Maria
Gombotz
*a,
Sarah
Lunghammer
a,
Stefan
Breuer
a,
Ilie
Hanzu
ab,
Florian
Preishuber-Pflügl‡
a and
H. Martin R.
Wilkening
*ab
aChristian Doppler Laboratory for Lithium Batteries, and Institute for Chemistry and Technology of Materials, Graz University of Technology (NAWI Graz), Stremayrgasse 9, 8010 Graz, Austria. E-mail: wilkening@tugraz.at; gombotz@tugraz.at; Fax: +43 316 873 32332; Tel: +43 316 873 32330
bAlistore-ERI European Research Institute, 33 rue Saint Leu, 80039 Amiens, France
First published on 9th January 2019
Diffusion of small ions in materials with confined space for translational dynamics can be quite different to isotropic (3D) diffusion, which is found in the majority of solids. Finding credible indications for 2D diffusion is not as easy as it looks at first glance, especially if only powder samples are available. Here we chose the ternary fluoride RbSn2F5 as a new model system to seek out low-dimensional anion diffusion in a nanocrystalline material. We prepared RbSn2F5via mechanochemically-assisted solid state synthesis and used both ac conductivity spectroscopy and spin-lock NMR relaxation measurements to find evidence that the fluorine ions preferably diffuse between the Rb-rich layers. In both cases the diffusion induced spin-lock NMR rates are only consistent with conductivity data if they are analyzed with the semi-empirical spectral density function for 2D jump diffusion as introduced by P. M. Richards [Solid State Commun., 1978, 25, 1019].
Besides such application-oriented challenges, fluorine ion conductors offer attractive model systems to study ultra-rapid jump processes8,25 from a fundamental point of view. To develop new functional materials we need to improve our understanding about the influence of crystal structure and morphology on ion transport. The highly reversible insertion and de-insertion processes in rocking-chair batteries, relying on insertion compounds, take advantage of layered materials that offer fast 2D diffusion pathways. From an atomic scale point of view it is, if only powder samples are available, a challenge to undoubtedly show that 2D diffusion prevails.26–29 The same troubles hold for 1D diffusion taking place along or inside the channels.30,31 Intrachannel and intralayer hopping processes may water the reasoning for 1D or 2D diffusion down.
Only few examples have been presented in literature26,27 for which rapid 2D diffusion of Li, Na or F ions has either unequivocally been shown or at least strongly anticipated. These examples include Na-beta′′-alumina,33–35 cathode materials based on LiyCoO2 (y ≤ 1) and LiFePO4,36,37 Li containing transition metal sulfides (or selenides), such as LixTiS2 (0 < x ≤ 1)38–41 or LizNbS2 (0 < z ≤ 1),29 hexagonal LiBH4,27 polycrystalline ZrBe2H1.4,26,42 and the ternary fluorides PbSnF4, see ref. 43 and 44, and BaSnF4.45 In some of these examples only a multi-method approach turned out to be successful to show that low-dimensional diffusion is present. For LixMS2 (M = Ti, Nb) and ZrBe2H1.4 the use of variable-frequency nuclear magnetic resonance (NMR) relaxation measurements turned out to be successful to prove low-dimensional diffusion as the main motional process. For instance, McDowell et al.26 impressively showed via variable-frequency 1H NMR spin–lattice relaxation measurements that the protons in the metal-hydride ZrBe2H1.4 are indeed subjected to 2D diffusion.
