Interface-dependent Phase Transitions and Ultrafast Hydrogen Superionic Diffusion of H2O Ice

By integrating artificial neural networks with large-scale molecular dynamics simulations, this study demonstrates that the diamond-ice interface significantly alters high-pressure water behavior by lowering the superionic transition temperature, inducing spontaneous bcc-to-fcc phase transitions via the inverse Bain mechanism, and redefining the stability fields of ice phases, thereby resolving discrepancies between theoretical predictions and experimental observations.

The Gist

Imagine you are trying to study how ice behaves when you squeeze it incredibly hard—so hard that it turns into a strange, super-hot, super-dense material that acts like a mix between a solid and a liquid. Scientists have been doing this for decades using a tiny, powerful tool called a Diamond Anvil Cell.

Think of a Diamond Anvil Cell like a pair of microscopic, diamond-tipped pliers. You put a tiny drop of water between the tips and squeeze. The pressure becomes so intense that the water turns into exotic forms of ice that don't exist anywhere else on Earth.

The Big Problem: The "Dirty" Hands
For a long time, scientists noticed something weird. When they squeezed the ice, the results didn't always match what computer simulations predicted.

• The Theory: Computers said, "Ice should melt at this temperature."
• The Experiment: Real-life diamonds said, "No, it melts way earlier!" or "It changes shape at a different pressure!"

The authors of this paper realized the missing piece of the puzzle: The Interface.

In the experiment, the ice isn't floating in a vacuum; it's squished directly against the diamond. It's like pressing your hand against a hot stove. The part of your hand touching the stove behaves differently than the part of your hand in the air. The scientists realized that the "touching" part (the interface) was messing up the data, but nobody had looked closely at it before.

The Superhero Team: AI and Giant Simulations

To solve this, the team didn't just run a standard computer simulation. They built a Neural Network Potential (NNP).

• The Analogy: Imagine trying to teach a robot how to play soccer. If you only show it 100 games, it might get confused. But if you show it 66 million games and let it learn from its mistakes (Active Learning), it becomes a grandmaster.
• They trained an AI on 66 million snapshots of atoms interacting between diamond and ice. This AI became so smart it could predict how atoms would move without needing to do the heavy math every single time. This allowed them to simulate a giant chunk of ice (20,000+ atoms) instead of a tiny, unrealistic speck.

The Three Big Discoveries

1. The "Super-Hot" Ice Slide (Superionic Diffusion)

Under extreme pressure, ice doesn't just melt; it becomes superionic.

• The Metaphor: Imagine a crowded dance floor where the dancers (oxygen atoms) are frozen in place, but the kids running around them (hydrogen atoms) are moving so fast they blur into a liquid. The ice is solid, but the hydrogen is flowing like water.
• The Discovery: The scientists found that when this ice touches the diamond, the "kids" (hydrogen) start running even faster. The interface acts like a slippery slide, making the hydrogen move much more easily. This means the ice becomes "superionic" at a much lower temperature than we thought. The diamond isn't just a wall; it's a catalyst that speeds things up.

2. The Spontaneous Shape-Shifter (bcc to fcc)

Ice has different "shapes" (crystal structures). One is called bcc (like a cube with a dot in the middle), and another is fcc (like a tightly packed stack of oranges).

• The Metaphor: Think of the bcc ice as a messy, loose stack of bricks. The fcc ice is a neat, tight pyramid. Usually, you need to push the bricks very hard and heat them up to get them to rearrange into the neat pyramid.
• The Discovery: When the ice touches the diamond, it doesn't need as much help. The interface acts like a mold. The ice spontaneously rearranges itself from the messy bcc shape into the neat fcc shape, even at lower pressures than scientists previously thought possible. It's like the diamond "whispers" to the ice, "Hey, stand up straight," and the ice listens.

3. Solving the Melting Mystery

Because the interface makes the ice change shape and move faster, it explains why experiments and theories disagreed.

• The "Ghost" Melting Point: In the past, scientists thought they were seeing ice melt. But the new study suggests they were actually seeing the ice change shape or turn superionic because of the diamond contact. The "melting" they saw was actually the ice reacting to the diamond, not just getting hot.
• By accounting for the diamond, the scientists drew a new map (a phase diagram) that finally matches what we see in the lab.

The Takeaway

This paper is a reminder that context matters. You can't study a material in isolation. If you are studying how ice behaves under pressure, you have to remember it's touching the diamond holding it.

The interface isn't just a boundary; it's an active participant. It changes how the ice moves, how it changes shape, and when it melts. By using AI to look at this "touching" zone, the scientists fixed a decades-old mystery and gave us a clearer picture of how water behaves in the most extreme environments in the universe—from the deep interiors of icy planets to the core of our own experiments.

In short: The diamond anvil isn't just a tool; it's a partner in the dance, and it changes the steps the ice takes.

Technical Summary

1. Problem Statement

High-pressure experiments on water ice (H₂O) using diamond-anvil cells (DAC) have revealed complex phase behaviors, including superionic states and solid-solid transitions. However, significant discrepancies exist between theoretical predictions and experimental observations regarding:

• Melting Curves: Reported melting temperatures at 20 GPa vary by up to 300 K, increasing to 800 K at 40 GPa.
• Phase Stability: Theoretical models predict face-centered cubic (fcc) ice stability only at extremely high pressures (>100 GPa), whereas experiments detect it at much lower pressures (~29–45 GPa).
• The Interface Gap: In DAC experiments, ice samples are in direct contact with diamond anvils (typically the (100) plane). The extent to which this diamond/ice interface alters intrinsic properties (such as phase transition temperatures, diffusion rates, and melting points) has been largely unknown, potentially explaining the inconsistencies between theory (often modeling bulk homogeneous ice) and experiment.

