![](/logos/buttonlogos/cross.png)
![](/logos/buttonlogos/cross.png)
X-ray spectroscopy for functional materials
![Open Access](/logos/open.png)
Effectiveness of ab initio in simulating spectra from layered systems
aConsiglio Nazionale delle Ricerche, Istituto Officina dei Materiali – OGG, c/o ESRF, Grenoble, France, and bDepartment of Chemical and Materials Engineering, New Uzbekistan University, Tashkent, Uzbekistan
*Correspondence e-mail: francesco.dacapito@cnr.it
The simulation of ab initio methods is discussed. The procedure for producing the spectra is presented as well as an application to a two-dimensional material (WSe2) where the effectiveness of this method in reproducing the spectrum and the linear dichroic response is shown. A series of further examples in which the method has been employed for the structural determination of materials are given.
spectra of thin films viaKeywords: EXAFS; ab initio molecular dynamics; density functional theory; thin films.
1. Introduction
The study of functional materials requires detailed knowledge of the structure of the systems under investigation. ab initio structural simulations can be helpful, as modern density functional theory (DFT) methods allow structures to be simulated to an accuracy of a few % for cell dimensions (Csonka et al., 2009) or interatomic distance values that match well with the accuracy of the In several cases, a simple comparison between the numerical results from DFT (namely, shell radii) is not sufficient due to the complexity of the site under study. If we consider a dopant in a matrix there are several possibilities for its incorporation site: substitutional, interstitial, coupled to vacancies, etc. These structures can be simulated but, in order to assess the occurrence of one of them in the sample under study, a description of the site with several coordination shells is necessary. This would be complex to carry out using classical methods (too many parameters like N, R and σ2 for the many shells to compare numerically) whereas direct comparison with the related simulated spectra could easily solve the task. Indeed, a site can be identified by the various features that appear in the spectrum and that result from the complex superimposition of all the (most significant) scattering paths of the photoelectron. The problem is that for a reliable calculation the dynamics of the structure under study also need to be considered as the damping of single contributions critically depends on atomic motions and their mutual correlations.
Molecular dynamics (MD) has been used for simulating et al., 1994; Palmer et al., 1996
; Kuzmin et al., 1997
) on ions in liquid that were based on classical MD using interatomic pair potentials. This approach is particularly effective in terms of the required computational resources, and permits the use of cells containing hundreds of atoms and long thermalization times (tens of picoseconds). On the other hand, the availability of appropriate pair potentials is needed or even their optimization/refinement on the system under study (Kuzmin & Evarestov, 2009
; Anspoks et al., 2012
), and the level of agreement between theory and experiment is remarkable.
For treating the general case, ab initio DFT simulations can be used by various approaches. A method has been proposed (Vila et al., 2007, 2008
, 2012
, 2018
) that theoretically calculates the Debye–Waller factors of the paths to be used for the simulation of the spectrum with a considerable level of success (Veronesi et al., 2010
). Another approach consists of running an ab initio (AIMD) cycle and then calculating the spectra from a series of snapshots of the cycle (Bocharov et al., 2016
, 2020
). A comparison of ab initio generated spectra can also be used to validate ad hoc generated potentials as shown by Shapeev et al. (2022
). Also, in these cases the quality of reproduction of the experimental data can be impressive even if the computational resources needed are considerable. MD can also be used to simulate theoretical spectra from tentative structures to interpret experimental data. In this case a close agreement is not required, as long as the candidate structures bear clear differences and the requirements on computation resources can be relaxed. An example of a structural choice made by comparing experiment and AIMD has been given by Cartechini et al. (2011
) where the focus of the study was the incorporation site of Zn in lead antimonate used for the production of the widespread historical paint pigment Giallo Napoli. The spectrum at the Zn K-edge [see Fig. 6(a) of Cartechini et al. (2011
)] showed features that are well reproduced by the simulated spectrum of Zn substituting for Sb and were in anti-phase with respect to the simulated spectrum of Zn substituting for Pb. This showed a counter-intuitive result, as it would be expected that the divalent Zn ion would substitute for the divalent Pb rather than Sb. Further examples of structural determination obtained from comparison of experimental spectra and AIMD have been given by d'Acapito et al. (2016
) regarding the incorporation site of Er3+ ions in CaF2 crystals and by Kopula Kesavan et al. (2020
) regarding the local structure around Ag in AgxBi1–xS2 nanoparticles.
