Structural inhomogeneities in fast ion conducting glasses

0 downloads 0 Views 495KB Size Report
Structural inhomogeneities in fast ion conducting glasses. J. Swenson,1 C. Karlsson,1 L. Börjesson,1 and R. K. Heenan2. 1Department of Applied Physics, ...
PHYSICAL REVIEW B, VOLUME 64, 134201

Structural inhomogeneities in fast ion conducting glasses J. Swenson,1 C. Karlsson,1 L. Bo¨rjesson,1 and R. K. Heenan2 1

Department of Applied Physics, Chalmers University of Technology, S-412 96 Go¨teborg, Sweden 2 Rutherford-Appleton Laboratory, Chilton, Dicot OX11 0QX, United Kingdom 共Received 19 March 2001; revised manuscript received 18 June 2001; published 28 August 2001兲 The structures of the ion conducting glasses LiCl-Li2O-2B2O3, (AgI兲0.6-共Ag2O-B2O3兲0.4, CsI-4AgPO3, and PbI2-9AgPO3 have been investigated using small- and wide-angle neutron diffraction experiments. We have applied the reverse Monte Carlo modeling method to the experimental data with the aim to reveal the intermediate-range structure on a length scale of 5–50 Å. The (AgI兲0.6-共Ag2O-B2O3兲0.4 glass shows exceptionally high scattering intensity at low Q values 共⬍0.5 Å⫺1兲 due to the existence of chemical inhomogeneities on length scales up to at least 50 Å. Both the salt ions and the network atoms are inhomogeneously distributed. The first diffraction peak, located at 0.46 Å⫺1 in the total structure factor, is caused by a characteristic distance between B-O segments separated by salt ions. The LiCl-Li2O-2B2O3 glass shows inhomogeneities only in the distribution of salt ions, where particularly the chlorine ions are very inhomogeneously distributed. The ion concentration fluctuations occur on a wide range of length scales. In contrast to the investigated borate glasses, the two phosphate glasses show only minor structural inhomogeneities on length scales above 10 Å. DOI: 10.1103/PhysRevB.64.134201

PACS number共s兲: 61.43.Fs, 61.12.⫺q, 61.43.Bn

I. INTRODUCTION

Glasses with high ionic conductivity presently attract considerable scientific interest because of their potential applications as solid electrolytes in various electrochemical devices such as solid-state batteries, fuel cells, memory devices, chemical sensors, and ‘‘smart windows.’’ Furthermore, they are of interest as model materials for investigations of diffusion in disordered systems. In some of the glasses the diffusion can be extraordinary fast, comparable to that in liquid electrolytes 共the ionic conductivity may be as high as 10⫺2 S/cm at room temperature1兲, and occurs in an otherwise frozen environment. Both for technical and fundamental reasons it is of interest to try to understand the fast ion diffusion on the basis of the microscopic properties. However, despite considerable experimental and theoretical efforts, the conduction mechanism is not yet fully understood, which is partly due to an incomplete knowledge of the microscopic structure, especially on an intermediate length scale of 5–50 Å. Several models, such as the random site model,2 the dynamic structure model,3 the diffusion pathway model,4 the cluster model,5–10 and the cluster-bypass model,11 have been proposed to explain the high ionic conductivity, and most of them involve specific or indirect assumptions about the intermediate-range structure of the glass. Given the various assumptions of the different models, it is of interest to seek a better understanding of the misoscopic structure of fast ion conducting glasses. Previous wide-angle diffraction experiments, in combination with reverse Monte Carlo 共RMC兲 modeling,12,13 on metal-halide-doped fast ion conducting glasses have provided interesting indications concerning the intermediaterange order and chemical heterogeneities on length scales in the range 10–50 Å. The RMC results14 –18 gave evidence for the structure on this length scale being very different for different glass compositions and metal-halide salts. The RMC-produced models suggest that particularly the alkali0163-1829/2001/64共13兲/134201共7兲/$20.00

