The Structure of Alkali Silicate Glasses and Melts: a Multi-Spectroscopic Approach

Home , Lithium disilicate, Lithium metasilicate, Lithium peroxide

The structure of alkali silicate glasses and melts: A multi-spectroscopic approach

Cedrick A. O’Shaughnessy

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy in Geology Graduate Department of Earth Sciences University of Toronto

The structure of alkali silicate glasses and melts: A multi-spectroscopic approach

Cedrick A. O’Shaughnessy Doctor of Philosophy in Geology Graduate Department of Earth Sciences University of Toronto 2019

The structure of alkali silicate glasses and melts is investigated using a multi-spectroscopic approach.

Raman spectroscopy is used to characterize the local to intermediate range order within the glasses. We show that the distribution of rings varies as a function of composition, with 3-membered rings gaining importance with increasing alkali content. We apply a newly developed model for the ﬁtting of the high

n frequency envelope related to SiO4 symmetric stretch vibrations of Q species. These fits are interpreted using the idea of modifier bound bridging oxygen. The proportions of the different Qn species vary with alkali concentration with Q4 species breaking down to form lower order Qn species with increasing alkali

2 content. The Q peak appears at increasingly higher concentrations of M2O with increasing cation size. This leads us to believe that cations with a higher charge density cluster more readily than cations with a lower charge density. At the ∼20 mol. % composition we see a change in the silicate network, as shown by the absence of a Q4 peak and the proportion of 3-membered rings.

The unique behaviour of lithium was investigated using X-ray absorption near-edge structure spectroscopy. We conducted experiments on a variety of lithium-bearing salts, minerals and glasses (LS: lithium silicate and LMS: lithium alkaline-earth silicate glasses) in order to characterize the lithium bonding environment. The Li K -edge position depends on the electronegativity of the element to which it is bound. The intensity of the ﬁrst peak varies, depending on the existence of a 2p electron and can be used to evaluate the degree of ionicity of the bond. Crystalline lithium metasilicate has a sharp, strong intensity absorption edge whereas the lithium silicate glasses all have a weak intensity edge feature, similar to that of lithium carbonate. The area of the absorption edge peak increases with the lithium content of the LS glasses. The LMS glasses edge peak changes drastically depending on the alkaline- earth present. The peak area of the LMS glasses decreases with increasing charge density from barium to magnesium. We believe that the presence of Mg leads to more covalent-like Li−O bonds.

ii To my loving wife and best friend, Joanna.

iii Acknowledgements

This thesis would not have been possible without the generosity of the researchers and laboratory facilities where I carried out this research. Daniel Neuville from the Geomaterials laboratory of the Institut de Physique du Globe de Paris who hosted me for months during the summer time. Wayne Nesbitt and Michael Bancroft from the University of Western Ontario were so helpful and generous of their time in which we had great discussions. Lucia Zuin and Tom Reiger were fantastic help on all our visits to the Canadian Light Source.

My fellow graduate students, Ben, Heidi, Alex, Simon, and many others I thank you for all the help and friendship. From the oﬃce to the GSU, it really was a great time!

A huge thank you to my family, Mom & Dad, you’ve always been there to guide me and more importantly, believe in me. Jola i Mariusz, dziekuje bardzo!

Joanna, countless times you have been there to help me get through this. I cannot express how grateful I am to you for your never ending support. I love you so very much! We did it!

Finally, to Grant Henderson, who gave me all of these opportunities, and pushed me to keep plugging away. From your gentle encouragements to your more forceful ones, I truly appreciate all the help and advice you’ve given me. Thanks a lot Mr. Sunshine!

Sincerely, thank you all.

iv Contents

1 Introduction 1 1.1 Structural models of silicate glasses...... 1 1.1.1 Continuous Random Network (CRN) model...... 2 1.1.2 Modiﬁed Random Network (MRN) Model...... 3 1.1.3 Central-force network models...... 3 1.1.4 Phase separation...... 4 1.2 Raman spectroscopy...... 5 1.2.1 The Raman spectra of silicate glasses...... 5 1.3 X-ray absorption spectroscopy...... 8 1.3.1 XANES of silicate glasses...... 9 1.4 Contribution of this thesis...... 10

2 Structure-property relations of caesium silicate glasses from room temperature to 1400 K: implications from density and Raman spectroscopy 11 2.1 Introduction...... 12 2.2 Methods...... 14 2.2.1 Glass synthesis...... 14 2.2.2 Raman spectroscopy...... 15 2.3 Results...... 16 2.3.1 Room temperature Raman spectroscopy...... 16 2.3.2 High-temperature Raman spectroscopy...... 18 2.4 Discussion...... 19 2.4.1 The Boson peak...... 19 2.4.2 The D1 and D2 bands: 495 and 606 cm−1 ...... 20 2.4.3 The C peak (770 cm−1)...... 22 2.4.4 The high frequency region (850-1250 cm−1)...... 24 2.4.5 Insights from high-temperature Raman spectra...... 29 2.5 Conclusions...... 32

3 The structure of high-silica alkali silicate glasses: Revisited 34 3.1 Introduction...... 35 3.2 Methods...... 41 3.2.1 Glass synthesis...... 42 3.2.2 Raman spectroscopy...... 42

v Experimental measurements...... 42 The ﬁtting of Raman spectra...... 42 3.3 Results...... 44 3.3.1 Density and molar volume of alkali silicate glasses...... 44 3.3.2 Raman spectra of alkali silicate glasses...... 45 The behaviour of the D1 and D2 bands (495 and 600 cm−1)...... 45 The symmetric stretch region (Qn species: 850-1300 cm−1)...... 45 3.4 Discussion...... 52 3.4.1 The D1 and D2 bands (495 and 600 cm−1)...... 52 The relative intensity of the D1-D2 bands in alkali silicate glasses...... 52 The splitting of the D1-D2 bands in alkali silicate glasses...... 54 Implications from Percolation theory...... 54 3.4.2 The high frequency peak (800-1300 cm−1)...... 61 The polarizability of Q4 species...... 61 The asymmetry of the Q3 band...... 61 The appearance and increase of Q2 species: implications for the silicate glass network 62 3.5 Conclusions...... 63 3.5.1 Defect bands and their behaviour...... 63 3.5.2 Q4 band properties...... 63 3.5.3 The Q3 band asymmetry: M−BO interactions & Qn,ijkl clustering...... 63 3.5.4 Q2 species and the clustering of alkalis in silicate glasses...... 63 3.5.5 Implications for the structural model of alkali silicate glasses...... 64 3.6 Acknowledgments...... 64

4 A Li K -edge XANES study of salts and minerals 76 4.1 Introduction...... 77 4.2 Methods...... 78 4.2.1 Mineral samples...... 78 4.2.2 X-ray Absorption Near-Edge Structure (XANES) Spectroscopy...... 78 4.3 Results...... 80 4.3.1 Lithium salts and metasilicate...... 81 Lithium chloride (LiCl)...... 81

Lithium sulphate (Li2SO4)...... 81

Lithium metasilicate (Li2SiO3)...... 82

Lithium carbonate (Li2CO3)...... 82 4.3.2 Lithium aluminosilicates minerals (LAS)...... 83

Eucryptite (LiAlSiO4)...... 83

Spodumene (LiAlSi2O6)...... 83

Petalite (LiAlSi4O10)...... 85

4.3.3 Lithium phosphates (PO4)...... 85

Montebrasite (LiAl(PO4)(OH))...... 85 2+ Lithiophilite (LiMn PO4)...... 85 4.3.4 Complex lithium-bearing silicates: phyllosilicates and cyclosilicates...... 86

Elbaite (Na(Li,Al)3Al6(Si6O18)(BO3)(OH)3(OH))...... 87

vi Lepidolite (KLi2Al · (Si4O10)(F,OH)2)...... 88 2+ Zinnwaldite (KLiFe Al(AlSi3O10)(F,OH)2)...... 88 2+ 2+ Neptunite (LiNa2K(Fe ,Mn )2Ti2(Si8O24))...... 88 4.4 Discussion...... 89 4.4.1 Lithium salts and metasilicate...... 89 4.4.2 Bond length distortion in lithium aluminosilicate minerals...... 90 4.4.3 Complex silicates...... 91 4.4.4 Phosphates...... 92 4.5 Conclusions...... 92 4.6 Acknowledgements...... 92

5 The eﬀect of alkaline-earth substitution on the Li K -edge of lithium silicate glasses 94 5.1 Introduction...... 95 5.2 Methods...... 96 5.2.1 Glass synthesis...... 96 5.2.2 X-ray Absorption Near-Edge Structure (XANES) Spectroscopy...... 96 5.3 Results...... 98 5.3.1 Lithium silicate glasses (LS)...... 98 5.3.2 Lithium alkaline-earth silicate glasses (LMS)...... 98 5.4 Discussion...... 99 5.5 Conclusions...... 101 5.6 Acknowledgments...... 102

6 Conclusions 103 6.1 Highlights of the Raman spectroscopy investigation on alkali silicate glasses...... 103 6.2 Highlights of the Li K -edge XANES investigation on lithium silicate minerals and glasses 104 6.3 Implications for the structural model of alkali silicate glasses...... 105

References 106

vii List of Tables

−1 −1 2.1 The Raman shift, ωc (± 1 cm ) and FWHM (± 2 cm ) values of the various peaks in the Cs-silicate glasses spectra...... 16 −1 2.2 The Raman shift, ωc (± 2cm ) for all the curve-ﬁt peaks...... 27 −1 2.3 The Raman shift, ωc (± 4cm ) for all the curve-ﬁt peaks of the high temperature spectra of Cs5, Cs10 and Cs15...... 32

3.1 The maximum positions for the D1, D2 and Qn peaks of the alkali silicate glasses..... 51 3.2 The average density of alkali silicate glasses as calculated from 10 independent measurements per sample. The error on the measurements is equal to ± 0.005 g/cm3...... 65 3.3 The fitting results for the lithium silicate glasses...... 67 3.4 The fitting results for the sodium silicate glasses...... 69 3.5 The fitting results for the potassium silicate glasses...... 71 3.6 The fitting results for the rubidium silicate glasses...... 73 3.7 The fitting results for the caesium silicate glasses...... 75

4.1 Symmetry, coordination number (C.N.) & the mean bond length (hLi−Xi; X=O, S)... 79 4.2 Peak positions in lithium-bearing salts and minerals...... 81

5.1 Electron microprobe analysis of lithium-bearing glasses...... 96 5.2 Peak positions in lithium-bearing compounds and minerals...... 97

viii List of Figures

1.1 A modiﬁed random network (MRN) for a 2-dimensional oxide glass with composition

Na2Si4O9 (after Greaves(1985) and adapted from Nesbitt et al.(2015)). Na −BO and Na−BO−Na were added by Nesbitt et al.(2015) and the percolation channels running through the covalent network are highlighted in white...... 3

1.2 Raman spectra of SiO2 glass with both the parallel and perpendicular components (after Matson et al.(1983))...... 5 1.3 (a) Raman spectra of glass and crystalline Cs disilicates (after Matson et al.(1983)). (b) Raman spectrum from this study with higher resolution down to very low wavenumber..7 n 1.4 (a) High temperature Raman spectra of crystalline Na2Si205 (b) Curve ﬁtting for Q

species distribution of Na2Si205 melt at 1473 K. Each deconvolved peak represents a speciﬁc Qn species unit (after You et al.(2001))...... 8 1.5 Mg K -edge spectra of haplobasaltic glasses and standards. A) Stacked intensity normalized spectra. B) Area normalized haplobasaltic spectra overlain to compare the subtle changes in edge intensities. The reference dashed line is at 1310 eV in both A and B. C) Increase in intensity of peaks A and B with haplobasaltic composition for area normalized spectra in B. Adapted from Moulton et al.(2016)...... 9

2.1 Example of the Gaussian curve-ﬁtting of Cs10 with the residual. Area normalized spectra are ﬁt using a minimum number of bands (5) with similar FWHM. Due to the area normalization the areas of the Gaussian components can be used to estimate the proportion of one band to the other. Observing the change in the relative areas of the bands with

increasing Cs2O composition provide a way to estimate the proportions of the diﬀerent chemical species present in the glasses...... 15 2.2 Raman spectra of caesium silicate glasses. The numbers on the right side of the spectra

represent the value of x for x · Cs2O−(1−x) · SiO2 (mol %). The spectra are vertically shifted in order to distinguish the different features present...... 17 2.3 High-temperature spectrum of Cs10 (640 ◦C, 800 ◦C, 950 ◦C, 1057 ◦C & 1140 ◦C). The spectra are vertically shifted in order to distinguish the different features present...... 18 2.4 The peak position (a) and FWHM (b) of the Boson peak as a function of composition... 19 2.5 (a) The measured density of caesium silicate glasses (our study: empty triangles, Tischen- dorf et al.(1998); Doweidar et al.(1999): filled triangles.) plotted with the calculated molar volumes (empty squares). (b) Ratio of the intensities of the D1 and D2 Raman

bands (black circles). (c) The D1-D2 peak splitting, ∆D...... 21

ix 2.6 (a) Peak centre of peak C as a function of composition. (b) Peak C intensity as a function of composition. The intensities of all spectra are normalized to the intensity of the C peak in sample Cs5...... 23 2.7 The barycentre of the high frequency envelope of the binary caesium silicate glasses as a function of the composition. Note a negative shift of more than 35 cm−1...... 25 2.8 The Gaussian curve-ﬁtting of a) Cs15 and b) Cs30. Note the change in the high frequency side of the envelope combined with the development of a new band at 927 cm−1...... 26 2.9 The relative areas of the Gaussian bands used to described the diﬀerent populations of Qn species present in the glasses. The peaks for Qn and QnM species are grouped together

in order to see the total change in the species with increasing Cs2O content...... 28 2.10 The barycentre calculated for the high frequency envelopes of the caesium silicate glasses and melts (Cs5, Cs10 & Cs15) at multiple temperatures. The same trends in the barycentre position are observed for both the composition and temperature...... 29 2.11 Gaussian curve-ﬁtting and band assignments at diﬀerent temperatures for Cs10 at a) 640 ◦C and b) 1140 ◦C...... 30 2.12 Area of the Qn species as a function of temperature and composition...... 31

3.1 The spectra of v-SiO2 and Na5. They are shown here in order to highlight the differences which occur when a small amount of alkali cations are added to the silicate glass network. Additionally, the main regions of interest are labeled and further discussed in the text... 36 3.2 (a) Illustrates the relation between the alkali oxide content of some crystals, glasses and melts, and the wavenumber of each Qn species observed in these phases as reported by McMillan(1984). The solid and dashed lines represent respectively the least squares best fits to the crystal and to the glasses/melt data. (b) The frequencies of the Qn species reported by McMillan(1984). The error bars indicate ± 25 cm−1 and are taken as the approximate uncertainty associated with each frequency based on the data of (a) and comments by (McMillan, 1984). The straight line is the linear least squares best fit to the data. This figure is adapted from (Nesbitt et al., 2017b)...... 38 3.3 (a) The measured values for the densities of the alkali silicate glasses. (b) A closer look at the densities of the lithium, sodium and potassium silicate glasses. It is important to

observe the apparent density cross over which occurs between 20 and 25 mol. % M2O for the sodium and potassium silicate glasses. Note that the errors were calculated by using ten independent density measurements for each sample and these errors lie within the symbols...... 41 3.4 The variation in pseudo-Voigt lineshapes at a constant Γ=2 (FWHM) and incremental mixing factor η ranging from 0 (Gaussian) to 1 (Lorentzian). We highlight the diﬀerence in the tails of the distributions and discuss these implications in the text...... 44 3.5 The Raman spectra of the lithium silicate glasses...... 46 3.6 The Raman spectra of the sodium silicate glasses...... 47 3.7 The Raman spectra of the potassium silicate glasses...... 48 3.8 The Raman spectra of the rubidium silicate glasses...... 49 3.9 The Raman spectra of the caesium silicate glasses...... 50 3.10 (a) The calculated molar volume of the alkali silicate glasses. (b) Ratio of the intensities

of the D1 and D2 Raman bands (c) The D1-D2 peak splitting, ∆D...... 53

x 3.11 The relative areas of the different Qn species for the lithium silicate glasses as a function of alkali content. (a) The areas for the Q4 bands. (b) The areas for the Q3 bands. (c) The area of the Q2 band...... 55 3.12 The relative areas of the different Qn species for the sodium silicate glasses as a function of alkali content. (a) The areas for the Q4 bands. (b) The areas for the Q3 bands. (c) The area of the Q2 band...... 56 3.13 The relative areas of the different Qn species for the potassium silicate glasses as a function of alkali content. (a) The areas for the Q4 bands. (b) The areas for the Q3 bands. (c) The area of the Q2 band...... 57 3.14 The relative areas of the different Qn species for the rubidium silicate glasses as a function of alkali content. (a) The areas for the Q4 bands. (b) The areas for the Q3 bands. (c) The area of the Q2 band...... 58 3.15 The relative areas of the different Qn species for the caesium silicate glasses as a function of alkali content. (a) The areas for the Q4 bands. (b) The areas for the Q3 bands. (c) The area of the Q2 band...... 59 3.16 The relative areas of the different Qn species for the alkali silicate glasses as a function of alkali content. (a) The toal area for the Q4 bands. (b) The total area for the Q3 bands. (c) The area of the Q2 bands...... 60 3.17 The curve-fitting for the lithium silicate glasses (peak assignments listed in Table 3.3)... 66 3.18 The curve-fitting for the sodium silicate glasses (peak assignments listed in Table 3.4)... 68 3.19 The curve-fitting for the potassium silicate glasses (peak assignments listed in Table 3.5). 70 3.20 The curve-fitting for the rubidium silicate glasses (peak assignments listed in Table 3.6).. 72 3.21 The curve-fitting for the caesium silicate glasses (peak assignments listed in Table 3.7)... 74

4.1 The Li K -edge spectra of our sample suite. The standards black, the L-A-S minerals

(blue) with increasing SiO2 content, the phosphate minerals (pink), tourmaline (green) and the phyllosilicates (orange) are intensity normalized. The dotted lines are located at 61 eV and 65 eV and serve as a guide for the eyes when comparing the diﬀerent spectra

(same locations for all ﬁgures). Note that the proximity of the Fe M2,3- and Mn M2,3- to the Li K -edge creates overlapping edge contributions and these complications will be discussed in the following sections...... 80 4.2 The Li K -edge spectra of standard compounds used to calibrate and interpret the results

for the mineral suite. Note the sharp absorption edge in LiCl, Li2SO4 and Li2SiO3 and

how it contrasts with the lower energy pre-edge (peak p) observed in Li2CO3...... 82 4.3 The Li K -edge spectra of lithium aluminosilicate minerals. Eucryptite appears to be composed of three diﬀerent peaks (B, C, and D). These three peaks most certainly exist in all three samples but simply overlap too much in spodumene and petalite (peak C) though a second peak can be seen in spodumene (peak C and D)...... 84 4.4 The Li K -edge spectra of phosphate minerals. Montebrasite demonstrates a well-deﬁned, sharp absorption edge as was the case with our standards as well as petalite. Lithiophilite

conversely is diﬃcult to interpret due to the overlap of the Mn M 2,3-edge. Nevertheless, several features from lithiophilite have similar energies to those observed in montebrasite. 86

xi 4.5 The Li K -edge spectra of a Li-bearing tourmaline and three phyllosilicates. Elbaite displays a main peak (D) with two lower intensity pre-edge features. The main peak in

lepidolite (D) resembles that of elbaite but is broader, they both are similar to Li2CO3. Zinnwaldite has an Fe M 2,3-edge contribution which renders the low energy part of the spectrum diﬃcult to interpret but has a main peak (D) position similar to that of lepi-

dolite and elbaite. Neptunite has both an Fe M 2,3-edge and a Mn M 2,3-edge overlapping the low energy portion but also has a main peak (D) located at a similar position as the aforementioned minerals...... 87 4.6 (a) The maximum peak position is the location of the maximal peak intensity within the absorption edge triplet, peak c (observed in the LAS minerals). The minerals are plotted

against the SiO2 content and are labeled above. Additionally, the coordination of lithium in the respective mineral is listed as it is important to the calculation of the bond-length distortion (BLD). (b) The BLD was determined using the formula described in Wenger and Armbruster(1991) and their values for the mean bond lengths of 4- and 6-coordinated lithium minerals. The values for the individual bond lengths were obtained from Daniels and Fyfe(2001) (eucryptite), Clark et al.(1968) (spodumene) and Ross et al.(2015) (petalite)...... 90

5.1 Graphical depiction of the area calculation of the main peak of the lithium glasses using two hinge-points and a trapezoidal method...... 98

5.2 The Li K -edge spectra of standard compounds (Li2CO3 and Li2SiO3) and three lithium silicate glasses (LS1.5, LS2 and LS4)...... 99

5.3 The Li K -edge spectra of standard compounds (Li2CO3 and Li2SiO3) and four lithium alkaline-earth silicate glasses (LiMg, LiCa, LiSr and LiBa). The spectrum of LS1.5 is also

reported for comparison since it also contains 60 mol. % SiO2...... 100 5.4 a) Normalized area of peak A of the lithium silicate glasses as a function of Li content. b) Normalized area of peak A of the lithium alkaline-earth silicate glasses as a function of their charge density (Z/r2) as calculated using the atomic radii of Shannon and Prewitt (1969)...... 101

xii Chapter 1

Introduction

Melting and crystallization are the two most important phase transitions in the differentiation of Earth and many other astronomical bodies. As they dictate the concentric structure of the Earth with a core, mantle and a crust. The existence of SiO2-rich buoyant continents and, therefore, the subsequent emer- gence of life are processes that were all ultimately determined by melting and crystallization reactions (Nesbitt et al., 2017a). Melting occurs in many different geological settings including; (i) mid-oceanic ridges by decompression; (ii) hot spots by localized upwelling of mantle plumes, and; (iii) subduction zones by fluxing of fluids. Historically, spectroscopic experiments to investigate the properties of molten silicates were conducted on their quenched (glassy) products due to the difficulty of in-situ high temperature experiments (cf., Warren, 1934; Warren and Biscce, 1938; Warren and Biscob, 1938; Greaves et al., 1981; Matson et al., 1983; Dupree et al., 1986; Henderson et al., 2006). In this introduction I will discuss the scientific relevance of silicate glasses and their respective liquids, as well as, our current state of understanding on the topic. I will also outline the contributions to the thesis and describe the context for each of the chapters. A multi-spectroscopic approach is used to characterize the atomic network present in simple glasses at both room and high temperature. The sample suite is an exhaustive compilation of alkali silicate glasses with composition of xM2O-(1-x)SiO2; M=(Li, Na, K, Rb and Cs). The proportion of alkalis ranges from 5 to 30 mol % in increments of 5 mol %. The ionic radius dramatically changes throughout the group. The range of radii is also different from one cation to the other due to their coordination within a given material: Li1+=0.59-0.74; Na1+=0.99-1.32; K1+=1.38-1.60; Rb1+=1.49-1.73; Cs1+=0.70-1.88 (Shannon and Prewitt, 1969, and references therein). We will also pay specific attention to lithium by probing the Li K -edge XANES due to the interesting results from Raman spectroscopy (see Chapter3).

1.1 Structural models of silicate glasses

Our understanding of high temperature liquids predominantly comes from using glasses as analogs. If a liquid does not crystallize as it is cooled below its liquidus temperature, its viscosity increases and eventually reaches a point where ﬂow no longer occurs. The liquid thus transitions to glass; a non- periodic, disordered solid (Mysen and Richet, 2005; Henderson et al., 2006). Certain liquids have a higher propensity to form glasses than others, for instance, SiO2 is a very good glass former (Angell et al., 2000). There are three main factors which deﬁne how easily a liquid will form a glass: (i) the

1 Chapter 1. Introduction 2 complexity of the atomic arrangement of the crystalline phase (must be high); (ii) the viscosity (the mobility of atoms must be low); (iii) the rate at which the liquid is cooled (fast quenching) (Angell et al., 2000).

A number of structural models have been developed to better characterize the location of specific atoms within the network and their position relative to other atoms of interest. Order exists in different ranges around atoms: short range order (SRO) is within the first coordination sphere and nearest neigh- bour (∼2 A),˚ intermediate range order (IRO) which encompasses the connection of adjacent structural units and network topology (10-20 A)˚ and possible long range order (LRO) defined by larger groupings of basic structural units particularly quasi-periodic or symmetric arrangements (Wright, 1990; Henderson, 2005).

1.1.1 Continuous Random Network (CRN) model

Zachariasen(1932) postulated the continuous random network (CRN) model in which he proposed that silica glass was composed of a network of corner-sharing SiO4 tetrahedra akin to that observed in crystalline counterparts: α-quartz, β-quartz, coesite, cristobalite, stishovite and tridymite. Silicon atoms act as network formers (T for tetrahedral) and are bonded to four oxygen with every oxygen atom bonded to two silicon; oxygen species which are bonded to two silicon atoms are termed bridging oxygen (BO) as they are the bridges between adjacent silica tetrahedra (Zachariasen, 1932). The structure of the silicate glass network is destabilized by the addition of network modiﬁer cations (M) such as; K2O or MgO. The premise of the model is that addition of modiﬁers destroys BO to form non-bridging oxygen (NBO) through the creation of M−O bonds; this process may be written using oxygen species as:

O2− + BO −−→ 2 NBO− (1.1) where O2– refers to free oxygen (Zachariasen, 1932). The formation of NBO leads to different types of bonding environments for Si atoms. The various types of Si (or any network former) are referred to as Qn species, where n is the number of BO atoms per silicon tetrahedron and has integer values between 0 and 4 (Henderson, 2005). The diffraction study by Warren and Biscce(1938) demonstrated that in sodium silicate glasses some oxygens were BO and others were NBO with the sodium atoms being six-fold coordinated to oxygen in random pockets in the glass network. They also correlated an increase in the quantity of NBO, with the addition of modifier cations. Warren and Biscce(1938) found that all of their experimental work was consistent with the CRN. Mozzi and Warren(1969) calculated the Si −O bond distance to be 1.62 A˚ and the Si−O−Si bond angle, α, was found to have a wide distribution function, V(α), with values ranging from 120 ◦-180 ◦. The large variation in bond angle reflected a large difference in the structure of glasses with respect to their crystalline forms, which have very narrow V(α). This also supports the idea of random orientation of the Si−O bond direction. Mozzi and Warren(1969) also noted the importance of finding the fraction of five, six and seven membered rings (see following discussion) in the structure and that the CRN was still a suitable model for vitreous silica. Chapter 1. Introduction 3

1.1.2 Modiﬁed Random Network (MRN) Model

The modified random network (MRN) theory was put forth by Greaves and contributors ((Greaves et al., 1981) and (Greaves, 1985)) in order to explain how silicate networks accomodate cations into their structure. They performed extended X-ray absorption fine structure experiments on glasses of a variety of compositions including Na2Si2O5. The proposed model for alkali silicate glasses consists of two interpenetrating crystal-like sub-lattices. The first sub-lattice is a continuous network of disordered

SiO2, penetrated by a second sub-lattice which holds the modifying cations. The two sub-lattices are linked by NBO atoms. Cations cluster around NBO atoms which leads to the formation of percolation channels (Fig. 1.1).

