Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E...

60
Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing Ming (CAS), Weiqing Ren (NYU) and Zhijian Yang (RIT). Supported by ONR, DOE, AFOSR, ARO and NSF.

Transcript of Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E...

Page 1: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Multiscale Modeling and Hybrid Algorithms

Weinan E

Princeton University

Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton),Pingbing Ming (CAS), Weiqing Ren (NYU) and Zhijian Yang(RIT).

Supported by ONR, DOE, AFOSR, ARO and NSF.

Page 2: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Hybrid, multi-physics algorithms have had a lot ofsuccesses

I Quantum mechanics-Molecular mechanics (QM-MM)methods (1975, Warshel and Levitt)

I Quasicontinuum method (1996, Tadmor, Ortiz and Phillips)

I Kinetic-hydrodynamic methods

I ......

Page 3: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

There are still many fundamental issues

I Interface conditions

I Consistency

I Stability

Major difficulty:

I Microsopic models, such as molecular dynamics and electronicstructure models, are very poorly understood.

I Effect of noise or fluctuations?

I In some cases, connection between macro and micro models.

Page 4: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Coupled kinetic-hydrodynamic simulation

Dynamics of cracks

Error at the atomistic-continuum interface

Stability

Page 5: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Strategy:

I Use kinetic equation (e.g. Boltzmann) to resolve the shocks

I Use hydrodynamic (e.g. Euler) elsewhere.

t

x

t

x

Page 6: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Kinetic theory: f = f (x, v, t)

∂t f + v · ∇f =1

εC (f ).

ρ =

∫fdv , ρu =

∫f vdv , E =

∫f|v|22

dv .

∂t

ρρuE

+∇ · F = 0

F =

R3

f

vv ⊗ v12 |v|2v

dv

No approximations yet!

Page 7: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

When ε ¿ 1, the distribution function f is close to localequilibrium states, or the local Maxwellians,

M(x, v, t) =ρ(x, t)

(2πθ(x, t))3/2exp

(−(v − u(x, t))2

2θ(x, t)

)(1)

with θ being the absolute temperature.

Page 8: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Coupled scheme:

I Kinetic-based finite volume method over the whole region – toguarantee conservation.

I Local Maxwellian approximation inside hydrodynamic region –kinetic scheme.

I Appropriate interface condition.

Page 9: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Macro side: Kinetic scheme – finite volume scheme based onkinetic equations (similar in spirit to Lattice Boltzmann methods(LBM))

(ρ, ρu,E ) = (ρj , ρjuj ,Ej), x ∈ (xj−1/2, xj+1/2]

The finite volume scheme takes the form:

ρn+1j −ρn

j

∆t +

(F

(1)j+1/2

−F(1)j−1/2

)∆x = 0,

(ρu)n+1j −(ρu)nj

∆t +

(F

(2)j+1/2

−F(2)j−1/2

)∆x = 0,

En+1j −En

j

∆t +

(F

(3)j+1/2

−F(3)j−1/2

)∆x = 0

Fj+1/2 = (F(1)j+1/2,F

(2)j+1/2,F

(3)j+1/2)

T

Page 10: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Fj+1/2 = F+j+1/2 + F−j+1/2,

F±j+1/2 =

R±f (x∓j+1/2, v , t)

vv2

12v3

dv .

To leading order, we have:

f (x , v , t) ∼ M(x − vt, v , tn)

Page 11: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

F± =

ρuA±(S)± ρ

2√

πβB(S)

(p + ρu2)A±(S)± ρu

2√

πβB(S)

(pu + ρue)A±(S)± ρ

2√

πβ(

p

2ρ+ e)B(S)

A± =1 + erf(S)

2, B(S) = e−S2

, S =u√2RT

, p = ρRT .

Page 12: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Inside the kinetic region – use solutions of the kinetic equation:

Fk+1/2 =1

τ

∫ tn+τ

tn

dt∫

R+

f (x−k+1/2, v , t)

vv2

12v3

dv

+

R−f (x+

k+1/2, v , t)

vv2

12v3

dv

.

