Super-Arrhenius temperature dependent viscosity due to liquid-liquid phase separation in the super-cooled Kob-Andersen model

This study investigates the liquid-liquid phase separation in the supercooled Kob-Andersen model using a weighted coordination number order parameter to reconstruct binodal lines and verify local equilibrium, ultimately modeling the transition to the glass transition region and its super-Arrhenius temperature-dependent viscosity through a Markov Network Model.

The Gist

Imagine you have a giant jar filled with two types of marbles: big, heavy red ones (Type A) and small, bouncy blue ones (Type B). You shake the jar, and the marbles swirl around happily, mixing together like a liquid. This is the "normal" state.

Now, imagine you suddenly freeze the jar. As the temperature drops, the marbles slow down. Usually, we expect them to just get stuck in place, turning into a solid block of glass (like a frozen puddle). But this paper suggests something much more interesting is happening before they freeze.

Here is the story of what the scientists found, explained simply:

1. The "Secret Identity" Detector (The Order Parameter)

In a normal liquid, all the marbles look the same to a camera. But in this "super-cooled" state, the marbles start to feel different. Some are surrounded by neighbors in a tight, cozy hug; others are in a loose, scattered group.

The scientists invented a special tool called the Weighted Coordination Number (WCN). Think of this as a "secret identity detector." Instead of just counting how many neighbors a marble has, it looks at who those neighbors are and how they are arranged.

• The Magic: This tool can tell the difference between two types of "liquids" that look identical to the naked eye. It's like being able to tell the difference between a crowd of people wearing identical gray suits just by looking at how they are standing in line.

2. The Great Split (Liquid-Liquid Phase Separation)

As the jar gets colder, the WCN detector reveals a shock: the marbles aren't just slowing down; they are splitting into two different teams.

• Team 1: The "Loose" marbles (lower density).
• Team 2: The "Tight" marbles (higher density).

They separate into distinct islands within the jar, like oil and water, but both are still liquid! This is called Liquid-Liquid Phase Separation. The scientists mapped out exactly where this happens, creating a "weather map" (a phase diagram) that shows when the split occurs.

3. The Sticky Border (Surface Tension and Coarsening)

Where these two liquid teams meet, there is a border. Imagine a soap bubble separating two rooms.

• The scientists found that this border is wobbly. Sometimes it curves inward (concave), sometimes outward (convex).
• The Analogy: Think of the border like a rubber band. The system wants to shrink the border to save energy. This causes the "islands" of liquid to merge and grow larger over time. This process is called coarsening.
• The Problem: As the temperature drops, the marbles get so sluggish that moving the border becomes incredibly hard. The "rubber band" gets stuck.

4. The Traffic Jam (Viscosity and Glass Transition)

This is the big discovery. Why does glass become so hard and sticky (viscous) just before it freezes?

• Old Idea: We thought the marbles just got too tired to move.
• New Idea: The marbles are actually trying to rearrange their "islands" (the liquid-liquid separation), but the border between them is fighting back.

The scientists used a Markov Network Model (think of it as a traffic simulation) to calculate how hard it is to move. They found that the "traffic jam" isn't because the marbles are frozen; it's because the process of merging the liquid islands is slowing down.

• As the temperature drops, the time it takes for these islands to merge grows exponentially. It's like trying to merge two traffic lanes where the cars are moving at a snail's pace. The "stickiness" (viscosity) skyrockets because the system is stuck waiting for the islands to rearrange.

5. The Big Picture

The paper argues that the "Glass Transition" (when a liquid turns into glass) isn't just a random freezing. It is actually a traffic jam caused by the struggle between two different liquid states.

• The Metaphor: Imagine a dance floor. At first, everyone is dancing freely. Then, the music slows down. Two groups of dancers form: the "Slow Dancers" and the "Fast Dancers." They try to separate into different corners of the room. But as the music gets slower, they can't move fast enough to get to their corners. They get stuck in the middle, bumping into each other, unable to finish the dance. The whole floor becomes a rigid, frozen mess.

Why Does This Matter?

This study provides a clear, physical reason for why glass gets so sticky. It suggests that if we understand how these "liquid islands" separate and merge, we can better understand how to make better glasses, plastics, and even understand how water behaves at extreme cold (which is a mystery in itself).

In short: Glass isn't just frozen liquid; it's a liquid that got stuck trying to sort itself out.

Technical Summary

1. Problem Statement

The glass transition is characterized by a dramatic, non-Arrhenius (super-Arrhenius) increase in viscosity as a liquid is supercooled, often spanning many orders of magnitude. While the phenomenon is well-documented, the underlying physical mechanism remains a subject of debate. Specifically, the interplay between thermodynamically driven liquid-liquid phase separation (LLPS) and dynamic heterogeneity in glass-forming liquids is not fully understood.

Traditional methods for calculating viscosity, such as the Green-Kubo method, fail in the deep supercooled regime because the relaxation times become prohibitively long. Furthermore, conventional order parameters (like density) often fail to distinguish between different amorphous liquid states that have similar thermodynamic variables but distinct local structures. This study aims to:

1. Investigate the existence and nature of LLPS in the Kob-Andersen (KA) binary Lennard-Jones model using a novel order parameter.
2. Determine if the kinetics of LLPS (specifically interface coarsening) can explain the super-Arrhenius viscosity behavior observed near the glass transition.