Regarding layered systems, an example of an investigation has been given by d'Acapito et al. (2020), focusing on the structure of [(GeTe)2/(Sb2Te3)m]n superlattices used for an innovative class of materials for non-volatile memories called interface phase change random access memories (IPCRAMs). In particular, the debate concerned the real structure (i.e. piling order of Ge, Te and Sb layers) occurring in the system among the different ones proposed in the literature (Tominaga et al., 2014
) and on the possible presence of inter-diffusion in the GeTe layer with Sb/Ge substitutions (Casarin et al., 2016
). The method was first validated by reproducing the spectrum of the GeTe bulk structure which revealed a considerably close agreement. Then, the structures proposed were simulated by AIMD, and the related calculated spectra were compared with experiment [see supplementary Fig. S9 in d'Acapito et al. (2020
)]. A candidate structure could be identified but the agreement was still not satisfactory due to the presence of intermixing. As a second step, an inter-diffused structure (obtained by randomly exchanging Ge with Sb) was created and a second spectrum was derived that better matched the experiment with a 30% intermixing. Due to strong intermixing and the loss of good order in the direction perpendicular to the surface, no dichroic analysis was carried out in that case.
In the present contribution, AIMD has been used to interpret data on a well oriented sample of WSe2 consisting of a few layers deposited on a silicon substrate. The dichroic signal has been computed and compared with experiment with good agreement, showing that AIMD can be effectively used on surface systems.
2. Computation strategy
In order to reproduce the ab initio methods, both for the static and for the dynamic part. This could be done in principle with appropriate pair potentials for selected cases but these are not widely available, and for a general approach it is more advisable to adopt the ab initio method. In the examples that follow, the simulations were carried out at the ESRF user's computing cluster (https://www.esrf.fr/Infrastructure/Computing/NICE/Implementation) using the VASP code (Kresse & Furthmüller, 1996) with the PBE-sol functional (Csonka et al., 2009
), non-relativistic calculations and unpolarized spin densities. A model of the structure under investigation is created first and a static DFT relaxation is carried out in order to deal with a cell that already possesses a minimized energy. This step can be carried out with a small cell, and a 10 × 10 × 10 k mesh in the is used. Once relaxed, a is created in order to start the AIMD cycle. The size of the is critical, as on one hand it would be desirable for it to be as large as possible to simulate the spectrum with the longest paths possible, but on the other the use of ab initio methods requires a considerable computational effort and, apart from using dedicated machine-time on supercomputing centres, cells beyond 150 atoms can barely be treatable on smaller machines. It is worth pointing out that the maximum distance that can be reproduced in the spectrum is half the size of the as beyond that value replicas of the same atoms in the structure will create unphysically intense signals. The supercells of the examples presented here have a size of 6–10 Å and 80–120 atoms. The AIMD cycle is carried out in DFT mode at the Γ point using the same functional as for the static relaxation and a Nosé–Hoover thermostat (Nosé, 1984
; Hoover, 1985
) for controlling the temperature. This parameter requires some attention as AIMD is intrinsically semi-classic and does not take into account zero-point vibrations. To treat this aspect, a useful rule-of-thumb can be derived from the discussion presented by Kuzmin (2017
) and based on the result of Yang & Kawazoe (2012
) on monoatomic lattices where it was found that the amplitude of the zero-point vibrations equals that of the excited vibrations at a temperature T0 corresponding to about one-third of the Debye temperature TD. For reproducing spectra of common crystals at room temperature (RT) or above with TD of the order of 300–500 K (namely, Cu, Ge, Zn) the actual temperature can be used. For the same systems when reproducing spectra collected at liquid-nitrogen temperature (LNT) it is more appropriate to use a somewhat higher value in the calculation, ideally 1/3TD. However, when a close agreement between simulation and experiment is needed at low temperature the path-integral Monte-Carlo/MD approach is recommended (Kuzmin et al., 2016
; Beccara & Fornasini, 2008
).
