Berry pseudorotation enabled photodissociation of Fe(CO)5 observed by ultrafast electron diffraction Authors: Haoran Ma, J. Pedro F. Nunes, Ambar Banerjee, Martin Centurion, Kareem Hegazy, Renkai Li, Yusong Liu, Xiaozhe Shen, Xijie Wang, Stephen Weathersby, Philippe Wernet, Thomas J. A. Wolf, Michael Odelius, Jie Yang Published: 2026 ABSTRACT The structural dynamics of metal carbonyls are central to processes ranging from catalysis to organometallic synthesis. Here we investigate the photodissociation of a prototypical transition metal carbonyl, Fe(CO)5, using mega-electron-volt ultrafast electron diffraction. By separately tracking structural evolution along the axial and equatorial directions, we provide an atomistic, angle-resolved view of the nuclear motions preceding CO dissociation and infer key features of the excited-state potential energy surface from the experimental observations. We further show that vibrational coupling before reaching the conical intersection facilitates the loss of a random carbonyl ligand via the Berry pseudorotation mechanism. Introduction Carbon monoxide (CO) is a strong field ligand that binds strongly to metal centers, capable of stabilizing low-valent metal and forming metal carbonyls1–4. However, the strong coordination ability of CO is a double-edged sword. On one hand, it enhances the stability of metal carbonyls and organometallic complexes. On the other hand, it hinders the release of CO, limiting the flexibility and reactivity of metal-centered complexes, particularly in catalytic and bioinorganic systems. Thus, it is essential to study the metal-CO dissociation reaction, as it plays a crucial role in synthesizing organometallic compounds from metal carbonyls5–7, activating catalysts by creating unsaturated coordination sites8–10, and regenerating metal complexes affected by CO poisoning11–13. Despite its importance, the non-adiabatic dynamics—particularly the detailed structural changes leading to dissociation remain poorly understood. Gaining deeper insights into these dynamics would enhance our understanding of the mechanisms behind this type of dissociation reaction and contribute to the development of more effective strategies for controlling CO binding and dissociation in catalysis. As a model system for studying metal–CO bond dissociation, the photodissociation of iron pentacarbonyl, Fe(CO)5, has attracted significant attention. Earlier spectroscopic studies generally revealed a sequential CO elimination process, Fe(CO)5 → Fe(CO)4 → Fe(CO)3, with the second dissociation step occurring on the few-picosecond timescale 14–16. Ultrafast spectroscopy with sub-ps resolution has provided direct insight into the initial excited-state dynamics 17–21. They’ve consistently reported rapid internal conversion from metal-to-ligand charge transfer (MLCT) states to metal-centered (MC) states within ~ 100 fs. Many of these studies also suggested preferential axial CO dissociation during or immediately following this relaxation. Thus, while there is consensus on the ultrafast MLCT → MC conversion, significant differences remain in the subsequent structural dynamics and ligand selectivity, underscoring the sensitivity of the photodissociation pathway to excitation conditions and vibronic preparation. A theoretical study employing excited state molecular dynamics (ESMD) has shed light on the early stages of the photochemistry of Fe(CO)522. Upon MLCT excitation, the photodissociation predominantly affects the axial CO ligand, followed by vibration relaxation leading to a dissociative MC state. This process leads to periodic bursts of CO at intervals of ~90 fs, as shown in Fig. 1B. It also suggests a minor dissociation channel, where the initial dynamics involve the closing of the axial C-Fe-C angle and the opening of one of the equatorial C-Fe-C angles, resembles the concept of Berry pseudorotation23,24. Finally, the pseudorotation leads to a loss of identity of equatorial and axial CO ligands, as shown in Fig. 1C.Fig. 1UED experiment and proposed mechanisms of the first-step dissociation of Fe(CO)5.A schematic representation of UED experiment and real-space analysis of diffraction patterns. B the periodic burst mechanism of the first-step dissociation of Fe(CO)5. C, the Berry pseudorotation mechanism of the first-step dissociation of Fe(CO)5. UED experiment and proposed mechanisms of the first-step dissociation of Fe(CO)5. A schematic representation of UED experiment and real-space analysis of diffraction patterns. B the periodic burst