For many cathode materials, which are used in insertion batteries,37,46–48 the NMR technique will, however, be of very limited use as paramagnetic centers cause short relaxation times which drastically narrow the time window needed for dimensionality analyses. In addition, strong spin-electron interactions govern the NMR relaxation rates rather than diffusive motions. Focussing here on rapid anion dynamics, ternary fluorides provide some very encouraging materials useful as model substance to test NMR relaxation models proposed for 2D diffusion.49–51
Here, RbSn2F5 attracted our attention. It belongs to the family of the pseudo-binary system MF-SnF2 where M is a monovalent cation such as Rb+, K+, Na+ or Cs+.52 Within this family RbSn2F5 offers the highest ion conductivity53 followed by KSn2F5 and NaSn2F5.54,55 Up to now it has only been prepared by solid-state reaction52 or hydrothermal synthesis in aqueous solution as first reported by Donaldson and O'Donoghue in 1964.56 In 1987 the structure of KSn2F5, to which RbSn2F5 is isomorphous, was resolved as being trigonal, i.e., crystallizing with the space group P.57 Later structure refinement was also carried out by Yamada et al.32 In Fig. 1 the crystal structure of RbSn2F5 is illustrated, which clearly shows the two-dimensional nature of this compound. In 1991 Hirokawa et al.58 suggested to interpret NMR data by using a 2D spectral density. Later, in 2004, Yamada et al.32 pointed out that also conductivity isotherms are in line with low-dimensional transport. A direct comparison of results from both NMR and conductivity spectroscopy is, however, still missing. While earlier studies focussed on coarse-grained polycrystalline samples with μm-sized crystallites, F self-diffusion in nanocrystalline RbSn2F5 has not been reported yet. Thus, the present study also aims at discussing possible effects on 2D diffusion in RbSn2F5 when the mean crystallite size is reduced to ca. 20 nm.
Fig. 1 Crystal structure of RbSn2F5 below the phase transition at approximately T = 368 K. The illustration shows the arrangement of atoms according to the structure refinement of Yamada et al.32 (space group P, no. (147); a = 7.3857(4) Å, c = 10.104(1) Å, Z = 3.) (a) view to illustrate the fluorine ions in the Sn-rich layers. (b) View along the b-axis. (c) View along the b-axis to visualise the hexagonal arrangement of the Rb+ cations. The F sites between the Rb-rich layers (labelled F2, F3, and F4) are only filled by 90%; F1 sites are fully occupied. |
For this purpose, that is, to compare F diffusion in microcrystalline RbSn2F5 with that in the nanocrystalline form, we synthesized the layer-structured fluoride employing a mechanochemically-assisted ceramic route. Using a one pot synthesis, nanocrystalline RbSn2F5 was directly obtained after high-energy ball milling59 the binary starting materials RbF and SnF2. Through soft annealing we converted the as-prepared material into a microcrystalline sample. Spin-lock 19F NMR60 instead of ordinary spin–lattice relaxation NMR61 was used to record purely diffusion-controlled relaxation rates. For non-nanocrystalline RbSn2F5 the NMR rates are consistent with results from conductivity measurements only if we analyse the NMR rate peak with the model of Richards49,62 introduced for 2D jump diffusion. This result is also consistent with the electrical responses seen by conductivity spectroscopy. These findings, being in close agreement to results from literature,32 build the basis for the analysis of F anion dynamics in nanocrystalline RbSn2F5.
The milled and annealed samples were characterised by X-ray diffraction (XRD) under atmospheric pressure and at room temperature. A Bruker D8 Advance diffractometer with Bragg Brentano geometry and CuKα radiation (1.5406 Å) was used to collect the diffractograms. Data points were recorded from 10 to 100° 2θ with a stepsize of 0.02° 2θ; the measuring time for each step was 1 second. Rietveld Refinement was performed with X-PertHighScorePlus (PANalytical).
Magic angle spinning (MAS) NMR was carried out on a Bruker Avance III 500 spectrometer at a nominal magnetic field of 11.7 T. This external magnetic field translates in resonance frequencies of 186.40 MHz for 119Sn, 470.30 MHz for 19F and 163.60 MHz for 87Rb. We used a Bruker MAS probe designed for 2.5 mm rotors (ZrO2) to acquire free induction decays (FIDs) at a spinning speed of 25 kHz (target gas flow: 400 L h−1, frame cooling: 35% of target gas flow). The bearing gas temperature was 293 K.
