Chapter 02:

Numerical methods for microfluidics

Xiangyu Hu
Technical University of Munich
 Possible numerical approaches

• Macroscopic approaches
– Finite volume/element method
– Thin film method
• Microscopic approaches
– Molecular dynamics (MD)
– Direct Simulation Monte Carlo (DSMC)
• Mesoscopic approaches
– Lattice Boltzmann method (LBM)
– Dissipative particle dynamics (DPD)
Possible numerical approaches

• Macroscopic approaches
 Macroscopic approaches
Finite volume/element method
• Solving Navier-Stokes (NS)
equation
Continuity equation
∇⋅v = 0
Interface/surface force
∂v 1 η 1
+ ∇ ⋅ vv = − ∇p + g + ∇ 2 v + Fs
∂t ρ ρ ρ
Momentum equation
Pressure gradient Gravity Viscous force

Pressure
– Eulerian coordinate used Velocity
– Equations discretized on a
mesh
– Macroscopic parameter and
states directly applied
 Macroscopic approaches
Finite volume/element method

• Interface treatments • Complex geometry

– Volume of fluid (VOF)
– Structured body fitted
• Most popular mesh
– Level set method • Coordinate
– Phase field transformation
• Matrix representing
– Unstructured mesh
1.0 1.0 0.64 0
• Linked list representing

1.0 0.95 0.32 0

0.11 0.07 0 0

Unstructured mesh

VOF description
 Macroscopic approaches
Finite volume/element method

• A case on droplet formation (Kobayashi et al 2004, Langmuir)

– Droplet formation from micro-channel (MC) in a shear flow
– Different aspect ratios of circular or elliptic channel studied
– Interface treated with VOF
– Body fitted mesh for complex geometry
 Macroscopic approaches
Finite volume/element method

• Difficulties in micro-fluidic simulations

– Dominant forces
• Thermal fluctuation not included
– Complex fluids
• Multi-phase
– Easy: simple interface (size comparable to the domain size)
– Difficult: complex interficial flow (such as bubbly flow)
• Polymer or colloids solution
– Difficult
– Complex geometry
• Easy: static and not every complicated boundaries
• Difficult: dynamically moving or complicated boundaries
 Macroscopic approaches
Thin film method

• Based on lubrication approximation of NS equation

Viscosity Film thickness

∂h
η − ∇ ⋅ (m(h)∇p ) = 0
∂t Mobility coefficient depends of
p = −σ∆h + V (h) boundary condition

Effective interface potential

Surface tension

h(x) Film
Solid
 Macroscopic approaches
Thin film method

• A case on film rapture (Becker et al. 2004, Nature materials)

– Nano-meter Polystyrene (PS) film raptures on an oxidized Si
Wafer
– Studied with different viscosity and initial thickness
 Macroscopic approaches
Thin film method

• Limitation

– Seems only suitable for film dynamics studies.

Possible numerical approaches

• Microscopic approaches
 Microscopic approaches
Molecular dynamics (MD)

• Based on inter-molecular forces

Potential of a molecular pair
∂u (rij )
Fi = ∑ Fij = ∑ − e ij
j ≠i j ≠i ∂rij
dp i Total force acted on a molecule
= Fi
dt
Molecule velocity
p
vi = i
mi

u ( rij ) Lennard-
Jones Fji
potential

j
Fij
i
rij
 Microscopic approaches
Molecular dynamics (MD)

• Features of MD
– Lagrangain coordinates used
– Tracking all the “simulated” molecules at the same time
– Deterministic in particle movement & interaction (collision)
– Conserve mass, momentum and energy
• Macroscopic thermodynamic parameters and
states
– Calculating from MD simulation results
• Average
• Integration
 Microscopic approaches
Molecular dynamics (MD)

• A case on moving contact line (Qian et al. 2004, Phys. Rev. E)

– Two fluids and solid walls are simulated
– Studied the moving contact line in Couette flow and Poiseuille
flow
– Slip near the contact line was found
 Microscopic approaches
Molecular dynamics (MD)

• Advantages
– Being extended or applied to many research fields
– Capable of simulating almost all complex fluids
– Capable of very complex geometries
– Reveal the underline physics and useful to verify physical
models
• Limitation on micro-fluidic simulations
– Computational inefficient computation load ∝ N2, where N is
the number of molecules
– Over detailed information than needed
– Capable maximum length scale (nm) is near the lower bound
of liquid micro-flows encountered in practical applications
 Microscopic approaches
Direct simulation Monte Carlo (DSMC)