mechanism of the first-step dissociation of Fe(CO)5. C, the Berry pseudorotation mechanism of the first-step dissociation of Fe(CO)5. Mega-electron Volt ultrafast electron diffraction (MeV-UED), first demonstrated in 201525, is now a mature experimental method that is capable of providing femtosecond temporal resolution and sub-Ångström spatial resolution simultaneously. While it has successfully been applied to various small organic molecules26–32, it is yet to be applied to transition metal complex. In this study, we present a comprehensive study of both the pre-photodissociation and photodissociation dynamics of Fe(CO)5 using MeV-UED and ground state ab initio molecular dynamics(AIMD) simulations.(Trajectories from the previous study22 have been reanalyzed for the purpose of comparing to experimental data.) Figure 1 depicts a schematic representation of the UED experiment and anisotropic analysis of diffraction patterns. The experiment was carried out in SLAC MeV-UED facility, and the experimental setup is described in detail elsewhere33. Briefly, Fe(CO)5 molecules were introduced to the interaction point through a pulsed nozzle, excited by a femtosecond laser with 266 nm wavelength, and probed by a 3.7 MeV electron beam. The temporal resolution of the setup is ~ 150 fs. The anisotropic analysis, previously demonstrated in small organic molecules such as CF3I and CH2I230,31 is capable of tracking angular-resolved interatomic distances, which is crucial for the identification of axial versus equatorial dynamics in this work. Earlier keV-UED studies by A.H.Zewail and co-workers reported different outcomes for Fe(CO)5 photodissociation34,35. In the 1997 work, two-photon excitation at 310 nm led to sequential CO elimination, yielding Fe(CO)2, Fe(CO), and eventually bare Fe within ~ 20 ps. In contrast, their 2001 study using two-photon 620 nm excitation revealed a rigid Fe(CO)4 intermediate that persisted for over 200 ps without further CO loss. These results already indicated a pronounced sensitivity of the dynamics to excitation conditions, yet they differ from more recent ultrafast measurements. More recently, a study using ultrafast X-ray scattering, which offers better temporal resolution but lower spatial resolution compared to MeV-UED, has unveiled the dissociation dynamics in detail, but wasn’t able to resolve the structural dynamics leading to the first dissociation36. In this work, we observe a clear two-step photodissociation process. Prior to the first step, we report a pre-dissociation dynamics involving vibrational energy redistribution prior to internal conversion, a phenomenon that has never been observed in ESMD simulations, time-resolved spectroscopy or time-resolved scattering studies. Results Figure 2 A shows the electronic ground state pair distribution function (PDF) of Fe(CO)5. The peaks in the PDF report the internuclear distances in the molecule, scaled by the atomic scattering amplitudes. We compared the experimental results with simulated PDFs derived from two sources: the average of ambient temperature structures obtained through AIMD calculations, and a hypothetical rigid structure without any vibration. The experimental and both simulated PDFs are in good agreement within the range of 0 to 3.5 Å, exhibiting three dominant peaks centered at 1.2 Å, 1.8 Å and 3.0 Å, which correspond to carbonyl C-O, Fe-C and Fe-O atom pair distances, separately. For pair distances beyond 3.5 Å, while the ambient temperature structure is still in good match with the experiment, the hypothetical rigid structure becomes obviously inadequate. This rigid PDF displays two distinct peaks centered at 4.2 Å and 5.1 Å, which correspond to non-bonded equatorial-equatorial (eq-eq) distances for the C-O and O-O pairs, respectively. This observation aligns with the fact that the coordination bond between Fe and C is relatively weak, with an average bond energy of 28.8 kcal/mol37. Even at room temperature, the Boltzmann distribution can induce strong molecular vibrations, leading to a broader distribution of interatomic pair distances, especially for atom pairs across different ligands.Fig. 2Ground-state diffraction signals and time-dependent modified molecular diffraction differences of Fe(CO)5.A Experimental (blue), AIMD-simulated (orange), and hypothetical non-vibrational (grey) electronic ground-state PDFs. The dominant peaks and their corresponding atom pairs are indicated by arrows of matching colors. B Experimental modified molecular diffraction differences (ΔsM) at