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

arXiv:2604.00709 · physics.chem-ph · Submitted 2026-04-01 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "Role of anisotropic electronic friction in laser-driven hydrogen recombination on copper".

Mira: Here's the gist of the paper. Ultrafast light-driven chemical dynamics at surfaces are governed by energy transfer from excited electrons to vibrational degrees of freedom <ref:2604.00709#pg2> This paper compares isotropic and anisotropic models of electronic friction during laser-driven hydrogen evolution on the (111) facet of copper <ref:2604.00709#pg6> While anisotropic electronic friction determines the rate of energy transfer into the adsorbate and the fluence dependence of desorption probabilities, final translational, vibrational and rotational energy distributions are mainly governed by the potential energy landscape at the barrier and insensitive to nonadiabatic effects <ref:2604.00709#pg2>.

Kai: Context for Laser-Driven Dynamics Adsorption and desorption of molecular hydrogen from metal surfaces are important steps in heterogeneous catalysis and electrochemical processes <ref:2604.00709#pg2>.

Mira: Plasmonic metals such as copper, aluminium or gold are often of particular interest due to their efficiency in converting incident light into electron-hole pair excitations beyond thermal excitation thresholds <ref:2604.00709#pg4>.

Kai: These non-thermal excitations rapidly thermalise into a ”hot” electron distribution through electron-electron scattering with an effective temperature above the surrounding lattice While hot electron thermalisation typically occurs within several fs, electron-phonon scattering occurs over multiple ps.

Title and authors: Kai: So, to get into the specifics of what this paper, "Role of anisotropic electronic friction in laser-driven hydrogen recombination on copper," actually does, they perform full-dimensional TTM-MDEF simulations of laser-driven hydrogen evolution from a copper (one hundred eleven) surface <ref:2604.00709#pg1,Role of anisotropic electronic friction in laser-driven hydrogen recombination on copper>. They are using a coverage of two over nine monolayer ML.

Mira: That setup is specific, and it sets the stage for testing the models under controlled conditions. They’re comparing the isotropic Local Density Friction Approximation, LDFA, with this new anisotropic Orbital-Dependent Friction approximation, ODF, which they calculate from first-principles electron-phonon coupling

thirty–thirty-three: <ref:2604.00709#pg2,from first-principles electron-phonon coupling 30–33>.

Lev: The point is that they aren't just plugging in an arbitrary friction number; they are calculating the tensor directly. That’s what allows them to see how the friction changes depending on where the adsorbate atom is sitting on that copper surface, which is what anisotropy means.

Kai: And because of this calculation, they find that ODF accurately reproduces things like vibrational lifetimes and those spatially anisotropic energy dissipation effects in reactive scattering of nitrogen on ruthenium thirty-four and hydrogen on copper thirty-five <ref:2604.00709#pg2,spatially anisotropic energy dissipation effects in reactive scattering of>. That’s a strong validation point for the methodology itself.

Mira: But they also acknowledge that previous LDFA-based simulations were able to predict the non-equipartition of energy for hydrogen evolution from copper <ref:2604.00709#pg2>, which makes this study important because it directly addresses that missing piece of information.

Lev: So, what they’re showing is that by moving from LDFA to ODF, you get a more accurate picture of the energy dissipation during the process on copper <ref:2604.00709#pg2>. But we have to remember this is still a simulation setup within TTM-MDEF, which has its own assumptions about how it handles the electronic and phononic subsystems.

The paper's summary: Kai: Okay, so putting it all together, the paper summarizes that while anisotropic electronic friction determines the rate of energy transfer into adsorbates and how the fluence dependence of desorption probabilities looks, the final translational, vibrational, and rotational energy distributions are mainly governed by the potential energy landscape at the barrier.

Mira: That’s a crucial distinction they make. They say that even though friction affects how fast things happen—the rate of energy transfer into adsorbates—the actual final states of those molecules don't really depend on the nonadiabatic effects in terms of translational, vibrational, and rotational energy distributions <ref:2604.00709#pg2>.

Lev: That implies that the friction model is most important for steering where you go—like how fast you reach the barrier—rather than dictating the final resting place of the molecule once it’s off it. For our hardware, if we're worried about fidelity in energy measurement, this helps us focus our validation on those dynamics right near the transition state.