2. Methodology

To address these issues, the authors employed a multi-scale computational approach combining machine learning and molecular dynamics:

• Neural Network Potential (NNP) Development:

  • Utilized an Active Learning scheme (DP-GEN and DeePMD-kit) to train a high-fidelity NNP.
  • The training dataset included 17,076 Density Functional Theory (DFT) data points derived from 66.41 million snapshots across 13 iterations.
  • The NNP covers diamond, body-centered cubic (bcc) ice (Ice VII), and their interface models (diamond (100)/bcc ice (100)) over a wide pressure-temperature (P-T) range (20–120 GPa, 400–2100 K).
  • Validation showed high accuracy with average atomic force deviations of ~1.06%.

• Large-Scale Molecular Dynamics (MD) Simulations:

  • Performed NPT MD simulations using the GPUMD package with the trained NNP.
  • System Sizes: Compared a homogeneous bulk bcc-ice model (6,000 atoms) against a large-scale interface model (20,648 atoms, including a thick ice layer sufficient to distinguish bulk from interface effects).
  • Conditions: Simulations were run at pressures from 20 to 120 GPa and temperatures up to 2100 K.
  • Analysis: Calculated hydrogen and oxygen diffusion coefficients, mean square displacements (MSD), hydrogen rotation rates, and structural evolution (using Common Neighbor Analysis). Path-integral MD (PIMD) was also used to verify that Nuclear Quantum Effects (NQEs) were negligible compared to thermal effects at these temperatures.

3. Key Contributions

1. Quantification of Interface Effects: Demonstrated that the diamond/ice interface is not a passive boundary but actively drives phase transitions and alters diffusion mechanisms.
2. Mechanism of Ultrafast Diffusion: Identified that the interface induces water molecule dissociation and C-H/O bonding, leading to a "long-range" activation of superionic diffusion that propagates from the interface into the bulk.
3. Spontaneous bcc-to-fcc Transition: Discovered a thermally driven, spontaneous phase transition from bcc to fcc ice mediated by the interface, occurring at pressures and temperatures previously thought impossible for bulk ice.
4. Revised Phase Diagram: Proposed a new high-pressure phase diagram for ice that incorporates interface effects, reconciling the gap between theoretical predictions and experimental data.

4. Key Results

A. Ultrafast Hydrogen Superionic Diffusion

• Temperature Reduction: The presence of the diamond interface significantly lowers the hydrogen superionic transition temperature. At 100 GPa, the transition temperature drops by approximately 104 K (using a diffusion threshold of $10^{-5}$ cm²/s). This reduction ranges from 44 K (at 20 GPa) to 118 K (at 120 GPa).
• Mechanism: Hydrogen diffusion is driven by molecular rotation. The interface causes water molecules to dissociate and bond with surface carbon atoms, creating a region of high hydrogen delocalization.
• Long-Range Effect: The enhanced diffusion is not limited to the immediate interface. Simulations show that the "superionic" state activates at the interface and propagates inward, increasing the diffusion coefficient of the center region over time.

B. Spontaneous bcc-to-fcc Phase Transition

• Inverse Bain Mechanism: The interface induces a spontaneous transition from bcc to fcc ice via an inverse Bain transformation. Oxygen atoms shift from bcc equilibrium positions to fcc positions, accompanied by c-axis expansion.
• Conditions: This transition occurs spontaneously at 20 GPa and 900 K in the interface system, whereas homogeneous bulk ice remains bcc under these conditions.
• Reversibility: The process is reversible; cooling the system returns the fcc phase to bcc.
• Stability Field: The study predicts a stability field for fcc ice at much lower pressures (<60 GPa) than previously thought, explaining experimental detections at ~29–45 GPa.

C. Melting Point and Phase Diagram

• Melting Point Accuracy: Homogeneous bcc ice simulations exhibit significant superheating (melting >1200 K at 20 GPa). Introducing the interface lowers the melting point to ~962 K, which aligns closely with experimental results (~937–985 K).
• Resolution of Discrepancies: The study suggests that the wide scatter in experimental melting data is due to the coexistence of different phases (superionic bcc, superionic fcc, and liquid) near the melting line.
  • The "lower" experimental melting lines likely correspond to the bcc-fcc transition or superionic transitions rather than true melting.
  • The "higher" experimental lines correspond to the actual liquid-solid coexistence, which the interface-coupled model successfully reproduces.

5. Significance

• Bridging Theory and Experiment: This work provides a definitive explanation for long-standing inconsistencies in high-pressure ice physics. It proves that the "sample-anvil interface" in DAC experiments is a critical variable that cannot be ignored.
• Methodological Advancement: It demonstrates the power of combining Active Learning, Neural Network Potentials, and large-scale MD to simulate complex interfacial phenomena that are inaccessible to standard DFT due to system size limitations.
• Broader Implications: The findings suggest that interface effects may be a universal factor in high-pressure experiments involving other materials (e.g., diamond/Xe, diamond/hydrogen, diamond/ammonia), potentially inducing new phases or altering reaction pathways.
• Planetary Science: The revised phase diagram and understanding of superionic diffusion at lower pressures have implications for modeling the interiors of ice giants (Uranus and Neptune), where such conditions prevail.

In conclusion, the paper establishes that interface effects are fundamental to understanding the phase behavior of water ice under extreme pressure, necessitating a re-evaluation of experimental data and theoretical models that assume bulk homogeneity.