Spectra were obtained after Fourier transformation of the FIDs and referenced to the isotropic chemical shifts δiso of LiF (δiso(19F) = −204.3 ppm), SnO2 (δiso(119Sn) = −604.3 ppm) and RbNO3 (δiso(87Rb) = −30 ppm site 3), respectively. The relaxation rates 1/T1 and 1/T1ρ were determined with a Bruker Avance III 300 NMR spectrometer at a magnetic field of 7.04 T, i.e., at a 19F NMR resonance frequency of 281.79 MHz. Prior to the relaxation measurements the powder sample was fire-sealed in a glass cylinder with a length of approximately 3 cm and 4 mm in diameter. The rates were recorded with the saturation recovery pulse sequence where at first a train of 10 closely spaced 90° pulses destroys any longitudinal magnetization Mz before its recovery as a function of delay time td was then detected with a single 90° pulse. The transients Mz(td) were analyzed with stretched exponentials Mz(td) ∝ 1 − exp(−(t/T1ρ)γ) with 0 < γ ≤ 1. While T1 NMR is sensitive to fast F anion dynamics, slower motional processes were measured with the spin-lock technique introduced by Slichter and Ailion64–66 utilizing the pulse sequence 90°(tlock) − acq.67–70 The locking frequency ω1 was set to ω1/2π ≈ 100 kHz and the duration of the locking pulse tlock was varied from 100 μs to 10 ms. Note that the recycle delay for the spin-lock experiments was at least 5 × T1 to ensure complete longitudinal relaxation between each scan. Once again, stretched exponentials served to parameterize the spin-lock transients Mρ(tlock) ∝ exp(−(tlock/T1ρ)γρ) (0 < γρ ≤ 1) to extract T1ρ(1/T).
To carry out the impedance spectroscopy measurements, approximately 60 mg of the sample powder was pressed to cylindrical pellets with a final diameter of 5 mm, using a hand press. As for the annealing step, we applied a uniaxial pressure of 0.5 t for 2 min. Au electrodes, which block ion transport, with a thickness of 100 nm were sputtered onto both sides of the pellets using a Leica sputter device (EM SCD050).
Alternating current (ac) impedance measurement were then performed with a broadband spectrometer (Novocontrol, Concept 80) in combination with an active BDS 1200 cell and a ZGS interface.71 The temperature was varied from 173 K to 473 K controlled by a QUATRO cryosystem. Our conductivity measurements covered a frequency range from 10−2 Hz to 107 Hz. All experiments were performed under a constant flow of dry, freshly evaporated N2 gas. ZView (Princeton Applied Research) and IGOR Pro (Wavemetrics) software were used to analyze the data.
For Hebb–Wagner-type polarisation measurements under N2 atmosphere we used pellets with 5 mm in diameter and employed a Parstat MC potentiostat (Princeton Applied Research) equipped with a low-current option. All preparation steps, including the metallisation, were carried out in Ar filled gloveboxes with an O2 and H2O content of less than 1 ppm.
Fig. 2 X-ray diffractograms of all RbSn2F5 samples prepared. XRD patterns were recorded at room temperature before and after the annealing steps indicated. The vertical bars shows the expected positions of the reflections for the low-T and high-T modification, respectively. While the reflections for the phase being stable above 368 K were taken from literature (ICSD no. 247178, recorded at 538 K); those for the low-T modification were constructed according to the structure model proposed by Yamada et al.32 |
Soft annealing, i.e., holding the sample for 8 h at 473 K under dry, oxygen-free inert gas atmosphere, clearly narrows the reflections because of crystallite growth and/or the release of strain. Lower annealing temperatures also lead to narrow reflections but the narrowing turned out to be less pronounced. Worth noting, if we simply store the ball-milled sample for several month in dry Ar atmosphere, we recognize a significant growth in average grain size. Therefore, crystallization also takes place under ambient conditions. At his temperature the crystallization kinetics are, of course, much slower than at high T. Obviously, crystal growth is driven by the high F anion conductivity of the samples. For comparison, at 293 K the ionic conductivity σdc is 6.24 × 10−5 S cm−1, see below.
The final reflections seen after soft annealing are in agreement with the structure refinement proposed by Yamada et al.32 assuming P symmetry and vacancies on the F1, F2 and F3 positions, see also Fig. 1. In Fig. 2 the positions of the main reflections of both the high-T and low-T phase of RbSn2F5 are indicated by vertical lines. We also included a simulation of the XRD pattern of the low-T form of RbSn2F5.
Annealing the samples at temperatures higher than 473 K, e.g., at 568 K for several hours, causes decomposition of the ternary fluoride. Light microscopy revealed small regions that shine metallic. We assume that, due to disproportionation of Sn4+, metallic tin is formed at sufficiently high T; cf. also Fig. S1 (see ESI†). Scanning electron microscopy showed a mixture of needle-like crystallites and rectangular tubes with a length of up to 20 μm, see also Fig. S1 (ESI†).
