Magnetic anisotropy of endohedral lanthanide ions: paramagnetic NMR study of MSc2N@C80-I h with M running through the whole 4f row† †Electronic supplementary information (ESI) available: Details of relaxation time measurements, ligand field splitting, additional correlation between data points, Cart

Paramagnetic and variable temperature 13C and 45Sc nuclear magnetic resonance studies are performed for nitride clusterfullerenes MSc2N@C80 with icosahedral I h(7) carbon cage, where M runs through all lanthanides forming nitride clusters.


Introduction
The molecules of endohedral metallofullerenes (EMFs) comprise the fullerene cage encapsulating one to three metal ions or a more complex hybrid cluster. 1 Shielding magnetic ions (such as lanthanides) inside the carbon cage protects their spin states from the environment making EMFs potentially useful for spintronics applications. Numerous studies of magnetic properties of EMFs reported before 2012 did not reveal any deviation from paramagnetic behaviour, although the inuence of the magnetic anisotropy of endohedral lanthanide ions on the measured magnetic moments was proposed. 2 However, in 2012 it was discovered that DySc 2 N@C 80 exhibits slow magnetization reversal at helium temperatures. 3 The presence of hysteresis in the magnetization curves measured below 5 K indicated that this EMF can be classied as a single molecule magnet (SMM). In subsequent studies, the SMM behaviour was also found in Dy 2 ScN@C 80 (ref. 4) and HoSc 2 N@C 80 , 5 and more EMF-based SMM are likely to be discovered in the near future.
Importantly, both DySc 2 N@C 80 and Dy 2 ScN@C 80 exhibit unusually long attempt times, which are orders of magnitude longer than in other lanthanide-based SMMs.
One of the most important factors determining the SMM behaviour of lanthanide molecular magnets is the single ion magnetic anisotropy. 6 Reliable elucidation of the ligand-eld (LF) splitting of the 4f n total momentum states is crucial (although not sufficient) for the correct description of magnetic properties of lanthanide SMMs. Despite the understanding that the magnetic anisotropy is an important factor determining magnetic properties of EMF, a consistent description of the LF effects in EMF has not been provided yet. The rst dedicated studies of the LF in nitride clusterfullerenes have been published only recently and were motivated by the discovery of SMM behaviour in DySc 2 N@C 80 and Dy 2 ScN@C 80 . According to the ab initio computations at the CASSCF level, 7 the LF splitting in DySc 2 N@C 80 is so large that its magnetization behavior at low temperature is determined solely by the ground state of the 6 H 15/2 manifold. Namely, the gap between the ground (m J ¼ AE15/2) and the rst excited (m J ¼ AE13/2) magnetic states was predicted to be 373-415 cm À1 (ref. 7a) or 485 cm À1 (ref. 7b). Likewise, magnetization curves of DySc 2 N@C 80 and HoSc 2 -N@C 80 measured by SQUID below 10 K are also described by only one state with m J ¼ AE15/2 (Dy) or m J ¼ AE8 (Ho). 4,5 The information about excited magnetic states might be revealed from magnetizations studies at higher temperatures (e.g., the population of the m J ¼ AE13/2 state in DySc 2 N@C 80 near room temperature is expected to be ca. 10%). However, with the increase of the temperature the total magnetization of the sample is decreasing dramatically and can hardly be measured reliably when only limited amount of sample is available (as it is oen the case with EMF SMMs). An alternative might be the use of spectroscopic techniques, such as analysis of the ne structure in the lanthanide-based luminescence spectrum. 8 Unfortunately, the fullerene cages of EMFs absorb light in the visible range and hence block the possibility to observe the lanthanidebased luminescence for a majority of 4f elements. So far, metalbased luminescence in EMFs could be detected only for endohedral Er 3+ ions emitting in the near-IR range. 9 Solution nuclear magnetic resonance (NMR) studies can provide complimentary information on the magnetic anisotropy of lanthanide ions for the temperatures not well accessible by direct magnetization measurements. The chemical shi of a nuclei in a paramagnetic compound can be described as a sum of diamagnetic and paramagnetic terms, d exp ¼ d dia + d para . In due turn, the paramagnetic shi has two major contributions, contact d con and pseudocontact d pc . 10 The contact (Fermi) shi results from the interaction between the nuclear spin (of 13 C or 45 Sc atoms in this work) and the spin-polarized electron density of the molecule (in particular in the region close to the nuclei of interest). As such, the contact shi is proportional to the hyperne coupling constant weighed with the expectation value of the spin operator S z of the lanthanide. The pseudocontact shi is caused by dipolar through-space interactions of the nuclear and electronic magnetic dipoles. For an i-th atom in a lanthanide-containing molecule the pseudocontact shi can be computed as: where c Ln aa are components of the magnetic susceptibility tensor of the lanthanide, whereas R i , q i , and 4 i are polar coordinates of the i-th atom in the coordinate system centred on the lanthanide ion. In particular, R i is a distance between the atom of interest and lanthanide ion, and q i is an angle between quantization axis z and the vector connecting the lanthanide ion and the i-th atom. If c Ln xx ¼ c Ln yy (i.e. the ligand eld is uniaxial), eqn (1a) is simplied to: As can be seen in eqn (1), the pseudocontact shi contains information on the molecular structure and on the magnetic properties of the paramagnetic center. Paramagnetic NMR is therefore a popular structure elucidation tool for metal complexes, polymers or biomolecules. 10b,c,11 At the same time, it can be also used to determine or verify LF parameters in lanthanide complexes. 12 The prerequisite for both types of application of NMR spectroscopy in the studies of paramagnetic molecules is the possibility to separate contact and pseudocontact contributions, 13 since only the latter brings necessary information about molecular structure and magnetic anisotropy.
In the eld of endohedral metallofullerenes, paramagnetic NMR was mainly used for the structural studies. Several Ce-EMFs, 14 anion of Pr@C 82 , 15 three isomers of Tm@C 82 , 16 Sm@C 80 , 17 and Sm@C 82 (ref. 18) were characterized by 13 C NMR. The variable-temperature 45 Sc NMR study was also reported for CeSc 2 N@C 80 . 19 The dominant contribution of the pseudocontact shi in Ce-EMFs was postulated based on the analysis of the temperature dependence in the framework of Bleaney's theory. The size of the contact term contribution in the paramagnetic NMR spectra of EMFs remains unclear. Our group recently reported paramagnetic NMR study of HoM 2 N@C 80 and Ho 2 MN@C 80 (M ¼ Sc, Y, Lu). 20 13 C and 45 Sc signals of these molecules could be identied in spite of the severe broadening induced by Ho 3+ ions. This study showed that paramagnetic NMR spectroscopy when combined with the analysis of the ligand eld splitting, can provide information on the magnetic state of lanthanide ions in EMFs.