Figure 1.1: A modiﬁed random network (MRN) for a 2-dimensional oxide glass with composition Na2Si4O9 (after Greaves(1985) and adapted from Nesbitt et al.(2015)). Na −BO and Na−BO−Na were added by Nesbitt et al.(2015) and the percolation channels running through the covalent network are highlighted in white.

1.1.3 Central-force network models

Sen and Thorpe(1977) studied the vibrational density of states of glassy AX 2 systems (e.g. SiO2 and

GeO2) using a simple nearest-neighbor central-force (α) between A and X atoms in order to theoretically understand the observations seen in experimental vibrational spectra. They constructed their model ◦ ◦ using corner-sharing AX4 tetrahedral units with cation-anion bond angles between 90 and 180 and a Chapter 1. Introduction 4 potential energy (V) given by α X V = [(~u − ~u ) · ~r ]2 (1.2) 2 i j ij hiji with the summation being over all nearest-neighbor pairs hiji. The displacements of either type of atom

A and X are defined as ~ui,j and ~rij corresponds to a unit vector along the AX bond. This potential, V, may be thought of as a random network of Hooke springs (Jacobs and Thorpe, 1996). The main vibrational bands within the glasses are well explained using this formulation. The central-force model also demonstrated that the character of these modes ranged from pure anion bending to pure anion stretching which sparked much research interest (Galeener, 1979; Galeener and Mikkelsen, 1981; Sharma et al., 1981; Galeener, 1982; Galeener and Thorpe, 1983). Their work underlined the importance of ring structures in AX2 and other networks which has led to the assignment of many vibrational bands of vitreous SiO2 (see discussion in section on Raman spectroscopy). Percolation theory is a simple, yet powerful concept which has been used to solve multiple problems concerning connected pathways in solids or aggregates (Kirkpatrick, 1973). Typical applications of these ideas has consisted of fluid flow through porous media, the conductance of metal/insulator composites and magnetism in insulating alloys (Thorpe, 1985; Jacobs and Thorpe, 1996). These contributors applied percolation theory to describe the geometrical aspects of elasticity in generic central-force networks where the property of interest was rigidity. Rigidity percolation is similar to connectivity percolation except that rigidity is a vector quantity and has an inherent long range aspect to it; that is, a single bond on one end of the network can affect the rigidity of the entire network (Jacobs and Thorpe, 1996). The theory takes into account the coordination (constraint counting) of specific areas within the network which is then used to classify regions as floppy (underconstrained or low coordination) or rigid (overconstrained or high coordination) (Thorpe, 1985). More recently, Micoulaut and Phillips(2003) used a combinatorial algorithm to count clusters (rings) of different sizes with constraint counting algorithms to show that there actually exists two transitions, or, the equivalent of three elastic phases: floppy, isostatically rigid and stressed-rigid. The transition from one regime to another will tremendously affect the properties of the material and has implications for phase seperation in alkali silicate glasses since they go through a rigid to floppy transition with increasing alkali content at a critical value of 20M2O−80SiO2 (Micoulaut and Phillips, 2003).

1.1.4 Phase separation

Phase separation within alkali silicate glasses can be present as pockets of segregated modiﬁer cations or as channels as described above. This behaviour has been observed in the Na2O−SiO2 and K2O−SiO2 systems using a variety of techniques including MD simulations, Raman spectroscopy and inelastic neutron scattering in both the glasses and melts as outlined by Meyer et al.(2004), Maehara et al. (2005) and Bauchy and Micoulaut(2011). Similarly, the properties of mobile ions in ion conducting glasses (e.g. Li2O−SiO2) is controlled by phase seperation (Voigt et al., 2005). By correlating NMR with MD simulations Voigt et al.(2005) proposed that low-lithia ( ≤ 33 Mol. % Li2O) glasses exhibit nanophase-segregation of lithium-enriched domains. The average Q3−Li bond distances obtained from the MD simulations were used to calculate the average coordination numbers which are interpreted as disconnected channel-like structures (Voigt et al., 2005). Chapter 1. Introduction 5

1.2 Raman spectroscopy

Vibrational spectroscopy, in this case Raman spectroscopy, involves the use of light to probe the vibrational behaviour of molecular systems (McMillan and Hofmeister, 1988). Since the vibrational spectrum depends on the interatomic forces of the particular sample, it is a sensitive tool for observing the microscopic structure and bonding (McMillan and Hofmeister, 1988). The scattering of light by solid matter is governed by the excitation of phonons when interacting with a monochromatic beam of light, in most cases, a laser (νlaser). The elastic scattering of light is called Rayleigh scattering and has ν ∼ νlaser whereas the inelastic scattering of light is termed Raman scattering and has ν = νlaser ± νvibration (Gouadec and Colomban, 2007). All collective vibrations which occur in solids can be viewed as the superposition of plane waves that virtually propagate to infinity (Gouadec and Colomban, 2007). These plane waves, or normal modes of vibration, are commonly modelled by quasi-particles called phonons. Modes can be classified as either stretching (ν), bending (δ), torsional (τ), and librational (i.e., lattice modes which involve the relative displacement of unit cells which do not occur in glasses) (Gouadec and Colomban, 2007). For instance, the water molecule has three normal modes of vibration which are associated with a vibrational frequency −1 −1 −1 (νvibration: ν1 = 3652cm ; ν2 = 1595cm ; ν3 = 3756cm ) which is the result of a specific type of relative motion between the oxygen and hydrogen atoms (McMillan and Hofmeister, 1988).

1.2.1 The Raman spectra of silicate glasses

Figure 1.2: Raman spectra of SiO2 glass with both the parallel and perpendicular components (after Matson et al.(1983)).

Raman spectroscopy has been used on numerous occasions to try to decipher both the SRO and IRO of glasses (Matson et al., 1983; Dupree et al., 1986; McMillan, 1984; McMillan, 1989; Malfait,

2009; Koroleva et al., 2013). The spectra of vitreous SiO2 contains many vibrational bands which as mentioned previously correspond to modes of a certain character. The area from 100-400 cm−1 is caused Chapter 1. Introduction 6 from motion due to the silicate network and is usually associated with (5,6,7-membered) rings (Matson et al., 1983).

The most pronounced band is broad and located at ∼440 cm−1 with a secondary sharp peak adjacent to it at ∼492 cm−1 (Fig. 1.2). The 440 cm−1 band is the result of symmetric motion of BO relative to the Si atoms in the 3-dimensional network (νs(Si−O−Si)) (Galeener, 1979; Galeener and Mikkelsen, 1981). The sharp peak at 492 cm−1 is attributed to the breathing motions of oxygen associated with

4-membered rings composed of SiO4 tetrahedra which are vibrationally isolated from the silica structure (Sharma et al., 1981).

The weak band located at 606 cm−1 corresponds to the breathing motions of oxygen associated with planar 3-fold rings (Galeener, 1982). The breathing motions of oxygen provide an explanation for the frequency, sharpness and strong polarization of both bands (492 and 606 cm−1) (Fig. 1.2). The weak band at ∼800 cm−1 results primarily from Si motion within the tetrahedron (Galeener and Mikkelsen, 1981).

−1 The last feature of the SiO2 glass spectrum is two broad, low intensity bands situated at ∼1050 cm and ∼1200 cm−1. These bands have long been interpreted as the longitudinal-optic-transverse-optic (LO-TO) splitting induced by the long-range nature of the Coulomb fields present in ionic materials, as observed in α-quartz (McMillan and Wolf, 1995, and references therein). More recently, Sarnthein(1997), investigated the vibrational properties of vitreous silica using first-principles density functional theory in order to resolve the origin of the high-frequency double peak. Many quantities of interest can be obtained through these simulations: equation of state, thermodynamics, atomic and electronic structures, self- diffusion and viscosity (Karki, 2010). By calculating the total density of states, the doublet was ascribed to the A1 and T2 modes belonging to the Td symmetry group of the SiO4 tetrahedron. The A1 mode corresponds to the in-phase motion of all oxygen atoms toward the Si atom in the tetrahedron, whereas, the triply T2 modes are due to two oxygens moving toward the central Si atom while the other two move away (Sarnthein, 1997).

Raman spectroscopy has more recently been used in conjunction with density functional theory to provide information about the medium-range structure of vitreous SiO2 (Giacomazzi et al., 2009). A comparison of three model structures with differing Si−O−Si bond angle distributions and ring statistics was used to investigate the effects on the vibrational spectra. A correlation was found between a decreasing average Si−O−Si bond angle and a positive shift in the position of the ∼800 cm−1 band. Their findings showed that the medium-range structure was characterized by small SiO4 ring (4,5,6,7-membered) concentrations. They also state that the large number of six-membered rings can be explained through the silica phase diagram; when the system is cooled from the liquid phase (at atmospheric pressure) the crystalline phase corresponds to β-cristobalite which consists only of rings of size 6 (Giacomazzi et al., 2009).

The addition of modiﬁer cations to the silicate network alters the Qn species distribution via Eq. 1.1. The greatest change in the spectrum is the formation of a broad peak between 900 cm−1 and 1200 cm−1 (Fig. 1.3). This peak is a result of the bands produced by every type of Qn species present in the glass (Matson et al., 1983). It is possible to ﬁt the spectrum in order to extract the contribution from each individual Qn species and monitor their relative proportions with varying alkali content, atomic radius and temperature (Maehara et al., 2005; Malfait, 2009; Le Losq et al., 2012). Chapter 1. Introduction 7

Figure 1.3: (a) Raman spectra of glass and crystalline Cs disilicates (after Matson et al.(1983)). (b) Raman spectrum from this study with higher resolution down to very low wavenumber.

As demonstrated in the data of many Raman experiments of alkali silicate glasses (Matson et al., 1983; Dupree et al., 1986; Malfait, 2009; Le Losq et al., 2012; Koroleva et al., 2013) the intensity of the Qn species distribution band increases with increasing alkali content and shifts to lower wavenumbers. This is caused by the increase in lower Qn species at the expense of Q4 and perhaps Q3. These changes in the network aﬀect the overall ring size distribution of the glass causing larger, 5,6,7-membered rings, −1 to break apart. Evidence for this is the decrease in intensity of the 100-400 cm region from SiO2 to alkali silicate glass with increasing alkali content. We also observe an increase in 3-membered rings (606 Chapter 1. Introduction 8

−1 −1 cm ), whose peak intensity transitions from being lower than 4-membered rings (492 cm ) in SiO2 glass to a higher intensity in alkali disilicate (M2O−2 SiO2) glasses.

(a) (b)

n Figure 1.4: (a) High temperature Raman spectra of crystalline Na2Si205 (b) Curve ﬁtting for Q species n distribution of Na2Si205 melt at 1473 K. Each deconvolved peak represents a speciﬁc Q species unit (after You et al.(2001)).

Raman spectra of silicate liquids have now been acquired via high-temperatre in-situ experiments (You et al., 2001; Maehara et al., 2005; Malfait, 2009; Koroleva et al., 2013). The results differ slightly from one study to the other but there is a definite effect of temperature on the spectrum as a whole, as outlined by the broadening of the spectral envelope from the onset of melting and the creation of more Q1 species (Fig. 1.4).

1.3 X-ray absorption spectroscopy

X-ray absorption spectroscopy is a synchrotron characterization technique which is generally split up into two regions (i) X-ray absorption near-edge structure (XANES) and (ii) Extended X-ray absorption fine structure (EXAFS). This brief overview will focus only on the XANES part of the absorption spectrum. For a theoretical treatment of XANES, Rehr and Albers(2000), de Groot(2001), de Groot (2005) and Henderson et al.(2014) provide reviews. The X-ray absorption process involves a core electron being excited to an empty state (de Groot, 2001). XANES is element sensitive because each element has different binding energies for each of its electrons. The binding energy of core electrons increases with increasing atomic number and decreases in energy with increasing quantum number (Henderson et al., 2014). The incoming X-ray intensity (I0) is chosen specifically to probe a certain energy range in which the absorption edge of a chosen element is located. The XANES intensity of these core level transitions is described by Fermi’s Golden Rule:

2 IXANES ∝| hi | eˆq · r | fi | ρ (1.3) where i and f are the initial and ﬁnal electronic states, ρ is the density ande ˆq · r is the electric dipole operator (Henderson et al., 2014). Chapter 1. Introduction 9

The variation in the absorption of X-rays due to the photoelectric effect is measured by a variety of different detection methods; for this study we will use fluorescence yield (FLY) rather than total electron yield (TEY) detectors because glasses are poor conductors which result in noisy TEY data (Henderson et al., 2014). The spectrum is then a result of the measured absorption versus the incoming intensity,

I0 in eV.

1.3.1 XANES of silicate glasses

Figure 1.5: Mg K-edge spectra of haplobasaltic glasses and standards. A) Stacked intensity normalized spectra. B) Area normalized haplobasaltic spectra overlain to compare the subtle changes in edge intensities. The reference dashed line is at 1310 eV in both A and B. C) Increase in intensity of peaks A and B with haplobasaltic composition for area normalized spectra in B. Adapted from Moulton et al.(2016).

Many scientists have chosen to use a variety of XANES edges in order to investigate the behaviour of particular elements in the glass network (Henderson et al., 2014, see references therein). A good example of such a study is the work of Moulton et al.(2016) where glasses from the anorthite-diopside compositional join (CaAlSi2O8−CaMgSi2O6) were investigated using XANES. All the elements from this system were probed at multiple edge energies: Si, Al, Mg, and O K -edges and Ca, Si, and Al L2,3-edges (Moulton et al., 2016). In Fig. 1.5, Moulton et al.(2016) illustrate that the coordination of Mg atoms can be inferred from the intensities of peak A and B as shown from their standards of diopside ([6]Mg) and spinel ([4]Mg). They highlight that [5]Mg is likely the most important environment for diopside rich Chapter 1. Introduction 10 glasses, whereas [4]Mg becomes increasingly important for compositions with higher amounts of Al. Additionally, the modelling of XANES spectra based on charge transfer multiplet theory (de Groot, 2005) and ﬁrst-principles density functional theory (Cabaret et al., 2007, 2010); is a very important component for understanding diﬀerent spectral features.

1.4 Contribution of this thesis

This thesis is organized into six chapters, including this introduction as well as a final concluding chapter that highlights the contributions. The additional four chapters each entail unique investigations into different important aspects related to the structure of alkali silicate glasses and melts. Chapter 2 is an in depth investigation of the structure caesium silicate glasses and melts using Raman spectroscopy on a range of compositions and temperatures. Very few studies have covered the entire spectral range in such detail. Every Raman peak is described and discussed. The room- and high- temperature measurements are also used to estimate the changes in the Qn species distribution as a function of cation concentration and temperature. Chapter 3 elaborates upon Chapter 2 by investigating the structure of alkali silicate glasses covering a wide compositional range. The most important parts of the spectra, the “Defect” bands and the Qn species envelope are investigated in great detail. The “Defect” bands are used to estimate the population distribution of 3- and 4-membered rings, something which has previously only been done in Chapter 2 n and for v-SiO2. The most influential part of this study is the use of different lineshapes to fit the Q species envelope to estimate the proportion of Qn species as a function of alkali type and concentration. Chapter 4 and 5 are the first Li K -edge XANES experiments conducted on geologically relevant samples. Chapter 4 is an analysis of a large suite of lithium-bearing salts and minerals in order to better understand the behaviour of the Li K -edge in a variety of crystalline samples ranging from simple to complex lithium bonding environments. Chapter 5 uses the results from Chapter 4 to investigate a series of lithium silicate and lithium-bearing alkaline-earth silicate glasses. The change in the peak area is tracked as a function of lithium content and alkaline-earth type. This thesis would not have been possible without the training and guidance of many collaborators. I must explicitly thank Daniel Neuville for allowing me to spend a few summers working in the Raman spectroscopy laboratory of the Geomaterials group at the Institut de Physique du Globe de Paris (IPGP) and for his input on all our papers. Wayne Nesbitt and Michael Bancroft from the University of Western Ontario also contributed many ideas which are present throughout the Raman spectroscopy chapters of this thesis, our discussions on the fitting of Raman spectra were especially enlightening. Lucia Zuin and Tom Regier were incredibly available and helpful during all the beamtime at the Canadian Light Source. Finally, I would like to thank Benjamin Moulton for his helpful discussion and assistance in gathering the XANES data presented in the two Li K -edge chapters. Chapter 2

Structure-property relations of caesium silicate glasses from room temperature to 1400 K: implications from density and Raman spectroscopy

This chapter has been published in the journal Chemical Geology as:

• O’Shaughnessy, C., Henderson, G. S., Nesbitt, H. W., Bancroft, G. M., Neuville, D. R. (2017) Structure-property relations of caesium silicate glasses from room temperature to 1400 K: Implications from density and Raman spectroscopy. Chemical Geology 461, 82–95.

Additionally, the melting mechanism described in this chapter was elaborated in:

• Nesbitt, H. W., Bancroft, G. M., Henderson, G. S., Richet, P., O’Shaughnessy, C. (2017a) Melting, crystallization, and the glass transition: Toward a uniﬁed description for silicate phase transitions. American Mineralogist 102, 412–420.

I have carried out all of the experiments at the Raman laboratory of the Geomaterials group at the Institut de Physique du Globe de Paris, France, completed the data processing, ﬁgure preparation and wrote the article. All other authors have provided training and given feedback during the publishing process.

11 Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 12

Abstract

This study investigated a series of caesium bearing silicate glasses with compositions ranging from 5 to 35 mol % Cs2O using Raman spectroscopy. The proportions of 4- and 3-fold SiO4 rings are estimated by using the relative intensities of the D1 and D2 vibrational bands in each spectra. An increase in the abundance of 3-fold SiO4 rings is apparent and confirms previous studies which proposed such a mechanism to be present with increasing alkali content in alkali silicate glasses. The behaviour of the D1 and D2 bands correlate with the density measurements lending credence to the importance of the formation of 3-fold SiO4 rings with increasing molar volume in silicate glasses. The high frequency envelopes (∼800-1300 cm−1) were modeled using Gaussian lineshapes in order to isolate the separate contributions to the convoluted peak shape. These Gaussian bands are interpreted as being representative of the Qn species and are linked to the depolymerization reaction. We observe 2 4 an increase in Q species coupled with a decrease in Q species with increasing Cs2O content. We propose that proximity to modifier cations drastically changes the vibrational character of the different Qn species and due to this effect we have assigned our curve-fit bands to both Qn and QnM species. High-temperature measurements were also conducted to investigate pre-melting and melting effects on the structure of these glasses. We observed similar changes in the high frequency envelope and Qn species distribution with increasing temperature as we did for increasing alkali content. The creation of Q2 species above the melting point leads us to conclude that depolymerization occurs at the solid-liquid transition.

2.1 Introduction

The structure of silicate glasses and melts is fundamental to our understanding of the physical properties of magmas. It is of interest to both Earth and material scientists. Silicate magmas are considered to be admixtures of liquid (melt), crystals and gases. Melts play a key role in the transfer of both mass and heat within the Earth and other terrestrial planets and they have received a considerable amount of experimental research interest since the 1930s (cf., Warren, 1934; Warren and Biscce, 1938; Warren and Biscob, 1938; Brawer and White, 1975; Greaves et al., 1981; Matson et al., 1983; Henderson et al., 2006). These eﬀorts have focused on understanding the relation between mass, energy transfer and melt structure. Many physical properties, such as, thermal or electrical conductivity, density and viscosity are directly related to the atomic structure of the melt (Stebbins et al., 1995, chapters therein). These properties are important on many levels, for example volcanic eruptive types are dependent upon the viscosity of the eruptive magma (Dingwell, 1989). Due to the diﬃculties which arise to study the atomic properties of melts, glasses are used as analogs.

Binary alkali silicate glasses and their melts (M2O−SiO2) oﬀer compositionally simple systems to investigate the structure of both magmatic and technologically signiﬁcant counterparts (Henderson, 2005). Many groups have investigated the structure of these glasses using a wide range of spectroscopic techniques (Brawer and White, 1975; Matson et al., 1983; McMillan, 1984; Greaves, 1985; Dupree et al., 1986; Maekawa et al., 1991; Nesbitt et al., 2011) and others via molecular dynamic (MD) simulations (Huang and Cormack, 1991; Du and Cormack, 2004; Du and Corrales, 2006; Gedeon et al., 2010; Bauchy and Micoulaut, 2011). Additionally, with experimental furnace designs, many have also obtained Raman spectra of glasses or melts in situ at high temperature (McMillan, 1984; Mysen and Frantz, 1992; Mysen, Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 13

1995; Frantz and Mysen, 1995; Maehara et al., 2005).

In addition, the structure of Cs-silicate glasses (xCs2O−(1−x)SiO2) is of interest for nuclear waste glasses due to diﬃculty in immobilizing 135Cs, which has a long half-life and is an abundant component of nuclear waste products (Malfait, 2009). Additionally, Cs is an important component in lithium-caesium- tantalum (LCT) pegmatite style deposits from which the majority of Cs is mined and its behaviour at higher temperatures as well as high water contents is of particular interest in the formation environments of these deposits (London, 2005).

In pure SiO2 glass, silicon atoms are bonded to four oxygen atoms to form a tetrahedron. These tetrahedra then form a corner-sharing network with all oxygens shared by two silicons (Zachariasen, 1932; Warren, 1934). Where alkalis are introduced they rupture Si-O-Si bonds to produce Si-O-M bonds, thus causing depolymerization of the network according to the Continuous Random Network (CRN) model (Zachariasen, 1932):

O2− + BO −−→ 2 NBO− (2.1) where BO and NBO are bridging and non-bridging oxygen and O2− is referred to as free oxygen. Bridging oxygens (BO) are bound to two silicons, non-bridging oxygens (NBO) are bound to a silicon and an alkali (Zachariasen, 1932). In order to describe the structure around each silicon atom to its second coordination sphere the Qn species notation is applied, where n, corresponds to the number of bridging oxygen per tetrahedron. Greaves et al.(1981) proposed that the alkali silicate glass network would be an extension of the crystalline counterpart, being made up of two interconnected sub-lattices: (i) a continuous, disordered SiO2 network, (ii) an interpenetrating alkali sub-lattice. A consequence of this model is cation clustering around NBO atoms leading to the formation of percolation channels (Greaves et al., 1981; Greaves, 1985). These observations appear to be consistent with ionic and channel diffusion measurements (Ingram, 1989) as well as with molecular dynamics (MD) simulations of alkali silicate glasses which show silica rich and modifier rich zones within each sample and as more modifier is added the zones get larger and form NBO rich channels (Huang and Cormack, 1991; Du and Cormack, 2004; Bauchy and Micoulaut, 2011). In the last 40 years there have been many Raman spectroscopic studies of binary silicate glasses containing alkali and alkali-earth elements (Neuville et al., 2014, for a recent review). In most cases, peaks are present in the 700-1100 cm−1 region and should be able to be assigned to the different Qn species (Q0,1,2,3,4) contained in the glass. However, there has been a great deal of controversy over the assignment of these peaks. It is generally accepted that the ∼850 cm−1, ∼900 cm−1, ∼970 cm−1, ∼1100 cm−1 and the ∼1200 cm−1 bands are interpreted as the Q0,Q1,Q2,Q3 and Q4 bands, respectively (Rossano and Mysen, 2012, references therein). Additional considerations from results obtained in crystals is needed in order to attempt to model these bands in a more consistent manor. The most important issue to solve is that of the Raman cross-section of the different Qn species and how it varies with composition and temperature. Raman linewidths in gas phase molecules may be affected by Doppler broadening, by collisional lifetime and by natural Raman lifetime, the last being an intrinsic property of the Raman transition. Doppler broadening results from atoms moving toward or away from the radiation source and band broadening increases as the square root of temperature (Atkins, 1998, p. 462). Band widths affected by collisional lifetime are dependent upon the collision frequency of a central atom with adjacent atoms. For a crystalline silicate solid, Richet et al.(1996, Table A4) measured the linewidth of the ∼950 cm−1 Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 14

2 −1 ◦ (Q ) band of crystalline Na2SiO3 (from the Si−O symmetric stretch) as 6 cm at 25 C broadening to 39 cm−1 near its melting point 1348 ◦C. The full width at half maximum (FWHM) increases with temperature, whereas its position decreases with increased temperature. The relationship between the position and the temperature is non- linear thus proving that neither Doppler broadening nor collisional lifetime makes a major contribution to the Q2 linewidth of crystalline Na metasilicate. The 6 cm−1 width is probably close to the natural linewidth; and the temperature broadening is probably mainly due to variations in local SiO4 structure on heating with perhaps some contribution from Doppler broadening. For glasses and melts, Richet et al.(1996) showed that the linewidths from the Si −O stretching peaks for all Qn species increased above the pre-melting value of 39 cm−1. Again, these linewidths must be due to the variations in local structure. But most importantly, recent results (Bancroft et al., 2018) strongly indicate that all Qn species’ linewidths should be very similar. The idea that individual Qn species have comparable linewidths in the Raman spectra of glasses is extremely important for our ensuing analysis. The natural lifetime of the transition giving rise to the Q2 band imposes a minimum linewidth on the band due to uncertainty related to the lifetime of the excited state (Atkins, 1998, p. 463). The FWHM 2 −1 of the Q band of crystalline Na2SiO3 is ∼6 cm but for the glass of the same composition the band is much broader. Low frequency bands give rise to narrow linewidths and high frequency transitions (i.e., Q0 to Q4 bands) display broader bands due to the shorter lifetime of the excited state (Atkins, 1998, p. 463). The highly localized, high frequency Q0,Q1,Q2 and Q3 bands have a common origin, the Si-ONBO symmetric stretch so that they likely will display similar natural (minimum) linewidths. Nevertheless, the Qn bands span ∼400 wavenumbers and depending on the sensitivity of the lifetime to the Raman transition frequency, the Qn bands located at very high wavenumbers (e.g., 1150 cm−1) may be somewhat broader than those positioned at lower wavenumbers (e.g., 800 cm−1). In this paper we discuss the various features of the Raman spectra of caesium silicate glasses and melts. Due to the size diﬀerence between caesium and silicon we believe to see the most pronounced changes to the network as alkali is added. A model for the high frequency envelope is proposed and results from both ambient and high temperatures are discussed with respect to current research.

