Simulations of dislocation dynamics on an atomic lattice: the effect of collision rules

This paper uses numerical simulations to demonstrate that while discrete dislocation dynamics models with annihilation rules consistently converge to a PDE accounting for annihilation, models without collision rules exhibit inconsistent convergence behavior, highlighting the critical importance of carefully treating dislocation collisions in such simulations.

The Gist

Imagine a crowded dance floor made of a giant, repeating grid of tiles. On this floor, there are many dancers. Some dancers wear red shirts (representing positive dislocations), and some wear blue shirts (representing negative dislocations).

This paper is a scientific experiment to figure out how to predict the movement of this entire crowd. The scientists want to know: If we watch every single dancer move one by one, can we predict the overall flow of the crowd using a simple set of rules (a "macroscopic" model)?

Here is the breakdown of their experiment, the rules they tested, and what they found.

The Two Rules of the Dance

The scientists ran two different versions of this simulation, changing only one rule about what happens when a red dancer and a blue dancer bump into each other.

1. The "Ghost" Rule (Conservation Model):
   In this version, if a red dancer and a blue dancer collide, they don't disappear. They just pass right through each other or stand on top of one another. They keep dancing. The total number of red and blue dancers stays exactly the same forever.

   • The Expectation: The scientists thought this would lead to a smooth, predictable flow of the crowd where the total number of red and blue dancers is always conserved.

2. The "Vanishing" Rule (Annihilation Model):
   In this version, if a red dancer and a blue dancer collide, they instantly cancel each other out and leave the dance floor. They vanish.

   • The Expectation: The scientists thought this would lead to a different kind of flow, where the crowd gets smaller over time, but the net difference between red and blue dancers remains constant.

The Experiment

The researchers used powerful computers to simulate thousands of these dancers moving randomly but influenced by each other (like magnets pushing and pulling). They ran these simulations with increasing numbers of dancers (from 20 up to 200) to see if the chaotic individual movements would eventually settle into a predictable pattern that matched their mathematical formulas.

The Surprising Results

1. The "Vanishing" Rule worked perfectly.
When the dancers were allowed to disappear upon collision, the chaotic individual movements perfectly matched the smooth, predictable mathematical formula the scientists had written down.

• The Analogy: It's like watching a crowd of people leave a concert. Even though every person walks a different path, the overall flow of the crowd leaving the building matches the traffic model perfectly. The math predicted exactly how the crowd thinned out.

2. The "Ghost" Rule failed (mostly).
When the dancers were not allowed to disappear (they just passed through), the results were messy and unpredictable.

• The Analogy: Imagine a traffic model that assumes cars never crash or disappear, they just drive through each other like ghosts. The scientists found that in certain conditions, the actual traffic didn't follow the "ghost" math at all. Instead, the crowd behaved as if the cars were vanishing, even though the rules said they shouldn't.
• The Twist: In some scenarios, the "Ghost" crowd started acting exactly like the "Vanishing" crowd. The mathematical model that assumed people stayed on the floor was actually a bad description of reality. The model that assumed people left the floor was the one that actually described the "Ghost" crowd's behavior.

The Big Takeaway

The main lesson of this paper is that how you handle collisions matters immensely.

If you are trying to build a computer model to predict how materials (like metal) bend and break, you have to be very careful about what happens when defects in the material crash into each other.

• If you assume they just pass through, your big-picture math might be completely wrong.
• Even if you assume they don't disappear, the physics of the situation might make them act like they do.

The authors conclude that for these specific types of simulations, the "Vanishing" rule provides a much more accurate map of reality than the "Ghost" rule, even if the microscopic rules say the dancers shouldn't actually vanish. It suggests that in the real world of metal physics, collisions are a critical event that changes the entire story, and ignoring them leads to the wrong predictions.

Technical Summary

Technical Summary: Simulations of Dislocation Dynamics on an Atomic Lattice

Problem Statement
This work investigates the connection between microscopic models of interacting dislocations and their macroscopic continuum limits described by partial differential equations (PDEs). Specifically, the authors focus on the stochastic dynamics of screw dislocations on a one-dimensional periodic lattice. Dislocations are modeled as particles with topological charges (Burgers vectors) of $+1$ or $-1$. A critical aspect of the problem is the treatment of collisions: when dislocations of opposite sign meet, do they annihilate (remove from the system), or do they pass through/collide without removal?

The authors study two discrete models:

1. $(P^n_{csv})$: A conservative model where collisions do not result in annihilation; the total number of dislocations and the absolute Burgers vector sum are conserved.
2. $(P^n_{ann})$: An annihilation model where colliding dislocations of opposite sign are instantaneously removed, conserving only the net Burgers vector.

The core objective is to determine whether these discrete Markov chain models converge to specific macroscopic PDEs as the number of dislocations ($n$) increases and the lattice spacing ($\varepsilon$) decreases, and to analyze how the microscopic collision rule affects the macroscopic limit.

