Imaging CF3I conical intersection and photodissociation dynamics with ultrafast electron diffraction

See allHide authors and affiliations

Science  06 Jul 2018:
Vol. 361, Issue 6397, pp. 64-67
DOI: 10.1126/science.aat0049

Motion picture of a conical intersection

In most chemical reactions, electrons move earlier and faster than nuclei. It is therefore common to model reactions by using potential energy surfaces that depict nuclear motion in a particular electronic state. However, in certain cases, two such surfaces connect in a conical intersection that mingles ultrafast electronic and nuclear rearrangements. Yang et al. used electron diffraction to obtain time-resolved images of CF3I molecules traversing a conical intersection in the course of photolytic cleavage of the C–I bond (see the Perspective by Fielding).

Science, this issue p. 64; see also p. 30


Conical intersections play a critical role in excited-state dynamics of polyatomic molecules because they govern the reaction pathways of many nonadiabatic processes. However, ultrafast probes have lacked sufficient spatial resolution to image wave-packet trajectories through these intersections directly. Here, we present the simultaneous experimental characterization of one-photon and two-photon excitation channels in isolated CF3I molecules using ultrafast gas-phase electron diffraction. In the two-photon channel, we have mapped out the real-space trajectories of a coherent nuclear wave packet, which bifurcates onto two potential energy surfaces when passing through a conical intersection. In the one-photon channel, we have resolved excitation of both the umbrella and the breathing vibrational modes in the CF3 fragment in multiple nuclear dimensions. These findings benchmark and validate ab initio nonadiabatic dynamics calculations.

Light-induced molecular dynamics usually cannot be described within the framework of the Born-Oppenheimer approximation. The picture of nuclear motion on a single adiabatic potential energy surface (PES), determined by treating the fast-moving electrons separately from the slower nuclei, breaks down wherever two or more adiabatic PESs come close in energy (1, 2). At the crossing point of PESs, the degeneracy is lifted along at least two internal degrees of freedom, and the resultant conical intersection guides efficient radiationless transitions between electronic states at specific nuclear configurations (3). Examples of important nonadiabatic reactions include photosynthesis (4), retinal isomerization in vision (5), ultraviolet-induced DNA damage (6), and formation of vitamin D (7).

Several experimental methods have been developed for studying nonadiabatic dynamics through conical intersections (813). Among these, time-averaged photofragment imaging can identify distinct spectral features of nonadiabatic coupling (8, 9) but does not allow the observation of dynamics in real time. Time-resolved laser spectroscopy is the most widely used real-time method for following electronic dynamics, but nuclear dynamics can only be inferred on the basis of an indirect comparison with simulated transient spectroscopic features (1114). In addition, comparison with theoretical predictions requires explicit modeling of the probing process, which can be more complex than the nonadiabatic dynamics in question. Recent developments in both x-ray (15, 16) and electron-based (17, 18) time-resolved diffraction techniques open an opportunity for direct imaging of conformational changes during chemical reactions—molecular movies with atomic resolution in space and time. Despite the great importance of nonadiabatic dynamics through conical intersections, spatiotemporal resolution has not been sufficient to image a coherent nuclear wave packet traversing a conical intersection with time-resolved diffraction techniques.

The nonadiabatic transitions of molecules between different PESs are inherently quantum mechanical. A wide variety of computational methods can be used to simulate dynamics through conical intersections. For small systems, nonadiabatic dynamics can be treated with exact full quantum dynamics (19) and the highly accurate multiconfigurational time-dependent Hartree approximation (20). For larger systems, semiclassically motivated approaches such as Tully’s surface hopping (21), Meyer-Miller formalism (22), or ab initio multiple spawning (AIMS) (23) are routinely used. Although simulations can provide rich details of the dynamics through conical intersections, nontrivial approximations at many different stages of the calculations are required, even for relatively small systems. Therefore, confirmation with experimental measurements is crucial.