• Combination of MD and Monte Carlo method

Translate a molecular Collision probability
Same as MD
ri → ri + v i ∆t proportional to velocity only

Number of pair trying

ρ 2πd 2 v ij Nc
for collision in a cell
M trial = max
Vc ∆t , ρ = , v ij = v i − v j
2 Vc
1 1
v i → ( v i + v j ) + v ij e
Molecular velocity 2 2 A uniformly
after a collision 1 1 distributed unit vector
v j → ( v i + v j ) − v ij e
2 2

cell
 Microscopic approaches
Direct simulation Monte Carlo (DSMC)

• Features of DSMC
– Deterministic in molecular movements
– Probabilistic in molecular collisions (interaction)
• Collision pairs randomly selected
• The properties of collided particles determined statistically
– Conserves momentum and energy
• Macroscopic thermodynamic states
– Similar to MD simulations
• Average
• Integration
 Microscopic approaches
Direct simulation Monte Carlo (DSMC)

• A case on dilute gas channel flow (Sun QW. 2003, PhD Thesis)
– Knudsen number comparable to micro-channel gas flow
– Modified DSMC (Information Preserving method) used
– Considerable slip (both velocity and temperature) found on
channel walls

Velocity profile Temperature profile

 Microscopic approaches
Direct simulation Monte Carlo (DSMC)

• Advantages
– More computationally efficient than MD
– Complex geometry treatment similar to finite volume/element
method
– Hybrid method possible by combining finite volume/element
method
• Limitation on micro-fluidic simulations
– Suitable for gaseous micro-flows
– Not efficiency and difficult for liquid or complex flow
Possible numerical approaches

• Mesoscopic approaches
 Mesoscopic approaches
• Why mesoscopic approaches?
– Same physical scale as micro- Macroscopic
fluidics (from nm to µm)
– Efficiency: do not track every N-S ρ r
molecule but group of molecules u
T
– Resolution: resolve multi-phase
fluid and complex fluids well
– Thermal fluctuations included Mesoscopic
Mesoscopic particle
– Handle complex geometry without
difficulty Increasing
• Two main distinguished methods LBM or DPD scale
– Lattice Boltzmann method (LBM)
– Dissipative particle dynamics
(DPD)
Microscopic
Molecule

r
v
MD or DSMC
 Lattice Boltzmann Method (LBM)
Introduction

• From lattice gas to LBM

– Does not track particle but distribution function (the
probability of finding a particle at a given location at a given
time) to eliminates noise
• LBM solving lattice discretized Boltzmann equation
– With BGK approximation
– Equilibrium distribution determined by macroscopic states

LBM D2Q9 lattice structure

Example of lattice gas collision
indicating velocity directions
 Lattice Boltzmann Method (LBM)
Introduction

• Continuous lattice Boltzmann equation and LBM

– Continuous lattice Boltzmann equation describe the
probability distribution function in a continuous phase space
– LBM is discretized in:
• in time: time step δt=1
• in space: on lattice node δx=1
• in velocity space: discrete set of b allowed velocities: f ⇒ set of
fi, e.g. b=9 on a D2Q9 Lattice

Equilibrium
distribution
Time step
Discrete velocities

Df ∂f ⎛ ∂f ⎞ f i (x, t ) − f i eq (x, t )
= + c ⋅ ∇f = ⎜ ⎟ f i (x + c iδ t , t + δ t ) = f i (x, t ) −
Dt ∂t ⎝ ∂t ⎠coll. τ
Continuous Boltzmann equation Lattce Boltzmann equation
i=0,1,…,8 in a D2Q9 lattice
Relaxation time
 Lattice Boltzmann Method (LBM)

• A case on flow infiltration (Raabe 2004, Modelling Simul. Mater. Sci. Eng.)
– Flows infiltration through highly idealized porous microstructures
– Suspending porous particle used for complex geometry
 Lattice Boltzmann Method (LBM)
Application to micro-fluidic simulation

• Simulation with complex fluids

– Two approaches to model multi-phase fluid by Introducing
species by colored particles
• Free energy approach: a separate distribution for the order
parameter
• Particle with different color repel each other more strongly than
particles with the same color
– Amphiphiles and liquid crystals can be modeled
• Introducing internal degree of freedom
– Modeling polymer and colloid solution
• Suspension model: solid body described by lattice points, only
colloid can be modeled
• Hybrid model (combining with MD method): solid body modeled
by off-lattice particles, both polymer and colloid can be modeled
 Lattice Boltzmann Method (LBM)
Application to micro-fluidic simulation

