Enhanced Kinetic Monte Carlo
T. P. Schulze Department of Mathematics University of Tennessee
Research supported by: NSF-DMS-0707443 & DE-FG02-03ER25586
Kinetic Monte-Carlo Simulations
• Configuration space is typically discrete, e.g. occupation arrays for a fixed lattice. • Dynamics are imposed via a Markov Chain model • Transition rates are based on configuration changes • Example: qi = ν exp
− ∆φ kb T
,
∆φ = ES + mEN
Early Stages of Growth
≈ 104 and 105 atoms colored by coordination Nijk
Effect of Surface Diffusion
K = 106
K = 105
K = 104
KMC with Elastic Interactions Collaborator: P. Smereka (Michigan) Reference: Lam, Lee & Sander Phys. Rev. Lett. 2002 • 1+1 SOS with height hi • Nearest & next nearest neighbor linear springs: long springs if short springs if • Hopping rate:
j>0 j < 0.
ri = r0 exp [(−γn + ∆W + E) /kB T ] ,
• Elastic energy barrier: ∆W = W (with atom i)−W (without atom i), 1 X W = wi,j , 2 (i,j)∈Ω
• Equilibrium configuration:
∂W ∂ui,j
=0
and
∂W ∂vi,j
= 0.
Energy Localization Method
• The exact elastic energy barrier is ∆W = W (u; Ω) − W (ua ; Ωa ). • The energy localization is approximation is ∆WL = W (u; Ωρ ) − W (uaρ ; Ωaρ ), where the displacement of the atom-off solution uaρ is constrained to agree with the atom-on solution u on the boundary.
Energy Barrier and Film Profile
Continuum Analog—Displacement Field Within this framework, the displacement field is give by ∂i2 uj + 2∂j ∂k uk Teij nj ui
=
x∈Ω
0,
= σ1 n1 δi1 , x ∈ ∂Ω, → 0, |x| → ∞.
The energy density can be expressed as w = 1/2Eij Tij , —consistent with the continuum limit of our discrete model. For a finite region Ωρ ⊂ Ω is the total energy is Z W (u; Ωρ ) = w dx. Ωρ
The elastic energy barrier is then ∆W = lim W (u; Ωρ ) − W (ua ; Ωaρ ) . ρ→∞
Energy Localization Approximation
For the local solution, we define the atom-off solution on a domain Ωaρ with lower boundary constrained by
uaρ = u,
x ∈ Γρ .
Our approximation for atom-off displacement field is then a uρ if |x| < ρ w= u if |x| ≥ ρ. and our approximation for the elastic energy barrier is ∆WL = W (u; Ωρ ) − W (uaρ ; Ωaρ ).
(1)
Energy Localization Approximation Theorem 1 - Residual Estimate. Suppose that h(x) is a compactly supported function whose support includes x = 0. Further, suppose that h(x) is modified by a localized change centered at x = 0 then the following is true RL = O(ρ−2 )
as
ρ → ∞.
Theorem 2 - Principle of Energy Localization. Under the same hypotheses of Theorem 1 the following is true ∆W = ∆WL(1 + O(ρ−2 ))
as
ρ → ∞.
This explains the high accuracy of our method.
Energy Truncation Approximation Naively, one might expect an approximation based on a simple truncation of the energy integrals, ∆WT = W (u; Ωρ ) − W (ua ; Ωaρ ), would be better. However, we also have Theorem 3 - Nonlocality of the Energy Density. Under the same hypotheses of Theorem 1 the following is true ∆W = ∆WT (1 + O(Hρ−1 ))
as
ρ → ∞.