WorksPhysics and Astronomy
Bond Anharmonicity in a Caged Molecule Probed by Vibronic Core Spectroscopy
Abstract
X-ray absorption spectroscopy and resonant inelastic X-ray scattering at the nitrogen K-edge show that the NN bond in N@C is substantially more anharmonic than in free N, with the ground-state () anharmonic constant a factor of two larger (~meV versus ~meV) despite an essentially unchanged harmonic frequency. The same enhancement is found across all three vibronic potentials accessible to the experiment—the electronic ground state, the core-excited intermediate state, and the spectator-decay final state—pointing to a common origin rather than an effect specific to one electronic configuration. High-level coupled-cluster calculations of the N@C potential energy surface, however, find no significant change in anharmonicity upon encapsulation, ruling out a simple static distortion of the NN bonding potential by the fullerene cage. We propose instead that the dramatic increase in above the gas-phase value – which, taken literally, would imply an unphysical halving of the N dissociation energy - reflects the breakdown of the single-molecule Morse description, as previously invoked to explain anomalous anharmonicities in N clusters, and supported by the observation of shifts in the fullerene host's Raman modes.
N2C60.pdf
Endohedral fullerenes enclose an atomic or molecular guest within a hollow carbon cage, providing a unique form of nanoscale confinement that isolates the encapsulated species from its surroundings while leaving it chemically intact [1, 2, 3]. For species that interact only weakly with the buckminsterfullerene cage, the ground-state electronic structure is comparable to that in the gas phase – there is minimal orbital hybridisation with, or charge transfer to, the surrounding buckyball [2, 3, 4, 5, 6]. The intramolecular degrees of freedom are therefore well isolated, providing a novel platform to study the coupling of a localised quantum state to the external environment [7, 8]. A prominent example is N@C₆₀, in which the encapsulated nitrogen atom retains its atomic quartet ground state () [9, 10] and exhibits an exceptionally long electron-spin coherence time [11, 12], motivating proposals for endofullerene-based qubit architectures [13–15].
The extent to which the intragage isolation is dynamically complete, however, remains an open question. The anharmonicity of the guest’s intramolecular potential is, in principle, a sensitive probe of the host-guest coupling dynamics: it captures the variation in the bond potential away from its minimum, and is therefore responsive to interactions that can leave the static electronic structure essentially unperturbed. Confinement effects – in particular, the quantisation (and coupling) of rotational and translational motion – have been studied extensively by inelastic neutron scattering [16, 17], terahertz [18, 19] and infrared spectroscopy [20, 21], nuclear magnetic resonance (NMR) [2, 3] [22, 23], and via computational and analytical calculations [24–26]. Moreover, it is now well established that the guest-cage interaction potential can have significant anharmonic character [18, 19, 27]. By contrast, the anharmonicity of the intramolecular potential of the encapsulate itself has received comparatively little attention, and its sensitivity to the host-guest coupling dynamics has not, to the best of our knowledge, previously been explored experimentally.
Smucker and Pérez-Ríos Smucker and Pérez-Ríos [28] recently calculated the spectroscopic constants for N₂@C₆₀ from a rotationally-averaged potential energy surface, finding only a modest reduction in the first anharmonic correction term, , relative to the free molecule. Our own frozen-cage coupled-cluster calculations (Supp. Mat.) likewise find no significant change in anharmonicity upon encapsulation. High-level quantum chemical theory therefore predicts minimal change, at most, in the anharmonic constant. As we show here, however, this is at odds with the results of vibronic core spectroscopy of N₂@C₆₀. We suggest that the discrepancy between theory and experiment arises from coupling between the N≡N stretch and the modes of the C₆₀ host: coupled oscillators do not retain the vibrational levels of their isolated components [29–31].For N₂@C₆₀, this implies that the anharmonicity observed in core spectroscopy is a property of the coupled guest–host system, not solely of the N≡N bond. Near-edge X-ray absorption fine structure (NEXAFS) spectroscopy and resonant inelastic X-ray scattering (RIXS) provide element-specific access to the vibronic structure of the guest across multiple electronic states: by tuning to the N K-edge, the response of the encapsulated N is isolated from that of the C host. This combination of element specificity and access to excited-state potentials is not available from infrared, Raman (see End Matter) or neutron-based techniques. N@C is particularly well suited to this approach: the nitrogen molecule is closed-shell with no orbital hybridisation with the cage [28, 32], and the N≡N triple bond produces well-resolved vibronic features that have allowed the intramolecular potential to be determined with high precision in both gas-phase and adsorbed N [31, 33–36].
N@C (Figure 1(a)) was synthesised using molecular surgery [37] and deposited as thick films/multilayers onto Si(111) or Ag(111) substrates via thermal sublimation under ultrahigh vacuum conditions (Supp. Mat.). RIXS measurements were acquired using the AERHA spectrometer of the SEXTANTS beamline at the SOLEIL synchrotron [38, 39]. The incident photon energy was scanned through the N 1s → π* resonance (Figure 1(b)), and the emitted-photon spectrum (spanning a 12 eV range below the elastic line) was recorded at each excitation energy. (We adopt a localised core-hole description throughout and denote the core-excited state as 1sπ*.)

