Photo-ionization and fragmentation of Sc₃N@C₈₀ following excitation above the Sc K-edge

Endohedral (or internally doped) fullerenes are intriguing systems that bridge the gap between molecular and nanosystems.^1,2 Like C₆₀, there is much to learn about their behavior when they are excited with photons, in particular, in the hard x-ray regime. Endohedral fullerenes are of great interest to study due to their unique properties, including electron transfer between the encaged species and the carbon cage,² and their potential use in molecular electronics and organic photovoltaics. The understanding derived from the photoionization of carbon nanomaterials can provide insight toward optimizing their properties for use in these applications.³ Endohedral fullerenes have been studied with optical lasers^4,5 and synchrotron radiation.^6–9 Synchrotron studies, which allow for core-level ionization of the encapsulated species, have indicated that there is a much higher fragmentation propensity if the inner species becomes highly excited,⁷ especially compared to the level of fragmentation observed following excitation/ionization of valence electrons by optical lasers.⁵ The most intriguing aspect of the endohedral systems compared to empty fullerenes is the mutual influence of the electrons from the cage and the electrons from the encapsulated moiety, which may significantly modify the dynamics.¹⁰ 

Like synchrotron radiation, x-ray free electron lasers (FELs) can access the core levels of the inner moiety of the endohedral fullerene.¹¹ The brilliance and photon energy tunability of the FEL allow for site-specific photon absorption and, because they have short pulse durations, also allow, in principle, for following the dynamics of the molecule.^11,12 Most of the early experiments involving FELs have been on understanding the nature of the interaction of light atoms, such as neon,¹³ and small molecules, such as N₂.^14,15 Recently, polyatomic molecules¹² and fullerenes¹⁶ have been studied, with an emphasis on understanding the ionization dynamics involved when multiple x-rays are absorbed causing ionization and subsequent Auger decay, which contribute substantially to what is known as “radiation damage”.¹⁷ 

The aim of the present experiment was to investigate Sc₃N@C₈₀ by site-selectively ionizing the Sc (1s) with an approximately 10 fs duration hard x-ray pulse at 4.55 keV photon energy inducing core-ionization followed by an Auger cascade. This investigation allows us to address how electronic rearrangement affects (i) the nuclear motion leading to atomic rearrangement, (ii) bond elongation and breaking, (iii) fragmentation of the moiety (inner species) and the cage, and (iv) bond (re)-forming. For Sc, it is known that for a K-shell vacancy, the radiative to nonradiative branching ratio is 0.18:0.82, while for the L-shell vacancy, the radiative yield is negligible compared to nonradiative relaxation which primarily occurs via the Coster-Kronig type decay through the LMM Auger process.¹⁸ Thus, following removal of a K-shell electron of Sc, further ionization of the whole system is expected to proceed via the Auger cascade or impact ionization of the cage. It is expected from the binding energy of the electrons in Sc that KLL (2s⁻²/2s⁻¹2p⁻¹/2p⁻²) Auger decay will produce a free electron having a kinetic energy (KE) of about 3.5–4.0 keV. Likewise, LMM (3s⁻¹3p⁻¹/3p⁻²) decay creates ∼330–400 eV electrons.¹⁹ The high energy electrons are expected to escape the cage with a low probability for further excitation of the molecule, while the few-hundred electron volt electrons are expected to have a higher probability to undergo inelastic collisions with the cage.^20,21 A cartoon schematic of this mechanism is shown in Fig. 1(a). We aimed to study transient structural changes by using a delayed second x-ray probe pulse, at a photon energy of 5.0 keV, through monitoring the production of fragment ions.