halide-doped diborate glasses17 and the metaborate glass (AgI兲0.6-共Ag2O-B2O3兲0.4 共Ref. 18兲 should exhibit considerable scattering intensity at low-Q values below 0.5 Å⫺1. In the case of (AgI兲0.6-共Ag2O-B2O3兲0.4 the structural model suggests that the glass contains relatively large clusters of AgI and that the structure shows similarities to the ␣-AgI composites (AgI) x -共3Ag2O-B2O3 ) 1⫺x (x⬎0.6), where structural inhomogeneities on length scales up to 60–70 Å have been observed.19 Such an interpretation has furthermore been supported by Raman,7,8 specific heat,20 and NMR 共Ref. 21兲 results, which have indirectly been interpreted as that metalhalide-doped borate glasses may contain microheterogeneities in the range 10–100 Å. In order to test the structural indications given above, we have to extend the structural investigations to much larger length scales. The aim of this study is therefore to combine small- and wide-angle neutron scattering with RMC modeling in order to produce large structural models of fast ion conducting glasses. This is the first study where the RMC method has been applied to small-angle neutron diffraction data. Using this approach, we are particularly interested to elucidate whether the glasses exhibit structural inhomogeneities on length scales considerably larger than that indicated by the position of the prepeak observed in wide-angle diffraction experiments. The present results partly support the earlier indications of heterogeneities on an intermediaterange length scale and show that these are mainly caused by an inhomogeneous distribution of the salt ions, although the prepeak at about 0.46 Å⫺1 in the total structure factor of (AgI兲0.6-共Ag2O-B2O3兲0.4 is almost entirely due to a characteristic distance of approximately 2 ␲ /0.46⬇14 Å between borate segments separated by salt ions. The small-angle diffraction data on the CsI- and PbI2-doped silver phosphate glasses show that only minor structural heterogeneities are present in these glasses on length scales ⬎10 Å. The experimental results are in agreement with our earlier findings from RMC modeling based on only wide-angle diffraction data of the two glasses.22,23 Due to the lack of pronounced structural

64 134201-1

©2001 The American Physical Society

¨ RJESSON, AND HEENAN SWENSON, KARLSSON, BO

PHYSICAL REVIEW B 64 134201

inhomogeneities on longer length scales, there was no need to produce any larger structural models of these glasses. II. EXPERIMENT

The ion conducting glasses LiCl-Li2O-2B2O3, (AgI兲0.6-共Ag2O-B2O3兲0.4, CsI-4AgPO3, and PbI2-9AgPO3 were prepared using melt quenching according to procedures described previously.22–25 In the case of the borate glasses, boron was isotopically enriched in 11B(⬎99%) in order to minimize the influence of the high neutron absorption of 10B present in natural boron. The mass density of the samples was measured by the Archimedes method using methanol. The number densities were 0.905, 0.059, 0.0546, and 0.066 Å⫺3 共with error bars of ⫾2%兲 for the LiCl-Li2O-2B2O3, (AgI兲0.6-共Ag2O-B2O3兲0.4, CsI-4AgPO3, and PbI2-9AgPO3 glasses, respectively. The neutron diffraction experiments were carried out on the low-Q time-of-flight diffractometer 共LOQ兲 共Ref. 26兲 and the liquid and amorphous materials diffractometer 共LAD兲,27 both situated at ISIS, Rutherford Appleton Laboratory, UK. The samples for the LOQ experiments were 1-mm-thick disks with a diameter of about 15 mm. The sample disks were mounted perpendicular to the incident beam without any use of sample container. LOQ uses wavelengths in the range 2.2–10 Å by the time-of-flight technique. There are two detector banks: a main detector located 4.1 m from the sample covering the Q range 0.008 – 0.25 Å⫺1 and a high-angle detector at 0.5 m from the sample covering 0.15⬍Q⬍1.6 Å ⫺1 . The incident beam had a diameter of 8 mm. The LOQ data were corrected for absorption by measuring the transmissions of the samples. The experimental procedures and data corrections of the LAD data are described in previous publications.17,18,22,23 The LOQ and LAD data were merged into a total structure factor somewhere in the Q range 0.2–1 Å⫺1, where the two data sets overlapped. Since the LAD data are rather easy to normalize correctly, this procedure helped us to also normalize the LOQ data, which in general are more difficult to normalize. The combined total structure factors should therefore be quantitatively reliable, containing only comparably small systematic experimental errors of the order of ⫾10% or less 共mainly due to the uncertainty in the normalization of the LOQ data兲. Such error bars have negligible influence on the present qualitative analysis of the RMC-produced structural models. III. REVERSE MONTE CARLO MODELING

