Systematic analysis of electron energy-loss near-edge structures in Li-ion battery materials

Electrical conductivity, state of charge and chemical stability of Li-ion battery materials all depend on the electronic states of their component atoms, and tools for measuring these reliably are needed for advanced materials analysis and design. Here we report a systematic investigation of electron energyloss near-edge structures (ELNES) of Li-K and O-K edges for ten representative Li-ion battery electrodes and solid-state electrolytes obtained by performing transmission electron microscopy with a Wien-filter monochromator-equipped microscope. While the peaks of Li-K edges are positioned at about 62 eV for most of the materials examined, the peak positions of O-K edges vary within a range of about 530 to 540 eV, and the peaks can be categorised into three groups based on their characteristic edge shapes: (i) double peaks, (ii) single sharp peaks, and (iii) single broad peaks. The double peaks of group (i) are attributable to the d electronic configuration of their transition metal ions bonded to O atoms. The origin of the different peak shapes of groups (ii) and (iii) is more subtle but insights are gained using density functional theory methods to simulate O-K ELNES edges of group (ii) material LiCoO2 and group (iii) material LiFePO4. Comparison of their densities of states reveals that in LiCoO2 the Co–O hybrid orbitals are separated from Li–O hybrid orbitals, resulting in a sharp peak in the O-K edge, while Fe–O, Li–O and P–O hybrid orbitals in LiFePO4 partially overlap each other and produce a broad peak.


Introduction
Li-ion batteries (LIBs) are widely used in a myriad of applications such as portable devices, electric vehicles and renewable energy storage systems. In an effort to develop superior LIBs, many candidate electrode materials have been investigated in terms of energy density, output voltage, charge-discharge rates and cycle lifetimes. [1][2][3] At the same time, efforts to improve battery safety, compactness and stability have led to the development of solid electrolytes that exhibit good Li-ion conductivities and structural integrity after many charge-discharge cycles when used in all-solid-state LIBs. 4 At the microscopic level, charging and discharging of LIBs involves diffusion of Li atoms, conduction of electrons and changes in valence states of the transition-metal (TM) atoms. Characterisation of the structural and chemical changes that take place during battery operation is thus of fundamental importance in the development of improved LIB materials.
Electrodes and electrolytes in LIBs typically contain TM and Li cations bonded to anions such as O, S and sometimes F. Energy-loss near-edge structure (ELNES) spectra of these elements, obtained by performing electron energy-loss spectroscopy (EELS) within a transmission electron microscope (TEM), can provide useful data about their energetics and coordination environments with a spatial resolution of a few nanometres. [5][6][7][8][9][10][11][12] In this work we focus on a number of actively investigated oxide LIB materials, some of whose EELS spectra have not yet been reported.
TM-L 2,3 ELNES spectra, ranging over energies between 450 and 900 eV depending on the element in question and its coordination environment, 13 are widely used to characterise TM valence states because TM-L spectra change markedly when the valence state changes. [14][15][16] Similarly, the shapes of O-K ELNES at around 530 eV and Li-K ELNES at around 60 eV depend strongly on the local crystal environments of these atoms and the nature of the interatomic bonding. 11,17,18 In these cases, ELNES simulations using density functional theory (DFT) are indispensable for identifying the electronic transitions producing the fine structures observed experimentally. 7,11,19,20 In the low-energy region (less than 100 eV) in which Li-K edges appear, high energy resolution is necessary to obtain peak shapes accurately. While measurements with resolutions up to 0.3 eV (as determined by the full width half maximum (FWHM) of the zero-loss peak) can be obtained using cold-FE guns, 21 to obtain resolutions better than 0.1 eV requires use of monochromators. [22][23][24] The latter allows Li-K edge spectra to be used to measure Li concentrations and distributions quantitatively with nanoscale resolution, as reported for materials LiCoO 2 , 25,26 LiFePO 4 27,28 and Li 2 MnO 3 . 29 One drawback of this method, however, is that TM-M 2,3 (or TM-N 2,3 ) edges lie near to Li-K edges, making their peaks difficult to deconvolute. Although measurements of Li-K and O-K edges of many LIB materials have been reported, 14,15,[30][31][32][33][34][35][36] to the best of our knowledge, the dependence of the ELNES spectra on the type of TM, the crystal structure, and the bonding environment has not been studied systematically. This lack of systematic comparison has sometimes led to conflicting results in the literature, and hampers deeper understanding of the relationship between electronic structure, chemical environment and electrochemical properties of these materials.