Below 368 K the ternary fluoride crystallizes with P symmetry (Z = 3); at higher temperatures RbSn2F5 reversibly transforms to P with Z = 1. The same transition has been reported for K2SnF5.57 Yamada et al. investigated this contraction along the c-axis. The contraction is accompanied by an expansion along the ab-axis.32 Defect disorder in the vacancy-rich (F1, F2, F3)-sublattices, caused by the phase transition, leads to an easily measurable increase in F anion conductivity. This increase is also seen here; we used broadband conductivity spectroscopy to study electrical responses over a wide range of temperature and a broad frequency region.
We will see later on that this view is supported by diffusion-induced nuclear spin relaxation. It is also in line with earlier reports on 2D ionic conduction in coarse-grained RbSn2F5 investigated by both impedance spectroscopy and NMR.32,58
A closer look at the data, especially if we include conductivity data recorded up to the GHz range, reveals that the situation is, however, more complex than can be recognized at first glance. At very low temperatures the isotherms, irrespective if we look at those from micro- or nanocrystalline RbSn2F5 (see Fig. 3a and b) deviate from the σ′ ∝ ν0.6 behaviour. Instead the dispersive regimes merge into a classical nearly constant loss (NCL) regime,74,75 being characterized by κ → 1. At sufficiently low T it is thus difficult to separate the two contributions from each other. On the other hand, at high temperatures, i.e., before the phase transformation of RbSn2F5 occurs, we can recognize that σ′ passes into a second dc plateau at frequencies above 106 Hz, see the arrow in Fig. 3a.
To clarify this feature we extended our measurements to frequencies in the GHz range. Fig. 4 shows the conductivity isotherm recorded at 333 K. It is clearly composed of two dc plateaus, the one occurring at high frequencies definitely belongs to bulk ion dynamics, while the one extending over a large frequency range at lower ν might also be influenced by grain boundary contributions. Importantly, the dispersive regime of the bulk plateau, whose conductivities obey an Arrhenius law with a very similar activation energy as σdc (see Fig. 3c) (0.52 eV), is characterised by κ = 0.54. This value is even closer to what we expect for 2D ionic conduction.72
The two contributions to the overall electrical response of σ′ can also be distinguished in the Nyquist plots of Fig. 5, where the imaginary part −Z′′ of the complex impedance Ẑ is plotted versus its real part Z′. The higher the temperature the better the separation; at low T the two semicircles merge into each other. Together with the NCL-type response, showing up at even lower T, they cannot be separately analyzed at T ≪ 293 K.
We analyzed the complex impedance data by using suitable equivalent circuits consisting of individual resistances R and constant phase elements (CPEs) representing the responses from bulk, grain-boundaries (g.b.) and electrodes. The complex impedance ẐCPE of a constant phase element can be simply expressed by
(1) |
C = R(1−n)/nQ1/n | (2) |
The two electrical relaxation processes are also seen in permittivity spectra. Whereas isotherms recorded at high T are less suitable to separate them (see the inset of Fig. 4a), at low T the real part, ε′, of the complex permittivity clearly reveals two contributions, see ESI† (Fig. S2). We attribute these two processes to electrical relaxation in the bulk and caused by g.b. regions. Similar features are also seen for the nanocrystalline sample, for which σdc almost coincides with that of the annealed sample, see Fig. 3c. In both cases, below 370 K the product σdcT follows an Arrhenius line characterized by 0.52 eV. This activation energy perfectly agrees with that published by Yamada et al.32 and Hirokawa et al.,58 both studies report on an activation energy of 0.54 eV.