The reverse Monte Carlo modeling technique has been extensively described elsewhere28,29 and will therefore not be discussed here. Here we will only give some specific details on the present RMC modeling, which was only performed on the LiCl-Li2O-2B2O3 and (AgI兲0.6-共Ag2O-B2O3兲0.4 glasses since neither the previous structural models of CsI-4AgPO3 and PbI2-9AgPO3, based on only wide-angle diffraction data, nor the present LOQ data show any indications of substantial structural inhomogeneities on length scales ⬎10 Å. Since the structural models have to be twice as large as the maximum length scale one wants to study, this means that a very large

number of atoms has to be used for investigations of length scales up to 40–50 Å. However, in the present study we are only considering length scales ⬎5 Å, which implies that we can neglect the exact short-range correlations. Therefore, by treating each ‘‘BO1.75 group’’ as one particle in the modeling of the LiCl-Li2O-2B2O3 glass and each ‘‘BO2 group’’ as one particle in the case of (AgI兲0.6-共Ag2O-B2O3兲0.4, we obtained box lengths of 85 and 95 Å, respectively, using only 32 000 particles in the simulations. Another advantage with the simplified network structure is that we could use a rather soft constraint for the network connectivity, in contrast to the more detailed connectivity constraints we had to use in our previous RMC modeling of the same glasses,17,18 where B and O were treated as individual atoms. In the case of the LiCl-Li2O-2B2O3 glass, we let 45% of the borate units 共i.e., BO1.75兲 be four coordinated and the remaining 55% three coordinated, between 1.25 and 3.6 Å, in accordance with 10 B NMR data on Li2O-2B2O3, 30 suggesting that 55% and 45% of the borons are three and four coordinated, respectively, to oxygens and that basically all oxygens are bridging 共1.25 Å was chosen as the lower limit because it is approximately equal to the absolutely nearest B-O distance, taken into account the vibrational displacements; 3.6 Å is a rough estimate of the largest B-O distance between two neighboring ‘‘BO2 groups’’兲. The network structure of the (AgI兲0.6-共Ag2O-B2O3兲0.4 glass is based on magic angle spinning NMR data on Ag2O-B2O3, 31 which has indicated that approximately 53% of the borons are four coordinated to oxygens. Provided that these oxygens are bridging and that no BO3 units, with all oxygens bridging, are present, the remaining borons must 共from a stociometric consideration兲 be located in metaborate structural groups, in accordance to what has been found for Li2O-B2O3. 30 These metaborate groups form chains of interconnected BO3 triangles with one nonbridging oxygen per BO3. Thus, for this glass, we let 53% of the borate units be four coordinated and the remaining 47% two coordinated within the distance 1.25–3.6 Å. This simplification of the B-O network should not have any significant effect on intermediate-range structural correlations, but provides a realistic network structure simultaneously as the formation of any unrealistically dense network cluster is prevented. In order to test the influence of the exact bonding constraint and the initial configuration for the final structural model, we produced an additional configuration of the AgI-doped metaborate glass based on a slightly different bonding constraint. Reassuringly, the two structural models of the glass showed only minor differences on length scales ⬎5 Å. To avoid the formation of similarly dense clusters of salt ions, we have also included closest allowed atomatom distances, i.e., to avoid overlap of atoms, in the modeling. The following closest particle-particle distances were used in the simulations: 1.25 Å for BO-BO, 1.75 Å for LiBO, 2.1 Å for Ag-BO, 2.3 Å for Li-Cl, 2.4 Å for Li-Li, 2.5 Å for Ag-I, 2.6 Å for Cl-BO, 2.7 Å for Ag-Ag, 2.8 Å for I-BO, 3.0 Å for Cl-Cl, and 3.6 Å for I-I. ‘‘BO’’ denotes the borate groups used in the simulations 共BO1.75 and BO2, respectively兲. The values given above were determined from experimentally obtained 共total and partial兲 pair correlation functions of these and similar glass compositions 共see, for

134201-2

STRUCTURAL INHOMOGENEITIES IN FAST ION . . .

