Ultramicroscopy 111 (2011) 847–853 Contents lists available at ScienceDirect Ultramicroscopy journal homepage: www.elsevier.com/locate/ultramic Experimental and theoretical study of electron density and structure factors in CoSb3 R. Sæterli a, E. Flage-Larsen b, J. Friis c, O.M. Løvvik b, J. Pacaud d, K. Marthinsen e, R. Holmestad a,n a Department of Physics, Norwegian University of Science and Technology (NTNU), 7491 Trondheim, Norway Department of Physics, University of Oslo, 0316 Oslo, Norway c Department of Synthesis and Properties, SINTEF Materials and Chemistry, 7491 Trondheim, Norway d Insitut Pprime, UPR 3346 CNRS, Université de Poitiers, SP2MI, Bd P et M Curie, F-86962 Chasseneuil cedex, France e Department of Materials Science and Engineering, Norwegian University of Science and Technology (NTNU), 7491 Trondheim, Norway b a r t i c l e in fo abstract Available online 27 August 2010 We refine two low-order structure factors of the skutterudite CoSb3 using convergent beam electron diffraction. The relatively large unit cell of this material causes the disks to overlap and introduces a series of challenges in the refinement procedure. These challenges and future work-arounds are discussed. The refined structure factors F200 and F600 are compared to X-ray diffraction and density functional calculated values, the latter calculated using two different functionals. Both relaxed and experimental lattice parameters are tested to explicitly highlight the impact of the lattice geometry and atomic position on the structure factors. & 2010 Elsevier B.V. All rights reserved. On the occasion of John Spence’s 65 years anniversary. Keywords: CBED DFT Skutterudite Electronic structure 1. Introduction The importance of accurate experimental structure factor measurements for determination of the electron density of materials is well known and a lot of effort has been put into this task. The advent of powerful calculation schemes such as density functional theory (DFT) to calculate the electron density and thus X-ray structure factors from first principles have contributed significantly to the field [1–3]. It is well known that the accuracy of DFT calculations relies on the choice of functionals, potentials, basis sets and technical implementation. Ultimately, it is also limited by computational resources. However, if the electron density is accurately determined, a wealth of electronic, optical and vibrational properties can be calculated, which is difficult, time consuming and in some cases impossible to obtain from experiments. In addition, electron transfer analysis can be done to explicitly investigate the electron rearrangement due to the setup of bonds [4]. Such electron rearrangement determines most of the properties of materials and it is thus very important to not only understand the difference between ionic and covalent bonded structures, but also the strength, local geometry and mixing of such bonds. Due to the array of approximations introduced in these schemes, experimental verification of different functionals, basis sets and implementation issues is, however, still necessary (see e.g. [5,6]). n Corresponding author. E-mail address: randi.holmestad@ntnu.no (R. Holmestad). 0304-3991/$ - see front matter & 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.ultramic.2010.08.002 Employed experimental techniques for accurate structure factor determination comprise both X-ray methods [7,8] and different methods based on transmission electron diffraction (see e.g. [9]). The strength of X-ray synchrotron diffraction lies in its capability to accurately measure a large number of structure factors [10]. Generally, this method is very accurate for high-order structure factors, while for the strong low order reflections it is severely affected by extinction due to multiple scattering. To overcome this problem, it has been proven very successful to combine X-ray diffraction with convergent beam electron diffraction (CBED) studies of the strong low order reflections (see e.g. [11–15]). The introduction and application of CBED for the measurement of crystal structure amplitudes and phases are very much due to the enthusiasm and pioneering work of Spence and Zuo [16] and Spence [17] in this field in the late eighties and early nineties. As a result of this and the accelerating development in hardware (energy filters and detection systems), a lot of work was done in the field in the nineties, with increasing complexity and precision (e.g. [18–21]). One of the most versatile methods is the use of a systematic row of CBED reflections, with high sensitivity both to the structure factor in question and the critical sample thickness parameter, and also well developed refinement