Almost the same activation energy is obtained if we analyze tanδ peaks (0.48 eV, see Fig. 3c, right axis) or electric modulus peaks M′′ (0.49 eV), not shown here for the sake of brevity. The slight increase from 0.48 eV to 0.52 eV, which is seen when we compare the temperature behaviour of σdcT with that of νmax from tanδ, points to a very slight increase of the charge carrier concentration N with increasing temperature. For comparison, Yamada et al. reported that N did not change much with T.32
If we look at electrical properties recorded at fixed frequency but variable temperature further information about the energy landscape the F anions sense can be extracted. In Fig. 4b the temperature dependence of the resistivity, expressed as M′′/ω, is shown for two frequencies, viz. ν = 100 kHz and ν = 1 MHz. Asymmetric relaxation peaks are obtained with the high-T flanks yielding 0.56 eV in very good agreement with σdcT(1/T). Importantly, the low-T flank is characterized by a much smaller value of approximately 0.2 eV. According to Ngai's coupling concept of ionic transport,75–77 such values can be identified as the activation barriers characterising short-range ion hopping processes. By using 19F spin–lattice relaxation NMR we extracted very similar values, see below. Comparing results from conductivity spectroscopy with those from NMR will help us to describe the shape of the underlying motional correlation function that governs the electrical and nuclear magnetic responses of 2D dynamics in RbSn2F5.
Variable-temperature 19F NMR spectra recorded under static conditions also reveal this coalescence, see Fig. 7. In the rigid-lattice regime, which is reached at sufficiently low T, we deal with a dipolarly broadened signal composed of several contributions as expected from MAS NMR. With increasing T dipole–dipole interactions are averaged; in the extreme narrowing regime, i.e., above 313 K, a single line governs the spectra, once again showing that nearly all F anions participate in fast self-diffusion processes. Only the very shallow signal at −75 ppm reveals some F anions that do not contribute to the coalesced signal, see the arrow in Fig. 7.
Comparing the spectra for microcrystalline RbSn2F5 with those of nanocrystalline RbSn2F5 we see that heterogeneous motional narrowing starts at slighter lower T for the nanocrystalline sample, see Fig. 7. While the ionic conductivity of microcrystalline and nanocrystalline RbSn2F5 reveals no significant difference, in defect-rich ball-milled RbSn2F5 (local) structural disorder slightly increases ionic motion – at least on the length scale to which NMR line shapes are sensitive. Obviously, the defect structure of nano-RbSn2F5 is not of detrimental nature for long-range ion transport. Ions at the grain boundary or interfacial regions might benefit from such disorder, while those having access to long-range 2D pathways along the inner surfaces already participate in fast exchange processes. However, in contrast to other systems the enhancement seen for nanocrystalline RbSn2F5 turned out to be marginal. For comparison, for poorly conducting oxides with 3D pathways, the introduction of defects usually leads to an enhancement in ion dynamics by several orders of magnitude.78–80 As an example, this behaviour has also been observed for nanocrystalline LiNbO3 and LiTaO3 prepared by high-energy ball milling.78,79
Diffusion-induced 19F NMR spin–lattice relaxation rates R1(ρ) for both annealed and as-prepared RbSn2F5 are shown in Fig. 8. Whereas spin–lattice relaxation in the laboratory frame (R1), i.e., being measured at Larmor frequencies in the MHz range, is sensitive to short-range F motions on the angstrom scale, with the rates recorded at locking frequencies in the kHz range (R1ρ) we sense long-range ion transport. Below 220 K the rates R1 are dominated by non-diffusion induced processes. 19F spins couple to phonons or paramagnetic impurities. Above 220 K spin–lattice relaxation gets increasingly controlled by F motional processes. The flanks seen in Fig. 8a result in activation energies of ca. 0.25 eV. This value agrees well with those seen on the low-T flank of the M′′/ω-curves shown in Fig. 4b. As RbSn2F5 reversibly transforms into a different crystal structure at 368 K, the R1 rates show deviations from Arrhenius behaviour at this temperature. Spin–lattice relaxation in nanocrystalline (ball-milled) RbSn2F5 is faster by approximately 1 order of magnitude if we compare R1 rates measured at 330 K. Obviously, local F dynamics is somewhat faster in the non-annealed form; this increase in local diffusivity, however, does not affect long-range ion transport in RbSn2F5 as σdc is the same for the two samples.