2.2 Methods

2.2.1 Glass synthesis

A variety of caesium silicate glasses were synthesized with concentrations ranging from 5-35 mol %

Cs2O in 5 mol % increments. The samples bare the mol % Cs2O as a name, e.g. Cs5, and will now be referred to in this manner. The samples were prepared by combining the stoichiometric amounts ◦ of Cs2CO3 to SiO2 to make approximately 3 g batches. The mixtures were then calcined at 900 C for 24 hours in order to drive oﬀ the CO2. The mixtures were placed in Pt crucibles and heated to 200 ◦C above their respective melting temperatures for 1 hour and then quenched. The samples were mechanically crushed in a mortar and pestle and the melting and quenching process was repeated. All glasses were transparent and showed no macroscopic signs of defects or crystals. Samples were stored in a ◦ desiccator or 100 C furnace to avoid interaction with atmospheric H2O. The densities of the Cs-bearing silicate glasses were measured using a Berman density balance and all data presented are a result of 10 independent measurements. All samples have densities which are consistent with previously published data (Tischendorf et al., 1998; Doweidar et al., 1999). The molar volume (Vm) was also calculated using Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 15

the density measurements for the series of glasses. The Vm were calculated using stoichiometries with an equivalent amount of oxygen (i.e. Cs4O2 and SiO2) along the compositional join via the relation: PN Vm= i=1 xiMi/ρmix where xi is the proportion of phase i, Mi is the molar mass of i and ρmix is the measured density of the sample.

2.2.2 Raman spectroscopy

Raman spectroscopic experiments were conducted on the glasses at the Raman Laboratory of the Institut de Physique du Globe de Paris (IPGP). The spectra were recorded using a T64000 Jobin-Yvon triple grating Raman spectrometer equipped with a confocal system, a 1024 CCD detector cooled by liquid nitrogen and an Olympus microscope (Le Losq et al., 2012). A Coherent 70-C5 Ar+ laser with a wavelength of 488 nm is used for the excitation line with a laser power of 100-250 mW. The measurement resolution was better than 0.5 cm−1. Scans were taken in triple-subtractive mode which averages three spectra per sample. The background continuum was removed using a polynomial spline. Intensity and area normalizations of the spectra were applied depending on the region and the quantity of interest.

Cs10

Figure 2.1: Example of the Gaussian curve-fitting of Cs10 with the residual. Area normalized spectra are fit using a minimum number of bands (5) with similar FWHM. Due to the area normalization the areas of the Gaussian components can be used to estimate the proportion of one band to the other. Observing the change in the relative areas of the bands with increasing Cs2O composition provide a way to estimate the proportions of the different chemical species present in the glasses.

Additional measurements were taken in situ at high temperatures using a Pt-Rh alloy wire heating system designed for the acquisition of spectroscopic data (Raman, XAS, etc.) of silicate melts (Neuville et al., 2009). The wires are calibrated using substances with known melting points in order to achieve Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 16 reliable temperature determination. Three samples were chosen for the melting experiments and spectra were taken at various temperature steps (640 ◦C, 800 ◦C, 950 ◦C, 1057 ◦C & 1140 ◦C). Spectra were taken before and after the melting in order to compare them and assure no loss of material. Gaussian curve-fitting was applied to the high frequency envelope in order to model the different vibrational bands which compose this complex feature. The high frequency part of the spectra were area normalized to allow for band area calculations. A procedure similar to that of Mysen et al.(1982a) was then employed using a Levenberg-Marquadt fitting algorithm on the Fityk software package (Wojdyr, 2010). Due to the theoretical consideration that Qn species should have similar linewidths, the FWHM were constrained to the range of 45-65 cm−1. This range was chosen because it reflects a range of 2 −1 linewidths larger than that of the Q species (39 cm ) in molten Na2SiO3 (Richet et al., 1996). This takes into consideration the change in linewidth with temperature and to accomodate for our larger cation (Cs instead of Na). The number of Gaussian curves chosen was the smallest quantity required to minimize χ2 and yield a random residual (Mysen et al., 1982a). We determined the minimal number of bands by fitting all spectra with 3-7 bands and evaluating which number gave the best fit, in all cases the increase in the quality of the fit (minimized and random residual) from 5 to 6 or 7 bands was negligeable. Thus, five bands were chosen to reflect the vibrational bands representative of the Qn species. An example of a fit of Cs15 with the corresponding residual is shown in Fig. 2.1. The model describes quite accurately the main features of the high frequency envelope for all samples in the suite including the high-temperature spectra.

2.3 Results

2.3.1 Room temperature Raman spectroscopy

Sample BP Main D1 D2 C Qn

ωc FWHM ωc ωc FWHM ωc FWHM ωc ωc

Cs5 29 46 454 488 48 599 30 802 1099 Cs10 29 46 456 492 50 599 32 797 1099 Cs15 29 50 459 497 50 598 32 790 1097 Cs20 30 54 - 506 44 595 32 783 1098 Cs25 31 56 - 514 38 592 35 770 1100 Cs30 34 60 - 519 34 589 37 768 1099 Cs35 35 60 - 519 44 588 40 772 1099

−1 −1 Table 2.1: The Raman shift, ωc (± 1 cm ) and FWHM (± 2 cm ) values of the various peaks in the Cs-silicate glasses spectra.

The spectra of the alkali silicate glasses can be subdivided into four main regions: 0-100 cm−1, 400-600 cm−1, 750-800 cm−1, and 900-1300 cm−1 (Fig. 2.2). Seven spectra ranging from 5-35 mol %

Cs2O were used to compare the eﬀect of composition on the diﬀerent regions of the spectra. At the low frequency end of the spectra (0-100 cm−1) lies the Boson peak, which varies negligibly in intensity but both the peak centre and FWHM vary with increasing Cs2O content. In the region spanning 400-600 Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 17

−1 −1 −1 cm there are three peaks present at low Cs2O content: the main band (454 cm ), the D1 (495 cm ), and D2 peaks (606 cm−1). The intensity of the peak at 454 cm−1 decreases steadily until it is no longer visible at Cs20 and higher Cs2O compositions. The peak centres (ωc) of the D1 and D2 peaks vary with composition, shifting to higher and lower wavenumber respectively, with increasing caesium composition. The peak located between 750-800 cm−1 (Peak C) displays a 35 cm−1 negative peak shift, as well as, a decrease in intensity with increasing Cs content, it is almost undetectable in sample Cs35. Finally, the high frequency portion (900-1300 cm−1) of the spectra represents the Qn species envelope. This envelope seems to consist of three main contributions; a broad shoulder on the high wavenumber side, the main peak, and a minor shoulder on the low frequency side of the envelope at high Cs2O content. The shape of this envelope undergoes substantial changes with increasing alkali content.

BP Main D1 D2 C Qn

Figure 2.2: Raman spectra of caesium silicate glasses. The numbers on the right side of the spectra represent the value of x for x · Cs2O−(1−x) · SiO2 (mol %). The spectra are vertically shifted in order to distinguish the diﬀerent features present. Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 18

2.3.2 High-temperature Raman spectroscopy

Raman heating experiments were conducted on three glasses (Cs5, Cs10, Cs15) with a focus on the high frequency envelope in an attempt to better understand the eﬀects of temperature on reactions among Qn species (McMillan, 1984; Mysen and Frantz, 1992; McMillan and Wolf, 1995; Mysen, 1995; Frantz and Mysen, 1995; Maehara et al., 2005; Henderson et al., 2009). It is worth noting that the heating experiments are reversible, that is, once the high-temperature experiment was complete the sample was quenched at room temperature and another spectrum was acquired. In all cases, there were negligible diﬀerences between the two room temperature spectra (before and after heating).

Cs10

1140o C

1060o C

950o C

800o C

640o C

25o C

Figure 2.3: High-temperature spectrum of Cs10 (640 ◦C, 800 ◦C, 950 ◦C, 1057 ◦C & 1140 ◦C). The spectra are vertically shifted in order to distinguish the diﬀerent features present.

The high temperature spectra of Cs10 are shown in Fig. 2.3. All three samples show the same behaviour with increasing temperature, the broad shoulder on the high frequency side of the envelope diminishes and the low frequency peak or shoulder increases in intensity. The overall envelope seems to shift to lower frequency and the main peak width appears to increase with increasing temperature.

This behaviour is very similar to that of the Li2O−SiO2, Na2O−SiO2 and K2O−SiO2 glasses studied by Mysen and Frantz(1992), who observed similar changes from 25-1400 ◦C. Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 19

2.4 Discussion

The features in the Raman spectra of the Cs2O−SiO2 glasses vary systematically and allow for the identification of different molecular vibrations which are taking place with increasing modifier composition.

2.4.1 The Boson peak

Figure 2.4: The peak position (a) and FWHM (b) of the Boson peak as a function of composition.

The ﬁrst feature on the low frequency portion of the spectra is referred to as the Boson peak (BP) because its intensity distribution can be approximated by Bose-Einstrein statistics (Elliott, 1992; Surovt- sev and Sokolov, 2002; Gurevich et al., 2003). The BP is observed in both inelastic scattering of light, neutrons or X-rays (Buchenau et al., 1984) or inferred by the heat capacities of glasses as they do not obey Debye T3 laws at very low temperatures (Elliott, 1992). This band stems from an excess density of modes with respect to the low-frequency end of the vibrational densities of states, g(ω), which would Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 20 represent only the Debye longitudinal and transverse acoustic modes (Richet, 2009). There is still debate surrounding the origin of the Boson peak but two main theories seem to be the most supported. The

Boson peak of vitreous SiO2 has been assigned to: (i) rotational motions of almost perfectly rigid SiO4 tetrahedra by inelastic light or neutron scattering (Buchenau et al., 1986; Hehlen et al., 2000); (ii) a van Hove singularity associated with the Brillouin zone limit of a transverse acoustic band by inelastic X-ray scattering (Duval et al., 2004). For silicate glasses investigated using Raman spectroscopy the BP lies between 25-60 cm−1 and its peak position varies systematically with composition (Richet, 2009; Kalampounias et al., 2009; Le Losq and Neuville, 2013; Hehlen and Neuville, 2015). We recorded the values of the peak centre and FWHM of the Boson peak in all seven room temperature spectra of the caesium silicate glasses. The values are shown in Fig. 2.4 and a clear correlation between the increase in Cs2O content and a positive shift in peak centre frequency can be observed. The FWHM of the BP displays a similar trend with increasing alkali concentration. The peak centre varies by approximately 10 cm−1 whereas the FWHM spans a range of 30 cm−1. Both of these trends are in agreement with previous Raman experiments on Li-, Na-, and K-bearing silicate and Na-, Ca-bearing aluminosilicate glasses (Richet, 2009; Hehlen and Neuville, 2015). The changes to the location and breadth of the BP with alkali content are present, but minor, in comparison to the effect of the type of cation. As shown by Richet(2009), the excess heat capacities and vibrational densities of states depends more strongly on the type of alkali than on the degree of polymerization of the silicate network. Finally, we believe that the increase in FWHM also agrees strongly with the theory that the modifier cations play an active role in the localized vibrations associated with the Boson peak (Richet, 2009, and references thein). Thus, silicon tetrahedra involved in floppy modes seem to have both bridging and non-bridging oxygen. There is a need for more experiments and atomistic simulations to help better understand this excess in the vibrational density of states of amorphous materials.

2.4.2 The D1 and D2 bands: 495 and 606 cm−1

The D1 and D2 peaks correspond to breathing modes of small SiO4 rings in the glass. The vibrations correspond to 3-membered (D2) and 4-membered (D1) SiO4 rings (Sharma et al., 1981; Galeener, 1982; Barrio et al., 1993; Pasquarello et al., 1998). These contributions of in-phase pure -O- breathing motions in the 3, 4-membered SiO4 rings found in pure silica glasses are most certainly also present in alkali silicate systems (Hehlen and Simon, 2012). With increasing alkali content the proportions of 3- and

4-membered SiO4 rings will change and should be identiﬁable in the Raman spectra. The D1 peak centre migrates to higher wavenumber and decreases in intensity with increasing alkali content. Whereas, the D2 band centre shifts to lower frequency while its intensity increases with added

Cs2O (Table 2.1). The intensity ratio of the D1 and D2 bands as a function of composition is plotted in

Fig. 2.5(b) and it shows a transition from a regime dominated by 4-membered SiO4 rings (D1>D2) to that of a distribution with a higher number of 3-membered SiO4 rings (D1

The increase in the abundance of 3-membered SiO4 rings coupled with a decrease in 4-membered SiO4 rings was previously noted for Cs-silicate glasses, though an argument using the Qn species distribution was used as justiﬁcation (Malfait, 2009). An increase in the intensity or area of the D2 band is also observed in the densiﬁcation of vitreous silica and is often proposed as a simple mechanism by which silicate glasses accommodate increasing pressure (Sonneville et al., 2012; Deschamps et al., 2013). A similar process must occur when a large cation such as Cs is introduced into the silicate network. Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 21

Figure 2.5: (a) The measured density of caesium silicate glasses (our study: empty triangles, Tischendorf et al.(1998); Doweidar et al.(1999): ﬁlled triangles.) plotted with the calculated molar volumes (empty squares). (b) Ratio of the intensities of the D1 and D2 Raman bands (black circles). (c) The D1-D2 peak splitting, ∆D.

Conversely to the densiﬁcation of vitreous silica, the change in chemistry requires us to consider the molar volume. The change in the molar volume tracks well with the change in density and the ratio of

ID1/ID2 which is reﬂective of the proportion of smaller SiO4 rings (3, 4-membered). The peak centres of both the D1 and D2 bands vary with increasing alkali content. The D1-D2 Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 22

splitting, ∆D = |ωD1 − ωD2|, is a measure of the relative change of the two peak centres as a function of the composition (Fig. 2.5(c)). The value of ∆D decreases with alkali content as the D1 peak shifts to higher wavenumber with the median value located at Cs20 (∼ 90 cm−1). This may demonstrate a decrease in the Si-O-Si bond angle for 4-membered SiO4 rings which are closing as Cs is added to the network (Sharma et al., 1981; Galeener, 1982). This is consistent with the formation of 3-membered

SiO4 rings caused by the breakdown of higher order SiO4 rings (≥4) due to the increasing concentration of NBO with the introduction of modiﬁers in the glass network (Greaves et al., 1981). The value of

∆D varies with composition and similar behaviour has been observed in both crystalline germanate and silicate pyroxenes (Wang et al., 2001; Tribaudino et al., 2012). The splitting of the 670 cm−1 band in (Ca,Mg)MgSi2O6 has been interpreted as a phase transition from C2/c to P21/c with increasing clinoenstatite (Mg2Si2O6) content and was also associated with splitting in the high frequency band −1 (∼1015 cm )(Tribaudino et al., 2012). Perhaps, the ∆D value reflects a similar structural change in the glasses and that splitting of the high frequency bands should also be expected. Percolation theory is a simple, yet powerful concept which has been used to solve multiple problems concerning connected pathways in solids or aggregates (Kirkpatrick, 1973). Typical applications of these ideas has consisted of fluid flow through porous media, the conductance of metal/insulator composites and magnetism in insulating alloys (Thorpe, 1985; Zhang and Boolchand, 1994; Jacobs and Thorpe, 1996). These contributors applied percolation theory to describe the geometrical aspects of elasticity in generic central-force networks where the property of interest was rigidity. Rigidity percolation is similar to connectivity percolation except that rigidity is a vector quantity and has an inherent long range aspect to it; that is, a single bond on one end of the network can affect the rigidity of the entire network (Jacobs and Thorpe, 1996). The theory takes into account the coordination (constraint counting) of specific areas within the network which is then used to classify regions as floppy (underconstrained or low coordination) or rigid (overconstrained or high coordination) (Thorpe, 1985). More recently, Micoulaut and Phillips(2003) used a combinatorial algorithm to count clusters (rings) of different sizes with constraint counting algorithms to show that there actually exists two transitions, or, the equivalent of three elastic phases: floppy, isostatically rigid and stressed-rigid. The transition from one regime to another will tremendously affect the properties of the material and has implications for phase seperation in alkali silicate glasses since they go through a rigid to floppy transition with increasing alkali content at a critical value of 20M2O−80SiO2 (Micoulaut and Phillips, 2003). The transition from one rigidity domain to another must be apparent in the Raman signal because of the change in vibrational modes associated with a more rigid framework. At approximately 20 mol % which coincides with both the deviation from a linear increase in the molar volume (Fig. 2.5(a)) as well as the critical transition value of rigid to floppy behaviour of the network (Micoulaut and Phillips, 2003) we observe the cross-over of ID1/ID2 from greater than to less than unity. Potentially, there is an effect on the density and molar volume of glasses which is attributed simply to the difference of mass between caesium and silicon but, there may also be an effect of packing which is seen by the increase of 3-fold SiO4 rings and the way in which Cs is configured within the network. This may lead to the transition from a rigid to a floppy network.

2.4.3 The C peak (770 cm−1)

The isolated, low intensity, asymmetrical C peak (770 cm−1) is the least pronounced feature in the spectra of silicate glasses. Galeener and Lucovsky(1976) demonstrated that the peak was a consequence Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 23

of TO-LO (tranverse optic-longitudinal optic) splitting in vitreous SiO2. Many contributors have attempted to describe the origin of the C peak: (i) the 3-fold degenerate rigid cage vibrational mode of

TO2 units in tetrahedral glasses using the central force model (Sen and Thorpe, 1977; Galeener, 1979); (ii) the motion of Si in its oxygen cage in vitreous silica and both alkali and alkaline-earth silicate glasses (Galeener and Mikkelsen, 1981; Mysen et al., 1982a); (iii) the Si-O stretching involving oxygen motion in the Si-O-Si plane in alkali silicate glasses (McMillan et al., 1994). The C peak remains problematic because the band exhibits similar behaviour, a peak shift to lower wavenumbers, with both increasing temperature (McMillan et al., 1994; Kalampounias et al., 2006) and increasing pressure (or density) (Hemley et al., 1986).

Figure 2.6: (a) Peak centre of peak C as a function of composition. (b) Peak C intensity as a function of composition. The intensities of all spectra are normalized to the intensity of the C peak in sample Cs5.

−1 In vitreous SiO2, the 770 cm band demonstrates clear correlation with the average Si−O−Si bond angle, shifting to higher frequencies when the average angle decreases as demonstrated by the local- density approximation to density-functional theory (Giacomazzi et al., 2009). The asymmetry of the Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 24 peak is interpreted as being due to multiple overlapping peaks which can be modeled using Gaussian analysis (Kalampounias et al., 2006; Giacomazzi et al., 2009). These Gaussian bands were assigned to Si−O stretching vibrations (Giacomazzi et al., 2009) and more specifically to two different substructures (i) cristobalite-like and (ii) supertetraherdra configurations of bridged tetrahedra (Kalampounias et al., 2006).

As seen in Fig. 2.6(a) the peak centre shifts to lower frequency with increasing alkali content, which is indicative of an increase in the average intertetrahedral angle. This perhaps reflects a change in the substructures reflected in the alkali silicate glasses with increasing alkali content. Due to the different aforementioned interpretations it is difficult to assign the C peak to a specific vibration without conducting more experiments and simulations on more relevant compositions. As seen in Figure 2.6(b), the peak intensity decreases with increasing alkali content which is consistent with the assignment of the band to Si−O specific stretching vibrations which become attenuated at high modifier content (Mysen et al., 1982a; Matson et al., 1983).

2.4.4 The high frequency region (850-1250 cm−1)

The entirety of the 1085 cm−1 peak is commonly attributed to the diﬀerent abundances of Qn species which are governed by the amount of modiﬁer added to the network (Matson et al., 1983; Malfait, 2009). The shoulder on the high frequency side is most pronounced at low alkali concentration and no longer exists at compositions > 20 mol % Cs2O. This has led Matson et al.(1983) to conclude that this feature is characteristic of Q4 species and that the disappearance of this peak at higher Cs concentrations is indicative of the conversion of Q4 species to lower order Qn species. That being said, the high frequency side of the peak is associated with higher order Qn species (n=3, 4) whereas the low wavenumber side of the envelope is related to lower order Qn species (n=0, 1, 2) (Matson et al., 1983; Mysen and Frantz, 1992; McMillan and Wolf, 1995).

To a ﬁrst order, we can quantify the evolution of the high frequency envelope (850 - 1200 cm−1) by calculating its barycentre, χb, with increasing alkali content: P ω ωI(ω) χb = P (2.2) ω I(ω) where ω is the wavenumber (cm−1) and I(ω) is the normalized intensity for a given wavenumber. The barycentre is the frequency below and above which there are equal integrated areas. In Fig. 2.7 we show χb as a function of composition, it is apparent that there is clear correlation between the increase −1 in alkali content and the negative shift in χb which spans over 40 cm . This observation corroborates results of previous studies on alkali silicate glasses (Li, Na, K, Rb, Cs) which have observed a shift in the overall envelope to lower wavenumbers (Matson et al., 1983; McMillan, 1984; Mysen and Frantz, 1992; Malfait, 2009; Le Losq et al., 2014). The negative shift in the barycentre of the high frequency envelope is consistent with the notion of higher order Qn species becoming less abundant with increasing modiﬁer cation content (Rossano and Mysen, 2012, references therein). Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 25

Figure 2.7: The barycentre of the high frequency envelope of the binary caesium silicate glasses as a function of the composition. Note a negative shift of more than 35 cm−1.

Traditionally, the high frequency peak of alkali and alkaline-earth silicate glasses has been investigated using a curve-fitting model composed of Gaussian bands (Mysen et al., 1982a; McMillan, 1984; Mysen, 1995; Frantz and Mysen, 1995; Le Losq et al., 2014). Despite fitting peaks for the Qn species there are always additional peaks required to correctly model the high frequency envelope, these peaks have been assigned to a variety of different structural species. As mentioned previously, the number of bands was the minimum required to minimize χ2 and provide a random residual (Mysen et al., 1982a). Setting the FWHM to vary within a range, due to the probable similarity of the natural inherent linewidths, yielded a total of five (5) bands to fit all the spectra of this study. The same number of bands was recently used on sodium silicate glasses of a similar composition Na20 (Le Losq et al., 2014). The peak positions of our bands are reported in Table 2.2. At low Cs2O content (<20 mol. %), bands B through F are present while at higher Cs2O concentration bands A to E are present. The transition is approximately at the same composition as the transition from dominantly 4-membered to 3-membered SiO4 rings. The peak positions of the bands do not seem to vary systematically with composition, rather the bands appear to be situated at roughly the same frequency (± 20 cm−1) (see Nesbitt et al., 2016, this volume). This is very similar to the findings of Mysen et al.(1982a) and Matson et al.(1983) which described envelopes with multiple bands with peak centres which did not seem to vary with composition. There have been many interpretations concerning the significance of these Gaussian bands. Through- out this debate there are certain assignments upon which there is general agreement. We compared the data for the peak positions of our bands to the values typically assigned to similar bands in alkali silicate glasses using Gaussian curve-fitting (Brawer and White, 1975; Mysen et al., 1982a; McMillan, 1984; Mysen, 1995; Frantz and Mysen, 1995; Le Losq et al., 2014). The most intense band is located at ∼1100 −1 cm in most alkali silicate glass spectra containing 5-35 mol % M2O and has previously been assigned to the stretching of Q3 species (Brawer and White, 1975; McMillan, 1984; Mysen and Frantz, 1994). They also identified a band for Q2 units located at approximately 950 cm−1 (Brawer and White, 1975; Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 26

McMillan, 1984; Mysen and Frantz, 1994). A band at 1150 cm−1 corresponds to fully polymerized Q4 (Bell et al., 1968; Phillips, 1984). These band assignments appear to be consistent regardless of the nature of the modifier cation (Matson et al., 1983). Recent, X-ray photoelectron spectroscopy (XPS) work on both sodium and potassium silicate glasses have underlined the potential importance of alkali proximity to BO and the ways in which this may affect bonding (Nesbitt et al., 2011; Sawyer et al., 2012; Nesbitt et al., 2015). The abundances of BO and NBO have been quantified using O 1s XPS spectra (Nesbitt et al., 2011; Sawyer et al., 2012; Nesbitt et al., 2015). They clearly identify two distinct populations of BO, (i) oxygen bound to two silicon atoms (BO) and (ii) BO bound to two silicons and a modifier cation (M−BO). These M−BO are most likely going to be found near the channels where the modifier cations are predominantly located (Greaves et al., 1981). This is not the first mention of modifier cations clustering near BO, as they have been identified in molecular dynamics simulations of both sodium silicate and borosilicate glasses (Ispas et al., 2005; Pedesseau et al., 2015a,b). We speculate that the bands present in the high frequency envelope are related to the BO and M−BO, thus creating the possibility of having two distinct bands for each of the Qn species.

B C D E F A B C D E a) b)

Figure 2.8: The Gaussian curve-ﬁtting of a) Cs15 and b) Cs30. Note the change in the high frequency side of the envelope combined with the development of a new band at 927 cm−1.