In this study, using a monochromator-equipped TEM, we carried out a systematic comparison of the effect of different TM elements and crystal environments on Li-K and O-K ELNES spectra for ten typical LIB-related materials. The structure types of each material and the coordination numbers (CN), valence states and d-electron configurations of their component TM elements are listed in Table 1, with representative crystal structures depicted in Fig. 1. The materials examined include six cathode materials, two solid electrolyte materials, one anode material, and one non-conductive material (as reference). The cathode materials include two layer-type materials, LiCoO 2 and Li 2 MnO 3 ; two olivine-type materials, LiFePO 4  These materials all exhibit differently shaped EELS spectra, especially in the case of O-K ELNES. To understand the origins of these differences, we carried out DFT-based calculations of LiCoO 2 and LiFePO 4 , which exhibit representative spectrum shapes. Although several theoretical investigations have been reported for both compounds, 20,47-49 the reported results are not sufficiently detailed to provide explanations for the differences in ELNES peak positions and shapes. In particular, electronic structures of excited states have not yet been reported, so in this study we performed ELNES simulations of LiCoO 2 and LiFePO 4 taking into account the electronic structures of the excited state, and compared the results with experimental spectra.

Specimen preparation
Samples of the ten compounds were prepared by different methods, as described below: (1) LiFePO 4 . Powder was prepared by solid-state reaction between stoichiometric amounts of Li 2 CoO 3 , FeC 2 O 4 Á2H 2 O and (NH 4 ) 2 HPO 4 at 600 1C for 6 h, followed by annealing at 800 1C for 24 h under pure Ar gas flow. 50 (2) LiCoPO 4 . A thin film of LiCoPO 4 was prepared via an aqueous sol-gel route using citric acid as the chelating agent. 51 Stoichiometric amounts of CH 3 COOLi, Co(NO 3 ) 2 Á6H 2 O and NH 4 H 2 PO 4 were dissolved in distilled water and citric acid was added to the solution. The precursor solution was spread on an Au-coated Al 2 O 3 substrate by spin-coating and annealed first at 350 1C for 1 h and then at 700 1C for 1 h.  (3) LiCoO 2 . LiCoO 2 single crystals were prepared via a flux method using LiCoO 2 , Li 2 O and LiCl. The mixture was heated at 900 1C for 10 h, cooled to 700 1C at a rate of 2 1C h À1 and then allowed to cool naturally down to room temperature. Platelet-like LiCoO 2 single crystals were separated by rinsing the frozen flux in hot water.
(4) Li 2 MnO 3 . Li 2 MnO 3 nanocrystals were fabricated by a hydrothermal method. LiOHÁH 2 O was dissolved in distilled water and g-MnO 2 powder was added to the solution. The mixture was heated at 180 1C for 24 h in an autoclave, centrifuged, washed with water and ethanol, and dried at 120 1C.  53 The deposited film was annealed at 900 1C.
(8) LiNbO 3 . LiNbO 3 powder was prepared by a conventional solid-state reaction method from Li 2 CO 3 and Nb 2 O 5 by sintering at 1000 1C for 5 h. 54 (9) Li 0.33 La 0.56 TiO 3 . Li 0.33 La 0.56 TiO 3 powder was prepared by a solid-state reaction method by sintering mixtures of Li 2 CO 3 , TiO 2 and La 2 O 3 . 55 (10) Li 0.14 La 0.29 NbO 3 . A single crystal of Li 0.14 La 0.29 NbO 3 was fabricated by a conventional unidirectional solidification method. 54 LiNbO 3 and LaNb 3 O 9 precursor powders were mixed, compacted into a cylindrical shape and sintered at 1200 1C for 20 h. The sintered polycrystalline Li 0.14 La 0.29 NbO 3 rod that resulted was melted in a Pt crucible at 1380 1C and then gradually passed downward through a unidirectional solidification furnace.