In this work we report on a systematic paramagnetic NMR study of MSc 2 N@C 80 compounds with all lanthanides forming nitride clusterfullerenes (M ¼ La, Ce, Pr, Nd, Gd, Tb, Dy, Ho, Er, Tm) and concomitant calculations of the ligand eld splitting in these molecules. The main goal is to provide a consistent and uniform description of the magnetic states of lanthanide ions in these molecules based on the ligand eld computations veried by experimental NMR data. The manuscript is organized in several parts. First, we describe the results of experimental NMR measurements, including the temperature dependence and relaxation times. Then, we perform a thorough analysis of the NMR data to separate contact and pseudocontact contributions to paramagnetic shis. This section is followed by the brief analysis of the molecular structures. Then point-charge calculations of the ligand eld splitting are described, which give the desired description of the single ion magnetic anisotropy in the series of MSc 2 N@C 80 molecules across the whole 4f row. Reliability of the computed LF patterns is then veried by comparison of the computed and experimental pseudocontact shis.

Experimental and computational details
In this work we will consider exclusively MSc 2 N@C 80 compounds with the I h (7) carbon cage (Fig. 1). Hereaer the compounds will be abbreviated as 1M, where M denotes the corresponding lanthanide (e.g., LaSc 2 N@C 80 is denoted here-aer as 1La). All studied compounds were synthesized using the arc-discharge method in the presence of nitrogen source such as NH 3 (ref. 21) or guanidine thiocyanate, 22 and then separated using chromatography as described in the ESI Fig. S1-S5. † The synthesis of 1Ce, 23 1Nd, 24 1Pr, 23 1Dy, 24 1Ho, 20b and 1Lu 24 in our group was described earlier. 1La, 1Pr, 1Tb, 1Er and 1Tm were synthesized for this work using guanidine thiocyanate as a nitrogen source, whereas 1Tb and 1Y were synthesized using NH 3 as a reactive gas. Note that the synthesis and characterization of 1Y, 25 1La, 26 1Ce, 19 1Tb, 27 and 1Er 28 was also reported by other groups using somewhat different conditions. Gd also forms mixed-metal nitride clusterfullerenes with Sc, 29 but neither 13 C nor 45 Sc NMR signals could be detected for 1Gd, so this molecule will not be discussed hereaer.
The 125 MHz 13 C NMR and 121.5 MHz 45 Sc NMR measurements were performed on a Bruker Avance 500 spectrometer equipped with a multiprobe head 1152Z. The measurements were performed for compounds dissolved in CS 2 with d 6acetone placed in a coaxial tube as a lock; in special cases d 4ortho-dichlorobenzene was also used as a solvent. Typically the spectra were measured at each 10 K in the temperature rage form 268 to 308 K.
Longitudinal relaxation rates (R 1 ) were measured using inversion recovery pulse sequence with different delays between p and p/2 pulses. The values were determined from the slope of the linear t of ln(I max À I) versus s (delay). Transversal relaxation rates (R 2 ) were measured using Carr-Pucell-Meiboom-Gill (CPMG) pulse sequence. 30 The values were determined from slopes of linear t of Àln(I) versus s (delay). In case of broad line width, R 2 cannot be determined by CPMG-pulse sequence because the relaxation is too fast. Conservative estimation of R 2 values in such cases can be done by measuring the line width at half peak maximum. However, R * 2 determined by this procedure includes the line broadening by inhomogeneity of the magnetic eld (R * 2 ¼ yDB 0 2 þ R 2 ) and therefore the T 2 time determined from the line width is shorter than the real one. DFT optimization of molecular coordinates was performed at the B3LYP level using the Firey 31 code. The basis sets were def2-SVP {5,1,1/3,1/1} for carbon atoms, 32 def2-TZVP {6,2,1,1/4,1,1/1,1/1} for nitrogen, 32 and Stuttgart-Cologne effective core potentials for Sc (ECP10MDF) 33 and lanthanides (4f-in-core ECP-MWB-II) 34 with {3,1,1,1,1,1/2,2,1,1,1/4,1,1/1,1/1} and {3,1,1,1,1,1/3,1,1,1,1/2,1,1,1,1/ 1,1,1/1,1} valence parts, respectively. Spin density and hyperne coupling constants in GdSc 2 N@C 80 were computed at the PBE level using the specially tailored SARC basis set of TZVP quality 35 combined with either DKH or ZORA scalar relativistic corrections implemented in the ORCA package. 36 To compute the Bader (QTAIM) charges, full electron calculations were performed at the PBE/SARC-TZVP level with DKH scalar relativistic corrections using ORCA. To avoid poor reliability of the full electron DFT treatment of the 4f elements, Y was used to model the charges of the lanthanides from Tb to Tm, whereas La was used to model Ce, Pr, and Nd. QTAIM computations of the atomic charges were then performed with the AIMAll 37 code. Point charge calculations of the ligand eld splitting, magnetic susceptibility tensors, and 4f electron densities were performed using so1ion (Celd) routine in McPhase code. 38 Molecular structures and isosurfaces were visualized with the help of VMD. 39 Results and discussion NMR spectroscopy 13 C and 45 Sc NMR spectra at 288 K. Fig. 2 and 3 show 13 C and 45 Sc NMR spectra of 1M compounds measured at 288 K (the temperature is chosen as the center of the temperature interval used in the variable temperature NMR studies, see below).
Fast rotation of the MSc 2 N cluster inside the highly symmetric C 80 -I h carbon cage results in the averaging of the carbon signals, leaving only two 13 C NMR peaks with a 3 : 1 intensity ratio. The higher intensity peaks corresponds to the 60 carbon atoms at the pentagon-hexagon-hexagon junctions (PHHJ), whereas the lower intensity peak is due to the 20 atoms at the triple hexagon junctions (THJ, see Fig. 1). The chemical shi of the PHHJ carbons in diamagnetic 1M compounds is bracketed by the 1La and 1Lu values, 144.76 and 143.99 ppm, respectively. Likewise, the THJ signal varies from 138.08 ppm in 1La to 136.90 ppm in 1Lu. Chemical shis of the THJ and PHHJ Fig. 1 DFT-optimized molecular structure of representative MSc 2 N@C 80 -I h (1Pr). Praseodymium ion (shown green) is at the distance of 2.225Å from the central nitrogen atom (blue), whereas two Sc-N distances are 1.939 and 1.936Å. Carbon atoms are shown in grey (pentagon-hexagon-hexagon junctions, PHHJs) and orange (triple hexagon junctions, THJs). In (b), the plane of the cluster is normal to the paper. Thin cyan lines denote shortest Pr-C distances: 2.508Å (C1), 2.553Å (C2), 2.591Å (C3), and 2.658Å (C4).   Table 1). Thus, depending on the size of the endohedral cluster, the diamagnetic 13 C values are subject to change within the range of ca. 1 ppm (see also ref. 40) which should be taken into account in the analysis of the paramagnetic shis in other 1M compounds. The d dia values listed in Table 1 and used to compute paramagnetic shis were obtained by approximating the 13 C chemical shis by a quadratic function of the Shannon radii of metal ions (see ESI Fig. S6 †).
