Collective vibrational strong coupling effects on molecular vibrational relaxation and energy transfer: Numerical insights via cavity molecular dynamics simulations

Tao E. Li,1, ∗ Abraham Nitzan,1, 2, † and Joseph E. Subotnik1, ‡ 1Department of Chemistry, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA 2School of Chemistry, Tel Aviv University, Tel Aviv 69978, Israel

For a small fraction of hot CO2 molecules immersed in a liquid-phase CO2 thermal bath, classical cavity molecular dynamics simulations show that forming collective vibrational strong coupling (VSC) between the C O asymmetric stretch of CO2 molecules and a cavity mode accelerates hot- molecule relaxation. The physical mechanism underlying this acceleration is the fact that polaritons, especially the lower polariton, can be transiently excited during the nonequilibrium process, which facilitates intermolecular vibrational energy transfer. The VSC effects on these rates (i) resonantly depend on the cavity mode detuning, (ii) cooperatively depend on molecular concentration or Rabi splitting, and (iii) collectively scale with the number of hot molecules, which is similar to Dicke’s superradiance. For larger cavity volumes, due to a balance between this superradiant-like behavior and a smaller light-matter coupling, the total VSC effect on relaxation rates can scale slower than 1/N, and the average VSC effect per molecule can remain meaningful for up to N ∼ 104 molecules forming VSC. Moreover, we find that the transiently excited lower polariton prefers to relax by transferring its energy to the tail of the molecular energy distribution rather than equally distributing it to all thermal molecules. Finally, we highlight the similarities of parameter dependence between the current finding with VSC catalysis observed in Fabry–Pérot microcavities.

I. INTRODUCTION In order to narrow the gap between theory and ex- periment, here we numerically investigate VSC effects on Collective vibrational strong coupling (VSC) can oc- two nonequilibrium processes — molecular vibrational cur if a macroscopic number of liquid-phase molecules energy relaxation and intermolecular vibrational energy are confined to a Fabry–Pérot microcavity and a molecu- transfer. These vibrational processes has been exten- lar vibrational mode is near resonant with a cavity mode sively studied both experimentally and theoretically out- [1, 2]. Under collective VSC, experimental reports indi- side a cavity, and the rates of which have been known cate not only a peak splitting, i.e., a Rabi splitting within to play an important role in many physical and chemi- molecular infrared (IR) spectroscopy, but also the mod- cal processes, including chemical reactions [15]. Inside a ification of chemical reaction rates [3–6] and crystalliza- cavity, a recent experiment [16] has studied the effect of tion processes [7] under thermal conditions. As pioneered VSC on intermolecular vibrational energy transfer rates first by Ebbesen [3] and co-workers, these observations by quantifying the response of hybrid light-matter states suggest that collective VSC might meaningfully modify (polaritons) after pumping the upper polariton (UP) for 12 13 individual molecular properties without external pump- a liquid mixture of W( CO)6 and W( CO)6. ing — although these intriguing experimental findings For the sake of simplicity, here our numerical study fo- cannot yet be well explained by current theory [8–13]. cuses on a pure liquid CO2 system when the C O asym- A simple example illustrating how conventional the- metric stretch forms VSC with a single optical cavity ory fails to explain the Ebbesen experiments is to con- mode (where two polarization directions are included). sider the case of N molecules forming VSC with a Rabi In such a system, instead of exciting polaritons, we will −1 splitting ΩN = 2g0√N 100 cm , where g0 denotes consider the case when a small fraction of uncorrelated the light-matter coupling∼ for individual molecules. Be- hot CO2 molecules dissipates and transfers vibrational cause g0 (= ΩN /2√N) is negligible when N becomes energy to the remaining thermal CO2 molecules at room macroscopic, one would guess that individual molecular temperature. Unlike many experiments and theoreti- properties (such as chemical reaction rates) cannot be cal studies concentrating on the polaritonic response, we meaningfully modified by a Fabry–Pérot microcavity, a will mainly focus on how individual molecules (which are theoretical prediction at odds with several experiments. arXiv:2103.06749v1 [physics.chem-ph] 11 Mar 2021 mostly composed of vibrational dark modes) relax and Recent efforts [10, 14] also suggest that, within a clas- transfer energy under VSC. In detail, we will extensively sical description of cavity photons and molecules, static study how vibrational energy relaxation and transfer de- properties of individual molecules during thermal equi- pend on cavity mode detuning, molecular concentration librium are entirely unchanged under usual VSC setups, (or Rabi splitting), and the number of hot molecules. Be- indicating a nonequilibrium (or perhaps quantum) origin cause there is no external polariton pumping, our investi- of the Ebbesen experiments. gation of how a cavity affects vibrational relaxation and energy transfer will hopefully yield insight into the VSC modifications of individual molecular properties (such ∗ [email protected] as chemical reaction rates) that are observed in experi- † [email protected] ments. In particular, by quantifying the asymptotic scal- ‡ [email protected] ing of VSC effects with molecular system size or effective 2

ps). Below, we will report how VSC affects vibrational energy relaxation and transfer using CavMD simulations.