PHYSICAL REVIEW B 64 134201

FIG. 1. Experimental structure factors S(Q) 共solid lines兲 for LiCl-Li2O-2B2O3 共a兲, (AgI兲0.6-共Ag2O-B2O3兲0.4 共b兲, CsI-4AgPO3 共c兲, and PbI2-9AgPO3 共d兲. The dotted lines are corresponding neutronweighted S(Q) from the RMC configurations of the two borate glasses.

nounced in the phosphate glasses, although one should note that the dopant concentration is lower in these glasses. The weak prepeaks and low scattering intensities of these glasses in the Q range 0.1–0.5 Å⫺1 suggest that only minor structural inhomogeneities are present on a length scale of 10–50 Å. Larger density fluctuations may, however, occur on longer length scales since the scattering intensity increases for the lowest Q values. The LiCl-Li2O-2B2O3 glass is different in the sense that is has no prepeak; however, it shows considerable scattering intensity in the low Q range 0.05–0.5 Å⫺1. The dotted lines in Fig. 1 are the computed neutronweighted S(Q) for the RMC configurations of LiCl-Li2O-2B2O3 and (AgI兲0.6-共Ag2O-B2O3兲0.4. As seen in Fig. 1, the agreement with the corresponding experimental data is excellent for both glasses. The combination of the good agreement to the experimental data and the applied network constraints should ensure that the configurations contain major structural features on the length scale ⬃5–50 Å. This is particularly the case for the intermediate-range order of the B-O network, due to the presence of the network

instance, Refs. 14, 17, and 32兲, geometrical considerations, and tabulated ionic radii. Before the actual RMC simulations were started, we ran hard-sphere Monte Carlo simulations to fulfill the applied constraints. First, we introduced only the borate groups 共randomly兲 and moved them around until these units were properly connected. Second, the salt ions were randomly added into the atomic computer configurations and moved apart from each other and from the borate groups in order to fulfill the closest atom-atom constraints. Only thereafter was the actual fitting procedure of the model structure factors to the experimental small- and wide-angle neutron diffraction data started. We used the experimental Q range of 0.05– 4 Å⫺1 for this fitting procedure. IV. RESULTS

The structure factors S(Q) of the investigated glasses are shown in Fig. 1. The narrow- and wide-angle neutron diffraction data are shown in the Q range of interest for the present study 共0.05–2 Å⫺1兲. From the figure it is evident that all glasses, except LiCl-Li2O-2B2O3, exhibit some kind of prepeak. In the case of the (AgI兲0.6-共Ag2O-B2O3兲0.4 glass, the prepeak is extraordinarily intense and located at an anomalously low Q value of 0.46 Å⫺1, giving an indication of density or chemical fluctuations on a length scale of about 2 ␲ /0.46⬇14 Å. This glass shows also a very high scattering intensity at lower Q values, indicating the presence of substantial structural inhomogeneities on length scales considerably longer than that corresponding to the position of the prepeak. The two phosphate glasses have much weaker prepeaks 共particularly the PbI2-doped glass兲 located at significantly higher Q values 共0.75 and 1.0 Å⫺1 for PbI2-9AgPO3 and CsI-4AgPO3, respectively兲. This indicates that the characteristic intermediate-range distance is considerably shorter 共approximately in the range 6 – 8 Å兲 and much less pro-

FIG. 2. Partial structure factors S i j (Q) calculated from the RMC configuration of 共AgI兲0.6-共Ag2O-B2O3兲0.4. Different scales are used in 共a兲 and 共b兲, and consecutive curves are shifted vertically by 10 in 共a兲 and 4 in 共b兲, for clarity.

