Heat transport in superionic materials via machine-learned molecular dynamics

This paper demonstrates that reliable thermal conductivity in superionic materials can be achieved by applying Onsager's reciprocal relations to account for coupled heat and mass transport, overcoming the model dependence and physical ambiguities inherent in conventional Green-Kubo methods.

The Gist

The Big Picture: The "Superionic" Dance Party

Imagine a material called a superionic conductor. You can think of this material as a crowded dance party with two types of people:

1. The Static Crowd: These people are standing still, holding hands in a rigid grid. They are vibrating a little bit (like shivering), but they aren't going anywhere.
2. The Runners: These people are running wild, weaving through the static crowd, changing partners, and moving freely like water flowing through a pipe.

This mix of "solid" (the grid) and "liquid" (the runners) is what makes the material superionic. Scientists love these materials because they are great for batteries and energy-saving devices. But to use them, we need to know exactly how well they conduct heat.

The Problem: The "Broken Thermometer"

For a long time, scientists used a standard method (called the Green-Kubo method) to calculate how heat moves through these materials. Think of this method as a thermometer that measures the total energy of the room.

However, the researchers in this paper discovered a major flaw: The thermometer was broken.

Here's why: To calculate the heat, the computer has to assign a specific amount of energy to every single atom in the simulation. But in a superionic material, there is no single "correct" way to split the total energy among the atoms. It's like trying to split a pizza among a group of friends where everyone agrees on the total size of the pizza, but they can't agree on how much slice belongs to whom.

• The Result: When the researchers used different computer models (different ways of slicing the pizza) to simulate the same material, they got wildly different answers for how well it conducts heat. One model said it was a great insulator; another said it was a great conductor. This made it impossible to trust the results.

The Solution: The "Onsager Correction" (The Traffic Cop)

The researchers realized that in these materials, heat and mass (the moving ions) are best friends. They travel together. When the "Runners" (ions) move, they carry heat with them. If you only measure the heat, you miss the fact that the moving people are dragging the heat along.

To fix the broken thermometer, they applied a rule called Onsager's Reciprocal Relations.

• The Analogy: Imagine you are trying to measure the speed of a river (heat). But the river is also carrying a lot of leaves (ions) downstream. If you just measure the water flow, you get a wrong answer because the leaves are pushing the water.
• The Fix: The new formula acts like a Traffic Cop. It looks at how fast the leaves are moving and subtracts that influence from the water flow measurement. This gives you the true speed of the water, regardless of how many leaves are floating by.

By using this "Traffic Cop" formula, the researchers found that all their different computer models finally agreed. They all gave the exact same answer for the thermal conductivity.

The Surprise: The "Unchanging" Heat

Once they fixed the math, they discovered something weird and wonderful about the material $\alpha$-Li$_3$PS$_4$ (a solid battery electrolyte).

• Normal Crystals: Usually, as a crystal gets hotter, it conducts heat worse (like a crowded hallway where people bump into each other more).
• Normal Glasses: Usually, as a glass gets hotter, it conducts heat better (the vibrations get stronger).
• This Superionic Material: It does something strange. No matter how much you heat it up (from room temperature to very hot), its ability to conduct heat stays exactly the same.

It's like a thermostat that refuses to change the temperature reading, no matter how hot the sun gets outside. The researchers think this happens because two opposing forces cancel each other out perfectly: the heat carried by vibrating atoms goes down, but the heat carried by the running ions goes up. They balance out to a constant.

The Takeaway: When Do You Need the Fix?

The paper ends with a simple rule for other scientists: Do you need this complicated "Traffic Cop" correction?

• Yes: If the material has ions that move around a lot and the energy assigned to them is "fuzzy" or uncertain.
• No: If the correction term is tiny compared to the main flow, you can skip it.

They created a simple "test" (a specific number involving energy and diffusion) that tells you if your calculation will be accurate or if it will be wildly wrong without the correction.

Summary

1. The Issue: Standard methods for calculating heat in "superionic" materials are unreliable because they depend on how you arbitrarily split energy between atoms.
2. The Fix: Use a special formula (Onsager correction) that accounts for the fact that moving ions carry heat with them.
3. The Discovery: Once fixed, the math works perfectly, and some materials show a weird, constant heat conductivity that doesn't change with temperature.
4. The Impact: This makes designing better batteries and energy devices much more reliable, as scientists can finally trust their computer simulations.

Technical Summary

1. Problem Statement

Superionic materials represent a unique state of matter where a subset of ions diffuses liquid-like through a rigid lattice of other ions. These materials are critical for next-generation energy storage (solid-state batteries) and thermoelectric devices. However, accurately modeling their thermal conductivity ($\kappa$) is challenging due to:

• Atomic Diffusion: The coexistence of solid-like and liquid-like subsystems complicates first-principles calculations.
• Ambiguity in Machine-Learned Potentials (MLPs): While MLPs combined with Green-Kubo (GK) molecular dynamics (MD) offer quantum-mechanical accuracy, the calculation of $\kappa$ via the conventional GK integral relies on the atomic site potential ($U_i$).
• Non-Uniqueness: The total energy in a system is unique, but the projection of this energy onto individual atoms (site potential) is not uniquely defined in different MLP models or even within the same model trained with different exchange-correlation functionals. This leads to significant, model-dependent variations in calculated $\kappa$, rendering standard approaches unreliable for superionic conductors.

