Water structuring at stacked graphene interfaces unveiled by machine-learning molecular dynamics

By combining machine-learning molecular dynamics with vibrational sum-frequency generation simulations, this study reveals that the apparent hydrophilicity of monolayer graphene on hydrophilic substrates stems from thermodynamically favorable intercalated water molecules causing signal cancellation, rather than wetting transparency, while multilayer graphene lacks this intercalation.

The Gist

Imagine you have a piece of graphene. It's a single layer of carbon atoms, thinner than a human hair, and it's often called a "wonder material" because it's incredibly strong and conductive. But for years, scientists have been arguing over one simple question: Is graphene wet or dry?

If you put a drop of water on it, does it spread out (hydrophilic/wet) or bead up like a raindrop on a duck's back (hydrophobic/dry)?

The answer seemed to change depending on who you asked and what experiment they did. Some said it was dry; others said it was wet. This paper acts like a high-tech detective story, using super-computers and artificial intelligence to solve the mystery.

Here is the story of what they found, explained simply.

The Mystery: The "Ghost" Water

Think of graphene as a very thin, invisible sheet of plastic wrap.

• The Floating Sheet: When graphene floats freely in water, it acts like a duck's back. The water beads up. It's hydrophobic (water-repelling).
• The Supported Sheet: But when you stick that same graphene sheet onto a glass-like surface (called Calcium Fluoride or CaF2), something weird happens. The water suddenly spreads out. It looks hydrophilic (water-loving).

Scientists used to think this was because the graphene was "transparent" to wetness. They imagined the water could "see" through the graphene and feel the wet glass underneath, so it spread out. They called this "Wetting Transparency."

The Investigation: The AI Detective

The authors of this paper didn't just guess; they built a digital world. They used Machine Learning (a type of AI) to simulate the movement of billions of water molecules around graphene. Think of this AI as a super-fast, ultra-accurate camera that can see every single water molecule dancing and spinning.

They simulated three main scenarios:

1. Graphene floating in water.
2. Graphene stuck to a wet glass surface (with no water trapped underneath).
3. Graphene stuck to a wet glass surface with water trapped underneath.

The Big Reveal: It's Not Transparency, It's a "Sandwich"

The AI simulations revealed that the "Wetting Transparency" theory was wrong. The graphene isn't letting the water see the glass. Instead, the water is sneaking underneath.

Here is the analogy:
Imagine you are standing on a stage (the glass substrate). You put a very thin, clear curtain (the graphene) in front of you.

• The Old Theory: People thought the audience (the water) could see you through the curtain, so they reacted to you.
• The New Reality: The audience didn't see you through the curtain. Instead, a few audience members slipped behind the curtain and stood right next to you!

Because the glass (substrate) is naturally "sticky" to water, it pulls water molecules into the tiny gap between the glass and the graphene. This creates a sandwich:

• Top: Water above the graphene.
• Middle: The graphene sheet.
• Bottom: A hidden layer of water trapped between the graphene and the glass.

Why Does This Matter?

The "hidden water" layer changes the signal.

• The Signal: Scientists use a special laser technique (called vSFG) to "listen" to how water molecules vibrate.
• The Confusion: When the hidden water layer is present, it cancels out the "dry" signal from the top layer and creates a "wet" signal.
• The Result: The experiment looks like the graphene is wet, but it's actually a trick caused by the water hiding underneath.

The Layer Count Rule

The paper also discovered a rule about how many layers of graphene you have:

• 1 Layer (Monolayer): The gap is tiny, but the "sticky" glass is strong enough to pull water in. The sandwich forms. The graphene looks wet.
• 4+ Layers (Multilayer): The graphene stack becomes too thick and heavy. The "sticky" glass can't pull water through the whole stack to get underneath. No sandwich forms. The water stays on top, beads up, and the graphene acts dry (hydrophobic).

The Takeaway

This paper solves a long-standing debate. Graphene is intrinsically dry (hydrophobic). It doesn't care if it's on glass or floating in air.

The reason it looks wet on glass is that water molecules sneak underneath it, creating a hidden layer that tricks our sensors.

Why should you care?
If you are building future devices like super-fast computers, water filters, or medical sensors using graphene, you need to know this. If you don't control the environment perfectly, water might sneak under your graphene layers, ruining the device's performance. You have to seal the edges or keep the air super dry to stop the "water sandwich" from forming.

In short: Graphene isn't transparent to water; it's just really good at hiding water underneath it!

Technical Summary

1. Problem Statement

The wettability of graphene (whether it is intrinsically hydrophobic or hydrophilic) remains a subject of long-standing debate. Experimental measurements of the water contact angle (WCA) and vibrational sum-frequency generation (vSFG) spectra have yielded conflicting results:

• Floating graphene on water appears hydrophobic.
• Graphene supported on hydrophilic substrates (e.g., CaF₂) often exhibits hydrophilic behavior in vSFG spectra.

Existing explanations, such as "wetting transparency" (the idea that the substrate's wettability penetrates the graphene layer), have not been fully resolved. Furthermore, experimental vSFG spectra are difficult to interpret because they are macroscopic observables influenced by multiple factors: the substrate, the number of graphene layers, defects, and the presence of intercalated water (water trapped between the graphene and the substrate). Traditional ab initio molecular dynamics (AIMD) simulations are too computationally expensive to capture the nanosecond timescales required for statistical convergence of vSFG spectra in these complex, multi-layer systems.

2. Methodology

The authors developed a high-accuracy computational framework combining Machine Learning Interatomic Potentials (MLIP) with Molecular Dynamics (MD) simulations to overcome the limitations of AIMD.