Figure 1. N K-edge NEXAFS of N@C. (a) Structure of N@C (C, purple; N, blue). (b) Energy-level diagram for the N 1s → π* resonance (black solid arrow) and a decay to the electronic ground state (dashed arrow) with the emission of an X-ray photon of energy . The molecule may be left vibrationally excited, so that, in this case, the energy loss equals the vibrational excitation energy of the state. Other decay channels that leave the molecule electronically excited are discussed below. (c) Fluorescence-yield (FY) NEXAFS spectrum fitted with Voigt components assigned to vibrational levels of the 1sπ* core-excited state. Lower panel: residuals; grey band: ±2σ. Upper inset: total-electron-yield spectrum of a thick N@C film and the spectrum of gas-phase N, both measured at I09 (Diamond Light Source) and aligned at . Lower inset: fitted peak energies, with Morse fit (solid line) and the progression expected for the core-excited state of gas-phase N (green dashed line).
Complementary NEXAFS spectroscopy (upper inset to Figure 1(c)) and core-level and valence-band photoemission (Supp. Mat.) were carried out at Beamline I09 at the Diamond Light Source [40]. All NEXAFS and RIXS spectra acquired at SOLEIL and the Diamond Light Source were calibrated by aligning the fitted 0-0 component of the N 1s → π* vibronic progression to the literature gas-phase value of 400.88 eV [35, 36]. Further details of sample preparation and photoemission characterisation are provided in the Supplemental Material.
The upper inset of Figure 1(c) is a comparison of nitrogen K-edge NEXAFS spectra measured for a multilayer of N@C (see Supp. Mat. for photoemission spectra of this sample) and for gaseous N, both acquired on the same beamline (I09, Diamond Light Source) where the additional anharmonicity for N@C is clear from the lower energies of the higher-lying vibronic peaks. The nitrogen K-edge NEXAFS spectrum shown in the main panel of Figure 1(c) was obtained from a RIXS map (at SEXTANTS) by integrating the emitted-photon intensity over the measured emission-energy window at each incident photon energy [41, 42]. A clear vibronic progression associated with the 1sπ* core-excited state is again observed. We have fitted the spectrum using a series of Voigt components assigned to successive vibrational levels of the 1sπ* state. In the lower inset to Figure 1(c) open circles show the fitted vibronic peak positions plotted as a function of , the vibrational quantum number, while the solid line shows the Morse energy
eigenvalue expression [43, 44]
fitted to those points and plotted continuously as a guide to the eye. Here, is the minimum of the potential, is the harmonic vibrational term, and is the anharmonic correction. Fitting the NEXAFS peak positions as a quadratic function of yields a harmonic energy of meV and an anharmonic constant, , of meV. The harmonic term is close to the gas-phase value for the core-excited state, while the anharmonicity is substantially larger than the gas-phase value of approximately 2.1 meV [36]. For comparison, the inset also shows the vibrational level spacings calculated from the known gas-phase harmonic and anharmonic constants [35, 36] (dashed green line).
Figure 2 shows the complete nitrogen K-edge RIXS response of as the incident photon energy is scanned through the N 1s resonance. At higher energy loss (Figure 2(a)), a set of features is observed due to the spectator-decay channel. In this process, the core hole that is present for NEXAFS measurements is filled while the electron promoted into the orbital remains in the valence shell. This leaves the molecule in a valence-excited final state. In contrast to the dispersive spectator features, the electronically elastic decay channel (Figure 2(c)) returns the molecule to the electronic ground state (dashed arrow in Figure 1(b)), producing a vibrational progression at approximately fixed energy loss. Figure 2(b,Figure 2d) are the corresponding simulated RIXS maps for the spectator- and ground-state-decay features, calculated using the Kramers-Heisenberg method described in the Appendix.
Figure 2. Nitrogen K-edge RIXS maps and Kramers–Heisenberg model of . Experimental RIXS intensity maps are shown for (a) the spectator-decay channel and (c) the ground-state energy-loss region, with incident photon energy scanned through the N 1s resonance. The corresponding Kramers–Heisenberg (K-H) simulations are shown in (b) and (d), calculated using Morse potentials for the ground, core-excited, and final electronic states.
Integrating the RIXS intensity over the incident-energy axis yields one-dimensional energy-loss projections (Figure 3). As the spectator-decay projection (Figure 3(a)) is dispersive, individual vibronic components comprising the spectator envelope are shifted in energy loss with incident photon energy. The projection is therefore constructed from the K-H model as a sum of NEXAFS-weighted vibronic components evaluated at their dispersion-shifted positions.