Kai: Right. So they’re saying that for things like desorption probabilities and how fast you get there, the electronic friction theory is key, but for the final energy states of the desorbed hydrogen molecules themselves—their translational and vibrational energy—the shape of the potential energy surface at that barrier dictates those results.

Mira: They also show a comparison in their findings: LDFA gives them values of n = three point zero five one plus or minus zero point two zero one for the power law relationship between desorption probability and laser fluence, while ODF simulations give n = two point three three seven plus or minus zero point one nine seven <ref:2604.00709#pg3>.

Lev: That difference in the exponent is significant, so it means that choosing the friction model has a real impact on predicting how often you get a desorbed molecule, which is something we need to be careful about when designing error-correction codes or anything involving chemical reactions.

The paper's improvements: Kai: The authors suggest a few ways this work improves things. First, they use the anisotropic description of orbital-dependent friction, ODF, which is directly calculated from first-principles electron-phonon coupling.

Mira: That’s the methodological improvement—moving away from LDFA to ODF because it gives a more physically detailed friction term that captures spatial anisotropy and mode dependence better. This allows for a more accurate description of energy dissipation effects in reactive scattering, like hydrogen on copper thirty-five <ref:2604.00709#pg2,energy dissipation effects in reactive scattering>.

Lev: And I see that as an improvement because it moves the physics closer to reality. When we build real hardware, you can't just use a simplified friction term; you have to account for these coupling effects if you want reliable predictions of reaction rates.

Kai: They also show how they can distinguish between two stages of laser-driven desorption: the "Diffusion" region and the "Desorption" region, by analyzing work components from the MDEF equation of motion.

Mira: That’s interesting because it lets them isolate exactly where the nonadiabatic energy transfer is happening—in the diffusion region, where atoms are mobilized by energy transfer, versus in the desorption region when they hit that transition state and recombine to form hydrogen <ref:2604.00709#pg2>.

Lev: So if we can map out those regions using the work terms from the MDEF equation, it tells us exactly where our computational effort should land for optimizing the physics of energy transfer in our own simulations.

Conclusion: Kai: So to wrap up on "Role of anisotropic electronic friction in laser-driven hydrogen recombination on copper," both isotropic and anisotropic models show differences in energy transfer and diffusion rates, but they agree that the desorption process itself is dominated by the barrier and the potential energy landscape.

Mira: The main implication is that choosing between LDFA and ODF primarily determines how quickly you reach that barrier for recombinative desorption, whereas the translational, vibrational, and rotational energy distribution of those desorbed hydrogen molecules is mostly defined by the potential energy surface at that barrier.

Lev: From my side, it’s clear that the choice of electronic friction approximation dictates how fast you reach that barrier to recombination, which is what matters when you think about running these simulations on real hardware and managing the overhead of those nonadiabatic effects.

Kai: And we've seen that even though the energy distributions of desorbed hydrogen molecules show only subtle differences between the two models, it confirms that the potential energy surface is still the main driver for those final states.

Mira: So, to summarize, this paper on "Role of anisotropic electronic friction in laser-driven hydrogen recombination on copper" shows that we need a more detailed friction model like ODF to get better predictions for reaction yields and rates when you have light driving the process.

Lev: For now, it’s clear that the choice of electronic friction approximation dictates how fast you reach the barrier, and once you have those dynamics mapped out, we can focus on where our hardware needs to be most accurate.

Alexander Spears, Wojciech G. Stark, Reinhard J. Maurer

Department of Chemistry, University of Warwick · University of Vienna · Department of Chemistry, Imperial College London · Department of Physics, University of Warwick

physics.chem-ph

Submitted: 2026-04-01

Updated: 2026-10-28

Journal ref: Phys. Rev. Lett. 137, 178001 (2026)

DOI: 10.1103/sply-f2km

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 86/100

The gist: Ultrafast light-driven chemical dynamics at surfaces are governed by energy transfer from excited electrons to vibrational degrees of freedom <ref:2604.00709#pg2> This paper compares isotropic and…

Key concepts

Electronic Friction Tensor (EFT)
This tensor describes the dissipative force caused by nonadiabatic electronic effects during surface dynamics. It quantifies how much energy is lost from the system due to coupling between electrons and nuclear motion, which is crucial for modeling laser-driven processes.
Molecular Dynamics with Electronic Friction (MDEF)
This simulation method combines classical molecular dynamics with quantum corrections. It uses the EFT as a dissipative force term in the equation of motion to account for nonadiabatic effects that classical simulations miss when studying light-driven surface reactions.
Potential Energy Landscape (PES)
The PES represents the energy surface governing the positions and interactions of atoms on a surface. The paper argues that while friction models affect how fast energy is transferred, the shape of this landscape ultimately defines where the molecules end up after desorption.
Orbital-Dependent Friction (ODF)
ODF is an anisotropic friction model calculated directly from first-principles electron-phonon coupling. It accounts for different types of orbital interactions, providing a more detailed description of how energy is dissipated based on the specific electronic states involved in the reaction.