tools [18,22,23]. The systematic row approach has been demonstrated for a range of materials such as GaAs [24], TiAl [25], MgO [26], Cu2O [11], Cu [13] and SrTiO3 [14]. During the last years, also two specialised methodologies based on the zone axis approach [27–30] have been further developed, and represent today’s ‘state-of-the-art’ in structure factor determination by electron diffraction. The method developed by Tsuda includes refinement 848 R. Sæterli et al. / Ultramicroscopy 111 (2011) 847–853 of atom positions as well as structure factors, and has recently been used to study orbital ordering in the spinel oxide FeCrO2 [30]. In all methodologies, complex structures with large unit cells are avoided due to the increased overlap of reflections. To the authors’ knowledge, the systematic row approach has not been demonstrated on unit cells larger than about 4–5 Å. In this work, we examine the possibility of refining low-order structure factors from the skutterudite CoSb3, having unit cell parameters of 9 Å, using the systematic row approach. Some skutterudites are promising thermoelectric materials, as their body-centered cubic unit cells can be filled by so-called ‘‘rattler’’ atoms introduced into the structure to reduce thermal conductivity while at the same time not altering the favorable electronic structure of the unfilled structures [31]. As some of the highest performing thermoelectric skutterudites are found among the filled CoSb3 variants [32], the electronic structure of CoSb3 has been thoroughly studied both experimentally and theoretically in the literature [10,33–38] and constitutes a basis on which to test whether the CBED technique is applicable also to such large unit cell materials. We refine two bond sensitive low-order structure factors from CoSb3, and show that the systematic row approach is close to its useful limit in this material. Comparison to structure factors retrieved from X-ray diffraction and DFT is made. To shed light on the sensitivity of the structure factors, the changes in the lattice geometry and atom positions were studied by relaxing these parameters from the experimental values. Structure factors from both the local density and generalized gradient approximation were also calculated. higher order Laue zone (HOLZ) lines. Simulated intensities were found through the EXTAL [23] program, which is based on the Bloch wave method. A goodness of fit parameter GoF ¼ X ðIexp cItheory Ibackgr Þ2 1 i i i nm1 i s2i ð1Þ was minimized by refining chosen parameters. The sum runs over all pixels i ¼1, ... ,n, m is the number of refined parameters, Iiexp , Iitheory and Iibackgr are experimental, theoretical and background intensity values, respectively, with c being a scaling factor and si the standard deviation of Iiexp . The refined parameters include the amplitude of the desired structure factor(s) Uhkl and corresponding absorption component(s) U0 hkl, c and Iibackgr (assumed constant for each disk), beam direction (or conversely, sample orientation) and sample thickness. Lattice parameters, atomic positions and anisotropic atomic displacement parameters (ADPs) were not refined, but taken from X-ray diffraction data, recorded at a temperature of 100 K [41]. Lattice parameters and atom positions used in the refinement are given in Table 1. Due to the large unit cell, a large number of beams had to be included in the calculations, chosen from the following three criteria (for an explanation of the terms, see Zuo and Weickenmeier [42]): (1) proximity to the Ewald sphere 92KSg9max ¼3.0 Å, (2) length of reciprocal scattering vector 9gmax9¼ 3.0 Å and (3) perturbation strength 9Ug/2KSg9min ¼25. Strong beams were treated exactly, while weak beams were treated by Bethe perturbation [43,44]. 2. Experimental 2.4. Density functional theory calculations 2.1. Material synthesis The sample was prepared by melting the elements (Co: SigmaAldrich, 99.995%; Sb: J.T. Baker, 99.8%) in evacuated, sealed silica glass ampoules, thereafter annealed at 800 1C for four days, crushed, remelted and further heat treated at 800 1C for 17 days. The sample was single phase according to powder X-ray diffraction. 