Resolving Structural Avalanches in Amorphous Carbon with Arclength Continuation

By employing a pseudo-arclength numerical continuation framework on machine-learned models of amorphous carbon, the authors demonstrate that structural avalanches can be decomposed into individual shear transformations and resolved through their underlying energy landscape, revealing a latent structure prior to onset and providing an event-driven method for studying avalanche dynamics.

The Gist

The "Domino Effect" in Invisible Glass: A Simple Guide

Imagine you are pushing a heavy, slightly uneven bookshelf across a carpeted floor. Most of the time, you push, and it moves smoothly. But every once in a while, the bookshelf hits a snag, catches on a piece of lint, and suddenly jumps forward with a loud thud.

That "jump" is what scientists call an avalanche. In the world of tiny, invisible materials (like amorphous carbon, which is a cousin to diamond), these jumps happen at the atomic level. Instead of a bookshelf jumping, a group of atoms suddenly rearranges themselves, causing a tiny "quake" in the material.

This paper describes a new, high-tech way to watch these tiny atomic quakes in slow motion to understand exactly why they happen.

---

The Problem: The "Blurry Snapshot" Method

Until now, scientists have been studying these atomic jumps using a method called AQS (Athermal Quasi-Static).

Think of AQS like trying to watch a movie by taking a photo every 10 minutes. If a massive explosion happens at minute 4, you won't see it. You’ll just see a photo of a peaceful room at minute 0, and then a photo of a destroyed room at minute 10. You know something happened, but you missed the actual explosion, the sparks, and the way the walls fell.

Because the "photos" (the time steps) were too far apart, scientists couldn't tell if one big explosion happened, or if it was actually ten small pops happening in a row. This made their data "blurry" and slightly inaccurate.

The Solution: The "Smooth Video" Method (Arclength Continuation)

The researchers used a new mathematical tool called Arclength Continuation (AC).

Instead of taking snapshots every 10 minutes, imagine if you had a camera that could follow the path of every single flying spark during that explosion. This method doesn't just wait for the "jump" to happen; it follows the "energy landscape"—the invisible hills and valleys that atoms live in.

When an atom is about to jump, it’s like a ball sitting on a hill that is slowly tilting. The old method waited for the ball to roll down. This new method follows the ball as the hill tilts, tracing the exact path the ball takes, even as it rolls through tiny little dips and bumps along the way.

The Big Discovery: The "Hidden Chain"

By using this "smooth video" approach, the researchers discovered something amazing: Avalanches aren't just one big, messy crash. They are organized chains of tiny events.

They found that a large avalanche in carbon is actually like a row of dominoes.

1. One single bond between two atoms breaks (the first domino falls).
2. That tiny movement nudges a second bond (the second domino falls).
3. This continues in a specific, predictable sequence.

Before the "big jump" happens, there is a "latent structure"—a hidden blueprint of these dominoes just waiting to be tipped over.

Why Does This Matter?

Understanding these tiny, invisible "domino effects" is like learning the secret language of how materials break.

If we can predict exactly how these atomic chains react, we can design better materials—things that are tougher, more flexible, or more durable. Whether it's the carbon in high-tech electronics or the glass in your smartphone, knowing how the "dominoes" fall helps us build a world that doesn't shatter unexpectedly.

Technical Summary

Technical Summary: Resolving Structural Avalanches in Amorphous Carbon with Arclength Continuation

1. Problem Statement

Plastic deformation in amorphous (glassy) solids is driven by localized rearrangements known as Shear Transformation Zones (STZs). Because STZs interact through long-range elastic fields, the activation of one can trigger a cascade of others, resulting in structural avalanches.

Current computational methods to study these avalanches, primarily Athermal Quasi-Static (AQS) simulations and Nudged Elastic Band (NEB) calculations, suffer from significant limitations:

• AQS Limitations: The results are highly dependent on the chosen strain step size ($\Delta\gamma$). Large steps merge distinct, closely spaced events, leading to inaccurate statistical distributions (e.g., underestimating the power-law exponent of stress drops). AQS also fails to provide the inherent energetics (energy barriers) of the avalanche.
• NEB Limitations: While NEB can find minimum energy paths (MEPs), it is computationally expensive for large systems, often fails to converge near instability points, and is highly sensitive to the initial guess, making it unsuitable for resolving complex, multi-bond avalanches.

2. Methodology

The authors propose a novel framework using Hessian-free pseudo-arclength numerical continuation (AC) to map the potential energy landscape (PES) of amorphous carbon.

• System Model: Amorphous carbon (a-C) is modeled using a high-fidelity, machine-learned Atomic Cluster Expansion (ACE) potential, which provides realistic interatomic interactions without the prohibitive cost of traditional quantum mechanical methods.
• Arclength Continuation (AC): Unlike AQS, which parameterizes the solution path by applied strain ($\gamma$), AC uses an arclength parameter ($s$). This allows the algorithm to traverse bifurcation points (where a local minimum becomes a saddle point) seamlessly, moving from minima to index-1 saddles without the singularities encountered in strain-based methods.
• Hessian-Free Approach: To avoid the massive computational cost of calculating the second derivative (Hessian) of the ACE potential, the authors employ a Krylov-subspace approach. This approximates the necessary matrix-vector products using finite differences, making the method scalable for realistic atomistic systems.
• Avalanche Dissection: The authors developed a recursive algorithm (Algorithm 1 & 2 in the supplement) that combines AC with "biased relaxations." By traversing a saddle point and then relaxing the system at lower strains, they can "dissect" a large avalanche into its constituent, individual bond-breaking or bond-forming events.

3. Key Contributions

• Development of a Tailored AC Framework: A robust numerical method that overcomes the step-size dependence of AQS and the convergence issues of NEB.
• Discovery of Latent Avalanche Structure: The identification that avalanches in a-C are not monolithic "jumps" but are composed of a sequence of well-separated local minima and index-1 saddle points.
• Event-Driven Simulation Framework: A method to resolve the intrinsic statistics of plastic flow by eliminating the artifacts introduced by finite strain increments.

4. Results

• Resolution of Latent Structure: The AC method revealed that prior to the onset of a full avalanche, the system possesses a "latent structure." For a six-bond avalanche, the authors successfully decomposed the event into six unique, sequential single-bond transformations.
• Energetic Validation: The energy barriers and minima identified via the AC method were validated against NEB calculations, showing excellent agreement.
• Elimination of Step-Size Bias: In comparison to AQS, the AC method produced stress-drop distributions that were independent of the strain step. While AQS underestimated the power-law exponent $\alpha$ of stress drops due to event merging, AC yielded a steeper, more accurate, and clearly defined scaling.
• Computational Efficiency: The AC method was found to be highly efficient; for a specific stress-strain curve, it required approximately 73% of the force calls compared to a high-resolution ($\Delta\gamma = 10^{-4}$) AQS simulation, while providing superior resolution.

5. Significance

This work represents a significant advancement in the computational physics of disordered materials. By providing a way to "look inside" an avalanche, the researchers have moved beyond merely observing macroscopic stress drops to understanding the microscopic, causal chain of atomic rearrangements.

The ability to resolve the energy landscape of realistic, machine-learned potentials opens new avenues for:

1. Building Coarse-Grained Models: Using the extracted energy barrier data to inform elasto-plastic constitutive models.
2. Predicting Failure: Understanding how local structural changes lead to catastrophic phenomena like shear banding or fracture.
3. Statistical Mechanics of Glasses: Providing a rigorous way to study the universal scaling laws of plasticity in amorphous systems without the numerical artifacts of traditional simulation protocols.