Molecular photodissociation enabled by ultrafast plasmon decay

Polaritonic chemistry and molecular polaritonics are the fields inspired by the possibility to manipulate and control molecular structure and dynamics through strong interactions with confined electromagnetic (EM) modes. Interest in these possibilities has increased strongly over the last few years from a range of different and previously independent disciplines. General overviews over the possibilities afforded in such systems can be found in a series of reviews published over the last few years.^1–6 Most initial interest was focused on “macroscopic” settings where a (large) collection of molecules interacts with optical cavity modes (e.g., in Fabry–Perot planar mirror cavities), in which the so-called strong-coupling regime can be reached relatively easily with organic molecular materials.⁷ This regime is reached when the (collective) interaction strength between light and matter excitations becomes larger than the relevant dissipation rates in the system. The excitations of the coupled system then become polaritons, i.e., hybrid light–matter excitations that are separated in energy by the so-called Rabi splitting. In particular, the possibility to manipulate (photo)chemical reactions in such settings has been demonstrated experimentally^8–12 and analyzed theoretically.^13–20 

While collective light–matter coupling involving a macroscopic number of molecules can be used to modify a variety of molecular properties, its influence at the single-molecule level is limited due to the delocalized nature of polaritons in such settings.^13,19–21 Much larger and more direct changes at the single-particle level can be achieved by increasing the single-molecule coupling strength, which requires increasing the confinement of the electromagnetic field modes. In order to approach the single-molecule strong-coupling limit at room temperature and with the dipole transition strengths available in organic molecules, strong subwavelength field confinement is required, which, in turn, requires that the real part of the dielectric permittivity of the “cavity” material becomes negative.²² At optical frequencies, this essentially means that the only currently available setups with significant single-molecule coupling strengths are based on metallic structures supporting plasmonic resonances (i.e., oscillations of the free electrons in the metal). In such systems, strong coupling can be reached with relatively few emitters,²³ even down to the single-molecule level.^24,25 It has also been shown that such setups can be used to modify molecular reactions such as photo-oxidation.²⁶ However, plasmonic resonances unavoidably come with large losses due to a significant fraction of their energy being stored in the motion of free electrons in the metal.²² The resulting resonance lifetimes are typically on the order of 10 fs, short compared to the time needed for most chemical reactions to take place. Very recently, theoretical efforts have thus started to be directed toward making virtue out of necessity, i.e., toward exploiting the ultrafast loss in plasmonic nanocavities and nanoantennas to achieve desired functionalities.^27–31 

Along similar lines, in this article, we devise a strategy for using ultrafast decay induced by coupling to plasmonic modes to permit photodissociation in molecules that do not normally do so upon photoexcitation. The general idea is to induce ultrafast decay at a specific position on the excited-state potential energy surface (PES) in such a way that the kinetic energy accumulated by the nuclei after photoexcitation at the Franck–Condon point is sufficient to overcome the dissociation barrier on the ground-state electronic surface after photon loss, even when photodissociation is not energetically allowed on the excited-state PES. We demonstrate the concept using the simplest neutral molecule, hydrogen (H₂). It turns out that the potential energy curves for the two lowest electronic states of H₂ exactly fulfill the conditions for which the process we propose takes place efficiently. Additionally, aluminum nanoantennas can provide plasmonic resonances with the required characteristics (energy and decay rate).

The system we consider, sketched in the inset of Fig. 1, consists of a single hydrogen molecule close to an aluminum nanoparticle. The molecular electronic structure is approximated by the two lowest electronic states within the Born–Oppenheimer approximation, the ground state $X1Σg+$ and first excited state $B1Σu+$⁠. In the following, we will refer to these states as simply the X and B states, respectively. Rotational motion is neglected, and the potential energy curves V_X(R) and V_B(R) as a function of the one remaining nuclear coordinate, internuclear distance R, are shown in the main panel of Fig. 1. The transition dipole moment between them is aligned along the molecular axis, and its value along that axis, μ_XB(R), is shown in the upper panel of Fig. 1. The values of V_X(R), V_B(R), and μ_XB(R) are taken from accurate reference calculations.^32–34 Additionally, both the nuclear position and the energy of the Franck–Condon point (corresponding to a vertical transition from X to B at the minimum position in X, R_FC ≈ 1.4 a.u.) are shown as black dashed lines. As its energy is significantly below the dissociation limit of the B state, V_B(R → ∞), photoexcitation will typically not lead to dissociation in the bare molecule (under instantaneous excitation, the dissociation probability is ∼1%).