II. RESULTS

A. VSC effects on vibrational energy relaxation and transfer

Fig. 2a plots the IR spectrum outside the cavity (black line; ε = 0) or inside the cavity (red line; ε = 2 10−4 −×1 a.u.) when the cavity mode (at ωc = 2320 cm ; the FIG. 1. Sketch of the simulation setup where a large num- dashede vertical blue line) forms VSC withe the C O ber of liquid-phase carbon dioxide molecules forms VSC with asymmetric mode (peaked at ω = 2327 cm−1) of liquid a single cavity mode. The left cartoon shows that the vi- 0 CO2. Inside the cavity, a pair of lower (LP; peaked at brational energy relaxation and transfer from hot (upper) to −1 −1 thermal (bottom) molecules inside a cavity can be accelerated 2241 cm ) and upper (UP; peaked at 2428 cm ) polari- relative to that outside a cavity due to polariton-accelerated tons form and these polaritons are separated by a Rabi −1 intermolecular vibrational energy transfer. splitting of 187 cm . The IR spectrum is calculated by evaluating the Fourier transform of the dipole autocor- relation function from equilibrium trajectories; see Ap- pendix S-III for details. cavity volumes, our study will also partly address if VSC We now consider a nonequilibrium process where effects can persist and affect the properties of individ- Nhot = 10 uncorrelated hot molecules are immersed in a ual molecules in the limit that a very large number of thermal CO bath at room temperature (where in total molecules are present in a cavity. 2 there are Nsub = 216 molecules in the simulation cell); see The theoretical approach we will take is classical cav- Appendix S-II for details. Fig. 2b plots the average time- ity molecular dynamics (CavMD) simulations [14, 17], a resolved C O bond potential energy per hot molecule newly developed numerical tool implemented by the au- outside (black line) or inside (red line) the cavity, where thors to classically propagate the coupled dynamics be- a thermal energy kBT = 300 K has been subtracted from tween realistic molecules (assumed to stay in their elec- the C O bond potential energy; note that here we use tronic ground-state) and cavity photons in the dipole kBT instead of kBT/2 since each CO2 molecule contains gauge. Since the self-dipole term is included in the light- two C O bonds. As shown in Fig. 2b, the initial po- matter Hamiltonian of CavMD simulations, this numeri- tential energy in the two C O bonds per hot molecule cal approach preserves gauge invariance and maintains 3 is roughly 2~ωc ( 6 10 K), i.e., the initial tempera- numerical stability [18]. Compared with VSC experi- ture of the hot molecules≈ × is 3 103 K. At later times, ments, this approach has reliably captured many VSC- the vibrational energy relaxation∼ × of the hot molecules in- induced phenomena, including an asymmetric Rabi split- side the cavity is accelerated compared with that outside ting [14, 19], polariton relaxation to vibrational dark the cavity. Meanwhile, as shown in Fig. 2c, the average modes on a time scale of ps and sub-ps [17, 20], and C O bond potential energy per thermal molecule inside a delay of population gain in the singly excited mani- (red line) the cavity increases faster than that outside fold of vibrational dark modes after pumping the lower (black line) the cavity. Here, "thermal molecules" refer polariton (LP), a process which stems from polariton en- to molecules that were prepared at thermal equilibrium. hanced molecular nonlinear absorption [17, 21]. Hence, During this nonequilibrium process, the total system en- CavMD simulations appears to be a promising tool to ergy is conserved: the simulation is performed under a study VSC-related phenomena. NVE (constant number, volume, and energy) ensemble; A brief introduction of CavMD is given in Appendix see simulation details in Appendix S-I. S-I and S-II; see Ref. [17] for more details regarding During the energy relaxation and transfer process, in- CavMD simulations of a liquid CO2 system and how the side the cavity, Fig. 2d plots the total (kinetic + po- −1 CO2 force field is defined. In short, as shown in Fig. 1, tential) energy of the cavity photons (ωc = 2320 cm CavMD simulates a system with Nsub CO2 molecules in and with two polarization directions) subtracted by the a periodic cell coupled to a single cavity mode (with two thermal background 2kBT . Because cavity photons con- polarization directions x and y). The effective coupling tribute half of the polaritons, Fig. 2d indicates that po- strength between each molecule and the cavity mode is laritons can be transiently excited during this nonequilib- ε. Note that, during nonequilibrium CavMD simulations, rium process. Note that, at long times, the cavity photon we have disregarded cavity loss. This simplification is energy does not decay back to zero because the relaxation valide because in Fabry–Pérot microcavities, the dominant of the hot molecules will increase the system temperature channel for polaritons to relax is through vibrational dark to above 300 K. From Figs. 2b-c, we can conclude that modes (with a lifetime of ps or sub-ps with our parameter the cavity acceleration of vibrational energy relaxation setting [17]) and cavity loss takes a longer lifetime ( 5 stems from cavity-accelerated intermolecular vibrational ∼ 3

4 ] ] c c

(a) (b) (c) ] (d)

hot molecules 0.08 c cavity photons ¯ ¯ 3 hω hω ¯ 2 hω 0.4 T[ T[ 0.06 B B T[ k k [arb. units] 2 B k

) cavity on 0.04 ω - 2 ( 1 0.2

α 1 ) ph

ω 0.02 E ( thermal molecules cavity off V(C=O) - V(C=O) - n 0 0 0.00 0.0 2000 2200 2400 2600 0 20 40 0 20 40 0 20 40 1 frequency [cm− ] time [ps] time [ps] time [ps]

FIG. 2. Rabi splitting and VSC effects on vibrational energy relaxation and transfer when Nsub = 216 and Nhot = 10. (a) Simulated IR spectrum for liquid CO2 outside (black) or inside (red) the cavity. For parameters, the cavity mode frequency −1 −4 is set to ωc = 2320 cm (denoted as the vertical blue line) and the effective coupling strength εe = 2 × 10 a.u. (inside the cavity) or zero (outside the cavity). (b,c) The corresponding average C O bond potential energy (per molecule) dynamics for the (b) hot or (c) thermal molecules outside (black) or inside (red) the cavity. (d) The corresponding photonic (kinetic + potential) energy dynamics inside the cavity, where two polarization directions of the cavity mode are taken into account. In the y-axis of Figs. b-d, a thermal energy (i.e., kB T for Figs. b,c and 2kB T for Fig. d) has been subtracted and all energies are in units of ~ωc. See Appendix S-II for other simulation details. Note that polaritons play an important role during the process of vibrational energy relaxation and transfer as evidenced from the high transient photonic energy as compared with the vibrational energy transferred to the thermal molecules.

the cavity mode detuning δ = ωc ω0. Compared with

] inside cavity − 1 the fitted decay rates outside the cavity (black squares),

− 0.100 outside cavity the rates inside the cavity (red circles) show a resonant dependence on the detuning δ: when δ 0, the maxi- 0.075 mum rate inside the cavity is roughly four≈ times the rate outside the cavity. Because the cavity mode is decoupled 0.050 from C O asymmetric stretch under a large detuning, 0.025 this resonance behavior again points to the importance of forming polaritons as far as modifying relaxation rates. fitted decay rate [ps 0.000 We have also found (not shown here) that the VSC effect 200 0 200 on relaxation rate depends only weakly on the tempera- − 1 cavity mode detuning [cm− ] ture of the hot molecules.