(2)

Page 13: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

At the interface – hybrid:

f (x , v , t) = M(x−k+1/2, v , t), v > 0

Fk+1/2 =1

τ

∫ tn+τ

tn

dt∫

R+

M(x−k+1/2, v , t)

vv2

12v3

dv

+

R−f (x+

k+1/2, v , t)

vv2

12v3

dv

.

Page 14: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

0 0.2 0.4 0.6 0.8 10

0.5

1

1.5

2

2.5

0 0.2 0.4 0.6 0.8 1−0.2

−0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Figure: solid line – kinetic equation; ‘o’ – coupled kinetic/gas dynamicssimulation; ‘+’ – gas dynamics equation.

Page 15: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Coupled kinetic-hydrodynamic simulation

Dynamics of cracks

Error at the atomistic-continuum interface

Stability

Page 16: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Multiscale Computational models

Continuum

Atomistic

Atomistic and continuum models:

I models at the atomic scale: molecularstatics, molecular dynamics

I continuum models: elasticity,elastodynamics, etc.

Main ideas:

I Atomistic models to resolve local defects

I Continuum models away from the defects

I Matching at the interface

Page 17: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Existing Coupling Methods

Some existing methods:

I quasicontinuum method (1996)

I finite temperature quasicontinuum (2005)

I coarse-grained molecular dynamics (1998,2005)

I bridging scale decomposition (2003)

I heterogeneous multiscale methods (2003)

I bridging domain method (2004)

I many others ...

Page 18: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

The continuum model

Elastodynamics:ρu = ∇ · σ.

I u: displacement.

I σ = σ(∇u): stress.

Formulation of Conservation Laws:

wt + fx + gy = 0.

w =

ε11

ε12

ε21

ε22

ρ0v1

ρ0v2

, f =

−v1

0−v2

0−σ11

−σ21

, g =

0−v1

0−v2

−σ21

−σ22

,

Page 19: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

The atomistic model

Reference position and displacement:

ui = ri − Ri .

Molecular dynamics

mui = −∇ui V .

Embedded atom potential:

V =1

2

∑φ(rij) +

∑E (ρi ), ρi =

j

ρ(rij).

Page 20: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Coupling of the two models

j j+1

k

k+1

t= nσ .

t= nσ .

t= nσ .

t= Σ f ij

Continuum Region Atomistic Region

Coupling method:

I In the continuum region:constitutive relation

I at the atomistic/continuuminterface: boundary conditions.

Page 21: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Finite volume formulation

Average velocity:

ρ0vJ+ 12,K+ 1

2=

1

|AJK |∑

i∈AJK

mivi ,

Momentum balance:

ρ0d

dtvJ+ 1

2,K+ 1

2=

∂AJK

t.

Computing the traction at the cell edges:

I In the continuum region: approximated by a continuum solver(Roe scheme) t = σ · n.

I At the atomistic/continuum interface: computed based on theatomic position t =

∑i∈AJKj /∈AJK

fij .

Page 22: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Overall computational procedure

∆ x δ x

∆ tδ t

Page 23: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Interface condition I: Linearizing the atomic interaction

Partition of the system:

u = (uA,uC ).

I uA: displacement in the atomistic region.

I uC : displacement in the continuum region.

Mechanical equilibrium:

∇uCV (uA,uC ) = 0, uC = RCA(uA).

Linearization:

V ≈ V (uA,RCA(uA)) +1

2

(uC − RCA(uA)

)TDCC

(uC − RCA(uA)

).

Page 24: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Interface condition II: boundary condition for molecularstatics

Harmonic approximation:

DCAuA + DCCuC = 0, uC = BCAuA.

Boundary map:

BCA = −D−1CCDCA = GCAG−1

AA .

Lattice Green’s function:

−∑

j

Di−jGj ,k = δik .

Di ,j ,Gj ,k are elements of matrix D,GDi−j = Di ,j due to translational invariance

Page 25: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Interface condition III: approximate solution at theinterface

Hamilton’s equation:

muA = −∇uAV (uA,BCAuA)− DACuC ,

muC = −DCCuC − DCAuA.

Initial condition is interpolated from the continuum solution:

uC (0) = uC , vC (0) = vC .