X-ray diffraction measurements were used to confirm that materials (1) to (10) were each prepared successfully as a single phase with the appropriate crystal structure under ambient conditions (see Fig. S1 in ESI †).

EELS measurements
TEM specimens of thin films of LiCoPO 4 , LiMn 2 O 4 and Li 4 Ti 5 O 12 were prepared using low-temperature (liquid N 2 -cooled to below À100 1C) and low-voltage (3.0 kV to 0.2 kV) ion-milling (PIPS, Gatan Inc.). Special care was taken to minimise damage to the samples, and surfaces were cleaned using a plasma cleaner (Solarus, Gatan Inc.) to remove residual amorphous phases. TEM specimens of the other bulk samples were prepared by careful crushing, and the resulting powders dispersed and deposited onto lacey carbon films (above copper substrates).
EELS measurements were carried out using a JEM-2400FCS atomic resolution microscope (JEOL Ltd.) equipped with a spherical-aberration corrector (CEOS GmbH), an EELS spectrometer (GIF Tridiem ERS, Gatan Inc.) and a Wien-filter type monochromator. The microscope was installed in a room designed and constructed to minimise AC magnetic fields, floor vibrations, air-flow volume, temperature fluctuations and acoustic vibrations. In this stable environment, a high energy resolution below 30 meV was achieved. 24 EELS spectra were recorded with an accelerating voltage of 60 kV to reduce irradiation damage and a convergence/collection semi-angle of 30/35 mrad. The energy resolution, DE, specified in terms of the FWHM of the zero-loss peak, was controlled by changing the width of an energy selection slit in the monochromator. To examine the effect of energy resolution on ELNES spectra, we measured Li-K edges of LiFePO 4 , LiCoPO 4 , LiCoO 2 and Li 4 Ti 5 O 12 under various slit conditions, viz., no slit, slit of width 4 mm, and slit of width 1 mm, keeping the dispersion value constant at 0.05 eV per pixel. The resulting DE were 0.7 eV, 0.35 eV and 0.1 eV, respectively.
Measurements to compare Li-K edges of all materials were performed using a dispersion value, slit width, and measurement exposure time of 0.05 eV per pixel, 1 mm and 0.2 s, respectively. Similarly, for O-K edges, values of 0.1 eV per pixel, 4 mm and 1 s, respectively, were used. The dispersion value of 0.1 eV per pixel used for O-K edge measurements was larger than that for Li-K edges, but is comparable to the natural line width of the O-K edge, 0.12 eV. 56 These conditions provided resolutions of DE = 0.1 and 0.3 eV for Li-K and O-K edges, respectively. Integration times per measurement for both edges were 10-20 s. Energy alignment was carried out by measuring the zero-loss peak just before and after recording edge spectra. The standard deviation of the zero-loss peak position was 0.03 eV over a time span of 10 minutes (see Fig. S2 in ESI †).

ELNES simulations
Theoretical O-K ELNES spectra of LiCoO 2 and LiFePO 4 were generated from first-principles calculations using the DFT+U method including spin polarisation. 57 In all calculations, the GGA-PBE functional was used for the exchange-correlation term. 58 Values of U of 4.91 eV and 3.71 eV, respectively, were applied to d electrons in Co 3+ in LiCoO 2 and Fe 2+ in LiFePO 4 . 59 In LiFePO 4 , although the most stable spin configuration is antiferromagnetic, the temperature of the anti-ferromagnetic-toferromagnetic transition is below room temperature, 49 so we assumed LiFePO 4 to be in the ferromagnetic state for the calculations.
Before generating ELNES spectra, lattice parameters of LiCoO 2 and LiFePO 4 were determined by performing structural optimisation using the VASP code 60 with a planewave cut-off energy of 500 eV. The hexagonal R% 3m unit cell (12 atoms) of LiCoO 2 was optimised using a 15 Â 15 Â 3 Brillouin zone k-point mesh, and the orthorhombic Pnma unit cell (28 atoms) of LiFePO 4 using a 6 Â 9 Â 15 k-point mesh.