The behaviour of σdc is also seen in R1ρ. Obviously, R1ρ measurements are able to detect the same long-range ion dynamics as is sensed by σdc. The R1ρ rates of annealed and non-annealed RbSn2F5 almost coincide and follow the expected behaviour for a diffusion-induced relaxation processes when R1ρ is plotted vs. 1/T. In general, R1ρ (and R1) will pass through a diffusion-induced rate peak whose maximum shows up when the jump rate τ−1 reaches the order of the locking (or Larmor) frequency: ω1(0)τ−1. While in the limit ω1τ ≫ 1 the flank of the R1ρ(1/T) peak is influenced by correlation effects and structural disorder, the flank in the regime ω1τ ≪ 1 should sense dynamic parameters being comparable to long-range ion dynamics. Here, the low-T flanks yield 0.29 eV and 0.24 eV, respectively. Once again, these values agree well with those obtained from R1 measurements and the analysis of M′′ω. The high-T flank, however, results in 0.39 eV for nano-RbSn2F5 and 0.41 eV for micro-RbSn2F5. These activation energies are clearly smaller than those seen by σdc. Obviously, simply analyzing the high-T flanks of the 19F NMR relaxation peaks, which does not take into account any effects from low-dimensional diffusion, yields activation energies being inconsistent with that seen by conductivity spectroscopy. Ea extracted from the slope of high-T flanks is only of use when 3D diffusion processes are to be analyzed.
If we take into account 2D diffusion and use the semi-empirical relaxation model introduced by Richards49,62 to analyze the R1ρ(1/T) NMR peaks we obtain a different result. The solid lines in Fig. 8a show fits with spectral density functions J(ω1)2D ∝ R1ρ that include a logarithmic frequency dependence of the R1ρ rates in the limit ω1τ ≪ 1. This dependence is characteristic for 2D diffusion. J(ω1)2D, for uncorrelated motion, results in an asymmetric rate peak with the high-T slope being smaller than that in the low-T regime. For correlated motion, instead, the low-T flank is lower than that expected for 3D diffusion. Correlation effects are taken into account by the factor β in the following expression for J(ω1)2D
J(ω1)2D ∝ τcln(1 + 1/(2ω1τc)β) ∝ R1ρ | (3) |
J(ω1)3D ∝ τc/(1 + (2ω1τc)β) | (4) |
It is valid for homonuclear 19F–19F spin fluctuations; here we restricted ourselves to a single term for a good approximation of the rates measured. The limiting cases of eqn (3) for the high- and low-T slopes of the corresponding R1ρ(1/T) relaxation peak are the following ones:82
J(ω1)2D ∝ τcln(1/(2ω1τc)), for ω1τc ≪ 1 | (5) |
J(ω1)2D ∝ τc1−β(2ω1)−β, for ω1τc ≫ 1 | (6) |
For β = 2, eqn (5) is identical with the result for 3D uncorrelated motion. τc−1 is expected to be identical with the jump rate τ−1 within a factor of 2–3. τ−1 is assumed to be thermally activated according to an Arrhenius law:
τ−1 = τ0−1exp(−Ea/(kBT)) | (7) |
Using J(ω1)2D, see eqn (3), to analyse the rate peaks of Fig. 8, yields fits that properly agree with the temperature behaviour of the rates measured, see the solid lines in Fig. 8. The activation energies for micro- and nanocrystalline RbSn2F5 turn out to be 0.52 eV and 0.48 eV, respectively. Especially the value for micro-RbSn2F5 perfectly agrees with that from conductivity measurements (0.52 eV). In contrast, eqn (4), i.e., using a 3D diffusion model, yields 0.39 eV (almost the same value is obtained if only the high-T slope is analyzed, as mentioned above). Although the quality of the two types of fits are very similar, only the spectral density of Richards developed for 2D diffusion is in agreement with the expectation that Ea from this model equals that deduced from σdc.
Here, as mentioned above, we have restricted J(ω1)iD (with i = 2, 3) to a single term. Including higher-order terms does not change the quality of the fits and has no effect on Ea and τ0−1. The pre-factors τ0−1 of the fits shown in Fig. 8b are in the order of 2 × 1015 s−1. Such a value is relatively high but in agreement with similar observations for surface diffusion.83 Obviously, F diffusion along the buried or inner planes84 of RbSn2F5 resembles ion dynamics on surfaces.