2.2. CBED experiments TEM specimens were prepared by dimpling and subsequent Ar ion milling of the CoSb3 polycrystal at liquid nitrogen temperature. Defect-free areas, as judged from bright field images and HOLZ lines, were chosen for analysis. CBED experiments were performed on a JEOL 2200FS equipped with a 2K 2K CCD camera and the patterns were filtered using an Omega energy filter tuned on to the zero-loss peak with an energy width of 10 eV to ensure contributions only from elastically and, unavoidably, thermally scattered electrons. Further, the sample was cooled to approximately 120 K and allowed to stabilize in order to minimize thermal diffuse scattering and also to avoid contamination. The patterns were further deconvoluted using the LUCY algorithm [39] to correct for the point-spreading of the CCD. This is normally sufficient for structure factor refinement, although it is acknowledged that the use of direct methods [40] might be better. The voltage of the microscope was calibrated in advance to 203.5 kV. The electronic structure of CoSb3 was calculated using the Vienna Ab-initio Simulation Package (VASP) [45,46]. Two functionals, the local-density approximation LDA-PZ [47] and generalized gradient approximation GGA-PBE [48] were employed together with the projector augmented wave (PAW) method [49]. Experimental lattice parameters were used as input to a quasi-Newton residual minimization scheme with direct inversion in the iterative subspace (RMM-DIIS) relaxation. An energy cutoff of 550 eV and Monckhorst– Pack k-point sampling of 8 8 8 were sufficient to obtain converged results to within a few meV of the total energy. The allelectron charge density was regenerated on a dense 480 480 480 real-space grid to improve the resolution of localized charges and was passed to a fast Fourier transform routine in the FFTW package [50] to obtain the structure factors. To improve the resolution of the all-electron density and thus the structure factors, a dense real-space grid is necessary. However, such grids have an upper bound limited by computational resources, in particular memory usage. In this work, we opted for a dense real-space grid such that relative errors in the 000 structure factor from the chemical electron count of the valence electrons were within 10 4. Relaxed lattice parameter and atom positions from VASP are given in Table 1 together with the experimental parameters of Krebs Larsen [41]. Table 1 Experimental (at 100 K) and relaxed DFT lattice parameter a from [41] and atomic positions. Experimental values 2.3. Refinement of CBED data Refinements of three low-order non-extinct structure factors, U110, U200 and U600, were attempted from systematic row orientations chosen on the basis of relatively few intersecting a (Å) Sb y Sb z a Ref. [41]. a Relaxed DFT values 9.02 9.11 0.3354 0.1579 0.3334 0.1595 R. Sæterli et al. / Ultramicroscopy 111 (2011) 847–853 849 2.5. Comparison of CBED and DFT data The electron structure factors Uhkl are not directly comparable to the structure factors Fhkl obtained from DFT and X-ray diffraction. The two quantities are however related by the Mott formula [16], taking into account the ADPs for the given temperature and the core contribution to the electron structure factors: Fhkl ¼ X 16p2 eh2 Os2 Zj Tj ðsÞe2pigrj Uhkl m0 gjej2 j ð2Þ here, Tj(s) is the temperature factor, s ¼ sin y=l is the scattering angle and other notations follow that of Spence and Zuo [16]. 0 X-ray structure factors at 0 K are referred to as Fhkl and are 0 120 K 0 100 K calculated as Fhkl ¼ Fhkl ðFSC =FSC Þ, assuming that the ADPs are sufficiently similar for 120 K and 100 K. One can define the bonding charge density, rB, as the charge density of the real crystal, rC, minus the charge density of a reference crystal, rR. This reference should be free of bonding features and is often referred to as an independent atom model (IAM). The IAM model can be found from Hartree–Fock calculations as in the parameterized version of Su and Coppens [51] (SC), or calculated by for example DFT by considering the procrystal (PC), a superposition of free atomic electron densities embedded in the crystal unit cell. The quantities FB, Fhkl and FR can then be defined as the Fourier transforms Z ri ðrÞe2pigr dr ð3Þ Fi ¼ Fig. 1. (a) Experimental CBED image of systematic 110 row of reflections. (b)–(g) Bloch wave simulated images from the same conditions as in (a) using approximately 25, 50, 125, 300, 500 and 850 beams, respectively. of rB, rC and rR, respectively. The direct comparison of experimental and calculated Fhkl values can be challenging due to the limitation of the plane-wave methodology to represent core states and/or all-electron densities accurately. However, as core contributions to the electron density do not usually contribute in the self-consistent procedure, we compare the experimental and theoretical values of FB rather than Fhkl. 3. Results Fig. 1(a) shows one of the experimental CBED patterns of the systematic 110 row of disks. The disk overlap is substantial due to the relatively large lattice parameter, although a small beam convergence is used. Any further reduction in beam convergence causes the disk area to be unsatisfactorily small, and would in addition extend the coherence fringes now seen at the edges of