O-K ELNES spectra were generated using the core-hole method implemented in the WIEN2k code with an all-electron basis set (APW + lo method). 61 In this method, the transition probabilities were calculated using a supercell with an electron removed from the 1s orbital of an O atom (core hole) and placed in the conduction band. 62 According to the spin selection rule, we need to consider the probability of each spin state of the excited electron. In the case of an up-spin excited electron, only up-spin orbitals are relevant to the transition calculation, and vice versa for down-spin excited electrons. Because LiCoO 2 in its ground state is non-magnetic, the probabilities of up-spin excitation and down-spin excitation are identical, so we only performed calculations for an up-spin excitation. In the case of LiFePO 4 , although up-and down-spin excitations are not equivalent, our test calculations revealed that the differences between the two excited states are small. The theoretical ELNES spectra of LiFePO 4 reported here were generated by averaging contributions of up-and down-spin orbitals from calculation of a down-spin excitation.
Muffin tin radii (R MT ) for Li, O, P, Co and Fe were 1.70, 1.60 (LiCoO 2 )/1.45 (LiFePO 4 ), 1.45, 2.00 and 2.10 bohr, respectively. The cutoff parameter that determines the maximum wave number of plane waves, R k max , was set at 7.0, and O-1s states were treated as core levels for both LiCoO 2 and LiFePO 4 . A 4 Â 4 Â 1 (192 atom) supercell was constructed from the LiCoO 2 unit cell and a 1 Â 2 Â 2 (112 atom) supercell from the LiFePO 4 unit cell, with 3 Â 3 Â 2 and 4 Â 3 Â 3 k-point meshes, respectively, for integrating over the Brillouin zone. In LiFePO 4 , as there are three non-equivalent O sites, we calculated the contribution of each type of O separately and took the weighted average when calculating the projected density of states (PDOS) of O-p orbitals and ELNES spectra. The core level of the most distant O atom from the excited O atom in each supercell was used as reference to adjust the PDOS position. The transition energy obtained from the difference in total energies between ground and excited states was used to adjust ELNES edge positions.

Effect of energy resolution on Li-K ELNES
The effect of different energy resolutions, i.e., DE = 0.7, 0.35 and 0.1 eV, on EELS spectra in the low-loss region around Li-K edges was examined for LiFePO 4 , LiCoPO 4 , LiCoO 2 and Li 4 Ti 5 O 12 (Fig. 2). As can be seen in Fig. 2(c) and (d), the positions and shapes of the strongest peaks change little with DE in the cases of Li 4 Ti 5 O 12 and LiCoO 2 . In the case of LiCoPO 4 , when DE = 0.35 eV, a small trough appears at higher energy than the largest peak, around 64 eV, and when DE = 0.1 eV, small peaks appear on both sides of this trough as indicated by the blue arrow in Fig. 2(b). The splitting of the peak on the left side of the trough from the largest peak is accompanied by a shift of the largest peak to slightly lower energy, allowing the edge position to be determined with higher accuracy. In the case of LiFePO 4 , the single peak at 58.2 eV apparent when DE = 0.7 eV changes into a main peak and shoulder peak at 57.6 and 58.3 eV, respectively, when DE = 0.1 eV, as indicated by the red arrows in Fig. 2(a).