In this contribution theoretical spectra were generated using the Feff8 code (Ankudinov & Rehr, 2000) with self-consistent calculation of the electron density (self-consistent field cluster of about 40 atoms corresponding to a radius of 4–5 Å depending on the cases) and using the Hedin–Lunqvist exchange correlation potential for the inelastic effects. The full calculation of the electron density is carried out on the initial structure, stored and used for all the other frames. Considering that the atomic oscillations around the equilibrium positions are small with respect to the interatomic distances at the temperature value considered, this density function can be safely transferred to all the structures in the sequence. For building up the signal, multiple-scattering events of up to four legs for the photoelectron were considered and the maximum path length was set at half of the in order to avoid strong signals generated by atomic replicas in the periodic cell. Global amplitude S0 2 and edge energy position E0 were left at the values determined by the calculation.
A short description of the codes developed for extracting the instantaneous structures from the MD cycles and calculating the ).
3. Examples
3.1. A benchmark system: metallic Cu
In this section an example of ab initio on a well known system will be given, in order to assess the reliability of the method. The aim of this step is to verify that the overall shape and features of the spectrum can be reproduced with the tentative structure (face-centred cubic Cu in this case) and with the methods chosen (the PBE-sol functional). The scope here is to have available a method that permits the data to be related to one structure among several, as shown by d'Acapito et al. (2020). Obtaining a close agreement between experiment and theory making use of advanced and resource-consuming methods is a successive step in this analysis that is not considered at this stage. All experimental spectra presented hereafter have been collected at the LISA beamline at the European Synchrotron Radiation Facility (d'Acapito et al., 2019
). For the simulation a cubic of 108 atoms (10.7 Å in size) derived from a static relaxed cell was used for the AIMD cycle at 150 K. Time-steps of 3 fs were used and frames were saved every three time-steps. A first cycle of 1.8 ps was used for thermalization and a successive cycle of 1.8 ps was used for the simulation obtained with the last 0.9 ps (100 spectra). This time duration is limited by the computation resources available. However, it corresponds to several typical atomic oscillation periods, whose value is around 10−13 s, so provides a fair estimation of the signal damping, although remaining a limiting factor of the correct reproduction of the details of the signal amplitude. The convergence of the simulation was checked by looking at the residual ξ, defined as the sum of the difference of the squared spectra ΘN(k) obtained for N and N − 1 frames,
(d'Acapito et al., 2016) and was found to be a few 10−3 units (see supporting information). Figs. 1
and 2
show the results.
![]() | Figure 1 Comparison of the experimental (circles) and simulated (line) EXAFS spectra of metallic Cu at LNT. The lattice parameter of the theoretical cell was expanded by 1% before calculating the EXAFS spectra in order to better show the agreement with the experimental data. It is worth noting that 1% is a typical absolute inaccuracy of DFT calculated structures. |
![]() | Figure 2 Comparison of the experimental (circles) and simulated (line) Fourier transforms (FTs) of the EXAFS spectra of metallic Cu at LNT. |
The features of the spectrum in k space, namely the amplitude and position of the crests and valleys of the oscillations (bearing information about amplitude and phase of the signals) as well as the peaks in the Fourier Transform are correctly reproduced, meaning that the method can simulate both the structure and the associated dynamics fairly well though with a less accurate agreement on the signal damping at high k values that nonetheless does not prevent the recognition of the phase. A more accurate modelling of the dynamics would lead to a better agreement (see Shapeev et al., 2022) but would require larger cells and a more sophisticated description of the interatomic potentials that is beyond the scope of this contribution.