In both cases, i.e., for nanocrystalline and microcrystalline RbSn2F5, the correlation factor β turned out to be 1.51 (nano) and 1.57 (micro), respectively. β directly influences the slope on the low-T side, the smaller value of 1.45 leads to a slightly smaller slope for nano-RbSn2F5 as seen in Fig. 8b. Considering the frequency limits of J(ω1)3D (R1 ∝ τc and R1 ∝ τc1−βω0−β) we see that the activation energies on the low-T and high-T side are connected to each other via
Elow-Ta = (β − 1)Ehigh-Ta | (8) |
For 2D ionic motion essentially the same equation holds true, we have, however, to replace Ea(high-T) with the activation energy from the fit with J(ω1)2D.29 If we simply take the slope of the R1(ρ)(1/T) peak we must consider that E2Da is related to E3Da = Ea(high-T) via E3Da = 3/4E2Da;49 with the values obtained here we have 0.39 eV = 3/4 × 0.52 eV. With β = 1.57 the relationship E2Da(low-T) = (β − 1)E2Da is also fulfilled as we indeed have found 0.29 eV ≈ 0.57 × 0.52 eV. The same also holds for nanocrystalline RbSn2F5 with β − 1 ≈ 0.5 and E2Da(low-T) = 1/2E2Da.
In summary, while long-range ion transport in microcrystalline RbSn2F5 is characterized by 0.52 eV, short-range motions, taking into account correlation effects such as strictly confined hopping processes, local disorder and Coulomb interactions are best described by an activation energy of 0.29 eV. For nanocrystalline RbSn2F5 this value is somewhat reduced to 0.24 eV; this slight change is also reflected in 19F NMR line narrowing, see above. This view on ion transport is also in agreement with the model introduced by Ngai. Independent of any dimensionality effects in Ngai's coupling concept dispersive regions in σ′(ν) would be expected that follow a νn=β−1 behaviour, which is fulfilled in the present case.
To further compare results from NMR with those from conductivity spectroscopy, we can use the rate R1ρ(1/T) peak to estimate a self-diffusion coefficient for F− hopping in RbSn2F5. At the peak maximum, which is seen at 273 K, the condition ω1τc ≈ 0.5 is valid; here, we have 1/τc = 2.5 × 106 s−1. Using the Einstein–Smoluchowski equation85,86D = a2/(2dτ) with d = 2 for 2D diffusion we obtain D = 2.8 × 10−14 m2 s−1 if we insert a mean jump distance of a ≈ 3 Å. The jump distances of the F anions labelled F2, F3 and F4 (see Fig. 1) range from 2.89 to 2.97 Å. The distance from F1 to F2 is 2.95 Å. For comparison, at 273 K the bulk ionic conductivity is in the order of 1.4 × 10−5 S cm−1. This value translates, according to the Nernst–Einstein equation87,88 using a charge carrier density of 3 × 1027 m3, into a solid-state diffusion coefficient Dσ of 6.7 × 10−14 m2 s−1. Thus, as the two diffusion coefficients only differ by a factor of 2, 19F NMR spin–lattice relaxation, if the maximum of the rate peak is used to estimate D, and conductivity spectroscopy sense the same motional process in layer-structured RbSn2F5.