In the cases of LiFePO 4 , LiCoPO 4 and Li 4 Ti 5 O 12 , pre-edge peaks become more distinct with decreasing DE as indicated by green arrows in Fig. 2(a), (b) and (d). In LiCoO 2 and LiFePO 4 small bumps appear in the tails of the largest peaks at higher energies when DE r 0.35 eV, as indicated by blue arrows in Fig. 2(a) and (c). Although such peaks are small, they are useful for checking the accuracy of calculated spectra. For example, the spectral shape including tail peaks of LiCoO 2 has been verified by DFT calculations using the Z + 1 method by Kikkawa et al. 30 In the case of LiFePO 4 , while the largest peak in Fig. 2(a) at around 58 eV corresponds to the Fe-M 2,3 edge, the peaks above 60 eV indicated by the blue arrows are thought to be components of the Li-K edge because it has been reported that the intensities of these peaks decrease with decreasing Li content. 32 In this case, performing EELS measurements with high resolution becomes critically important as such features may otherwise be missed.  peak positions of TM-M 2,3 (or N 2,3 for Nb) edges and Li-K edges are indicated by grey and black inverted triangles, respectively. The weak peaks assumed to be Li-K edges in the spectra of Li 0.14 La 0.29 NbO 3 and Li 0.33 La 0.56 TiO 3 are indicated by open inverted triangles. Fig. 3(b) shows a magnified view of the energy region around the Li-K edges after background subtraction. TM-M 2,3 (or N 2,3 for Nb) edges have also been subtracted, except for the Fe-M 2,3 edge for LiFePO 4 . Table 2 compares these peak positions with data available in the literature. 13,14,30,[32][33][34]36 Agreement to within 1 eV was obtained in all cases.

Comparison of Li-K ELNES
TM-M 2,3 (Nb-N 2,3 ) edges are known to correspond to transitions from 3p 3/2,1/2 (or 4p 3/2,1/2 ) core states to the unoccupied conduction bands. A systematic comparison of the spectra reveals peak positions shift to higher energies as the atomic number increases. This trend reflects the shift to lower core energy levels with increasing atomic number.
In LiNi 0.5 Mn 1.5 O 4 , LiFePO 4 , LiCoPO 4 and LiCoO 2 , the M 2,3 edges of Ni, Fe and Co are close to and partially overlap the Li-K edges, complicating detection of the latter. As mentioned in the previous section, in the case of LiFePO 4 , we cannot assign the weak protrusion at around 62 eV to the Li-K edge without comparison with measurements from delithiated samples. Similarly, the Li-K edge in LiCoO 2 has been identified by comparison with spectra from delithiated samples. 30 In the case of LiCoPO 4 , no such data are available, so we here assume the peak at around 62 eV to be the Li-K edge, similar to the case for LiCoO 2 . The Ni-M 2,3 edge of LiNi 0.5 Mn 1.5 O 4 is assumed to occur at energies higher than the Li-K edge by several electronvolts because in Ni metal it appears at 68 eV. 13 Based on this, the small protrusion at 61.9 eV to the right of the Mn-M 2,3 edges was assigned to the Li-K edge.
In the magnified spectra of Li 0.14 La 0.29 NbO 3 and Li 0.33 La 0.56-TiO 3 , very slight peaks are visible at around 60 eV. However, it is difficult to distinguish these Li-K edges from Nb-N 1 and Ti-M 1  edges which also exist at around 60 eV, 34 so it is difficult to identify the Li-K edges with certainty. The chief reason for this is the lower Li concentrations in these solid electrolytes compared to the other materials; the volume densities of Li atoms in these two materials are approximately 2.3 and 5.7 atoms per nm 3 , respectively, while those of the other materials are between 13 and 40 atoms per nm 3 . Previous studies of other Li-containing materials by ELNES or XANES showed that the positions of Li-K edges can vary by more than 5 eV depending on the particular chemical environment of the Li atoms, as reflected in the Mulliken electronegativities or oxidation numbers of neighbouring atoms. 18,63 In our work, the main Li-K peaks were observed to lie at around 61-62 eV for all materials except Li 0.33 La 0.56 TiO 3 and Li 0.14 La 0.29 NbO 3 . This suggests that, although there are differences between the compounds, such as the CN of Li, Li-O bond lengths, and the types of second nearest neighbour cations, the positions of the main peaks are only slightly affected by the local environments around Li.
For some materials, the fine structures of Li-K edges contain small pre-edge peaks and/or sub-peaks above the main peak. These structures are expected to contain valuable information about the chemical bonding states in these compounds. It remains difficult, however, to extract these details because of the overlap with TM-M 2,3 edges mentioned above. Furthermore, although DFT calculations can help explicate these fine structures, simulation of a Li-K edge requires a calculation scheme that includes electron-hole coupling effects, e.g., the Bethe-Salpeter equation (BSE) method, 64 which is beyond the scope of the present work.  Table 3, the peak positions are listed for all ten materials and compared with available reported values. In all cases, differences in energy are less than 2 eV.