3.2. A real-life system: two-dimensional WSe2
The case shown above deals with a polycrystal, thus providing a spherically averaged spectrum. However, 2 deposited on Si for opto-electronic applications is presented. WSe2 represents an interesting family of two-dimensional materials and has been proposed as a component of photosensitive hetero-junctions (Choi et al., 2020) stacked with MoSe2 or MoxW1–xSe2 layers. The samples were grown by using a chemical vapour deposition system. The Si substrates were first cleaned by the RCA cleaning standard method and then placed in a two-inch vertical cold-wall chamber, where WSe2 growth was conducted. The W and Se precursors, in the gaseous phase, were fed into the chamber as tungsten hexacarbonyl [THC; W(CO)6] and diethyl sulfide [DES; (C2H5)2Se], respectively. The parameters were set for THC at 0°C and for DES at −15°C, for the vapour phase growth of WSe2. Additionally, Ar and H2 were pumped into the chamber to supply the precursors for Se and W, respectively, and their reaction. A total pressure of 50 Torr, a growth temperature of 600°C and a growth time of 140 min were found to be the ideal experimental parameters. In the case of THC, only 3 s.c.c.m. [where s.c.c.m. is standard cubic centimetres per minute] gas flow was used; however, for the Ar gas it was 10 s.c.c.m. (Rehman et al., 2023
). The layers have been characterized by as shown in the supporting information.
Layered systems of the class (transition-metal)–(chalcogen)2 can be effectively studied by comparison of the MD-simulated spectra and In a recent paper, Pudza et al. (2023) studied the MoS2 system in powdered samples. The comparison between experimental and theoretical spectra obtained by MD and reverse Monte Carlo methods permitted deep details on the intra- and inter-layer structures to be retrieved as well as the mean-square displacements Mo–O and Mo–S as a function of temperature.
This topic can be extended by noting that the hexagonal structure (typical of these systems) leads to a strongly (linear) dichroic ab initio methods. Measurements were carried out at RT at the Se K-edge with the beam polarization vector parallel or perpendicular to the sample surface, i.e. with respect to the hexagonal plane of the sample. data were collected using an Si(111) monochromator and Si-coated mirrors for harmonic rejection. The sample was placed at grazing incidence with respect to the incoming beam at an angle of ∼0.1° in both cases and data were collected in fluorescence mode using a four-element silicon drift detector (Hafizh et al., 2019).
The see supporting information) derived from the structure presented by Schutte et al. (1987). This model mimics a free-standing layer, i.e. a layer with a negligible interaction with the substrate. A first cycle for thermalization (6 ps) was followed by a successive 3 ps cycle for calculating the 300 spectra were averaged in this case for a total time of 900 fs and they were calculated for a linear polarization either parallel or perpendicular to the surface. In the first case spectra were calculated with the polarization vector parallel to the c-axis; for the latter case a circular average on the a–b plane was carried out using the related tools available in the Feff8 code. The results are shown in Figs. 3
and 4
.