selected time delays (0 ps, 1 ps, and 50 ps; bottom) and the corresponding experimental ground-state (t < − 0.2 ps) sM (top). C Simulated modified molecular diffraction differences (ΔsM) of the hot-ground-state Fe(CO)5, Fe(CO)4, and Fe(CO)3 (bottom) and the corresponding simulated 300 K sM (top). Red vertical dashed lines in B and C extend across both the upper and lower panels, passing through several zero-crossing points of the ΔsM curves for visual guidance. D Integrated experimental ΔsM over 0.6–1.0 Å−1 (blue) and the corresponding first-order reaction-model fit (orange). The shaded areas in B and the error bars in D represent 68% confidence intervals obtained from bootstrap analysis; details are provided in Supplementary Note 6. Arb. units arbitrary units. Ground-state diffraction signals and time-dependent modified molecular diffraction differences of Fe(CO)5. A Experimental (blue), AIMD-simulated (orange), and hypothetical non-vibrational (grey) electronic ground-state PDFs. The dominant peaks and their corresponding atom pairs are indicated by arrows of matching colors. B Experimental modified molecular diffraction differences (ΔsM) at selected time delays (0 ps, 1 ps, and 50 ps; bottom) and the corresponding experimental ground-state (t < − 0.2 ps) sM (top). C Simulated modified molecular diffraction differences (ΔsM) of the hot-ground-state Fe(CO)5, Fe(CO)4, and Fe(CO)3 (bottom) and the corresponding simulated 300 K sM (top). Red vertical dashed lines in B and C extend across both the upper and lower panels, passing through several zero-crossing points of the ΔsM curves for visual guidance. D Integrated experimental ΔsM over 0.6–1.0 Å−1 (blue) and the corresponding first-order reaction-model fit (orange). The shaded areas in B and the error bars in D represent 68% confidence intervals obtained from bootstrap analysis; details are provided in Supplementary Note 6. Arb. units arbitrary units. Before analyzing the pump–probe dynamics, we first examine the reference ground-state scattering. In both the experimental ground-state sM (t < 0) and the 300 K simulated sM (top panels of Fig. 2B, C), the signal amplitude is strongest at small momentum transfer (s ≲ 1.5Å−1). This behavior arises because the dominant pair correlations in Fe(CO)5 are heavy-atom pairs, specifically Fe–O and Fe–C, whose large scattering factors and pair populations concentrate intensity at low s. In real space, the characteristic distances of these pairs are ~ 3.0 Å (Fe–O) and ~ 1.8 Å (Fe–C), corresponding in reciprocal space to oscillation frequencies of roughly 2.1 and 3.5Å−1, respectively. Within the first half-periods of these oscillations, both the Fe–O and Fe–C terms are positive for s < 1.0Å−1 and add constructively, producing a strong positive baseline. Between 1.0 and 1.7Å−1, the Fe–O term becomes negative while the Fe–C term remains positive, resulting in partial cancellation and a zero crossing near 1.5Å−1, as demonstrated by the two-component model in SI Fig. S8. For the time-resolved data, we first focus on the dissociation kinetics. The lower panel of Fig. 2B shows the experimental modified molecular diffraction differences (ΔsM) at three time points: 0 ps, 1 ps, and 50 ps. These time points correspond to the structures before dissociation, after the first CO dissociation, and after the second CO dissociation, respectively. The lower panel of Fig. 2C presents the simulated ΔsM signals of the hot-ground states of Fe(CO)5, Fe(CO)4, and Fe(CO)3 relative to the parent ground state Fe(CO)5 in ambient temperature. Both the experimental and simulated results clearly show that the small-s region reflects the degree of CO loss, consistent with the baseline behavior discussed above for the ground-state sM. Notably, for the latter two species, discernible differences are confined to s < 1.0Å−1, where Fe(CO)3 exhibits a more negative response; at larger s, both experimental and simulated curves nearly overlap. Therefore, the negative band near 1Å−1 in both experiment and simulation can be attributed to Fe–CO bond cleavage, providing a direct and reliable signature of dissociation progression. To quantify the dissociation kinetics, we integrated the experimental ΔsM(s) signal over the range of 0.6–1.0 Å−1 and fitted it with a delayed bi-exponential kinetic model (Supplementary Information 8.1). The best-fit results indicate that measurable structural changes in ΔsM emerge after an onset delay of t0 = 266 ± 52 fs, which we attribute to the pre-dissociation period. After this onset, two