Here, we report the direct imaging of both conical intersection dynamics and photodissociation dynamics of gas-phase CF3I molecules with atomic resolution by use of ultrafast gas-phase electron diffraction (UGED). A 264.5-nm pump laser pulse initiates two photoexcitation channels: a one-photon transition to the dissociative A band and a two-photon transition to the [5pπ3, 2Π1/2](7s) (24) Rydberg state (referred to as 7s below) (25), as illustrated in Fig. 1. The adiabatic dissociation dynamics through A-band excitation of CF3I and its analog, CH3I, have been studied extensively (2629). We created a multidimensional movie of the structural changes in the CF3 fragment immediately after iodide dissociation, with a precision of ±0.01 Å in bond length and ±1° in bond angle. Various groups have studied the two-photon transition into the 7s channel by using pump-probe photoelectron and photoion spectroscopy (25, 3032). These studies identified the decay time scale and anisotropy of fragment ions, but the reaction pathway remained elusive. Specifically, it was only a speculation that a nearby ion-pair state might be involved in the reaction dynamics (31). We have mapped out the nuclear wave-packet trajectory in real space, which directly shows wave-packet bifurcation though a conical intersection. Through cross-verification with AIMS simulations, we have clarified that the σσ* state correlates asymptotically to a CF3+–I ion-pair state (referred to as IP) at large C–I separation, and that this state plays a key role in this channel. The reaction pathway is predominantly determined by the nonadiabatic coupling between IP and multiple states: the 7s and [5pπ3, 2Π1/2](6s) Rydberg states (referred to as 6s below) and valence states.

Fig. 1 Two-channel excitation in CF3I.

PES along the C–I bond length coordinate, with major states labeled and reaction pathways marked by arrows. Color coding indicates different electronic states: yellow, 7s; red, ion pair (IP); blue, 6s; green, valence open-shell states. C3v symmetry is retained in this plot. The two relevant states in the A band are 3Q0+ and 1Q1 (using Mulliken notation). The 6s, 7s, and 1Q1 states are of E symmetry, as indicated by two closely spaced parallel surfaces.

The UGED experimental setup is shown in Fig. 2A, which is described in detail in (33, 34) and the supplementary materials. For diffraction pattern analysis, we used a two-dimensional (2D) Fourier transform followed by Abel inversion to convert data from momentum space to real space. This procedure returns a pair-distribution-function (PDF) that reports all the interatomic distances, as explained in Fig. 2, B to E.

Fig. 2 Real-space analysis of diffraction patterns.

(A) Schematic drawing of the experiment. (B) A model for a CF3I molecule oriented along the laser polarization axis. (C) Simulated PDF for molecular ensemble with a cos2θ angular distribution. (D) A model for a CF3I molecule oriented perpendicular to the laser polarization. (E) Simulated PDF for molecular ensemble with a sin4θ angular distribution. In (B) to (E), the laser polarization is indicated by the double-headed arrow. Gray, light green, and purple represent carbon, fluorine, and iodine, respectively.

The one-photon channel preferentially excites molecules with the C–I axis aligned along the laser polarization. This results in a cos2θ angular distribution of excited-state molecules, where θ is the angle between the C–I bond and the laser polarization (Fig. 2, B and C). In this case, C–I and F–I pairs mostly appear in the parallel direction (PDF||), and the C–F and F–F pairs preferentially appear in the perpendicular direction (PDF). The two-photon channel corresponds to a perpendicular excitation (sin4θ distribution) (Fig. 2, D and E). In this case, C–I and F–I pairs preferentially appear in PDF, whereas C–F and F–F slightly favor PDF||. This analysis simultaneously yields information about atom pair distances and their corresponding angular distribution, which is critical for assigning the reaction channels and acquiring multidimensional structural reconstructions of the target molecule during the reaction.