O-K ELNES
The observed trends can be partly interpreted in terms of the effect of TM-d orbitals overlapping with neighbouring O-2p orbitals. As reported in Table 1, TM atoms in group (i) materials have a d 0 configuration, while TM atoms in the other two groups have partially occupied d orbitals. Since all TM atoms of the materials examined here are in octahedral coordination environments, their d orbitals are split into t 2g and e g bands.
In group (i) materials, the t 2g and e g bands hybridise with unoccupied O-2p orbitals to form part of the conduction band. Such hybridisation results in splitting of the O-2p states into which an O-1s electron is excited, resulting in the double peak in the O-K ELNES spectra.  In the case of group (ii) materials, the origin of the single sharp peak and small sub-peaks can also be explained in terms of the TM-d orbitals, with d orbitals in the conduction band splitting by different amounts as a result of their partial occupation (for details see Section S3 in ESI †).
O-K ELNES spectra of group (iii) materials contain only one broad peak, and are distinctly different to those of group (ii) materials despite also containing partially occupied TM-d orbitals. In addition, the peak positions of group (iii) materials are energetically higher than those of the other groups. These differences cannot be explained satisfactorily using a simple model of the electronic structure, so we simulated the ELNES of a group (ii) material, LiCoO 2 , and a group (iii) material, LiFePO 4 , using the core-hole method within the framework of DFT, analysing and comparing their electronic structures in detail. In ground-state LiCoO 2 , the valence band consists of Co-3d (t 2g ) and O-2p orbitals and the conduction band of Co-3d (e g ), O-2p and Li-2p orbitals. 30,47 The octahedral environment of the Co atom causes Co-d orbitals to split into fully occupied t 2g states and unoccupied e g states. The calculated band gap of 2.5 eV is consistent with previous DFT calculations. 48 Formation of an O-1s hole reduces the symmetry reduction of CoO 6 octahedra resulting in splitting of Co-d (e g ) orbitals hybridised with 2p orbitals of the excited O atom. This splitting lowers the conduction band minimum by about 1.5 eV, resulting in a lowering of the energy of the O-K edge, and produces the very small shoulder-like sub-peak in the tail of the main peak in the ELNES. Other changes in the shape of the broad hybridisation band between O-2p and Li-2p orbitals at energies higher than 7 eV in the excited state are also noticeable in the PDOS plots in Fig. 5(b). Since the O-K ELNES spectrum is produced by the transition of an O-1s electron to unoccupied O-p states, Fig. 5(b) shows that the single sharp ELNES peak at around 532 eV and the broad peak at energies higher than 536 eV originate from hybridisation of O-p orbitals with Co-d and Li-p orbitals, respectively.