The measurement of the 13 C NMR spectra of paramagnetic EMFs is complicated by the considerable line broadening. However, thanks to the simple two-line spectrum of the C 80 carbon cage, the signals are detectable even when they are severely broadened as in 1Tb or 1Dy. In 1Ce, 1Pr, 1Nd, 1Tb, 1Dy, and 1Ho the 13 C signals are shied up-eld with respect to the diamagnetic values. The magnitude of the paramagnetic shi varies from few ppm in 1Ce, 1Pr, and 1Nd to tens of ppm in 1Tb, 1Dy, and 1Ho ( Fig. 2 and Table 1). On average, lanthanideinduced shis of the THJ carbons are more pronounced than those of the PHHJ carbons.
In 1Er and 1Tm the lanthanide-induce shi is positive (down-eld). The PHHJ and THJ 13 C signals of 1Er are coinciding near the room temperature and appear at 153.2 ppm, but can be resolved at higher or lower temperatures as discussed below. In 1Tm, the stronger lanthanide-induced shi of the THJ signal pushes it to a lower eld than that of the PHHJ (in all other 1M compounds the PHHJ signal appears at lower eld than that of the THJ carbons). Direction of the lanthanideinduced paramagnetic shi is determined by a magnetic anisotropy of the lanthanide ion (see ref. 41 for recent examples), and hence experimental NMR data gives information at least on the sign of the anisotropy. This question will be discussed in more details below.
In 45 Sc NMR spectra, all 1M compounds exhibit one relatively broad peak (Fig. 3). The 45 Sc chemical shis of the diamagnetic 1La and 1Lu are 198 and 200 ppm, respectively, whereas in paramagnetic 1M molecules the 45 Sc signal position spans the range from À233 ppm in 1Er to 1892 ppm in 1Dy. The peak width varies from 30-40 ppm for the diamagnetic 1Lu and 1La and paramagnetic 1Ce, 1Pr, 1Nd and 1Er, to ca. 70-80 ppm for the 1Tb, 1Dy, 1Ho, and 1Tm.
Variable temperature NMR studies. Temperature dependence of paramagnetic NMR shis can be used as an additional parameter to distinguish the contact and pseudocontact contributions as well as to clarify the extent of the magnetic anisotropy. Fig. 4 and 5 show variations of 13 C and 45 Sc NMR spectra in the temperature range 268-308 K. To quantify the temperature variation, we will use the D 308 268 (d) parameter, which is the difference of the chemical shis measured at 308 and 268 K.
Temperature variation of the peak positions is in line with the single-temperature d para values. Namely, the compounds with strong paramagnetic shis exhibit enhanced variation of the peak position with the temperature and vice versa. Likewise, the sign of the D 308 268 (d) values is opposite to the sign of the d para shis. The only exclusion is the PHHJ signal of 1Tm, for which both d para and D 308 268 (d) parameters are positive. Since this signal  has the smallest d para value (0.79 ppm), we cannot exclude that uncertainties in the estimation of the d dia value may play a role here. More likely, however, is that the pseudocontact and contact contributions have opposite signs and compensate each other in this molecule. Line width and relaxation times. Variation of the line width, which is especially well pronounced in the 13 C NMR spectra, points to the substantial change of relaxation times across the lanthanide row. A transverse relaxation time T 2 can be estimated using the relation T 2 ¼ (pDn 1/2 ) À1 , where Dn 1/2 is the peak width at its half maximum. The Dn 1/2 values determined for PHHJ signals at 288 K are listed in Table 1. For 1Tm we have also measured T 2 and T 1 times directly using Carr-Purcell-Meiboom-Gill pulse sequence and inversion recovery method, which gave 50 ms and 66 ms at 288 K, respectively (see ESI Fig. S7 and S8 † for other temperatures). As might be anticipated, T 1 is longer than T 2 , and the T 2 * value of 19 ms determined by the line width method underestimates the real T 2 time but gives correct order of the value. Direct measurements of shorter relaxation times is beyond the possibilities of our spectrometer. In diamagnetic 1M compounds, the nuclei spin relaxation times are much longer. 42 For instance, T 2 and T 1 times of the PHHJ carbons in 1Y are 17 and 23 s, respectively. Note that T 2 determined from the CPMG pulse sequence for diamagnetic 1Y is signicantly longer than might be estimated from the line width (T 2 * ¼ 0.2 s), which shows that the line width for such narrow signals is limited by the instrument.
Comparison of the Dn 1/2 values for 13 C peaks in 1M series ( Fig. 1 and Table 1) reveals dramatic inuence of the lanthanide on the nuclei spin relaxation. For instance, 13 C nuclei spins in 1Dy (T 2 z 0.3 ms) relax ca. 5 Â 10 4 times faster than in diamagnetic 1Y (T 2 ¼ 17 s). The plot of Dn 1/2 versus d para for the PHHJ carbons along the 1M series can be well t by a parabola (see ESI Fig. S9 †). This fact points to the prevailing contribution of the pseudocontact shi in the total paramagnetic shi since the relaxation rate in paramagnetic molecules has R À6 dependence on the distance between the nuclei and paramagnetic center, whereas the pseudocontact shi scales as R À3 (eqn (1)).
The broadening of 45 Sc signals along the 1M series is not so pronounced as for carbon signals because the 45 Sc quadrupole moment broadens the peaks even in diamagnetic molecules. The 45 Sc signals in 1Dy, 1Tb, 1Ho, and 1Tm are roughly two times broader than in the 1Y (Fig. 3), which can be ascribed to a paramagnetic effect. At the same time, in 1Ce, 1Pr, 1Nd, and 1Er the line width is comparable to that of diamagnetic 1Y, which means that the spin relaxation rate in these molecules is still determined by a quadrupole effect.