In ﬁgure 2.8(a), bands E and F are both situated on either side of the typical band assignment for fully polymerized Q4 units at 1150 cm−1. We propose that there are two populations of Q4 species: those BO which have no alkali cations in their second coordination sphere (designated BO) and those which have one or more alkali cations situated in their second coordination sphere (designated M−BOn where n represents the number of alkalis). With increasing alkali content the proportion of M−BO to BO increases (Nesbitt et al., 2011), thus augmenting the signal for the Q4M and diminishing that of Q4. This is precisely the behaviour displayed by bands E (Q4M) and F (Q4) respectively, as the F band eventually disappears by 25 mol % Cs2O. We can observe the F band migrate towards band E with −1 −1 increasing Cs2O content with a splitting of 52 cm decreasing to 38 cm . Similar curve-ﬁtting was carried out on sodium silicate glass (25 mol % Na2O) and both high frequency bands (1143 and 1151 cm−1) were assigned to Q4 species, though they do not make the distinction of having two populations of these units (Le Losq et al., 2014). Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 27

Sample A (cm−1) B (cm−1) C (cm−1) D (cm−1) E (cm−1) F (cm−1)

Cs5 - 987 1062 1098 1146 1198 Cs10 - 993 1062 1096 1141 1188 Cs15 - 988 1062 1095 1137 1179 Cs20 - 970 1046 1095 1144 1182 Cs25 922 976 1051 1096 1136 - Cs30 927 990 1052 1096 1126 - Cs35 921 986 1048 1097 1138 -

−1 Table 2.2: The Raman shift, ωc (± 2cm ) for all the curve-ﬁt peaks.

The most prominent band present in the high frequency spectra is the ∼1100 cm−1 band as mentioned above. This corresponds well to the D band (∼1098 cm−1) in our caesium silicate glasses, which leads us to assign this band to the vibrations associated with the Q3 species as has been done previously for similar compositions (Brawer and White, 1975; McMillan, 1984; Mysen and Frantz, 1994; Le Losq et al., 2012; Le Losq et al., 2014). The C band located at ∼1050 cm−1 is undoubtedly the most controversial peak to be assigned, with multiple contributors ascribing it to a plethora of vibrations in a variety of silicate glasses (Mysen et al., 1982b; McMillan et al., 1992; Mysen, 1995; Le Losq et al., 2014). The main assignments are to (i) vibrations involving bridging oxygen atoms in structural units which do not need to be fully polymerized (Mysen et al., 1982b), (ii) vibrations of Si−O doublets associated with modifier cations (e.g. alkali and alkaline-earth metals) (McMillan et al., 1992), and (iii) the stretching T2 vibrational mode of SiO4 tetrahedra (Le Losq and Neuville, 2013), which corresponds to two oxygen atoms moving closer to the central Si atom while the two others oxygen atoms are moving away as demonstrated by vitreous SiO2 (Sarnthein, 1997). The peak centre for band C does not seem to vary dramatically with composition but its area does increase. This leads us to be skeptical of the previous assignments because there should be either no effect, or a decreased signal intensity with increased alkali content. Instead, we propose that the ∼1050 cm−1 is reflective of Q3M band since its intensity increases with increased alkali content. The idea of two structurally and vibrationally distinct populations of Q3 species has been previously 0 underlined in a comprehensive study of alkali silicate glasses (Q3 and Q3 )(Matson et al., 1983). The C band is also shifting away from the D band, increasing the splitting between them, indicating that the vibrational character of the two bands are increasingly dissimilar with composition. Similarly, bands A and B are located on both sides of the position normally ascribed to Q2 units (∼950 cm−1) (Fig. 2.8a). This appears to be another splitting of the band into the contributions of the Qn species bound to BO and those bound to BO and M−BO. Using a similar argument as that presented for the increase in area in Q4M species with increasing caesium content, we assigned peak A to Q2M and B to Q2. Overall, we have identified a total of six different bands with a maximum of five present in a given 4 3 spectrum. At low alkali content (≤25 mol % Cs2O), the bands for both Q and Q pairs are present as 2 well as the Q band. There are still large regions of the glass which are rich in SiO2 (quartz-like zones) where there is the possibility of having Q4 units which are isolated from any alkali which are present in pocket or small channel structures (Greaves et al., 1981; Bauchy and Micoulaut, 2011; Hodroj et al., Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 28

2013). These alkali pockets are small zones in the structure which are not connected to other alkali zones. Upon depolymerization these isolated pockets may become connected in order to form channel structures. The Q4M units are in closer proximity to the zones with alkalis present. The location of the Q3 species is near an alkali pocket, whereas, Q3M species are more likely to be found near multiple pockets, thus favoring M−BO bonding. As for the Q2 units present, the alkali concentration is still low enough, that even with their proximity to the pockets we still do not expect the BO to convert to 4 M−BO. At high alkali content (>25 mol % Cs2O), we observe the disappearance of the Q band which points to a conversion of all remaining Q4 to Q4M. On the low frequency side of the envelope we see the nascent Q2M band which increases in abundance with composition. Estimating the absolute abundances of the different Qn species in glasses and melts using spectral curve-fitting of the high frequency envelope has been employed by many previous contributors (Brawer and White, 1975; Mysen et al., 1982a; McMillan, 1984; Mysen, 1995; Frantz and Mysen, 1995; Le Losq et al., 2014). The techniques have either used band intensity in intensity normalized spectra or band area in area normalized spectra as a proxy for the abundance of the different Qn species.

Figure 2.9: The relative areas of the Gaussian bands used to described the diﬀerent populations of Qn species present in the glasses. The peaks for Qn and QnM species are grouped together in order to see the total change in the species with increasing Cs2O content.

In order to portray the changes observed in our glasses we have combined the contributions (areas) from Qn and QnM species and displayed these relative abundances as a function of the alkali content (Fig. 2.9). It is clear that there are almost systematic changes in the Qn species distribution, we observe quasi-linear trends for all species as a function of composition displayed by the relative decrease in Q4 units and an increase in both the Q3 and Q2 units. This behaviour correlates with the notion of depolymerization with increasing alkali content, breaking down higher order Qn species in order to create Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 29 lower order ones (Zachariasen, 1932). Though these trends appear to be consistent with depolymerization of the network they do not seem to be governed by the typical reaction 2 Q3 )−−−−* Q4+Q2 (Matson et al., 1983; McMillan, 1984; Mysen, 1995). This may point to the fact that this is not the only reaction taking place which governs the abundances of Qn species in alkali silicate glasses. It is important to note that the relative abundances as calculated in both our work and that of other contributors can only be seen as qualitative until the exact Raman cross-section of each individual Qn species is known (McMillan et al., 1992). Attempts have been made to correlate the data obtained from these abundances to that of other techniques (e.g. Stebbins, 1987) but assumptions need to be made concerning the conversion of one type of data to the other. Nonetheless, we consider that even if the relationship between the relative area and the real Qn species abundance is not one-to-one, the insight acquired from the relative areas is still valuable and correlate well with the trends displayed by the relative abundances presented in previous works and can be used as a proxy for the variation in the Qn species (Mysen et al., 1982a; McMillan, 1984; Mysen, 1995; Frantz and Mysen, 1995; Le Losq et al., 2014).

2.4.5 Insights from high-temperature Raman spectra

Figure 2.10: The barycentre calculated for the high frequency envelopes of the caesium silicate glasses and melts (Cs5, Cs10 & Cs15) at multiple temperatures. The same trends in the barycentre position are observed for both the composition and temperature.

Many diﬀerent Raman spectroscopic studies have been conducted on alkali silicate glasses and their high temperature melts (Mysen et al., 1982b; Mysen and Frantz, 1992, 1994; Maehara et al., 2005). Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 30

These efforts were focused on monitoring the changes in the high frequency envelope with increasing temperature in order to extract the speciation (Qn species distribution). As we did for the room temperature (RT) samples, we calculated the barycentre, χb, for the three samples at five different temperature steps (Fig. 2.10). At a given temperature, the samples follow the same trend as described previously for the room temperature samples, that is, χb shifts to lower frequency with increasing alkali content.

A similar eﬀect is also observed with increasing temperature, where the value χb decreases progressively with shifts on the order of ∼15-20 cm−1. The negative shift in the barycentre of the high frequency envelope is consistent with the notion of lower order Qn species becoming more abundant with increasing modiﬁer cation content at the expense of higher order Qn units (McMillan and Wolf, 1995; Rossano and Mysen, 2012, references therein).

B C DEF B C DEF a) b) Cs10 Cs10 T=640o C T=1140o C

Figure 2.11: Gaussian curve-ﬁtting and band assignments at diﬀerent temperatures for Cs10 at a) 640 ◦C and b) 1140 ◦C.

The Gaussian curve-fitting model employed for the room temperature spectra of our caesium silicate glasses was also used on the high temperature spectra in order to attempt to track the changes in network connectivity as a function of temperature. Obtaining Qn species distributions from glasses and following their evolution with temperature is important to our understand of the melting and crystallization processes, not to mention the implications it has for the glass transition in different silicate glass compositions. The high temperature spectra were also fit with the same total of five bands (Fig. 2.11). The band assignments are equivalent to those at room temperature as judged by the near-constancy of peak maxima (Table 2.3). There is more variation in the peak position of each band due to the inherent difficulties associated with taking measurements at high temperature but the general trends are very similar to those observed in the room temperature spectra: (i) the peak centres do not vary systematically with composition but rather tend to be located about a constant frequency (±25 cm−1), (ii) the splitting between related peaks (Qn and QnM) remains approximately constant (50±15 cm−1). The modeled relative abundances of the pairs of Qn species were calculated for Cs5, Cs10 and Cs15 for all temperatures (Fig. 2.12). We observe similar behaviour to the trends displayed by the relative abundances from the room temperature spectra (Fig. 2.9). A general observation at each temperature increment shows abundance of Q4 units is decreasing with composition whereas Q3 and Q2 units are Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 31 increasing. Once again, this is consistent with the theory that depolymerization increases with increasing alkali content, breaking down higher order Qn species in order to create lower order ones (Zachariasen, 1932). The most important feature is the decreased intensity of the pair of Q4 peaks and the concomitant increase in the two Q2 peaks. Apparently increased temperature favours the lower Qn numbers (Fig.

2.11). This transition is caused by the fact that sample Cs15 underwent melting (Tm) just below 900 ◦ C as shown in the theoretical calculations of the Cs2O−SiO2 phase diagram (Kim and Sanders, 1991, their Fig. 3). A similar change is observed in the room temperature spectra when going from Cs20 to Cs25 (Table 2.2). It seems that both increasing the alkali content and raising the temperature have comparable eﬀects on the shape and centre of mass (χb) of the high frequency envelope and in turn, on the areas of the diﬀerent Gaussian bands used to model the distribution of Qn species.

Figure 2.12: Area of the Qn species as a function of temperature and composition.

The act of melting, both in silicates and other materials, is one which is not particularly well un- derstood with many notable scientists investigating the problem over the past century (Born, 1939; Lennard-Jones and Devonshire, 1939; Eyring et al., 1958). We suggest that the creation of lower order Qn species when melting occurs is most likely part of the dissociation mechanisms which are at play when transitioning from a solid to a liquid. If the depolymerization reaction were to be considered in this respect, we would no doubt be forced to create free oxygen (O2– ) as a result of melting, which has been recently observed in sodium silicate glasses (Nesbitt et al., 2017a). This has long been described in the Toop-Samis model (Toop and Samis, 1962a,b; Ottonello and Moretti, 2004) and its recent imple- mentation has shown that polymeric models apply on binary silicate systems, showing that with both increasing temperature and alkali content, increases depolymerisation (Ottonello and Moretti, 2004). Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 32

Perhaps, the high frequency envelope of the high temperature Raman spectra could hold the key to unlocking an actual melting reaction through the information provided by the relative abundances of the diﬀerent structural units (Qn species).

Sample T ( ◦C) A (cm−1) B (cm−1) C (cm−1) D (cm−1) E (cm−1) F (cm−1)

Cs5 640 - 989 1045 1094 1145 1195 800 - 985 1040 1093 1148 1199 950 - 957 1026 1093 1149 1201 1060 - 964 1028 1090 1142 1192 1140 - 995 1051 1095 1146 1208

Cs10 640 - 998 1056 1094 1143 1192 800 - 980 1035 1095 1148 1197 950 - 994 1045 1095 1149 1208 1060 - 994 1053 1096 1145 1205 1140 - 956 1036 1091 1141 1203

Cs15 640 - 957 1024 1093 1148 1201 800 - 961 1031 1092 1145 1198 950 913 964 1035 1085 1129 - 1060 912 972 1036 1085 1131 - 1140 900 967 1032 1095 1157 -

−1 Table 2.3: The Raman shift, ωc (± 4cm ) for all the curve-ﬁt peaks of the high temperature spectra of Cs5, Cs10 and Cs15.

2.5 Conclusions

The Raman spectra of binary caesium silicate glasses displayed many trends from 5-35 mol % Cs2O. Our main conclusions from these results are:

1. With increasing alkali content:

(a) The behaviour of the Boson peak resembles that of previous studies on similar systems, with the peak centre shifting to higher frequencies and the FWHM increasing.

(b) The abundance of 3- and 4-membered SiO4 rings varies with composition and this is reﬂected in the relative intensities of the D1 and D2 bands. Increase in the proportion of 3-membered

SiO4 rings is a known densiﬁcation mechanism and the ratio of ID1/ID2 tracks extremely well with glass density and molar volume. (c) The high frequency envelope shifts to lower wavenumber with increasing Cs composition, as shown by both the barycentre trend and Qn species relative abundances. Thus, following the Chapter 2. Raman spectroscopy of Cs-Silicate Glasses and Melts 33

depolymerization reaction. (d) Splitting of Qn species bands was observed for all species present between a peak associated with BO and one associated with M−BO. (e) The splitting observed in both the Qn species bands and the defect bands (D1 and D2) are most likely related, as is the case in both silicate and germanate crystalline pyroxenes (Wang et al., 2001; Tribaudino et al., 2012). (f) It is worth noting that many of the important changes observed in the Raman spectra seemed

to be around the 20 mol % Cs2O concentration which corresponds to the transition from a rigid to a ﬂoppy network (Thorpe, 1985; Zhang and Boolchand, 1994; Micoulaut and Phillips, 2003).

2. With increasing temperature:

(a) The three diﬀerent compositions (Cs5, Cs10, and Cs15) displayed a negative shift in the high frequency envelope as shown by both the barycentre trend and the Qn species relative abundances. (b) Lower Qn species appear to be created, lending credence to the possibility that the melting reaction taking place is similar in nature to the depolymerization reaction: O2– + BO −−→ 2 NBO– . These ﬁndings agree with recent studies using the Toop-Samis model (Ottonello and Moretti, 2004). Chapter 3

The structure of high-silica alkali silicate glasses: Revisited

This chapter will be submitted to the Journal of Non-crystalline solids and has been the foundation of the following publications:

• Nesbitt, H. W., Henderson, G., Bancroft, G. M., O’Shaughnessy, C. (2017b) Electron densities over Si and O atoms of tetrahedra and their impact on Raman stretching frequencies and Si-NBO force constants. Chemical Geology 461, 65–74

• Bancroft, G. M., Nesbitt, H. W., Henderson, G. S., O’Shaughnessy, C., Withers, A. C., Neuville, D. R. (2018) Lorentzian dominated lineshapes and linewidths for Raman symmetric stretch peaks (800-1200 cm−1) in Qn (n=1-3) species of alkali silicate glasses/melts. Journal of Non-Crystalline Solids 484, 72–83

• Nesbitt, H. W., O’Shaughnessy, C., Henderson, G. S., Michael Bancroft, G., Neuville, D. R. (2019) Factors aﬀecting line shapes and intensities of Q3 and Q4 Raman bands of Cs silicate glasses. Chemical Geology 505, 1–11

I have carried out all of the experiments at the Raman laboratory of the Geomaterials group at the Institut de Physique du Globe de Paris, France, did the data processing, ﬁgure preparation and wrote the article. All other authors have provided training and given feedback during the writing process.

34 Chapter 3. The structure of alkali silicate glasses 35

Abstract

The Raman spectra of silicate glasses containing 0 to 30 mol % M2O (M = Li, Na, K, Rb, Cs) have been fit successfully with line shapes of dominantly Lorentzian character for the Q3 species, allowing quantification of Q3 and Q4 species intensities. This differs from the practice of using Gaussian lineshapes 4 which have been used for the past four decades. The intensity of the Q species A1 symmetric stretch is exceptionally weak in vitreous silica (v-SiO2) but it increases dramatically with addition of small 4 amounts of M2O to the glass. We propose that alkalis, where in close proximity to BO of Q species, promote the polarizability of Si−O bonds of the Q4 tetrahedra and these “primed” Q4 species (Q4-p) produce a strong signal. Thus, there are, two variants of Q4 species, a Q4-p species which produces a strong Raman signal due to proximity to alkali cations and an “unprimed” species (Q4-u) which yields a very weak Raman intensity due to the absence of alkali cations in close proximity. 3 Similarly, there is an increase in asymmetry in the Q band, with increasing M2O content which appears to result from the weakened Si−O force constants of some Q3 bands due to charge transfer via M−BO bonds. We cite evidence from Si 2p and O 1s X-ray photoelectron spectroscopic studies which demonstrate that the electron density over Si and BO atoms of Q4 species increases with alkali content. With charge transfer to tetrahedra, the negative charge accumulates preferentially on Si atoms 3 thus decreasing Si−O Coulombic interactions, weakening Si−O force constants, and shifting the Q A1 symmetric stretch vibrational frequencies to lower values (e.g., from ∼ 1100 cm−1 to ∼ 1050 cm−1). The fraction of affected Q3 species increases with alkali content, as does the Q3 peak asymmetry. We believe that this extends through all the Qn species and postulate that there are potentially multiple vibrational modes for each Qn species which are dictated by their proximity to network modifier cations.

The proportions of 4- and 3-fold SiO4 rings are estimated by using the relative intensities of the

D1 and D2 vibrational bands in each spectrum. An increase in the abundance of 3-fold SiO4 rings is apparent and confirms previous studies which proposed such a mechanism to be present with increasing alkali content. This finding is consistent with the depolymerization of the network and is also reflected in the results for the abundances for the Qn species present in alkali silicate glasses.

3.1 Introduction

Silicate glasses are widely used as structural models for their molten counterparts. Knowledge of the structure of silicate melts is important in order to understand the chemical and physical properties of magmas and melt systems, such as viscosity and density, which occur naturally within the earth and dominate geological processes (Henderson, 2005). As described in the previous chapter, the most widely accepted model for the structure of pure oxide glasses is that of Zachariasen(1932), which describes vitreous SiO2 (v-SiO2), as a continuous three-dimensional random network of silicon-oxygen tetrahedra which are linked together by bridging oxygen (BO) bonds.

The structure of v-SiO2 has been studied in considerable detail (e.g. Zachariasen, 1932; Warren and Biscce, 1938; Mozzi and Warren, 1969; Sharma et al., 1981; Galeener, 1982; Sarnthein, 1997; Pasquarello et al., 1998). The vibrational spectrum of v-SiO2 was initially described by Bell, Dean and contributors (Bell et al., 1968; Bell and Dean, 1970). The Raman spectrum is comprised of many notable features which we will discuss individually (Fig. 3.1).

• The spectrum is dominated by a large broad peak (also called band) named the Main band, which Chapter 3. The structure of alkali silicate glasses 36

−1 is located at ∼ 440 cm and is commonly associated to the vibrations of BO in large SiO4 rings (n>5) (Sharma et al., 1981).

• There are two weak, sharp, polarized peaks at 492 cm−1 and 606 cm−1 called the “Defect bands”. These are labeled as the D1 (492 cm−1) and D2 (606 cm−1) bands. These bands correspond to

breathing modes of oxygen in small (D1: 4-membered and D2: 3-membered) planar SiO4 rings (Sharma et al., 1981; Galeener, 1982; Barrio et al., 1993; Pasquarello et al., 1998). The proportion of these two peaks varies upon densiﬁcation (Sonneville et al., 2012; Deschamps et al., 2013). In these studies, it is noted that the proportion of 3-membered rings increases with respect to 4-membered rings as a function of densiﬁcation.

• The low intensity, asymmetrical C peak located at ∼ 770 cm−1 and has been assigned to many

diﬀerent origins: (i) the 3-fold degenerate rigid cage vibrational mode of TO2 units (Sen and Thorpe, 1977); (ii) the motion of Si in its oxygen cage (Galeener and Mikkelsen, 1981); (iii) the Si−O stretching involving oxygen motion in the Si−O−Si plane (McMillan et al., 1994).

−1 −1 • The high frequency, low intensity doublet (1060 cm and 1200 cm ) corresponds to the A1 and

T2 vibrational modes of SiO4 tetrahedra (Sarnthein, 1997). The A1 mode is due to the in-phase motion of all oxygen atoms toward the Si atom in the tetrahedron, whereas, the triply degenerate

T2 modes are due to two oxygen moving toward the central Si atom while the other two oxygen move away (Sarnthein, 1997).

D1 Main

Boson D2 Qn C T 2 A1

Figure 3.1: The spectra of v-SiO2 and Na5. They are shown here in order to highlight the diﬀerences which occur when a small amount of alkali cations are added to the silicate glass network. Additionally, the main regions of interest are labeled and further discussed in the text. Chapter 3. The structure of alkali silicate glasses 37

Binary alkali and alkaline-earth silicate glasses and their melts (MO−SiO2 for M = Mg, Ca, Sr, Ba

& Fe or M2O−SiO2 for M = Li, Na, K, Rb & Cs) differ from pure oxide glasses due to the presence of network modifiers (M) which break the BO bonds (Si−O−Si) (Henderson, 2005). The network modifiers create non-bridging oxygen (NBO) and Qn species (where n is the number of BO atoms per silicon-oxygen tetrahedron and ranges from 0 to 4). In order to take these network modifiers into consideration, a new structural model was developed by Greaves et al.(1981) on sodium silicate glasses. The modified random network (MRN) model proposed that the alkali silicate glass network would be an extension of the crystalline counterpart, being made up of two interconnected sub-lattices: (i) a continuous, disordered SiO2 network akin to that in v-SiO2, (ii) an interpenetrating alkali sub-lattice. A consequence of this model is cation clustering around NBO atoms leading to the formation of percolation channels (Greaves et al., 1981; Greaves, 1985).

Additional models for the structure of silicate glasses have been developed using percolation theory in a central-force model (e.g. Sen and Thorpe, 1977; Thorpe, 1985; Micoulaut and Phillips, 2003). These theories make use of the interatomic interactions and mechanical constraints similar to those used in the analysis of the stability of macroscopic trusses (Micoulaut, 2008). If a network contains a high amount of cross-linkages (i.e. high connectivity), as is the case in v-SiO2 where these linkages are BO, it is found to be rigid. Whereas, if the network has low connectivity, it is considered to have a more ﬂexible structure and would be described as ﬂoppy (Micoulaut, 2008). In the case of alkali silicate glasses, Micoulaut and Phillips(2003) pointed out that areas of large connectivity corresponded to the silica-rich regions, whereas those with low connectivity were associated to depolymerized regions of the glass.

Many groups have investigated the structure of alkali silicate glasses using a wide range of analyti- cal techniques including nuclear magnetic resonance spectroscopy (NMR), Raman spectroscopy, X-ray absorption near-edge structure (XANES) spectroscopy, Extended X-ray absorption ﬁne structure spectroscopy (EXAFS) and X-ray photoelectron spectroscopy (XPS) (Matson et al., 1983; McMillan, 1984; Greaves, 1985; Dupree et al., 1986; Maekawa et al., 1991; Henderson, 1995; Nesbitt et al., 2011; Sawyer et al., 2015; O’Shaughnessy et al., 2017) and others via molecular dynamic (MD) simulations (Huang and Cormack, 1991; Du and Cormack, 2004; Du and Corrales, 2006; Gedeon et al., 2010; Bauchy and Micoulaut, 2011).

Raman spectroscopy of the symmetric stretch region (800-1200 cm−1) of alkali and alkaline earth silicate, and aluminosilicate glasses and melts may be used to identify the Qn species (n=0-4) and potentially may be used to determine their abundances over a wide temperature range (O’Shaughnessy et al., 2017). We investigated the Raman behaviour of the diﬀerent Qn species for a variety of alkali silicate glasses in order to see if they changed with alkali type and concentration (Fig. 3.1). We found that the positions, as shown in Fig. 3.2(a) and (b), are remarkably similar irrespective of which alkali was present for both crystals and glasses. The Raman shift of the Qn species decreases such that Q4 is located at a higher wavenumber than Q0. The average values are as follows: Q4 ∼ 1200 cm−1;Q3 ∼ 1098 cm−1;Q2 ∼ 950 cm−1;Q1 ∼ 900 cm−1; and Q0 ∼ 850 cm−1. These wavenumber values form the basis for our subsequent Qn species assignments as outlined in O’Shaughnessy et al.(2017), Bancroft et al.(2018) and Nesbitt et al.(2019). Chapter 3. The structure of alkali silicate glasses 38

Figure 3.2: (a) Illustrates the relation between the alkali oxide content of some crystals, glasses and melts, and the wavenumber of each Qn species observed in these phases as reported by McMillan(1984). The solid and dashed lines represent respectively the least squares best fits to the crystal and to the glasses/melt data. (b) The frequencies of the Qn species reported by McMillan(1984). The error bars indicate ± 25 cm−1 and are taken as the approximate uncertainty associated with each frequency based on the data of (a) and comments by (McMillan, 1984). The straight line is the linear least squares best fit to the data. This figure is adapted from (Nesbitt et al., 2017b).

Many previous studies have conducted Raman spectroscopy on a variety of alkali silicate and alkaline earth silicate glasses but they did not ﬁt the high frequency area, that has been attributed to NBO vibrations associated with diﬀerent Qn species (e.g. Brawer and White, 1975; Matson et al., 1983). The most complete description of low alkali silicate glasses was produced by Matson et al.(1983) where they Chapter 3. The structure of alkali silicate glasses 39

compared Raman spectra ranging from 5-30 mol. % M2O (M=Li, Na, K, Rb & Cs). They described the entire range of wavenumbers from 100-1400 cm−1 and vibrational assignments for all major peaks. They discussed the importance of the 1100 cm−1 band, characteristic of the Si-NBO stretching modes of Q3 species, and concluded that it is a result of two distinct Q3 species which become increasingly similar with increasing alkali content which in turn depolymerizes the silicate network (Matson et al., 1983). Additionally, they assigned the 950 cm−1 band to Q2 species and show that intensity of this band increases with alkali content. They mention that by comparing the relative intensities of the 950 cm−1 and 1100 cm−1 bands, they have shown that lithium ions have the greatest tendency to form Q2 species at low alkali concentrations and that this tendency decreases with increasing alkali cation size (Matson et al., 1983). Subsequent studies attempted to fit spectra using Gaussian bands which did not bear any real physical significance (McMillan, 1984; Mysen and Frantz, 1992; Mysen, 1995; Frantz and Mysen, 1995). The vast majority of previous studies did not attempt to control the linewidth of their fitted peaks which led to a wide array of possible fits and ambiguities regarding the interpretation of their spectra. Recent O 1s

XPS results demonstrated that the linewidths are controlled by the SiO4 symmetric stretch of the ion state and that these linewidths are similar for all silicate phases (Bancroft et al., 2009). As proposed in n Bancroft et al.(2018), the Raman linewidths for the Q species (resulting from SiO4 symmetric stretch) in Raman spectra of silicate crystals, glasses and melts must also be similar. In a previous study on crystalline silicates, phosphates and sulfates Nesbitt et al.(2018) listed some important concepts which are relevant to the data processing and interpretation of the Qn species area in Raman spectra of glasses and melts as described in Bancroft et al.(2018).