The aluminum nanoparticle is modeled as a perfect sphere and represented by a single bosonic mode describing a plasmonic pseudomode.^35–37 Such a pseudomode corresponds to a highly localized excitation arising as a coherent superposition of many high-order multipole modes with similar frequencies. For sufficiently short distances between the molecule and the nanoparticle (on the order of a few nanometers and below), the pseudomode is by far the most strongly coupled mode. When treated perturbatively (i.e., under the assumption of weak light–matter coupling), it is responsible for quenching,^38–41 but this simple picture breaks down at short emitter-nanoparticle distances where the coupling becomes comparable to the decay rates and the strong-coupling regime is entered.³⁶ Although such pseudomodes are nonradiative and cannot be excited (and thus detected or characterized) by far-field radiation, they can still be used as effective quantized “cavity” modes with large coupling strengths that can be useful for applications in polaritonic chemistry and in molecular polaritonics, such as photoprotection.²⁹ In order to quantize the pseudomode, we calculate the electromagnetic spectral density⁴² $J(ω)=μ2ω2πℏε0c2n→⋅ImG(r→e,r→e,ω)⋅n→$⁠, where $G(r→1,r→2,ω)$ is the dyadic Green’s function solving the (macroscopic) Maxwell equations at frequency ω, $r→e$ is the emitter position, $n→$ is the orientation of its transition dipole moment, and μ is its amplitude. We then use the well-known fact that Lorentzian peaks in J(ω)/μ² correspond to lossy quantized modes with mode energy ω_p, loss rate κ, and single-photon field strength E_1ph determined by the peak position, bandwidth, and amplitude of the peaks.^{35,36,43–45} While this approach can be extended to more complex spectral densities by allowing interactions between the quantized modes,⁴⁶ this is not necessary here as the relevant spectral densities are dominated by a single peak.

The Hamiltonian resulting from the above considerations is (ℏ = 1 here and in the following)

where a, a^† are the plasmon annihilation and creation operators, ω_m(R) = V_B(R) − V_X(R) is the position-dependent molecular excitation energy, and σ = |X⟩⟨B| is the molecular de-excitation operator. We have here treated the light–matter interaction within the dipole approximation and additionally used the rotating wave approximation in which the total number of excitations (molecular excitations + photons) is conserved. This is a good approximation as long as the coupling strength is small compared to the transition energies, i.e., the system does not enter ultrastrong coupling.⁴⁷ We also note that the pseudomode (and nanoparticle plasmons in general) is very well approximated as a quasistatic excitation that only interacts with the molecule through the longitudinal electric field (i.e., Coulomb interactions). This has the consequence that irrespective of whether the quantized EM field is treated using minimal coupling or in the Power–Zienau–Woolley picture, the light–matter interaction within the dipole approximation is simply given by $E→⋅μ→$⁠, and no dipole self-energy term is present in the Hamiltonian.^48,49

In the following, we will use the notation |i, n⟩, with i = (X, B) and n = (0, 1, …), for the combined electronic-photonic state. The main panel in Fig. 1 additionally shows the potential energy surface for the state |X, 1⟩, corresponding to the molecule in its ground state and a single plasmon excitation. The associated potential energy is simply V_X(R) + ω_p, i.e., the ground-state potential energy surface shifted upward by that amount.^2,13 The plasmon energy ω_p = 7.6 eV is chosen such that the curves corresponding to plasmonic and molecular excitations touch tangentially (at R ≈ 4 a.u.), while the dissociation limit of the |X, 1⟩ curve lies below the Franck–Condon energy for excitation of the bare molecule (the |B, 0⟩ state). This already implies that from purely energetic considerations, the presence of the plasmonic nanoparticle can open a photodissociation channel that does not exist in the bare molecule. In the following, we will show that this cavity-enabled photodissociation channel, indeed, is opened and will discuss its properties. Taking into account the ultrafast decay of the plasmonic pseudomode, we have to explicitly treat dissipation in the system in order to obtain an accurate representation of the dynamics of the coupled system. As mentioned above, a mode corresponding to a Lorentzian peak in the spectral density leads to dynamics described by a Lindblad master equation^50,51 for the system density matrix ρ,