FIG. 3. Fitted vibrational energy relaxation rates as a func- tion of the cavity mode detuning. All parameters are the same C. Rabi splitting dependence as in Fig. 2b except that we now change the cavity mode fre- quency (ωc). Rates are obtained by fitting the signals in Fig. 2b to a simple exponential function: y = A exp(−kt). Note We next investigate how vibrational energy relaxation that the VSC effect on vibrational energy relaxation reso- rates depend on the Rabi splitting by introducing an iso- nantly depends on the cavity mode frequency. topic liquid mixture of carbon dioxide and changing the relative molecular concentration of each isotopic form. With all other parameters the same as Fig. 2 (where a energy transfer from the hot to the thermal molecules. pure CO , or 12C16O system is studied), Rabi splitting is Furthermore, compared with thermal molecules (see Fig. 2 2 tuned by replacing some 12C16O molecules by 14C18O . 2c red line), cavity photons can be excited more meaning- 2 2 Fig. 4a plots the equilibrium IR spectrum inside the cav- fully at short times. This fact emphasizes the importance 12 16 ity under an increased concentration of C O2 (c = 20% of forming polaritons and the interaction between polari- 14 18 to 100% from bottom to top). Because C O2 is rela- tons and individual molecules (which are predominately 14 18 composed of vibrational dark modes) in modifying these tively heavy, the C O asymmetric stretch (the left- 12 16 rates. est peak in Fig. 4a) is well separated from the C O −1 asymmetric stretch (peaked at ω0 = 2327 cm ), and 14 18 C O2 molecules effectively do not participate in the B. Detuning dependence formation of polaritons (LP and UP in Fig. 4a) between the cavity mode (peaked at 2320 cm−1) and the 12C 16O Consider now that case where the cavity photon fre- asymmetric stretch. The inset of Fig. 4a plots the Rabi quency is changed but all other parameters are the same splitting ΩN between the UP and LP as a function of as in Fig. 2. Fig. 3 plots the fitted vibrational en- c/c0, where c0 = 100% denotes the concentration of ergy relaxation rates (using an exponential function y = pure 12C16O . As in many experiments, a linear scaling p 2 A exp( kt) to fit Fig. 2b) of the hot molecules against between Ω and c/c is observed. − N 0 p 4 ]

(a) 1 (b) − ]

150 1 [cm −

15 N 100 Ω 0.50 0.75 1.00 0.10 c/c 0 inside cavity [arb. units] 10 LP p UP

) c=100% outside cavity ω

( c=80% 0.05 α

) 5 c=60% ω ( c=40% n 14C=18O fitted decay rate [ps 0 c=20% 0.00 2000 2200 2400 2600 0.25 0.50 0.75 1.00 1 12 16 frequency [cm− ] C O2 concentration c/c0

FIG. 4. VSC effects on vibrational energy relaxation as a function of Rabi splitting. (a) Simulated IR spectrum for a liquid 12 16 14 18 −1 12 16 mixture of C O2 and C O2 inside the cavity. The cavity mode (ωc = 2320 cm ) forms polaritons with the C O −1 14 18 asymmetric stretch mode (ω0 = 2327 cm ), while the C O asymmetric stretch mode (the leftest peak) is largely decoupled 12 16 from the cavity. Note that when the C O2 concentration increases from c = 20% to 100% (bottom to to top), the Rabi splitting (ΩN ) is also increased proportionally; see the inset. (b) The corresponding fitted vibrational relaxation rates for the 12 16 12 16 hot C O2 molecules plotted against the C O2 concentration (c/c0), where c0 = 100% denotes the concentration for a pure 12 16 12 16 C O2 system. The rates inside the cavity (red circles) show a sensitive dependence on the C O2 concentration (or Rabi splitting), while the rates outside the cavity (black squares) show a weak dependence on the concentration.

12 16 Under different concentrations of C O2, Fig. 4b molecular concentration or Rabi splitting, we next study plots the fitted vibrational energy relaxation rates when how vibrational relaxation rates depend on the number 12 16 10 hot C O2 molecules (Nhot = 10) are immersed of hot molecules (Nhot). Going beyond Fig. 2 (where in the liquid mixture. The outside-cavity results (black Nsub = 216 molecules are confined in a periodic sim- squares) show a weak dependence on the molecular con- ulation cell), here we simulate Nsub = 2160 molecules centration. By contrast, inside the cavity (red cir- while keeping all other macroscopic variables — such as cles), we observe an obvious acceleration of the relax- molecular density (1.101 g/cm3) and the Rabi splitting 12 16 ation rates when the C O2 concentration is increased — unchanged. Note that we maintain a constant Rabi from c = 20% to 60% and then a plateau region above splitting by reducing the effective light-matter coupling c = 60%. This acceleration of the relaxation rates (ε) for each molecule. Physically speaking, increasing (with a monotonic dependence on molecular concentra- the number of molecules while adjusting the coupling so tion) shows that, inside a cavity, the relaxation of a few ase to keeping the Rabi splitting constant corresponds to molecules indeed depends strongly on the total molecular increasing the effective volume of the cavity at constant number (or concentration). molecular density. Interestingly, experiments outside a cavity [22] have Fig. 5a plots relaxation rates versus the number of shown that vibrational relaxation rates in hydrogen- hot molecules (Nhot) inside the large molecular system bonded liquids (X H/X D mixture) demonstrate simi- with Nsub = 2160. Both the inside- (red circles) and lar sensitive dependence on isotope concentration as we outside-cavity (black squares) rates show a linear rela- have found inside a cavity in Fig. 4b. In Ref. [22], tionship (fitted with a linear function; see lines with re- the authors argued that such isotopic dependence can spective colors) against Nhot. Outside the cavity, the be explained by noting that, for a system with hydrogen linear scaling against Nhot is understandable because in- bonding, intermolecular vibrational energy transfer can creasing the number of hot molecules increases the tem- be facilitated by forming a delocalized intermediate state perature of the system, enhances intermolecular colli- between two neighboring molecules. In an analogous pat- sions and strengthens dipole-dipole interactions, all of tern, Fig. 4b implies that polaritons can similarly serve which can lead to an acceleration of the relaxation of hot as a "delocalized intermediate state" and facilitate in- molecules. More interestingly, the inside- and outside- termolecular vibrational energy transfer even in weakly cavity rates show different slopes against Nhot. The dif- interacting liquids. ference between these rates is plotted with green stars and represents a pure cavity effect. This pure cavity ef- fect scales roughly linearly against Nhot, demonstrating D. Superradiant-like collective relaxation that polariton-accelerated vibrational energy relaxation collectively depends on Nhot. After demonstrating that VSC leads to cooperative ef- Another example demonstrating the collective behav- fects on vibrational energy relaxation rates against the ior of vibrational relaxation is shown in Fig. 5b, where 5