![]() | Figure 3 Comparison of the experimental (circles) and simulated (line) EXAFS spectra at the Se K-edge of WSe2 at RT with the polarization parallel (bottom) and perpendicular (top) to the surface. |
![]() | Figure 4 Comparison of the experimental (circles) and simulated (line) FTs of the EXAFS spectra at the Se K-edge of WSe2 at RT with the polarization parallel (bottom) and perpendicular (top) to the surface. |
The residuals have been calculated using equation (1) and are shown in Fig. 5
. The value of the residuals saturates at an oscillating behaviour at about 300 spectra before the maximum number of elements on average demonstrating the conversion of the procedure. A good agreement between theory and experiment is visible in Fig. 3
, showing that by using AIMD it is possible to reproduce not only the spectral features but also the differences in the parallel (PAR) and perpendicular (PER) polarizations. It is worth noting as an example the difference in the PAR and PER spectra in k space at around 5 Å−1 where a single peak present in the PER spectrum that splits in two in the PAR spectrum is correctly reproduced by the simulation. The same can be stated for the series of oscillations at around 9 Å−1 which exhibit a different amplitude in the PAR and PER spectra that is again well reproduced by the simulation. The convergence of the procedure can be verified by considering the evolution of the residual ξ as a function of the evolution time that reached a minimum in the interval considered here (Fig. 5
). This demonstrates that AIMD can be used not only for reproducing spherically averaged spectra but also for analysing the features derived from the linear dichroic response of the sample. In order to further test the reliability of the simulations a quantitative analysis of the experimental and AIMD-derived spectra in both PAR and PER geometries has also been carried out using theoretical signals from the WSe2 structure generated using the Feff84 code as described previously. Data were reproduced for the first two intra-layer coordination shells, Se–W [Rcryst = 2.526 Å (Schutte et al., 1987
)] and Se–Se. In the PAR and PER cases the latter corresponded to the Se–Se in-plane coordination (Rcryst = 3.282 Å) or to the Se–Se out-of-plane coordination (Rcryst = 3.346 Å). The results are shown in Table 1
. The bond distances in the two datasets compare well and the crystallographic data (Schutte et al., 1987
) and the Debye–Waller factors of the Se—W bond are in good agreement with the determination of Saisopa et al. (2023
) (∼0.0030 Å−2) carried out at the W L3-edge.
|
![]() | Figure 5 Residual ξ as defined in equation (1) ![]() |
4. Discussion
The present study on layered WSe2 shows that AIMD permits the experimental data (containing both the structure and the dynamics at the atomic scale) of the system to be reproduced correctly. The aim here was to correlate a structural model of the experiment and in this case it was possible to confirm the presence of a 2D hexagonal layer. The correct reproduction of the dichroic signal demonstrated that the layer was well oriented on the surface and that there were no detectable traces of inter-diffusion with the substrate or additional phases. A less accurate agreement experiment/simulation is visible in the PER spectrum which can be tentatively explained by the fact that the simulation was a single layer whereas the sample actually consisted of a multilayer that could affect the dynamics of the structure and/or add signals from inter-layer correlations.
This study demonstrates that, by using simple and well established models (in this case the MD–DFT method with the PBE-sol functional), it is possible to simulate with fair accuracy the et al. (2014) and d'Acapito et al. (2020
)] and a choice (and possible modifications) is to be made on the bases of the experimental data. This procedure would be extremely complex in a conventional data analysis as several fitting parameters would enter into play making their interpretation difficult. In this formulation the method is oriented to a structural recognition task: once a structure is identified, obtaining its fine details will require more refined simulation methods [see, namely, Pudza et al. (2023
)].
This method could be further applied in materials for optoelectronics in studies of stacked layers proposed for the realization of devices (Choi et al., 2020) in order to reveal proper stacking or possible growth defects. Neat interfaces and obtaining the correct structure are aspects of paramount importance for acquiring the desired electro-optic properties.
5. Conclusion
In this study, AIMD has been used to study the structure of a WSe2 multilayer deposited on silicon. The Se K-edge spectra corresponding to two different orientations of the X-ray beam polarization vector (parallel and perpendicular to the sample surface) relating to a free-standing monolayer have been simulated. The simulated spectra show good agreement with the experimental ones, even in the correct reproduction of the signal linear dichroism demonstrating that the layer is well oriented with the c-axis perpendicular to the surface and that no interdiffusion occurs with the substrate. This shows how AIMD can be used for validating a structural hypothesis on a material and that, in the case of non-cubic surface systems, the linear dichroism can be effectively exploited.
Supporting information
Supplementary info. DOI: https://doi.org/10.1107/S1600577524005484/up5001sup1.pdf
Acknowledgements
The beamline LISA is a project funded by the Consiglio Nazionale delle Ricerche (project DFM.AD006.072).
