Direct observation of photocarrier electron dynamics in C60 films on graphite by time-resolved two-photon photoemission

Time-resolved two-photon photoemission (TR-2PPE) spectroscopy is employed to probe the electronic states of a C60 fullerene film formed on highly oriented pyrolytic graphite (HOPG), acting as a model two-dimensional (2D) material for multi-layered graphene. Owing to the in-plane sp2-hybridized nature of the HOPG, the TR-2PPE spectra reveal the energetics and dynamics of photocarriers in the C60 film: after hot excitons are nascently formed in C60 via intramolecular excitation by a pump photon, they dissociate into photocarriers of free electrons and the corresponding holes, and the electrons are subsequently detected by a probe photon as photoelectrons. The decay rate of photocarriers from the C60 film into the HOPG is evaluated to be 1.31 × 1012 s−1, suggesting a weak van der Waals interaction at the interface, where the photocarriers tentatively occupy the lowest unoccupied molecular orbital (LUMO) of C60. The photocarrier electron dynamics following the hot exciton dissociation in the organic thin films has not been realized for any metallic substrates exhibiting strong interactions with the overlayer. Furthermore, the thickness dependence of the electron lifetime in the LUMO reveals that the electron hopping rate in C60 layers is 3.3 ± 1.2 × 1013 s−1.

In this article, we present a spectroscopic study of photocarrier electron dynamics of C 60 deposited on a highly-oriented pyrolytic graphite (HOPG) as a model for multi-layered graphene by means of time-resolved two-photon photoemission (TR-2PPE) spectroscopy. The TR-2PPE unveils the ultrafast dynamics of electronically excited states on a surface system with femtosecond (fs = 10 −15 s) time resolution 31,32 , and it has lately begun to be applied to graphene-related surfaces 33,34 .
The experimental results reveal that the energy levels of C 60 , which are independent of the C 60 coverage, are little influenced by the 2D electronic nature of the in-plane sp 2 carbon network; the overlayer molecular C 60 film is electronically decoupled from the HOPG substrate. The photoexcited electron behavior in the lowest unoccupied molecular orbital (LUMO) of C 60 suggests that free photocarrier electrons are generated by the dissociation of hot excitons without forming lower-lying excitons. The decay rate of photocarrier electrons into the HOPG substrate is very low (1.31 × 10 12 s −1 ) compared with that on metallic substrates, which is also a characteristic of the 2D substrate. Moreover, thanks to the 2D nature of the material, it is revealed that the photocarrier electron hopping rate in C 60 layers is 3.3 ± 1.2 × 10 13 s −1 , from the C 60 thickness dependence of the TR-2PPE measurements.

Results and Discussion
Energy levels of C 60 monolayer on HOPG. Figure 1a shows 2PPE spectra for a C 60 monolayer (1 ML, see Supplementary Figs S1 and S2 for the coverage calibration) formed on HOPG by changing the incident hν at ~0.05 eV intervals. The photoelectrons emitted with surface normal (k // = 0) were recorded with a limited acceptance angle of ± 0.6°. The horizontal axis shows the final state energy (E final ) relative to Fermi level (E F ). Six spectral features labeled as L0, L2, F0, F1, F2, and H appear by the formation of C 60 film. The vacuum level (E vac ) is located at 4.60 eV above E F obtained from the energy of the low energy cut-off of the spectra with lower photon energy, while the intensity around the cut-off becomes stronger at hν ≥ E vac because of signals from direct (one-photon) photoemission. When the peak positions in the 2PPE spectra, including a threshold of H (red open triangle), are plotted in Fig. 1b   (a) hν dependences of 2PPE spectra for 1 ML C 60 film on HOPG aligned with E final , where the hν range was 4.13-4.77 eV. (b) Energy positions of 2PPE structures including the onset of H against hν, which are assignable to the slope of "0 hν", "1 hν", or "2 hν". Factors obtained by liner fits of the plots are indicated nearby the data. The dotted line shows the "1 hν" slope by an unoccupied level at E F + 1.8 eV (see text). (c) Total 2PPE yield integrated whole energy region versus hν in which the maximum is corresponding to that optical absorption ( 1 T 1u ) of C 60 film. Note that recovering intensity at the highest hν is due to a contribution of one photon photoemission. fact, the larger symbols are consistently fitted by a slope of 1 hν as a dotted line superimposed in Fig. 1b. Note that, the L1 structure appears more clearly in the 2PPE spectra with a wider acceptance angle ( Supplementary Fig. S1) or in angle-resolved 2PPE ( Supplementary Fig. S3), because it has an electronic resonance with dispersive F0-F2, as discussed below. Including the state at E F + 1.8 eV, the 2PPE spectra unveil seven electronic states; one occupied state located at E F − 2.2 eV (H), three unoccupied states between E F and E vac at E F + 0.7 (L0), + 1.8 (L1), and + 2.7 eV (L2), and three unoccupied states above E vac at E F + 5.8 (F0), + 6.0 (F1), and + 6.5 eV (F2). Although the peak H (red solid triangle) is broadened below hν = 4.6 eV, which is assignable to an occupied state, judging from the plot of the slope aligning with 2 hν, including that of higher energy thresholds (red open triangle).