0.05 Nhot = 10 ] Nsub = 2160, change Nhot 5 1 (a) 0.3 (b) Nhot = 30 − Nhot = 90 0.04 inside cavity 4

0.03 3

0.2 area [arb. units] 5 10 √N 0.02 outside cavity hot 0.1 0.01 difference (cavity effect) fitted decay rate [ps

0.00 photonic spectrum0 [arb.. units] 0 0 50 100 2200 2300 2400 2500 1 Nhot frequency [cm− ]

FIG. 5. (a) Fitted vibrational energy relaxation rates for the hot molecules inside (red circles) or outside (black squares) the cavity are plotted versus Nhot, with Nsub = 2160 CO2 molecules in the simulation cell. The difference between the inside- and outside-cavity results (green stars) is a pure VSC effect.Lines with different colors denote the respective linear fits of the rates. (b) Corresponding photonic spectrum (which effectively represents a polaritonic spectrum) during the relaxation process√ when Nhot = 10 (black), 30 (red), and 90 (orange). The inset plots the integrated area of the photonic spectrum versus Nhot. Note that the intensity of the spectrum increases monotonically as the number of hot molecules increases. See Appendices S-II and S-III for simulation details and methods to calculate the spectrum.

2 we study the frequency distribution of the transiently as O(ε ) = O(1/Nsub). Below we will examine the scaling excited photons (which effectively represents the polari- behavior for realistic CavMD simulations. tons). See Appendix S-III for details regarding the cal- a.e Standard O(1/Nsub) scaling When the simula- culation of the polariton spectrum. As shown in Fig. tion system is enlarged by increasing Nsub from 216 to 5b, the polaritonic spectrum broadens (especially for the 12960 and keeping the number of hot molecules (Nhot = LP) and red-shifts when Nhot increases [from Nhot = 10 10) fixed, Fig. 6a plots the fitted average vibrational en- (black line) to Nhot = 90 (orange line)]. In the inset ergy relaxation rates for the hot molecules inside (red cir- of Fig. 5b, we show that the total, integrated intensity cles) or outside (black squares) the cavity versus 1/Nsub. of the polaritonic spectrum increases monotonically ver- Outside the cavity, the rates decreases when Nsub in- sus √Nhot. The inset implies that when Nhot increases, creases. This observation is understandable because in- the LP grows in intensity and can interact more strongly creasing the system size while keeping Nhot = 10 effec- with the hot molecules in the system; the end result is tively decreases the temperature of the system and sup- an acceleration of the hot-molecule relaxation by what presses the relaxation rate, which is consistent with the one might call polariton-enhanced decay. This collective outside-cavity scaling in Fig. 5a. More importantly, the behavior is reminiscent of the Dicke’s superradiance phe- difference between the inside- and outside-cavity rates nomenon [23], where the spontaneous emission rates of N — which is a pure VSC effect — scales linearly with electronic two-level systems can be collectively enhanced 1/Nsub, which confirms the standard perturbative result. by a factor of N when all two-level systems interact with Also as shown in Fig. 6b, when Nsub increases, the in- the same electromagnetic field. Here, we observe a simi- tensity of the polaritonic spectrum decreases. This de- lar behavior because all molecules interact with the same crease arises because under a fixed Rabi splitting, the polaritons. light-matter coupling ε for each molecule decreases when Nsub increases, thus leading to a negligible polaritonic effect on the relaxatione rates for the hot molecules. E. Asymptotic scaling of system size b. Slower-than-O(1/Nsub) scaling Rather than studying VSC with a fixed number of hot molecules Let us now address the asymptotic behavior of VSC (Nhot = 10) and a variable number of molecules in effects for different molecular system sizes. Here, a simulation cell (Nsub), another approach is to keep we change the number of molecules in the simulation fixed Nhot/Nsub = 10/216 = 4.63%. This approach cell (Nsub), while keeping the molecular density (1.101 captures the physical reality that, as extensive proper- 3 g/cm ) and the Rabi splitting the same. As mentioned ties, both Nhot and Nsub should scale proportional to above, this change corresponds to investigating different one another as a function of system size. In Fig. 6c, effective cavity volumes. As discussed in Appendix S- we plot vibrational energy relaxation rates for the hot I, under these conditions (and especially the fixed Rabi molecules with different Nsub. Here, the outside-cavity splitting), second-order perturbative calculations suggest rate (black squares) is independent of the system size. that the VSC effects on individual molecules should scale In other words, the rate is an extensive property versus 6

4 Nsub = 216 ]

1 (a) Nhot = 10, change Nsub 0.3 (b) Nsub = 864 − Nsub = 4320 0.10 3

inside cavity 0.2 area [arb. units] 50 100 √Nsub 0.05 difference (cavity effect) 0.1

fitted decay rate [ps outside cavity

0.00 photonic spectrum0 [arb.. units] 0 0.000 0.002 0.004 2200 2300 2400 2500 1 1/Nsub frequency [cm− ]