Page 26: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Interface condition IV: boundary condition for MD

Decomposition of the displacement at the interface:u

C= u(1)

C+ u(2)

C.

u(1)C

= D−1/2CC sin(D

1/2CC t)(v

C− BCAv

A) + cos(D

1/2CC t)(u

C− BCAu

A)

u(2)C

=

∫ t

0α(τ)u

A(t − τ)dτ.

Displacement in the atomistic region:

muA = −∇uAV

(uA,uC

)−∫ t

0θ(τ)vA(t − τ)dτ − θ(t)u(0)

Page 27: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Boundary conditions at finite temperature

Partition of the system:

u = (uA,uC ), v = (vA, vC )

The generalized Langevin equation:

muA = −∇uAV −

∫ t

0Θ(τ)uA(t − τ)dτ + R(t).

R(t) : stationary Gaussian process.

⟨R(t)R(s)T

⟩= kBTΘ(t − s).

Page 28: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

1D shock propagation

−40 −30 −20 −10 0 10 20 30 40

0

0.005

0.01

0.015

0.02

0.025

Str

ain

Before the shock leaves the atomistic region

−40 −30 −20 −10 0 10 20 30 40

0

0.005

0.01

0.015

0.02

0.025

Str

ain

After the shock leaves the atomistic region

Solution in the atomistic regionSolution at the continuum region

Solution in the atomistic regionSolution at the continuum region

−40 −30 −20 −10 0 10 20 30 40−0.005

0

0.005

0.01

0.015

0.02

0.025Before the shock enters the atomistic region

−40 −30 −20 −10 0 10 20 30 40−0.005

0

0.005

0.01

0.015

0.02

0.025After the shock enters the atomistic region

Solution in the atomistic regionSolution at the continuum region

Solution in the atomistic regionSolution at the continuum region

Page 29: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Crack propagation: Preparation

Displacement from the anisotropic elasticity solution (Sihand Liebowitz 1968):

u(x , y) = KI

[p1s2s2−s1

φ(z) + p2s1s1−s2

ψ(z).]

K cI =

√πν2

1√(b11b22/2d0)(

√b22/b11+(2b12+b66)/(2b11))

.

I K cI is chosen based on the Griffith criterion.

I s1, s2 are roots of characteristic equation reduced from thestress function.

I bij : compliance parameters of corresponding plane stress state.

I p1 = a11s21 + a12 − a16s1, p2 = a11s

22 + a12 − a16s2.

I aij : compliance parameters of corresponding plane strain state.

Page 30: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Relaxation of the crack KI = K cI

Page 31: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Relaxation of the crack KI = 1.2K cI

Page 32: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Crack front waves in the continuum region

Mesh size ∆x = ∆y = 6√

2a0.Time step ∆t = 8δt.

Page 33: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Scattering of the elastic wave off the crack tip

crack under mode I loading.

Page 34: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Velocity and displacement at the cross sections

Displacement and velocity ahead of the crack tip.

Page 35: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Pulse

Dynamic loading condition.

Page 36: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Coupled kinetic-hydrodynamic simulation

Dynamics of cracks

Error at the atomistic-continuum interface

Stability

Page 37: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Quasicontinuum method

I An adaptive mesh and model refinement procedureI Based on linear finite elementsI Representative atoms define the triangulationI Near defects, the mesh becomes fully atomisticI Local (continuum) and nonlocal (atomistic) regions

Page 38: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Consistency

I Consistency in the bulk: For simple systems, the two modelsshould produce consistent results.

I Consistency at the local-nonlocal interface.

Page 39: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Consistency between atomistic and continuum models

Nonlinear elasticity theory:

E (u) =

ΩW (∇u)dx

W (·) = stored energy density.In linear elasticity, W = a quadratic function of ∇u.

E (y1, · · · , yN) =∑

i ,j

V2(yi , yj) +∑

i ,j ,k

V3(yi , yj , yk) + · · ·

Question: Can we relate W to the atomistic model?

Page 40: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

The Cauchy-Born rule

Given A, a 3× 3 matrix, W (A) =?