References
Acapito, F. d', Kowalczyk, P., Raty, J.-Y., Sabbione, C., Hippert, F. & Noé, P. (2020). J. Phys. D Appl. Phys. 53, 404002. Google Scholar
Acapito, F. d', Lepore, G. O., Puri, A., Laloni, A., La Manna, F., Dettona, E., De Luisa, A. & Martin, A. (2019). J. Synchrotron Rad. 26, 551–558. Web of Science CrossRef IUCr Journals Google Scholar
Acapito, F. d', Pelli-Cresi, S., Blanc, W., Benabdesselam, M., Mady, F., Gredin, P. & Mortier, M. (2016). J. Phys. Condens. Matter, 28, 485301. Web of Science PubMed Google Scholar
Ankudinov, A. L. & Rehr, J. J. (2000). Phys. Rev. B, 62, 2437–2445. Web of Science CrossRef CAS Google Scholar
Anspoks, A., Kalinko, A., Kalendarev, R. & Kuzmin, A. (2012). Phys. Rev. B, 86, 174114. Web of Science CrossRef Google Scholar
Beccara, S. & Fornasini, P. (2008). Phys. Rev. B, 77, 172304. Web of Science CrossRef Google Scholar
Bocharov, D., Anspoks, A., Timoshenko, J., Kalinko, A., Krack, M. & Kuzmin, A. (2020). Radiat. Phys. Chem. 175, 108100. Web of Science CrossRef Google Scholar
Bocharov, D., Krack, M., Kalinko, A., Purans, J., Rocca, F., Ali, S. E. & Kuzmin, A. (2016). J. Phys. Conf. Ser. 712, 012009. CrossRef Google Scholar
Cartechini, L., Rosi, F., Miliani, C., D'Acapito, F., Brunetti, B. G. & Sgamellotti, A. (2011). J. Anal. At. Spectrom. 26, 2500–2507. Web of Science CrossRef CAS Google Scholar
Casarin, B., Caretta, A., Momand, J., Kooi, B. J., Verheijen, M. A., Bragaglia, V., Calarco, R., Chukalina, M., Yu, X., Robertson, J., Lange, F. R. L., Wuttig, M., Redaelli, A., Varesi, E., Parmigiani, F. & Malvestuto, M. (2016). Sci. Rep. 6, 22353. Web of Science CrossRef PubMed Google Scholar
Choi, W., Akhtar, I., Kang, D., Lee, Y., Jung, J., Kim, Y. H., Lee, C. H., Hwang, D. J. & Seo, Y. (2020). Nano Lett. 20, 1934–1943. Web of Science CrossRef CAS PubMed Google Scholar
Cresi, J. S. P. (2015). Studio della struttura locale di droganti in quantum dots mediante spettroscopia di assorbimento di raggi X. Masters thesis, University of Bologna, Bologna, Italy (in Italian) (https://core.ac.uk/download/pdf/31158383.pdf. Google Scholar
Csonka, G. I., Perdew, J. P., Ruzsinszky, A., Philipsen, P. H. T., Lebègue, S., Paier, J., Vydrov, O. A. & Ángyán, J. G. (2009). Phys. Rev. B, 79, 155107. Web of Science CrossRef Google Scholar
D'Angelo, P., Di Nola, A., Filipponi, A., Pavel, N. V. & Roccatano, D. (1994). J. Chem. Phys. 100, 985–994. CAS Google Scholar
Hafizh, I., Bellotti, G., Carminati, M., Utica, G., Gugiatti, M., Balerna, A., Tullio, V., Lepore, G., Borghi, G., Ficorella, F., Picciotto, A., Zorzi, N., Capsoni, A., Coelli, S., Bombelli, L. & Fiorini, C. (2019). J. Instrum. 14, P06027. Google Scholar
Hoover, W. G. (1985). Phys. Rev. A, 31, 1695–1697. CrossRef CAS PubMed Web of Science Google Scholar
Kopula Kesavan, J., d'Acapito, F., Scardi, P., Stavrinadis, A., Akgul, M. Z., Burgués-Ceballos, I., Konstantatos, G. & Boscherini, F. (2020). Nanomaterials, 10, 316. Web of Science CrossRef PubMed Google Scholar