subsequent exponential components describe the sequential CO-loss processes, with time constants of τ1 = 200 ± 94 fs and τ2 = 8.0 ± 9.9 ps, respectively. Therefore, the first CO dissociation starts at ~266 fs and proceeds with an effective exponential decay time of ~ 200 fs, followed by a slower second-step dissociation on the few-picosecond timescale. In previous measurements, the two time constants are ~ 200 fs and ~ 3 ps, respectively14,15,17. The second time constant from our measurement appears to be longer, but has much larger uncertainties. We attribute it to the limited signal-to-noise ratio of experimental data and the limited number of available delay points. When the fit is restricted to the − 0.2 to 1 ps time window to isolate the first dissociation step, we obtain t0 = 273 ± 46 fs and τ1 = 186 ± 78 fs, in quantitative agreement with the corresponding parameters from the full fit. This confirms that inclusion of the long-time component does not affect our conclusions regarding the early-time dynamics. Now we focus on the internal conversion dynamics prior to the first dissociation. The anisotropy of the diffraction pattern, resulting from the preferential excitation of molecules by a linearly polarized excitation laser38, contains angular-resolved structural information. We extract the difference pair distribution function(ΔPDF) signal in both the parallel(ΔPDF∥) and the perpendicular(ΔPDF⊥) directions to the laser polarization direction using a 2-D inverse Fourier transform followed by an inverse Abel transform39. Assuming, based on calculated transition dipole moments, that the transition dipole moment of the photo-excited MLCT state is aligned along with the axial direction in Fe(CO)5, ΔPDF∥and ΔPDF⊥ preferentially report structural changes in the axial and the equatorial direction, respectively. Figure 3A, B show experimental ΔPDF∥ and ΔPDF⊥, respectively. Both ΔPDF∥ and ΔPDF⊥ present two prominent peaks, centered around 1.8 Å and 3.0 Å, which correspond to the loss of Fe-C atom pair and Fe-O atom pair. The comparison between the two panels unambiguously reveals that the negative features at around 1.8 Å appear earlier and more pronounced in ΔPDF∥, showing that the axial motion of Fe-C atom pair takes place prior to the equatorial. Figure 3C, D show the ESMD simulated ΔPDF∥ and ΔPDF⊥, respectively, based on the reanalysis of 116 trajectories from the previous study22. It is clear that the experiment and simulation are in qualitatively good agreement after dissociation begins (beyond the dashed black line). However, the experimental data show a distinct 250 fs pre-dissociation process (between the red and black dashed lines), characterized by features that are clearly different from those observed in the dissociation region. Although the ESMD simulations do not explicitly account for laboratory-frame polarization selection, any Wigner-sampled geometry with non-zero oscillator strength can still contribute in a randomly oriented gas-phase ensemble upon suitable orientation relative to the pump polarization. Thus, the main consequence is expected to be a change in the relative weighting of axial and equatorial motions in the measured anisotropic channels, rather than a qualitative breakdown of the experiment-simulation comparison or an explanation for the distinct 250 fs pre-dissociation feature, which instead points to intrinsic early-time dynamics not captured by the ESMD description.Fig. 3Experimental and simulated angle-resolved pair-distribution changes.A, B False-color plots of ΔPDF∥(A) and ΔPDF⊥(B) extracted from the diffraction data. C, D represent the corresponding ESMD simulation results relative to A and B. The black dashed lines represent the beginning of dissociation, and the red dashed lines in A and B represent time zero. Experimental and simulated angle-resolved pair-distribution changes. A, B False-color plots of ΔPDF∥(A) and ΔPDF⊥(B) extracted from the diffraction data. C, D represent the corresponding ESMD simulation results relative to A and B. The black dashed lines represent the beginning of dissociation, and the red dashed lines in A and B represent time zero. One important but so far unanswered question is whether the dissociation is preferentially axial or equatorial. The angular-resolved ΔPDF from our measurement is the most straightforward method for answering this question. We define anisotropic parameter η(r) by the following equation: 1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\eta (r)=\frac{{\Delta {{\rm{PDF}}}}_{\parallel }({{{\rm{r}}}})-{{\Delta {{\mathrm{PDF}}}}}_{\perp }({{{\rm{r}}}})}{\left|{{\Delta {{\mathrm{PDF}}}}}_{\parallel }({{{\rm{r}}}})\right|+\left|{{\Delta {{\mathrm{PDF}}}}}_{\perp }({{{\rm{r}}}})\right|}$$\end{document}η(r)=ΔPDF∥(r)−ΔPDF⊥(r)ΔPDF∥(r)+ΔPDF⊥(r) At a pair distance of r = 3Å (i.e., the Fe-O pair), a higher η value indicates more parallel (axial) dissociation, whereas a lower value indicates more perpendicular (equatorial) dissociation. Pure axial, random, and pure equatorial dissociation correspond to η = − 0.52, 0.12, and 0.73, respectively, according to a simple independent-atom model simulation (Supplementary Information S8). The experimental η values, together with those derived from ESMD trajectories for the periodic axial-burst and Berry-pseudorotation pathways, are shown in Fig. 4. The experimental average over 0.3-0.7 ps, spanning from the onset to the completion of the first dissociation step (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\eta }=-0.00\pm 0.22$$\end{document}η¯=−0.00±0.22), agrees quantitatively with the random-dissociation limit (η = 0.12) and the ESMD-simulated Berry pseudorotation channel (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\eta }=0.02\pm 0.08$$\end{document}η¯=0.02±0.08), within experimental uncertainty. This agreement indicates that the dissociation is neither purely axial nor purely equatorial, but statistically mixed within the experimental uncertainty. The underlying origin of this apparent randomization will be discussed in the next section.Fig. 4Experimental and simulated anisotropy parameter η during the first-step dissociation.The anisotropic parameter (defined by Eq. (1)) for ESMD simulated axial dissociation via the periodic burst mechanism (blue), ESMD simulated equatorial dissociation via the Berry pseudorotation mechanism (orange), and experimental data (green). To account for differences in dissociation initiation times, the ESMD simulation data is shifted by +250 fs. The horizontal dashed black lines represent three extreme cases: pure equatorial dissociation (top), random dissociation (middle), and pure axial dissociation (bottom). The bottom black-grey box represents the dissociation ratio calculated using the kinetic model discussed above. Error bars represent 68% confidence intervals obtained from bootstrap analysis; details are provided in Supplementary Note 6. Experimental and simulated anisotropy parameter η during the first-step dissociation. The anisotropic parameter (defined by Eq. (1)) for ESMD simulated axial dissociation via the periodic burst mechanism (blue), ESMD simulated equatorial dissociation via the Berry pseudorotation mechanism (orange), and experimental data (green). To account for differences in dissociation initiation times, the ESMD simulation data is shifted by +250 fs. The horizontal dashed black lines represent three extreme cases: pure equatorial dissociation (top), random dissociation (middle), and pure axial dissociation (bottom). The bottom black-grey box represents the dissociation ratio calculated using the kinetic model discussed above. Error bars represent 68% confidence intervals obtained from bootstrap analysis; details are provided in Supplementary Note 6. In the pre-dissociation region below the black dashed line in Fig. 3A, the negative feature at around 1.8 Å and the positive feature at a slightly shorter pair distance than that originates from Fe-C stretching vibration, as shown in S13 in Supplementary Information. To explain this phenomenon quantitatively, we employ a genetic fitting algorithm to the experimental ΔPDF∥ and ΔPDF⊥ data40. The vibration of the Fe-C bond plays a crucial role in the relaxation pathway of the excited-state dynamics, as suggested by the ESMD simulation. In the fitting we treat the Fe-C vibration as a non-synchronous harmonic oscillation characterized by two critical parameters: the mean bond length difference(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \bar{{{\rm{R}}}}$$\end{document}ΔR¯) and the variance(Var(R)) of the Fe-C atom pair distribution. The detailed fitting results are shown in Fig. 5. Within the 250-fs pre-dissociation area, it is observed that the mean bond length of the axial Fe-C bond remains unchanged within our experimental precision( ~ 0.03Å). However, a significant axial vibrational motion characterized by the Var(R) of ~0.04 Å2(corresponding to a bond length distribution with ~0.5 Å at FWHM) is evident. This suggests that while the average distance between the Fe and C atoms