Figure 3. Energy-loss projections of the spectator-decay and ground-state RIXS channels. (a) Projection of the spectator-decay region, obtained by summing the RIXS intensity over the incident-energy range of the N 1s → 1π* resonance. The experimental spectrum is compared with the total Kramers–Heisenberg projection and the individual NEXAFS-weighted slices used to reproduce the dispersive spectator-decay envelope. (b) Projection of the ground-state decay region, showing the vibrational progression of the X state. The fitted inelastic peaks correspond to increasing vibrational excitation of the N≡N stretch. Inset: fitted ground-state vibrational excitation energies plotted against , with the solid line showing the Morse-level fit used to extract and . Green dashed line: gas-phase N.
By contrast, the ground-state projection (Figure 3(b)) involves an electronically elastic channel that is non-dispersive so that each peak falls at an approximately fixed energy loss; the vibrational level spacings of are therefore visible directly in the integrated spectrum. The vibrational constants extracted from the K-H analysis are listed in Table I. Note that while the harmonic vibrational frequency is essentially unaffected, there is a pronounced increase in the anharmonic term. For the electronic ground state, the fitted value of 3.8 ± 0.3 meV is more than twice the gas-phase value of approximately 1.8 meV [36, 45, 46]. The same trend is observed for the excited states.
To investigate the microscopic origin of the enhanced anharmonicity, the N≡N potential energy surface was computed for both free and -encapsulated using the domain-based local pair natural orbital coupled-cluster method, DLPNO-CCSD(T)/def2-SVP [47–49] (Supp. Mat.). Although the calculations reproduced the harmonic frequency of free to within a few percent of the experimental value, neither the equilibrium nor the displaced configurations yield any significant increase in anharmonicity upon encapsulation. This insensitivity of the computed anharmonicity to the guest–host geometry, combined with the near-unchanged seen in both experiment and calculation, suggests that the enhanced anharmonicity is not driven by a static modification of the N≡N potential by the cage environment.
Since [31, 43, 50], a doubling of with an essentially unchanged implies a halving of the dissociation energy. This is clearly physically untenable given the lack of hybridisation of and orbitals [28, 32]. As such, the factor of two increase in the anharmonicity constant for encaged as compared to that of its gas-phase counterpart cannot reflect a genuine static modification of the N≡N bond potential by the cage environment alone, and must instead arise from the breakdown of the single-molecule Morse description when applied to a coupled guest–host system. Anomalously large changes in have similarly been reported previously for nitrogen clusters: Doronin et al. [31] observed that free clusters exhibit a ground-state anharmonicity approximately half that of the gas-phase single-molecule value, with essentially unchanged. Drawing on the work of Redlich [29] and Heller and Gelbart [30], Doronin et al. attributed the “anomalous” anharmonicity constants they measured to interactions between neighbouring N≡N bonds within a coupled-oscillator framework. Our observation of dramatically altered (albeit increased rather than decreased) constants with near-unchanged is reminiscent of that of Doronin et al. [31] and similarly points to a breakdown of the Morse description when the N≡N stretch is coupled to additional degrees of freedom.
Direct experimental evidence of coupling between the encaged molecule and the host is provided by
Raman spectroscopy (Figure 4 and Table II, End Matter).

Figure 4. Micro-photoluminescence and Raman spectroscopy of C and N@C. (a) Normalised micro-photoluminescence spectra of bulk C and N@C, shown in grey and red, respectively. (b) Corresponding micro-Raman spectra over the full measured Raman-shift range. The highlighted regions indicate the H(1) and A(2) cage modes enlarged in panels (c) and (d), respectively. (c) Enlargement of the low-frequency H(1) mode, which shifts from 271.8 cm in C to 273.5 cm in N@C, corresponding to a blue shift of = +1.7 cm. (d) Enlargement of the high-frequency A(2) mode, which shifts from 1467.6 cm in C to 1465.4 cm in N@C, corresponding to a red shift of = −2.2 cm. The spectra are normalised and vertically offset for clarity. The mode-dependent shifts demonstrate that encapsulated N measurably perturbs the vibrational potential of the fullerene cage, providing evidence for guest–host coupling.
| Mode | C / cm | N@C / cm | / cm |
| H(1) | 271.8 | 273.5 | +1.7 |
| H(2) | 431.8 | 437.5 | +5.7 |
| A(1) | 495.7 | 495.9 | +0.2 |
| G(2) | 567.2 | 569.4 | +2.2 |
| H(4) | 772.1 | 771.7 | −0.4 |
| H(5) | 1099.6 | 1098.6 | −1.0 |
| H(6) | 1250.1 | 1246.0 | −4.1 |
| H(7) | 1422.7 | 1419.4 | −3.3 |
| A(2) | 1467.6 | 1465.4 | −2.2 |
| H(8) | 1573.7 | 1572.4 | −1.3 |
Table II. Assigned Raman-active cage modes of empty C and N@C.
Several low-frequency cage modes are blue-shifted relative to their positions for the empty C molecule; higher frequency modes are red-shifted. These shifts — in opposite directions for modes of radial and tangential character, consistent with observations for Ar@C [51] — provide strong evidence for cage-guest coupling. By the symmetry of any coupling Hamiltonian, the cage modes must also perturb the N≡N vibrational states (in a manner that is invisible to the frozen-cage calculations).