Kresse, G. & Furthmüller, J. (1996). Phys. Rev. B, 54, 11169–11186. CrossRef CAS Web of Science Google Scholar
Kuzmin, A. (2017). The use of Molecular Dynamics simulations for the interpretation of EXAFS spectra, http://www.dragon.lv/edaca/BNL_Course_2017_L1_A_Kuzmin.pdf. Google Scholar
Kuzmin, A., Anspoks, A., Kalinko, A. & Timoshenko, J. (2016). Z. Phys. Chem. 230, 537–549. Web of Science CrossRef CAS Google Scholar
Kuzmin, A. & Evarestov, R. A. (2009). J. Phys. Condens. Matter, 21, 055401. Web of Science CrossRef PubMed Google Scholar
Kuzmin, A., Obst, S. & Purans, J. (1997). J. Phys. Condens. Matter, 9, 10065–10078. CrossRef CAS Web of Science Google Scholar
Nosé, S. (1984). J. Chem. Phys. 81, 511–519. CrossRef CAS Web of Science Google Scholar
Palmer, B. J., Pfund, D. M. & Fulton, J. L. (1996). J. Phys. Chem. 100, 13393–13398. CrossRef CAS Web of Science Google Scholar
Pudza, I., Bocharov, D., Anspoks, A., Krack, M., Kalinko, A., Welter, E. & Kuzmin, A. (2023). Mater. Today Commun. 35, 106359. Web of Science CrossRef Google Scholar
Rehman, M. A., Kim, M., Pawar, S. A., Park, S., Nasir, N., Kim, D., -e Khan, M. F., Nguyen, V. H., Chabungbam, A. S., Seo, Y., Sakurai, T., Chun, S.-H., Koo, D. H., Lee, C.-H., Jun, S. C. & Park, H.-H. (2023). Int. J. Energy Res. 2023, 8195624. Web of Science CrossRef Google Scholar
Saisopa, T., Bunpheng, A., Wechprasit, T., Kidkhunthod, P., Songsiriritthigul, P., Jiamprasertboon, A., Bootchanont, A., Sailuam, W., Rattanachai, Y., Nualchimplee, C., Hirunpinyopas, W. & Iamprasertkun, P. (2023). Radiat. Phys. Chem. 206, 110788. Web of Science CrossRef Google Scholar
Schutte, W., De Boer, J. & Jellinek, F. (1987). J. Solid State Chem. 70, 207–209. CrossRef ICSD CAS Web of Science Google Scholar
Shapeev, A. V., Bocharov, D. & Kuzmin, A. (2022). Comput. Mater. Sci. 210, 111028. Web of Science CrossRef Google Scholar
Tominaga, J., Kolobov, A. V., Fons, P., Nakano, T. & Murakami, S. (2014). Adv. Mater. Interfaces, 1, 1300027. Web of Science CrossRef Google Scholar
Veronesi, G., Degli Esposti Boschi, C., Ferrari, L., Venturoli, G., Boscherini, F., Vila, F. D. & Rehr, J. J. (2010). Phys. Rev. B, 82, 020101. Web of Science CrossRef Google Scholar
Vila, F., Rehr, J. J., Kas, J., Nuzzo, R. G. & Frenkel, A. I. (2008). Phys. Rev. B, 78, 121404. Web of Science CrossRef Google Scholar
Vila, F. D., Lindahl, V. E. & Rehr, J. J. (2012). Phys. Rev. B, 85, 024303. Web of Science CrossRef Google Scholar
Vila, F. D., Rehr, J. J., Rossner, H. H. & Krappe, H. J. (2007). Phys. Rev. B, 76, 014301. Web of Science CrossRef Google Scholar
Vila, F. D., Spencer, J. W., Kas, J. J., Rehr, J. J. & Bridges, F. (2018). Front. Chem. 6, 356. Web of Science CrossRef PubMed Google Scholar
Yang, Y. & Kawazoe, Y. (2012). EPL (Europhysics Lett), 98, 66007. Google Scholar
This is an open-access article distributed under the terms of the Creative Commons Attribution (CC-BY) Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original authors and source are cited.
![Open Access](/logos/open.png)