The line width exhibits substantial decrease at lower temperatures (Fig. 4 and 5). For instance, 45 Sc signals measured at 268 K are roughly twice broader than at 308 K (Fig. 5). In some previous works, temperature dependence of the 45 Sc line width was used to estimate rotation barrier of the endohedral clusters. 19,43 If the line width is directly proportional to the rotational correlation time, the plot ln(Dn 1/2 )-versus-T À1 gives a straight line with the slope proportional to the activation energy. In this work, none of the 1M molecules gave straight line in the ln(Dn 1/2 )-versus-T À1 coordinates, preventing thus further analysis of the data. The broadening at low temperature should result from the interplay of the slowing of the cluster dynamics and the electron spin contribution, which can hardly be separated at this moment.

Separation of contact and pseudocontact shis
The localized "buried" nature of the 4f electrons in lanthanides implies that the contact contribution to the paramagnetic shi should be rather small. Typically, the values become negligible when the nucleus of interest is separated by more than four Fig. 4 Variable-temperature 13 C NMR spectra of selected 1M compounds measured in CS 2 in the 268-308 K range. Vertical dash lines denote chemical shifts at 288 K and are shown to guide an eye. At 288 and 308 K two carbon signals in 1Er are not resolved, and therefore the measurements were performed in o-DCB in a broader temperature range (268-328 K). Low solubility of 1Dy and 1Tb at 268 K resulted in poor signal-to-noise ratio (the noise features in the spectrum of 1Dy are marked by asterisks). bonds from the lanthanide ion. However, in the 1M molecules both Sc and carbon atoms are located rather close to the lanthanide, and hence the contact term cannot be ignored. With systematic NMR data along the whole lanthanide series in hand, we can address this problem. Separation of the contact and pseudocontact shis is a long standing problem in paramagnetic NMR, and several schemes were proposed to solve it. 11c, 13,44 Unfortunately, all methods allowing analytical deviation of linearizable equations are based on Bleaney's assumption 45 that LF splitting is less than the thermal energy and hence expansion of the magnetic susceptibility in reciprocal temperature series is limited to T À2 . This assumption is denitely not fullled for 1M molecules, and hence the results of these calculations may give only qualitative estimation at best (see also ref. 41a for a detailed discussion of the limitations of Bleaney's theory). Another assumption usually applied is that the system is either purely axial (LF operator described in phenomenological B q k parameters is limited to the B 0 2 term), or its rhombic anisotropy can be described solely by a B 2 2 parameter. The sum of contact and pseudocontact shis can be then described as: The rst term is the contact shi, where F i is proportional to a Fermi coupling constant, and hS z i j is the expectation value of the spin projection operator S z for a given lanthanide. The second term is the pseudocontact shi, C j is a numerical factor specic for each lanthanide and tabulated by Bleaney et al.,45b whereas G i and H i are structural factors (compare to eqn (1a)). If rhombic anisotropy is negligible, the second term reduces to d pc ij ¼ C j B 0 2 G i (compare to eqn (1b)).
Reilley's method. Following the method proposed by Reilley et al., 13 eqn (2) can be written in the form: The plot d para ij /hS z i j versus C j /hS z i j along the lanthanide series is expected to give a straight line, if all molecules are isostructural, and the F i value and LF parameters are not changing along the series. Contact and pseudocontact shis can be then computed from the intercept and the slope. Fig. 6 shows d para ij /hS z i j versus C j /hS z i j plots for two NMR signals measured for 1M molecules. 13 C PHHJ and 45 Sc shis can be roughly linearized if 1Tm values are excluded from the set; the d con and d pc shis computed from the linear ts are listed in Tables 2 and 3. The 13 C THJ shis cannot be tted by a straight line in these coordinates (see ESI Fig. S10 †).
The absolute values of d con shis for the PHHJ carbons are 2-4 times smaller than those of d pc . Furthermore, the d con and d pc shis of 1Ce, 1Pr, 1Nd, 1Er, and 1Tm, have opposite signs. For 1Nd this analysis predicts mutual compensation of d con and d pc shis, but uncertainties of 1-2 ppm make the analysis of the small values rather ambiguous.
The ratio between d con and d pc for the 45 Sc chemical shis is smaller than for the 13 C counterparts, and the contact term contribution drops to less than 10% for the majority of lanthanides; the largest relative d con values of ca. 20% are found in 1Nd and 1Ho. Thus, although uncertainty of the t is rather high, the dominance of the pseudocontact term is beyond any doubt.
Deviations from the straight line in eqn (3) are usually considered to be an indication of the loss of the isostructurality along the series. This can hardly be the case for the 1M series since molecular structures determined by single-crystal X-ray diffraction for some of them are very similar (subject to vary with lanthanide contraction, see below). 19,[26][27][28] More likely, other assumptions used in the model are not valid. In particular, Reilley's approach implies that the LF parameters are not changing along the lanthanide row, which is most probably not correct for the 1M molecules. 7b Parameter-free models. The requirement of the constant LF parameters is one of the obvious weak points of Reilley's approach, which can be circumvented by combining the data on two or more nuclei and excluding B 0 2 and B 2 2 terms in eqn (3).
For an uniaxial system (B 2 2 ¼ 0), the 2-nuclei parameter-free model is obtained by plotting d para ij /hS z i j versus d para kj /hS z i j , which should give a straight line for the isostructural series (ref . 46): The slope of the line is a ratio of the structural factors G i /G k , and the intercept is a difference of F i and F k G i /G k . For the molecules with a rhombic anisotropy (B 2 2 s 0), the parameterfree model can be obtained in a similar fashion by simultaneous analysis of chemical shis of three different nuclei. In the latter case linearization gives the equation of a plane (ref. 47): here coefficients D ikl and E ikl are functions of the structural parameters G i and H i from eqn (3), see ref. 47a for full details. Fig. 7 visualizes analysis of all chemical shis for 1M molecules within parameter free models. The 3D plot corresponds to the 3-nuclei method, whereas projections on the coordinate planes allow evaluation of the 2-nuclei method for three different combination of nuclei. In the 3-nuclei model, all data points except for 1Pr are close to the plane obtained by the least square tting of eqn (4b). Analysis of the 2D plots in Fig. 7 shows that a good linear correlation exists only for the PHHJ-vs.-Sc plot (1Pr value is deviating again). Substituting the d con values obtained for the PHHJ carbons using eqn (3) into eqn (4a) for the PHHJ-vs.-Sc correlation allows estimation of the 45 Sc d con shis. The values obtained by this way are much smaller than those obtained using Reilley's approach. It should be also noted that the results of the d con calculations for Sc using the 2nuclei method are rather erratic and change signicantly with small variation of the input parameters. At the same time, both methods agree in that the 45 Sc d con shis are considerably smaller than the pseudocontact contributions.