The series of F0-F2 states show large band dispersions parallel to the surface from an angle-resolved 2PPE measurement, while the others have little dispersion (see Supplementary Fig. S3). The result suggests that F0-F2 are assignable to a series of higher-lying superatomic molecular orbitals (SAMOs) [36][37][38][39][40][41] ; the wavefunctions of SAMOs generally diffuse outside the molecular framework, which would form highly dispersive band structures in a molecular film with a periodic surface structure. As shown in Fig. 1a, F0-F2 are enhanced only around non-dispersive L1 ( Supplementary Fig. S3), suggesting that the observation of SAMOs in 2PPE requires an initial or intermediate electronic state to allow the resonance. It is known that the lowest (1s-) SAMO is experimentally observed at around E F + 3.3-3.8 eV in a C 60 monolayer on noble metal substrates with a strongly delocalized nature 36,[38][39][40] . The absence of the 1s-SAMO in our study is attributable to the 2D electronic structure of the HOPG substrate; since the electronic structures of graphene below and above E F are governed by sp 2 hybridized orbitals oriented parallel to the graphene plane, the chemical interaction with the overlayer material is very weak compared to metallic materials 8 . In fact, the energetics of a C 60 film are almost independent (± 0.1 eV) of the C 60 coverage from 1 to 30 ML (see Supplementary Fig. S4), while metallic substrates considerably perturb the electronic states of C 60 , especially for the first layer, by strong metal-molecule coupling at the interface. Owing to the weak chemical interaction, an electron excitation through the conduction band of the substrate into the unoccupied states of C 60 film would be unlikely. For C 60 on HOPG, electron excitation into the 1s-SAMO is allowed only from an occupied level (e.g. the highest occupied molecular orbital (HOMO)) of C 60 with intra-or inter-molecular excitation. Since the onset of the HOMO is E F -1.8 eV from the ultraviolet photoelectron spectroscopy (UPS) ( Supplementary Fig. S5), the HOMO electron cannot access energetically to the 1s-SAMO with the current hνs (4 -5 eV). Indeed, the 1s-SAMO observed in a 1 ML C 60 film on Ag(111) with a similar hν quickly disappeared on additional C 60 deposition 38 , indicating that injection of electrons from a metal substrate is essential to resolve the 1s-SAMO with the hν range used here.
More importantly, we now focus on the observed energy levels of H and L0-L2, which are essential to understand the electronic and optical properties of the C 60 -HOPG system, rather than higher-lying C 60 SAMOs. Comparing with UPS result (see Supplementary Fig. S5), the peak H is assignable to HOMO of C 60 22,42 .
For the assignments of L0, Dutton et al. had initially assigned it to LUMO of C 60 28 , where they observed very similar 2PPE spectra with the present result for > 2 ML C 60 coverages of Cu(111) substrate. However, they subsequently reassigned the corresponding peak (L0) to the next LUMO (LUMO + 1) involved in Frenkel excitions 29,43 . The main reason of their change in the assignment relied on the energy matching with a theoretical result for the optical transition in solid C 60 44 . They have also argued that the LUMO (exciton) is expected to be located at slightly above E F , which cannot be resolved using their experimental conditions. Another 2PPE group using a pump photon of 3 eV has insisted on other assignments of 2PPE structures; peaks observed at HOMO + 2.2 eV, and + 3 eV are attributed to free photocarrier electron in LUMO, and LUMO + 1, respectively 26,27 . In spite of their previous (re)assignments, judging from our experimental findings on HOPG substrate, it should be more reasonable to assign L0 located at HOMO + 2.9 eV as a photocarrier electron in LUMO for the following reasons. C 60 has a HOMO (h u ) and LUMO (t 1u ), with an optical gap (E opt ) of 1.7-1.9 eV [45][46][47] . The UPS and inverse photoemission spectroscopy (IPES) for a C 60 film have shown that the LUMO is observed at E F + 0.6 eV, which provides the HOMO-LUMO gap (ΔE) of 2.35 eV 48 . The energy difference (ΔE− E opt ) is attributed to on-site Coulomb energy (exciton binding energy), U.