Nsub = 216

] 6 1 (c) Nhot/Nsub = 4.63% (10/216) 0.3 (d) Nsub = 864 − 0.10 Nsub = 4320

4

0.2 area [arb. units] 50 100 √N 0.05 sub 0.1 fitted decay rate [ps

0.00 photonic spectrum0 [arb.. units] 0 0.00 0.01 0.02 2200 2300 2400 2500 0.7 1 1/Nsub frequency [cm− ]

FIG. 6. The dependence of vibrational energy relaxation rates on molecular system size (Nsub). (a) Fitted vibrational energy relaxation rates for the hot molecules inside (red circles) or outside (black squares) the cavity against 1/Nsub when Nhot = 10 is fixed. We observe a linear scaling between the difference (green starts), which is a pure VSC effect, and 1/Nsub. (b) Corresponding polaritonic spectrum for the different molecular system sizes in√ Fig. a: Nsub = 216 (black), 864 (red), and 4320 (orange). The inset plots the integrated area of the photonic spectrum versus Nsub. (c) Fitted vibrational energy relaxation 0.7 rates for the hot molecules against 1/Nsub when Nhot/Nsub = 10/216 = 4.63%. (d) The polaritonic spectrum corresponding to the different molecular system sizes in Fig. c.

the system size, suggesting that increasing Nhot and reduced light-matter coupling ε and the increased num- Nsub at the same time is a more appropriate approach ber of hot molecules (Nhot) that arises when Nsub in- for studying the system size dependence than keeping creases. On the one hand, whene ε decreases proportion- Nhot = 10 fixed (as above). If we compare the inside- ally to 1/√Nsub, as mentioned below Fig. 6a, the cavity and outside-cavity rates, the average cavity effect (green effect on vibrational energy relaxatione rates tends to ex- stars) on vibrational energy relaxation rates remains hibit an O(1/Nsub) scaling. On the other hand, according meaningful (i.e., > 10% cavity effect compared with the to Fig. 6d where we plot the corresponding polaritonic rates outside the cavity) even when Nsub reaches up to spectrum for different molecular system sizes, the tran- 4 3 10 . For example, when Nsub = 8 10 , the cavity effect siently excited LP intensifies for larger molecular systems on the relaxation rates is 0.004 ps×−1, which is 14% of (which is similar to Fig. 5b). Hence, this intensified LP −1 the bare relaxation rate (0.028 ps ) outside the cavity. tends to accelerate the vibrational relaxation when Nhot increases. Overall, these two competing effects lead to a The most interesting feature of Fig. 6c is that the slower-than-O(1/Nsub) scaling. 0.7 cavity effect scales with 1/Nsub (instead of 1/Nsub); see lines with different colors which represent linear fits of 0.7 the corresponding rates versus 1/Nsub. Here, we note that the 0.7 in the exponent should not be regarded as F. Polaritonic energy redistribution a universal quantity and might vary by changing simu- lation parameters (e.g. the ratio Nhot/Nsub). The un- Another interesting and potentially significant obser- derlying mechanism behind this nontrivial slower-than- vation is the way transiently excited polaritons redis- O(1/Nsub) scaling comes from the opposing effects of the tribute their energy among molecules following the vi- 7

(a)101 (b) inside cavity 1.8 cavity on / cavity off 100 outside cavity 1.6

1 10− Nsub = 2000 1.4 Nhot/Nsub = 4.63%

2 Density ratio 1.2 10− Density distribution 1.0 3 10− 0.0 0.5 1.0 1.5 2.0 0.00 0.25 0.50 0.75 1.00 1.25 1.50 C=O bond potential energy in thermal CO [hω¯ ] C=O bond potential energy in thermal CO2 [hω¯ 0] 2 0

FIG. 7. (a) Logarithmic-scaled density distribution of the C O bond potential energy as found in the molecules prepared in thermal equilibrium ("thermal molecules") during the hot-molecule relaxation process. We fix Nsub = 2000 and Nhot/Nsub = 4.63%. Purple bins denote the inside cavity results; cyan bins denote the outside cavity results, which is hardly observed in the figure because the inside- and outside-cavity distributions largely overlap with each other (the overlap of purple and cyan is blue). (b) The corresponding density ratio between the inside- versus outside-cavity distribution of C O bond potential energy. Note that the polaritons prefer to transfer energy to a small subset of thermal molecules located at the tail of the distribution. brational energy relaxation process. For the same con- Finally, we remark that, since this polaritonic ef- ditions as in Figs. 6c,d, Fig. 7a plots the logarithmic- fect mostly takes place in the long tail of the thermal scaled density distribution of the C O bond potential molecule C O vibrational energy distribution, it is pos- −1 energy (in unit of ~ω0, where ω0 = 2327 cm ) for the sible that polariton-accelerated vibrational energy trans- molecules prepared at thermal equilibrium (or "thermal fer may still be meaningful for a small subset of ther- molecules"). For this calculations, we set Nsub = 2000 mal molecules even when Nsub is very large. In other and run 40 NVE nonequilibrium trajectories; for each words, event though the average VSC effects per molecule 4 trajectory we calculate the energy distribution by tak- will vanish once Nsub exceeds 10 (see Fig. 6), some ing snapshots every 1 ps during the 40-ps simulation, so molecules in the tail of the distribution∼ may feel the ef- 4 that overall we count 40 40 2000 CO2 configurations fect of the polaritons when Nsub is beyond 10 (e.g., in during the whole relaxation× and× transfer process. During Fabry–Pérot microcavities). Future work will∼ investigate this time, both the outside (cyan bins) and inside (purple this possibility. bins) cavity results demonstrate an exponential distribu- tion, which implies that the C O bond potential energy of thermal molecules roughly obey a Maxwell-Boltzmann distribution; recall that the y-axis is on a logarithmic III. CONCLUSION scale. Very interestingly, however, the tail of the dis- tributions of C O vibrational energy differ strongly in- We have studied the effect of VSC on vibrational en- side versus outside the cavity. This fact is more clearly ergy relaxation and transfer for a small fraction of hot shown in Fig. 7b which plots the ratio of the probability molecules immersed in a thermal bath of CO2 at room density of thermal-molecule C O bond potential energy temperature. Several important observations have been inside versus outside the cavity (the bins in Fig. 7a). Be- made: (i) During this nonequilibrium process with no cause both simulations start from exactly the same initial external pumping, polaritons, especially the LP, can be conditions and with ε switched on or off (see Appendix transiently excited and facilitate intermolecular vibra- S-II for details), this difference in the tail distribution tional energy transfer, which leads to an acceleration of is a pure polaritonice effect, i.e., the transiently excited vibrational energy relaxation of the hot molecules. (ii) polaritons are more likely to create vibrationally higher This acceleration resonantly depends on the cavity mode excited molecules rather than equally distributing energy detuning and can be enhanced by increasing Rabi split- to all thermal molecules. ting (or molecular concentration). (iii) The vibrational energy relaxation acceleration is superradiant-like and A possible explanation for the large difference in the collectively scales with the number of hot molecules. (iv) tail could come from the perspective of spectral overlap For large system sizes (or large effective cavity volumes), between polaritons and individual molecules (which are when the fraction between the number of hot and ther- mostly composed of vibrational dark modes). Due to mal molecules remains the same, the VSC effect on the anharmonicity, the molecules at the tail of energy distri- relaxation rates scales slower than 1/Nsub due to a com- bution have smaller vibrational frequencies, leading to a petition between the reduced light-matter coupling (ε) larger spectral overlap with the LP. Therefore, the tran- and an enhanced superradiant-like behavior of the hot siently excited LP would interact more strongly with the molecules. (v) Although our simulations suggest that thee molecules at the tail and transfer more energy to these effect of VSC on the average relaxation rate becomes neg- 4 molecules than molecules with small vibrational energy. ligible when Nsub exceeds 10 , polaritons are always ∼ 8 transiently and meaningfully excited, and the energy in- Appendix S-I METHOD fused into the polaritons transfers more strongly to the tail of the thermal-molecule energy distribution. Alto- CavMD propagates the following equations of motion gether, this work suggests that collective VSC effects in for the coupled photon-nuclei system: a cavity can significantly affect vibrational relaxation en- ergy relaxation and transfer. ¨ (0) cav MnjRnj = Fnj + Fnj (S1a)