Terminology

Summary

Ultrafast light-driven chemical dynamics at surfaces are governed by energy transfer from excited electrons to vibrational degrees of freedom <ref:2604. p. 0

Where the paper says it

Stark,1, 3 and Reinhard J. Maurer1, 2, 4, ∗ 1Department of Chemistry, University of Warwick, Gibbet Hill Road, CV4 7AL Coventry, UK 2University of Vienna, Faculty of Physics, Kolingasse 14-16, Vienna, Austria 3Department of Chemistry, Imperial College London, White City Campus, Molecular Sciences Research Hub, 82 Wood Ln, London W12 0BZ, UK 4Department of Physics, University of Warwick, Gibbet Hill Road, CV4 7AL Coventry, UK (Dated: August 25, 2026) Ultrafast light-driven chemical dynamics at surfaces are governed by energy transfer from excited electrons to vibrational degrees of freedom.

00709#pg2> This paper compares isotropic and anisotropic models of electronic friction during laser-driven hydrogen evolution on the (111) facet of copper <ref:2604. p. 0
Where the paper says it

When this nonadiabatic energy transfer is anisotropic, it can lead to dynamical steering effects that affect reaction probabilities or non-thermal final energy distributions in molecules. Here, enabled by a machine-learning-based simulation framework, we compare isotropic and anisotropic models of electronic friction during laser-driven hydrogen evolution on the (111) facet of copper.

00709#pg6> While anisotropic electronic friction determines the rate of energy transfer into the adsorbate and the fluence dependence of desorption probabilities, final translational, vibrational and rotational energy distributions are mainly governed by the potential energy landscape at the barrier and insensitive to nonadiabatic effects <ref:2604. p. 0
Where the paper says it

Here, enabled by a machine-learning-based simulation framework, we compare isotropic and anisotropic models of electronic friction during laser-driven hydrogen evolution on the (111) facet of copper. While anisotropic electronic friction determines the rate of energy transfer into the adsorbate and the fluence dependence of desorption probabilities, final translational, vibrational and rotational energy distributions are mainly governed by the potential energy landscape at the barrier and insensitive to nonadiabatic effects.

00709#pg2>.

Context for Laser-Driven Dynamics

Adsorption and desorption of molecular hydrogen from metal surfaces are important steps in heterogeneous catalysis and electrochemical processes <ref:2604.00709#pg2>. Plasmonic metals such as copper, aluminium or gold are often of particular interest due to their efficiency in converting incident light into electron-hole pair excitations beyond thermal excitation thresholds <ref:2604. p. 0

Where the paper says it

Although these steps are thoroughly studied in thermally activated catalysis, many open questions remain for photocatalytic processes on metal interfaces [2]. Plasmonic metals such as copper, aluminium or gold are often of particular interest due to their efficiency in converting incident light into electron-hole pair excitations beyond thermal excitation thresholds [3, 4].

00709#pg4>. These non-thermal excitations rapidly thermalise into a ”hot” electron distribution through electron-electron scattering with an effective temperature above the surrounding lattice While hot electron thermalisation typically occurs within several fs, electron-phonon scattering occurs over multiple ps. p. 0
Where the paper says it

These non-thermal excitations rapidly thermalise into a ”hot” electron distribution through electron-electron scattering with an effective temperature above the surrounding lattice [5]. While hot electron thermalisation typically occurs within several fs, electron-phonon scattering occurs over multiple ps [6].

Until the energy from electron-hole-pair (EHP) excitations is completely equilibrated with the surrounding lattice, energy can be transferred to adsorbates [5] driving chemical transformations [7]. p. 0
Where the paper says it

While hot electron thermalisation typically occurs within several fs, electron-phonon scattering occurs over multiple ps [6]. Until the energy from electron-hole-pair (EHP) excitations is completely equilibrated with the surrounding lattice, energy can be transferred to adsorbates [5], driving chemical transformations [7].