Deform the crystal uniformly: yj = xj + Axj = (I + A)xj

W (A) = energy density of deformed unit cell, computed accordingto the given atomistic or electronic structure model.

Page 41: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Validity of Cauchy-Born rule: ConsistencyOne dimension model: xk = kε.Assume: yk = xk + u(xk) and u is a smooth function.

V =1

2

i 6=k

V0(yi − yk) (3)

≈ 1

2

i

k 6=i

V0(1 +du

dx(xi ))kε

=∑

i

W

(du

dx(xi )

)ε ≈

∫W

(du

dx(x)

)dx ,

where

W (A) =1

k

V0((1 + A)kε)

X. Blanc, C. Le Bris and P. L. Lions (2002) considered generalcase, including some QM models.

Page 42: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Validity of Cauchy-Born rule: Counterexample

Example: Lennard-Jones potential, next nearest neighborinteraction

I Triangular lattice, Cauchy-Born rule is valid

I Square lattice, Cauchy-Born gives negative shear modulus(unstable), can’t speak of elasticity theory.

Page 43: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Validity of Cauchy-Born rule: Stability

I Continuum level (Born criteria) – Elastic stiffness tensor ispositive definite

I Atomic level (Lindemann criteria) – Phonon spectra(dispersion relation for the lattice waves) remain “positivedefinite”

I Electronic level – Disperson relation for the charge-densitywaves and spin waves

Page 44: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Phonon stability condition

Acoustic branch: dynamics of the Bravais lattice

Optical branch: relative motion of the internal degrees of freedom

1st Brillouin zone: Voronoi cell of the origin of the dual lattice

Page 45: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Consistency in the bulk

Under these conditions, the solutions to the Cauchy-Borncontinuum model and the atomistic model are close.

I E and Ming (2007), molecular mechanics models.

I E and Lu (2008), several classes of quantum mechanicsmodels.

These conditions are sharp!!Violation of the stability conditions signals onset of plasticdeformation or structural (or electronic) phase transformation.Related work of Ju Li, S. Yip et al. (Λ-criterion),R. Elliott et al.

Page 46: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Consistency at the loca-nonlocal interfaceThe issue of “ghost force” (e.g. in quasicontinuum methods)

−10 −8 −6 −4 −2 0 2 4 6 8 10−0.2

−0.15

−0.1

−0.05

0

0.05

0.1

0.15Ghost forces with 5th nearest neighbor interaction

Original QCQuasi−nonlocalLinear construction

I Left side: using continuum model based on the Cauchy-Bornrule (effectively a nearest neighbor model).

I Right side: Using full atomistic model, next nearest neighborinteraction.

Page 47: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Explicit solution for quadratic potential (Ming and Yang, 09)

0 5 10 15 20 25 30 35−0.3

−0.2

−0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

N

D+(y

qc−

x)

1. The deformation gradient has O(1) error at the interface.

2. The influence of the ghost force decays exponential fast awayfrom the interface.

3. Away from an interfacial region of width O(ε| log(ε)|), theerror in the deformation gradient is of O(ε) (see also recentwork of Dobson and Luskin).

Page 48: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Removing the ghost force

Ghost force may induce numerical artifacts (e.g. plasticdeformation) at the interface.

I Force-based approach (Tadmor et al., Miller, Dobson andLuskin)

I Quasi-nonlocal atoms (Jacobson et al.)

I Geometrically consistent scheme (E, Lu and Yang)

Classical numerical analysis viewpoint:

I Truncation error = O(ε) in a weak sense

I Stability conditions (similar to the ones discussed above)

Uniform O(ε) accuracy for smooth solutions.

Page 49: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Ghost force for the coupled OF-DFT/EAM method(Choly, Lu, E, Kaxiras)

-0.03

-0.02

-0.01

0

0.01

-0.03

-0.02

-0.01

0

0.01

−3

∆ρ (A−3

) ∆ρ (A−3

)

1 nm

0.14A−3

A

(d)

0.22

(c)

(b)(a)

Page 50: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Loss of fluctuations

Example: Coupled KMC-continuum models of epitaxial crystalgrowth (Schulze, Smereka and E):

I Around the step-edges, use KMC, since fluctuations areimportant

I On the terraces, use continuum (e.g. diffusion) models

Page 51: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Mean position and variance of step edge

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

0 200000 400000 600000 800000 1e+06

σe

D/F

+++++++++

+

33333333

3

3

+++++++++

+

+++++++++

+

Figure 6: The time and space averaged surface adatom density for the KMC simulations(diamonds) and the hybrid scheme (crosses) using cell-widths M = 20, 25, 40.