the disks further towards the center. Simulated images from the same thickness and orientation are shown in Fig. 1(b)–(g), using an increasing number of beams from about 25 to 850 in the calculations. The refined sample thickness was about 95 nm. As can be seen from the simulations, a high number of beams is clearly needed in order to reproduce the experimental intensity. Fig. 2(a) shows an experimental CBED image of a systematic 200 row of disks, used to refine U200 and U600. The best U200 and U600 structure factor refinement yielded values of 0.01549 and 0.03985 Å 2, respectively, with a GoF value of 1.59 for 368 intensities/pixels and 897 beams included in the refinement. The result of the refinement is shown in Fig. 2(b), portraying the intensity profiles of the linescans marked and numbered in Fig. 2(a). Table 2 summarizes the results of the U200 and U600 refinements. In addition to the structure factors Uhkl and absorption U0 hkl, the structure factors are also given as X-ray 0 structure factors Fhkl that have been corrected for the finite experimental temperature. These are compared to experimental Fig. 2. (a) experimental CBED image of systematic 200 row of reflections with the linescans used to refine U200 and U600. (b) Linescan intensity profiles of the linescans marked in (a). Red crosses mark experimental intensities while green/ light curves are best fit refined intensity values. Blue/dark curves mark the difference between theoretical and experimental intensities. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) X-ray diffraction values [41] and the SC IAM values. No reliable refinement of the 110 structure factor was obtained. To check the consistency of the refined structure factor values, the GoF was evaluated as a function of the two refined structure factors for the measurement yielding the lowest GoF. Fig. 3 plots the GoF versus ten values of U200 and U600 in an interval around 850 R. Sæterli et al. / Ultramicroscopy 111 (2011) 847–853 Table 2 Experimentally refined best fit U200 and U600 structure factors (in Å 2) with scattering angles s ¼ sin y=l and absorption U0 hkl(in Å 2), and the conversion to X-ray structure 0 factors Fhkl (in electrons per unit cell) at 0 K. Experimental X-ray structure factors [41] also with the temperature factors set to zero and the IAM values of Su and Coppens (SC) [51] are included for comparison. h k 2 6 0 0 a l 0 0 s 0.11 0.33 XRD a IAM (SC) CBED Fhkl Uhkl U0 hkl 0 Fhkl 0 Fhkl 149.50 663.22 0.01549 (4) 0.0398 (3) 6.71 10 4 3.31 10 3 153.2 (4) 670 (6) 153.8 666.2 Ref. [41]. Fig. 3. Contour plot of GoF versus the two refined structure factors U200 and U600 for an interval around the final refined values, calculated using about 900 beams. the obtained minimum values. All refinable parameters other than the structure factors are fixed. Table 3 shows FB from DFT calculations using the LDA and GGA functionals, and for both functionals using both SC and PC as reference IAM valence values. Additionally, for GGA, two sets of lattice parameters and atomic positions were used as input to the DFT calculations, one with the experimental parameters used in the refinements and one set of relaxed parameters. The results of the CBED refinements and X-ray diffraction values [41] are shown for comparison, also these with both SC and PC used as reference values. In Fig. 4, the electron density difference rB in the 130 and 200 planes from the relaxed GGA calculations is shown. The electron accumulations between Co and Sb, and between the Sb positions in the Sb4 ring caused by bonding are clearly visible. To compensate, electron depletion occurs in the areas not covered by these bonds. 4. Discussion The best U200 and U600 structure factor refinement yielded values 0.01549 and 0.03985 Å 2, respectively, with a GoF of 1.59. Several linescan positions and experimental images were tested, which all came out with higher GoF values. Using refinements which resulted in GoF values less than 2 gives intervals of 0 0.01549–0.01632 Å 2 and 152.5–153.2 for U200 and F200 , respectively. Corresponding numbers for the 600 structure factor are 0.03985–0.04123 Å 2 and 659.2–670.1. Using these intervals as 0 estimates of the standard deviation gives uncertainties in the Fhkl values are 0.23% and 0.81% for the 200 and 600 structure factors, respectively. In Table 2, we have chosen to give these intervals around the value with lowest GoF rather than the mean value (corresponding to the refinement run with the largest number of beams). One benefit of the large unit cell is that even with a relatively large error in Uhkl, the s2 factor in the second term of the Mott formula (Eq. (2)) ensures that for the low-order factors the error in Fhkl remains