1. All Qn species peaks of alkali and alkaline earth crystals are dominantly Lorentzian at all temperatures up to the melting point.

2. The Raman shifts are well-deﬁned for each Qn species and have remarkably similar temperature dependences regardless of the nature of the counter cation.

3. The full widths at half maximum (FWHM) are similar for all Qn species, and all Qn species display a similar temperature dependence.

4. The Lorentzian lineshapes, along with the temperature trends of the Raman shifts and FWHM can be rationalized using well established theories.

For instance, Richet et al.(1996, Table A4) measured the Lorentzian linewidth of the Q 2 (∼950 −1 −1 ◦ −1 cm ) band of crystalline Na2SiO3 as 6 cm at 25 C broadening to 39 cm near its melting point 1348 ◦C. The full width at half maximum (FWHM) increases with temperature, whereas its position decreases with increased temperature. The important aspects to note are that, even with increasing disorder (transition from solid to liquid) the lineshapes remain Lorentzian and only broaden as seen in the FWHM. Thus, in theory, both Infrared and Raman lineshapes of silicate glasses are Lorentzian (Loudon, 1963; Eﬁmov, 1999) but nearly no one has used Lorentzian lineshapes in the past 40 years. The argument which is typically made for the use of Gaussian lineshapes is that: a distribution of local environments could lead to a Gaussian distribution of Lorentzian lines (e.g. McMillan, 1984; Mysen and Frantz, 1992). To this eﬀect, we believe that a more realistic way of describing such environments would be by using Lorentzian lineshapes which are broadened by a Gaussian component (Bancroft et al., 2018; Nesbitt et al., 2019). Chapter 3. The structure of alkali silicate glasses 40

In order to evaluate the feasibility of fitting alkali silicate glass spectra with different band shapes, Bancroft et al.(2018) used various combinations of Gaussian, Lorentzian, Voigt or pseudo-Voigt profiles (see below for discussion of the different types) to fit a range of compositions with differing alkali type. At low alkali content (5 and 10 mol. % M2O), Voigt profiles with 92-98% Lorentzian components produced the best fit for Q3 species peaks and the overall high frequency envelope (Bancroft et al., 2018). While at higher alkali content (>30 mol. % M2O), Voigt profiles with 60-68% Lorentzian components were favoured to reproduce the Q3,Q2 and Q1 signals (Bancroft et al., 2018). Using this technique, Nesbitt et al.(2019), fit caesium silicate glasses (5-30 mol. % Cs 2O) and found that the % Lorentzian of the pseudo-Voigt functions used decreased with increasing Cs2O content from 95% in Cs5 to 80% in Cs30 which was consistent with the preliminary findings of Bancroft et al.(2018). Additionally, Nesbitt et al. (2019) found that the linewidths of all the pseudo-Voigt peaks were between 40 and 56 cm−1, again in agreement with Bancroft et al.(2018). The assignments of peaks in the high frequency envelope have also been contentious because there are typically more bands present than expected Qn species (e.g. McMillan, 1984; Mysen and Frantz, 1992). As mentioned in O’Shaughnessy et al.(2017), recent studies have shown that there exist multiple energetically distinct types of BO in potassium silicate glasses which arise from the interactions between alkali atoms and the second coordination sphere of BO (Nesbitt et al., 2011, 2017b). This concept has been seen in sodium metasilicate crystals where two distinct types of BO are present, those coordinated to a sodium atom and those which are not (Ching et al., 1983). Bridging oxygens have increased electron densities due to these ionic interactions with alkali atoms (M). Nesbitt et al.(2017b, their Fig. 5 (a) and (b)) explained that electronic charge must be redistributed across the SiO4 tetrahedron, through molecular orbitals. The Si atoms gain the majority of this which weakens the Coulombic force between Si and all four O atoms for both BO and NBO. This in turn, diminishes the Si−O bond strengths, associated force constants and Si−O vibrational frequencies (Nesbitt et al., 2017b). Therefore, tetrahedra which are affected by M−BO interactions are shifted to lower frequency compared to tetrahedra which are devoid of these interactions (Bancroft et al., 2018). We believe that these “extra peaks” in the high frequency region of the Raman spectra are manifestations of Qn species with varying degrees of M−BO interactions as shown in our previous studies (O’Shaughnessy et al., 2017; Bancroft et al., 2018; Nesbitt et al., 2019). Though the high frequency region is by far the most studied portion of alkali silicate glass Raman spectra, another area of interest exists at ∼ 490-600 cm−1 (Fig. 3.1). The variation in the in-phase pure

-O- breathing motions in the 3, 4-membered SiO4 rings is no doubt reflected in the behaviour of the D1-D2 peaks in alkali silicate glass systems (Hehlen and Simon, 2012). We have already shown that this relationship exists for caesium silicate glasses from 5-30 mol. % Cs2O in O’Shaughnessy et al.(2017) and should be present for all alkali silicate glasses. The work of Matson et al.(1983) also mentions the dependence of D1 band (∼ 500 cm−1) position on alkali type which they interpreted as reflecting the effect caused by the alkali cation size on the silicate structure. Furthermore, they discuss the relative insensitivity of the D2 band (∼ 600 cm−1) position on both alkali type and content and suggest that it results from a localized vibrational mode common to all alkali silicate glasses that may be characteristic of vibrationally isolated defect structures or 3-membered rings (Matson et al., 1983). In this paper, we aim to generalize and extend the findings of O’Shaughnessy et al.(2017), Bancroft et al.(2018) and Nesbitt et al.(2019) concerning the presence of M −BO interactions, linewidths and lineshapes of the Qn species to all low-alkali silicate glass samples with compositions ranging between Chapter 3. The structure of alkali silicate glasses 41

5-30 mol % M2O (M=Li, Na, K, Rb, Cs). Additionally, we hope to link them with the behaviour of the “Defect bands” which give an indication of the changes in small SiO4 ring statistics present in the glasses. A combination of these two approaches will be used to infer important conclusions regarding the structure of the alkali silicate glass network with varying alkali type and content.

3.2 Methods

(a)

(b)

Figure 3.3: (a) The measured values for the densities of the alkali silicate glasses. (b) A closer look at the densities of the lithium, sodium and potassium silicate glasses. It is important to observe the apparent density cross over which occurs between 20 and 25 mol. % M2O for the sodium and potassium silicate glasses. Note that the errors were calculated by using ten independent density measurements for each sample and these errors lie within the symbols. Chapter 3. The structure of alkali silicate glasses 42

3.2.1 Glass synthesis

For this study we have synthesized a suite of alkali silicate glasses with concentrations ranging from

5-30 mol % M2O (M=Li, Na, K, Rb, Cs) in 5 mol % increments. The samples are named by the alkali type and concentration (e.g. Li5 for 5Li2O-95SiO2) and will be referred to in this way for the remainder of the article. These samples were prepared by mixing stoichiometric amounts of M2CO3 and SiO2 powders (99% purity) in approximately 3g batches. We then placed the mixtures in Pt crucibles and calcined them by gradually increasing the temperature from 200 ◦C to 900 ◦C over a period of 6 hours, the samples then stayed at this temperature for 24 hours in order to remove the CO2. The crucibles were then placed in a drop-down glass making furnace and heated to approximately 200 ◦C above their respective melting temperature for 1 hour and then quenched in air by placing the base of the crucible in water (O’Shaughnessy et al., 2017). The glasses were then mechanically crushed under ethanol in a mortar and pestle, the ethanol was allowed to evaporate and the mixture underwent the melting and quenching process for a second time. All glasses showed no macroscopic signs of defects or crystals and were transparent. The samples densities were measured using a Berman density balance which compared the densities in air and in toluene (which was calibrated for the temperature). Each density value is a result of 10 independent measurements which are averaged (see Table 3.2 and Fig. 3.3). The densities have been compared to literature data and are consistent with previous ﬁndings (Tischendorf et al., 1998; Doweidar et al., 1999).

We then calculated the molar volume (Vm) using stoichiometries with an equivalent amount of oxygen PN (i.e. M4O2 and SiO2) using the relation: Vm= i=1 xiMi/ρmix where xi is the proportion of phase i, Mi is the molar mass of i and ρmix is the measured density of the sample (Fig. 3.10(a)).

3.2.2 Raman spectroscopy

Experimental measurements

We conducted Raman spectroscopy measurements on the alkali silicate glasses at the Raman laboratory of the Geomaterials group at the Institut de Physique du Globe de Paris (IPGP). The spectra were obtained using the T64000 Jobin-Yvon triple grating Raman spectrometer equipped with a confocal system, a 1024 CCD detector cooled by liquid nitrogen and an Olympus microscope (Le Losq et al., 2012; O’Shaughnessy et al., 2017). A Coherent 70-C5 Ar+ laser with a wavelength of 488 nm and a laser power of 100 mW on the samples. No degradation of the samples due to the laser was observed. The resolution of the measurements was better than 0.5 cm−1 and scans were taken in triple-subtractive mode (which averages three spectra per sample). The peak centre values reported in all tables correspond to peak maximums. The spectra were area normalized in order to be compared and the background continuum was removed using a polynomial spline (O’Shaughnessy et al., 2017). This spline is generated using a single pass binomial smoothing ﬁlter (Marchand and Marmet, 1983; Bancroft et al., 2018) in order to determine a convex hull and then using the lower portion of the hull as nodes for a cubic spine (using the CubicSpline() function in the SciPy library).

The ﬁtting of Raman spectra

A brief discussion on the nature of diﬀerent lineshapes is required as it is a main point in our n analysis of the Q species area (controlled by SiO4 symmetric stretch) in the spectra of our alkali silicate glasses. As mentioned in the introduction the use of Voigt functions is a better approximation to the Chapter 3. The structure of alkali silicate glasses 43 natural lineshapes of Raman vibrations in alkali silicate glasses. The Voigt function is described as the convolution of Gaussian (normal distribution) and Lorentzian (Cauchy distribution) functions and is an important descriptor of symmetric features in multiple types of X-ray spectroscopies and diﬀraction (e.g.

Sanchez-Bajo and Cumbrera, 1997; Ida et al., 2000). Thus, the Voigt function (fV ) is deﬁned by the following equations:

1/2 1/2 1/2 fV (x;ΓG, ΓL) = (2/ΓG)[(ln 2)/π] × K[2(ln 2) x/ΓG, (ln 2) ΓL/ΓG], (3.1) where ΓG and ΓL are the FWHM of the Gaussian and Lorentzian components respectively. And;

∞ Z K(x, y) = (y/π) exp(−t2)/[y2 + (x − t)2]dt (3.2) −∞ = <[w(x + iy)],

w(z) = exp(−z2) erfc(−iz), (3.3) where K(x, y) is usually referred to as the Voigt function, w(z) is a scaled complex error function (Faddeeva function), erfc(z) is the complementary complex error function and <[z] is the real part of z (from Ida et al., 2000). Although it is convenient that K(x, y) is specified only by the variable y, it is numerically unpleasant to solve as it requires both x → ∞ and y → ∞ to properly reproduce the Lorentzian profile. Due to this inconvenience, a numerical approximation to the Voigt profile, called the pseudo-Voigt function (fpV ), is often used instead. The pseudo-Voigt function uses a linear combination of Gaussian and Lorentzian components rather than their convolution (as is the case in the classical Voigt function) and is defined as follows (from Ida et al., 2000):

fpV = (1 − η)fG(x; γG) + ηfL(x; γL), (3.4) with η deﬁned as the mixing parameter (0 < η < 1) and where fG(x; γG) and fL(x; γL) are the Gaussian 1/2 and Lorentzian functions (with FWHM Γ = 2(ln 2) γG = 2γL) described as:

1/2 2 2 fG(x; γG) = (1/π γG) exp(−x /γG), (3.5) and 2 2 −1 fL(x; γL) = (1/πγL)(1 + x /γL) . (3.6)

The mixing factor, η, dictates the proportion of each component (1=Gaussian or 0=Lorentzian) to use, and the effect on the distribution is shown in Fig. 3.4. In the remainder of the paper, we will refer to η as % Lorentzian (where % Lorentzian = 100 ∗ η) for practicality and consistency with our previous studies (Bancroft et al., 2018; Nesbitt et al., 2019). The work of Efimov(1999) discusses the crucial role of band shapes (Lorentzian, Gaussian and intermediate functions) in our understanding of the processes which give rise to vibrational signals perceived by Raman and IR spectroscopies. In particular, the shortcomings of using Gaussian lineshapes is due to the fact that the wings of Gaussian contours are steeper than those of Lorentzian and convolution (Voigt, pseudo-Voigt and Pearson) contours with similar half widths at half maximum (HWHM = 1/2(FWHM)) (Efimov, 1999). Chapter 3. The structure of alkali silicate glasses 44

Gaussian

Lorentzian

Figure 3.4: The variation in pseudo-Voigt lineshapes at a constant Γ=2 (FWHM) and incremental mixing factor η ranging from 0 (Gaussian) to 1 (Lorentzian). We highlight the diﬀerence in the tails of the distributions and discuss these implications in the text.

−1 The high frequency area (800-1300 cm ) attributed to the vibrations of the SiO4 symmetric stretch of the different Qn species was fit using both Gaussian (Q4) and pseudo-Voigt (Q3,2,1) line shapes following Bancroft et al.(2018) and Nesbitt et al.(2019). A Levenberg-Marquardt non-linear least squares optimization algorithm was implemented using the LMFIT Python package (Newville et al., 2014). The linewidths were allowed to vary within a range of values as described in Bancroft et al. (2018) and Nesbitt et al.(2019). The best fit was selected using a typical χ2 parameter and the residuals calculated as the difference between the fit values and the experimental values. The fits are shown in Fig. 3.17-3.21 and all pertinent values (peak position, FWHM, area, % Lorentzian and vibrational assignment) are reported in Tables 3.3 through 3.7. The vibrational assignments follow Nesbitt et al. (2019) and are labeled as: (i) Qn-1 for species with no M−BO bonds; (ii) Qn-2 for species with one M−BO bond; and (iii) Qn-3 for species with two M−BO bonds.

3.3 Results

3.3.1 Density and molar volume of alkali silicate glasses

The density of the alkali silicate glasses increases with alkali content as observed in previous studies by Tischendorf et al.(1998) and Doweidar et al.(1999) and are reported in Table 3.2 and plotted in Fig. Chapter 3. The structure of alkali silicate glasses 45

3.3. There is an eﬀect of the relative atomic mass of the alkali versus that of silicon but it may be taken into consideration using the molar volumes of the glasses (Fig. 3.10a). These molar volumes behave very similarly to the densities (increase with increasing M2O content) except for the lithium bearing glasses. Due to the small atomic radius we believe that lithium occupies diﬀerent spaces within the glass network than the other alkalis and this will be addressed in further detail below.

3.3.2 Raman spectra of alkali silicate glasses

The spectra of the alkali silicate glasses can be divided into ﬁve main regions of interest: the Boson peak (∼ 0-100 cm−1), the Main band (∼ 200-460 cm−1), the D1-D2 peaks (∼ 495 cm−1 and ∼ 600 cm−1 respectively), the C peak (∼ 750-800 cm−1) and ﬁnally the Qn-species envelope (∼ 800-1300 cm−1). For this article we will focus on two areas of the spectra, the D1-D2 peaks and the Qn-species envelope. In doing so, we will be able to compare the behaviour of small rings to that of the Qn species and the implications for the local structure of alkali silicate glasses. The maximum peak positions of these regions of interest are listed below in Table 3.1.

The behaviour of the D1 and D2 bands (495 and 600 cm−1)

The D1-D2 peaks behave similarly across the different types of alkali elements but appear to be affected by the amount of alkali (M2O mol. %) present. For the samples with 5 mol. % M2O both peaks are distinct and at similar positions (D1 ∼ 495 cm−1 and D2 ∼ 600 cm−1). As the amount of alkali is increased, the intensity of the D2 peak increases while that of the D1 peak decreases. Additionally, the positions of the D1 and D2 peaks shift towards each other (D1 to higher wavenumber and D2 to lower wavenumber), thus decreasing the splitting (defined as: ∆D = |ωD1 − ωD2|) between the two. At high alkali contents (∼30 mol. %), the two peaks appear to merge and they are indistinguishable (See Table 3.1). The lithium silicate glasses are the most difficult to characterize as the peaks are less pronounced and merge at much lower alkali content. This is no doubt because of the uniqueness of lithium which has a very small ionic radius and relatively high charge, thus we will be discussing the general findings for the D1-D2 peaks of the sodium to caesium silicate glasses in more detail in the discussion below.

The symmetric stretch region (Qn species: 850-1300 cm−1)

The behaviour of the Qn-species envelope is also similar across all types of alkalis. We observe, that with increasing alkali content, the high wavenumber shoulder (∼ 1195 cm−1) begins to decrease in intensity and a new peak forms on the low wavenumber side (∼ 950 cm−1) of the envelope. The position of the high intensity peak (∼ 1095 cm−1) assigned to the Q3 band does not shift with additional alkali content. A description of the behaviour of the high frequency envelope for each alkali silicate glass series is included below and the ﬁtting results are listed in Tables 3.3 to 3.7. Lithium silicate glasses (Table 3.3) The spectra of the lithium silicate glasses are characterized by the presence of low intensity shoulders on both sides of the high intensity band located between 1075-1089 cm−1. The relative intensity of the intense band increases with increasing lithium as does the low wavenumber peak (∼ 950 cm−1). The high wavenumber shoulder (∼ 1150 cm−1) decreases and is no longer discernible at approximately 20 4 4 mol. % Li2O which coincides with the transition from two distinct Q peaks to only one (Q -2 located at (1144 cm−1)). The ﬁts for the lithium silicate glasses reveal a very unique feature, which is the presence Chapter 3. The structure of alkali silicate glasses 46 Lithiumsilicates

Figure 3.5: The Raman spectra of the lithium silicate glasses. Chapter 3. The structure of alkali silicate glasses 47 Sodiumsilicates

Figure 3.6: The Raman spectra of the sodium silicate glasses. Chapter 3. The structure of alkali silicate glasses 48 Potassiumsilicates

Figure 3.7: The Raman spectra of the potassium silicate glasses. Chapter 3. The structure of alkali silicate glasses 49 Rubidiumsilicates

Figure 3.8: The Raman spectra of the rubidium silicate glasses. Chapter 3. The structure of alkali silicate glasses 50 Caesiumsilicates

Figure 3.9: The Raman spectra of the caesium silicate glasses. Chapter 3. The structure of alkali silicate glasses 51

Sample D1 (cm−1) D2 (cm−1)Qn (cm−1)

Li5 484 598 1079 Li10 489 595 1075 Li15 485 597 1085 Li20 490 591 1085 Li25 481 593 1082 Li30 564 - 1089

Na5 490 597 1091 Na10 493 597 1093 Na15 516 592 1093 Na20 529 590 1094 Na25 543 590 1098 Na30 563 - 1100

K5 490 597 1100 K10 498 599 1099 K15 514 595 1098 K20 529 591 1103 K25 545 590 1103 K30 575 - 1103

Rb5 489 602 1101 Rb10 493 599 1099 Rb15 497 598 1101 Rb20 516 596 1100 Rb25 526 592 1105 Rb30 529 591 1102

Cs5 490 597 1101 Cs10 491 597 1097 Cs15 497 596 1099 Cs20 506 595 1099 Cs25 512 591 1099 Cs30 525 590 1100

Table 3.1: The maximum positions for the D1, D2 and Qn peaks of the alkali silicate glasses. Chapter 3. The structure of alkali silicate glasses 52 of Q2 species (∼ 950 cm−1) at very low alkali content (5 mol. %). This is at lower alkali concentration than any of the other glass series. The Q2 peak (∼ 950 cm−1) area also increases with increasing lithium content to a maximum of 14.5% in Li30, which is, substantially higher than any other value throughout the alkali silicate glasses in this study. Sodium silicate glasses (Table 3.4) Sodium silicate glasses display a shoulder (∼ 1150 cm−1) on the high wavenumber side of the high frequency envelope at low Na2O content. As sodium is added this shoulder begins to decrease in intensity and is no longer visible at 20 mol. %, similar to the lithium silicate glasses. The main difference between the lithium and sodium series is that the development of the low wavenumber shoulder (∼ 950 cm−1) begins at slightly higher alkali content, as the Q2 peak is only weakly seen in Na10. But, as in the 2 lithium series the area of the Q peak increases with Na2O concentration to a maximum value of 8.4% in Na30. Potassium silicate glasses (Table 3.5) In the potassium silicate glasses we observe the same trend for the high wavenumber shoulder (∼ 1150 cm−1) as in both the lithium and sodium series. It is very pronounced for K5 and K10 but gradually begins to decrease in area in sample K15. We see the Q4-1 peak is no longer present in sample K20 as was the case with the lithium and sodium samples (Li20 and Na20 respectively). Sample K30 also now loses the signal from the Q4-2 peak and has a slightly broader Q3-1 peak. The Q2 peak (∼ 950 cm−1) appears at 15 mol. % K2O and steadily increases to 8.8% in K30. Rubidium silicate glasses (Table 3.6) Our rubidium silicate glasses also have a high wavenumber shoulder (∼ 1150 cm−1) which decreases 4 with the eventual disappearance of the Q -1 peak at 20 mol. % Rb2O. This series is different because only three bands are needed to fit the envelope for samples Rb5 to Rb20 conversely with 4 to 5 bands 2 −1 with the Li, Na and K series of glasses. The Q peak (∼ 950 cm ) appears at 20 mol. % Rb2O and marginally increases from 1.8% in Rb20 to 2.3% in Rb30. Caesium silicate glasses (Table 3.7) Our final series, the caesium silicate glasses, also have a high wavenumber shoulder (∼ 1150 cm−1) 4 which is pronounced in Cs5, Cs10 and Cs15. The Q -1 peak no longer appears at 20 mol. % Cs2O confirming that this is a systematic change throughout all of the alkali silicate glass series. As in the 2 −1 rubidium series, we observe that the Q peak (∼ 950 cm ) appears at 20 mol. % Cs2O and marginally increases from 1.0% in Cs20 to 3.2% in Cs30.

3.4 Discussion

3.4.1 The D1 and D2 bands (495 and 600 cm−1)

The relative intensity of the D1-D2 bands in alkali silicate glasses

We notice that as alkali is added to v-SiO2 there is a marked difference in the relative intensity (ID1/ID2) of the D1 and D2 peaks. As mentioned in Section 3.3.2, the peak intensity of the D1 peak decreases and that of the D2 peak increases with increasing alkali content for all alkalis (Fig. 3.10(b)). This confirms our previous findings in the caesium silicate glasses and extends it to all alkali silicate glasses (O’Shaughnessy et al., 2017). In that study, we proposed that this shift in relative intensities is indicative of an increase in the abundance of 3-membered SiO4 rings coupled with a decrease in 4- Chapter 3. The structure of alkali silicate glasses 53

membered SiO4 rings (O’Shaughnessy et al., 2017). The transition from a population of SiO4 rings dominated by 4-membered to 3-membered rings was previously noted in caesium silicate glasses by

(a)

(b)

(c)

Figure 3.10: (a) The calculated molar volume of the alkali silicate glasses. (b) Ratio of the intensities of the D1 and D2 Raman bands (c) The D1-D2 peak splitting, ∆D. Chapter 3. The structure of alkali silicate glasses 54

n Malfait(2009) using an argument based on the abundances of Q species. The change in SiO4 ring population statistics is no doubt reﬂective of the space accommodation required as alkalis are added to the silicate glass network. This is reinforced by our density measurements (Fig. 3.3) and molar volume calculations (Fig. 3.10(a)) which both increase with increasing alkali content with the exception of the lithium bearing silicate glasses. The molar volume of the lithium series decreases as the lithium content grows due to the low mass and radius of lithium.

The splitting of the D1-D2 bands in alkali silicate glasses

We noted in Section 3.3.2 that the peak positions of the D1 (positive shift in wavenumber) and D2 (negative shift in wavenumber) bands vary with increasing alkali content (Table 3.1). In order to illustrate these changes we are using a splitting value, ∆D = |ωD1 − ωD2| (where ωD1 and ωD2 are the peak positions of the D1 and D2 bands respectively), which tracks the relative diﬀerence of the two peak centres as a function of the composition (Fig. 3.10(c)). The maximum change in the value of ∆D occurs in potassium silicate glasses with a change of ∼60 cm−1 from ∼105 cm−1 to ∼55 cm−1. The general behaviour is consistent throughout all alkalis, though the data for the lithium silicate glasses makes it diﬃcult to draw any conclusions since the peaks are no longer discernible above 15 mol. % Li2O. For −1 all other alkalis, we see an average change of ∼55 cm in ∆D values.

The ∆D of all alkali silicate glasses behave similarly to caesium silicate glasses (O’Shaughnessy et al.,

2017). We suggest that this indicates a decrease in the Si-O-Si bond angle for 4-membered SiO4 rings which are closing as M2O is added to the network (Sharma et al., 1981; Galeener, 1982; O’Shaughnessy et al., 2017). As previously noted in O’Shaughnessy et al.(2017), this is consistent with the formation of 3-membered SiO4 rings caused by the breakdown of higher order SiO4 rings (≥4). The breakdown is due to the increasing concentration of NBO with the introduction of modiﬁers in the glass network (Greaves et al., 1981).