In the 2PPE spectroscopy, occupied states in a cationic final state are probed similar to UPS, whereas unoccupied states reflect a neutral or cationic final state depending on the excitation processes. There are two photoemission pathways via an unoccupied state: (1) When electrons are supplied from a donor state with a pump photon, the molecule can be regarded as an anion, and then the electron occupying originally unoccupied states is emitted by a probe photon. A probe photon of 2− 6 eV can emit the photoelectron mainly from molecular anions in the top layer owing to the limited escape depth of electrons. In fact, the 2PPE spectra for 1 ML C 60 /Au(111), which show a strong molecule-substrate interaction are markedly changed at 2 ML 29 , suggesting that the photoelectron escape depth is around or smaller than the thickness of one monolayer.
The other pathway of (2) is via an intra-molecular excitation, in which a photo-excited e-h pair is localized at a molecule, resulting in a larger excitonic interaction between the electron and hole. After the intra-molecular excitation, the electron is emitted with the probe photon, where there are two channels of photoelectrons (2a) from the exciton and (2b) from photocarrier electrons. The latter is generated by the hot exciton dissociation into a free electron and hole.
For the C 60 in the outermost layer, furthermore, the number of surrounding C 60 molecules is smaller than that inside the C 60 , and then the polarization energy to stabilize a charged molecule becomes smaller for the surface molecules. In other words, the less crowded surroundings result in the increase of U, and the Δ E of the outermost layer is larger than that in the bulk. This effect should be paid attention to when the values are compared among the results of UPS, IPES, and 2PPE spectroscopy.
In bulk C 60 , ΔE has been estimated to be 2.35 eV from the thresholds of UPS and IPES 49 , while ΔE is 2.9 eV in the 2PPE spectroscopy obtained from H and L0 peaks. Based on the assignment of L0, indeed, the successive L1 and L2 peaks are assignable to LUMO + 1 and LUMO + 2 with reasonable energy positions observed by IPES 22 These consistent results show that L0 is assignable to the LUMO of a neutral C 60 , where the photoemission from L0 occurs via process (2b).
It is known that ΔE for a C 60 monolayer on a metal substrate is small compared to that in the bulk C 60 , because the on-site charge is more effectively screened by the substrate than by a C 60 film, resulting in a reduction of U 49-51 . As mentioned above, the energy positions are almost identical regardless of the C 60 coverage on HOPG (Supplementary Figs S1 and S4), which shows that the screening effect from HOPG is much smaller than that from a metal substrate, due to its low density of states near E F . Furthermore, as discussed below, electrons in L0 have a long lifetime of several picoseconds, also indicating that L0 does not originate from the LUMO + 1 involved in Frenkel excitions, as pervious reassignments 29,43 , otherwise it would decay rapidly into lower-lying states within a few hundred fs [52][53][54] . Another 2PPE group has observed a peak at HOMO + 2.2 eV using a 3 eV pump photon, and they have assigned it to the free photocarrier electron in the LUMO 26,27 . However, we believe the assignment is not appropriate, because the peak to peak energy separation of 2.2 eV is too small to assign to a free photocarrier in the LUMO, considering that the ΔE of 2.3 eV defined by the respective onset edges in the UPS and IPES spectra.