remains constant, the axial Fe-C bond undergoes strong stretching vibration.Fig. 5Genetic structural fitting results and corresponding ESMD simulation results.A the experiment’s and ESMD simulation’s dissociation ratio. The experiment’s dissociation ratio is calculated using the kinetic model described above and then convoluted with a 150 fs FWHM Gaussian kernel. To determine the simulated dissociation ratio, structures in each trajectory with Fe–C bond lengths exceeding 2.5 Å are considered dissociated. The dissociated trajectory ratio is then calculated against delay and convoluted with a 150 fs FWHM Gaussian kernel. Structural changes in the axial and equatorial directions are labeled in blue and orange, respectively. The vertical dashed red line marks the beginning of dissociation. B and C, the mean bond length (B) and bond length variance (C) from experimental fitting (dashed lines with error bars) and simulation (solid lines with shades). Axial and equatorial data are shown in blue and orange, respectively. To account for the difference in dissociation initiation times, all simulation data is shifted by +250 fs. Error bars represent 68% confidence intervals obtained from bootstrap analysis; details are provided in Supplementary Note 6. Genetic structural fitting results and corresponding ESMD simulation results. A the experiment’s and ESMD simulation’s dissociation ratio. The experiment’s dissociation ratio is calculated using the kinetic model described above and then convoluted with a 150 fs FWHM Gaussian kernel. To determine the simulated dissociation ratio, structures in each trajectory with Fe–C bond lengths exceeding 2.5 Å are considered dissociated. The dissociated trajectory ratio is then calculated against delay and convoluted with a 150 fs FWHM Gaussian kernel. Structural changes in the axial and equatorial directions are labeled in blue and orange, respectively. The vertical dashed red line marks the beginning of dissociation. B and C, the mean bond length (B) and bond length variance (C) from experimental fitting (dashed lines with error bars) and simulation (solid lines with shades). Axial and equatorial data are shown in blue and orange, respectively. To account for the difference in dissociation initiation times, all simulation data is shifted by +250 fs. Error bars represent 68% confidence intervals obtained from bootstrap analysis; details are provided in Supplementary Note 6. The transient ΔPDF features at short pair distances reveal oscillatory components consistent with Fe–C stretching motions, indicating that the Fe(CO)4 fragment produced at 266 nm carries significant vibrational energy. This contrasts with the long-lived, nearly rigid Fe(CO)4 structure reported in the keV-UED study using two-photon 620 nm excitation35, where little vibrational motion was observed. The presence of strong vibrational excitation in our case is consistent with the higher pump photon energy ( ~ 0.66 eV higher), which can efficiently populate vibronic levels of the Fe–C stretching and pseudorotation modes. Discussion To reconcile the apparent discrepancy between the ~ 100 fs timescale commonly reported in ultrafast spectroscopic studies17–20 and the ~ 266 fs onset extracted from our UED data, it is important to consider the different observables involved. Spectroscopy is primarily sensitive to the electronic relaxation (MLCT → MC internal conversion), whereas UED detects nuclear displacements associated with CO escape. According to the analysis by Schori et al. 36, the escape velocity of the dissociated CO is estimated to be ~11.2Å ps−1. Based on this velocity, the escaping motion would result in an apparent delay of about 117 fs, as shown in Fig. S8 of the Supplementary Information. Given that the dissociation process should involve an initial acceleration phase before the CO reaches its terminal velocity, this 117 fs delay likely represents a lower limit of the actual escaping timescale. Thus, the observed 266 fs onset can be decomposed into a 100 ~ 150 fs electronic internal conversion time followed by a > 120 fs nuclear escape time. Although our experimental data do not allow us to unambiguously resolve these two steps, this partition provides a compelling explanation for the apparent discrepancy in timescales between our work and previous ultrafast spectroscopic studies. Within the first ~ 250 fs, neither the mean nor the variance of the equatorial Fe-C pair distance shows significant change, indicating that the equatorial bonds remain relatively stable. This