains one hybrid solution interface
(HSI) has been considered. From the HIS toward continuum domain MD domain has been
extended large enough to screen the HSI from the effects of the termination. To have proper
coupling among the domains the averaged velocity of 0th bin of MD domain has been supplied
to the continuum domain as the velocity boundary condition. On the contrary, the momentum
flux from the continuum domain is supplied to the MD domain. This type of coupling is more
commonly known as Domain decomposition, where the overall domain is subdivided into sub-
regions of macroscopic and micro solver. For proper stable coupling of the regions, selection
of information to be exchanged is very important. In case of continuum solver Dirichlet or
Neumann type boundary information is extracted from the other solver (Figure 6). For this
information transfer averaging and smoothing is required. But on the contrary supplying
information from macro domain to the microdomain is much complicated. Because most of
the cases the macro scale properties are not valid on the smaller scale. In those cases, special
reconstruction operators or special distribution function are required.

Another popular strategy of multiscale coupling is Heterogeneous Multiscale Method. Here at

the nodes of the macroscale domain, microscale domain is placed. For this reason, this type
of modeling is also known as Embedded type multiscale modeling. This type of modeling is
mostly used for simulation of complex fluids and microchannel flow. For this type of simulation,
the macro scale solver uses the general governing equations but the stress terms and the flow
boundary conditions are supplied by the microscale simulations. Now for evaluation of these
parameters with the help of the micro-mesoscale solvers, the flow conditions need to be
supplied from the macro scale solver. Borg et al. [9] has utilized this type of strategy for the
simulation of converging-diverging nanochannel flow (Figure 4). Such modeling is formally
known as Internal-flow Multiscale Modelling (IMM). They have also placed the microdomains
at the nodes of the domain as shown in Figure 4. Macro-scale simulation is done to enforce
the mass and momentum conservation over the channel. They have reported a maximum
speed up to sixth order with multiscale simulation as compared to full domain molecular
dynamics simulation.
 Figure 4: Schematic for Multiscale simulation of nanochannel proposed by Borg et al. [9]

Another special type of multiscale strategy has been first proposed by Kevrekidis et al. [10]
where modeling where the macroscale solver hardly exists. This multiscale simulation strategy
is commonly known as Equation Free Method (EFM), which totally depends on the microscale
solvers. As performing microscopic simulation over the full domain is computationally to much
expensive, in this method microscopic simulation is performed over distributed patches of
space as well as time. Simulations are done for very small amount of time at a coarser time
interval. This is more formally known as time burst. Here the macroscopic solver does not
have any closed form of equation. It just does the interpolation, extrapolation, and integration
of these microscopic solvers to give the complete solution. Once the simulations of a time step
are completed, the macroscale solver after aggregating all the microscale simulation
extrapolates the data to generate the information about the next time step via extrapolation.
This can be done with the help of forward Euler’s scheme or any other higher-order scheme,
in case to case basis. After the extrapolation, this information is supplied to the microscale
solver to generate new configurations based on them.

EFM can help us to find the flaws in the macroscale governing equation, and guide us to make
the macroscale solver more efficient. This type of methodology is really helpful for the
comparatively newer domain of research, where stable macroscale solver is yet to be
established. But for that we need to know the type, order of unavailable governing equation
and also the parameters based on which the governing equation can be formed. This
information can be obtained from the experiments and the basic understanding of the
phenomena.
 Figure 5: Schematic of Equation Free Method proposed by Kevrekidis et al. [10]

Ongoing Work

Works has already been started to establish a platform for multiscale simulation of two phase
system. To start with initially a 1D heat conduction problem has been taken. At the center of
the domain a small SPH (Spherical Particle Hydrodynamics) domain has been taken and rest
domain has been simulated with Finite Difference Method. Domain length has been taken as
15 units. For data transfer a small overlapping region has also been kept at the two ends of
the SPH domain as shown in the figure 6.