134201-3

¨ RJESSON, AND HEENAN SWENSON, KARLSSON, BO

PHYSICAL REVIEW B 64 134201

FIG. 3. Partial structure factors S i j (Q) calculated from the RMC configuration of LiCl-Li2O-2B2O3. Consecutive curves are shifted vertically by 3, for clarity.

constraints and the fact that the neutron data are most sensitive to the network structure. The distributions of salt ions are less reliable due to the absence of contrasting data sets 共e.g., from x-ray diffraction兲. However, since we have realistic bonding constraints and reliable network structures, the distributions of salt ions must also be somewhat constrained. This implies that the main features of the distributions of salt ions should be at least qualitatively correct. In Fig. 2 we show the partial structure factors S i j (Q) of the RMC, produced configuration of (AgI兲0.6共Ag2O-B2O3兲0.4. It is seen in Fig. 2共a兲 that the partial structure factors either increase or decrease with decreasing Q for Q values less than about 0.15 Å⫺1. This implies, as also the total S(Q) suggested, that there are correlations between chemical inhomogeneities on length scales longer than the 14 Å corresponding to the position of the prepeak in the total S(Q). Since the partial structure factors involving correlations between the salt ions as well as between the BO2 groups increase for decreasing Q, whereas the cross correlations between the salt ions and the borate network are going in the opposite direction, it is evident that the origin is chemical fluctuations between AgI-rich domains and highdensity regions of network atoms. From Fig. 2共b兲 it is clear that the prepeak at 0.46 Å⫺1 in total S(Q) is almost entirely caused by correlations between the network atoms. Thus the real-space characteristic distance of approximately 14 Å is due to a typical distance between segments of boron-oxygen groups. Figure 3 shows the partial structure factors of the RMCproduced configuration of the LiCl-Li2O-2B2O3 glass. It can be seen that the low-Q features are much weaker for this glass, although some substantial longer-scale 共⬎10 Å兲 inhomogeneities are observed in the distribution of salt ions, particularly Cl⫺. The two partial structure factors S LiCl(Q) and S ClCl(Q) show peaks at about 0.14 Å⫺1, suggesting a typical distance of 40–50 Å between high-density regions of LiCl. However, one has to be careful in the interpretation of features on this length scale since they may be caused by the

FIG. 4. Partial reduced radial distribution functions d i j (r) of 共AgI兲0.6-共Ag2O-B2O3兲0.4 共a兲 and LiCl-Li2O-2B2O3 共b兲. Consecutive curves are shifted vertically by 0.2 in 共a兲 and 0.1 in 共b兲, for clarity.

finite size of the structural model. More conclusive results are observed for the correlations between the network atom, where S BOBO(Q) shows no evidence for density fluctuations within the B-O network on length scales ⬎10 Å. In order to interpret the nonobvious features in reciprocal space, we have also calculated the partial reduced radial distribution functions d i j (r) of the two RMC-produced configurations: see Figs. 4共a兲 and 4共b兲. d i j (r) is related to S i j (Q) through the following Fourier transformation:

d i j共 r 兲⫽

2 ␲





0

Q 关 S i j 共 Q 兲 ⫺1 兴 sin共 rQ 兲 dQ.

共1兲

The observation in reciprocal space of chemical fluctuations on rather long length scales in the (AgI兲0.6-共Ag2O-B2O3兲0.4 glass is confirmed in real space. Figure 4共a兲 shows also that several d i j (r), particularly d II(r), do not reach their average values until after about 40 Å, and typically they show a very broad peak in the range 15–25 Å. The prepeak at 0.46 Å⫺1 in total S(Q), as well as in S BOBO(Q), is observed in real space

134201-4

STRUCTURAL INHOMOGENEITIES IN FAST ION . . .

PHYSICAL REVIEW B 64 134201

as a small peak at about 12 Å in d BOBO(r). The peak in real space is, however, rather weak considering the high intensity of the peak in total S(Q). The partial reduced radial distribution functions shown in Fig. 4共b兲 for the LiCl-Li2O-2B2O3 glass are different from the corresponding d i j (r) for the AgI-doped metaborate glass. In this case it is evident from S BOBO(Q) that there are basically no density fluctuations within the B-O network on a length ⬎5 Å. The most pronounced density fluctuations are seen in d ClCl(r), whose intensity is significantly above its average value for r⬍12 Å. Furthermore, the peak at about 20–25 Å in d ClCl(r) is most likely due to a characteristic distance between these salt-rich regions. V. DISCUSSION