Despite the above, it seems reasonable that hot excitons are generated by pump photon in 2PPE spectroscopy. Figure 1c shows the hν dependence of the total 2PPE yield, which reaches a maximum at hν = 4.3-4.4 eV, implying a resonant photo-excitation in the C 60 film. This energy indeed corresponds to the optical absorption of a C 60 film (280 nm in wavelength) 23,55 , assignable to the third optically allowed resonance of 3 1 T 1u 56 marked in Fig. 1c. As seen in Fig. 1a, the intensity of LUMO + 2 at E F + 2.7 eV is much weaker than that of the LUMO and LUMO + 1, also implying that electrons are dominantly excited from the HOMO at E F − 2.2 eV with current hν of 4.8 eV at most; while the contribution of occupied states in HOPG is rather small. Therefore, it can be considered that hot excitons are nascently generated in the C 60 film by a pump photon and subsequently dissociate into a free electron and hole without relaxing into lower-lying excitons. The excited electron temporally occupies LUMOs of surrounding non-excited C 60 , and is finally detected as a photoelectron excited by a probe photon. One may wonder why the hot exciton (3 1 T 1u ) is not resolved as an individual peak in the 2PPE spectra. The absence of the initially-excited hot exciton in 2PPE spectra would imply a very fast (< sub 10 fs) process of hot exciton dissociation, which broadens the spectral feature of the excitonic state (lifetime broadening) 32 . The observed energy levels and the excitation and detection scheme in 2PPE process are schematically illustrated in Fig. 2a,b. Here, it is not exclusive for a charge transfer (CT) state of C 60 + C 60 − to be involved, and the CT state is actually predicted to be located at HOMO + 2.3 eV in theoretical calculations 57 . Formation of the lowest-lying excitons such as S 1 and T 1 would be another pathway of hot electron relaxation that may induce the photoemission by process 2a described above. The central energy of S 1 (0-0 transition) is known to be 2.0 eV, whereas the absorption band of S 1 distributes from 430 to 670 nm in wavelength 58 . Taking into account the energy of the HOMO (E F − 2.2 eV), the higher energy component of S 1 is expected to appear around the low energy cut-off (E F + 0− 0.7 eV in the intermediate energy (E final − hν)) and also to exhibit relaxation dynamics on the ps-ns time scale. However, such a spectroscopic feature in the low energy region and the dynamics could not be observed ( See Figs 1a and 3a shown below), showing that the formation of lower-lying excitonic states is a minor process in the current hν range employed of 4-5 eV. The formation of S 1 and T 1 has been actually reported in 2PPE with different experimental conditions 26,27,30 ; a 3 eV pump photon results in a lower efficiency of photocarrier generation than that with hν ~ 4.5 eV. Furthermore, they employed a higher laser intensity (~sub-μ J/pulse) 26 with low repetition rates (250 kHz) and measured for thicker C 60 film (20 nm ≈ 25 ML), which may favorably induce excitons by paring the free photocarriers in C 60 film due to a high photoexcitation density. It should be emphasized again that the enormous intensity around the cut-off with hν > E vac in Fig. 1a must be due to direct one photon photoemission from just below E F . Figure 3a shows TR-2PPE spectra taken with various pump-probe delays (Δ t), where a spectrum at Δ t = 9 ps is subtracted as a background. The LUMO peak survives even at Δ t = 3 ps, while LUMO + 1 decays more quickly within a few hundred fs. The energy positions of these peaks remain almost the same, which again suggests that free electron photocarriers are generated in the early regime. Otherwise, they would subsequently form lower-lying excitons, (e.g. S 1 or T 1 ), which recombine. Figure 3b shows the intensity of LUMO + 1 against Δ t. The electron lifetime in the level is 48 ± 6 fs, evaluated by a convolution of a Gaussian autocorrelation (AC) (116 fs in full width at half maximum (FWHM)) and a single exponential decay, where the AC is determined from the intensity trace of the TR-2PPE for bare HOPG at the energy of E F + 1.2-1.7 eV; the lifetime of the electron in the intermediate energy is very short (< 20 fs) 59 as compared to the true AC width. As shown in Fig. 3c, interestingly, the signal intensity for LUMO is once again recovered around Δ t = 100-250 fs after the excitation and gradually decreases. Assuming exponential rise and decay processes with time constants of τ r and τ d , the time evolution of 2PPE intensity, I(Δ t), can be described as

TR-2PPE results.