Simulation Methodology

Classical molecular dynamics simulations impose the Born-Oppenheimer approximation and are unsuitable to describe laser-driven dynamics, as they neglect nonadiabatic electron-nuclear coupling effects. A commonly employed mixed quantum-classical approach to study light-driven dynamics at surfaces is the Molecular Dynamics with Electronic Friction (MDEF) method. p. 0

Where the paper says it

While a fully ab-initio description of electron dynamics triggered by a laser pulse is possible [19], simulations of the ensuing coupled-electron nuclear dynamics of realistic interfaces are still computationally challenging [20]. A commonly employed mixed quantum-classical approach to study light-driven dynamics at surfaces is the Molecular Dynamics with Electronic Friction (MDEF) method [21–24].

The MDEF equation of motion includes nonadiabatic effects as a dissipative force mediated by an electronic friction tensor (EFT), Λ(R), and a coupled random force term, W˙ (T(t), t,Λ): mR¨ = −∇V − Λ(R)R˙ + W˙ (T(t),Λ) Light-driven surface dynamics can be simulated by coupling MDEF with the Two-Temperature Model (TTM) [25] a parametrised set of coupled heat transfer equations for the electronic (Tel) and phononic (Tph) subsystem temperatures. p. 0
Where the paper says it

In addition to the conservative dynamics governed by the potential energy surface (PES), V , the MDEF equation of motion includes nonadiabatic effects as a dissipative force mediated by an electronic friction tensor (EFT), Λ(R), and a coupled random force term, W˙ (T(t), t,Λ): mR¨ = −∇V − Λ(R)R˙ + W˙ (T(t),Λ). (1) Light-driven surface dynamics can be simulated by coupling MDEF with the Two-Temperature Model (TTM) [25], a parametrised set of coupled heat transfer equations for the electronic (Tel) and phononic (Tph) subsystem temperatures [26, 27] (see fig.

Comparison of Friction Models

In this Letter, we perform full-dimensional TTMMDEF simulations of laser-driven H2 evolution from a Cu(111) surface at a coverage of 2/9 ML using a tensorial and anisotropic description of orbital-dependent friction (ODF) that is directly calculated from first-principles electron-phonon coupling. ODF has been shown to accurately reproduce the mode dependence of vibrational lifetimes [12, 32] and spatially anisotropic energy dissipation effects in reactive scattering of N2 on Ru(0001) [34] and H2 on Cu [35]. p. 1

Where the paper says it

In this Letter, we perform full-dimensional TTMMDEF simulations of laser-driven H2 evolution from a Cu(111) surface at a coverage of 2/9 ML using a tensorial and anisotropic description of orbital-dependent friction (ODF) that is directly calculated from first-principles electron-phonon coupling [30–33]. ODF has been shown to accurately reproduce the mode dependence of vibrational lifetimes [12, 32] and spatially anisotropic energy dissipation effects in reactive scattering of N2 on Ru(0001) [34] and H2 on Cu [35].

However, LDFA is known to overestimate vibrational lifetimes of H atoms adsorbed on metals [9, 12, 32, 35]. p. 1
Where the paper says it

Previous LDFA-based simulations were able to predict the non-equipartition of energy for H2 evolution from Ru(0001) surfaces [22]. However, LDFA is known to overestimate vibrational lifetimes of H atoms adsorbed on metals [9, 12, 32, 35].

Here, we show that the non-equipartition of energy originates dominantly from the shape of the PES, while the rate of recombination and fluence dependence are determined by the employed electronic friction theory. p. 1
Where the paper says it

However, LDFA is known to overestimate vibrational lifetimes of H atoms adsorbed on metals [9, 12, 32, 35]. Here, we show that the non-equipartition of energy originates dominantly from the shape of the PES, while the rate of recombination and fluence dependence are determined by the employed electronic friction theory.

Impact on Desorption Outcomes

The choice of electronic friction approximation also leads to almost an order of magnitude difference in desorption yields, with LDFA showing significantly higher reactivity across the range of fluences investigated. The TTM-MDEF simulations find values of n = 3. p. 2

Where the paper says it

This relationship has not been measured experimentally for light-driven H2 recombination from Cu(111). The TTM-MDEF simulations find values of n = 3.