The fact that the data points for THJ carbons t well into the 3-nuculei model but do not give linear correlations in the 2nuclei model might indicate that the rhombic anisotropy contribution is considerable here, but the limited amounts of points does not allow more solid conclusions.
DFT calculations of contact shis. Independent computations of contact shis are possible using quantum chemical approaches. DFT computations of lanthanide-containing molecules for systems other than 4f 7 cannot be reliable (unless 4f-in-core effective core potentials are used, but this approach cannot be used if spin properties related to the 4f electrons are of interest). However, the F i value usually remains the same for the whole lanthanide row, and hence computations can be performed for Gd analogues. Recently, such calculations provided accurate estimation of the contact shis in a series of Tb complexes. 48 Fig . 8 shows the spin density distribution in 1Gd computed using the scalar-relativistic DKH2 approximation, PBE functional, and SARC-TZVP basis set. Negative spin polarization of the central nitrogen atoms and several carbon atoms close to the lanthanide can be well seen (red-colored lobes). At the same time, the sign of the spin density on the Sc atoms and carbon  a Obtained using the t of eqn (4a) together with 13 C-PHHJ shis, and then subtracting d con ( 13 C) values estimated by Reilley's approach; 1Pr value was not included in the t; inclusion of 1Pr shi worsened the t from R 2 ¼ 0.96 to R 2 ¼ 0.87, but gave the values closer to those obtained by Reilley's approach. b The values are obtained as d pc (2-nuc) ¼ d para (exp) À d con (2-nuc). c "PCM" stands for point charge model. d D 308 268 values are computed using PCM approach and include only d pc contribution. Fig. 7 Three-nuclei 3D plot in d para ij /hS z i j coordinates for 45 Sc, 13 C-THJ and 13 C-PHHJ chemical shifts in 1M series (T ¼ 288 K). Red spheres are 3D data; blue, green, and black circles are projections on coordinate planes. Red grid shows a fitted plane (1Pr was not included in the fit).
atoms further away from the Gd ion is the same as for the Gd itself.
Hyperne coupling constants for carbon atoms in GdSc 2 -N@C 80 span the range from À0.041 MHz to 0.096 MHz, and the averaging over the whole carbon cage yields the values of 0.015 and 0.010 MHz for the PHHJ and THJ carbons, respectively. The pseudocontact shis computed for 13 C atoms using these constants are listed in Table 2. Remarkably, the DFT-computed values for the PHHJ carbons are quite close to those obtained by Reilley's method. If we tentatively suggest that the agreement between the theory and experiment is equally well for the THJ carbons, their contact shis should be 35% smaller than those of PHHJs. Furthermore, relative contributions of the contact term in total paramagnetic shis of the THJ carbons are much smaller than for the PHHJ carbons.
For the 45 Sc hyperne coupling constant, the PBE-DKH approximation with tailored SARC-TZVP basis set gives the average value of À0. 27 MHz. An increasing the basis set to def2-TZVP or inclusion of the spin-orbit coupling corrections give the values of À0.26 MHz and À0.28 MHz, respectively, whereas switching to ZORA scalar relativistic method reduces the 45 Sc hyperne coupling constant to À0.21 MHz. If hybrid PBE0 functional is used instead of PBE with SARC-TZVP basis, predicted DKH and ZORA values are À0. 35 and À0.49 MHz, respectively. The use of PBE-DKH/TZVP constants for estimation of the 45 Sc contact shis in 1M molecules yields unrealistically high values (625 ppm for 1Tb, 560 ppm for 1Dy, etc.). It should be noted that computation of the 45 Sc hyperne coupling constants is less straightforward than for 13 C nuclei because of the subtle balance between the valence-and coreshell spin polarizations, which oen have opposite signs in transition metals. 49 Due to this reason, prediction of the 45 Sc hyperne coupling constants meets serious difficulties even for the Sc-based EMF radicals such as Sc 3 N@C 80 À (ref. 50), where coupling constants are large and dominated by SOMO contribution. Reliable prediction of the core-shell spin polarization becomes even more problematic for paramagnetic molecules, in which the spin density on the Sc atoms is largely induced by spin polarization (such as in 1M). We have to admit that DFT prediction of small 45 Sc coupling constants in 1M can hardly be reliable at this moment and refrain from further discussion of DFT-computed d con ( 45 Sc) values.
Temperature dependence of chemical shis. If Bleaney's conditions are satised, the temperature dependence of chemical shis can be also used to distinguish contact and pseudocontact terms: whereas the contact term scales as T À1 , the pseudocontact term has T À2 dependence. If one of the terms dominates, plot d para versus T À1 or T À2 should give a straight line, whose intercept is close to d dia . In due turn, a large deviation of the intercept from the reference d dia value signals that the corresponding contribution is small. This simple yet efficient approach was employed before to prove the dominance of pseudocontact shi in Ce-based EMFs. However, this approach can give reliable results only if the pseudocontact term has real T À2 dependence, and if one the terms dominates. If LF splitting is larger than the thermal energy, contribution of T À3 and higher terms cannot be ignored, 51 and results of the linear tting become inconclusive. In the 1M series, only for 1Ce is the intercept of the d para (T À2 ) plot close to d dia for all three nuclei (see ESI Table S1 †). For all other lanthanides, the intercepts of d para (T À1 ) and d para (T À2 ) linear ts give rather irregular set of data, sometimes with large deviations from d dia for both T À1 and T À2 plots (ESI Table S1 †). Thus, large magnetic anisotropy of lanthanide ions in 1M molecules prevents the use of temperature dependence for distinguishing contact and pseudocontact contributions to paramagnetic chemical shis. Good results obtained for the 1Ce in d para (T À2 ) ts are in part due to the small hS z i Ce value (Table 2), which makes the contact term negligible.