• Machine Learning Potential (MLIP):
  • Utilized the Atomic Cluster Expansion (ACE) method to train potentials on Density Functional Theory (DFT) data.
  • Employed an Active Learning (AL) strategy: The MLIP was iteratively refined by identifying configurations with high prediction errors during MD simulations, recalculating them with DFT, and retraining the potential.
  • Accuracy: Achieved energy RMSE < 0.50 meV/atom and force RMSE < 40 meV/Å, ensuring "chemical accuracy" necessary for describing subtle water-graphene interactions.
• System Models:
  • Simulated various configurations labeled as $S_nL_m$ (Substrate, $n$ intercalated water molecules, $m$ graphene layers).
  • Systems included: Monolayer ($L1$) and multilayer ($L5$) graphene without substrates; systems with CaF₂ substrates ($SL1, SL4, SL5$); and systems with varying amounts of intercalated water ($S_{24}L1, S_{48}L1$, etc., up to full monolayer ice).
  • Included ordered ice phases (hexagonal, pentagonal, square) and amorphous water layers.
• vSFG Spectra Simulation:
  • Calculated vSFG spectra from MD trajectories using many-body dipole moment and polarizability surfaces.
  • Applied screening functions to isolate signals from specific interfaces (water-graphene vs. water-substrate).
• Thermodynamic Analysis:
  • Calculated the Gibbs free energy ($\Delta G$) of water intercalation using internal energy from MD and entropy via the Two-Phase Thermodynamic (2PT) model.

3. Key Contributions

1. Development of System-Specific MLIPs: Created highly accurate, system-specific ACE potentials capable of simulating large-scale, long-timescale (nanosecond) dynamics of complex graphene-water-substrate interfaces, a task previously infeasible with AIMD.
2. Mechanistic Resolution of Wettability Debate: Provided a definitive explanation for the apparent hydrophilicity of graphene on CaF₂, refuting the "wetting transparency" hypothesis in favor of an intercalation mechanism.
3. Layer-Dependent Intercalation Energetics: Demonstrated that water intercalation is thermodynamically favorable for monolayer graphene but energetically unfavorable for multilayer graphene (4+ layers).

4. Key Results

A. Intrinsic Hydrophobicity of Graphene

• Simulations of pristine graphene (floating or supported without intercalated water) confirmed that both monolayer and multilayer graphene are intrinsically hydrophobic.
• Layer Effect: Hydrophobicity increases with the number of layers. Multilayer graphene ($L5$) showed a higher population of "dangling" O-H bonds (pointing away from the surface) compared to monolayer graphene ($L1$), resulting in stronger hydrophobic vSFG signatures (enhanced peak at ~3665 cm⁻¹).

B. Refutation of Wetting Transparency

• Simulations of graphene on CaF₂ without intercalated water ($SL1, SL4, SL5$) showed that the vSFG spectra remained hydrophobic and did not resemble the hydrophilic CaF₂ substrate spectrum.
• This contradicts the "wetting transparency" theory, proving that the substrate's wettability does not simply transmit through the graphene layer.

C. The Role of Intercalated Water

• Monolayer Graphene: The presence of intercalated water between monolayer graphene and the CaF₂ substrate drastically alters the vSFG spectrum.
  • Intercalated water molecules orient with O-H bonds pointing toward the graphene, creating a signal that cancels out the hydrophobic "dangling O-H" signal from the top interface.
  • This signal cancellation results in a spectrum that mimics a hydrophilic surface, explaining the experimental observations.
• Multilayer Graphene: As the number of layers increases, the vSFG spectrum retains its hydrophobic character even with intercalated water, because the top interface dominates the signal and the hydrophobicity of the top layers is stronger.

D. Thermodynamic Driving Forces

• Free Energy Calculations:
  • Monolayer ($n=1$): Water intercalation is thermodynamically favorable ($\Delta G < 0$). Water spontaneously enters the gap to minimize system free energy.
  • Multilayer ($n \ge 4$): Water intercalation is thermodynamically unfavorable ($\Delta G > 0$, > 1.5 eV). The energy cost to confine water between thick graphene layers and the substrate is too high.
• This energetic difference explains why experimental vSFG spectra show hydrophilic features for monolayers (due to spontaneous intercalation) but hydrophobic features for multilayers (where intercalation is suppressed).

5. Significance and Implications

• Reinterpretation of Experiments: The study suggests that previous interpretations of "wetting transparency" were likely misinterpretations caused by unaccounted-for intercalated water. The observed hydrophilicity is an artifact of the experimental setup (immersion in water) allowing water to seep under the graphene.
• Device Design: For graphene-based devices (e.g., field-effect transistors, sensors), controlling environmental humidity is critical. Unintentional water intercalation can alter electronic properties (bandgap, carrier mobility) and device stability.
• Experimental Protocols: To measure intrinsic wettability, the authors propose transferring graphene under high vacuum and sealing edges (e.g., with gold films) to prevent water ingress during vSFG measurements.
• Methodological Advance: The work demonstrates the power of MLIPs in solving complex interfacial problems where quantum accuracy and long timescales are simultaneously required, setting a precedent for future studies of 2D material interfaces.

In conclusion, the paper establishes that pristine graphene is hydrophobic, and the hydrophilic behavior observed in experiments on monolayer graphene is a result of spontaneous water intercalation driven by thermodynamics, not wetting transparency. This intercalation is suppressed in multilayer systems, leading to the observed layer-dependent wettability.