051±0.201 for LDFA and n = 2. p. 2
Where the paper says it

The TTM-MDEF simulations find values of n = 3. 201 for LDFA and n = 2.

337 ± 0.197 for ODF simulations. p. 2
Where the paper says it

201 for LDFA and n = 2. 197 for ODF simulations.

H2 molecules formed through laser-driven desorption are known to desorb translationally hot, and with a preference for certain rotational and vibrational states, however it is unclear whether the topology of the PES or the energy transfer due to electronic friction (or both) dominate this behaviour [7, 54]. p. 2
Where the paper says it

197 for ODF simulations. H2 molecules formed through laser-driven desorption are known to desorb translationally hot, and with a preference for certain rotational and vibrational states, however it is unclear whether the topology of the PES or the energy transfer due to electronic friction (or both) dominate this behaviour [7, 54].

Analysis of the translational, vibrational and rotational energy of desorbing H2 molecules in our simulations reveals only subtle differences in the energy partitioning between the two friction models (fig. p. 2
Where the paper says it

H2 molecules formed through laser-driven desorption are known to desorb translationally hot, and with a preference for certain rotational and vibrational states, however it is unclear whether the topology of the PES or the energy transfer due to electronic friction (or both) dominate this behaviour [7, 54]. Analysis of the translational, vibrational and rotational energy of desorbing H2 molecules in our simulations reveals only subtle differences in the energy partitioning between the two friction models (fig.

3b).

Mechanism of Desorption

The stochastic equation of motion contains ∇V, the potential gradient informing classical dynamics. The velocity-dependent friction term Λ ∂/∂tR describes the frictional force caused by nonadiabatic effects, while W(Λ, Tel, R) describes stochastically distributed thermal random noise. p. 2

Where the paper says it

The stochastic equation of motion contains ∇V , the potential gradient informing classical dynamics. The velocity-dependent friction term Λ ∂ ∂tR describes the frictional force caused by nonadiabatic effects, while W(Λ, Tel, R) describes stochastically distributed thermal random noise.

We can distinguish between two stages of a laser-driven desorption event: the ”Diffusion” region, where chemisorbed H atoms are mobilised by nonadiabatic energy transfer from EHPs and the ”Desorption” region, where two H atoms enter the transition state region of the PES and recombine to H2. p. 2
Where the paper says it

For a given simulation, we can determine the work of each of the three force terms described above as a function of time, which allows us to estimate the relevance of each contribution throughout the trajectory, as shown in fig. (Further details on work calculations in the SM [39]) We can distinguish between two stages of a laser-driven desorption event: The ”Diffusion” region, where chemisorbed H atoms are mobilised by nonadiabatic energy transfer from EHPs and the ”Desorption” region, where two H atoms enter the transition state region of the PES and recombine to H2.

In the diffusion regime, where the distance between H atoms is large, off-diagonal elements of the ODF EFT are significantly lower than diagonal components, so differences between LDFA and ODF mostly depend on the magnitudes of diagonal EFT components. p. 3
Where the paper says it

dissipates energy out of the system, but at a much lower rate. In the diffusion regime, where the distance between H atoms is large, off-diagonal elements of the ODF EFT are significantly lower than diagonal components, so differences between LDFA and ODF mostly depend on the magnitudes of diagonal EFT components.

During the desorption stage, i. p. 3
Where the paper says it

This in turn determines the significant differences in the LDFA and ODF energy transfer rates. During the desorption stage, i.

e. the tens of femtoseconds before and after the onset of desorption (fig. p. 3
Where the paper says it

During the desorption stage, i. the tens of femtoseconds before and after the onset of desorption (fig.

4b), both contributions are eclipsed by the energy contribution from the PES (WV), consistent with adiabatic, classical dynamics. p. 3
Where the paper says it

the tens of femtoseconds before and after the onset of desorption (fig. 4b), both contributions are eclipsed by the energy contribution from the PES (WV ), consistent with adiabatic, classical dynamics.