Nsub ¨ 2 Finally, let us make a few remarks regarding the con- m q≈ = m ω q≈ ε d (S1b) k,λ k,λ − k,λ k,λ k,λ − k,λ ng,λ nection of this work to VSC catalytic effects on ground- n=1 state chemical reactions observed in Fabry–Pérot micro- X cavities. The rates of vibrational energy relaxation and Here, subscript nj denotes the j-the nucleus in molecular transfer can significantly modify ground-state chemical (0) cav n, Mnj, Rnj, Fnj , and Fnj denote the mass, position, reaction rates outside the cavity. For example, Kramers’ nuclear force, and cavity force for nucleus nj; see Refs. theory [24, 25] suggests that ground-state reaction rates [14, 17] for the exact forms of the forces. Subscripts k, λ can depend proportionally or inversely on the energy re- denote the cavity photon mode with wave vector k and laxation rate. Therefore, the observation of VSC effects polarization direction λ = x, y for a z-oriented cavity on vibrational energy relaxation and transfer might imply (see Fig. 1 for the simulation setup), mk,λ, q≈k,λ, ωk,λ de- the modification of ground-state chemical reaction rates. note the auxiliary mass, position, and frequency of cavity That being said, VSC catalysis is highly nontrivial as photon mode k, λ. The cavity photon mode k, λ interacts experiments suggest (at least) the following four criteria with the dipole moments of molecules (dng,λ) with an ef- [3, 4, 6]. (i) The chemical reaction rates are modified fective coupling strength εk,λ. in thermal conditions and without external polaritonic pumping. The cavity modification of chemical reaction e rates (ii) resonantly depends on the cavity mode detun- A System size dependence ing and (iii) collectively depends on molecular concen- tration (or Rabi splitting). (iv) The cavity modification One important feature of CavMD is the use of peri- can be observed in Fabry–Pérot microcavities (where the odic boundary conditions. In detail, cavity photons in- effective cavity volume is λ3 and λ takes units of mi- teract with Ncell identical simulation cells, each of which crometers), meaning that the∼ number of molecules form- contains Nsub molecules, so the total molecular number ing VSC can reach 109 1012. For VSC effects on vi- ∼ becomes N = NsubNcell. The replica of Ncell simulation brational energy relaxation and transfer, we have also cells has been reflected in the definition of observed the satisfaction of criterion (i)-(iii). However, when criterion (iv) (the number limit) is considered, al- εk,λ = Ncellεk,λ (S2) though we have observed a negligible average VSC effect 4 p per molecule once Nsub exceeds 10 , our simulation in Eq. (S1), where ε is the true coupling strength ∼ e k,λ suggests a larger polaritonic effect for molecules at the between a single molecule and the cavity mode k, λ. tail of the energy distribution. Because chemical reac- When studying how VSC effects can depend on the tions also occur at this same tail, such a similarity indi- molecular system size (or the molecular number), we cates that VSC effects on vibrational energy relaxation can take Ncell = 1 and study the molecular response and transfer could possibly play a significant role in VSC for different choices of Nsub while keeping the molecu- catalysis — a hypothetical premise that deserves further lar density and Rabi splitting the same. Here, the Rabi study. splitting is unchanged if we modify εk,λ according to εk,λ 1/Nsub. The corresponding CavMD results will reflect∝ the VSC response for a liquid system in cavities p e withe the same polaritonic frequencies but with different effective volumes [26]. When this system size dependence is studied, accord- ing to a second order perturbation calculation, the VSC 2 effect on individual molecules should scale as O(εk,λ) = O(1/Nsub). This O(1/Nsub) scaling will quickly remove any VSC effects on individual molecules once Ne sub is IV. ACKNOWLEDGMENTS large enough; as has been noted earlier [27], this approach cannot explain any collective cavity effect.