where $LA[ρ]=AρA†−12{A†A,ρ}$ is a Lindblad dissipator, with {A, B} = AB + BA being the anticommutator. The second term in L_A[ρ] is responsible for probability disappearing from the decaying state, while the first term makes it reappear in the new state that the decay leads to. The first term is, thus, also variously called the “refilling,” “feeding,” or “quantum jump” term. Equation (2) can be rewritten as

where $Heff=H−i2κa†a$ is an effective non-Hermitian Hamiltonian. While we solve the full Lindblad master equation here, we mention that in situations where the refilling term is negligible⁵² or the state reached by it is not of interest,²⁹ the dynamics can equivalently be described by using the Schrödinger equation of a wave function evolving with the effective Hamiltonian,⁵³ leading to significant savings in computational effort as only a wave function instead of a density matrix has to be propagated. However, this simplification is not applicable when the wave packet dynamics after decay are required,³¹ as is the case for the setup discussed here.

We note that the only incoherent process we treat here is the decay of the plasmon mode with rate κ. In particular, we do not take into account interactions of the nuclear motion along the reaction coordinate with other modes of the molecule or of the host material, which could lead to dephasing and/or intramolecular vibrational energy redistribution (IVR), i.e., dissipation of nuclear kinetic energy. For the specific case of H₂, this is simply because there are no other vibrational modes that could affect the dynamics, and in the gas phase, collisions with surrounding molecules are sufficiently rare that they have no effect on the timescales considered here. However, for bigger molecules and/or in solvents or solid host materials, it is well-known that IVR can lead to fast redistribution of energy,^54,55 and the effectiveness of the effect we describe would depend on the relative speed of the plasmon-induced decay and the dissipative processes leading to a loss of nuclear kinetic energy.

We numerically discretize the nuclear coordinate R using a finite element discrete variable representation (FEDVR)^56,57 in which the domain is split into finite elements and a high-order DVR basis is used in each element. This approach allows for high flexibility in adapting both the element size and the polynomial order to the local requirements of the problem, while still leading to a very sparse representation of derivative operators and potentials as block-diagonal and diagonal matrices, respectively. It typically achieves significantly higher precision than finite difference grids for the same computational effort. Most calculations shown below use 46 equally sized elements spanning from R = 0.5 a.u. to R = 17 a.u., with polynomial order 9 inside each element, leading to 369 nuclear basis functions. Some results are obtained with “better” grids, but we have checked that the results are visually indistinguishable from those obtained with the parameters given here and, thus, do not mention those cases explicitly. In order to prevent unphysical reflections at the end of the grid when the molecule dissociates, we additionally introduce a complex absorbing potential for R > R_abs, given by $Vabs(R)=C(R−Rabs)4$⁠, where R_abs = 12 a.u. and C = 10⁻⁴ a.u. determine the strength of the absorber. In order to maintain trace-preserving dynamics, we introduce this potential through Lindblad dissipators to a fictitious “dissipated” nuclear basis function $|RD$ that is not otherwise coupled to the system. The absorbing potential is then described by additional Lindblad dissipators in the master equation,

where |R_i⟩ is the FEDVR basis function corresponding to nuclear position R_i. These additional dissipators add a complex absorbing potential to the effective Hamiltonian, while also collecting the absorbed probability in the density matrix at the “grid point” $RD$⁠. This approach allows straightforward monitoring of the electronic-plasmonic potential energy curve on which absorption (i.e., dissociation) happened.