Methodology
The study employs a numerical approach combining Kinetic Monte Carlo (KMC) simulations for the discrete systems and finite volume discretization for the continuum PDEs.

• Discrete Models: The dynamics are modeled as continuous-time Markov chains on a lattice $\Lambda_\varepsilon$. Jump rates are determined by Kramers' law, driven by interaction forces derived from the Peach–Koehler force (modeled via a periodic Green's function derivative). The parameter $\beta$ represents the ratio of interaction energy to thermal energy. The authors operate in a low-temperature regime ($\beta \to \infty$) but with $\beta \ll 1/\varepsilon$, ensuring that randomness dominates the interaction forces on the atomistic scale.
• Continuum Models:
  • For the conservative case, the limit is postulated to be the Groma-Balogh equations $(P^\infty_{csv})$, a system of continuity equations for positive and negative densities.
  • For the annihilation case, the limit is hypothesized to be a conservation-law type equation $(P^\infty_{ann})$ for the net Burgers vector density $\kappa$, where annihilation is explicitly modeled.
• Numerical Implementation:
  • KMC: Trajectories are simulated using an event-driven algorithm. To reduce computational cost, jump rates are updated incrementally rather than recalculated from scratch.
  • PDE Solvers: Time-explicit upwind finite volume schemes are used to solve the PDEs. The singular interaction kernel is handled via principal value integration approximations.
  • Convergence Metric: To compare discrete particle positions with continuum densities, the authors utilize a modified 1-Wasserstein distance ($W$). This metric quantifies the distance between the signed empirical measure of the discrete system and the continuum solution.
  • Statistical Analysis: Since discrete trajectories are stochastic, the authors run ensembles of simulations ($M=50$) for various parameter sets. They estimate the median distance $w$ and the standard deviation of the log-distance to assess convergence as $n \to \infty$.

Key Contributions and Results
The authors systematically explore the asymptotic regime where $n \to \infty$, $\varepsilon \to 0$, and $\beta \to \infty$ under the scaling constraints $n \ll 1/\varepsilon$ and $\beta \ll 1/\varepsilon$.

1. Convergence of the Annihilation Model:
   The simulations provide strong numerical evidence that the discrete annihilation model $(P^n_{ann})$ converges to the continuum annihilation PDE $(P^\infty_{ann})$ within the studied parameter regime. The median distance $w$ between the discrete and continuum solutions decays approximately as $n^{-0.8}$. This convergence holds robustly for various choices of scaling exponents, provided the parameters remain within the interior of the defined asymptotic regime.

2. Failure of Convergence for the Conservative Model:
   Contrary to the expectation that the conservative discrete model $(P^n_{csv})$ would converge to the conservative Groma-Balogh equations $(P^\infty_{csv})$, the results are inconclusive and often indicate divergence.

   • In certain parameter regimes (specifically when the interaction strength relative to thermal noise is high), the discrete conservative model does not converge to $(P^\infty_{csv})$.
   • Instead, the numerical evidence suggests that for some parameters, the discrete conservative model behaves asymptotically like the annihilation model $(P^\infty_{ann})$. The statistical density of the conservative discrete system tends to align with the continuum annihilation solution rather than the conservative one.
   • The authors observe that even without explicit annihilation rules, "dipoles" (pairs of opposite-sign dislocations) formed during collisions in the conservative model effectively behave as if removed because the probability of them dissociating later is negligible.

3. Sensitivity to Collision Rules:
   The study highlights that the macroscopic limit is highly sensitive to the microscopic collision rule. The distinction between the PDEs $(P^\infty_{csv})$ and $(P^\infty_{ann})$ is significant, particularly in regions where positive and negative densities overlap. The conservative PDE predicts smooth transitions and conservation of absolute density, while the annihilation PDE predicts steep interfaces and a reduction in total density.

Significance and Claims
The paper claims that careful treatment of dislocation collisions is crucial in discrete dislocation dynamics models. The primary finding is that a density-conserving PDE model may not accurately describe the dynamics of a discrete system even if that system theoretically allows for the separation of colliding dipoles.

• Robustness of Annihilation Limits: The explicit treatment of annihilation at the microscopic scale leads to a robust convergence to a specific macroscopic PDE.
• Limitations of Conservative Models: The assumption that a conservative microscopic model naturally converges to a conservative macroscopic PDE is challenged. The authors suggest that in the low-temperature limit, the effective macroscopic behavior of a conservative system may be better approximated by an annihilation model due to the effective "trapping" of opposite-sign pairs.
• Scope: The authors note that these conclusions are derived for a one-dimensional model with a single active slip plane. They conjecture that the importance of collision rules likely extends to two- and three-dimensional modeling of macroscopic plastic behavior in metals, but rigorous proof for higher dimensions remains an open mathematical problem.

The work provides numerical evidence rather than rigorous mathematical proofs for the convergence/divergence of these specific limits, filling a gap where rigorous connections from the atomistic scale to PDE models for dislocation density remain "out of reach."