Role of anisotropic electronic friction in laser-driven hydrogen recombination on copper

This study utilizes machine-learning-enabled simulations to demonstrate that while anisotropic electronic friction significantly influences the rate of energy transfer and reaction probabilities during laser-driven hydrogen recombination on copper, the final molecular energy distributions are primarily governed by the potential energy landscape rather than the friction anisotropy.

The Gist

Imagine a crowded dance floor (the copper surface) where tiny dancers (hydrogen atoms) are waiting to pair up and jump off the stage (desorb as hydrogen gas). Usually, they need a lot of heat to get the energy to jump. But in this experiment, scientists hit the floor with a super-fast, intense laser flash. This laser acts like a sudden burst of energy that wakes up the "electrons" in the metal, turning them into a chaotic, super-hot crowd that bumps into the dancers.

The big question the scientists wanted to answer was: How exactly does this energy transfer happen, and does the direction of the bumps matter?

Here is the breakdown of their discovery using simple analogies:

1. The Two Ways of Pushing (Isotropic vs. Anisotropic)

To simulate this, the scientists used two different "rules" for how the hot electrons push the hydrogen atoms.

• The Old Rule (Isotropic Friction / LDFA): Imagine the electrons are like a giant, invisible fog. No matter which way the hydrogen atom tries to move, the fog pushes back with the same amount of resistance. It's like wading through thick, uniform mud. The push is the same whether you try to move up, down, left, or right.
• The New Rule (Anisotropic Friction / ODF): Imagine the electrons are like a complex, 3D maze of invisible springs and walls. The resistance changes depending on the direction. Maybe it's easy to slide sideways but hard to jump up, or vice versa. This is "anisotropic" friction—it's directional and depends on the specific shape of the atom's orbitals (its electron cloud).

2. The Race to the Exit (Diffusion and Reaction)

When the laser hits, the hydrogen atoms start vibrating and moving. They need to find each other, pair up, and climb a high energy "hill" (the barrier) to escape the surface.

• What the "Fog" (Old Rule) did: Because the fog pushed back equally in all directions, it gave the atoms a lot of energy very quickly, especially in the direction pointing away from the surface. It was like a strong wind blowing them straight up. This made them move around the dance floor very fast and pair up quickly. The simulation predicted that a lot of hydrogen would escape.
• What the "Maze" (New Rule) did: The directional maze was more realistic. It turned out that the resistance was actually weaker for moving up and away from the surface than the fog suggested. Because of this, the atoms didn't get as much "upward" energy as the old model thought. They moved slower and paired up less often. The simulation predicted much less hydrogen escaping.

The Analogy: Think of trying to run out of a building.

• The Old Model thought the doors were wide open and the floor was slippery, so everyone ran out fast.
• The New Model realized the doors were narrow and the floor was sticky in some spots, so fewer people made it out, even though they were running just as hard.

3. The Surprise: How They Jumped Off

The scientists expected that because the "Maze" model changed how many atoms escaped, it would also change how they jumped (how fast they were spinning or vibrating when they left).

But here is the twist: It didn't matter much!

Whether they used the "Fog" or the "Maze," the hydrogen atoms that did manage to escape had almost the exact same energy distribution. They were spinning and vibrating in the same way.

Why?
Think of the energy barrier (the hill they have to climb) as a funnel.

• The "Fog" and "Maze" models determined how many people got to the top of the funnel (the reaction rate).
• But once they were at the top, the shape of the funnel itself (the potential energy landscape) dictated exactly how they fell out the other side.

The "landscape" of the copper surface is so specific that it forces the hydrogen atoms to jump in a very particular way, regardless of how they got there. The "steering" of the electrons matters for getting to the jump, but the jump itself is dictated by the shape of the hill.

The Big Takeaway

This paper is a victory for precision.

1. Direction Matters for Speed: If you want to know how fast a chemical reaction happens or how much laser power you need to trigger it, you must use the complex, directional "Maze" model (Anisotropic). The old, simple "Fog" model overestimates how fast things happen.
2. Shape Matters for Outcome: If you want to know what the product looks like (how much energy the gas has when it leaves), the shape of the energy hill is the boss. The details of the electron friction don't change the final "dance moves" of the escaping gas.

In short: The scientists found that while the direction of the electron bumps controls the traffic (how many molecules escape), the shape of the terrain controls the destination (how the molecules fly away). To predict real-world chemical reactions driven by light, we need to stop treating electrons like a uniform fog and start treating them like a complex, directional landscape.

Technical Summary

1. Problem Statement

Ultrafast light-driven chemical dynamics at metal surfaces are governed by the transfer of energy from excited "hot" electrons to adsorbate vibrational degrees of freedom. While the Molecular Dynamics with Electronic Friction (MDEF) method is the standard approach for simulating these non-adiabatic processes, most existing studies rely on the Local Density Friction Approximation (LDFA).

• The Limitation: LDFA assumes that electronic friction is isotropic (the same in all spatial directions) and depends only on the local electron density.
• The Gap: Recent theoretical work suggests that friction is actually anisotropic and orbital-dependent (ODF), meaning the coupling between hot electrons and adsorbate vibrations varies significantly depending on the direction of motion and the specific electronic orbitals involved.
• The Question: How does the choice between isotropic (LDFA) and anisotropic (ODF) friction models affect the dynamics of laser-driven hydrogen recombination and desorption on a Cu(111) surface? Specifically, does it alter reaction probabilities, diffusion rates, or the final energy distribution of desorbed molecules?