For the numerical implementation, we rely on NumPy⁵⁸ and QuTiP.^59,60 Since the master equations become relatively large, we have implemented a custom master equation solver that runs on graphics processing units (GPUs) by exploiting the CUDA.jl^61,62 and DifferentialEquations.jl^63,64 packages for the Julia language.⁶⁵ All plots were prepared using matplotlib.^66,67

We initially set a plasmon mode frequency of ω_p = 7.6 eV with a decay rate κ = 0.476 eV and quantized field strength E_1ph = 70 mV/a₀, where a₀ is the Bohr radius.⁶⁸ These parameters correspond approximately to a hydrogen molecule at a distance of slightly less than 1 nm from an aluminum nanosphere with a radius of 20 nm embedded in a background dielectric material with refractive index n = 1.75. The corresponding polaritonic potential energy curves,² calculated using the effective Hamiltonian, are shown in Fig. 2. The dynamics are initialized assuming ultrafast vertical excitation from X to B, i.e., with an initial state that corresponds to the vibrational ground state on V_X(R) promoted to V_B(R). This initial wavepacket is shown in Fig. 2 in red. The vertical excitation energy at the equilibrium position R₀ ≈ 1.4 a.u. corresponds to ω_m(R₀) ≈ 12.75 eV, far detuned from the plasmon resonance. As the wavepacket starts oscillating and spreading on the B curve, within less than 10 fs, it reaches the region where the cavity and molecular transition are on resonance. There, the coupling to the plasmonic mode leads to population transfer into the plasmonically excited surface, which almost immediately decays to the ground state. As we have recently shown, the photons emitted from a plasmonic cavity in similar setups can give direct optical access to ultrafast nuclear motion.²⁷ However, in the current case, the plasmonic mode is nonradiative, and no photons are emitted into the far field. Nonetheless, the plasmonic decay conserves the kinetic energy accumulated by the nuclear wavepacket during its propagation on the B surface.⁶⁹ The nuclear wavepacket arriving on the ground-state surface V_X(R) (shown in green at t = 10 fs in Fig. 2) can thus keep propagating, with a significant fraction overcoming the dissociation barrier. At the same time, some fraction of the wavepacket remains trapped in the bound vibrational states of V_X(R), shown in brown for t = 27 fs in Fig. 2.

To get detailed insight into the wavepacket dynamics, we show the nuclear probability density on each of the three relevant states |B, 0⟩, |X, 1⟩, and |X, 0⟩ as a function of time in Fig. 3. This shows the nuclear wavepacket initially oscillating on the molecular excited state |B, 0⟩. The light–matter coupling then leads to efficient energy transfer to the plasmonic state |X, 1⟩ at the nuclear positions where they are close in energy, roughly between R = 3 a.u. and R = 7 a.u. The plasmonic state almost immediately decays after excitation such that the wavepacket is directly transferred to the |X, 0⟩ state. However, it is clearly visible in the third panel of Fig. 3 that a significant fraction of the wavepacket that decays to the ground state overcomes the dissociation barrier and propagates to large internuclear distances R, where it is absorbed by the absorbing potential. Since the nuclear wavepacket in the lower panels of Fig. 3 is shown in logarithmic scale, which inhibits easy visual interpretation of absolute probabilities, in the upper panels, we additionally show the integrated probabilities in the “active” nuclear region of each electronic-plasmonic state, $PA(t)$⁠, as well as the already dissociated component $PD(t)$⁠. This demonstrates clearly that the light–matter coupling through the plasmonic pseudomode induces ultrafast decay of the nuclear wavepacket from the initial excited state, with most of the probability disappearing within about 100 fs. Furthermore, it also shows that after decaying to the ground state |X, 0⟩, a significant portion of the nuclear wavepacket dissociates, reaching more than 50% for the present parameters. The remaining wavepacket ends up in bound vibrational levels on the ground-state surface.