O-K ELNES simulations of LiCoO 2 and LiFePO 4
Experimental and calculated O-K ELNES spectra of LiFePO 4 using DE = 0.3 eV are compared in Fig. 6(a). Fig. 6(b) shows PDOSs of O-p, Fe-d, P-p and Li-p orbitals in ground and excited states in the same manner as for LiCoO 2 in Fig. 5(b) where PDOSs of the excited O atom and Fe, P and Li atoms neighbouring it have been averaged over contributions from crystallographically independent sites. The calculated O-K ELNES spectrum in Fig. 6(a) has a single broad peak and appears more similar to the experimental spectrum compared to those previously calculated using the Z + 1 method 19 or from dielectric function calculations. 20 This result demonstrates the efficacy of using the core-hole method to simulate the ELNES spectra with reasonable accuracy. 62 In the ground state, while the top of the valence band of LiFePO 4 consists of O-2p and Fe-3d orbitals, in the energy range from 3 to 5 eV above the valence band maximum (VBM), the conduction band mainly consists of O-2p and Fe-3d orbitals and the region with energies higher than 5 eV consists of O-p, P-p and Li-p orbitals ( Fig. 6(b)). The relatively large differences between up-and down-spin contributions in PDOSs of Fe-d and O-p orbitals reflect the high-spin states of Fe 2+ ions with a d 6 electronic configuration, i.e., (t 2g ) 4 (e g ) 2   When an electron is elevated from an O-1s shell in LiFePO 4 , the PDOSs of O-p, Fe-d and P-p orbitals at the bottom of the conduction band (around 3 to 7 eV above the VBM) change dramatically, while their PDOSs at higher energies undergo relatively little change. Introduction of the core hole shifts the bottom of the conduction band, which in the ground state consists mainly of Fe-d electrons between 3 and 5 eV above the VBM, higher by about 1.5 eV. The PDOS of P-p orbitals between 5 and 7 eV above the Fermi level also increases dramatically, a consequence of the strong covalent bonding between P and O atoms. The band gap, however, remains essentially unchanged because Fe-d orbitals at the top of the valence band in the ground state are elevated to the bottom of the conduction band in the excited state. These changes enhance the energetic overlap between the Fe-O, Li-O and P-O hybrid orbitals, resulting in the characteristic broad peak in the O-K ELNES of LiFePO 4 , in contrast to that of LiCoO 2 . The broad single peaks observed for group (iii) materials (LiFePO 4 and LiCoPO 4 ) can thus be attributed to the presence of tetrahedral PO 4 units in addition to the TM-O polyhedra. Group (ii) materials, which lack PO 4 tetrahedral units, exhibit single, sharp O-K edges regardless of the overall crystal structure, e.g., layered-rocksalt in the case of LiCoO 2 or spinel in the case of LiMn 2 O 4 .
Although the band gaps of LiCoO 2 and LiFePO 4 are almost the same, the peak positions of O-K edges in LiCoO 2 and LiFePO 4 differ by about 6.7 eV. Our ELNES simulations also demonstrated a similar trend, with a calculated difference in peak positions of 4.9 eV. Analysis of the electronic structures showed that this difference arises from the difference in changes in the conduction bands in the excited state. As mentioned above, the peak in the O-K edge of LiCoO 2 corresponds to the conduction band minimum composed of hybrid O-Co orbitals, and decreases by about 1.5 eV upon introduction of a core hole. In the case of LiFePO 4 , the conduction band minimum is not decreased when a core hole is introduced, and neither are there isolated hybrid O-TM orbitals. This difference provides one explanation for why the main peak of the O-K edge in LiFePO 4 appears at higher energy than in LiCoO 2 .

Conclusions
Li-K and O-K ELNES spectra of ten different representative LIB materials were systematically investigated by EELS using an aberration-corrected TEM equipped with a monochromator. Measurements with a high energy resolution of DE = 0.1 eV revealed fine structures in Li-K ELNES spectra that can otherwise remain hidden. The peak positions of Li-K edges were found to be about 62 eV for most of the LIB materials because of the similar coordination environments of their Li ions. In contrast, the peak positions of O-K edges varied over a range of about 10 eV.
The shapes of O-K edges could be divided into three groups. Group (i) materials contained octahedrally coordinated TM ions with empty d orbitals, and these materials produced ELNES with double peaks at the O-K edges because of splitting of TM-d orbitals into t 2g and e g levels. Group (ii) and (iii) materials both contained TM ions with partially filled d orbitals, but the shapes of their O-K edges were found to be distinctly different, the former with a single sharp peak and the latter with a single broad peak.
ELNES simulations based on DFT revealed that the single sharp peak in the O-K edge of group (ii) material LiCoO 2 results from hybridisation of O-p and Co-d e g orbitals with energies close to the conduction band minimum. In contrast, the broad peak of the O-K edge of group (iii) material LiFePO 4 results from hybridisation between O-p orbitals with P-p, Fe-d and Li-p orbitals in the conduction band at similar energy levels.
Our systematic analysis of EELS spectra demonstrates how a combination of high-resolution ELNES measurements and simulations based on DFT calculations provides useful insights into the electronic structures and coordination environments of atoms in LIB materials.

Conflicts of interest
There are no conflicts to declare.