small. The plot in Fig. 3 shows that the minimum in GoF is broad, however with no other distinct minima than the one found. The variations in the refined values do, however, suggest that other local minima can be found in the parameter space comprising the other refinable parameters. The relative difference in the calculated FB values between the two lattice parameters and atom position sets is around 3–4% (see Table 3) for both the 200 and the 600 structure factor, most likely caused by the temperature difference for which the lattice parameters have been obtained (e.g. 100 K versus 0 K) and/or density functional related errors, stemming from the choice of functional, basis set or implementation issues associated with representing the all-electron density on a finite grid. The relative difference between the LDA and PBE functionals for both 200 and 600 structure factors is smaller, but similar in magnitude. Even though the fundamental approach of superpositioning the electron density at atomic positions is similar in the PC and SC models, the way the single atomic electronic density is addressed is different. The SC reference employs relativistic Hartree–Fock approximations, while LDA and PBE density functional calculations were used for the PC reference. Hartree–Fock calculations include exact exchange, while correlation effects are disregarded. The PBE functional mixes exchange and correlation in an approximate manner. Consequently, it is reasonable to expect that the SC and the calculated PC reference in this work yield different results. Thus, from a calculation point of view, since the same calculation methods are used, it can be argued that the PC reference is better founded, such that the difference between the electron density of the crystal and PC represent a minimum of approximation errors. Using the DFT structure factors of the crystal with the SC reference would likely introduce additional errors related to the different calculation schemes. From an experimental point of view it is difficult to argue which reference to use. This is also confirmed in Table 3, where both the SC and PC references for the experimental data yield results that are similar in magnitude to the calculations. The fact that the calculated FB is consistently lower than measured values (given the same reference) might be caused by effects not included in the calculations, for example dynamic core contributions, lack of proper description of the exchange and correlation effects, or particularly lattice dynamics related to the finite temperature during the experiments. Future studies with R. Sæterli et al. / Ultramicroscopy 111 (2011) 847–853 851 Table 3 Comparison of FB ¼ FCvalence FRvalence from LDA and GGA-PBE functionals, where FRvalence found using both the procrystal (PC) and Su-Coppens (SC) [51] approach, with the core core experimentally found FB ¼ FC FRvalence FSC values for the 200 and 600 structure factors. The core values FSC were obtained from SC as these are not obtainable through the PC approach and comprise the same core electrons, while the reference valence values FRvalence were obtained from both SC and PC approaches. Values are in electrons per unit cell. a IAM reference h k l LDA GGA-PBE GGA-PBE (relaxed) CBED XRD SC 2 6 2 6 0 0 0 0 0 0 0 0 3.313 2.395 1.596 2.283 3.288 2.379 1.639 2.308 3.228 2.305 1.582 2.240 3.70 6.88 2.05 6.81 4.30 2.98 2.65 2.53 PC a Ref. [41]. Fig. 4. (a) The unit cell of CoSb3 with the 200 (horizontal) and 130 (inclined) lattice planes. Co atoms are black, while Sb atoms are gray. (b), (c) CoSb3 electron density difference rB plots in the 130 and 200 planes, respectively, calculated using relaxed structural parameters in the GGA-PBE approach. The 200 plane is shifted horizontally by half a unit cell to encompass a whole Sb4 ring. Contour lines are drawn from 0.01 to 0.01 Bohr 3 in 0.002 increments, with black and gray lines denoting positive and negative contour lines, respectively. The electron buildup due to the Co–Sb bond in the octahedra is clearly visible between the Co and Sb position. Similarly visible is the electron buildup between the Sb positions due to the Sb–Sb bond in the Sb4 ring. different calculation schemes or a larger number of structure factors might reveal the source of the discrepancies. The comparison of experimental and theoretical structure factors is today burdened by the need of a common bond-free reference. By for example combining plane-waves and atomic orbitals, or using a continuous electron density model for the core region, better bond-free electron density references might be obtained, so that the absolute structure factors can be used as the common reference. The potential of electron density analysis is