Importantly, the coupling implied by the Raman shifts need not be large to produce the observed apparent anharmonicity enhancement. The anharmonicity constant is determined by the second finite difference of the vibrational level energies. A perturbation to the level energies modifies the fitted anharmonicity by
which depends only on the curvature of with respect to . For a perturbation of the form — the simplest form with non-zero curvature, and physically natural for any coupling to an external degree of freedom that grows with vibrational amplitude — the second finite difference is , giving . To double therefore requires meV. At , this corresponds to a perturbation of meV, which is less than 2% of meV — entirely consistent with a weak van der Waals-mediated guest-host interaction [28, 32]. A fully coupled quantum treatment of the guest-host vibrational eigenstates — analogous to that developed by Felker and Bacić [25] but extended to resolve the anharmonic structure of the vibrational ladder — is needed to account in full for our observations.
| State | (meV) | (meV) |
| (ground) | ||
| (core-excited) | ||
| (spectator) |
Table 1. Harmonic and anharmonic constants for ground, core-excited, and spectator states. Vibrational constants were obtained from the projection-constrained fitting procedure described in the End Matter section. The ground-state values were extracted from the near-elastic energy-loss projection, the core-excited values from the NEXAFS spectrum (Figure 1), and the spectator-state values from the spectator-decay projection and full K-H map.
Acknowledgements. We gratefully acknowledge both the SOLEIL synchrotron and Diamond Light Source (DLS) for the provision of beamtime via awards 20230252 and 99240058 (SOLEIL) and allocations SI-31574-1 and SI31574-2 (DLS). C.F. also thanks DLS for the award of PhD studentship STU0425. Endofullerene synthesis was supported by EPSRC grants EP/P009980/1 and EP/T004320/1. P.M. thanks the EPSRC for an Established Career Fellowship (EP/T033568/1). Python code for the handling, fitting, and plotting of RIXS, NEXAFS, and photoemission data was developed with the assistance of Open AI’s ChatGPT (ver 4.0 and 4.5) and Anthropic’s Claude Opus 5.5. The authors specified the analysis, checked the code and its outputs against the raw data, and verified all reported values.
Data availability—The data that support the findings of this article are openly available at https://repository.nottingham.ac.uk/handle/123456789/96741 (DOI: https://doi.org/10.17639/nott.47998.)
Kramers–Heisenberg modelling.
To extract quantitative information from the RIXS maps, the spectra were modelled using a Kramers–Heisenberg description of resonant scattering [36, 52, 53]. In this formalism, RIXS is treated as a second-order process in which the system is first excited from the electronic ground state into a core-excited intermediate state and then decays into an allowed final state. Within the Franck–Condon approximation, the electronic and nuclear parts of the transition matrix element can be separated [54, 55]. Each electronic state is therefore described by its own potential-energy curve, supporting a corresponding ladder of vibrational levels and nuclear wavefunctions. Through vibronic coupling, the change in electronic configuration during excitation and decay allows the system to be projected onto different vibrational levels and electronic states. These vibrational-state populations modify the emitted-photon energies and intensities, producing the features observed in the RIXS maps. Since the model considers a single dominant electronic excitation and decay channel, the electronic contribution can be absorbed into an overall intensity factor, leaving the relative RIXS intensities to be determined by Franck–Condon transitions [36, 44].
For each final vibrational state , the state-resolved scattering intensity was calculated from the coherent sum over intermediate vibrational levels [36, 52],
where is the incident photon energy, is the initial vibrational state on the electronic ground-state potential, denotes vibrational levels of the core-excited intermediate-state potential, and denotes vibrational levels of the final-state potential. The terms and are Franck–Condon overlaps between the corresponding nuclear wavefunctions, while accounts for the finite lifetime of the core-excited state.
The full RIXS intensity map was then obtained by placing each state-resolved contribution at the corresponding energy loss,
The Dirac delta function enforces energy conservation by placing each final vibrational state at its corresponding energy loss. In the numerical model used to generate the maps in Figure 2, this delta function was replaced by a peak broadening function, while the finite incident-energy resolution was included by convolution along the excitation-energy axis. In this form, Eq. (3) retains the coherent interference between pathways through different intermediate vibrational levels, while Eq. (4) maps the resulting final-state intensities onto the energy-loss axis of the calculated RIXS spectrum.
The values used to construct the Kramers–Heisenberg maps in Figure 2 are listed in Table I. These parameters were obtained through a multistep fitting procedure rather than by a direct unconstrained fit to the full two-dimensional maps. In brief, the RIXS-derived NEXAFS spectrum and the energy-loss projections of the ground-state and spectator-decay regions were first used to determine the vibronic peak positions, widths, and relative energy offsets. The resulting parameters were then used to generate the full Kramers–Heisenberg simulations shown in Figure 2.