Conclusion on Dominance

In summary, we have compared machine-learning accelerated TTM-MDEF simulations of H2 recombination from Cu(111) surfaces using an isotropic and an anisotropic friction model. Both models yield differences in energy transfer, diffusion rates and desorption probabilities, yet the desorption process itself is dominated by the barrier and the potential energy landscape. p. 3

Where the paper says it

In summary, we have compared machine-learning accelerated TTM-MDEF simulations of H2 recombination from Cu(111) surfaces using an isotropic and an anisotropic friction model. Both models yield differences in energy transfer, diffusion rates and desorption probabilities, yet the desorption process itself is dominated by the barrier and the potential energy landscape.

p. 3
Where the paper says it

In summary, we have compared machine-learning accelerated TTM-MDEF simulations of H2 recombination from Cu(111) surfaces using an isotropic and an anisotropic friction model. Both models yield differences in energy transfer, diffusion rates and desorption probabilities, yet the desorption process itself is dominated by the barrier and the potential energy landscape.

The choice of electronic friction approximation primarily determines how quickly the energy barrier to recombinative desorption will be reached, while the translational, vibrational and rotational energy distribution of the desorbing H2 molecules is mostly defined by the PES. p. 3
Where the paper says it

Both models yield differences in energy transfer, diffusion rates and desorption probabilities, yet the desorption process itself is dominated by the barrier and the potential energy landscape. The choice of electronic friction approximation primarily determines how quickly the energy barrier to recombinative desorption will be reached, while the translational, vibrational and rotational energy distribution of the desorbing H2 molecules is mostly defined by the PES.

p. 3
Where the paper says it

Both models yield differences in energy transfer, diffusion rates and desorption probabilities, yet the desorption process itself is dominated by the barrier and the potential energy landscape. The choice of electronic friction approximation primarily determines how quickly the energy barrier to recombinative desorption will be reached, while the translational, vibrational and rotational energy distribution of the desorbing H2 molecules is mostly defined by the PES.

Supporting Data for Model Validation

The MACE models are accurate enough to describe laser-driven desorption. To assess the accuracy of the ACEfriction. p. 7

Where the paper says it

The cutoff distance was set at 4. The ACEfriction.

jl6 model used to predict the ODF electronic friction tensor (EFT) for simulations in this work, we compare predicted EFTs to their reference values calculated using time-dependent perturbation theory (TDPT) for the structures sets previously described. p. 7
Where the paper says it

The electron density for different surface configurations used to train LDFA models was calculated using the output cube total density keyword. To determine the electronic friction tensor (EFT) from first principles electron phonon coupling (orbital dependent friction, ODF), a “light” basis set with additional “second-tier” improvements for hydrogen was found to yield sufficiently converged results.

The DFT settings used to calculate the EFT are described in section SI A. p. 9
Where the paper says it

S4 intensity maximum of the laser pulse corresponds to t = 0. The electronic friction tensor was calculated using one of the two models described in section SI C.

Uncertainty and Future Work

By resampling the desorption probabilities 4000 times, we can quantify the uncertainty in the resulting fitting coefficients. The difference in power law exponents obtained for LDFA and ODF is statistically significant using the above method. p. 2

Where the paper says it

This is slightly more pronounced for ODF, where the component of friction perpendicular to the surface is lower than for LDFA. Considering the significant difference in magnitude of friction between the two models, the resulting differences in the quantumstate-resolved energy distributions of desorbed molecules are surprisingly small.

Vibrational lifetimes of adsorbates are sensitive to coverage [9] and further detailed studies at higher coverages as well as a comparison with experimental observables, e. p. 3
Where the paper says it

However, our comparison only holds in the low coverage regime, where off-diagonal components in ODF are small. Vibrational lifetimes of adsorbates are sensitive to coverage [9] and further detailed studies at higher coverages as well as a comparison with experimental observables, e.

g. as previously conducted on ruthenium surfaces [7], will be needed to show whether frictional steering and nonadiabatic energy transfer differ in highcoverage scenarios. p. 3
Where the paper says it

Vibrational lifetimes of adsorbates are sensitive to coverage [9] and further detailed studies at higher coverages as well as a comparison with experimental observables, e. as previously conducted on ruthenium surfaces [7], will be needed to show whether frictional steering and nonadiabatic energy transfer differ in highcoverage scenarios.

Energy Partitioning Details

For classical H2 in the gas phase, we can partition the total energy into the following contributions: Etot = Epot + Etrans + Erot + Evib. Mean energies are determined through bootstrapping with 1000 re-sampling steps. p. 2

Where the paper says it

Statistical uncertainties from bootstrapping analysis are indicated where they are larger than the markers used. (b): Mean internal energies of desorbing H2 molecules determined from LDFA and ODF simulations for a Cu(111) surface with an initial temperature of 100 K and a laser fluence of 120 J m−2 .