The neutron diffraction results of LiCl-Li2O-2B2O3, (AgI兲0.6-共Ag2O-B2O3兲0.4, CsI-4AgPO3, and PbI2-9AgPO3 show that the borate glasses exhibit more pronounced structural inhomogeneities than the phosphate glasses on length scales in the range 10–100 Å. Particularly strong heterogeneities were found for the (AgI兲0.6-共Ag2O-B2O3兲0.4 glass, where two interesting features were observed in the d i j (r) involving correlations between the salt ions: see Fig. 4共a兲. First, these partials do not reach their average values until after about 40 Å and, second, they show a broad peak in the range 15–25 Å. The combination of these two findings implies that there are heterogeneities of widely different length scales. Simultaneously, as approximately 20 Å must be one kind of typical distance between salt-rich regions, the comparably high intensities 共relative to the average intensities兲 for 5⬍r⬍25 Å in combination with the rapidly decreasing intensities in the r range 25– 40 Å implies that some AgI-rich regions must be as large as 15–20 Å in radius. Due to the finite box length of 95 Å for the RMC configuration, it is not possible to determine whether there is any correlation between these larger salt-rich regions. From the partial d ClCl(r) shown in Fig. 4共b兲 for the LiCl-Li2O-2B2O3 glass, it is evident that the inhomogeneities in the distribution of salt ions, particularly the anions, are much more rapidly varying in space than was found for the AgI-doped metaborate glass. This is partly because the average number concentration c of salt ions is lower in the LiCldoped glass than in the highly network modified and AgIdoped glass (c LiCl /c AgI⬇0.58), which tends to make the size of the LiCl-rich regions smaller and comparably 共relative to the average density兲 more dense. However, there are also similarities to the (AgI兲0.6-共Ag2O-B2O3兲0.4 glass, since the most pronounced longer-scale heterogeneities are observed in the distribution of anions, suggesting that the glass structure consists of salt-rich and salt-poor regions. The chemical fluctuations that are evident from the partial S i j (Q) and d i j (r) can, in fact, be observed by a direct inspection of the RMC-produced structural models. Figures 5共a兲 and 5共b兲 show the distribution of BO2 groups and the distribution of silver and iodine ions, respectively. The figures show a 6-Å-thick slice of the RMC configuration. The origin of the prepeak in the total S(Q) is clearly seen in Fig. 5共a兲 as a common distance of approximately 10–15 Å be-

FIG. 5. A 6-Å-thick slice of the RMC configuration of 共AgI兲0.6-共Ag2O-B2O3兲0.4. 共a兲 shows the distribution of BO2 network units and 共b兲 the distribution of silver and iodine ions. The radii of the chemical components are BO2 ⫽1.2 Å, Ag⫹ ⫽0.8 Å, and I⫺ ⫽1.6 Å.

tween neighboring borate segments. By comparing Figs. 5共a兲 and 5共b兲, it is evident that the salt ions occupy the empty spaces between the borate segments, which means that some ‘‘tubelike’’ domains of AgI 共with a size of ⬃10–15 Å兲 are present in the largest voids of the B-O structure. However, as can be seen in Fig. 5共b兲, there are no voids of significant sizes in the distribution of salt ions. This is partly because some of the Ag⫹ ions are involved in the network formation, coordinating to different oxygens, but also because there are no particularly large aggregates of borate groups. The chemical fluctuations observed in Fig. 4共b兲 for the LiCl-Li2O-2B2O3 glass are also evident from a visual inspection of the RMC-produced structural model. Figures 6共a兲 and 6共b兲 show 6-Å-thick slices of the distribution of BO1.75 groups and the distribution of lithium and chlorine ions, respectively. The inhomogeneities in the distribution of salt

134201-5

¨ RJESSON, AND HEENAN SWENSON, KARLSSON, BO

PHYSICAL REVIEW B 64 134201

with two nonbridging oxygens.33 Thus it is obvious, particularly for high salt concentrations, that it is easier to have the comparably thin phosphate chains homogeneously dispersed in the glass than the interconnected superstructural borate units, which unambiguously must cause high local densities of network atoms. The distance between these high-density regions increases with increasing dopant concentration, and for a glass composition such as (AgI兲0.6-共Ag2O-B2O3兲0.4 the distance is large enough to produce chemical fluctuations on length scales ⬎10 Å. Finally, we believe the present results on the AgI-doped metaborate glass also explain why it is very close to the limit of glass formation. It is evident that the size of the AgI domains grows with increasing salt concentration, and for high enough concentrations the link to the surrounding network segments will be too weak to prevent crystallization. Instead, with increasing AgI concentration the regime of the ␣-AgI composites is approached.19