The discrete vibronic transitions generated by the Kramers–Heisenberg model were broadened to account for the experimental resolution before comparison with the data. In the incident-energy direction, the linewidth reflects both the finite lifetime of the core-excited state and the excitation-energy resolution, and was therefore described using a Voigt profile. In the energy-loss direction, the linewidth is dominated by the spectrometer response. This response was also modelled empirically using a Voigt function. This does not imply an additional lifetime-broadening contribution along the energy-loss axis; rather, the Voigt profile provides a flexible description of the non-trivial spectrometer response, which is not captured adequately by either a purely Gaussian or purely Lorentzian lineshape [56–58]. The same broadening convention was used for the RIXS-derived NEXAFS spectrum and the full RIXS maps.
Mode coupling evidenced through Raman spectroscopy.
Micro-photoluminescence (PL) and micro-Raman spectroscopy (RS) measurements were performed on bulk C and N@C using a commercial optical spectrophotometer (LabRAM Evolution HR, HORIBA, France). The instrument was equipped with a 633 nm He–Ne laser (Pacific Lasertec, USA) and a 785 nm semiconductor diode laser (Innovative Photonic Solutions, USA), operated at incident powers of 0.07 mW and 0.6 mW, respectively. The scattered light was collected using a 100×, 0.9 numerical-aperture metallurgical objective (Olympus, Japan) and dispersed using a 600 lines mm diffraction grating. Edge filters with a low-frequency cut-off of approximately 50 cm were employed, and the confocal pinhole was set to 200 m. The single-pixel spectral dispersions at zero Raman shift were better than 0.8 and 0.5 cm for the 633 nm and 785 nm configurations, respectively. The spectrometer was calibrated before measurement using a Standard Pro-Raman Calibration Objective. Samples were dispersed in methanol and deposited by drop casting.
Direct experimental evidence of the interaction between the encapsulated N molecule and the C host is provided by the Raman spectroscopy data presented in Figure 4 and Table II. Several low-frequency cage modes are blue-shifted following encapsulation, including the H(1) and H(2) modes, by 1.7 and 5.7 cm, respectively. In contrast, several higher-frequency modes are red-shifted, including H(6), H(7) and A(2), by 4.1, 3.3 and 2.2 cm, respectively. These mode-dependent changes demonstrate that the presence of N measurably modifies the cage vibrational potential and provide evidence for guest–host coupling.
Supp_info.pdf
Supplemental Material for:
Bond Anharmonicity in a Caged Molecule
Probed by Vibronic Core Spectroscopy
SAMPLE PREPARATION
Milligram quantities of powder, synthesised via molecular surgery[1] with subsequent purification by HPLC and sublimation, were provided by the Southampton group (Whitby et al.) For the NEXAFS measurements at the Diamond Light Source (beamline I09) shown in Fig. 1(c) of the main paper, the endofullerene sample was prepared as follows. Following degassing for a minimum period of twelve hours, a multilayer film of was grown by sublimation at a nominal evaporator temperature of K onto a clean Ag(111) surface held at 180 K. (Although not discussed here, submonolayer and monolayer coverages of that produced a low energy electron diffraction (LEED) pattern were also deposited and spectroscopically analysed prior to the multilayer studies.)
We note that although the Ag(111) surface prior to the deposition of was free of an O 1s photoemission signal, in the initial stages of the first deposition of a small oxygen contribution was observed, which we attribute to the presence of interstitial water and/or oxidised fullerene at the surface of the material in the evaporator crucible. Growth of a sufficiently thick film “buried” the oxygen contaminant such that it was located outside the escape depth of the photoelectrons. Alternatively, evaporating off the multilayer (by heating the Ag(111) substrate) followed by sputter-anneal cycles restored a clean, fullerene-free Ag(111) surface. Subsequent depositions of from the same evaporator showed no evidence of the contaminant. To circumvent beam damage, all measurements at I09 were made at a sample temperature of 180 K and with the undulator flux detuned by approximately a factor of ten from the peak output.
A different sample preparation strategy was used for the RIXS measurements at SOLEIL (SEXTANTS beamline). In that case, samples were prepared ex situ and transported to the beamline end-station through air. To ensure that the films were microns thick at a minimum, all of the mg source material was sublimed from the evaporator. The samples were subsequently transferred into the SEXTANTS end-station UHV chamber (base pressure mbar) and cooled using a liquid-helium flow cryostat to approximately 40 K during the measurements. For the data shown in Fig. 2 of the main paper four RIXS maps were acquired and averaged to improve signal-to-noise ratio.
VALENCE BAND AND CORE-LEVEL PHOTOEMISSION SPECTROSCOPY
Core-level and valence-band photoelectron spectra were recorded at the soft X-ray branch of beamline I09 at the Diamond Light Source [2], using a Scienta EW4000 hemispherical analyser. Binding energies are referenced to the Fermi level of the Ag(111) substrate (extracted from a fit of a Fermi-Dirac function). Rough estimates of film thickness based on the ratio of C 1s to Ag 3d intensity suggest that the valence band spectrum shown in Figure S1 arises from a film that is of order 4 to 5 layers thick. Given the substrate temperature during deposition (180 K), the growth was unlikely to be ideally layer-by-layer, however. In line with this suggestion, the lack of a “drop to baseline” between the HOMO and HOMO-1 peaks in Figure S1 arises from photoemission from the Ag(111) substrate. Nonetheless, the valence band spectrum is wholly characteristic of the C cage and, as for many other endofullerenes synthesised by molecular surgery[3, 4, 5], shows that encapsulation leaves the frontier orbitals of the cage essentially unchanged.