We first concentrated on the experimental evidence for nonadiabatic dynamics in the two-photon channel. The experimental ΔPDF as a function of pump-probe time delay is shown in Fig. 3A, with blue indicating loss and red indicating gain of atom pair distances as compared with unexcited molecules. This signal contains structural information from both two-photon (C–I and F–I pairs) and one-photon (C–F and F–F pairs) channels. They can be roughly separated by time scales: The one-photon channel dominates the signal at time delay (Δt) < 200 fs, and the two-photon channel dominates the signal at Δt > 200 fs, at which point the only contribution from the one-photon channel, once dissociation is complete, is a smoothly decaying signal due to rotational dephasing (supplementary materials). The evolution of ΔPDF is plotted in Fig. 3B at three specific pair distances: the initial C–I distance (2.14 Å), a position between the initial C–I and F–I distances (2.52 Å), and the initial F–I distance (2.90 Å). After a time delay of Δt = 100 fs, the signals corresponding to 2.14 and 2.52 Å oscillate out of phase. This result clearly indicates that the C–I bond is vibrationally excited, and the ~200-fs period matches well with the documented C–I stretching mode on the 7s surface (35). The 2.9-Å signal shows oscillatory decay up to Δt = 400 fs, with a surprisingly strong recurrence at Δt = 500 fs. This recurrence time scale cannot be explained by any previously reported vibrational mode on the 7s surface.

Fig. 3 Nuclear wave-packet conical intersection dynamics in the two-photon channel.

(A) Experimental ΔPDF, smoothed by an 80-fs Gaussian kernel. The dashed line at 200 fs shows a rough separation between contributions from the one-photon and the two-photon channels. (B) Experimental time evolution of ΔPDF at 2.14, 2.52, and 2.90 Å, error bars corresponding to 1 SD of a bootstrapped dataset (supplementary materials). A comb illustrates the first two periods of the C–I stretching vibration. (C) Simulated ΔPDF of the two-photon channel. (D) Simulated time evolution of ΔPDF at 2.14, 2.52, and 2.90 Å. Curve width represents 1 SD of a bootstrapped simulation dataset. (E) Two-photon PDF generated by removing a common decaying signal from ΔPDF. Black dots indicate identified ridges, and blue arrows indicate trajectories generated by connecting nearby ridges. The axes are the same as (F) for easy comparison. (F) Simulated nuclear wave packet along the C–I coordinate. Blue dots indicate identified ridges. (G) Simulated nuclear wave packet along the C–I coordinate as in (F), with color-coding to reflect diabatic state character, as shown in the legend. The dashed box in (F) and (G) matches the region captured in (E). The simulation results are generated by averaging over more than 2000 spawned trajectories from 50 initial conditions. In (B) and (D), the 2.90 Å curve is shifted by –1 for visibility.

The real-space reaction trajectory is encoded in ΔPDF and can be extracted by using a ridge-detection algorithm. First, we extracted the two-photon PDF by removing a common decaying signal from ΔPDF. Second, we used a ridge-detection algorithm to locate 1D local maxima, or ridges. Last, we generated the reaction trajectory by connecting nearby ridges (supplementary materials). The excited-state PDF is shown in Fig. 3E together with identified ridges (Fig. 3E, black dots) and trajectory (Fig. 3E, blue arrows). In this trajectory map, at least two wave-packet bifurcation events can be identified: One occurs at ~2.7 Å/300 fs, and the other at ~2.4 Å/420 fs. In addition, a two-branch crossover event can be seen at ~3.3 Å/400 fs. These features serve as strong evidence for the involvement of multiple electronic states and nonadiabatic coupling through conical intersections.

We support our experimental results by using AIMS simulations. The nuclear wave-packet density is shown in Fig. 3G projected along the C–I distance and color-coded according to diabatic state character. The color mixing reflects the composition of the population on these diabatic states. For example, orange indicates that 7s and IP states are dominantly populated, and magenta indicates a dominant population in 6s and IP. Upon excitation, the wave packet takes ~100 fs to reach the 7s-IP conical intersection seam, where electronic transitions cause the wave packet to bifurcate into two branches. The branch remaining on the 7s surface has a strong C–I stretching character; the center of the wave packet oscillates with a period of about 200 fs, with relatively little dispersion. For the branch transferred to IP, a large fraction of the wave-packet amplitude is further transferred to the 6s surface through the IP-6s conical intersection seam and returns to the Franck-Condon region at ~500 fs. The vibrational wave packet on 7s reaches the 7s-IP conical intersection seam again at ~280 and 480 fs, giving rise to the second and the third population transfer events to the 6s surface through IP. The third outgoing wave packet on IP transiently overlaps with the returning wave packet on 6s, causing a strong recurrence of population at 500 fs at ~3Å.