The experiment was carried out at beamline BL3, EH4c hutch²² of the SPring-8 Angstrom Compact free electron LAser (SACLA)²³ using a time-of-flight ion spectrometer.²⁴ Both the “pump” and the “probe” pulses were produced by the same electron bunch using a variable length undulator scheme.²⁵ For this experiment, a pump pulse with 4.55 keV photon energy (FWHM = 13.11 eV) and a probe pulse with 5.0 keV photon energy (FWHM = 69.79 eV) were used. Since the Sc atoms have an average charge state of 2.4+ in the endohedral complex, the pump pulse was tuned to be well above the K-edge of Sc (4492.8 eV²⁶) so that Sc³⁺ could be ionized by single photon absorption. Meanwhile, the probe pulse was chosen to be about 5.0 keV for two reasons: (1) the photon energy of 5.0 keV sufficiently allows for ionization to a highly charged states, above 4+, and (2) the difference in photon energies between the two pulses needed to be large enough such that the pump and the probe could be monitored independently by an in-line spectrometer.²⁷ Both beams were focused by a Kirkpatrick–Baez (KB) mirror system to a focal spot size of about 1.7 × 1.8 µm² (FWHM). Just prior to the experiment, the beam sizes were individually measured by using an edge scanning method that utilizes a 200 µm gold wire to create an intensity profile.²⁸ During the experiment, the spatial overlap of the pump and the probe was checked for the two delay points by using another 200 µm gold wire at the interaction region, since the scheme²⁵ used for the production of the pump and the probe pulses did not guarantee an automatic spatial overlap following changes in the undulator gap of the electron bunch. The repetition rate of the pulses was 60 Hz, and the duration of these two pulses was estimated to be about 10 fs each, as measured using a spectrometer consisting of an analyzer flat crystal of silicon²⁹ prior to the experiment. Since these measurements require their own chamber, the pulse duration was measured ex situ before the beamtime. The temporal jitter between the two pulses was estimated to be a few femtoseconds due to the electron bunch spacing, as measured previously for this scheme.²⁵ Using an in-line spectrometer,²⁷ the shot-to-shot pulse energy was monitored to be about 105 µJ for the pump and 110 µJ for the probe, and the energy fluctuation between the two pulses was found to be about 20% of the mean for each arm, as also shown in Fig. S1.

The experimental setup is shown schematically in Fig. 1. The Sc₃N@C₈₀ sample (97% purity) was procured from SES Research. The sample was converted to an effusive vapor by using a sample dispenser oven¹⁶ which evaporated Sc₃N@C₈₀ into its vapor phase at 970–1000 K. The orientation of the oven was horizontal with respect to the spectrometer axis and along the polarization of the x-ray pulses. We estimate the target density from the oven to be about 2 × 10⁸ cm⁻³.

The ions created were extracted by a uniform electric field of 550 V/cm at the interaction region and were subsequently detected on a microchannel plate (MCP) detector. The ion hits on the MCP were recorded using a digitizer and a software discriminator.³⁰ More details of the ion TOF spectrometer are described elsewhere; see Ref. 24.

At a fluence of about 40 μJ/μm² for each pulse (corresponding to an intensity of about 4 × 10¹⁷ W/cm²), the ionization mechanism is expected to be step-wise. First, there is a photoionization (P) event followed by an intra-atomic Auger (A) decay until a stable charge state is reached. This mechanism was previously reported for FEL interactions with atoms, molecules, viruses, and weakly bound clusters.^{11–14,16,31–37} At the photon energies used in the experiments, the absorption cross section of the three scandium ions inside the cage is about 2.0 × 10⁻¹⁹ cm² (0.2 Mb) for the 4.55 keV pump and 1.5 × 10⁻¹⁹ cm² (0.15 Mb) for the 5.0 keV probe pulse. For the cage itself, with 80 carbon atoms, the cross section is about 3.0 × 10⁻²⁰ cm² (0.03 Mb) for both of the pulses. The stability of the Sc₃N@C₈₀ molecular complex depends on the oxidation state of the cage. It has been determined that electron sharing with the Sc₃N stabilizes the C₈₀ cage, which attains the stable icosahedral form by accepting about 6.3e from the inner moiety.³⁸ This induces a partial charge of 2.4+ on each Sc atom inside the cage, for the overall neutral Sc₃N@C₈₀.