We note that due to the highly lossy nature of aluminum, the decay rate of the pseudomode (and other plasmonic resonances) is quite large (κ = 0.476 eV), with an associated lifetime of only τ = ℏ/κ ≈ 1.38 fs, even shorter than for the more commonly used metals such as gold or silver, which have plasmon resonances in the visible spectrum. The strong-coupling regime is entered when g > κ/4 if the condition is taken that at resonance ω_p = ω_m(R), the real part of the eigenenergies of the effective Hamiltonian starts to split, and for slightly larger g if one requires that the corresponding peaks can be spectrally distinguished.⁷⁰ The maximum value of the transition dipole moment μ_XB(R) is ∼1.61 e a₀, reached at R = 3.1 a.u. For realistic parameters, with E_1ph reaching up to about 100 mV/a₀, the system we treat will thus stay within the weak coupling regime or at best barely enter strong coupling. However, since the coupling is quite strong even when smaller than the decay rate, the plasmonic pseudomode serves as an effective decay channel at internuclear distances where ω_m(R) is sufficiently close to ω_p. The ultrafast decay we see above can thus be understood, to a good approximation, as an extreme Purcell effect, with the emissive lifetime decreasing from about ≈0.5 ns for the bare molecule down to the femtosecond scale due to the pseudomode. However, it should be stressed that an important aspect of the strategy we propose is the energy selectivity of the plasmonic pseudomode, i.e., the fact that it does not “just” enhance emission but does so at a specific energy. Within a temporal picture, this allows us to enhance decay at nuclear positions where the initial wavepacket has already accumulated enough kinetic energy to subsequently dissociate. Energetically, the approach can be understood as selecting the energy difference between the initially absorbed photon and the subsequently emitted one. This difference is necessarily converted into vibrational energy (like in Raman scattering), which can then be used to dissociate the molecule in its ground state. A further important point is that this process happens on ultrafast timescales so that, e.g., intramolecular vibrational energy redistribution (which would only occur in more complex molecules with multiple nuclear degrees of freedom) does not have time to dissipate the accumulated kinetic energy between different modes.

We thus next explore the energy sensitivity of the loss-induced photodissociation process in more detail. To do so, we scan the frequency of the plasmonic pseudomode ω_p over a large interval (from 4 eV to 11 eV) while keeping all other parameters fixed. Since for the given coupling strength the system is mostly in the weak coupling regime, a useful quantity to plot is the induced effective decay rate of the (cavity-modified) excited molecular state B^*. Here, the asterisk denotes polaritonic states obtained by diagonalizing the effective Hamiltonian H_eff within the Born–Oppenheimer approximation (i.e., separately for each R), and B^* is the state that is closest in character to the bare molecular excited state B. In the effective Hamiltonian, the energies (and thus, the polaritonic PES) become complex, and the local decay rates are given by $κB*(R)=−2 ImVB*(R)$⁠. Within the (almost) weak coupling regime considered here, the modifications of the real part of the surface are much smaller, $ReVB*(R)≈VB(R)$⁠. We thus show $κB*(R)$ as a function of plasmon frequency ω_p in Fig. 4(a). The white dashed line is given by ω_m(R), i.e., the position-dependent molecular excitation energy. The plasmon-induced molecular decay rate is most strongly enhanced when the plasmon is on resonance with the molecular transition, ω_p ≈ ω_m(R), as could be expected and has been previously proposed as a tool for probing molecular dynamics by ultrafast emission.²⁷ The cavity-induced decay rate for non-zero detuning depends on both the detuning and the coupling strength. For g(R) = E_1phμ_XB(R) ≪ κ/4, it can be well approximated by a Lorentzian function, $κB*(R)≈4g(R)24δ(R)2+κ2$⁠, where δ(R) = ω_m(R) − ω_p.

We now focus on how this frequency dependence of the pseudomode-induced decay affects the photodissociation process. To that end, Fig. 4(b) shows the total dissociation probability as a function of time for the same range of plasmon frequencies. We first note that at short times, there is a step-wise behavior that stems from the initially localized character of the nuclear wavepacket, which creates a burst of dissociation every time it passes the nuclear distances where decay is efficient²⁷ and then proceeds toward dissociation. At later times, the nuclear wavepacket delocalizes due to the non-harmonic character of the involved potentials, and the step-wise behavior is washed out.