2. Methodology

A. Simulation System

• Model: The study utilizes the binary Kob-Andersen (KA) Lennard-Jones potential, a standard model for glass formers that avoids crystallization.
• Composition: 80% Type A particles and 20% Type B particles.
• Ensemble: Simulations were performed in NVT and NPT ensembles using periodic boundary conditions, Nose-Hoover thermostats/barostats, and the LAMMPS package.
• Protocol: Systems were equilibrated at high temperature ($T^* = 1.2$) and then instantaneously quenched to target temperatures ranging from $T^* = 0.3$ to $0.58$.

B. Classification Scheme: Weighted Coordination Number (WCN)

To distinguish between different liquid states where density fails, the authors employed the Weighted Coordination Number (WCN) as an order parameter.

• Mechanism: WCNs are constructed by assigning values to a particle's solvation features based on normalized Gaussian distributions centered on the peaks of the radial distribution function $g(r)$.
• Dimensionality Reduction: The high-dimensional WCN vectors are reduced using Principal Component Analysis (PCA) to visualize the feature space.
• Clustering: K-means clustering (with $k=2$) is applied in the PCA space to classify particles into two distinct thermodynamic states. A "co-learning" strategy involving Gaussian Mixture Models (GMM) and real-space tagging is used to iteratively refine the classification of interfacial particles, ensuring a soft assignment that respects thermal fluctuations.

C. Phase Diagram Construction

• The authors reconstructed the gas-liquid binodal line to validate the WCN method against literature.
• They extended the method to map the liquid-liquid binodal line by quenching systems at various bulk densities and identifying the coexistence densities of the two liquid states.
• The lever rule was verified to confirm local thermodynamic equilibrium.

D. Viscosity Modeling: Markov Network Model (MNM)

To calculate viscosity in the deep supercooled regime where Green-Kubo fails, the authors adapted the Markov Network Model.

• Concept: The system is modeled as a network of basins (states). In the context of LLPS, the two basins correspond to the two coexisting liquid domains.
• Mechanism: Particle transitions between domains are treated as activated processes. The activation energy barrier ($Q$) is derived from the interfacial free energy (surface tension $\times$ area) between the two liquid phases.
• Calculation: The viscosity $\eta(T)$ is calculated using a two-state approximation where the activation energy is the sum of local surface tensions of the interface. The local curvature of the interface is calculated using a 2D quadric fit on interfacial particle neighbors.

3. Key Results

A. Phase Diagram and Thermodynamics

• LLPS Confirmation: The WCN analysis successfully identified a clear liquid-liquid phase separation in the supercooled KA model.
• Binodal Line: A liquid-liquid binodal line was constructed, with a critical point estimated at $T^*_c \approx 0.6$ and $\rho^*_c \approx 1.1$.
• Equilibrium Verification: The densities of the coexisting liquid states obeyed the lever rule across different bulk densities, confirming that the system reaches local thermodynamic equilibrium.
• Pressure Profiles: Analysis of density and pressure profiles across the interface revealed mechanical equilibrium for convex interfaces but a slight breakdown (local inhomogeneity) for concave interfaces, driving coarsening kinetics.

B. Surface Tension and Curvature

• The study calculated surface tension ($\sigma_s$) for both convex and concave curvatures.
• Results showed that surface tension acts as a driving force for interface coarsening. The total interfacial free energy ($\phi$) was found to be independent of system size ($N$), validating the scaling of the method.

C. Viscosity and Super-Arrhenius Behavior

• Viscosity Calculation: Using the MNM and the calculated interfacial free energy, the temperature-dependent viscosity was computed for $T^* \in [0.36, 0.55]$.
• Super-Arrhenius Scaling: The viscosity increased by approximately 16 orders of magnitude as temperature dropped from 0.55 to 0.36.
• Agreement: The results matched well with literature data (Green-Kubo) in the overlapping temperature range and extended successfully into the deep supercooled regime where traditional methods fail.
• Mechanism: The drastic slowdown is attributed to the coarsening kinetics of the liquid-liquid phase separation. As the system cools, the activation energy required for particles to cross the interface (driven by surface tension) increases, leading to the observed non-Arrhenius behavior.

4. Significance and Conclusions

• Unified Mechanism: The paper provides a compelling physical mechanism linking liquid-liquid phase separation directly to the glass transition. It suggests that the dramatic slowdown in dynamics is not merely a kinetic arrest but is governed by the thermodynamics and coarsening kinetics of a first-order LLPS.
• Methodological Advancement: The use of WCNs combined with machine learning (PCA/K-means) offers a robust framework for identifying structural heterogeneities in amorphous systems where density-based methods fail.
• Viscosity Prediction: The adaptation of the Markov Network Model to use LLPS interfacial properties allows for the accurate calculation of viscosity in regimes previously inaccessible to standard molecular dynamics.
• Broader Implications: The authors suggest this mechanism may be universal for glass formers, noting that similar LLPS has been observed in water models (ST2). If confirmed, this offers a "simple picture" for understanding the glass transition as a consequence of competing liquid phases.

In summary, the study demonstrates that the super-Arrhenius viscosity of the Kob-Andersen model is a direct consequence of the thermodynamics of liquid-liquid phase separation, where the energy barrier for structural relaxation is determined by the surface tension of the separating liquid domains.