behavior can be understood if the MLCT-state potential energy surface (PES) is approximately parabolic along both axial and equatorial coordinates. The genetic-fitting results, which reveal strong oscillations only along the axial direction, suggest that the Franck-Condon geometry lies near a local minimum on the MLCT PES but in an axially excited vibrational state. Along the axial coordinate, the Franck-Condon point may lie slightly below the conical intersection (CI), leading to a delayed access to the CI and the ~ 250 fs pre-dissociation period. Initially, the nuclear wavepacket is confined mainly along the axial coordinate due to the polarization-selective excitation of the Fe-C(axial) mode. As the system evolves, anharmonic coupling and intramolecular energy redistribution rapidly transfer vibrational energy into other degrees of freedom, including equatorial Fe-C stretches and bending motions. This energy flow gives rise to coupled axial-equatorial vibrations—corresponding to the “scissoring motion” described in earlier theoretical and spectroscopic studies 19,22. In our data, the anisotropy parameter η approaches zero even before the onset of dissociation, implying that axial and equatorial Fe-C motions have already become dynamically mixed within Fe(CO)5. From a geometric perspective, this vibrational mixing manifests as a Berry pseudorotation, in which the identity of the ligands continuously exchanges during the energy redistribution process. Within this picture, the Berry pseudorotation is not a distinct subsequent step but rather a natural outcome of vibrational energy delocalization across the molecule. The overall reaction sequence can be summarized as: photoexcitation initiates an axial Fe-C vibration, followed by intramolecular energy redistribution and mode mixing that drive Berry pseudorotation, eventually resulting in random CO dissociation. Figure 6A displays a hypothesized PES that support the observed dynamics.Fig. 6Schematic illustrations of the excited-state potential energy surfaces and possible wavepacket pathways.A Schematic illustration of excited-state PES involved in the dissociation dynamics of Fe(CO)5. B Schematic illustration of two possible wavepacket pathways on a 1D PES. Schematic illustrations of the excited-state potential energy surfaces and possible wavepacket pathways. A Schematic illustration of excited-state PES involved in the dissociation dynamics of Fe(CO)5. B Schematic illustration of two possible wavepacket pathways on a 1D PES. The branching ratio between the two dissociation pathways appears to differ between the present work and previously reported ultrafast X-ray scattering results36. This difference might originate from variations in the initial vibrational state distribution of the excited ensemble, determined by the Franck-Condon factors and sensitive to subtle differences in excitation conditions. Figure 6B presents one hypothesized scenario illustrating how such variations could influence the dynamics. If the Franck-Condon point lies slightly above the conical intersection, the wavepacket can quickly reach the CI during the initial Fe-C stretching, leading to a faster and predominantly axial dissociation pathway, similar to that observed in excited-state molecular dynamics simulations. Conversely, if the Franck-Condon point lies marginally below the CI, the wavepacket may first transfer energy into equatorial vibrational modes before reaching the CI, favoring a more randomized, Berry-pseudorotation-enabled dissociation. Although this description is qualitative, it provides a plausible explanation for the different dissociation preferences observed under comparable excitation wavelengths. Future investigations combining wavelength-dependent measurements with multidimensional nonadiabatic dynamics simulations will be crucial for quantitatively mapping how the initial vibronic preparation governs the branching between these pathways. Our time- and angle-resolved MeV-UED measurements provide a three-dimensional picture of the structural dynamics accompanying the first-step photodissociation of Fe(CO)5. By resolving the axial and equatorial components of motion, we observe that the dissociation involves both directions and is consistent with a Berry-pseudorotation-mediated random CO loss. The anisotropy and kinetic analyses further indicate that an initial axial Fe-C vibration precedes the onset of bond breaking and couples to other modes during energy redistribution. These results offer a direct structural view of