As a function of frequency, there is a dominant peak centered just below ω_p = minω_m(R), i.e., the minimum of the molecular transition frequency (indicated by a thin white dotted line). This reinforces that, as discussed above, the molecular decay rate shown in panel (a) does not by itself determine the dissociation probability. In addition to inducing ultrafast decay on the excited-state surface, the decay has to happen after the wavepacket acquires enough vibrational energy to dissociate after reaching the ground state. This consideration implies that smaller ω_p should be favored, as less of the total available energy is then spent on creating the plasmon excitation, and the remainder can be used for overcoming the dissociation barrier. Combining this with the competing requirement of being close to resonance to achieve efficient decay explains the results observed here. However, it should additionally be noted that while at the maximum times considered here, both very low and very high plasmon frequencies appear similarly efficient in inducing dissociation, and their long-time behavior is markedly different: While at high frequencies, the wavepacket decays efficiently [at least as long as the cavity frequency is not significantly above the molecular excitation frequency at the Franck–Condon point, ω_m(R_FC) ≈ 12.75 eV, higher than the frequencies considered here], it primarily does so with little vibrational energy and thus stays in the bound vibrational states on the electronic ground state curve V_X(R). In contrast, the induced decay slows down when the plasmon frequency is below the minimum of ω_m(R), but the wave packet after decay acquires considerable vibrational energy and can thus dissociate with high probability. The end result is that at high frequencies, the dissociation probability is already essentially saturated for the latest time shown in Fig. 4(b), t_f = 1000 fs, while at low frequencies, it keeps rising even at longer times. For the lowest frequency considered, ω_p = 4 eV, the dissociation probability rises to over 45% when propagation is performed until t = 10 ps, and fitting to an exponential saturation curve implies that the limiting value is $PD(t→∞)≈49%$⁠. This implies that the plasmon-induced photodissociation efficiency at lower plasmon frequencies can also be significant, but the reduced speed means that this case is more sensitive to effects not considered here (such as collisions or IVR in more complex molecules).

We remark that the fact that the plasmon-induced photodissociation does not depend too strongly on the cavity frequency implies that in situations where several plasmonic resonances are significantly coupled to the molecular transition, their overall effect would be cumulative. This means that taking into account all plasmonic modes of the nanoparticle instead of just the pseudomode we treat here would increase the decay rate even further.

We next explore the influence of the nuclear mass on the process described here. To do so, we repeat the calculations shown in Fig. 4 for the case of D₂, which has twice the nuclear mass of the H₂ molecule. This increase in mass has two main consequences: On the one hand, the nuclear wave packets become more localized in space (i.e., more “classical”), and on the other hand, nuclear motion is slowed down. As shown in Fig. 5, as a result, the dissociation occurs at slightly later times, although the overall efficiency is hardly affected and even slightly increased at long times (note that the color bar in Fig. 5 extends to slightly higher values than in Fig. 4). This increase can be understood by considering the limit of extremely high mass. In that limit, the quantum nature of the nuclear wave function would stop playing any role (the nuclear wave packet becomes delta-function-like), and motion would thus follow classical trajectories. Since photodissociation is energetically allowed starting from the Franck–Condon point, the high-mass limit would thus give 100% dissociation probability in the absence of nuclear energy dissipation (which is not included in the present simulations). We have verified this trend with an additional calculation where the nuclear mass is nine times larger than in H₂, with all other parameters as in Fig. 3. The photodissociation yields for H₂, D₂, and this “super-heavy” H₂ are ∼60%, 63%, and 68%, respectively. Of course, this simple picture breaks down if nuclear energy dissipation and redistribution are taken into account. In more complex molecules where such processes play an important role, the efficiency of the plasmon-induced photodissociation would thus depend on the detailed interplay between dissipation and the speed and “classicality” of the nuclear motion on the concrete potential energy surfaces.