Figure S1. Photoelectron spectroscopy of a multilayer film of on Ag(111). The main panel is the valence band spectrum measured with a photon energy of 110 eV. The prominent, characteristic leading features at eV and eV below the Fermi level (0 eV binding energy) arise from the highest occupied molecular orbital (HOMO) and HOMO-1 of the cage. Lower inset: 1s shake-up features. The shake-up spectrum is essentially indistinguishable from that of empty . Upper inset: Open circles: 1s core-level spectrum measured with a photon energy of 520 eV. A single-component Voigt fit is shown as a red line.
N 1s
The N 1s spectrum (Figure S1, upper inset; ) consists of a single, symmetric line, well described by one Voigt component on a linear background. The fit gives a binding energy of 403.5 0.1 eV and a FWHM of 0.65 eV (Gaussian 0.47 eV, Lorentzian 0.29 eV). The two N atoms of the encapsulated molecule are therefore chemically equivalent, consistent with negligible orbital mixing with the cage. Assuming a typical work function of 4.6–4.8 eV for a C multilayer on Ag(111), the vacuum-referenced N 1s binding energy is 408.2–408.4 eV. This is 1.5–1.7 eV below that of free N ( 409.9 eV), a shift consistent with polarisation screening of the core hole by the surrounding cage and film.
C 1s shake-up spectrum
The C 1s shake-up spectrum (Figure S1, lower inset) for N@C shows a number of well-defined satellite features at energies above the bulk C 1s line. As we have previously discussed in the context of noble gas endofullerenes[5], these coincide with the shake-up energies of empty C[6, 7]. Because the satellite energies directly reflect excitations between the cage’s frontier orbitals, this agreement, together with the valence-band spectrum, again indicates that the electronic structure of the cage is not measurably perturbed by the encapsulated .
We also note that the 1s spectrum was recorded at an earlier, thinner stage of deposition than that from which the 1s spectrum was acquired. Its main line contained, in addition to the bulk component, a component shifted by eV to lower binding energy arising from first-layer molecules chemisorbed on Ag(111) (see, for example, the SI for [4] and references therein). This interface contribution would at most broaden the lowest-energy satellite and does not affect the comparison with empty .
COUPLED-CLUSTER CALCULATIONS FOR AND
We analysed using high-level coupled-cluster calculations in order to obtain a more accurate description of the potential energy surface than was previously possible with DFT methods.
Isolated molecule
First, we constructed the PES for isolated at several levels of theory and analysed the data using both Morse and polynomial fits (Table SI). Although the Morse fits visually reproduced the PES curves well, we found that the extracted vibrational parameters depended strongly on the fitting range. We therefore decided to proceed with a polynomial fit. With the polynomial fits, we could monitor convergence by increasing the polynomial order. In the fit we kept the number of PES points fixed, while varying the degree of the polynomial expansion. Each value of corresponds to a polynomial of order . The polynomial fits were constructed such that the lower-order coefficients, and , remained close to the values obtained from the initial Morse estimate. This constraint was important because unconstrained high-order polynomial fits were very flexible and produced unphysical values of and . As a result, the converged values matched experiment reasonably well, which gave us more confidence in the reliability of the fit.
The CCSD(T)/aug-cc-pVQZ level of theory produced results that were in the closest agreement with the experimental data. However, calculations for at this level of theory become computationally demanding. Therefore, we used the DLPNO-CCSD(T)/def2-SVP approach, which still provides a reliable approximation and is considered a high-level quantum-chemical approach.[8]
Using DLPNO-CCSD(T)/def2-SVP we calculated the PES for encapsulated within for three orientations, aligned with the cage’s , and axes (Figure S2). In all DLPNO-CCSD calculations, the cage was frozen and only the N-N bond length was varied to trace out the PES. Although this takes into account any coupling in the electronic structure between and the cage (e.g. charge transfer or hybridisation), it cannot account for vibrational coupling. A full vibrational coupling at the DLPNO-CCSD level (involving VSCF/VCI based on mode-coupled PES generated by the Adaptive Density-Guided Approach (ADGA) with DLPNO-CCSD as the underlying electronic-structure method) is currently too computationally expensive to run. Hence, our mitigation was to physically move the encapsulated molecule to different positions and orientations inside the cage to probe for any vibrational coupling effects. Thus, to investigate whether stronger cage–molecule interactions could induce noticeable changes in the harmonic frequency or the anharmonicity, we placed the molecule closer to the fullerene wall in three orientations (, , and ) (Figure S2) and also calculated the PES for this configuration (Table SII). Figure S3 shows the PES calculated at the DLPNO-CCSD(T)/def2-SVP level of theory together with the fitted (a) Morse potentials and (b) polynomial fits for @ in the , , and orientations (left-hand side), and for positioned close to the fullerene wall in the , , and orientations (right-hand side). All calculations were performed using the DLPNO-CCSD(T)/def2-SVP method.