the early nuclear motions that govern CO release in a prototypical transition-metal carbonyl. Beyond demonstrating the capability of MeV-UED for such complex molecular systems, this work provides experimental insights into how vibrational excitation shapes ligand dynamics in photoinduced reactions. Methods Megaelectronvolt gas-phase ultrafast electron diffraction The details of the experimental setup have been described elsewhere25,41. In summary, we utilized an 800 nm output from a Ti:sapphire laser system operating at a repetition rate of 120 Hz and split the beam into two paths. The pulses in both paths were frequency tripled. The pulses from the probe beam path were directed onto the photocathode of a radiofrequency gun, generating an ultrashort pulse containing ~104 electrons. These electrons were then accelerated to a kinetic energy of 3.7 MeV in a rf-type photoelectron gun and focused to a spot size of 200 μm full-width at half-maximum using a solenoid in the interaction region of a gas-phase experimental chamber. The pump pulses (30 μJ) were focused into the experimental chamber to a diameter of 250 μm full-width( ~ 0.5TW/cm2) at half-maximum and overlapped with the electron pulses at a 5∘ angle. The compound Fe(CO)5 was obtained from Sigma-Aldrich and used without further purification. We introduced Fe(CO)5 vapor into the interaction region of the experiment using a pulsed nozzle with a 100 μm orifice. Diffracted electrons were detected using a combination of a phosphor screen and an electron multiplying charge coupled device (EMCCD) camera. The rough time zero was determined using a single-crystal Si sample by tracking the transient change of a specific Bragg peak, while the final time zero used in data analysis was refined from the time-resolved diffraction data (Supplementary Information S18). Based on the dynamic signal and the relative AIMD simulation signal levels, we estimate that ~13.2% of the molecules were excited (Supplementary Information S7). Additionally, we conducted pump pulse energy titration to confirm that we were operating in the linear absorption regime (Supplementary Information S9). Anisotropic diffraction pattern analysis This diffraction pattern analysis method is explained graphically elsewhere30,31. In summary, the diffraction pattern is first transformed using an inverse 2-D fast inverse Fourier transform (2D-IFFT), which yields a z-projection of the pair distribution function (PDF). The outcome is then transformed using Abel inversion with the rBasex algorithm42, in which the blurring due to z-projection is removed. We split the angular-resolved PDF into 9 sections, with each section covering a 10∘ angular range. The two important sections— ΔPDF∥ (0∘–10∘) and ΔPDF⊥ (80∘–90∘) are selected for discussion. In our experimental data, a few additional filters are applied to suppress noise, including Legendre filtering43, a Gaussian filter in momentum space (0.229 Å−1 FWHM) and in temporal domain(150fs FWHM). In the 2D FFT, momentum transfer up to s = 10.2Å−1 is used, and a zeropadding technique is applied. The low s (s < 0.8Å−1) signal is unavailable in the experiment due to a center hole in the phosphor screen that transmits the unscattered electron beam. We compensate the isotropic term with a simple sinc function, and the anisotropic term is manually set to zero. AIMD computational details As a basis for the theoretical analysis of the experimental data, we performed Born-Oppenheimer ab initio molecular dynamics (AIMD) simulations of ambient temperature (300 K) 1 Fe(CO)5(g) and of hot 1 Fe(CO)5(g) (1500 K), 1 Fe(CO)4(g) (1500 K), and 1 Fe(CO)3(g) (1000 K) in the CP2K set of programs44. Density functional theory (DFT) was used with the dispersion corrected BLYP functional45–47 and the mixed Gaussian and plane wave (GPW) method48 with a plane wave cutoff of 600 Ry in the description of the electron density. A cubic simulation cell of 20 Å was used with a Poisson solver for isolated systems49, and the simulations were run for 9 ps with a time-step of 0.1 fs. The GTH pseudopotentials50 (with explicit inclusion of semi-core 3s3p electrons for Fe) were employed in combination with the GTH-DZVP (Fe) and GTH-TZVP (C, O, H) basis sets. Supplementary information Supplementary Information Description of Additional Supplementary File Supplementary Data 1 Supplementary Data 2 Transparent Peer Review file Supplementary Information Description of Additional Supplementary File Supplementary Data 1 Supplementary Data 2 Transparent Peer Review file