We finally check how strongly the effects studied here depend on the coupling strength to the plasmon mode, which is strongly dependent on the distance to the nanoparticle. The total dissociation probability at time t_f = 1000 fs is shown as a function of plasmon frequency ω_p and quantized single-photon field strength E_1ph in Fig. 6. The considered values for E_1ph, ranging from 10 mV/a₀ to 100 mV/a₀, correspond to distances of roughly 0.5 nm–2 nm, which lies within the experimentally accessible range for nanoplasmonic antennas.^24,25 The effective mode volume $Veff=ℏωp/(2ε0E1ph2)$ for the largest couplings considered here is on the order of 20 nm³. While we stay within a single-mode description, it should be noted that for increasing distances, the pseudomode stops being the most strongly coupled mode, and the dipole mode (which is radiative and has lower frequency) becomes dominant.³⁶ Figure 6 shows that, as could be expected, increasing coupling strengths lead to increasing dissociation efficiency. It can also be appreciated that the frequency range over which dissociation is efficient increases with an increase in field confinement (i.e., larger E_1ph). However, we also find that there is an optimal value for the coupling strength, i.e., the photodissociation probability starts dropping again as the coupling strength increases above an optimal value. Consistent with results found in the photoprotection efficiency of uracil molecules,²⁹ the optimal value is close to the onset of the strong-coupling regime. While the situation in a molecule with nuclear motion is somewhat more complicated than in two-level emitters, the underlying reason for this can be understood by taking into account that plasmon-induced decay is most efficient just at the limit between weak and strong coupling,⁷¹ where the effective Hamiltonian has an exceptional point.⁷² 

To summarize, we have theoretically investigated a strategy for using highly lossy plasmonic modes to enable and enhance photodissociation in molecules for which the excited-state surface reached by photoexcitation is non-dissociative. We used the H₂ molecule next to an aluminum nanoparticle as a model system. For short distances, the extremely localized and nonradiative plasmonic pseudomode then functions as an effective “cavity” mode with extreme sub-wavelength field confinement. The basic idea of the strategy we proposed is to use a resonantly enhanced Raman-like process where part of the energy absorbed from an exciting photon is resonantly emitted into a lower-energy plasmon resonance, with the remaining energy converted to vibrational excitation on the molecular ground state. If this energy is high enough, the nuclear wavepacket can then overcome the dissociation barrier. Making use of the unavoidable high losses of plasmonic nanoantenna modes, the induced ultrafast decay rates ensure that the whole process takes place on femtosecond scales, faster than other relevant processes leading to energy dissipation.

We have studied in detail how the process depends on the properties of the plasmonic modes, in particular their frequency and the achieved field confinement, and have found that there are competing effects leading to a pronounced maximum as a function of cavity frequency: On the one hand, lower plasmon frequencies leave more of the initial energy available as vibrational energy for the wavepacket after decay, while on the other hand, resonance between the plasmon and molecular excitation enhances the overall decay rate on the plasmon-modified potential energy curve. The competition between these two effects leads to a maximum of dissociation efficiency close to the minimum excitation energy of the molecule, i.e., at the point where the ground- and excited-state surfaces are closest to each other.

The strategy we have presented adds to the growing repertoire of loss-induced phenomena in molecular polaritonics and in polaritonic chemistry.^27–31 This field has enjoyed increasing interest recently as a way to make virtue out of necessity in nanoplasmonic systems, where extreme subwavelength confinement can provide coupling strengths close to or within the single-molecule strong-coupling regime at room temperature, but losses on the femtosecond scale are unavoidable.

We thank Alicia Palacios for insightful discussions and for providing the PES of the hydrogen molecule in digitalized form. This work was funded by the European Research Council through Grant No. ERC-2016-StG-714870 and by the Spanish Ministry for Science, Innovation, and Universities–Agencia Estatal de Investigación through Grant Nos. RTI2018-099737-B-I00, PCI2018-093145 (through the QuantERA program of the European Commission), and CEX2018-000805-M (through the María de Maeztu program for Units of Excellence in R&D).

The data that support the findings of this study are available from the corresponding author upon reasonable request.