0

0.005

0.01

0.015

0.02

0 200000 400000 600000 800000 1e+06

σρ

D/F

++++++++++

33

33333

33

3

++++++++++ ++++++

+++

+

Figure 7: The standard deviation (in time) of the surface averaged adatom density as afunction of the ratio D/F . The solid curve (diamonds) are from the KMC simulations andthe remaining curves are for the hybrid scheme with cell-widths M = 20, 25, 40 sites percell.

21

Other examples: See work of Garcia, Bell and Donev, et al.

Page 52: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Coupled kinetic-hydrodynamic simulation

Dynamics of cracks

Error at the atomistic-continuum interface

Stability

Page 53: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Stability of coupled continuum/MD methods

Work of Weiqing Ren (NYU)

I Example of fluids

I General strategy: Domain decomposition (with overlap)

I Coupling:

Page 54: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

III

z=−az=−b

z=az=b

z=1

z=−1u=0

u=0

I

II

In continuum region (I, III): ρut − µuzz = 0

In particle region (II): md2xj

dt2 = f j

Four coupling schemes:

1. velocity(MD)-velocity(C),

2. velocity(MD)-flux(C),

3. flux(MD)-velocity(C),

4. flux(MD)-flux(C)

Page 55: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Particular features as a domain decomposition method

I The MD (molecular dynamics) domain is very small.

I Statistical error cannot be avoided

Page 56: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Numerical solutions for equilibrium states

0 0.5 1 1.5 2

x 104

0

0.02

0.04

t / T

||u|| 2

0 0.5 1 1.5 2

x 104

0

0.02

0.04

t / T

||u|| 2

0 0.5 1 1.5 2

x 104

0

0.02

0.04

t / T

||u|| 2

0 0.5 1 1.5 2

x 104

0

0.2

0.4

t / T

||u|| 2

Upper panel: velocity-velocity Upper panel: velocity-fluxLower panel: flux-velocity Lower panel: flux-flux

Page 57: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

0

1

2

u

−100 −50 0 50 1000

1

2

z

u

0 2000 4000 6000 80000

0.1

0.2

t / Tc

e u

Page 58: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Steady-state calculation: Tc = ∞Amplification factors g for the four schemes:

I velocity-velocity, flux-velocity

un(z) =n∑

i=1

gn−iξi1− z

1− a, 〈‖un‖2〉 ≤

(1

3(1− g2)

)1/2

σv

ξi : Statistical errors in velocity BC; σv =⟨ξ2i

g = a(1−b)b(1−a) for velocity-velocity; g = a

a−1 for flux-velocity

I velocity-flux, flux-flux

un(z) =n∑

i=1

gn−iξ(z − 1)

g = b−1b for velocity-flux; (g > 1 → Diverge)

g = 1 for flux-flux → 〈‖un‖2〉 ≤ 3−1/2(1− a)n1/2στ

Page 59: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Stability: Finite Tc

0 200 400 600 800 10000

0.5

1

Tc

k vv ,

k fv

10 30 50 70 900.8

0.9

1

1.1

Tc

k vf ,

k ff

Figure: The amplification factor g versus Tc/∆t for the four schemes:VV (squares), FV (diamonds), VF (triangles) and FF (circles).

Page 60: Multiscale Modeling and Hybrid Algorithms · Multiscale Modeling and Hybrid Algorithms Weinan E Princeton University Collaborators: Xiantao Li (PSU), Jianfeng Lu (Princeton), Pingbing

Remarks:

I There are many different variants of atomistic/continuumcoupling schemes.

I Errors and artifacts are difficult to understand.

I What I have described are examples of efforts to try to putthings on a solid foundation.