illustrated in Fig. 4, where the bonds can be seen qualitatively through electron accumulation and depletion. It is also possible to quantify the magnitude of these accumulation and depletion areas by studying the inter-atomic line scans of the electron density difference. Such analyses have previously been used to study the real-space geometry and quantitative determination of the bond strength in skutterudites and Zintl compounds [4,52]. The task of generating the electron density with reasonable resolution from experiments is demanding and sometimes impossible. It is thus more foreseeable to calculate the electron density and structure factors and then compare the calculated and experimental structure factors, which are most sensitive to the bond rearrangements. Choosing these structure factors might be difficult, but a general idea can be obtained from the geometry of the lattice and atomic positions. The systematic row of 110 reflections, with lattice spacing of 6.4 Å, in Fig. 1(b) illustrates nicely the disk overlap challenges and the need for a high number of beams in such large unit cell systems. We note that the difference between the two images simulated using 500 beams and 850 beams is as much as 4% in certain areas. It is possible that an even higher number of beams has to be included to obtain converged structure factor values. Also, the lack of results can be attributed to the very small area available for refinement due to the disk overlap. For the CBED refinements, the large number of excited beams leads to large calculations that need to consider several hundred beams, so that the result depends on the exact knowledge of several structure factors. Also the refinement of parameters other than the structure factors is severely affected by the large unit cell, as the small area encompassed in each disk complicates the refinement of the exact direction of the incoming beam. In addition, the line scan intensities are likely to be influenced by overlapping weak background disks. The issue with the background is addressed by Nakashima and Muddle [30], who propose that their differential approach partly solves this problem. An interesting future prospect would be to try this approach on large unit cell systems to see whether any improvement can be obtained. Experimentally, the closeness of the reflections calls for a small beam convergence in order to avoid overlap of the disks, so that the reciprocal area spanned by each disk is limited. The number of HOLZ lines is large due to the large number of excited beams, and areas without two or more intersecting HOLZ lines are rare. The small convergence angle of the beam also introduces the 852 R. Sæterli et al. / Ultramicroscopy 111 (2011) 847–853 coherence fringes seen around the perimeter of each disk, limiting the useful area of quantitative analysis even further. The fringes could also be due to some slight instrumental misalignment of apertures or lenses, and are not present at the convergence angles regularly used in CBED measurements. The existence of these fringes has been verified to be present in patterns from JEOL 2010F and JEOL 2200FS instruments. The large angle CBED (LACBED) technique [53] has been applied to materials characterization in order to overcome the overlap of reflections, however the introduction of a shadow image of the sample overlaid with the reciprocal space information makes it less useful to the determination of structure factors, where precise measurement of intensities is vital. With everincreasing computational power, a more likely path to accurate low-order structure factors is the consideration of interference effects and the overlapping disk areas, however the theoretical refinement tools to achieve this still awaits invention. The invention of such tools could be the next challenge in the area of quantitative electron diffraction for great minds such as John Spence or those following in his steps. 5. Summary We have refined the 200 and 600 structure factors of CoSb3 using convergent beam electron diffraction (CBED). These results were compared to structure factors obtained from plane-wave based density functional calculations and also X-ray diffraction. The low-order structure factors are important both for verification of various theoretical approaches, and also to complement X-ray diffraction results that suffer from the effects of extinction for these reflections. Technical aspects regarding the calculations of first principle structure factors were discussed. Both experimental and relaxed lattice parameters were included in the calculations to shed light on the sensitivity of the structure factors to geometry changes in the lattice. Two functionals were also compared and