Figure S2. Tested structures of .

Figure S3. Potential energy surfaces (PES) for N@C calculated at the DLPNO-CCSD(T)/def2-SVP level of theory together with fitted (a) Morse potentials and (b) polynomial fits. The left panels show the , , and symmetry orientations of the fullerene cage, while the right panels correspond to configurations in which the N molecule is positioned close to the fullerene wall along the , , and orientations. Again, both fits appear to be good; however, the vibrational parameters calculated using the polynomial fit (Table SII) were more consistent than those obtained from the Morse fit.
Different symmetry orientations of the fullerene cage do not significantly affect either the vibrational frequency or the anharmonicity. If anything, the orientation in the structures
| Method | (cm) | (cm) | ||||
| Morse | Polynomial | Morse | Polynomial | Morse | Polynomial | |
| DLPNO-CCSD/def2-SVP | 2463.19 | 2437 23 | 13.65 | 13.80 0.3 | 0.006 | 0.006 0.0001 |
| DLPNO-CCSD/aug-cc-pVDZ | 2396.00 | 2442 21 | 16.00 | 14.12 0.4 | 0.006 | 0.006 0.0001 |
| DLPNO-CCSD(T)/def2-SVP | 2403.29 | 2423 23 | 14.23 | 14.13 0.3 | 0.006 | 0.006 0.0001 |
| DLPNO-CCSD(T)/aug-cc-pVDZ | 2329.51 | 2340 20 | 14.60 | 14.30 | 0.006 | 0.006 0.0002 |
| CCSD(T)/aug-cc-pVDZ | 2319.66 | 2323 15 | 14.75 | 14.98 0.3 | 0.006 | 0.006 0.0001 |
| CCSD(T)/aug-cc-pVTZ | 2340.78 | 2345 18 | 14.45 | 14.65 0.3 | 0.006 | 0.006 0.0001 |
| CCSD(T)/aug-cc-pVQZ | 2355.75 | 2363 16 | 14.41 | 14.52 0.2 | 0.006 | 0.006 0.0001 |
| Experiment | 2358.52 | 14.35 | 0.006 |
Table S1. Vibrational parameters of gas phase obtained from Morse and polynomial fits at different levels of theory.
“ by the wall”, which is the closest to the fullerene wall, showed the largest change in bond distance and harmonic frequency , while the anharmonicity changed very little.
Tip-enhanced Raman spectroscopy measurements
Tip-enhanced Raman spectroscopy (TERS) measurements were performed to investigate whether the spatial distribution of could be correlated with the topographic and vibrational features of a mixed layer deposited on Au. The resulting AFM and TERS measurements are presented in Figure S4. The topographic image in Figure S4 was acquired using a SmartSPM system (HORIBA, France) operated in tapping mode. During acquisition, the lateral drift was measured to be approximately 0.82 nm min.

Figure S4. AFM and tip-enhanced Raman spectroscopy of a mixed layer on Au. (a) Tapping-mode AFM topographic image showing the formation of monolayer (1L) and bilayer (2L) fullerene domains on the Au surface. The displayed height range is to nm. The dashed orange box marks the region over which the TERS maps in panels (b)–(d) were acquired. (b)–(d) TERS intensity maps obtained by integrating the Raman signal over the spectral ranges – cm (b), – cm (c), and – cm (d). These ranges contain contributions from the and modes of , the mode of , and the stretching mode of encapsulated , respectively. The displayed intensity ranges are to counts in (b), to counts in (c), and to counts in (d). (e) Representative TER spectra acquired at the monolayer and bilayer positions marked by the black and red boxes, respectively, in panel (b). The spectra are dominated by the , , and modes of , whose intensities are generally greater within the bilayer region. No distinct signal attributable to the intramolecular vibration of encapsulated was resolved. The available spectral resolution was also insufficient to determine small changes in the fullerene-mode peak positions that might otherwise provide indirect confirmation of the presence of within the cage.
Reflection-mode TERS measurements were performed using a commercial LabRAM Nano system (HORIBA, France), comprising the same AFM and optical spectrometer used for the AFM and micro-Raman spectroscopy measurements, respectively. The two instruments were optically coupled using an OmegaScope platform (HORIBA, France) equipped with a 100, 0.7-NA metallurgical objective lens (Mitutoyo, Japan), providing side illumination at an angle of relative to the sample-surface normal. A 150-lines-mm diffraction grating was used, providing a spectral resolution better than 6.6 cm with a pixel-binning value of 2.