2. Methodology

The authors propose a rigorous framework to resolve these ambiguities by integrating Onsager reciprocal relations into the GK formalism for multicomponent systems.

• Theoretical Framework:

  • Instead of treating the system as a single component, they treat it as a multicomponent system with coupled heat and mass fluxes.
  • They utilize the Onsager coefficient matrix ($\Lambda_{ij}$), where $L_{00}$ represents the energy flux autocorrelation, $L_{11}$ relates to mass diffusion (ionic conductivity), and $L_{01}$ captures heat-mass coupling.
  • The corrected thermal conductivity is derived as:
    $$\kappa = \frac{1}{T^2} \left( L_{00} - \frac{L_{01}^2}{L_{11}} \right)$$
  • This formulation satisfies gauge invariance (independence from the definition of atomic site energy) and convective invariance, unlike the conventional $\kappa = L_{00}/T^2$.

• Computational Setup:

  • Materials: Two representative superionic conductors were studied: $\alpha$-Li$_3$PS$_4$ (solid-state electrolyte) and $\beta$-Cu$_{1.98}$Se (thermoelectric).
  • MLP Training: Six different MLP models (A–F) for Li$_3$PS$_4$ and four for Cu$_{1.98}$Se were trained using the Neuroevolution Potential (NEP) framework on Density Functional Theory (DFT) datasets.
  • Simulation: Equilibrium MD simulations were performed using the GPUMD package.
  • Validation: Results were cross-validated using Multivariate Cepstral Analysis (MCA) via the SporTran code.

3. Key Contributions

1. Identification of Model Dependence: The study demonstrates that the conventional GK integral ($\kappa = L_{00}$) yields thermal conductivity values that vary significantly (up to a factor of 3) depending on the specific MLP model used, due to the non-uniqueness of the atomic site potential.
2. Onsager Correction Implementation: The authors establish that applying the Onsager correction term ($-L_{01}^2/L_{11}$) eliminates this model dependence, yielding reliable, model-independent $\kappa$ values that satisfy physical invariance principles.
3. Critique of Energy Flux Decomposition: The paper challenges the physical interpretation of decomposing $\kappa$ into kinetic ($\kappa_k$), potential ($\kappa_p$), and cross ($\kappa_{kp}$) terms. While mathematically valid, these terms are not gauge invariant and exhibit strong model dependence in superionic phases, making mechanistic interpretations based on them ambiguous.
4. Discovery of Anomalous Temperature Dependence: In $\alpha$-Li$_3$PS$_4$, the corrected $\kappa$ is found to be temperature-invariant over a wide range (300–850 K), a behavior distinct from the $T^{-1}$ dependence of crystals or the gradual increase in glasses.
5. Criterion for Correction Necessity: A simplified criterion ($U^2/L_{11}$) is proposed to predict when the Onsager correction is necessary. If the ratio of the squared site potential magnitude to the diffusion coefficient is small, the conventional approach may suffice; otherwise, the correction is mandatory.

4. Key Results

• $\alpha$-Li$_3$PS$_4$:

  • Model Variance: Using the conventional $L_{00}$ method, different MLP models produced $\kappa$ values ranging from ~1 W/mK to ~3 W/mK at 650 K.
  • Invariance: After applying the Onsager correction, all models converged to a consistent $\kappa$ value.
  • Temperature Trend: The corrected $\kappa$ remained nearly constant from 300 K to 850 K. This is attributed to a competition between decreasing phonon contributions and increasing tunneling/diffusion transport.
  • Decomposition Failure: The individual components ($\kappa_k, \kappa_p, \kappa_{kp}$) showed wild fluctuations across different MLP models, confirming they lack physical robustness in this regime.

• $\beta$-Cu$_{1.98}$Se$:

  • The model dependence of $L_{00}$ was smaller (8% variation) compared to Li$_3$PS$_4$, and results were closer to experimental data.
  • However, the Onsager correction was still required for theoretical consistency. The smaller correction magnitude was attributed to a much larger $L_{11}$ (ionic conductivity) in Cu$_{1.98}$Se, which suppresses the $L_{01}^2/L_{11}$ term.

• Physical Mechanism:

  • The correction arises from the competition between thermally driven mass flux (Soret effect) and chemically driven mass flux (Fick's law). A larger diffusion coefficient ($L_{11}$) allows the system to reach a steady state with a smaller concentration gradient, thereby reducing the back-flow of heat (Dufour effect) and minimizing the correction needed.

5. Significance

• Methodological Advancement: This work provides a definitive protocol for calculating thermal conductivity in superionic and diffusive materials using MLPs, resolving a long-standing issue of gauge non-invariance.
• Reliability for Applications: By ensuring model-independent results, the study enables more accurate thermal management designs for solid-state batteries and thermoelectric devices.
• New Physics: The discovery of temperature-invariant thermal conductivity in superionic phases suggests a new class of thermal transport behavior driven by the interplay of phonons and diffusive ions, challenging traditional crystal/glass paradigms.
• Practical Guidance: The proposed $U^2/L_{11}$ criterion offers a practical tool for researchers to determine whether the computationally expensive Onsager correction is necessary for a specific material system.