reasonable agreement with experiments was shown for all calculations. Reasons why other CBED low-order structure factor refinements are hard to converge in such large unit cell systems are discussed, together with possible future improvements. Acknowledgements The authors would like to thank Ole-Bjørn Karlsen for synthesizing the material and Bjørn G. Soleim for preparation of the TEM sample. Further, we are indebted to Finn Krebs Larsen who has been kind enough to supply the two relevant X-ray structure factors as well as temperature factors, lattice parameters and atom positions. In addition, we would like to thank Krebs Larsen, Øystein Prytz and Johan Taftø for valuable discussions. The NOTUR consortium should be acknowledged for computational resources and The Research Council of Norway for financial support. References [1] Z.W. Lu, A. Zunger, M. Deutsch, Electronic charge distribution in crystalline diamond, silicon, and germanium, Phys. Rev. B 47 (1993) 9385–9410. [2] J.R. Trail, D.M. Bird, Accurate structure factors from pseudopotential methods, Phys. Rev. B 60 (1999) 7875–7880. [3] J. Pere, M. Gelize-Duvignau, A. Lichanot, Comparison of Hartree-Fock and density functional theory structure factors and charge density in diamond, silicon and germanium, J. Phys. Condens. Matter 11 (1999) 5827–5843. [4] E. Flage-Larsen, O.M. Løvvik, Ø. Prytz, J. Taftø, Bond analysis of phosphorus skutterudites: elongated lanthanum electron buildup in LaFe4P12, Comput. Mater. Sci. 47 (2010) 752–757. [5] J.M. Zuo, P. Blaha, K. Schwarz, The theoretical charge density of silicon: experimental testing of exchange and correlation potentials, J. Phys. Condens. Matter 9 (1997) 7541–7561. [6] J. Friis, G.K.H. Madsen, F.K. Larsen, B. Jiang, K. Marthinsen, R. Holmestad, Magnesium: comparison of density functional theory calculations with electron and X-ray diffraction experiments, J. Chem. Phys. 119 (2003) 11359–11366. [7] N. Kato, The determination of structure factors by means of Pendellösung fringes, Acta Cryst. A 25 (1969) 119–128. [8] M. Hart, A.D. Milne, Absolute measurement of structure factors using a new dynamical interference effect, Acta Cryst. A 26 (1970) 223–229. [9] J.M. Zuo, Measurements of electron densities in solids: a real-space view of electronic structure and bonding in inorganic crystals, Rep. Prog. Phys. 67 (2004) 2053–2103. [10] A. Ohno, S. Sasaki, E. Nishibori, S. Aoyagi, M. Sakata, B.B. Iversen, X-ray charge density study of chemical bonding in skutterudite CoSb3, Phys. Rev. B 76 (2007) 064119. [11] J.M. Zuo, M. Kim, M. O’Keeffe, J.C.H. Spence, Direct observation of d-orbital holes and Cu–Cu bonding in Cu2O, Nature 401 (1999) 49–52. [12] V.A. Streltsov, P.N.H. Nakashima, A.W.S. Johnson, Charge density analysis from complementary high energy synchrotron X-ray and electron diffraction data, J. Phys. Chem. Solids 62 (2001) 2109–2117. [13] J. Friis, B. Jiang, J.C.H. Spence, R. Holmestad, Quantitative convergent beam electron diffraction measurements of low-order structure factors in copper, Microsc. Microanal. 9 (2003) 379–389. [14] J. Friis, B. Jiang, J. Spence, K. Marthinsen, R. Holmestad, Extinction-free electron diffraction refinement of bonding in SrTiO3, Acta Cryst. A 60 (2004) 402–408. [15] L.J. Wu, Y.M. Zhu, T. Vogt, H.B. Su, J.W. Davenport, J. Taftø, Valence-electron distribution in MgB2 by accurate diffraction measurements and firstprinciples calculations, Phys. Rev. B 69 (2004) 064501. [16] J.C.H. Spence, J.M. Zuo, in: Electron Microdiffraction, Plenum, New York, 1992. [17] J.C.H. Spence, On the accurate measurement of structure-factor amplitudes and phases by electron diffraction, Acta Cryst. A 49 (1993) 231–260. [18] J.M. Zuo, Automated structure-factor refinement from convergent-beam electron diffraction patterns, Acta Cryst. A 49 (1993) 429–435. [19] C. Deininger, G. Necker, J. Mayer, Determination of structure factors, lattice strains and accelerating voltage by energy-filtered convergent beam electron diffraction, Ultramicroscopy 54 (1994) 15–30. [20] M. Saunders, D.M. Bird, N.J. Zaluzec, W.G. Burgess, A.R. Preston, C.J. Humphreys, Measurement of low-order structure factors for silicon from zone-axis CBED patterns, Ultramicroscopy 60 (1995) 311–323. [21] K. Tsuda, M. Tanaka, Refinement of crystal structural parameters using two-dimensional energy-filtered CBED patterns, Acta Cryst. A 55 (1999) 939–954. [22] J.M. Zuo, Accurate structure refinement and measurement of crystal charge distribution using convergent beam electron diffraction, Microsc. Res. Tech. 46 (1998) 220–233. [23] J.M. Zuo, Quantitative convergent beam electron diffraction, Mater. Trans. JIM 39 (1998) 938–946. [24] J.M. Zuo, J.C.H. Spence, M. O’Keeffe, Bonding in GaAs, Phys. Rev. Lett. 