where a and b are weight factors. In addition to the rise and decay processes of LUMO electrons, the second term of the AC component is taken into account, because photoelectrons at the corresponding energy may include two components of (i) coherent 2PPE from occupied states of HOPG and HOMO− 1 of C 60 (see Supplementary  Fig. S5) and (ii) short-lived hot electrons in the HOPG and C 60 film, both of which are not regarded as the photocarrier electrons in the LUMO. The TR-2PPE trace for these components should be very similar to the shape of the true AC for the current pulse duration. Eq. (1) reproduces the evolution of the LUMO with τ r,1ML = 60 ± 9.8 fs and τ d,1ML = 760 ± 24 fs. Note that the rise time τ r,1ML is close to the lifetime of LUMO + 1 electron (48 ± 6 fs), suggesting that the LUMO electrons are mainly supplied from highly excited states via LUMO + 1. The LUMO lifetime on HOPG is significantly longer than those for C 60 on metal substrates 29,43 , e.g. Dutton and co-workers have reported that the lifetime increases with the C 60 thickness on Au(111) from 79 fs at 2 ML to 340 fs at 30 ML, while the corresponding peak has been assigned to a LUMO + 1* exciton 29,43 . Since 2D material substrate of HOPG exhibits weak electronic coupling toward adsorbed species, the decay rate into HOPG is less compared to metal substrates. In fact, the energy positions in the 2PPE spectra for C 60 films on HOPG show little changes with the C 60 coverage ( Supplementary Figs S1 and S4). One should also be concerned about the effect of the excitation density on the TR-2PPE results; the recombination of photocarriers generated at different C 60 molecules (bimolecular recombination) would contribute to the relaxation dynamics. However, the effect is negligible because of the sufficiently low excitation density used in the current study. In the TR-2PPE measurement, assuming a photon flux of ~3 × 10 10 photons·cm −2 at most (76 MHz, 10 mW, and 0.1 mm spot diameter), three photons are irradiated in an area of 10 × 10 nm 2 area per pulse (an area consisting of about a hundred of C 60 molecules). Therefore, the excitation density would be too low to cause bimolecular recombination.
Thickness dependence of LUMO lifetime. Since the sp 2 hybridized nature of the HOPG substrate reduces photocarrier relaxation, the photocarrier hopping rate in a C 60 film can be evaluated by the C 60 thickness dependence of the LUMO lifetime. Figure 4a shows the time evolution of LUMO intensities obtained for various coverages of C 60 films. The recovery of LUMO intensity around Δ t = 100-250 fs becomes unclear at ≥ 2 ML, while the lifetime of LUMO + 1 is independent of the coverage (see Supplementary Fig. S6). As discussed in the previous section for spectral assignments, 2PPE probes the photocarrier electrons remaining in the topmost C 60 layer even for the multilayered films. On the other hand, the photocarrier electrons are generated by the pump photon penetrating into whole C 60 layers. When the photocarrier electrons at the topmost layer initially originate not only from the topmost layer but also at the deeper layers inside, the rise time behavior should become vaguer in the multilayer films, as seen in Fig. 4a. The result therefore indicates that the created photocarriers move around in the C 60 film with a finite hopping rate. To evaluate the photocarrier hopping rate in the C 60 film, the thickness dependence of the decay phenomena was modelled for the LUMO electrons.
As shown in Fig. 4a, the LUMO lifetime lengthens with the thickness, and the fitting with eq. (1) shows that τ d for 1, 2, 3, and 5 ML are 730 ± 22, 1520 ± 120, 2060 ± 270, and 2510 ± 210 fs, respectively. With increases in the C 60 thickness, a uniform layer-by-layer structure grows without the presence of non-uniform islands (see Supplementary Fig. S7). Since the lifetime of a free electron carrier in solid C 60 is much longer (ns or μs) than the observed τ d 60 , the decay of photocarriers mainly occurs at the first C 60 layer interfaced with the substrate. When the photocarrier electron hopping rate from an original C 60 site to the nearest neighboring C 60 is defined as k C60 , the rate to the specific neighboring C 60 should be 1/12 k C60 , because of the fcc crystal structure of a C 60 film exhibiting a coordination number of twelve. Considering that the fcc's C 60 (111) surface faces to the vacuum, the interlayer hopping rate from original C 60 site at the i th layer to the i + 1 or i − 1 th layer is 3/12 k C60 each (six neighboring C 60 molecules exist at the i th layer). In addition, it can be assumed that the interlayer hopping from top and bottom layers are modified to be 3/9 k C60 owing to the absence of a layer to one side. Based on these concepts, here, the electron population of each layer, N n,m , is evaluated, in which n is the total number of C 60 layers, and m is the targeted C 60 layer (1 ≤ m ≤ n). First, we simply evaluate the value of k C60 for the result for 2 ML: the number of LUMO electrons at the first interface layer (N 2,1 ) and the top layer (N 2,2 ) can be simply described by where k d is the decay rate of the photocarrier electron density by relaxing into the substrate from the first C 60 monolayer, i.e. k d = 1/τ d,1ML = 1.37 × 10 12 s −1 . Since the irradiated photon penetrates through the C 60 films, it is reasonable to assume that all of C 60 layers are uniformly photoexcited as N 2,1 = N 2,2 = 1 at t = 0, whereas the time trace of the TR-2PPE reflects the photocarrier at the top layer (N 2,2 ). Then, the above differential equations can be solved numerically for the variable k C60 . The blue solid line in Fig. 4b is the simulated result for N 2,2 , where k C60 is evaluated to be 3.3 ± 1.2 × 10 13 s −1 . So far, the electron hopping rate in unoccupied energy levels has been evaluated by analysis of the resonant autoionization spectrum which has been established as core-hole-clock-spectroscopy 61 .