Figure 2(a) shows the time-of-flight spectrum for Sc₃N@C₈₀, for 0 fs delay between the two x-ray pulses, with the different charge states of the parent molecular ions indicated. Multiple charge states of the parent molecule up to 4+ are observed, as seen previously for the case of the single photon ionization of the Sc 2p orbital⁶ for the same target. The lack of singly charged parent molecules shows that the Sc (1s) photo-ionization plus Auger decay efficiently produces multiple charge states of Sc₃N@C₈₀. It is striking that there is no evidence in the mass spectrum for the occurrence of C₂ fragmentation from the carbon cage, i.e., Sc₃N@C₇₈, Sc₃N@C₇₆ and so on. Such typical “shrink-wrapping” behavior has been observed in previous studies where the Sc 2p orbital was photoionized⁶ and also in intense fs laser studies of Ho₃N@C₈₀.⁵ The lack of C₂ fragmentation from the cage is a clear indication that the parent molecular ions visible in the mass spectrum have low internal energies. Note that the MCP detector was not floated in these experiments to provide high impact velocity, and thus, the detection efficiency of the parent molecular ions is expected to be significantly lower than that for the small carbon species in the experiments. The other striking observation is the lack of highly charged fragment ions. The highest fragment charge state that can be clearly distinguished in the spectra is C²⁺ with possibly a small amount of Sc²⁺. The low level of charging and corresponding lack of Coulomb explosion is also manifested by the rather narrow ion peaks and the dominance of the detection of single ions per shot (∼96% of all shots that yield an ion signal), as illustrated in Fig. 1(b). The most stable charge on the parent molecule is 2+ and 3+. This is due to the stabilizing nature of the π electron delocalization in the system following ionization, which corresponds to a larger gap between the HOMO and LUMO orbitals for the resultant multiply charged parent molecule.³⁹ Figure 2(b) shows the time-of-flight spectrum for the various singly charged carbon fragments. We observe carbon fragments, $Cn+$⁠, for n < 20. Some doubly charged atomic carbon is also observed, indicating that a portion of the carbon atoms in the cage lost multiple electrons either through their transfer from the cage to the moiety or through electron collisions from the Auger electrons. What is surprisingly missing in the time-of-flight spectra is the observation of highly charged (>2+) states of Sc ions which one may have expected following Sc (1s) photo-ionization and subsequent Auger decay. That we mainly observe Sc⁺, and only a small amount of Sc²⁺, seems to be indicative of charge being efficiently transferred from the cage to the Sc ions.

In Fig. 2(c), we show the yield of the singly charged carbon fragments for the two delay points, normalized to the number of shots and pulse energy per shot (see supplementary material, Fig. S1 for the distribution of pulse energies for the two pump-probe experiments). The inset shows the Sc-containing fragment ion yields, which include Sc⁺, Sc$C2+$⁠, ScCN⁺, and Sc$C4+$⁠. The latter three fragments are products requiring new bond formation since in the initial system, no direct bond existed between the three Sc and the C₈₀ cage. Such new bonds are typically observed following ionization and multifragmentation of the cage, as seen previously for the case of FEL and femtosecond optical laser-induced fragmentation of Ho₃N@C₈₀^5,32 as well as photo-ionization of Sc (2p) in Sc₃N@C₈₀⁶ and intense nanosecond optical laser photoionization/fragmentation of La@C₈₂.⁴ On comparison of the two sets of data, the normalized ion yield is slightly higher for the small carbon species and, more significantly, the scandium-containing small fragments at 100 fs delay compared to 0 fs. Due to technical reasons of the two pulse scheme,²⁵ the only way single pulse spectra could be obtained is by filtering out the probe pulse using an Al filter. This resulted in some absorption of the pump pulse, and thus, the pulse energy for single pulse mode was different than what was used for the two pulse experiment. Single pulse spectra obtained for pulse energies of 80 µJ and 520 µJ show mostly similar trends in the relative intensities of the fragment ion peaks (see supplementary material, Figs. S2–S4). The pulse energy variation between the two delays (0 fs and 100 fs) was about 5%, which should not change the fact that what we are seeing is predominantly single photon absorption (see supplementary material, Fig. S5, which shows that the effect of this pulse energy variation is negligible for multiple ionization of parent molecules). Inside the cage, each Sc has contributed about 3e to the cage and the nitrogen atom. This implies that the 3d and 4s levels are not occupied. The x-ray absorption first leads to a KLL process, leaving two L-shell holes. This can be followed by subsequent decay of two M-shell electrons to fill the holes. Since direct double ionization has a much lower probability than single Auger decay,⁴⁰ this would leave the system with the 4e removed from the Sc. Electrons from the cage would then flow to the ionized Sc. If very little energy is transferred to the cage on the way out or via the charge rearrangement process, then one could expect to see up to Sc₃N@$C804+$⁠. If the system absorbed more than one photon, then we would expect to see higher charge states. However, the stability of endohedral fullerenes is also dependent upon the π-orbital delocalization of the system³⁹ and does not support high charge states beyond 5+. For the case of single photon absorption, one can definitely create multiple charge states on the Sc site. However, as also observed in previous studies of Sc 2p ionization,⁶ competing processes of cascade ionization and charge rearrangement from the cage to the moiety lead to fragmentation, as it becomes more and more unfavorable for the system to remain stable with increasingly higher charge states. Therefore, further ionization of the cage would result in instability of the system, manifested by cage opening/breakup. This ensures that the products which will show time-dependent features are fragments ensuing from charge rearrangements by the cage. Such fragments are Sc⁺, ScCN⁺, Sc$C2+$⁠, Sc$C3+$⁠, and so on. It is expected that if the two photon processes are initiated by the absorption of the pump and the probe pulses by the same system, then the newly formed Sc$Cn+$ would further ionize and fragment.