The classical H2 molecule has six total degrees of freedom: Three translational, two rotational and one vibrational. p. 0
Where the paper says it

In addition, laser-desorbed H2 molecules generally have a non-thermal partitioning of energy across vibrational, rotational and translational degrees of freedom [7]. However, it is still not clear if this nonequipartition of energy in the molecule results from nonadiabatic effects or from details of the barrier landscape during recombinative desorption.

Work Contribution Analysis

For a given simulation, we can determine the work of each of the three force terms described above as a function of time, which allows us to estimate the relevance of each contribution throughout the trajectory. The thermal fluctuation force is defined as FW(t) = p2mΛkBTel(t) ∗ Normal(0, 1) √∆t. p. 2

Where the paper says it

(Further details on work calculations in the SM [39]) We can distinguish between two stages of a laser-driven desorption event: The ”Diffusion” region, where chemisorbed H atoms are mobilised by nonadiabatic energy transfer from EHPs and the ”Desorption” region, where two H atoms enter the transition state region of the PES and recombine to H2. In the first region, the thermal fluctuation work WW continuously transfers energy from hot electrons into the molecule and induces H atom diffusion.

The force of friction is defined as Λ ∂/∂tR, where Λ is the friction tensor. p. 0
Where the paper says it

A commonly employed mixed quantum-classical approach to study light-driven dynamics at surfaces is the Molecular Dynamics with Electronic Friction (MDEF) method [21–24]. In addition to the conservative dynamics governed by the potential energy surface (PES), V , the MDEF equation of motion includes nonadiabatic effects as a dissipative force mediated by an electronic friction tensor (EFT), Λ(R), and a coupled random force term, W˙ (T(t), t,Λ): mR¨ = −∇V − Λ(R)R˙ + W˙ (T(t),Λ).

Desorption Probability Definition

H2 molecules are counted as desorbed when the centre of mass of the molecule reaches a distance higher than 6.0 ˚A from the average position of the top layer of surface slab Cu atoms. p. 2

Where the paper says it

Due to the high symmetry and lack of steps or edges on the Cu(111) surface, collision events between hydrogen atoms are most likely to occur when both atoms are at similar heights over the surface, and the MEP favours a concerted motion away from the surface. These two factors likely discourage cartwheel rotations on desorption from a terrace, leading to desorption with, on average, low rotation energy.

The transition state to desorption is determined based on the mean H-H distance after desorption has begun. p. 2
Where the paper says it

(b): Mean internal energies of desorbing H2 molecules determined from LDFA and ODF simulations for a Cu(111) surface with an initial temperature of 100 K and a laser fluence of 120 J m−2 . tions [53], the desorption probability follows a power law relationship of the form |A| · F n with the laser fluence.

Searching backward in time from the point of desorption for at least 20 fs, the onset of desorption is reached when the H-H bond distance increases beyond one standard deviation of the mean (µ + σ).

Energy Distribution Comparison

In ODF simulations, H2 molecules desorb with lower overall energy than in LDFA simulations. However, the absolute vibrational energy remains roughly equal for both methods. p. 2

Where the paper says it

In ODF simulations, H2 molecules desorb with lower overall energy than in LDFA simulations. However, the absolute vibrational energy remains roughly equal for both methods.

This may be a weak indication of the stronger nonadiabatic coupling along the bond axis for ODF (fig. p. 2
Where the paper says it

However, the absolute vibrational energy remains roughly equal for both methods. This may be a weak indication of the stronger nonadiabatic coupling along the bond axis for ODF (fig.

1b) [14, 55].

Friction Tensor Comparison

The magnitude of ODF only exceeds LDFA in the transition state region and only along the recombination mode (intramolecular stretch), shown as Λrr in fig. 1b. The off-diagonal components of ODF are considerably lower during on-surface diffusion and only reach higher values around the transition to desorption. p. 1

Where the paper says it

The ODF is weaker perpendicular to the surface compared to components parallel to the surface (Figure S13 in the SM). The off-diagonal components of ODF are considerably lower during on-surface diffusion and only reach higher values around the transition to desorption.

Phonon Thermostatting Effect

Overall, the choice of phonon thermostatting scheme does not significantly impact desorption probabilities for LDFA or ODF simulations (see fig. S7 and tables SIII, SIV). p. 6

Where the paper says it

Energy partitioning across translation, rotation and vibration S7 G. Minimum Energy Paths for desorption S8 SIII.