Finally, we shall discuss the change in ion dynamics when going from microcrystalline RbSn2F5 to its structurally disordered, nanocrystalline form. Many studies, dedicated to comparing diffusion properties of nanocrystalline ceramics with those from coarse-grained materials of the same chemical composition, report on huge differences at least if poorly conducting oxides or fluorides are considered, see above.17,59,78–80 For materials with rapid ion dynamics from the start, the influence of structural disorder or size effects on ion transport is expected to be moderate or marginal if not without any positive effect on ion transport. If we consider layered materials structural disorder might even slow down ionic transport rather than increase it. For nanocrystalline LixTiS2 (0 < x ≤ 1), crystallizing with a layered structure, Winter and Heitjans did not found any large increase in Li diffusivity when compared to its coarse-grained form.89,90 Li diffusion in the TiS2 host is governed by very similar activation energies for the two forms. The incorporation of Rb for Ba in the fast 2D ionic conductor BaSnF4 increases ionic conductivity;59 most likely aliovalent doping affects, however, the slower interlayer exchange process rather than fast intralayer diffusion.59,91 For the layer-structured form of LiBH4,27,92 crystallizing with hexagonal symmetry and showing 2D diffusion, downsizing the mean crystallite diameter through ball-milling has only small effect on Li ion conductivity.92 On the other hand, the poorly conducting orthorhombic form of LiBH4 with 3D diffusion, clearly benefits from nanostructuring. In such a sample, NMR relaxation measurements clearly revealed a subset of rapidly diffusing Li ions.93
Here, the nanocrystalline form of RbSn2F5 does not reveal strongly enhanced ion dynamics as compared to the annealed material with sharpened X-ray reflections. Conductivity spectroscopy clearly shows that ionic conductivities are the same for the two morphologies. On the contrary, one might even expect a slight decrease if defects disturb the fast 2D diffusion pathways that are already equipped with a large number fraction of vacant F anion sites. While we cannot recognize any strong effect on long-range ion transport from conductivity measurements, 19F NMR line shapes point to slightly faster ion dynamics in the nanocrystalline form as seen by the motionally averaged line emerging at lower temperatures for non-annealed RbSn2F5. Presumably, disordered grain-boundary structures act as hosts for these mobile ions being able to contribute to dipole–dipole averaging already at slightly lower temperatures. This observation is consistent with enhanced R1 relaxation rates of nano-RbSn2F5 if compared to those of microcrystalline RbSn2F5. Moreover, the corresponding activation energy on the low-T side of the spin-lock NMR peak is reduced from 0.29 eV to 0.24 eV.
Although the effects are small, on the short-range or angstrom length scale we see enhanced dynamics for the as-prepared form. Also magnetic field fluctuations of spatially confined jump processes might contribute to faster relaxation processes communicated to other 19F spins via spin-diffusion. At least, the F anions residing near the defect sites do not become trapped. Such trapping effects could easily explain any decrease in overall ion transport of layer-structured materials.
In other cases, fast localized dynamics are able to trigger through-going macroscopic transport.25 These dynamic processes might be involved in also causing the huge increase in conductivity seen for the aforementioned nanocrystalline oxides, fluorides and carbonates prepared by milling.91,94 Prominent examples include, for example, LiXO3 (X = Ta, Nb),78,79 LiAlO2,80 BaF2,95 CaF217,96,97 and Li2CO3,94 as mentioned above. Here, the subtle changes when going from the micro scale to nm-sized crystallites are, however, hardly comparable with the boosts in 3D ionic conductivity seen for the disordered systems. To sum up, for the fast 2D ionic conductor RbSn2F5 site disorder is of limited help to further enhance the already rapid long-range ion transport along the internal or buried interfaces.
Careful analysis of both the conductivity isotherms and variable-temperature Nyquist curves, which are composed of bulk and grain boundary regions, revealed indications that F dynamics in RbSn2F5 is of 2D nature. Importantly, only if analyzed with a spectral density formalism which was developed for 2D diffusion the 19F NMR data are consistent with dynamic parameters from conductivity spectroscopy. NMR relaxation measurements point to highly correlated F− exchange processes for both the nanocrystalline and microcrystalline form. Local jump processes in micro-RbSn2F5 are to be characterized by an average activation energy as low as 0.24 eV; an even lower value of only 0.16 eV can be inferred from dielectric spectroscopy. An activation energy of 0.52 eV, on the other hand, determines long-range ion transport.
By comparing our results on microcrystalline and nanocrystalline RbSn2F5 we see that structural site disorder does not lead to any drastic enhancement of ion dynamics as it has been recognized for other ionic conductors. Obviously, for layer-structured materials, as it is also known for LiCoO2 used as cathodes in lithium-ion batteries, the structurally perfect layers guiding the ions along the internal interfaces. This guidance is crucial to ensure long-range ion transport being fast enough that additional disorder does not lead to any further enhancement. Quite the contrary, one might also expect the opposite behaviour. Against this background, it seems worth considering the effect of dimensionality as a design principle for fast ion conductors.
Footnotes |
† Electronic supplementary information (ESI) available: Further NMR spectra and conductivity data. See DOI: 10.1039/c8cp07206j |
‡ Present address: Schunk Carbon Technology GmbH, Au 62, 4822 Bad Goisern, Austria. |
This journal is © the Owner Societies 2019 |