In the present case, differences between the two delay points include a significant nonlinear increase in the $C3+$ intensity relative to the other small carbon species and the Sc⁺ intensity relative to the other Sc-containing fragments at the higher pulse energy, which may be indicative of 2-photon processes. In order to obtain more evidence for this, we consider the ions that are detected in coincidence.

Figure 3(a) shows a portion of the photo-ion photo-ion coincidence (pipico) map for ions relevant to Sc⁺ and small carbon fragment ions. Since the MCP detector count rate was ≪1/shot (about 5% of the FEL shot produced any hits on the MCP for the two pulses), we assume the ions coincident with each other evolved from the same molecule. The prominent partner ions detected in coincidence with Sc⁺ are $Cn+$ with n spanning n = 1–18. Note that it is not possible to rule out the contribution of doubly or triply charged species with the same mass/charge ratio as the singly charged species. However, a consideration of the peak shapes in the mass spectra would indicate that the contribution of multiply charged species is very low.

In Fig. 3(b), we also see a larger intensity for coincidences between Sc⁺ and $Cn+$ for the 100 fs delay data, particularly for the smallest carbon species. There is a more obvious difference in the coincidences between Sc⁺ and scandium-containing fragment ions with a very significant decrease in the Sc⁺–Sc⁺ coincidence rate and increase in the Sc⁺–Sc$C2+$ and Sc⁺–ScCN⁺ intensities at a delay of 100 fs between the two pulses. From the pipico data, we also observe indications of Sc²⁺ coincident with Sc⁺, with the prominent cage fragment $C3+$⁠, and with ions where pieces of the cage have formed new bonds with atoms originating inside the cage, i.e., Sc$C2+$ and ScCN⁺. As there are large neutral fragments from the original molecule not accounted for, these pipico structures do not form sharp lines, but rather small blobs, as long as the remaining fragment(s) do not carry the majority of the dissociation momentum. Somewhat surprisingly, we do not observe a noticeable amount of Sc²⁺ coincident with species other than Sc⁺, although we also expect these coincidences to be difficult to identify due to the higher level of multibody fragmentation, further blurring out the ion-ion coincidence islands. There are also some indications of carbon fragment ions being produced in coincidence with other carbon fragment ions, such as C⁺ + $C2+$⁠, C⁺ + $C3+$⁠, C⁺ + $C4+$⁠, C⁺ + $C5+$⁠, C⁺ + $C7+$⁠, and others.

A consideration of both the full mass spectra and the coincidence data allows us to draw some interesting conclusions about the ionization and fragmentation behavior of the Sc₃N@C₈₀ molecule. The full mass spectra, which we believe to be dominated by single-photon absorption, provide evidence for two distinct processes. First, the strong signal from intact parent molecular ions with charge states from 2+ to 4+ is indicative of Auger cascades involving predominantly the Sc. The doubly charged species arises from a single KLL Auger transition from the Sc (2p) with the emission of a high energy electron. If we consider the role of the cage which encapsulates the moiety, then we would see that $C804−$⁠, which is the charge state of the cage after charge rearrangement followed by removal of 2e from a single Sc, is as stable as $C806−$⁠, which is the charge states attained by the cage in the neutral system.³⁹ This is analogous to stating that Sc₃N@$C802+$ is as stable as the neutral system due to the interplay of the charge stabilization by the π orbital and the strain effect due to Coulomb repulsion. Triply and quadruply charged species most likely arise from LMM Auger decays within Sc. Considering that, within the cage, Sc exists on average in a 2.4+ charge state, no further Auger decays within the Sc system to produce higher charge states are possible. For the highest observable charge state, this would leave a Sc ion with an electron configuration of 1s²2s²2p⁴3s²3p⁶. Any subsequent rearrangement of electrons within the endohedral fullerene system to compensate for the high charge on the Sc ion would transfer electrons either from neighboring species inside the cage or from the carbon atoms in the cage. The lack of C₂ evaporation from these parent molecular ion species, such as Sc₃N@C₇₈, Sc₃N@C₇₆, and so on, [see Fig. 2(a)] is a clear indication that energy transferred to internal vibrational energy of the cage is insufficient to produce any fragmentation on the microsecond time scale in which the ions reside in the extraction region of the mass spectrometer. This would be expected for sequential transfer of electrons from the carbon cage to the Sc (n = 3) shell. It can therefore be assumed that to produce these intact parent molecular ions, the energetic electrons emitted from the KLL and LMM cascades do not collide inelastically with the carbon cage while exiting the molecule. If, however, there is an inelastic collision between the 300–400 eV electrons from the LMM cascades with the carbon cage, this will provide enough internal excitation to induce a phase transition in the hot molecule,⁴¹ leading to rapid cage breakup and the characteristic mass spectrum of small carbon fragment ions that we observe. Extensively studied fragmentation patterns using multiphoton laser excitation⁴² or collisional excitation⁴³ transfer a wide range of energies to the fullerenes and typically produce a bimodal fragment distribution. Here, we either transfer sufficient energy via inelastic transfer of energy from electrons produced within the cage to cause rapid statistical cage breakup and the formation of singly charged ring and chain fragments or there is insufficient energy transferred to evaporate C₂ from the cage on the time scale of the mass spectrometry detection, leaving intact parent molecular ions.