In addition, the energy distributions of desorbing H2 molecules across translational (fig. p. 0
Where the paper says it

In laser-induced dynamics, isotope effects can be enhanced [15], trapping in precursor states can be promoted [16], or direction-specific diffusion can be induced [17]. In addition, laser-desorbed H2 molecules generally have a non-thermal partitioning of energy across vibrational, rotational and translational degrees of freedom [7].

S8), rotational (fig. p. 2
Where the paper says it

H2 molecules formed through laser-driven desorption are known to desorb translationally hot, and with a preference for certain rotational and vibrational states, however it is unclear whether the topology of the PES or the energy transfer due to electronic friction (or both) dominate this behaviour [7, 54]. Analysis of the translational, vibrational and rotational energy of desorbing H2 molecules in our simulations reveals only subtle differences in the energy partitioning between the two friction models (fig.

S9) and vibrational (fig. p. 2
Where the paper says it

H2 molecules formed through laser-driven desorption are known to desorb translationally hot, and with a preference for certain rotational and vibrational states, however it is unclear whether the topology of the PES or the energy transfer due to electronic friction (or both) dominate this behaviour [7, 54]. Analysis of the translational, vibrational and rotational energy of desorbing H2 molecules in our simulations reveals only subtle differences in the energy partitioning between the two friction models (fig.

S10) degrees of freedom are not significantly affected by the phonon thermostatting approach. p. 6
Where the paper says it

ACE-friction ODF models S10 SIV. Impact of phonon thermostatting S11 SV.

Final Summary

In summary, we have compared machine-learning accelerated TTM-MDEF simulations of H2 recombination from Cu(111) surfaces using an isotropic and an anisotropic friction model. Both models yield differences in energy transfer, diffusion rates and desorption probabilities, yet the desorption process itself is dominated by the barrier and the potential energy landscape. The choice of electronic friction approximation primarily determines how quickly the energy barrier to recombinative desorption will be reached, while the translational, vibrational and rotational energy distribution of the desorbing H2 molecules is mostly defined by the PES.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Anisotropic friction models can be integrated into machine learning based simulations to determine reaction probabilities in laser-driven hydrogen evolution on copper (111) facets, allowing for a more accurate prediction of desorption yields by comparing results from isotropic and anisotropic friction theories.

  2. The improved system can accurately model the nonadiabatic effects governing the rate of energy transfer into adsorbates by employing orbital dependent friction (ODF) which is directly calculated from first-principles electron-phonon coupling, leading to better descriptions of mode dependence and spatially anisotropic energy dissipation effects.

  3. The AI system can distinguish between two stages of a laser driven desorption event, specifically the Diffusion region where H atoms are mobilized by nonadiabatic energy transfer and the Desorption region, by analyzing work components from the Molecular Dynamics with Electronic Friction (MDEF) equation of motion.

  4. The simulation can be used to quantify how much of the energy barrier is overcome during desorption by analyzing the work terms in the MDEF equation, allowing researchers to estimate the relevance of each contribution throughout the trajectory.

  5. The system can accurately predict the power law relationship between desorption probability and laser fluence for hydrogen recombination on Cu(111) surfaces, yielding distinct exponents for different friction approximations (nLDFA = 3.051 ± 0.201 and nODF = 2.337 ± 0.197).

  6. The AI system can provide detailed analysis of the final translational, vibrational, and rotational energy distributions of desorbed hydrogen molecules by revealing only subtle differences in the energy partitioning between the two friction models, which helps clarify whether PES topology or electronic friction dominates this behavior.

Abstract

Ultrafast light-driven chemical dynamics at surfaces are governed by energy transfer from excited electrons to vibrational degrees of freedom. When this nonadiabatic energy transfer is anisotropic, it can lead to dynamical steering effects that affect reaction probabilities or non-thermal final energy distributions in molecules. Here, enabled by a machine-learning-based simulation framework, we compare isotropic and anisotropic models of electronic friction during laser-driven hydrogen evolution on the (111) facet of copper. While anisotropic electronic friction determines the rate of energy transfer into the adsorbate and the fluence dependence of desorption probabilities, final translational, vibrational and rotational energy distributions are mainly governed by the potential energy landscape at the barrier and insensitive to nonadiabatic effects.

Related papers