The simulated ΔPDF of the two-photon channel is shown in Fig. 3C, and its evolution at 2.14, 2.52, and 2.90 Å is shown in Fig. 3D. Comparison between Fig. 3, A and C, and Fig. 3, B and D, shows that the ~200-fs vibration and the strong recurrence at 2.90 Å/500 fs match very well. The vibration is more pronounced in the experimental data, possibly because of various approximations adopted in the simulation. The result of the ridge-detection algorithm on the simulated nuclear wave packet is shown in Fig. 3F. The dashed box in Fig. 3F shows a very similar trajectory as that in Fig. 3E: Two wave-packet bifurcation events can be found at 2.6 Å/270 fs and 2.5 Å/460 fs, and a two-branch crossover event is seen at ~3.6 Å/420 fs. All three events are within a 0.3-Å/40-fs spatiotemporal displacement in comparison with experimental data. Comparison between Fig. 3E and Fig. 3F shows that the real-space reaction trajectory, including nonadiabatic events through conical intersections, is directly captured in the experimental data.

We next concentrated on photodissociation after single-photon excitation to the 3Q0 state and explored the ensuing structural changes in multiple nuclear coordinates. After the breaking of the C–I bond, the most obvious diffraction signature is the loss of C–I and F–I atom pairs. This is reflected by the two strong bleaching bands in ΔPDF|| around 2.14 and 2.9 Å in Fig. 4A. The time dependence of these two bleaching signals is plotted in Fig. 4B. The C–I bleaching signal starts ~30 fs earlier than the F–I signal on account of comparatively fast recoil of the lighter carbon relative to the heavier iodine. Both the iodine and the three fluorine atoms move on a slower time scale, leading to an observable delay between the loss of C–I pair and F–I pair. The AIMS simulation shows a 16-fs separation of the two bleaching signals, which is in reasonable agreement with the experiment. The simulated ΔPDF|| of the one-photon channel is given in Fig. 4C. Three main features in Fig. 4A are reproduced in Fig. 4C: The two bleaching bands correspond to the loss of C–I and F–I atom pairs, and a positive feature ~1.3 Å after 300 fs is caused by the rotational dephasing of CF3 radicals.

Fig. 4 Multidimensional structural evolution during photodissociation in the one-photon excitation channel.

(A) Experimental ΔPDF||, smoothed by an 80-fs Gaussian kernel. (B) C–I and F–I bleaching signal in ΔPDF||. Different onset time reflects the transient recoil of the carbon atom, as shown in the inset. Error bars correspond to 1 SD of a bootstrapped dataset. (C) Simulated ΔPDF|| of the one-photon channel. (D) Structural evolution of CF3 group from experiment (circles with error bars representing the standard error of the fit), plotted together with temporal-blurred structural evolution from AIMS simulation (solid curves). RC–F and ∠FCF are color-coded blue and red, respectively. In (D), the vertical axis ranges for length and angle are adjusted to match each other with a 1.33-Å (ground state C–F distance) radius.

More details about the very early motion after photodissociation can be extracted from the C–F and F–F pairs encoded in ΔPDF for Δt < 200 fs. We performed a χ2fit in ΔPDF in order to extract the change of molecular structure (supplementary materials) and the fitted C–F bond length change (ΔRC–F) and F–C–F bond angle change (Δ∠FCF) are given in Fig. 4D. The dynamics assembled from the data are shown in movie S1. Upon dissociation, the ∠FCF immediately opens up by ~4°, followed by RC–F elongating by ~0.03 Å with a ~50-fs delay.