The statistical breakup of internally excited fullerene cages had been modeled previously using both a percolation model⁴⁴ and a maximum entropy model.⁴¹ In the bond percolation model, a power-law distribution for the small cage fragment species is predicted. The fragment ion intensities for both the 0 fs and 100 fs delay data have been plotted on a ln-ln plot in Fig. 4. Figures 4(a) and 4(b) show that the data fit reasonably well to the expected power law behavior for all fragments from $C3+$ to $C15+$⁠, as given by S_n ∝ n^−γ. Here, S_n is the yield of carbon fragments, $Cn+$⁠, and n is the number of carbon atoms, while γ is the fit parameter. The goodness of the fit, R², for the blue fit lines for Figs. 4(a) and 4(b) are 0.63 and 0.82, respectively. For the fragmentation of a cage system such as C₆₀, it has been shown previously that due to finiteness and the periodic nature of the system, γ ≈ 1.3 is obtained for a fullerene system undergoing a phase transition by emitting small carbon molecular fragments.⁴⁴ This value of γ is a consequence of the amount of energy transferred to the fullerene in the highly charged ion collisions⁴⁴ after averaging over the impact parameter dependence of the energy transfer. The gradient will change depending on how much internal energy is present in the system. For very high amounts of internal energy, γ will be higher since the fragment distribution will shift more to smaller mass ions. Since, neither from the consideration of C⁺ and $C2+$ nor from their omission in the overall ln-ln fit, we did not obtain the value of γ higher than 1.3 for any of the delays, it can be said that the measured distribution is consistent with the power law behavior expected from statistical breakup as predicted by the percolation model, in analogy with nuclear multifragmentation.^45,46

The maximum entropy model as applied to C₆₀ shows similar behavior but is able to reproduce variations in the relative intensities, including the lower than predicted yield for C⁺ and $C2+$ by taking the ionization energies and binding energies of all possible fragments into consideration.⁴¹ For C₆₀, the internal energy leading to the observation of the small carbon ring and chain fragment ions was shown to be at about 230 eV,⁴¹ while for the larger endohedral system La@C₈₂, it was predicted to be at about 300 eV.⁴ A simple extrapolation based on the number of degrees of freedom for Sc₃N@C₈₀ would lead to an expected threshold for the phase transition of about 320 eV. A lower energy would be expected if the cage was multiply charged. This energy is consistent with the electron energies produced by the LMM Auger transitions in Sc. The observed behavior is thus supporting a model in which the cage is efficiently excited (possibly further ionized) by interaction with Auger electrons as they exit the molecule, leading to a rapid energy equilibration followed by cage breakup into predominantly singly charged fragments. Application of the maximum entropy model to La@C₈₂ fragmentation was also shown to reproduce the formation of La@$Cn+$ fragments, similar to the scandium-containing ions observed in the current experiments.⁴ The statistical nature of the breakup will also discriminate against highly charged ions in the maximum entropy model due to their relatively higher ionization energies. The consideration of established results from both these models provides further support for our assertion that the single photon absorption process from the pump pulse alone initiates the cage breakup in the present study.