VI. CONCLUSIONS

FIG. 6. A 6-Å-thick slice of the RMC configuration of LiCl-Li2O-2B2O3. 共a兲 shows the distribution of BO1.75 network units and 共b兲 the distribution of lithium and chlorine ions. The radii of the chemical components are BO1.75⫽1.2 Å, Li⫹ ⫽0.7 Å, and Cl⫺ ⫽1.6 Å.

ions, as indicated by d ClCl(r) in Fig. 4共b兲, are clearly seen in Fig. 6共b兲 as salt-rich and salt-poor regions. In Fig. 6共a兲 it can be seen that there are no voids of significant sizes 共⬎10 Å兲 in the borate network, which means that we cannot have any larger domains of LiCl 关similar to the AgI clusters found in (AgI兲0.6-共Ag2O-B2O3兲0.4兴, only fluctuations of the salt concentration on a length scale ⬎10 Å. From the small-angle diffraction data, it is clear that the salt-doped silver phosphate glasses show considerably weaker structural inhomogeneities compared to the AgI and LiCl-doped borate glasses. This is an interesting result since NMR measurements30 have shown that the network structures of network-modified borate and phosphate glasses are very different. The borate glasses seem to consist of relatively large interconnected superstructural units,30 whereas the metaphosphate composition contains chains of PO4 units

In this study we have performed small- and wide-angle diffraction measurements on the fast ion conducting glasses LiCl-Li2O-2B2O3, (AgI兲0.6-共Ag2O-B2O3兲0.4, CsI-4AgPO3, and PbI2-9AgPO3. The experimental results show that the two borate glasses exhibit structural inhomogeneities on length scales considerably larger than indicated by the position of the first diffraction peak, in contrast to the CsI-4AgPO3 and PbI2-9AgPO3 glasses, which show no evidence for substantial structural heterogeneities on length scales ⬎10 Å. In order to elucidate the nature of the structural inhomogeneities in the borate glasses, we have used the reverse Monte Carlo method and produced considerably larger and more accurate structural models of the intermediate-range order 共5–50 Å兲 than previously has been done on any fast ion conducting glass. From the RMC modeling it is clear that the origin of the extraordinary intense prepeak at the anomalously low Q value of 0.46 Å⫺1 in the total neutron structure factor of (AgI兲0.6-共Ag2O-B2O3兲0.4 is due to a characteristic distance between borate segments separated by salt ions. However, this glass shows also chemical inhomogeneities on considerably longer length scales, which is most evident from the distribution of salt ions where density fluctuations on, at least, two different length scales were observed. The LiCl-Li2O-2B2O3 glass has a more homogeneous structure than the AgI-doped metaborate glass, but exhibits clear fluctuations in the concentration of salt ions on a length scale 10–50 Å. No prepeak is observed for this glass due to a very homogeneous distribution of the network atoms and the rather uncorrelated density fluctuations in the distribution of salt ions. To conclude, the present results show the difficulties in describing the intermediate-range structure of glasses in general and chemically complicated multicomponent glasses, such as the present fast ion conducting glasses, in particular. The glasses show chemical fluctuations on different length scales, and generally these fluctuations are very weak and the length scales weakly defined. Thus we are observing weak, but still significant, structural correlations on an otherwise

134201-6

STRUCTURAL INHOMOGENEITIES IN FAST ION . . .

PHYSICAL REVIEW B 64 134201

completely disordered structure. In this study we have elucidated these weak structural inhomogeneities, which indeed are clearly different for different glasses, more elaborately than in previous studies of fast ion conducting glasses.