2. Methodology

The authors employed a high-dimensional, machine-learning-enabled simulation framework to compare the two friction models under laser excitation.

• System: Recombinative desorption of $H_2$ from the Cu(111) surface. This system features a high activation barrier (0.592 eV) and strong friction anisotropy.
• Simulation Framework:
  • MDEF + TTM: The simulations coupled MDEF with the Two-Temperature Model (TTM). The TTM describes the non-equilibrium heating of the electron ($T_{el}$) and phonon ($T_{ph}$) subsystems following a femtosecond laser pulse.
  • Equation of Motion: The nuclear dynamics were governed by a Langevin equation including a conservative potential force, a dissipative friction force ($\Lambda \dot{R}$), and a stochastic random force ($\dot{W}$) representing thermal fluctuations.
• Machine Learning Potentials:
  • Potential Energy Surface (PES): Trained on Density Functional Theory (DFT) data using the SRP48 functional and the MACE (Machine Learning Atomic Cluster Expansion) architecture. This allowed for full-dimensional simulations including all surface degrees of freedom.
  • Electronic Friction Tensors (EFT):
    • LDFA Model: Based on the isotropic approximation using atomic cluster expansion.
    • ODF Model: Calculated directly from first-principles electron-phonon coupling using time-dependent perturbation theory, capturing the tensorial (anisotropic) nature of the friction.
• Simulation Parameters:
  • Laser fluences ranging from 10 to 120 J m$^{-2}$.
  • Initial surface temperatures from 100 K to 300 K.
  • Thousands of trajectories simulated for statistical convergence.

3. Key Contributions

1. First Full-Dimensional Comparison: This is one of the first studies to compare isotropic and anisotropic friction models in a full-dimensional, laser-driven desorption scenario using machine learning potentials.
2. Decoupling Mechanism from Rate: The study successfully decouples the factors governing the rate of reaction from those governing the final energy distribution of products.
3. Validation of ODF: It demonstrates that ODF models, which are computationally more expensive, are necessary for accurately predicting reaction yields and diffusion dynamics, even if they do not drastically change the final internal energy states of the products.

4. Key Results

A. Diffusion and Energy Transfer Rates

• Kinetic Energy Gain: Adsorbed hydrogen atoms gain kinetic energy roughly twice as fast in LDFA simulations compared to ODF simulations. This is because LDFA overestimates the friction coefficient for motion perpendicular to the surface.
• Diffusion Behavior: Consequently, the diffusion rate of hydrogen atoms is approximately 20% slower in ODF simulations. The population of high-energy bridge sites (crucial for recombination) is less pronounced in ODF compared to LDFA.
• Desorption Timing: The peak desorption rate occurs later in ODF simulations (0.75 ps) compared to LDFA (0.5 ps), reflecting the slower energy accumulation.

B. Reaction Probabilities (Fluence Dependence)

• Significant Discrepancy: The choice of friction model drastically alters the predicted desorption probability. LDFA simulations show significantly higher reactivity (up to an order of magnitude difference) across the fluence range.
• Power Law Scaling: The desorption probability follows a power law $P \propto F^X$.
  • LDFA: $X \approx 2.86$
  • ODF: $X \approx 2.30$
• Interpretation: The LDFA model likely overestimates the non-adiabatic coupling for perpendicular motion, leading to an overestimation of desorption yields. The ODF model provides a more realistic, albeit lower, reaction rate.

C. Final Energy Distributions (Translational, Vibrational, Rotational)

• Surprising Similarity: Despite the large differences in reaction rates and diffusion, the final energy partitioning of the desorbed $H_2$ molecules is remarkably similar between the two models.
• Dominance of PES: The final translational, vibrational, and rotational energies are primarily governed by the topology of the Potential Energy Surface (PES) at the reaction barrier, not by the details of the electronic friction.
• Subtle Differences: ODF desorbing molecules have slightly lower translational and rotational energy than LDFA, but vibrational energy remains high in both due to strong non-adiabatic coupling along the bond axis.

D. Timescales

• Diffusion Phase (Hundreds of fs): Electronic friction is the dominant driver, mobilizing atoms toward the recombination barrier. Here, the anisotropy of friction matters immensely.
• Desorption Phase (Tens of fs): Once the barrier is crossed, the process is ultrafast. The electronic friction does not significantly "steer" the molecule during the actual bond formation and detachment; the PES dictates the trajectory.

5. Significance and Conclusion

• Mechanism vs. Yield: The study concludes that while the mechanism of desorption (final energy states) is robust and determined by the PES, the reaction yield and fluence dependence are highly sensitive to the accuracy of the non-adiabatic coupling description.
• Model Selection: For predicting whether a reaction occurs (yields), anisotropic (ODF) friction is essential. For predicting how the energy is distributed in the products, isotropic (LDFA) may suffice, provided the PES is accurate.
• Future Directions: The presented workflow is fully scalable. The authors note that at higher surface coverages, off-diagonal friction components (which LDFA cannot capture) become more significant, suggesting that anisotropic models will be critical for studying high-coverage scenarios and complex surface facets.

In summary, the paper establishes that anisotropic electronic friction is critical for accurately predicting reaction rates and diffusion dynamics in laser-driven surface chemistry, but the final product energy distributions are largely dictated by the potential energy landscape.