If we again turn our attention to the coincidence data in Fig. 3(b), we see small but significant differences between the 0 fs and the 100 fs data which may be indicative of two-photon processes and the time scale of intramolecular electron transfer events. The increase in the intensity of coincidences between Sc⁺ and small scandium-containing fragment ions compared to coincidences between two singly charged scandium ions could be due to the dynamics of electron transfer from the carbon cage to the photoionized Sc ions. At 0 fs delay, all direct or Auger induced photoionization can be expected to occur on the 10 fs time scale. Any two-photon absorption is likely to produce two charged Sc species within the cage giving a relatively high probability to detect a Sc⁺–Sc⁺ coincidence signal for 0 fs delay (taking into consideration the energy and charge equilibration that will take place as the system is undergoing the rapid phase transition and breakup). After a delay of 100 fs, there is time for electron transfer from the cage carbon atoms to the highly charged Sc to occur, transferring the positive charge to the cage prior to the absorption of the second photon and therefore giving a higher probability to detect more small charged carbon species in coincidence with Sc⁺ on the second (P)-(A) cycle. The higher yield of coincidences between Sc⁺ and scandium-containing fragment ions for 100 fs delay may be caused by the initiation of the cage breakup on this time scale after the first photon absorption. In the total count of Sc⁺, as seen in Fig. 2(c), we do not see a significant difference between the two delays. K-shell ionization of two neighboring Sc is more likely to give Sc⁺–Sc⁺ coincidences than in the single photon case. We can see from Fig. S3 that the single pulse Sc⁺–Sc⁺ peak is much lower in intensity than the pump-probe coincidence data in Fig. 3. This is reasonably convincing evidence that a majority of the intensity of the Sc⁺–Sc⁺ coincidence peak in Fig. 3 is coming from the pump-probe. At 100 fs, Fig. 3 distribution is much closer to that of Fig. S3, and we can interpret as the holes in the first ionized Sc being refilled due to transfer from the cage on a sub-100 fs time scale so that for a delay of 100 fs the Sc is basically similar to that of the pristine molecule with the charge transferred to the carbon cage.

We have experimentally probed the photo-ionization and fragmentation behavior of Sc₃N@C₈₀, following K-shell core ionization of the encapsulated Sc. The mass spectra provide evidence for two processes leading to distinct signatures in the ion distributions: (i) multiple ionization of the parent molecular ion via Auger cascades that transfer sufficiently little energy to vibrational excitation of the molecule to survive intact during the microsecond time scale of the mass spectrometer and (ii) high-energy transfer to the cage followed by cage breakup and the production of small, predominantly singly charged fragment ions. Pump-probe measurements at delays of 0 fs and 100 fs provided additional evidence for a “slow” (>10 fs) electron transfer between the carbon cage and the multiply ionized Sc ions. Due to the “all or nothing” nature of the energy transfer to the cage, the mass spectra look quite different to the typical bimodal fragment distributions that are normally observed for fragmenting fullerene species. The absence of highly charged species is also an unusual feature for K-shell excitation studies and is thought to be a consequence of the particular geometry of the endohedral species and the highly statistical nature of fullerene fragmentation. Although the current data are of a rather preliminary nature and requires further experimental and theoretical study to fully unravel the complex dynamics of the studied system, we have shown the feasibility of x-ray FEL pump-probe experiments to probe the complex electron and nuclear dynamics of large molecular systems.

See supplementary material for Figs. S1–S5 mentioned in the manuscript.

This work was funded by the Chemical Sciences, Geosciences, and Biosciences Division, Office of Basic Energy Sciences, Office of Science, U.S. Department of Energy, Grant No. DE-SC0012376. T.T. gratefully acknowledges support from the JSPS KAKENHI, Grant No. JP16J02270. H.F., K.N., and K.U. were supported by “Dynamic Alliance for Open Innovation Bridging Human, Environment and Materials” from the Ministry of Education, Culture, Sports, Science and Technology of Japan (MEXT), by the Research Program of “Dynamic Alliance for Open Innovation Bridging Human, Environment and Materials” in “Network Joint Research Center for Materials and Devices,” by the Japan Society for the Promotion of Science (JSPS) KAKENHI, Grant No. JP15K17487, and by the IMRAM project. D.Y. was supported by a Grant-in-Aid of Tohoku University Institute for Promoting Graduate Degree Programs Division for Interdisciplinary Advanced Research and Education. K.K. and E.K. acknowledge financial support from the Academy of Finland.

We would like to thank the SACLA Accelerator scientists and beamline staff for the technical and experimental help provided during the beamtime.