Geometry parameters of 1M molecules
For the further analysis of the magnetic state and the LF splitting of lanthanide ions in the 1M compounds it is necessary to take into account a variation of the molecular geometry parameters along the series. Namely, a signicant decrease of the ionic radii of lanthanides from La to Lu cannot be ignored, especially when magnetic anisotropy is large. Therefore, to obtain a consistent set of the molecular geometry parameters along the whole 4f row for the subsequent LF calculations (see next section), we performed DFT optimization of all structures using the B3LYP functional and 4f-in-core basis set for lanthanides. The M-N and averaged Sc-N bond lengths are listed in ESI Table S2 † and are plotted versus Shannon ionic radii of M 3+ ions in Fig. 9. As expected, the decrease of the ionic radii from La to Lu results in the gradual decrease of the M-N distance (from 2.241Å in 1La to 2.133Å in 1Lu). At the same time, since the carbon cage remains the same, the increase of the lanthanide size also leads to the simultaneous decrease of the Sc-N bond lengths (compare 1.929Å in 1La to 1.979Å in 1Lu) so that the cluster size remains more or less the same along the whole series. Single-crystal X-ray structures were reported earlier for 1La, 26 1Ce, 19 1Gd, 27 1Tb, 27 and 1Er, 28 and experimental bond length values also conrm this trend. Namely, in the 1La-1Ce-1Gd-1Tb-1Er the lanthanidenitrogen bond is decreasing as 2.196(4)-2.184(2)-2.149(10)-2.126(11)-2.089(9)Å. The averaged Sc-N distance is varying as 1.932(7)-1.938(2)-1.918(9)-1.949(8)-1.968(6)Å, respectively, i.e. there is a gradual decrease from 1La/1Ce to 1Tb to 1Er (note that 1Gd deviates from the trend). A comparison to the computed data shows that DFT overestimates the M-N distances by ca. 0.04-0.06Å, whereas the error for the Sc-N distances is smaller than 0.01Å. Overall, DFT-optimized bond lengths give reasonable estimation of the experimental values.
Signicant variation of the bond lengths in nitride clusters can be also inferred from the analysis of the FTIR absorption spectra. In the frequency range of 500-800 cm À1 , all nitride clusterfullerenes have a characteristic antisymmetric metalnitrogen stretching vibration with medium to strong absorption intensity, whose frequency correlates with the metal-nitrogen bond length and therefore can be used for the analysis of the structural correlations. 29c,52 This mode corresponds to the inplane motion of the nitrogen atom and is two-fold degenerate for the M 3 N cluster with the frequency of (3k M-N ) 0.5 m N À0.5 , where k M-N is the force constant of the metal-nitrogen bond, and m N is the mass of the nitrogen atom. In the mixed-metal cluster MSc 2 N, the degeneracy is lied and the frequencies of two resulting vibrations are: (5b) Fig. 9c compares the IR spectra for 1La, 1Pr, 1Tb, and 1Tm (IR spectra of other 1M compounds were reported by our group earlier 20b, 24,29c ). In all spectra both the n M-N and n Sc-N components can be clearly identied in the 600-800 cm À1 range. Comparison of the n M-N values along the 1M series is not straightforward, because the frequency versus bond length correlation does not hold exactly for different metals and also because of the admixture of the k Sc-N force constant in eqn (5b). In fact, the n M-N frequency remains in the range of 635-650 cm À1 in the whole 1M series (Fig. 9b). The n Sc-N frequencies are better suited for the structural correlations: Fig. 9b shows that the n Sc-N frequencies vary from 652 cm À1 in 1Lu to 740 cm À1 in 1La and exhibit perfect linear correlation with the lanthanide ionic radii. Very good correlation also exists between the DFTpredicted Sc-N bond lengths and experimental n Sc-N frequencies (not shown), thus conrming reliability of the DFTcomputed structural parameters.

Ligand eld in 1M molecules
Reliable determination and verication of the cluster bond lengths discussed in the previous section is crucial for the analysis of the LF in nitride clusterfullerenes. Recently we showed that the point charge model employing scaled Bader charges can reasonably reproduce the ab initio (CASSCF) computed ligand eld splitting in DySc 2 N@C 80 , 7a,20a which enables the use of this simple approach for other 1M compounds. To calculate the LF splitting in this work we used B3LYP-optimized coordinates discussed above and Bader charges for nitrogen, scandium, and four closest carbon atoms (denoted C1-C4 in Fig. 1; since the cage carbon atoms have very small contribution to the LF, 20a it is sufficient to consider only few nearest atoms, their QTAIM charges are listed in the ESI Table S2 †). The charges were scaled by a factor of 0.754 to match the ab initio LF splitting in 1Dy from ref. 7a (see ESI Table S3 †). Fig. 10 shows an overview of the m J levels in all studied 1M molecules (the values and assignment are listed in the ESI Table  S4 †).
The central nitrogen atom gives the main contribution to the LF in nitride clusterfullerenes because of its large negative charge (Bader charge À1.7) and close distance to the lanthanide Fig. 9 (a) Correlation between DFT-computed M-N and Sc-N bond lengths in 1M molecules and ionic radii of lanthanides; (b) correlation between the metal-nitrogen stretching mode frequencies and ionic radii of lanthanides (green -M-N bonds, purple -Sc-N bonds); (c) FTIR spectra of selected 1M molecules: 1La, 1Pr, 1Tb, and 1Tm; arrows mark the n Sc-N modes. ion (see ESI Table S3 † for the LF point charge calculations using different parts of the MSc 2 N@C 80 molecule). Therefore, correct description of the molecular geometry is crucial since even small variations of the metal-nitrogen distance lead to considerable changes in the ligand eld. 20a Two Sc ions produce the second largest contribution to the LF aer the nitride ion. Their positive charges (Bader charge of ca. +1.7) reduce the effect of the nitride ion (e.g., in 1Dy the LF splitting is reduced by ca. 35% when Sc contribution is "turned on") and introduce a deviation from the uniaxial symmetry. The charges of carbon atoms are much smaller than those of N and Sc (below À0.1), and hence only few carbons close to the lanthanide ion have non-negligible contribution to the LF (in comparison to the M + N + 2Sc system, addition of four carbons increases the splitting energies by ca. 12%, see ESI Table S3 † for 1Dy and ref. 20a for a more detailed analysis of 1Ho). Thus, the LF experienced by lanthanide ions in 1M molecules has roughly a uniaxial character induced by a large negative point charge of the nitride ion, which results in rather simple LF splitting pattern despite the low molecular symmetry. A similar situation was found for a series of other Dy-based SMMs with low molecular symmetry. 53 Lanthanide ions with oblate 4f-density. The ground magnetic state of the ions with oblate shape of the 4f electron density (Ce, Pr, Nd, Tb, Dy, Ho; see ref. 54 and Fig. 11a for 4f density in 1Dy) has the largest J z value (5/2 in 1Ce, 4 in 1Pr, 9/2 in 1Nd, 6 in 1Tb, 15/2 in 1Dy, and 8 in 1Ho), and the energy of the other m J states is gradually increasing with the decrease of the |m J | values. The gaps between the neighboring m J states are also getting smaller with the decrease of the |m J | value. The deviation from the uniaxial symmetry induced by the two Sc ions is well seen for non-Kramers ions (Pr, Tb, Ho). For these ions, degeneracy of the AEm J states is not enforced, but remains rigorous in the uniaxial ligand eld produced by a single nitride ion. When the contribution of the Sc ions is also taken into account, an approximate degeneracy of the AEm J states holds only for large projections of the total momentum, whereas for the levels with |m J | ¼ 1-2 the splitting becomes signicant. A mixing of the m J states is also substantial for small |m J | values showing that the m J is not a "good" quantum number anymore.