Figure 6: Domain description and Algorithm for 1D FDM-SPH coupled heat conduction
problem.
FDM domain has been discretised in 7+7 nodes and the SPH domain has been discretised in
25 particles. For the 4 end particles (2 from each end) of SPH temperature data has been
taken from the FDM as boundary condition. Similarly, for the Last node of FDM at temperature
boundary condition is taken from SPH simulation. At the extreme end of the domain periodic
boundary condition is applied. Computation steps of both the solver are shown in the figure 6.
Both the solver codes are written in C++ language. For information exchange here Multiscale
Universal Interface has been used.

Figure 7: Result for 1D FDM-SPH coupled heat conduction problem.

In figure 7 temperature profile for the whole domain has been shown combining the results of
FDM and SPH simulation. Initially the left part of the FDM was kept at 0 temperature and right
part was kept at temperature 1. Different particles of the assigned with different temperatures
ranging from 0 to 1.5. After few simulation steps the temperature profile took a smooth profile
as shown in the figure 7. This result confirms proper coupling and information exchange
between the solvers.

For the simulation of proposed problem statement also both micro and macro scale
simulations are required. For micro scale simulation at present most popular open source
software is LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator) and on the
other hand OpenFOAM is one of the most common CFD solver package. For this reason,
LAMMPS-OpenFOAM has been selected to be coupled for the multiscale simulations with the
help of a coupler library.
Figure 8: Flow chart of Coupled Macro-Micro Simulation steps
 Figure 9: Equations of FVM and MD simulations

In figure 9 major equations for both the solvers are enlisted. These equations from the different
solvers are dependent on each other in case of multiscale simulation. Hence the solver codes
need to be modified to enable communication among them and ensure proper data transfer
for smooth running of the multiscale simulation. A flow chart (Figure 9) has been prepared to
show which data need to be transferred to and from the coupling library. This also shows the
whole sequence of the computational steps of both the solvers and at which step the data
transfer should occur.

References

[1] Wei, X.R., Wang, G.X., Massarotto, P., Golding, S.D. and Rudolph, V., 2007. Numerical
simulation of multicomponent gas diffusion and flow in coals for CO2 enhanced coalbed methane
recovery. Chemical engineering science, 62(16), pp.4193-4203.

[2] Groen, D., Zasada, S.J. and Coveney, P.V., 2013. Survey of multiscale and multiphysics
applications and communities. Computing in Science & Engineering, 16(2), pp.34-43.

[3] Tang, Y.H., Kudo, S., Bian, X., Li, Z. and Karniadakis, G.E., 2015. Multiscale universal interface:
a concurrent framework for coupling heterogeneous solvers. Journal of Computational
Physics, 297, pp.13-31.

[4] Smith, E. and Trevelyan, D., 2016. cpl-library.

[5] Cooper, M.G. and Lloyd, A.J.P., 1969. The microlayer in nucleate pool boiling. International
Journal of Heat and Mass Transfer, 12(8), pp.895-913.

[6] Koffman, L.D. and Plesset, M.S., 1983. Experimental observations of the microlayer in vapor
bubble growth on a heated solid.

[7] Cox, R.G., 1986. The dynamics of the spreading of liquids on a solid surface. Part 2.
Surfactants. Journal of Fluid mechanics, 168, pp.195-220.
[8] O’connell, S.T. and Thompson, P.A., 1995. Molecular dynamics–continuum hybrid computations:
a tool for studying complex fluid flows. Physical Review E, 52(6), p.R5792.

[9] Borg, M.K. and Reese, J.M., 2017. Multiscale simulation of enhanced water flow in
nanotubes. MRS Bulletin, 42(4), pp.294-299.

[10] Kevrekidis, I.G., Gear, C.W., Hyman, J.M., Kevrekidid, P.G., Runborg, O. and Theodoropoulos,
C., 2003. Equation-free, coarse-grained multiscale computation: Enabling mocroscopic simulators
to perform system-level analysis. Communications in Mathematical Sciences, 1(4), pp.715-762.