This material is based upon work supported by the U.S. National Science Foundation under Grant No. Appendix S-II SIMULATION DETAILS CHE1953701 (A.N.); and US Department of Energy, Of- fice of Science, Basic Energy Sciences, Chemical Sciences, We simulate a model yet realistic molecular system Geosciences, and Biosciences Division (J.E.S.). under VSC: the C O asymmetric stretch mode (peaked 9

−1 at 2327 cm ) of a large ensemble of liquid-phase CO2 tories. Inside the cavity, we start from exactly the same molecules forms collective VSC with a near resonant cav- nonequilibrium initial configurations as the outside cav- ity mode. The detailed procedure to perform CavMD ity case but reset ε = 2 10−4 a.u. to switch on the × simulations for such a liquid CO2 system is given in light-matter coupling and run NVE trajectories. Note Ref. [17] and all input files to generate results in this that the use of thee NVE ensemble implies that we have manuscript are available at Github [28]. Therefore, be- neglected any cavity loss, which simplifies the interpre- low we only briefly outline the simulation details. tation of results. Appendix S-III ON CALCULATING As shown in Fig. 1, the simulation setup consists POLARITONIC SPECTRUM of Nsub = 216 CO2 molecules in a cubic simulation cell (with cell length 24.292 Å, which corresponds to a Because polaritons are composed of a molecular bright molecular density 1.101 g/cm3) confined within a pair of mode and cavity photons, a polaritonic spectrum can be metallic mirrors along the z-direction. An anharmonic obtained from either the molecular or the photonic side. force field [17] is used to propagate the dynamics of CO . 2 From the molecular side, a polaritonic spectrum can be During the simulation, only a single cavity photon mode obtained by calculating the molecular infrared (IR) ab- (with two polarization directions: x and y) is considered sorption spectrum, which can be evaluated by Fourier and the effective coupling strength is set as ε = 2 10−4 transforming the dipole auto-correlation function [25, 29– a.u.. × 31]: We are interested in the cavity modificatione of both the vibrational energy relaxation of the hot CO molecules πβω2 1 +∞ 2 n(ω)α(ω) = dt e−iωt and the subsequent intermolecular vibrational energy 2 V c 2π 0 Z−∞ transfer to the thermal CO2 molecules. Before simulating (S3) this nonequilibrium dynamics, outside a cavity (ε = 0), (µ (0) e )(µ (t) e ) × S · i S · i we first run 150 ps NVT (constant molecular number, *i=x,y + X volume, and temperature) simulations at 300 K toe equili- brate the system and then run 40 consecutive NVE (con- Here, α(ω) denotes the absorption coefficient, n(ω) de- stant molecular number, volume, and energy) trajectories notes the refractive index, β = kBT , V is the volume of with duration 20 ps. Starting from the initial configura- the system (i.e., the simulation cell), ei denotes the unit tions (with both position and velocity information) of the vector along direction i = x, y, and µS(t) denotes the above 40 equilibrium NVE trajectories outside a cavity total dipole moment of the molecules at time t. Fig. 2a (ε = 0), we prepare a nonequilibrium initial condition and Fig. 3a are calculated by evaluating Eq. (S3) from from equilibrium configurations by resampling the initial equilibrium NVE trajectories. velocitiese of Nhot = 10 arbitrary CO2 molecules (in total Similar as Eq. (S3), in order to obtain the polaritonic Nsub = 216 CO2 molecules are simulated). The initial spectrum, we can also define a photonic coordinate auto- conditions of the hot molecules are reset so that their ki- correlation function: netic energy in each degree of freedom obeys a uniform +∞ distribution in an interval 6000 250 K. Such a random 2 −iωt ( ) ( ) ≈ (0)≈ ( ) resampling of velocities has been± chosen to mimic the n ω αk ω ω dt e qk,λ qk,λ t ∝ −∞ * + preparation of uncorrelated hot molecules in a thermal Z λ=x,y X (S4) bath at room temperature, where the effective temper- where ( ) denotes the absorption coefficient for cavity ature of these hot molecules would be 3000 K (since αk ω photon mode . Fig. 5b and Figs. 6b,d are calculated the initial positions of the hot molecules∼ are not modified k by evaluating Eq. (S4) from nonequilibrium NVE trajec- and still obey a thermal distribution). tories during the whole simulation period (40 ps). Note Starting from each of the 40 nonequilibrium initial con- that when nonequilibrium trajectories are used to cal- ditions, we then run nonequilibrium NVE simulations for culate Eq. (S4), the resulting spectrum is a transient 40 ps and calculate physical properties outside the cavity spectrum which reflects the average dynamic behavior of (ε = 0) by averaging over the 40 nonequilibrium trajec- photons during the nonequilibrium trajectories.

e

[1] J. George, A. Shalabney, J. A. Hutchison, C. Genet, and C. Genet, J. A. Hutchison, and T. W. Ebbesen, Ground- T. W. Ebbesen, Liquid-Phase Vibrational Strong Cou- State Chemical Reactivity under Vibrational Coupling to pling, J. Phys. Chem. Lett. 6, 1027 (2015). the Vacuum Electromagnetic Field, Angew. Chemie Int. [2] J. George, T. Chervy, A. Shalabney, E. Devaux, H. Hiura, Ed. 55, 11462 (2016). C. Genet, and T. W. Ebbesen, Multiple Rabi Splittings [4] A. Thomas, A. Jayachandran, L. Lethuillier-Karl, R. M. under Ultrastrong Vibrational Coupling, Phys. Rev. Lett. Vergauwe, K. Nagarajan, E. Devaux, C. Genet, J. Moran, 117, 153601 (2016). and T. W. Ebbesen, Ground state chemistry under vibra- [3] A. Thomas, J. George, A. Shalabney, M. Dryzhakov, tional strong coupling: dependence of thermodynamic S. J. Varma, J. Moran, T. Chervy, X. Zhong, E. Devaux, parameters on the Rabi splitting energy, Nanophoton. 10