In this method, however, the accessible frequency range of 10 16   1.7 × 10 14 s −1 , while they could not evaluate it for the LUMO 62 . The difference in the hopping rates depending on the energy levels can be explained by the width of the density of states 63 ; the hopping rate would become higher for a wider band width. From the resonance window of LUMO + 1 (~0.5 eV) with SAMOs (supplemental material, S3), the band width is larger than that of LUMO (~0.2 eV in FWHM).
The above simulation can also be applied to the result for 3 ML. By using the central value, 3.3 × 10 13 s −1 , for k C60 as obtained above, however, the simulation for 3 ML slightly overestimates the experimental data (not shown). This could be caused by the polarization effect; owing to the reduction of U at the topmost C 60 , it can be qualitatively deduced that the hopping rate of electrons from an interior C 60 layer to the topmost layer should be smaller than that from the topmost to an interior layer. Indeed, the result for 3 ML could be well simulated when the former and latter rates are modified to be 2.1 × 10 13 and 4.5 × 10 13 s −1 (pink solid line in Fig. 4b), respectively, as upper and lower limits of the error for k C60 (± 1.2 × 10 13 s −1 ). The reasonable modification falling within its uncertainty suggests the validity of our interlayer hopping model where hopping within an in-plane intralayer does not contribute explicitly. Note that the simulation for 5 ML (N 5,5 ) overestimates the experimental result (τ d,5ML ), possibly due to thickness inhomogeneity (see Supplementary Fig. S7 for scanning tunneling microscopy images) that could open a decay channel through an edge state of C 60 layers.
We think that the hopping rate obtained with the above analysis is intrinsic for organic species, but independent of the substrate. The analysis will therefore be also applicable universally for the determination of k d at the interface between an organic film and a metallic substrate, where k d is typically too large to be resolved experimentally because of the strong substrate-molecule interaction.

Summary
We observed the hot exciton dissociation and decay dynamics of generated photocarrier electrons in a C 60 film on a HOPG substrate through the use of TR-2PPE spectroscopy. Unlike metal substrates, the 2D material substrate of HOPG substrate suppresses the electronic interaction with overlayered C 60 film, which enables us to detect clear time required for photocarrier generation within a hundred fs. The time evolutions of the 2PPE intensity and their C 60 thickness dependence reveal that the generated photocarriers randomly and frequently hop in the film until decaying into the substrate, where the electron hopping rate in C 60 is independently and directly evaluated to be 3.3 ± 1.2 × 10 13 s −1 . The results have clarified fundamental aspects of photocarrier generation and carrier transportation in organic materials with time-domain measurements, and will provide a key to improving the performance of graphene-based organic devices.

Methods
Sample preparation and 2PPE measurements. Prior to C 60 deposition, an HOPG substrate was cleaved in air and cleaned by heating (670 K, 50 h) in an ultrahigh vacuum (UHV) chamber, where the base pressure was better than 1 × 10 −8 Pa. C 60 was then deposited in the UHV system with a rate of 0.07 ML/min, monitoring with a quartz microbalance. The C 60 coverage was calibrated from 2PPE measurements (Supplementary Figs S1 and S2). TR-2PPE measurements were carried out using the third harmonics (hν = 4.13-4.77 eV, p-polarization) of a Ti:sapphire laser (~100 fs, 76 MHz), which was separated into pump and probe photons with a beam splitter. Both pump and probe photons are recombined at the sample in UHV with a concave mirror (f = 400 mm). In this optical configuration ("skew configuration") the autocorrelation of incident photons can approximate a Gaussian peak function without interference oscilation 31 . Photoelectrons emitted in the normal direction were detected by a hemispherical electron energy analyzer (VGSCIENTA: R3000), where the energy of photoelectron was calibrated by UPS of an Au plate providing a clear Fermi edge as zero binding energy at E final = E F + hν (21.22 eV) for E final (Supplementary Fig. S5). A sample bias of − 3 V was applied to collect low-energy photoelectrons around the cut-off energy (E vac ). Energy and time resolutions of the TR-2PPE system were about 20 meV and 20 fs, respectively. The sample temperature was 293 K during the TR-2PPE measurement.