Implications from Percolation theory

Percolation theory can be used to describe the geometrical aspects of elasticity in generic central-force networks where the property of interest is rigidity (Thorpe, 1985; Zhang and Boolchand, 1994; Jacobs and Thorpe, 1996). Micoulaut and Phillips(2003) used a combinatorial algorithm to count clusters (rings) of different sizes with constraint counting algorithms to show that there exists two rigidity transitions in silicate glasses, thus, defining three elastic phases: floppy, isostatically rigid and stressed-rigid. They highlight that the transition from one elastic regime to another will drastically affect the properties of materials, and in this case, alkali silicate glasses (Micoulaut and Phillips, 2003). Specifically, this has implications for phase separation and the potential creation of channels in alkali silicate glasses. The transition from a rigid to a floppy network is calculated to occur at a critical value of 20M2O−80SiO2 (Micoulaut and Phillips, 2003). From the values of ID1/ID2 we can confirm that there is a change in the abundance of 3-membered SiO4 rings relative to 4-membered SiO4 rings between 20-25 mol. % for K-, Rb- and Cs-silicate glasses. We believe that a large change in ring statistics, coupled with the accompanied change in the distribution of Qn species may be indicative of a transition from a rigid to a floppy network where there are regions with higher amounts of alkali. Chapter 3. The structure of alkali silicate glasses 55

(a)

(b)

(c)

Figure 3.11: The relative areas of the diﬀerent Qn species for the lithium silicate glasses as a function of alkali content. (a) The areas for the Q4 bands. (b) The areas for the Q3 bands. (c) The area of the Q2 band. Chapter 3. The structure of alkali silicate glasses 56

(a)

(b)

(c)

Figure 3.12: The relative areas of the diﬀerent Qn species for the sodium silicate glasses as a function of alkali content. (a) The areas for the Q4 bands. (b) The areas for the Q3 bands. (c) The area of the Q2 band. Chapter 3. The structure of alkali silicate glasses 57

(a)

(b)

(c)

Figure 3.13: The relative areas of the diﬀerent Qn species for the potassium silicate glasses as a function of alkali content. (a) The areas for the Q4 bands. (b) The areas for the Q3 bands. (c) The area of the Q2 band. Chapter 3. The structure of alkali silicate glasses 58

(a)

(b)

(c)

Figure 3.14: The relative areas of the diﬀerent Qn species for the rubidium silicate glasses as a function of alkali content. (a) The areas for the Q4 bands. (b) The areas for the Q3 bands. (c) The area of the Q2 band. Chapter 3. The structure of alkali silicate glasses 59

(a)

(b)

(c)

Figure 3.15: The relative areas of the diﬀerent Qn species for the caesium silicate glasses as a function of alkali content. (a) The areas for the Q4 bands. (b) The areas for the Q3 bands. (c) The area of the Q2 band. Chapter 3. The structure of alkali silicate glasses 60

(a)

(b)

(c)

Figure 3.16: The relative areas of the diﬀerent Qn species for the alkali silicate glasses as a function of alkali content. (a) The toal area for the Q4 bands. (b) The total area for the Q3 bands. (c) The area of the Q2 bands. Chapter 3. The structure of alkali silicate glasses 61

3.4.2 The high frequency peak (800-1300 cm−1)

The polarizability of Q4 species

In v-SiO2 the high frequency doublet was ascribed to the A1 and T2 vibrational modes (1060 and −1 1200 cm respectively) belonging to the Td symmetry group of the SiO4 tetrahedron (Sarnthein, 1997). These modes have very low intensities compared to the remaining portions of the spectrum: Boson, Main and Defect bands (Fig. 3.1). A curious occurrence is that, with small amounts of alkali added to the silicate network, the intensity of these bands seem to increase (Nesbitt et al., 2019). In Nesbitt et al.(2019), we postulated that the presence of alkalis could prime Q4 species such that their intensities increase since the coordination number of Cs to O is between 10 and 11. The polarization of a symmetric molecule due to an ion (in this case M, an alkali) is likely at short range and decreases with distance (as a power of four) (Nesbitt et al., 2019). Thus, since we observe a similar behaviour for the Q4 bands in all alkali silicates, we propose that it is feasible for a cation to change the polarizability of Q4 molecules with BO located within the ﬁrst coordination sphere, while having no eﬀect on more distant Q4 species (Nesbitt et al., 2019). This is likely the reason for the disproportionate growth of Q4 peaks in alkali silicate glasses with respect to v-SiO2. Whilst having no relation to the priming of Q4 species, the behaviour of the Q4-1 and Q4-2 bands is still of note. For all alkali silicate glasses investigated in this study, we observe that the Q4-1 band decreases in intensity and shifts to lower wavenumber until it is no longer visible at 20 mol. % M2O. We believe that Q4-1 species are eventually so minor that they are no longer detectable and that Q4-2 become the only type Q4 species in the glasses due to increase in alkali content which leads at least one of their BOs to interact with an alkali cation. In all series, except the lithium silicate glasses, the Q4-2 band eventually decreases as well and in the case of the potassium silicate glasses it also disappears (Fig. 3.11(a)-3.15(a)). Additionally, when looking at the total Q4 areas we observe that all the alkali silicate glasses increase from 5-15 mol. % and then decrease from 15-30 mol. %, again with the exception of the lithium series (Fig. 3.16(a)). This data appears to contradict the expected values of total Q4 species from typical mass balance calculations which assume that only Q3 and Q4 species are present in the glasses (Nesbitt et al., 2019). These mass balance calculations do not take into consideration the random distribution of alkalis in the glass and ignore potential areas which are enriched in alkalis versus ones which are depleted.

The asymmetry of the Q3 band

The main band of the high frequency envelope for all alkali silicate glasses corresponds to the Q3 vibrational band as described by all previous investigators (e.g. Matson et al., 1983; McMillan, 1984; O’Shaughnessy et al., 2017; Bancroft et al., 2018; Nesbitt et al., 2019). But as discussed previously in the introduction, it has been shown that M−BO interactions with SiO4 tetrahedra will shift the Si−O vibrations to lower wavenumbers, thus creating multiple Q3 bands resulting from diﬀering amounts of alkalis bound to the BOs in the glass. The production of multiple bands from M−BO interactions leads to an asymmetry in the Q3 band (Nesbitt et al., 2019, their Fig. 6). We monitored the asymmetry in the Q3 band by plotting the areas of all Q3 species (Q3-1, Q3-2 and Q3-3) versus the total Q3 area (Fig. 3.11(b)-3.15(b)). We observed that the areas of Q3-2 and Q3-3 increased with increasing alkali content thus conﬁrming that the asymmetry of the Q3 band grows as alkali is added to the glass. Another second order process, may also produce similar asymmetry and is referred to as Qn,ijkl species where a Qn has Chapter 3. The structure of alkali silicate glasses 62

Qi,Qj,Qk and Ql as neighbouring structural units (Glock et al., 1998; Sen and Youngman, 2003). Using 29Si double quantum NMR, Sen and Youngman(2003) showed that both the Q 3 and Q4 peak shifted by ∼ 5 ppm (parts per million by frequency) if they were linked to adjacent Q3 species rather than Q4 (Glock et al., 1998; Sen and Youngman, 2003). This happens due to higher concentration of modiﬁer cations in certain regions of the glass, it is likely that M−BO interactions produce a stronger eﬀect on the electron density because of their proximity to the BO atoms but Qn,ijkl clustering may also decrease 3 the A1 symmetric stretch frequencies of the Q band rendering it asymmetric (Nesbitt et al., 2019). Furthermore, we show that the total area of Q3 bands increases in conjunction with the decrease in total Q4 area, which is expected with addition of alkalis and general thoughts on the depolymerization of the glass network (Fig. 3.16(b)). However, we note once again that, lithium does not follow the same trend as the other alkalis (Fig. 3.11(b) & 3.16(b)).

The appearance and increase of Q2 species: implications for the silicate glass network

As expected from the MRN model, lower order Qn species should be produced from the breakdown of Si−O−Si bonds with the addition of network modifier cations (Greaves et al., 1981). It is then no 2 surprise that the area of the Q peak increases with increasing M2O content (Fig. 3.11(c)-3.16(c)). The occurrence of Q2 species at such low alkali content in lithium silicate glasses points to the fact that lithium ions appear to have a predisposition to concentrate in areas which are rich in NBO and leave large areas of SiO2-rich glass. This is corroborated by the strange behaviour of the total area of 4 Q species which remains more or less unchanged from 5-30 mol. % Li2O. These results are consistent with the findings of Matson et al.(1983) which showed that lithium exhibits a greater preference for Q 2 species at low alkali concentrations relative to the other alkalis. In a study from Uhlig et al.(1996), neutron and X-ray diffraction were used to measure the Li −O bond distance as 1.96 A˚ for Li20 (close to 1.94 A˚ found for crystalline Li-silicates). They calculated the Li coordination number as 4.2. Additionally, the ab initio molecular orbital simulations of Uchino and Yoko(1999) showed that the coordination number of sodium was larger than that of lithium in silicate glasses. They reported that the coordination environments are affected by alkali electronegativity and that the Li−O bond overlap is much larger than that for Na−O bonds leading to stronger covalent bonds between lithium and oxygen (Uchino and Yoko, 1999). Due to this, lithium is more likely to interact with NBO atoms rather than BO atoms, whereas sodium most likely interacts with both types of oxygen, thus creating M−BO interactions (Soltay and Henderson, 2005). More recently, Ispas et al.(2010) investigated Li20 and Na20 (listed as LS4 and NS4 respectively, in their paper) using Car-Parinello MD simulations. The Li−O bond distance was calculated as 1.92 A˚ and the Na−O bond distance was 2.28 A˚ reflective of the charge density and bond strength of the two alkalis (Ispas et al., 2010). They also calculated the Li−NBO bond distance and obtained 1.87 A,˚ this value is smaller than the total Li−O bond length average. These results are indicative of some Li−BO bonds in these glasses (Ispas et al., 2010). Finally, when simulating NMR data, they obtained signals for Q2 species which are not discernible in their experimental Li20 spectra (Ispas et al., 2010). 2 2 The appearance of Q species occurs at increasing M2O content with increasing charge density (Z/r ) of modifier cation (5 mol. % for Li up to 20 mol. % for Cs). It appears that the tendency of alkalis to form Q2 species decreases with increasing alkali size as reflected in the small areas of Q2 species in both the rubidium and caesium silicate glasses, confirming the initial findings of Matson et al.(1983). Lithium atoms appear to have a higher affinity for each other, even at low concentrations, compared to Chapter 3. The structure of alkali silicate glasses 63 caesium. We therefore postulate that, though all alkali silicate glasses will form percolation channels (Greaves et al., 1981; Le Losq et al., 2017), they will form them at different alkali content depending on the size of the alkali in question; from lowest M2O content for Li to highest for Cs.

3.5 Conclusions

The Raman spectra of alkali silicate glasses have proven to be useful for deciphering the structural changes which occur in the silicate glass network as network modifying cations are added.

3.5.1 Defect bands and their behaviour

The change in relative intensity of the “Defect bands” relates to the formation of smaller 3-membered

SiO4 rings at the expense of the larger 4-membered SiO4 rings. Additionally, the shift of the D2 band to higher wavenumber is consistent with the closing of the intertetrahedral angle in 4-membered SiO4 rings as modiﬁer cations are incorporated into the silicate glass network. These track with the molar volume of the glasses as they do with densiﬁcation in v-SiO2.

3.5.2 Q4 band properties

The A1 and T2 vibrational mode peaks in v-SiO2 are weak in intensity, but, with only small amounts of alkali added, the Q4-2 species bands grow in intensity. We believe that the presence of alkali in the vicinity of Q4 species have a direct inﬂuence on this increase in band intensity and these species are referred to as primed Q4 species.The total area of the Q4 bands decreases with increasing alkali content with the exception of the lithium silicate glasses. This is indicative of network depolymerization in which the addition of modiﬁer cations ruptures Si−O−Si bonds and creates NBO.

3.5.3 The Q3 band asymmetry: M−BO interactions & Qn,ijkl clustering

3 3 3 We propose that the peaks labeled Q -2 and Q -3 result from A1 symmetric stretch of Q species aﬀected by M−BO interactions (Nesbitt et al., 2015, 2017b, 2019) and/or possibly the clustering of Qn,ijkl (Sen and Youngman, 2003). Weakening of the Si−O force constants decreases the value for the 3 3 Q species A1 symmetric stretch to lower wavenumber, thus, creating an asymmetric Q band, illustrated by the total combined areas of Q3-1, Q3-2 and Q3-3. This conﬁrms the increase in M−BO interactions of Q3 species with increasing alkali content in the glasses.

3.5.4 Q2 species and the clustering of alkalis in silicate glasses

The increase in the area of the Q2 peak across all alkalis is due to the depolymerization of the silicate 2 network with increasing M2O content. The Q peak appears at increasingly higher concentrations of

M2O with increasing cation size (from Li to Cs) as seen in Fig. 3.16(c). This leads us to believe that cations with a higher charge density (Z/r2) cluster more readily than cations with a lower charge density. It is then probable that caesium silicate glasses have a more random distribution of modiﬁer atoms than lithium silicate glasses. Chapter 3. The structure of alkali silicate glasses 64

3.5.5 Implications for the structural model of alkali silicate glasses

M2O where we observe a transition in the behaviour of the relative intensities of the “Defect bands”, as well as, the disappearance of the Q4-2 peak. This composition was highlighted by previous researchers as the potential transition from a rigid to ﬂoppy network using Percolation theory (Micoulaut and Phillips, 2003). This is a result of the destruction of Q4 species which would in turn render the network less rigid through breaking of Si−O−Si bonds. This is entirely consistent with the creation of percolation channels as expected by the MRN (Greaves et al., 1981) which has recently been extended to all glass compositions (Le Losq et al., 2017).

3.6 Acknowledgments

GSH acknowledges funding from NSERC in the form of a discovery grant. Chapter 3. The structure of alkali silicate glasses 65

Sample Density (g/cm3)

Li5 2.220 Li10 2.236 Li15 2.259 Li20 2.283 Li25 2.307 Li30 2.331

Na5 2.241 Na10 2.289 Na15 2.337 Na20 2.384 Na25 2.426 Na30 2.466

K5 2.258 K10 2.308 K15 2.351 K20 2.387 K25 2.423 K30 2.454

Rb5 2.381 Rb10 2.543 Rb15 2.704 Rb20 2.859 Rb25 3.087 Rb30 3.247

Cs5 2.521 Cs10 2.793 Cs15 3.025 Cs20 3.242 Cs25 3.434 Cs30 3.632

Table 3.2: The average density of alkali silicate glasses as calculated from 10 independent measurements per sample. The error on the measurements is equal to ± 0.005 g/cm3. Chapter 3. The structure of alkali silicate glasses 66

Li5 Li20

Li10 Li25

Li15 Li30

Figure 3.17: The curve-ﬁtting for the lithium silicate glasses (peak assignments listed in Table 3.3). Chapter 3. The structure of alkali silicate glasses 67

Sample Position (cm−1) FWHM (cm−1) % Lorentzian Area (%) Qn

Li5 1213 59 0 5.7 Q4-1 1164 59 0 7.6 Q4-2 1082 54 92 59.6 Q3-1 1050 52 92 23.9 Q3-2 937 33 92 3.2 Q2

Li10 1205 59 0 4.5 Q4-1 1156 59 0 6.2 Q4-2 1081 61 92 63.4 Q3-1 1039 62 92 19.2 Q3-2 944 46 92 6.7 Q2

Li15 1186 54 0 3.7 Q4-1 1135 58 0 6.9 Q4-2 1081 62 90 63.5 Q3-1 1031 52 90 15.3 Q3-2 950 58 90 10.6 Q2

Li20 1144 73 0 8.2 Q4-2 1083 62 90 59.7 Q3-1 1035 62 90 20.6 Q3-2 949 56 90 11.5 Q2

Li25 1143 70 0 9.1 Q4-2 1085 62 85 58.4 Q3-1 1034 62 85 20.5 Q3-2 950 56 85 12.0 Q2

Li30 1135 71 0 10.9 Q4-2 1084 62 80 53.6 Q3-1 1031 62 80 21.0 Q3-2 950 56 80 14.5 Q2

Table 3.3: The ﬁtting results for the lithium silicate glasses. Chapter 3. The structure of alkali silicate glasses 68

Na5 Na20

Na10 Na25

Na15 Na30

Figure 3.18: The curve-ﬁtting for the sodium silicate glasses (peak assignments listed in Table 3.4). Chapter 3. The structure of alkali silicate glasses 69

Sample Position (cm−1) FWHM (cm−1) % Lorentzian Area (%) Qn

Na5 1212 59 0 7.9 Q4-1 1161 61 0 16.3 Q4-2 1092 52 96 60.7 Q3-1 1055 52 96 15.1 Q3-2

Na10 1195 59 0 7.5 Q4-1 1148 61 0 15.9 Q4-2 1093 52 96 63.0 Q3-1 1059 49 96 12.0 Q3-2 950 52 96 1.6 Q2

Na15 1184 59 0 9.3 Q4-1 1139 59 0 16.7 Q4-2 1093 52 90 59.2 Q3-1 1058 52 90 12.8 Q3-2 950 52 90 2.0 Q2

Na20 1149 75 0 24.7 Q4-2 1095 62 90 65.7 Q3-1 1052 52 90 7.3 Q3-2 950 52 90 2.3 Q2

Na25 1140 70 0 22.3 Q4-2 1096 62 85 65.0 Q3-1 1047 52 85 8.9 Q3-2 950 52 85 3.8 Q2

Na30 1130 50 0 13.9 Q4-2 1099 52 80 59.7 Q3-1 1060 52 80 13.8 Q3-2 1018 43 80 4.2 Q3-3 950 52 80 8.4 Q2

Table 3.4: The ﬁtting results for the sodium silicate glasses. Chapter 3. The structure of alkali silicate glasses 70

K5 K20

K10 K25

K15 K30

Figure 3.19: The curve-ﬁtting for the potassium silicate glasses (peak assignments listed in Table 3.5). Chapter 3. The structure of alkali silicate glasses 71

Sample Position (cm−1) FWHM (cm−1) % Lorentzian Area (%) Qn

K5 1197 70 0 13.7 Q4-1 1155 58 0 16.9 Q4-2 1098 46 94 64.4 Q3-1 1060 52 94 5.0 Q3-2

K10 1186 52 0 12.4 Q4-1 1148 50 0 18.0 Q4-2 1098 52 90 62.4 Q3-1 1029 52 90 7.2 Q3-2

K15 1190 55 0 8.4 Q4-1 1152 56 0 18.4 Q4-2 1097 62 90 69.3 Q3-1 1035 62 90 2.2 Q3-2 950 52 90 1.7 Q2

K20 11530 67 0 27.7 Q4-2 1098 64 85 68.6 Q3-1 1027 40 85 2.0 Q3-2 950 51 85 1.7 Q2

K25 1139 56 0 18.1 Q4-2 1100 61 80 70.9 Q3-1 1037 50 80 7.7 Q3-2 950 50 80 3.3 Q2

K30 1104 54 68 75.5 Q3-1 1054 54 68 12.6 Q3-2 1008 52 68 3.1 Q3-3 943 41 68 8.8 Q2

Table 3.5: The ﬁtting results for the potassium silicate glasses. Chapter 3. The structure of alkali silicate glasses 72

Rb5 Rb20

Rb10 Rb25

Rb15 Rb30

Figure 3.20: The curve-ﬁtting for the rubidium silicate glasses (peak assignments listed in Table 3.6). Chapter 3. The structure of alkali silicate glasses 73

Sample Position (cm−1) FWHM (cm−1) % Lorentzian Area (%) Qn

Rb5 1205 51 0 6.1 Q4-1 1161 59 0 23.8 Q4-2 1099 45 92 70.1 Q3-1

Rb10 1200 51 0 4.4 Q4-1 1156 59 0 25.4 Q4-2 1099 46 90 70.2 Q3-1

Rb15 1200 46 0 2.9 Q4-1 1154 66 0 29.7 Q4-2 1098 50 85 67.4 Q3-1

Rb20 1152 60 0 30.1 Q4-2 1100 61 85 68.1 Q3-1 963 51 85 1.8 Q2

Rb25 1140 53 0 23.9 Q4-2 1102 52 80 65.8 Q3-1 1055 54 80 8.2 Q3-2 946 51 80 2.1 Q2

Rb30 1131 42 0 14.9 Q4-2 1101 43 75 71.2 Q3-1 1059 54 75 11.6 Q3-2 935 30 75 2.3 Q2

Table 3.6: The ﬁtting results for the rubidium silicate glasses. Chapter 3. The structure of alkali silicate glasses 74

Cs5 Cs20

Cs10 Cs25

Cs15 Cs30

Figure 3.21: The curve-ﬁtting for the caesium silicate glasses (peak assignments listed in Table 3.7). Chapter 3. The structure of alkali silicate glasses 75

Sample Position (cm−1) FWHM (cm−1) % Lorentzian Area (%) Qn

Cs5 1187 75 0 16.3 Q4-1 1150 57 0 14.9 Q4-2 1098 40.1 95 67.1 Q3-1 1019 40.2 95 1.7 Q3-2

Cs10 1188 59 0 11.4 Q4-1 1145 63 0 24.4 Q4-2 1097 41.3 90 61.4 Q3-1 1017 40.1 90 1.8 Q3-2 983 38 90 1.0 Q3-3

Cs15 1179 57.1 0 11.2 Q4-1 1142 63.6 0 26.1 Q4-2 1095 45.6 85 60.0 Q3-1 1004 45.6 85 1.7 Q3-2 965 45.6 85 1.0 Q3-3

Cs20 1147 63.5 0 32.2 Q4-2 1095 53.6 80 65.1 Q3-1 990 53.6 80 1.7 Q3-2 940 51.4 80 1.0 Q2

Cs25 1136 56 0 26.2 Q4-2 1096 54 80 68.1 Q3-1 1050 45 80 3.0 Q3-2 967 55 80 1.8 Q3-3 927 40 80 0.9 Q2

Cs30 1135 39 0 9.2 Q4-2 1099 50 80 73.3 Q3-1 1045 52 80 11.1 Q3-2 990 43 80 3.2 Q3-3 931 40 80 3.2 Q2

Table 3.7: The ﬁtting results for the caesium silicate glasses. Chapter 4

A Li K -edge XANES study of salts and minerals

This chapter has been published in the Journal of Synchrotron Radiation as:

• O’Shaughnessy, C., Henderson, G. S., Moulton, B. J. A., Zuin, L., Neuville, D. R. (2018b) A Li K -edge XANES study of salts and minerals. Journal of Synchrotron Radiation 25, 543–551.

I have carried out all of the experiments with help from fellow graduate student Dr. Benjamin Moulton at the VLS-PGM beamline at the Canadian Light Source, Saskatoon, Canada. I have done the data processing, ﬁgure preparation and written the article. All other authors have provided training, provided trouble shooting support during the experiments and given feedback during the publishing process.

76 Chapter 4. A Li K -edge XANES study of salts and minerals 77

Abstract

The first comprehensive Li K -edge XANES study of a varied suite of Li-bearing minerals is presented. We demonstrate how the bonding environment changes drastically for lithium and that it can be monitored by using the position and intensity of the main Li K absorption edge. The complex silicates confirm the assignment of the absorption edge to be a convolution of triply degenerate p-like states as previously proposed for simple lithium compounds. The Li K -edge position depends on the electronegativity of the element to which it is bound. The intensity of the first peak varies depending on the existence of a 2p electron and can be used to evaluate the degree of ionicity of the bond. The presence of a 2p electron results in a weak first peak intensity. The maximum intensity of the absorption edge shifts to lower energy with increasing SiO2 content for the lithium aluminosilicate minerals. The bond length distortion of the lithium aluminosilicates decreases with increasing SiO2 content, thus increased distortion leads to an increase in edge energy which measures lithium’s electron affinity.

4.1 Introduction

Lithium is the lightest alkali metal, but also one of the fastest diffusing elements in silicates (Jambon and Semet, 1978). Lithium isotopes in the outer layers of the Earth (hydrosphere, crust and lithospheric mantle) can be fractionated by up to 60 (Tomascak, 2004). The structural chemistry of lithium bearing minerals is of particular interest becauseh of the implications with respect to economic geology. Granitic pegmatite deposits contain lithium in a variety of exotic minerals which have diverse bonding environments (London, 2005). The main industrial use for lithium is in sacrificial anodes in batteries and the inorganic compounds which form on the surface of these anodes has motivated many studies (Kobayashi et al., 2007; Lu et al., 2011; Ogasawara et al., 2015; Wang and Zuin, 2017). There are also applications for lithium in ion-conducting glasses (Voigt et al., 2005) and nuclear waste glasses which typically contain 3.5 wt. % Li2O(Abraitis et al., 2000) and for Zerodur low expansion glass ceramics which are widely used for astronomy and many other technological fields. Using X-ray absorption near-edge structure (XANES) spectroscopy we have examined the Li K -edge of a variety of synthetic Li-bearing compounds as well as a suite of natural minerals from around the world (Table 4.1). We hope these results will be useful as a fingerprinting method for amorphous lithium materials (lithium silicate glasses, glass-ceramics or melts) as well as for help in identification of unknown compounds and highly irregular bonding environments. XANES is a synchrotron-based technique which can provide information on the electronic, magnetic and structural properties of matter (Henderson et al., 2014). It is a process in which a photon of light excites a core electron to an unoccupied molecular orbital state. The photon must have an energy which is larger than or equal to that of the binding energy of the electron. The binding energy of the electron is different for every element, which is why XANES is an element-specific technique which can be tuned to investigate any element of interest. When the electron is excited to a higher empty state it emits a photon and this fluorescence may be measured to yield the absorption spectrum (FLY). The Li K -edge is located between 55-65 eV depending on the environment of Li in the specific compound. Due to its very low energy, it is difficult to probe the Li K -edge, but over the last 15 years there has been an increase in the amount of work being performed using XANES, electron energy loss spectroscopy (EELS) and X-ray Raman scattering (XRS) (Tsuji et al., 2002; Kobayashi et al., 2007; Lu Chapter 4. A Li K -edge XANES study of salts and minerals 78 et al., 2011; Fister et al., 2011; Lee et al., 2014; Pascal et al., 2014; Kikkawa et al., 2014; Taguchi et al., 2015; Ogasawara et al., 2015). Additionally, there has been a small body of computational work using density-functional theory (DFT) and molecular dynamics (MD) simulations on a number of different Li-bearing compounds (Jiang and Spence, 2004; Mauchamp et al., 2006, 2008; Olovsson et al., 2009a,b; Yiu et al., 2013; Pascal et al., 2014).