61 (1988) 353–356. [25] R. Holmestad, J.M. Zuo, J.C.H. Spence, R. Høier, Z. Horita, Effect of Mn doping on charge-density in g–TiAl by quantitative convergent-beam electrondiffraction, Philos. Mag. A 72 (1995) 579–601. [26] J.M. Zuo, Charge density of MgO: implications of precise new measurements for theory, Phys. Rev. Lett. 78 (1997) 4777. [27] K. Tsuda, Y. Ogata, K. Takagi, T. Hashimoto, M. Tanaka, Refinement of crystal structural parameters and charge density using convergent-beam electron diffraction—the rhombohedral phase of LaCrO3, Acta Cryst. A 58 (2002) 514–525. [28] K. Tsuda, D. Morikawa, Y. Watanabe, S. Ohtani, T. Arima, Direct observation of orbital ordering in the spinel oxide FeCr2O4 through electrostatic potential using convergent-beam electron diffraction, Phys. Rev. B 81 (2010) 180102. [29] P.N.H. Nakashima, Thickness difference: a new filtering tool for quantitative electron diffraction, Phys. Rev. Lett. 99 (2007) 125506. [30] P.N.H. Nakashima, B.C. Muddle, Differential quantitative analysis of background structure in energy-filtered convergent-beam electron diffraction patterns, J. Appl. Cryst. 43 (2010) 280–284. [31] G.S. Nolas, J. Sharp, H.J. Goldsmid, in: Thermoelectrics: Basic Principles and New Materials Developments, Springer Verlag, Berlin, 2001. [32] B.C. Sales, D. Mandrus, R.K. Williams, Filled skutterudite antimonides: a new class of thermoelectric materials, Science 272 (1996) 1325–1328. [33] I. Lefebvre-Devos, M. Lassalle, X. Wallart, J. Olivier-Fourcade, L.R. Monconduit, J.C. Jumas, Bonding in skutterudites: combined experimental and theoretical characterization of CoSb3, Phys. Rev. B 63 (2001) 125110. [34] K. Koga, K. Akai, K. Oshiro, M. Matsuura, Electronic structure and optical properties of binary skutterudite antimonides, Phys. Rev. B 71 (2005) 155119. [35] A.P. Grosvenor, R.G. Cavell, A. Mar, X-ray photoelectron spectroscopy study of the skutterudites LaFe4Sb12, CeFe4Sb12, CoSb3, and CoP3, Phys. Rev. B 74 (2006) 125102. [36] H. Anno, K. Matsubara, T. Caillat, J.P. Fleurial, Valence-band structure of the skutterudite compounds CoAs3, CoSb3, and RhSb3 studied by X-ray photoelectron spectroscopy, Phys. Rev. B 62 (2000) 10737–10743. R. Sæterli et al. / Ultramicroscopy 111 (2011) 847–853 [37] D.J. Singh, W.E. Pickett, Skutterudite antimonides: quasilinear bands and unusual transport, Phys. Rev. B 50 (1994) 11235–11238. [38] J.O. Sofo, G.D. Mahan, Electronic structure of CoSb3: a narrow-band-gap semiconductor, Phys. Rev. B 58 (1998) 15620–15623. [39] J.M. Zuo, Electron detection characteristics of a slow-scan ccd camera, imaging plates and film, and electron image restoration, Microsc. Res. Tech. 49 (2000) 245–268. [40] P.N.H. Nakashima, A.W.S. Johnson, Measuring the PSF from aperture images of arbitrary shape—an algorithm, Ultramicroscopy 94 (2003) 135–148. [41] F. Krebs Larsen et al., Private communication, ongoing work, 2010. [42] J.M. Zuo, A. Weickenmeier, On the beam selection and convergence in the Bloch-wave method, Ultramicroscopy 57 (1995) 375–383. [43] W. Nüchter, A.L. Weickenmeier, J. Mayer, Inst. Phys. Conf. Ser. 147 (1995) 129–132. [44] C. Birkeland, R. Holmestad, K. Marthinsen, R. Høier, Efficient beam-selection criteria in quantitative convergent beam electron diffraction, Ultramicroscopy 66 (1996) 89–99. [45] G. Kresse, J. Hafner, Ab initio molecular dynamics for liquid metals, Phys. Rev. B 47 (1993) 558–561. [46] G. Kresse, J. Furthmuller, Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set, Comput. Mater. Sci. 6 (1996) 15–50. 853 [47] P. Perdew, A. Zunger, Self-interaction correction to density-functional approximations for many-electron systems, Phys. Rev. B 23 (1981) 5048–5079. [48] J.P. Perdew, K. Burke, M. Ernzerhof, Generalized gradient approximation made simple, Phys. Rev. Lett. 77 (1996) 3865–3868. [49] G. Kresse, D. Joubert, From ultrasoft pseudopotentials to the projector augmented-wave method, Phys. Rev. B 59 (1999) 1758–1775. [50] M. Frigo, S.G. Johnson, The design and implementation of FFTW3, Proc. IEEE 93 (2005) 216–231. [51] Z. Su, P. Coppens, Relativistic X-ray elastic scattering factors for neutral atoms Z¼ 1–54 from multiconfiguration Dirac-Fock wavefunctions in the 0–12 Å 1 sin y/l range, and six-Gaussian analytical expressions in the 0–6 Å 1 range. Erratum, Acta Cryst. A 54 (1998) 357. [52] E.S. Toberer, A.F. May, B.C. Melot, E. Flage-Larsen, G.J. Snyder, Electronic structure and transport in thermoelectric compounds AZn2Sb2 (A ¼Sr, Ca, Yb, Eu), Dalton Trans. 39 (2010) 1046–1054. [53] H.H. Liu, X.F. Duan, Q.X. Xu, B.G. Liu, Study of strained-silicon channel metaloxide-semiconductor field effect transistors by large angle convergent-beam electron diffraction, Ultramicroscopy 108 (2008) 816–820.