The practical importance of the states with small |m J | values in ions with oblate 4f density is rather limited because the LF splitting in 1M is very large (Fig. 10). The smallest overall LF splitting, 571 cm À1 , is found in 1Ho, whereas the largest LF splitting is reaching 1915 cm À1 in 1Ce. The gap between the ground state and the rst excited state in 1Ce (1075 cm À1 ), 1Pr (849 cm À1 ), 1Nd (364 cm À1 ), 1Tb (384 cm À1 ), and 1Dy (382 cm À1 ) is so large that even at room temperature the ground magnetic state with the largest J z projection is predicted to have predominant contribution to the magnetic properties. As a result, 4f density in such 1M molecules exhibits only small variation with temperature. For instance, 4f densities in 1Dy calculated at 2 K and 288 K can hardly be distinguished by eye (Fig. 11a).
The two-fold degenerate ground state with the large J z projection and a signicant gap to the rst excited state are prerequisites for the SMM behaviour, 54b and hence the data in Fig. 10 show that slow magnetization reversal can be expected for all 1M compounds (except for the 1Gd, 1Er and 1Tm). The SMM behaviour with exceptionally long attempt time is already discovered for 1Dy. 3,4 Field-induced SMM behaviour is also found in non-Kramers 1Ho, 5 albeit on a much shorter timescale than in 1Dy. The rigorous protection of the two-fold degeneracy of the ground magnetic state in Kramers ions makes them more attractive for single ion magnetism, and we expect 1Ce and 1Nd to be the next candidates for the SMM behaviour in nitride clusterfullerenes as already detected in some organometallic complexes of Ce 55 and Nd. 56 It should be however noted that the magnetization reversal barrier in 1Dy is smaller than expected from its LF splitting, 4 which indicates that the mechanism for the magnetization reversal in 1M compounds is more complex and may include, for instance, Raman-like process. 5 Lanthanide ions with prolate 4f-density. The LF splitting pattern in 1Er and 1Tm is completely different because of the prolate shape of the 4f electron density of these lanthanides (Fig. 11b). The ground state for these lanthanides has the lowest J z projection (m J ¼ AE1/2 in 1Er and m J ¼ 0 in 1Tm). The energy is increasing with the increase of the |m J | values, and the density of states is higher in the low energy part of the spectrum, whereas at higher energies the energy levels are distributed sparser. As a result of such level distribution, variation of the temperature has profound effect on the magnetic anisotropy of 1Tm and 1Er. As can be seen in Fig. 11b, an increase of the temperature dramatically changes 4f electron density distribution in 1Er and makes it almost isotropic. Thus, the room-temperature magnetic properties of 1Er and 1Tm result from the contribution of several m J states. Fig. 11 (a and b) isosurfaces of 4f electron density in 1Dy (a, green) and 1Er (b, red) computed at 2 and 288 K. Only the nitride cluster and four carbon atoms nearest to the lanthanide ion are shown. Whereas the shape of the lobe remains almost the same in 1Dy, in 1Er the increase of the temperature from 2 to 288 K changes the shape of the 4f density distribution to a more spherical one. (c and d) Pseudocontact shift isosurfaces in 1Dy (c) and 1Er (d) at 288 K computed using the pointcharge model (cyanpositive, yellownegative). Solid/transparent surfaces correspond, respectively, to AE6000/AE1500 ppm isovalues in 1Dy and AE2000/AE500 ppm isovalues in 1Er.
1Er is the only compound in the whole 1M series for which spectroscopic data on the lanthanide-based luminescence are available. At helium temperatures the luminescence spectra of 1Er exhibit the ne structure due to the Er-based transitions in the 4 I 13/2 / 4 I 15/2 manifold. 9b,57 Unfortunately, emission from several molecular/cluster sites complicates the ne structure and makes assignment less straightforward. The energy gap between the two lowest energy LF states determined in ref. 57, 28/37 cm À1 , compares reasonably to the 37 cm À1 gap between the m J ¼ AE1/2 and m J ¼ AE3/2 states in our calculations, however the overall splitting determined in the optical measurements, 330 cm À1 , is considerably smaller than the value of 554 cm À1 predicted by the point-charge model.
Computed versus experimental pseudocontact chemical shis LF calculations allow direct estimation of the magnetic susceptibility tensor at different temperatures using Van Vleck formulae, and therefore enable computations of the pseudocontact chemical shis using eqn (1a) or (1b) given the atomic coordinates are known (e.g. from DFT calculations) and they are not changing with temperature. Comparison between experimental and computed chemical shis can be used then to evaluate reliability of the computed ligand eld splitting levels. Rotation of the nitride cluster inside the carbon cage means that the values computed for individual carbon atoms should be averaged. Recently we found that the static model (simple averaging of the values) and the averaging of the values computed over molecular dynamics trajectory give comparable results for 1Ho, 20a and therefore only static model is applied in this work.
In the 1M molecules with oblate 4f-density eqn (1a) and (1b) give identical values of the pseudocontact shi because c xx and c yy components of the magnetic susceptibility tensor are virtually equal. In the 1Er and 1Tm molecules, the difference between c xx and c yy values is rather large, and hence eqn (1a) is to be used. The pseudocontact shis computed for the THJ and PHHJ carbons as well as for the Sc atoms at T ¼ 288 K are listed in Table 2 and 3 along with the computed D 308 268 (d) values. Fig. 11 plots pseudocontact shi isosurfaces in 1Dy and 1Er as examples of lanthanides with oblate and prolate 4f density shapes, respectively.
The reliability of the point charge model in calculations of the magnetic susceptibility can be evaluated comparing experimental and computed d pc shis (Fig. 12). As experimental values we use the results of Reilley's approach for 13 C-PHHJ signal and the values from the 2-nuclei tting for 45 Sc shis. Details for the 13 C-THJ shis are not discussed in the text but can be found in the ESI Fig. S11. † As can be seen in Fig. 12, reasonably good correlation exists between experimental and computed data. If 1Tm and 1Tb are not included in the set, the linear correlation with the intercept set to zero gives R 2 values of 0.98 and 0.99 for the PHHJ and 45 Sc shis, respectively. Note also that the 45 Sc d pc shis from the 2nuclei model are very close to the total d para shis. Thus, we can tentatively use 45 Sc chemical shis of paramagnetic mixedmetal nitride clusterfullerenes to evaluate the size of the magnetic anisotropy of the endohedral lanthanide ions without rather complicated and sometimes hardly possible estimation of the contact term.