Before each measurement, the laser was aligned with the probe apex by scanning the objective position and mapping the intensity of the weak plasmonic signal produced by the tip while it was retracted from the sample surface. TERS measurements were conducted using a 633-nm excitation laser because a greater near-field signal was observed at this
| Structure | (cm) Morse | (cm) Polynomial | (cm) Morse | (cm) Polynomial | Morse | Polynomial |
| N@C () | 2393.18 | 2378.31 17 | 16.72 | 13.55 0.4 | 0.007 | 0.006 0.0002 |
| N@C () | 2394.31 | 2376.89 17 | 16.75 | 14.65 0.3 | 0.007 | 0.006 0.0001 |
| N@C () | 2387.23 | 2375.98 20 | 16.80 | 14.38 0.7 | 0.007 | 0.006 0.0003 |
| N@C (by the wall, ) | 1987.12 | 2029.69 14 | 13.72 | 15.95 1.6 | 0.007 | 0.008 0.0007 |
| N@C (by the wall, ) | 2612.20 | 2567.43 11 | 15.94 | 15.32 0.8 | 0.006 | 0.006 0.0003 |
| N@C (by the wall, ) | 2450.56 | 2481.18 12 | 17.06 | 17.18 1.3 | 0.007 | 0.007 0.0005 |
| Experiment | 2348.70 | 30.64 | 0.013 |
Table SII. Vibrational parameters of N@C obtained from Morse and polynomial fits. Level of theory: DLPNO-CCSD/def2-SVP
wavelength than with the 785-nm laser for the commercial Omni-TERS-SNC-Au probes employed (AppNano, USA). The maximum incident laser power was limited to 0.07 mW to reduce the likelihood of thermal or chemical damage to either the probe or the sample.
Measurements were performed in “SpecTop” mode. At each measurement position, the probe was operated in contact mode to collect the combined near-field and far-field Raman signal, . The probe was subsequently operated in tapping mode to collect the far-field-only signal, , and while moving between measurement positions. The Raman contrast factor was defined as
Using the HORIBA reference sample, comprising carbon nanotubes supported on Au, a typical probe used in this study produced a Raman contrast factor of approximately 15. The spatial resolution, estimated using the same reference sample, was approximately 15 nm.
References
- [1]S. Bloodworth and R. J. Whitby, Synthesis of endohedral fullerenes by molecular surgery, Com- munications Chemistry 5, 121 (2022).
- [2]T.-L. Lee and D. A. Duncan, A Two-Color beamline for electron spectroscopies at Diamond Light Source, Synchrotron Radiation News 31, 16 (2018).DOI
- [3]S. P. Jarvis, H. Sang, F. Junqueira, O. Gordon, J. E. A. Hodgkinson, A. Saywell, P. Rahe, S. Mamone, S. Taylor, and A. Sweetman, Chemical shielding of H₂O and HF encapsulated inside a C₆₀ cage, Communications Chemistry 4, 135 (2021).DOI
- [4]C. Fields, A. Foerster, S. Ghaderzadeh, I. Popov, B. Huynh, F. Junqueira, T. James, S. Alonso Perez, D. A. Duncan, T.-L. Lee, Y. Wang, S. Bloodworth, G. Hoffman, M. Walkey, R. J. Whitby, M. H. Levitt, B. Kiraly, J. N. O’Shea, E. Besley, and P. Moriarty, Timing the escape of a photoexcited electron from a molecular cage, Nat. Commun. 16, 5062 (2025).DOI
- [5]A. Sufyan, T. James, C. Fields, S. Naseri, F. Junqueira, S. Alonso-Perez, S. Bloodworth, G. Hoffman, M. C. Walkey, E. S. Marsden, R. J. Whitby, Y. Wang, D. A. Duncan, T.-L. Lee, J. N. O’Shea, J. A. Larsson, B. Kiraly, and P. Moriarty, Charge transfer dynamics in noble gas endofullerenes: intra- and extramolecular tunnelling, Nanoscale Advances 7, 7913 (2025), https://pubs.rsc.org/na/article-pdf/7/24/7913/1887236/d5na00727e.pdf.DOI
- [6]C. Enkvist, S. Lunell, B. Sjögren, S. Svensson, P. A. Brü hwiler, A. Nilsson, A. J. Maxwell, and N. Mårtensson, c₁s shakeup spectrum of C₆₀: Global charge-transfer satellites and their relation to the x-ray threshold singularities in macroscopic systems, Phys. Rev. B 48, 14629 (1993).pubmed.ncbi.nlm.nih.gov/10007885
- [7]A. J. Maxwell, P. A. Brühwiler, A. Nilsson, N. Mårtensson, and P. Rudolf, Photoemission, autoionization, and x-ray-absorption spectroscopy of ultrathin-film c₆₀ on au(110), Phys. Rev. B 49, 10717 (1994).pubmed.ncbi.nlm.nih.gov/10009900
- [8]DLPNO is a method that retains only the important virtual orbitals associated with each occupied electron pair contributing to the correlation energy, rather than the full virtual space used in canonical CCSD(T).