4.2 Methods

4.2.1 Mineral samples

A suite of nine (9) mineral samples were collected from the UniversitéPierre-Marie Curie (UPMC), Paris, France (Table 4.1). The minerals were identified using Raman spectroscopy fingerprinting from the RRUFF database. Additionally, four synthetic Li-bearing compounds were used as standards. The mineral samples were broken down to ∼ 25 mm2 chips appropriate for XANES experiments. The lithium metasilicate is a synthetic lithium silicate (LS) glass sample which was recrystallized whereas the lithium sulphate, carbonate, and chloride are all high quality powders (>99%). All samples were kept in a desiccator before being transfered to the experimental chamber (detailed in the following section).

4.2.2 X-ray Absorption Near-Edge Structure (XANES) Spectroscopy

The Li K -edge XANES measurements were performed at Canadian Light Source Inc. (CLS), Saska- toon, Canada at the variable line spacing-plane grating monochromator (VLS-PGM) beamline (Hu et al., 2007). The spectra were collected in fluorescence yield (FLY) (Wang and Zuin, 2017). Many researchers have used FLY data because it is known to sample a larger volume of space which renders it more sensitive to the bulk (Jiang and Spence, 2006; Henderson et al., 2014; Moulton et al., 2016; Wang and Zuin, 2017). The spectra were collected in the range of 35-75 eV, some minerals also contain Mn, Fe and Al which cause overlap with the Mn M2,3-edge (∼ 47 eV), Fe M2,3-edge (∼ 53 eV) and Al L2,3-edge (∼ 73 eV). This is discussed on a sample by sample basis. The beamline slits were opened to 50 mm × 50 mm with a resolution E/∆E > 10000 and the pressure in the experimental chamber is maintained below 10−7 torr for all measurements (Wang and Zuin, 2017). The Li K -edge spectra were processed by the following routine: (i) the fluorescence signal was intensity normalized to the incoming intensity (I0), measured by a Ni mesh located upstream of the sample chamber, (ii) a two-step polynomial background subtraction was performed on the pre-edge region (∼ 35-58 eV) and post-edge region (∼ 65-75 eV) following the edge processing of Moulton et al. (2016), (iii) the resulting spectra were then intensity normalized setting the maximum intensity peak to a value of 1. After the normalization procedure, the final spectra of the Li-bearing salt standards (LiCl,

Li2CO3 and Li2SO4) were compared to previously published experimental spectra and were in accord (Tsuji et al., 2002; Handa et al., 2005; Fister et al., 2011; Pascal et al., 2014). Finally, the edge value of our lithium chloride (LiCl) sample has been calibrated to the edge value (60.8 eV) from Tsuji et al. (2002) XANES experiment, following the procedure of Wang and Zuin(2017) and the remaining spectra have all been shifted by an equal amount. The calibrated positions are the same as the edge positions obtained in previous XANES (Tsuji et al., 2002; Handa et al., 2005) and XRS (Fister et al., 2011; Pascal et al., 2014) studies with a maximum edge diﬀerence of ±0.1 eV. Chapter 4. A Li K -edge XANES study of salts and minerals 79 Synthetic Synthetic Synthetic Synthetic California, USA San Benito, USA Branchville, USA Bikita, Zimbabwe Montebras, France Zinnwald, Germany Minas Gerais, Brazil Minas Gerais, Brazil Minas Gerais, Brazil ; X=O, S) ˚ A) Locality i ( 4 5 6 7 1 2 3 8 9 X 11 10 12 13 i − X Li − h Li h 6 2.12 6 2.21 6 1.89 44 2.00 6 1.94 6 2.02 2.11 44 1.97 1.97 6 2.16 6 2.57 1 m /c ¯ 3 4 1.99 ¯ 1 6 2.11 ¯ 3 1 P R Cc C2 R3m P2/c C2/c C2/c C2/c Pmnb P2 Fm Cmc2 Triclinic Trigonal Trigonal Isometric Monoclinc Monoclinic Monoclinic Monoclinic Monoclinic Monoclinic Monoclinic Orthorhombic Orthorhombic (OH) 3 ) 24 O 2 8 )(OH) 3 (Si OH) 2 2 , Ti )(BO )(F 2 OH) ) 18 , 10 O 2+ O 6 )(F 3 10 Mn (Si , 6 O 4 4 2+ Al )(OH) 3 6 10 Al(AlSi (Si 4 4 PO · O O 3 3 Al) 2 4 4 2+ K(Fe , 2+ Al 2 2 SO CO SiO 2 2 2 Table 4.1: Symmetry, coordination number (C.N.) & the mean bond length ( KLi KLiFe LiAlSiO LiAlSi LiAlSi LiAl(PO LiMn LiNa Li Li Li Na(Li LiCl Lepidolite Zinnwaldite Eucryptite Spodumene Petalite Montebrasite Lithiophilite Neptunite Li sulphate Li carbonate Li metasilicate Elbaite MineralLi chloride Formula[1] ( 1963 );Wyckoﬀ [2] ( 1933 );Albright Symmetry [3] [7] ( 1957 );Zemann Ross [4] etHesse ( 1977 ); al. ( 2015 ); [5] [8] [12] DanielsGroatGuggenheim and et and Fyfe ( 2001 ););( 2003 al. Bailey( 1977 ); [6] [9] [13] ClarkHatertCannillo et et et al. ( 1968 ); al. ( 2012 ); al. ( 1966 ); [10] C.N. Gatta et al. ( 2012 ); [11] Sartori et al. ( 1973 ); Chapter 4. A Li K -edge XANES study of salts and minerals 80

Zinnwaldite Neptunite

Lepidolite

Elbaite Complexsilicates Montebrasite

Lithiophilite PO

Petalite Spodumene

LAS Eucryptite

Li2 CO 3

Li2 SiO 3

Li2 SO 4

LiCl Standards

Figure 4.1: The Li K-edge spectra of our sample suite. The standards black, the L-A-S minerals (blue) with increasing SiO2 content, the phosphate minerals (pink), tourmaline (green) and the phyllosilicates (orange) are intensity normalized. The dotted lines are located at 61 eV and 65 eV and serve as a guide for the eyes when comparing the diﬀerent spectra (same locations for all ﬁgures). Note that the proximity of the Fe M2,3- and Mn M2,3- to the Li K-edge creates overlapping edge contributions and these complications will be discussed in the following sections.

4.3 Results

The results are divided into sections according to the similarities the spectra have with one another.

We begin with the four salt standards (LiCl, Li2SO4, Li2SiO3, Li2CO3), followed by the lithium alum- niosilicate minerals (eucryptite, spodumene and petalite), complex silicates (elbaite and physllosilicates: lepidolite, neptunite and zinnwaldite), and ﬁnally the lithium phosphates (lithiophilite and montebrasite). For comparison, all the spectra are shown in Fig. 4.1 and their peak positions are compiled in Table 4.2. Chapter 4. A Li K -edge XANES study of salts and minerals 81

Table 4.2: Peak positions in lithium-bearing salts and minerals

Mineral p0 p A B C D E F G H Li chloride - 59.5 60.8 - 62.4 - 63.8 - - 69.3 Li sulfate - 59.0 - 61.7 - 63.0 64.0 - 67.1 69.1 Li metasilicate - - 60.5 61.8 - - - 66.4 68.0 69.6 Li carbonate - 58.8 - 61.8 - - - - 67.2 69.2 Eucryptite 57.2 - - 61.2 62.2 63.1 - 65.3 68.7 - Spodumene 56.5 58.7 - - 62.1 62.7 64.3 65.8 68.2 - Petalite 57.6 - - - 61.7 - 64.4 - 67.3 - Montebrasite - - 60.8 61.8 - 62.8 64.9 - 67.2 - Lithiophilite - - 60.9 61.8 - 63.0 - 66.1 67.9 - Elbaite - - 60.4 - 62.1 63.0 - 65.7 67.3 - Lepidolite - 59.3 - 61.6 - 62.9 - 65.6 67.7 - Zinnwaldite - - - - - 63.0 - 66.1 68.1 - Neptunite - - - - - 63.1 - 65.6 - -

4.3.1 Lithium salts and metasilicate

Lithium chloride (LiCl)

The ﬁrst standard to be considered is lithium chloride which is isometric (Fm3¯m) with Li ions occupying the Na site in the halite structure. The Na K -edge of halite (NaCl) has previously been investigated and displays a number of complex features which have been attributed to the high symmetry of the close packed B1 structure type using a combination of XANES and ab-initio calculations (Kasrai et al., 1991; Kikas et al., 2001; Prado and Flank, 2005; Neuville et al., 2004). The Li K -edge absorption spectrum of LiCl consists of a sharp main peak (peak A) at 60.8 eV similar to that observed in the Na K -edge of halite. The two spectra diﬀer due to a small number of additional features present in the spectrum of LiCl which only has a low intensity doublet located at 62.4 eV (peak C) and 63.8 eV (peak E) respectively (Fig. 4.2). Additonally, peak A is asymmetric on the low energy side and undergoes a change in slope between 59.5 eV to 60.4 eV, this pre-edge feature will be referred to as peak p. The spectrum of LiCl shows all the same features as those presented in previous XANES studies on lithium halides (Handa et al., 2005; Wang and Zuin, 2017).

Lithium sulphate (Li2SO4)

Lithium sulphate is an ionic compound with monoclinic symmetry (P21/c), with Li atoms surrounded by four oxygen atoms which belong to a sulphate group (Albright, 1933). The spectrum of Li2SO4 is comprised of a main absorption edge located at 61.72 eV (peak B) and a secondary broad peak with two features at 67.0 eV (peak G) and 69.1 eV (peak H) (Fig. 4.2). Peak B is also characterized by a small pre-edge feature at 59 eV (peak p) and a shoulder composed of a doublet similar to that found in the spectrum of LiCl at 63.0 eV (peak D) and 64.0 eV (peak E). The energy of all the features are in agreement with the X-ray Raman spectroscopy and ab initio molecular dynamics (AIMD) simulations of Pascal et al.(2014). Chapter 4. A Li K -edge XANES study of salts and minerals 82

G B H P

A F G Li2 CO 3 B H

B Li2 SiO 3 D G E H

A Li2 SO 4

C E p H LiCl

Figure 4.2: The Li K-edge spectra of standard compounds used to calibrate and interpret the results for the mineral suite. Note the sharp absorption edge in LiCl, Li2SO4 and Li2SiO3 and how it contrasts with the lower energy pre-edge (peak p) observed in Li2CO3.

Lithium metasilicate (Li2SiO3) The ﬁnal standard is a crystalline sample of lithium metasilicate which has orthorhombic symmetry

(Cmc21) and is composed of parallel chains of SiO4 tetrahedra in the [001] direction (Hesse, 1977). In ˚ Li2SiO3, lithium is tetrahedrally coordinated and has a mean Li−O distance of 2.00 A(Hesse, 1977). The spectrum of lithium metasilicate exhibits two main features: (i) the main absorption edge at 60.5 eV (peak A) with a shoulder at 61.8 eV (peak B); and, (ii) a triplet at 66.4 (peak F), 68.0 (peak G) and 69.6 eV (peak H) (Fig. 4.2). As is the case with the remaining spectra to be presented in the study, there are no previous XANES Li K -edge data available for comparison.

Lithium carbonate (Li2CO3) We present lithium carbonate which is a monoclinic compound (C2/c) with Li in tetrahedral coordination (Zemann, 1957). The XANES spectrum of lithium carbonate diﬀers quite drastically from that Chapter 4. A Li K -edge XANES study of salts and minerals 83

of both LiCl and Li2SO4 due to its very strong pre-edge feature located at 58.8 eV (peak p) (Fig. 4.2). The main edge is found at 61.8 eV (peak B) and appears to be the product of three contributions as the peak shows inﬂection points at both the low and high energy sides of the maximum intensity. The second broad feature is located at ∼ 67 eV and seems to be composed of at least two contributions, peak G and peak H at 67.2 eV and 69.2 eV). The features in the experimental spectrum are in agreement with those of Pascal et al.(2014) though their result does not resolve peak p. Conversely, their simulations predict this pre-edge feature in both the results for the AIMD simulations at 298 K and the excited electron and Core-Hole (XCH) formalism of density functional theory (DFT) for a static crystal at 0 K (Pascal et al., 2014). It appears that XANES is better suited at resolving higher resolution features in the Li K -edge.

4.3.2 Lithium aluminosilicates minerals (LAS)

A series of three aluminosilicates was analyzed in order to evaluate the eﬀect of increasing SiO2 concentration on the absorption edge of lithium. The lithium aluminosilicates are eucryptite (LiAlSiO4), spodumene (LiAlSi2O6), and petalite (LiAlSi4O10) and their silica concentration increases in this order.

These three phases fall along the tie-line between SiO2 and Li2O · Al2O3 in the Li2O−Al2O3−SiO2 phase diagram (Roy and Osborn, 1949) (their Fig. 1). It is worth noting that eucryptite and spodumene are two very important phases for glass ceramics, particularly for Zerodur low-expansion glass ceramics. The samples show systematic trends in the edge energy with changing composition which will be elaborated upon below (Fig. 4.3). It is important to note that due to the aluminum content the data at energies greater than 70 eV are subject to interference due to the presence of the Al L2,3 absorption edge which is located around 72 eV.

Eucryptite (LiAlSiO4)

α-Eucryptite has R3¯ symmetry and consists of three symmetrically non-equivalent tetrahedral sites with one site occupied solely by Li and the other two being partly occupied by both Si and Al (Hesse, 1985; Daniels and Fyfe, 2001). Hesse(1985) described α-eucryptite as being isostructural to willemite

(Zn2SiO4). All three types of tetrahedra (Li, Al, and Si) are connected to eight other tetrahedra via bridging oxygen atoms (which are all 3-fold coordinated) and form channels along [0001] with six- membered rings perpendicular to the channels (Daniels and Fyfe, 2001) (their Fig. 2). The Li K -edge spectrum of α-eucryptite displays a main absorption edge which is split into a doublet at 61.2 eV (peak B) and 62.2 eV (peak C) with the latter having higher intensity and a small shoulder (peak D) (Fig. 4.3). One must also note the presence of a pre-edge feature located at 57.2 eV (peak p0). Furthermore, it displays another major feature at 65.3 eV (peak F) before a few smaller features appear at 68.7 eV (peak G) and 71.5 eV (due to the Al L2,3-edge).

Spodumene (LiAlSi2O6)

Spodumene is a c-centered clinopyroxene which occurs in three diﬀerent modiﬁcations which depend on temperature (Li and Peacor, 1968; Arlt and Angel, 2000). α-Spodumene (monoclinic, C2/c) is the only naturally occurring room temperature phase while β-spodumene (tetragonal) and γ-spodumene (hexagonal, highly packed and related to β-quartz) are high-temperature polymorphs (Li and Peacor, Chapter 4. A Li K -edge XANES study of salts and minerals 84

1968). The spodumene sample used in these experiments is obviously the room temperature phase α- spodumene. The lithium atoms are located in the octahedral M2 site where other atoms such as Na and Ca may also be located depending on the sample (Clark et al., 1968).

C F B D Al G Eucryptite

p’

C D G E F Al

Spodumene

p’ p

Al E G

p’ Petalite

Figure 4.3: The Li K-edge spectra of lithium aluminosilicate minerals. Eucryptite appears to be composed of three diﬀerent peaks (B, C, and D). These three peaks most certainly exist in all three samples but simply overlap too much in spodumene and petalite (peak C) though a second peak can be seen in spodumene (peak C and D).

The spectrum of spodumene is also composed of a main absorption edge consisting of a doublet located at 62.1 eV (peak C) and 62.7 eV (peak D) respectively. Once again there is the presence of a pre-edge feature, but in the form of a doublet at 56.5 eV (peak p0) and 58.7 eV (peak p) (Fig. 4.3). The higher energy portion of the spectrum may be divided into three parts, this appears similar to eucryptite: (i) a low intensity doublet at 64.3 eV (peak E) and 65.8 eV (peak F); (ii) a narrow asymmetrical peak at 68.2 eV (peak G); and (iii) a broad peak at 71.2 eV (Al L2,3-edge). Chapter 4. A Li K -edge XANES study of salts and minerals 85

Petalite (LiAlSi4O10) The ﬁnal lithium aluminosilicate to be presented is petalite, a monoclinic mineral with space group P2/c (Liebau, 1961). It occurs as α-petalite at ambient pressure and undergoes a phase transition at 2.71 GPa to form β-petalite, this transition is isomorphic and results in the tripling of the unit-cell volume

(Ross et al., 2015). Petalite consists of sheets of corner-sharing SiO4 tetrahedra which are connected to one another via AlO4 tetrahedra (Liebau, 1961; Ross et al., 2015). The lithium atoms are also found in 4-coordinate polyhedra and their geometry varies between square planar and a typical tetrahedron (Ross et al., 2015). The absorption spectrum of petalite diﬀers from both that of eucryptite and spodumene due to the main edge being composed of one single peak located at 61.7 eV (peak C) though it has a shoulder on the high energy side (Fig. 4.3). There is also a pre-edge feature for petalite, though it is much broader than that found in eucryptite. It is found at 57.6 eV (peak p0) and is potentially similar to the doublet observed in spodumene but with both peaks overlapping making them indistinguishable. Furthermore, we observe similar features in the region between 63-73 eV as we did for both eucryptite and spodumene. This region is composed of two peaks at 64.4 eV (peak E) and 67.3 eV (peak G) as well as a third more pronounced peak located at 71.0 eV (Al L2,3-edge).

4.3.3 Lithium phosphates (PO4) Lithium bearing phosphate minerals, speciﬁcally those which contain Fe, Mn and Co have been proposed as alternative materials to LiCoO2 for use as a cathode in lithium-ion batteries (Yiu et al.,

2013) (and references therein). We present the XANES spectrum for montebrasite (LiAl(PO4)(OH)) as a means of isolating and understanding the Li K -edge contribution to absorption spectrum of lithiophilite 2+ (LiMn PO4) which is slightly obfuscated by the presence of the Mn M2,3-edge (Fig. 4.4).

Montebrasite (LiAl(PO4)(OH)) The structure of montebrasite is triclinic (P1)¯ (Baur, 1959; Groat et al., 2003). It often displays solid solution with amblygonite (LiAl(PO4)(F)). Octahedral aluminates form corner-sharing chains along the c axis which are cross-linked by tetrahedral phosphates forming a tetrahedral-octahedral network (Groat et al., 2003). Lithium sits in a six-fold coordinated site in pure montebrasite but if F is added to the structure it causes site disorder between Li in proximity of OH versus F and the occupancy is more accurately described by using a split Li site model (Groat et al., 2003). The spectrum of montebrasite displays a sharp absorption edge at 61.8 eV (peak B) with two shoulder features at 60.8 (peak A) and 62.8 eV (peak D) respectively (Fig. 4.4). Further from the edge lie three features: (i) a shallow peak at 64.9 eV (peak E) and (ii) a well-deﬁned doublet at 67.2 (peak G) and

70.0 eV (Al L2,3-edge).

2+ Lithiophilite (LiMn PO4)

2+ The second phosphate of interest is lithiophilite (LiMn PO4), an orthorhombic mineral (Pmnb) which is isostructural to olivine Mg2SiO4 (Geller and Durand, 1960). Lithium and manganese are both located in octahedral sites which are highly distorted whereas the phosphorous is found in discrete tetrahedral complexes (PO4)(Geller and Durand, 1960). The M1 (Li and vacancies) and M2 (Mn) octahedra form two distinct types of chains throughout the structure (Hatert et al., 2012). Chapter 4. A Li K -edge XANES study of salts and minerals 86

Mn A B D F G

Lithiophilite

D G Al

A Montebrasite

Figure 4.4: The Li K-edge spectra of phosphate minerals. Montebrasite demonstrates a well-deﬁned, sharp absorption edge as was the case with our standards as well as petalite. Lithiophilite conversely is diﬃcult to interpret due to the overlap of the Mn M2,3-edge. Nevertheless, several features from lithiophilite have similar energies to those observed in montebrasite.

We observe signiﬁcant contributions from the Mn M2,3-edge in the spectrum of lithiophilite akin to neptunite (Fig. 4.4). Using the information obtained from montebrasite we have isolated the peaks which are caused by the Li K -edge of lithiophilite. The main absorption edge is composed of a triplet at: 60.9 eV (peak A), 61.8 eV (peak B), and 63.0 eV (peak D). Similarly to montebrasite we also observe a well-deﬁned doublet at 66.1 (peak F) and 67.9 (peak G).

4.3.4 Complex lithium-bearing silicates: phyllosilicates and cyclosilicates

We now consider more complex lithium silicates which provide a very diﬀerent bonding environment for lithium than those mentioned above (Fig. 4.5). This suite consists of a cyclosilicate in the form of the tourmaline end-member elbaite, as well as three phyllosilicates which are lepidolite, zinnwaldite and neptunite. Chapter 4. A Li K -edge XANES study of salts and minerals 87

Mn Fe Fe D Neptunite F

Fe D F G Al

Zinnwaldite D B F p Al G

D Lepidolite C F G Al A

Elbaite

Figure 4.5: The Li K-edge spectra of a Li-bearing tourmaline and three phyllosilicates. Elbaite displays a main peak (D) with two lower intensity pre-edge features. The main peak in lepidolite (D) resembles that of elbaite but is broader, they both are similar to Li2CO3. Zinnwaldite has an Fe M2,3-edge contribution which renders the low energy part of the spectrum diﬃcult to interpret but has a main peak (D) position similar to that of lepidolite and elbaite. Neptunite has both an Fe M2,3-edge and a Mn M2,3-edge overlapping the low energy portion but also has a main peak (D) located at a similar position as the aforementioned minerals.

Elbaite (Na(Li,Al)3Al6(Si6O18)(BO3)(OH)3(OH))

The tourmaline group of minerals are complex borocyclosilicates (XY3Z6(T6O18)(BO3)3V3W where X=Ca, Na, K; Y=Al, Fe2+, Fe3+, Li, Mg, Mn; Z=Al, Cr3+, Fe3+,V3+; T=Si, Al, B; V=O, OH; and W=F, O, OH) and are found in a wide range of natural environments (Hawthorne and Henry, 1999; Gatta et al., 2012). We are speciﬁcally interested in the elbaite end-member as it is the sodium-rich Li-bearing tourmaline member (Donnay and Barton, 1972). It has a trigonal crystal structure with space group R3m and lithium is found in an octahedral site (ZO6) shared with aluminium (Gatta et al.,

2012). This octahedral site along with an Al speciﬁc octahedral site (YO6) link the 6-membered rings of tetrahedra (TO4, T = Si, Al, and B) to the triangular BO3 site (Gatta et al., 2012). Chapter 4. A Li K -edge XANES study of salts and minerals 88

Considering the complexity of elbaite one cannot be surprised by the number of features and shape of its XANES spectrum (Fig. 4.5). The main edge is located at 63.0 eV (peak D) and has two features at slightly lower energy. The ﬁrst is an apparent change in slope occurring at 60.4 eV (peak A) and the second is a shoulder appearing at 62.1 eV (peak C). Similar to the spectra of the lithium aluminosilicates, elbaite’s spectrum consists of three smaller features at higher energies: 65.7 ev (peak F), 67.3 eV (peak

G) and 70.9 eV (Al L2,3-edge).

Lepidolite (KLi2Al · (Si4O10)(F,OH)2)

The first phyllosilicate we present is lepidolite (KLi2Al · (Si4O10)(F,OH)2) which is monoclinic (C2/c) (Sartori et al., 1973). There are three polytypes of lepidolite (1M, 2M1 and 2M2) and they can be differentiated using techniques such as electron back-scattered diffraction (Guggenheim, 1981; Kogure and Bunno, 2004). Guggenheim(1981) described 1M lepidolites of symmetry C2/m as having the potential of octahedral ordering between M1 and two M2 sites along a mirror plane. As outlined by

Sartori et al.(1973), the features of 2M 2 polytype lepidolites to note, is the remarkable octahedral ordering (one of these sites is almost exclusively occupied by Li), and the distortion of the tetrahedra which make up the tetrahedral sheets (resembling trigonal pyramids). For the purpose of this study, we are focusing our efforts on the more common 1M polytype of lepidolite, though it may be interesting to collect the XANES spectra of different polytypes in order to differentiate them. Even though lepidolite is quite different from elbaite certain similarities are observed in both spectra (Fig. 4.5). The absorption edge is also composed of three contributions which are located at 59.3 eV (peak p, shoulder), 61.6 eV (peak B, shoulder) and 62.9 eV (peak D, main peak). On the high energy side of the main peak (D) we observe an inflection at 65.6 eV (peak F). The remaining portion of the spectrum is composed two additional peaks located at 67.7 eV (peak G) and 71.0 eV (Al L2,3-edge).

2+ Zinnwaldite (KLiFe Al(AlSi3O10)(F,OH)2)

2+ A trioctahedral mica, zinnwaldite (KLiFe Al(AlSi3O10)(F,OH)2), is monoclinic (C2 ) comparable to lepidolite (Guggenheim and Bailey, 1977). Lithium is found in the larger of two octahedral sites, shared with Fe2+, Mg and Mn and their respective quantities vary depending on polytype and formation environment (Guggenheim and Bailey, 1977). As seen in lepidolite there are both 1M and 2M polytypes of zinnwaldite (Rieder, 1968). The hypothesis of octahedral ordering was in fact developed in zinnwaldite by Rieder(1968) and subsequently extended to other micas, such as lepidolite.

The spectrum of zinnwaldite is slightly more difficult to interpret due to the overlap of the Fe M2,3- edge which is located at ∼ 53 eV (Fig. 4.5). The first band observed at 59.3 eV (peak Fe) corresponds to the main absorption band of the Fe M2,3-edge (van der Laan, 1991; Garvie and Buseck, 2004). The first observable Li feature is the main absorption edge located at 63.0 eV (peak D). Analogous to lepidolite and elbaite we observe three contributions at higher energies; a well defined peak on the high energy side of the main absorption band centered at 66.1 eV (peak F) followed by a small, asymmetrical peak at 68.1 eV (peak G) and finally the typical broad peak at 70.8 eV (Al L2,3-edge).