10.1515/nanoph-2019-0340 (2019). 094124 (2021), arXiv:2011.03192. [5] A. Thomas, L. Lethuillier-Karl, K. Nagarajan, R. M. A. [18] C. Schäfer, M. Ruggenthaler, V. Rokaj, and A. Ru- Vergauwe, J. George, T. Chervy, A. Shalabney, E. De- bio, Relevance of the Quadratic Diamagnetic and Self- vaux, C. Genet, J. Moran, and T. W. Ebbesen, Tilting Polarization Terms in Cavity Quantum Electrodynamics, a ground-state reactivity landscape by vibrational strong ACS Photonics 7, 975 (2020). coupling, Science 363, 615 (2019). [19] R. M. A. Vergauwe, A. Thomas, K. Nagarajan, A. Sha- [6] J. Lather, P. Bhatt, A. Thomas, T. W. Ebbesen, and labney, J. George, T. Chervy, M. Seidel, E. Devaux, J. George, Cavity Catalysis by Cooperative Vibrational V. Torbeev, and T. W. Ebbesen, Modification of En- Strong Coupling of Reactant and Solvent Molecules, zyme Activity by Vibrational Strong Coupling of Water, Angew. Chemie Int. Ed. 58, 10635 (2019). Angew. Chemie Int. Ed. 58, 15324 (2019). [7] K. Hirai, H. Ishikawa, J. HUTCHISON, and H. Uji-i, [20] B. Xiang, R. F. Ribeiro, A. D. Dunkelberger, J. Wang, Selective Crystallization via Vibrational Strong Coupling Y. Li, B. S. Simpkins, J. C. Owrutsky, J. Yuen-Zhou, 10.26434/CHEMRXIV.13191617.V1 (2020). and W. Xiong, Two-dimensional infrared spectroscopy of [8] J. Galego, C. Climent, F. J. Garcia-Vidal, and J. Feist, vibrational polaritons, Proc. Natl. Acad. Sci. 115, 4845 Cavity Casimir-Polder Forces and Their Effects in (2018). Ground-State Chemical Reactivity, Phys. Rev. X 9, [21] B. Xiang, R. F. Ribeiro, L. Chen, J. Wang, M. Du, 021057 (2019). J. Yuen-Zhou, and W. Xiong, State-Selective Polariton [9] J. A. Campos-Gonzalez-Angulo, R. F. Ribeiro, and to Dark State Relaxation Dynamics, J. Phys. Chem. A J. Yuen-Zhou, Resonant catalysis of thermally acti- 123, 5918 (2019). vated chemical reactions with vibrational polaritons, Nat. [22] D. J. Shaw, M. R. Panman, and S. Woutersen, Evi- Commun. 10, 4685 (2019). dence for Cooperative Vibrational Relaxation of the NH- [10] T. E. Li, A. Nitzan, and J. E. Subotnik, On the , OH-, and OD-Stretching Modes in Hydrogen-Bonded origin of ground-state vacuum-field catalysis: Equilib- Liquids Using Infrared Pump-Probe Spectroscopy, Phys. rium consideration, J. Chem. Phys. 152, 234107 (2020), Rev. Lett. 103, 227401 (2009). arXiv:2002.09977. [23] R. H. Dicke, Coherence in Spontaneous Radiation Pro- [11] J. A. Campos-Gonzalez-Angulo and J. Yuen-Zhou, Po- cesses, Phys. Rev. 93, 99 (1954). laritonic normal modes in transition state theory, J. [24] H. Kramers, Brownian motion in a field of force and Chem. Phys. 152, 161101 (2020). the diffusion model of chemical reactions, Physica 7, 284 [12] D. Sidler, C. Schäfer, M. Ruggenthaler, and A. Rubio, (1940). Polaritonic Chemistry: Collective Strong Coupling Im- [25] A. Nitzan, Chemical Dynamics in Condensed Phases: plies Strong Local Modification of Chemical Properties, Relaxation, Transfer and Reactions in Condensed Molec- J. Phys. Chem. Lett. 12, 508 (2021), arXiv:2011.03284. ular Systems (Oxford University Press, New York, 2006). [13] X. Li, A. Mandal, and P. Huo, Cavity frequency- [26] Note that periodic boundary conditions are always ap- dependent theory for vibrational polariton chemistry, plied to exclude the edge effect of the simulation cell. Nat. Commun. 12, 1315 (2021). [27] P. Pilar, D. De Bernardis, and P. Rabl, Thermodynamics [14] T. E. Li, J. E. Subotnik, and A. Nitzan, Cavity molecular of ultrastrongly coupled light-matter systems, Quantum dynamics simulations of liquid water under vibrational 4, 335 (2020), arXiv:2003.11556v5. ultrastrong coupling, Proc. Natl. Acad. Sci. 117, 18324 [28] T. E. Li, Cavity Molecular Dynamics Simulations Tool (2020), arXiv:2004.04888. Sets, https://github.com/TaoELi/cavity-md-ipi (2020). [15] M. Gruebele and P. G. Wolynes, Vibrational Energy Flow [29] D. A. McQuarrie, Statistical Mechanics (Harper-Collins and Chemical Reactions, Acc. Chem. Res. 37, 261 (2004). Publish- ers, New York, 1976). [16] B. Xiang, R. F. Ribeiro, M. Du, L. Chen, Z. Yang, [30] M.-P. Gaigeot and M. Sprik, Ab Initio Molecular Dynam- J. Wang, J. Yuen-Zhou, and W. Xiong, Intermolecular ics Computation of the Infrared Spectrum of Aqueous vibrational energy transfer enabled by microcavity strong Uracil, J. Phys. Chem. B 107, 10344 (2003). light–matter coupling, Science (80-. ). 368, 665 (2020). [31] S. Habershon, G. S. Fanourgakis, and D. E. Manolopou- [17] T. E. Li, A. Nitzan, and J. E. Subotnik, Cavity molecular los, Comparison of path integral molecular dynamics dynamics simulations of vibrational polariton-enhanced methods for the infrared absorption spectrum of liquid molecular nonlinear absorption, J. Chem. Phys. 154, water, J. Chem. Phys. 129, 074501 (2008).