We performed AIMS simulations on the 3Q0 state. Upon dissociation, both the umbrella and the breathing vibrational modes are strongly activated with a difference in phase. The angle ∠FCF immediately opens up and vibrates, whereas RC–F shrinks slightly before the strong lengthening. This difference in initial motion is again caused by the recoil of the carbon atom, and when blurred by the instrumental response, results in a measurable delay between the opening of ∠FCF and stretching of RC–F. The red and blue lines in Fig. 4D indicate the simulated changes in RC–F and ∠FCF convolved with an 80-fs Gaussian cross-correlation function so as to incorporate instrumental response. The simulation predicts umbrella opening, RC–F lengthening, and the delay between these, which is in agreement with the experimental observation. A small amount (10%) of intersystem crossing from the 3Q0 to the 1Q1 state has been reported in previous experiments (27), but this effect is not observable in our experiments owing to the spatiotemporal resolution limit.

We have shown that UGED can track a nuclear wave packet with atomic spatiotemporal resolution during nonadiabatic processes involving conical intersections, measuring multidimensional nuclear geometry changes, and simultaneously observing dynamics from different excitation channels in polyatomic molecules. In addition, UGED provides a direct probing method for nuclear degrees of freedom, complementing the standard ultrafast laser spectroscopic techniques that directly probe electronic degrees of freedom. Both the experiment and the data analysis of UGED are generally applicable to a wide range of systems in the gas phase. This approach opens the door for studying many important problems in fundamental photochemistry.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S17

Table S1

References (3763)

Movie S1

References and Notes

  1. The square bracket encloses the electronic configuration, symmetry, and spin-orbit state of the cation core, and the parenthesis indicates the configuration of the Rydberg electron.
Acknowledgments: We thank G. M. Stewart from SLAC National Accelerator Laboratory for the assistance with making movie S1. Funding: The experimental part of this research was performed at SLAC MeV-UED, which is supported in part by the U.S. Department of Energy (DOE) Office of Basic Energy Sciences (BES) SUF Division Accelerator and Detector R&D program, the LCLS Facility, and SLAC under contracts DE-AC02-05-CH11231 and DE-AC02-76SF00515. J.Y., T.F.H., J.P.C., R.C., T.J.A.W., X.Z., and T.J.M. acknowledge support by the AMOS program within the Chemical Sciences, Geosciences, and Biosciences Division, BES of the DOEunder contract DE-AC02-76SF00515. J.P.F.N. acknowledges support by the Wild Overseas Scholars Fund of Department of Chemistry, University of York. T.J.M. acknowledges support from Office of Naval Research grant N00014-12-1-0828 and the Global Climate and Energy Project. This work used the XStream computational resources supported by the National Science Foundation Major Research Instrumentation program (ACI-1429830). Z.L. acknowledges support by the Volkswagen Foundation. M.G. is funded via a Lichtenberg Professorship of the Volkswagen Foundation. K.J.W. and M.C. were supported by the DOE Office of Science, BES under award DE-SC0014170. Author contributions: J.Y., T.J.A.W., J.P.F.N., J.P.C., K.H., R.L., X.S., T.V., S.W., Q.Z., and X.W. carried out the experiments; R.C., J.P.C., J.Y., T.J.A.W, and S.W. developed the laser system; M.G., J.Y., K.J., C.Y., X.S., R.L., and K.J.W. constructed and commissioned the setup for gas phase experiments; J.Y. performed the data analysis with input from M.C., T.J.A.W, J.P.C, X.Z., T.F.H., and M.G; M.C., J.Y., M.G., X.Z., and Z.L. conceived the experiment; X.Z. and T.J.M. performed the AIMS simulations; Z.L. and T.J.M performed the 3D full quantum wave-packet simulation; and J.Y., X.Z., T.J.M, M.G., M.C., and X.W. prepared the manuscript with discussion and improvements from all authors. Competing interests: The authors declare no competing interests. Data and materials availability: Both the experimental data and the simulated trajectories, after basic noise removal and statistical averaging, are available on (36). The raw experimental data are archived at SLAC’s centrally managed GPFS storage, and the raw simulated data are stored at T.J.M.’s group at Stanford University. All the raw data will be made available upon request.

Stay Connected to Science

Navigate This Article