J. Kincs and S. W. Martin, Phys. Rev. Lett. 76, 70 共1996兲. A. M. Glass and K. Nassau, J. Appl. Phys. 51, 3756 共1980兲. 3 P. Maass, A. Bunde, and M. D. Ingram, Phys. Rev. Lett. 68, 3064 共1992兲. 4 G. N. Greaves, J. Non-Cryst. Solids 71, 203 共1985兲. 5 J. P. Malugani, M. Tachez, R. Mercier, A. J. Dianoux, and P. Chieux, Solid State Ionics 23, 189 共1987兲. 6 M. Tachez, R. Mercier, J. P. Malugani, and P. Chieux, Solid State Ionics 25, 263 共1987兲. 7 A. Fontana, F. Rocca, and M. P. Fontana, Phys. Rev. Lett. 58, 503 共1987兲. 8 A. Fontana, F. Rocca, and M. P. Fontana, Philos. Mag. B 56, 251 共1987兲. 9 C. Rousselot, M. Tachez, J. P. Malugani, R. Mercier, and P. Chieux, Solid State Ionics 44, 151 共1991兲. 10 C. Rousselot, J. P. Malugani, R. Mercier, M. Tachez, P. Chieux, A. J. Pappin, and M. D. Ingram, Solid State Ionics 78, 211 共1995兲. 11 M. D. Ingram, Philos. Mag. B 60, 729 共1989兲; Mater. Chem. Phys. 23, 51 共1989兲. 12 R. L. McGreevy and L. Pusztai, Mol. Simul. 1, 359 共1988兲. 13 D. A. Keen and R. L. McGreevy, Nature 共London兲 344, 423 共1990兲. 14 J. Swenson, L. Bo¨rjesson, R. L. McGreevy, and W. S. Howells, Phys. Rev. B 55, 11 236 共1997兲. 15 J. Wicks, L. Bo¨rjesson, R. L. McGreevy, W. S. Howells, and G. Bushnell-Wye, Phys. Rev. Lett. 74, 726 共1995兲. 16 J. Swenson, R. L. McGreevy, L. Bo¨rjesson, and J. D. Wicks, Solid State Ionics 105, 55 共1998兲. 1 2

ACKNOWLEDGMENT

This work was financially supported by the Swedish Natural Science Research Council.

17

J. Swenson, L. Bo¨rjesson, and W. S. Howells, Phys. Rev. B 57, 13 514 共1998兲. 18 J. Swenson, L. Bo¨rjesson, and W. S. Howells, J. Phys.: Condens. Matter 11, 9275 共1999兲. 19 C. Guy, N. Umesaki, N. Kamijo, M. Tatsumisago, N. Torata, T. Minami, and M. Furusaka, Physica B 213&214, 493 共1995兲. 20 A. Avogadro, S. Aldrovandi, F. Borsa, and G. Carini, Philos. Mag. B 56, 227 共1987兲. 21 M. Villa, G. Chiodelli, A. Magistris, and G. Licheri, J. Chem. Phys. 85, 2392 共1986兲. 22 J. Swenson et al., Phys. Rev. B 60, 12 023 共1999兲. 23 J. Swenson, A. Matic, C. Karlsson, L. Bo¨rjesson, and W. S. Howells, J. Non-Cryst. Solids 263&264, 73 共2000兲. 24 L. Bo¨rjesson, Phys. Rev. B 36, 4600 共1987兲. 25 L. Bo¨rjesson and L. M. Torell, Solid State Ionics 25, 85 共1987兲. 26 R. K. Heenan, J. Penfold, and S. M. King, J. Appl. Crystallogr. 30, 1140 共1997兲. 27 W. S. Howells 共unpublished兲. 28 R. L. McGreevy, Annu. Rev. Mater. Sci. 22, 217 共1992兲. 29 R. L. McGreevy, Nucl. Instrum. Methods Phys. Res. A 354, 1 共1995兲. 30 S. A. Feller, W. J. Dell, and P. J. Bray, J. Non-Cryst. Solids 51, 21 共1982兲. 31 C. Chiodelli, A. Magistris, M. Villa, and J. L. Bjorkstam, J. NonCryst. Solids 51, 143 共1982兲. 32 J. Swenson, L. Bo¨rjesson, and W. S. Howells, Phys. Rev. B 52, 9310 共1995兲. 33 R. K. Brow, C. C. Phifer, G. L. Turner, and R. J. Kirkpatrick, J. Am. Ceram. Soc. 74, 1287 共1991兲.

134201-7