• Simulation with complex No slip

geometry
– Simple bounce back algorithm
• Easy to implement WALL
• Validate for very complex
geometries
• Limitations of LBM
Free slip
– Lattice artifacts
– Accuracy issues
• Hyper-viscosity
WALL
• Multi-phase flow with large
difference on viscosity and density
 Dissipative particle dynamics (DPD)
Introduction

• From MD to DPD
– Original DPD is essentially MD with a momentum conserving
Langevin thermostat
– Three forces considered: conservative force, dissipative force
and random force
Translation
dri 1
= pi Random number with
dt mi Momentum equation
Gaussian distribution
dp i
= FiC + FiD + FiR
dt
FiC = ∑ αϖ ijC eij , Fi D = ∑ − γϖ ijD e ij ⋅ v ij e ij , Fi R = ∑ σξijϖ ijR eij
j ≠i j ≠i j ≠i

Conservative force Dissipative force Random force

 Dissipative particle dynamics (DPD)

• A case on polymer drop (Chen et al 2004, J. Non-Newtonian Fluid Mech.)

– A polymer drop deforming in a periodic shear (Couette) flow
– FENE chains used to model the polymer molecules
– Drop deformation and break are studied

1 2 5 6

3 4 7 8
 Dissipative particle dynamics (DPD)
Application to micro-fluidic simulation

• Simulation with complex fluids

– Similar to LBM, particle with different color repel each other
more strongly than particles with the same color
– Internal degree of freedom can be included for amphiphiles or
liquid crystals
– modeling polymer and colloid solution
• Easier than LBM because of off-lattice Lagrangian properties
• Simulation with complex geometries
– Boundary particle or virtual particle used
 Dissipative particle dynamics (DPD)
Application to micro-fluidic simulation

• Advantages comparing to LBM

– No lattice artifacts
– Strictly Galilean invariant
• Difficulties of DPD
– No directed implement of macroscopic states
• Free energy multi-phase approach used in LBM is difficult to
implement
• Scale is smaller than LBM and many micro-fluidic applications
– Problems caused by soft sphere inter-particle force
• Polymer and colloid simulation, crossing cannot avoid
• Unphysical density depletion near the boundary
• Unphysical slippage and particle penetrating into solid body
 Dissipative particle dynamics (DPD)
New type of DPD method

• To solving the difficulties of the original DPD

– Allows to implement macroscopic parameter and states
directly
• Use equation of state, viscosity and other transport coefficients
• Thermal fluctuation included in physical ways by the magnitude
increase as the physical scale decreases
• Simulating flows with the same scale as LBM or even finite
volume/element
– Inter-particle force adjustable to avoid unphysical penetration
or depletion near the boundary
• Mean ideas
– Deducing the particle dynamics directly from NS equation
– Introducing thermal fluctuation with GENRIC or Fokker-
Planck formulations
 Dissipative particle dynamics (DPD)
Voronoi DPD

• Features
– Discretize the continuum hydrodynamics equations (NS
equation) by means of Voronoi tessellations of the
computational domain and to identify each of Voronoi
element as a mesoscopic particle
– Thermal fluctuation included with GENRIC or Fokker-
Planck formulations
dρ
= − ρ∇ ⋅ v Voronoi tessellations
dt
dv 1 F (1)
= g − ∇p + F +
dt ρ ρ

Isothermal NS equation in Lagrangian coordinate

 Dissipative particle dynamics (DPD)
Smoothed dissipative particle dynamics
(SDPD)

• Features
– Discretize the continuum hydrodynamics equations (NS
equation) with smoothed particle hydrodynamics (SPH)
method which is developed in 1970’s for macroscopic
flows
– Include thermal fluctuations by GENRIC formulation
• Advantages of SDPD
– Fast and simpler than Voronoi DPD
– Easy for extending to 3D (Voronoi DPD in 3D is very
complicate)
• Simulation with complex fluids and complex
geometries
– Require further investigations
 Summary

• The features of micro-fluidics are discussed

– Scale: from nm to mm
– Complex fluids
– Complex geometries
• Different approaches are introduced in the situation
of micro-fluidic simulations
– Macroscopic method: finite volume/element method and
thin film method
– Microscopic method: molecular dynamics and direct
simulation Monte Carlo
– Mesoscopic method: lattice Boltzmann method and
dissipative particle dynamics
• The mesoscopic methods are found more powerful
than others