It should be also noted that the slope of the linear t in Fig. 12 is close to 0.9 for 45 Sc but is near 1.4 for 13 C-PHHJ. The slope for an analogous t for the 13 C-THJ signals (see ESI Fig. S11 †) is close to 0.9. As follows from eqn (1) and our earlier studies of the Ho-based NCFs, 20a the structural factor also plays an important role in determining the d pc values, and dynamic nature of the 13 C shis in 1M molecules (i.e. averaging over the cluster rotation) makes calculation less straightforward than for the 45 Sc shis. For instance, in 1Ho the term (3cos 2 q i À 1)/ 12pR i 3 (denoted hereaer as G i ) computed in this work at the B3LYP level is À3.30 Â 10 À4 /À1.29 Â 10 À3ÅÀ3 for the PHHJ/ THJ carbons, respectively. At the same time, our recent PBE computations using Y as a model for Ho gave the values of À7.37 Â 10 À4 /À5.59 Â 10 À4ÅÀ3 , respectively. Along the 1M series, the G i values for carbon atoms vary by 15-20%, from À3.91 Â 10 À4 /À1.13 Â 10 À3ÅÀ3 in 1Ce to À3.16 Â 10 À4 /À1.31 Â 10 À3ÅÀ3 in 1Er. Since the slope of the linear t between experimental and calculated d pc values for THJ carbons is closer to 1, we propose that B3LYP level gives correct values for the THJ carbons, whereas G i factors for the PHHJs are underestimated (and therefore the slope is close to 1.4). Thus, the use of paramagnetic 13 C chemical shis for analysis of the magnetic anisotropy of the endohedral lanthanide ions meets certain difficulties caused by the uncertainties in the structural factor when endohedral cluster is rotating on the NMR time scale. On the contrary, the G i term for 45 Sc is almost independent on the level of theory employed to calculate it. Moreover, its variation along the whole 1M series is below 1% because the shortening of the M-N distance in the cluster is balanced by the elongation of the Sc-N distance. Such indifference to the structural factor makes the 45 Sc chemical shis more suitable for the analysis of the magnetic anisotropy in mixed-metal nitride clusterfullerenes.
Computed D 308 268 values reasonably agree with experimental data for THJ carbon and Sc signals, but are rather far from the experimental values of the PHHJ carbons. The latter is likely to Fig. 12 Correlation between computed and experimental chemical pseudocontact shifts: 13 C-PHHJ (left) and 45 Sc (right). Experimental d pc ( 13 C) shifts are obtained by Reilley's approach, whereas 45 Sc values are estimated using 2-nucleus method, see text for further details. Solid lines is a linear fit for a complete set of data (R 2 ¼ 0.94 for both PHHJ and Sc), whereas red dashed lines were obtained for fitting without 1Tm and 1Tb values. The intercept was set to zero in linear fits. be due to the larger relative contribution of the contact shi for the PHHJ carbons, which is not taken into account in PCM computations of D 308 268 . This is especially well seen for the 1Tm: based on its magnetic anisotropy, negative D 308 268 value is expected for the pseudocontact shi, but positive for the contact shi. A small positive D 308 268 value observed experimentally for the 1Tm-PHHJ carbon is therefore due to the compensation of both terms. On the contrary, the pseudocontact contribution in the 1Tm-THJ chemical shi is much larger than the contact counterpart, and the experimental D 308 268 value is negative in good agreement with the results of computations.
1Tm and 1Tb are important deviations from the general trend (Fig. 12). Especially noticeable is the 1Tb value for the 45 Sc shi, which is predicted to be much more positive than experimentally observed (Fig. 12). In other words, whereas the point charge model predicts that the LF splitting in 1Tb and 1Dy is quite similar (Fig. 10), from the 45 Sc NMR spectra it follows that the LF splitting in 1Tb should be considerably smaller. Interestingly, recent ab initio CASSCF calculations predict that the LF splitting in 1Tb is indeed smaller than in 1Dy, 7b which would give better agreement with experimental NMR shis. Similarly, correct prediction of the experimental values for 1Tm require reliable estimation of a considerable number of the LF splitting levels (Fig. 10), rather than only 1-2 excited states needed for other 1M molecules. Inability of the point charge model to correctly treat all states may be also the reason of the noticeable deviations well seen for 1Tm values in Fig. 12. Therefore, both 1Tb and 1Tm values provide convenient training set to be considered in future computational studies. At the current moment we can conclude that the point charge model gives qualitatively correct predictions of the LF splitting in lanthanide EMFs but fails to describe more subtle effects. More advanced approaches would be needed to address this problem, 58 but this task goes beyond the scope of this work.

Conclusions
In this work we have performed the rst systematic paramagnetic NMR study of MSc 2 N@C 80 -I h molecules with M running through all lanthanides capable of forming nitride clusterfullerenes (M ¼ La, Ce, Pr, Nd, Tb, Dy, Ho, Er, Tm, Lu). Analysis of the whole set of data enabled separation of the contact and pseudocontact contributions to the paramagnetic shis. We showed that the contact shi of 13 C nuclei may be rather large, although pseudocontact term is still larger. For the 45 Sc shis, the pseudocontact term dominates. Since pseudocontact term is directly dependent on the magnetic anisotropy, the 45 Sc NMR spectra are found to be especially useful for the studies of magnetic properties of lanthanide ions in clusterfullerenes.
Interpretation of the paramagnetic NMR data required modelling of the ligand led splitting in MSc 2 N@C 80 molecules, which was accomplished using the point charge model. Although quite simple, this approach provided semi-quantitative agreement with experimental data. Considerable deviations between experimental and computed chemical shis are found only for TmSc 2 N@C 80 and TbSc 2 N@C 80 . We showed that the main contribution to the ligand eld is from the nitride ion, which results in an almost uniaxial ligand eld. For the lanthanides with oblate shape of the 4f density (i.e. all lanthanides except for Er and Tm) the ground magnetic state in MSc 2 N@C 80 has the largest J z projection with rather large gap to higher energy states (with smaller m J values). It shows that MSc 2 N@C 80 molecules with Kramers lanthanide ions are especially promising in the eld of single molecule magnetism. Yet, more rened treatment of the ligand eld splitting might be needed to fully account for all experimental data.