2+ 2+ Neptunite (LiNa2K(Fe ,Mn )2Ti2(Si8O24))

The ﬁnal phyllosilicate is neptunite, a piezoelectric mineral with space group Cc with general formula 2+ 2+ LiNa2K(Fe ,Mg,Mn )2Ti2(Si8O24)(Cannillo et al., 1966). The structure can be described as a cage Chapter 4. A Li K -edge XANES study of salts and minerals 89

made up of two 18-membered and four 14-membered rings of SiO4 tetrahedra, this cage is then repeated via translation (Cannillo et al., 1966). Though no previous XANES experiments have been conducted on the Li K -edge, the Ti K -edge of neptunite has previously been reported and correlated to both the coordination and size of the two Ti sites (Waychunas, 1987; Farges et al., 1996). The absorption spectrum of neptunite displays similar features as that of zinnwaldite due to the overlap of the Fe M2,3-edge but it is also complicated by the presence of the Mn M2,3-edge (Fig. 4.5). The contribution from the Li K-edge is still apparent with the main peak visible at 63.1 eV (peak D) as well as a broad feature located at 65.6 eV (peak F).

4.4 Discussion

4.4.1 Lithium salts and metasilicate

The lithium bearing standards provide an opportunity to understand features observed in the Li K -edge spectra of more complex mineral standards. They also have the added beneﬁt of having been previously investigated using a variety of experimental (XANES & XRS) and theoretical (DFT & AIMD) techniques (Tsuji et al., 2002; Handa et al., 2005; Fister et al., 2011; Pascal et al., 2014; Wang and Zuin, 2017). The combined XANES and Discrete Variational Xα molecular orbital (DV-Xα) simulations of Tsuji et al.(2002) showed a correlation between the existence of the 2 p electron and the shape of the core exciton peak. In compounds such as Li metal, Li2O, Li2CO3 and Li2S a 2p electron exists at the exciton level because the ionic bond is weak (Tsuji et al., 2002; Handa et al., 2005). This is manifested as a signiﬁcant pre-edge feature in the spectra rather than a sharp absorption edge as is seen in the lithium halides (Tsuji et al., 2002; Handa et al., 2005). Furthermore, a Li K -edge XANES study of lithium halides by Handa et al.(2005) demonstrated that the edge position was sensitive to the type of halide bound to lithium, they noted a negative shift in the edge energy with decreasing halide electronegativity.

More recently, similar results on lithium compounds with weak ionic bonds (e.g. Li2O, LiOH and

Li2CO3) have been obtained using XRS and band-structure calculations carried out using the Bethe- Salpeter equation (BSE) under DFT (Fister et al., 2011). They observed an overlap between the s- and p-DOS features in all the compounds they studied which points to s-p hybridization and can also be used to separate the symmetry of discrete features (Fister et al., 2011). In the case of Li2CO3 they showed a clear correlation between the pre-edge feature in the Li K -edge spectrum and the same feature in the C K -edge which is typically associated with the π∗ orbital in carbonate anions (Fister et al., 2011).

Furthermore, (Pascal et al., 2014), demonstrated that in both LiF and Li2SO4 the strong absorption peak was due to triply degenerate p-like states which are separated by ∼ 1 eV. Our results are not only consistent with the previously conducted experiments and simulations but have increased resolution around the absorption edge. The spectrum of Li2CO3 in particular resolves the important pre-edge feature which was observed in previous XANES experiments (Tsuji et al., 2002; Wang and Zuin, 2017) and predicted through DFT simulations (Fister et al., 2011; Pascal et al., 2014).

As previously mentioned, the XRS spectrum of Li2CO3 did not resolve the pre-edge feature which was seen in the simulations (Pascal et al., 2014), we can therefore conﬁrm the existence of this feature and agree with the interpretations of Tsuji et al.(2002) that it is present due to the existence of a 2 p electron causing a weak exciton process. Chapter 4. A Li K -edge XANES study of salts and minerals 90

The spectrum of lithium metasilicate (Li2SiO3) has a very sharp Li K -edge with features resembling those present in Li2SO4. As previously stated, this well-deﬁned edge is indicative of a bond with a more ionic character due to the absence of a 2p electron (Tsuji et al., 2002; Pascal et al., 2014). The features present in this spectrum will be useful to compare with lithium silicate glasses and glass ceramics.

4.4.2 Bond length distortion in lithium aluminosilicate minerals

Eucryptite

[4] Spodumene [6]

Petalite

[4]

[6] Spodumene

Eucryptite Petalite [4] [4]

Figure 4.6: (a) The maximum peak position is the location of the maximal peak intensity within the absorption edge triplet, peak c (observed in the LAS minerals). The minerals are plotted against the SiO2 content and are labeled above. Additionally, the coordination of lithium in the respective mineral is listed as it is important to the calculation of the bond-length distortion (BLD). (b) The BLD was determined using the formula described in Wenger and Armbruster(1991) and their values for the mean bond lengths of 4- and 6-coordinated lithium minerals. The values for the individual bond lengths were obtained from Daniels and Fyfe(2001) (eucryptite), Clark et al.(1968) (spodumene) and Ross et al. (2015) (petalite).

The lithium aluminosilicates provide the possibility of investigating the eﬀect of the Li2O : SiO2 ratio along a compositional tie-line. The position (xc) of peak C (the edge feature with the highest intensity) was compared to the Li2O : SiO2 ratio and demonstrate that at higher Li2O content the position of peak Chapter 4. A Li K -edge XANES study of salts and minerals 91

C is located at higher energy (Fig. 4.6(a)). The position reﬂects the strength of the Li−O bond which proves to be stronger with higher Li2O content. The bond strength is typically related to the distortion of the Li polyhedron (i.e. the local environment).

Wenger and Armbruster(1991) investigated the oxygen coordination of Li in a large dataset of lithium minerals and compounds. In doing so, they determined the mean bond lengths (hLi−Oi) for both tetrahedral (1.96 A)˚ and octahedral (2.15 A)˚ lithium (Wenger and Armbruster, 1991). Furthermore, they deﬁned the bond-length distortion (BLD) as:

n 100 X |di − dm| BLD = [%] (4.1) n d i=1 m where BLD is expressed as a percentage, n is the number of Li−O bonds, and di and dm are the individual and mean Li−O bond lengths (Wenger and Armbruster, 1991). The BLD for the three lithium aluminosilicate minerals were calculated using the average values (dm) for both tetrahedral and octahedral lithium from Wenger and Armbruster(1991) as well as the individual bond lengths (d i) from Daniels and Fyfe(2001) (eucryptite), Clark et al.(1968) (spodumene) and Ross et al.(2015) (petalite). The calculated BLD was 1.28 for eucryptite (tetrahedral), 4.17 for spodumene (octahedral) and 1.02 for petalite (tetrahedral) (Fig. 4.5). We observe that the compound with the highest Li2O content

(eucryptite - distorted tetrahedra) has the highest BLD whereas the compound with the lowest Li2O content (petalite - quasi-ideal tetrahedra) has the smallest BLD. Increased distortion shifts the position of peak C to higher energy that increased distortion may actually strengthen the Li−O bond or increase lithium’s electron aﬃnity. A possible explanation to the correlation between increased bond strength and site distortion could be charge compensation due to the proximity of Al (which is much higher in eucryptite than in petalite). The distortion of spodumene is much higher than the tetrahedral minerals but due to the fact that it is octahedral no comparison should be made regarding the strength of the Li−O bond.

4.4.3 Complex silicates

This group of minerals contains a range of Li environments but shows very little variation in the edge position. In the cases of elbaite and lepidolite we can observe signiﬁcant pre-edge features similar to those in materials with weaker ionic bonds (e.g. Li2CO3). Additionally, the main edge seems to be composed of three contributions as was the case in the lithium halides and lithium sulphate (Pascal et al., 2014). The splitting of the three features, that is, the diﬀerence between the highest and lowest energy peaks is 2.6 eV and 3.6 eV in elbaite and lepidolite respectively. These values are both larger than the splitting value described in Li2SO4 which is ∼ 2 eV or that of eucryptite at 1.9 eV. The increased complexity of the environments in elbaite and lepidolite with respect to Li2SO4 and eucryptite may be the cause for this increased splitting of the triply degenerate p-like states (Pascal et al., 2014). Elbaite and lepidolite both have two lower energy shoulders to the highest intensity peak which is reminiscent of peak P observed in Li2CO3 and perhaps points to more covalent type bonding between lithium and oxygen (Pascal et al., 2014; Wang and Zuin, 2017). Chapter 4. A Li K -edge XANES study of salts and minerals 92

4.4.4 Phosphates

The phosphates have very similar main peak positions though extracting information from lithiophilite is diﬃcult due to the overlapping Mn M2,3-edge. Montebrasite on the other hand shows a very sharp absorption edge similar to that of Li2SO4 and lithium metasilicate. There are interesting similarities in the chemistry of montebrasite (P-bearing) and eucryptite (Si-bearing). The main edge is located at 61.8 eV for montebrasite and 61.2 eV for eucryptite. The electronegativity of phosphorous is higher than that of silicon. The edge position shifts to lower energy with decreasing electronegativity of the network former. This behaviour seems to follow the same trend as lithium halides (Handa et al., 2005) but due to the next-nearest neighbor instead of their ﬁrst coordination shell.

4.5 Conclusions

The Li K -edge XANES spectra of a variety of compounds and minerals were reported. The cause of the absorption edge is mainly due to s→p transitions leading to triply degenerate p-like states as was shown for LiF (Pascal et al., 2014). Lithium sulphate also shows a sharp absorption edge which has been shown to be caused by triply degenerate p-like states reminiscent of LiF (Pascal et al., 2014). Lithium carbonate shows a strong pre-edge which has been noted by Tsuji et al.(2002) as an indication of occupancy in the 2p orbital as is the case for Li2O and Li metal. Additionally, the edge feature in ∗ Li2CO3 is very similar to the edge in the C K -edge spectrum which is typically associated with the π states (Fister et al., 2011). Thus, a feature such as this can be used to estimate the covalency of the bond in question in other lithium compounds and minerals. Lithium metasilicate has a pronounced edge similar to that of Li2SO4, which points to a strong exciton process and an ionic character to the Li−O bond. It will be interesting to compare this information to future spectra of lithium silicate glasses. The lithium aluminosilicates were used to compare the Li site distortion to the edge energy and how these factors changed with increasing lithium content. We demonstrate that with higher distortion lithium appears to have a higher electron affinity. Perhaps, this can be explained by the role lithium must play in charge compensation when proximal to AlO4 tetrahedra in eucryptite. Much less can be said about the complex silicates because of the influence of other edges but we do note that the edge position does not vary significantly. In elbaite and lepidolite we observed the presence of pre-edge features similar to those in Li2CO3 which may indicate that the covalency of the Li−O bond is higher (Tsuji et al., 2002; Fister et al., 2011). The Li K -edge XANES measurements are useful to study of the bonding environment of lithium in a variety of different chemical compounds. Combining the experimental data with new numerical simulations on minerals especially their density of states would immensely improve our understanding of the XANES spectra. Finally, investigating other edges within these systems (e.g. Si L-edge and O K -edge) would be valuable to corroborate the interpretations from the Li K -edge.

4.6 Acknowledgements

The research described in this paper was performed at the Canadian Light Source, which is supported by the Canada Foundation for Innovation, Natural Sciences and Engineering Research Council of Canada, the University of Saskatchewan, the Government of Saskatchewan, Western Economic Diversiﬁcation Canada, the National Research Council Canada, and the Canadian Institutes of Health Research. A Chapter 4. A Li K -edge XANES study of salts and minerals 93 special thanks to Jean-Claude Boulliard, curator of the mineralogical collection of Universit´ePierre and Marie Curie, who provided all the mineral samples. GSH acknowledges funding from NSERC in the form of a discovery grant. Chapter 5

The eﬀect of alkaline-earth substitution on the Li K -edge of lithium silicate glasses

This chapter has been published in the Journal of Non-Crystalline Solids as:

• O’Shaughnessy, C., Henderson, G. S., Moulton, B. J., Zuin, L., Neuville, D. R. (2018a) The eﬀect of alkaline-earth substitution on the Li K -edge of lithium silicate glasses. Journal of Non-Crystalline Solids 500, 417–421.

94 Chapter 5. The Li K -edge of lithium silicate glasses 95

Abstract

We present the ﬁrst Li K -edge XANES study of a varied suite of lithium silicate (LS) and lithium alkaline-earth silicate (LMS) glasses. The LS series contains three glasses with compositions LS4

(20 Li2O−80 SiO2), LS2 (33 Li2O−67 SiO2) and LS1.5 (40 Li2O−60 SiO2) and the LMS series is comprised of four glasses of compositions 20 Li2O−20 M2O−60 SiO2 where M=Mg, Ca, Sr and Ba. The spectra were compared to those of known salts and minerals, particularly to lithium metasilicate (Li2SiO3) and lithium carbonate (Li2CO3). Crystalline lithium metasilicate has a sharp, strong intensity absorption edge whereas the lithium silicate glasses all have a weak intensity edge feature, similar to that of lithium carbonate (Li2CO3). The intensity of the edge is governed by the existence of a 2p electron and can be correlated with the ionicity of the Li−O bond. Therefore, the bonding environment in LS glasses differs considerably from their crystalline counterparts. The area of the absorption edge peak increases with the lithium content of the LS glasses. As for the LMS glasses the edge peak changes drastically depending on the alkaline-earth present. The peak area of the LMS glasses decreases with increasing charge density (Z/r2) from barium to magnesium. The area of the absorption edge peak in the LS glasses is larger than that of the LMS glasses. Thus, the addition of alkaline-earth elements to lithium silicate glasses modifies the glass network differently depending on the specific element, which in turn creates an altered lithium bonding environment as evidenced by their Li K -edge XANES spectra.

5.1 Introduction

Lithium is a very important element in the cycling of volatiles in magmatic settings and can be strongly isotopically fractionated in both silicic differentiates and the resulting gas condensates (Vlastélic et al., 2011). The structure of lithium silicate glasses has been studied for many years because both lithium metasilicate and disilicate glasses are exceptional examples of homogeneous nucleation (Zanotto and Leite, 1996; Soares et al., 2003; Fokin et al., 2006; Nascimento et al., 2011; Zanotto et al., 2015). Homogeneous nucleation is preferred to heterogeneous nucleation when attempting to control physical properties of glass ceramics for industrial purposes due to a more uniform distribution of crystals (Zanotto et al., 2015). Heat-resistant ceramics are typically multicomponent systems which include both Li2O and MgO in significant quantities (∼10 wt. %) (El-Shennawi et al., 1990; Kichkailo and Levitskii, 2005). Similarly, a class of phosphate-bearing glass-ceramics containing lithium disilicate as the main crystalline phase has been established as a biomaterial in dental applications (Krügeret al., 2013). Additionally, lithium-bearing silicate glasses are an important component in a variety of industrial applications: nuclear waste glasses (Abraitis et al., 2000) and in ion-conducting glasses (Voigt et al.,

2005; Ross et al., 2015) which contain a minimum of 3.5 wt. % Li2O. Lithium is a difficult element to probe as it is nearly invisible to X-ray diffraction (XRD). Therefore, element-specific probes, such as, X-ray absorption near-edge structure (XANES) spectroscopy are an excellent tool to determine the bonding environment of elements in amorphous materials (Henderson et al., 2014). Many Li K -edge XANES spectroscopy studies have been performed on a variety of Li- bearing compounds including battery components, salts and minerals (Tsuji et al., 2002; Handa et al., 2005; Yiu et al., 2013; Wang and Zuin, 2017; O’Shaughnessy et al., 2018b). Additionally, some of these materials have also been probed using X-ray Raman Spectroscopy and their spectra reveal similar features for the Li K -edge (Fister et al., 2011; Pascal et al., 2014). Chapter 5. The Li K -edge of lithium silicate glasses 96

In this study, we probed the Li K -edge of seven (7) glasses and two standard compounds in order to assess structural changes in the glasses. The two series (lithium silicate and lithium alkaline-earth silicate) enable us to explore the eﬀect of lithium content and alkaline-earth substitution on the lithium- bearing silicate glasses.

5.2 Methods

5.2.1 Glass synthesis

A variety of lithium silicate (LS4 (20 Li2O−80 SiO2), LS2 (33 Li2O−67 SiO2) and LS1.5

(40 Li2O−60 SiO2)) and lithium alkaline-earth silicate (20 Li2O−20 M2O−60 SiO2 named LiM where M=Mg, Ca, Sr and Ba) were synthesized using a melt-quench technique. The samples were prepared by combining the stoichiometric amounts of Li2CO3, MCO3 and SiO2 to make approximately 100 g batches. ◦ The mixtures were then calcined at 900 C for 24 hours to remove the CO2. The mixtures were placed in Pt crucibles and heated to 200 ◦C above their respective melting temperatures for 1 hour and were then quenched by placing the bottom of the crucible in water. Following the ﬁrst quench, samples were then mechanically crushed in a mortar and pestle and a second melt-quench cycle was performed. All ◦ samples were stored in a desiccator or 100 C furnace to avoid interaction with atmospheric H2O. The glass samples were analyzed using the electron microprobe at IPGP with a Camebax SX50. The lithium content was calculated using the diﬀerence method and the abundances are reported in wt. % in Table 5.1. The densities of the LS glasses were measured using a Berman density balance and their values are consistent with previously published data (Tischendorf et al., 1998; Doweidar et al., 1999). Additionally, all samples were analyzed using Raman spectroscopy to verify that the samples had not undergone crystallization. Two crystalline standards were used to compare to the glasses, lithium metasilicate crystal and lithium carbonate. The spectra of these standards are described in detail in O’Shaughnessy et al.(2018b). The lithium metasilicate crystal was synthesized by recrystallizing a lithium silicate glass sample prepared as outlined above, whereas the lithium carbonate was a high purity powder (>99%).

Table 5.1: Electron microprobe analysis of lithium-bearing glasses

Sample SiO2 Li2O MgO CaO SrO BaO LS1.5 75.17 (6) 24.83 - - - - LS2 79.97 (9) 20.03 - - - - LS4 88.99 (7) 11.01 - - - - LiMg 71.97 (5) 11.93 16.13 (7) - - - LiCa 67.73 (9) 11.21 - 21.13 (5) - - LiSr 57.45 9.52 - - 32.98 (8) - LiBa 49.63 (7) 8.22 - - - 42.23 (7)

*The abundances are reported in wt. % and the standard deviation is in parentheses.

5.2.2 X-ray Absorption Near-Edge Structure (XANES) Spectroscopy

We performed the Li K -edge (54.7 eV) XANES measurements at the Canadian Light Source Inc. (CLS), Saskatoon, Canada using the variable line spacing-plane grating monochromator (VLS-PGM) Chapter 5. The Li K -edge of lithium silicate glasses 97 beamline (Kasrai et al., 1991; Hu et al., 2007). The spectra were collected in ﬂuorescence yield (FLY) due to its sampling of a large volume of space which renders it more sensitive to the bulk of the sample compared to the total electron yield (TEY) (Jiang and Spence, 2006; Henderson et al., 2014; Moulton et al., 2016; Wang and Zuin, 2017; O’Shaughnessy et al., 2018b). The microchannel plate detector is mounted at 90◦ to the incident beam with the sample at 45◦ to the incident beam which negates self- absorption eﬀects (Kasrai et al., 1991). The beamline slits were opened to 50 mm × 50 mm with a resolution E/∆E > 10000 and the pressure in the experimental chamber was maintained below 10−7 torr for all measurements (Wang and Zuin, 2017). The spectra were collected in the range of 35-75 eV.

The ﬂuorescence signal was normalized to the incoming intensity (I0), measured by a Ni mesh located upstream of the sample chamber. A two-step polynomial background correction was performed on the pre-edge region (∼ 35-58 eV) and post-edge region (∼ 65-75 eV) and the spectra were normalized to their maximum intensity (as in O’Shaughnessy et al., 2018b).

Table 5.2: Peak positions in lithium-bearing compounds and minerals

Sample p A B C F G H Li metasilicate - 60.5 61.8 - 66.4 68.0 69.6 Li carbonate 58.8 - 61.8 - 67.2 69.2 - LS1.5 - 60.0 - 63.1 - 67.3 - LS2 - 60.1 - 63.0 - 67.3 - LS4 - 60.1 - 63.9 - 67.3 - LiMg - 60.2 - 63.1 - 68.5 - LiCa - 60.2 - 62.8 - 67.9 - LiSr - 60.2 - 64.0 - 68.6 - LiBa - 60.1 - 63.0 - 68.3 -

*All peak positions represent the energy of peak maximums and are reported in eV.

All edge positions are calculated by using the maximum intensity of a specific peak and are reported as such in eV (Fig. 5.2). In order to compare the values of the edge positions in our samples to those of previously published data we calibrated the value of the main absorption edge of LiCl by shifting its value to 60.8 eV as reported in the experiments of Tsuji et al.(2002) following the procedure of Wang and Zuin (2017) and O’Shaughnessy et al.(2018b). The data presented in this paper was collected during the same beamtime as LiCl and the spectra were shifted accordingly. Additionally, the peak positions for lithium metasilicate and lithium carbonate are also taken from O’Shaughnessy et al.(2018b) to compare to the values of the lithium glass samples. The peaks have been labeled by their relative positions and not by the specific electronic interaction as in O’Shaughnessy et al.(2018b). The spectra of the previously studied minerals show many features which were labeled and these labels have been used for the spectra of the glass samples in this study (O’Shaughnessy et al., 2018b). The spectra of reference standards are in agreement with previous XANES studies (Tsuji et al., 2002; Handa et al., 2005; Wang and Zuin, 2017). In order to determine the area of the absorption edge peak (peak A, in the glass samples) the spectra were area normalized and two hinge-points were determined (using positions obtained from the first derivative) to calculate the area underneath the spectra (as graphically depicted in Fig. 5.1). Chapter 5. The Li K -edge of lithium silicate glasses 98

LS2

E1 E2

Figure 5.1: Graphical depiction of the area calculation of the main peak of the lithium glasses using two hinge-points and a trapezoidal method.

5.3 Results

5.3.1 Lithium silicate glasses (LS)

The lithium silicate glasses display almost identical spectra with small diﬀerences in peak positions and relative intensities (Fig. 5.2 and Table 5.2). The spectra have three main features peaks A, C and

G. Peak A is similar to peak p in Li2CO3 which has been described in previous studies (Tsuji et al., 2002; Handa et al., 2005). The positions of peak A (∼60 eV) for the lithium silicate glass samples are listed in Table 5.2 and its position shows no compositional variation. The position of peak A in lithium metasilicate is located at higher energy than for the equivalent peak in the glasses. The highest intensity peak in all three glass samples is peak C located at ∼63 eV (Table 5.2). Finally, peak G is located at ∼67 eV (for the LS glasses) and is almost identical in shape, position and intensity in all three samples (Fig. 5.2 and Table 5.2).

5.3.2 Lithium alkaline-earth silicate glasses (LMS)

The LMS glasses also have three main features but their shape and positions show a large amount of variability depending upon the alkaline-earth element present in each sample. Peak A is located at ∼60 eV as in the LS glasses but its shape is very different (from a pronounced peak in LiBa to a shoulder-like feature in LiMg). The peak is difficult to resolve in the magnesium glass and becomes more apparent with increasing field strength of the alkaline-earth present, being most visible in the Ba-bearing glass Chapter 5. The Li K -edge of lithium silicate glasses 99

(see Fig. 5.3 and Table 5.2). There is a shoulder on the low energy side of peak A which is noticeable in some of the LMS glasses. This peak is diﬃcult to resolve because of how faint it is in most samples but it may point to the existence of a triply degenerate p-like state of lithium as discussed in Fister et al. (2011). The highest intensity peak is peak C, which, changes in shape and peak position depending on the alkaline-earth cation present, though these changes are not systematic. Finally, peak G varies in breadth, intensity, and position which diﬀers from the behaviour of this peak in the LS glasses.

C G

LS1.5 C G

C LS2 G A

LS4 A F G B H

Li2 SiO 3

B G H p

Li2 CO 3

Figure 5.2: The Li K-edge spectra of standard compounds (Li2CO3 and Li2SiO3) and three lithium silicate glasses (LS1.5, LS2 and LS4).

5.4 Discussion

When comparing the Li K -edge XANES spectra of lithium metasilicate and lithium carbonate one notices the difference in intensity of their first peaks. Lithium metasilicate has a clear, strong peak (peak A) whereas the first peak of lithium carbonate (peak p) is weak. The presence of a strong first peak in lithium metasilicate is akin to the spectra of the lithium halides (Handa et al., 2005). Lithium carbonate, on the other hand, has a first peak which is similar to that seen in Li metal and Li2O(Tsuji Chapter 5. The Li K -edge of lithium silicate glasses 100 et al., 2002; Pascal et al., 2014). The reason for this change in relative intensities has been described using the exciton process by Tsuji et al.(2002). The exciton phenomenon is due to the excitation of the valence band electron to the exciton level which causes the remaining core-hole to interact with this excited electron, as shown in Fig. 12 of Tsuji et al.(2002). In the Li K -edge XANES spectra of lithium metasilicate and the lithium halides the X-rays excite the 1s electron to the 2p orbital (exciton level). In these compounds neither the 2s or 2p electron exist because the ionic bond between lithium and its coordination atom is strong and the charge transfer is approximately equal to 1 (Tsuji et al., 2002). This leads to the interaction between the core-hole and the excited electron which produces a clear and strong peak. Conversely, if the 2s or 2p orbital exists then the ionic bond is weak and the strong constraint does not exist, leading to a weak peak as seen in Li metal, Li2O, and lithium carbonate.

C G A LiBa C G

A LiSr C G A LiCa

C G A LiMg C G A LS1.5 A F G B H

Li2 SiO 3

B G H Li2 CO 3 p

Figure 5.3: The Li K-edge spectra of standard compounds (Li2CO3 and Li2SiO3) and four lithium alkaline-earth silicate glasses (LiMg, LiCa, LiSr and LiBa). The spectrum of LS1.5 is also reported for comparison since it also contains 60 mol. % SiO2.

The ﬁrst peak in the LS series of glasses is very similar to the one observed in Li2CO3, though the energy position of this peak varies by at least 1 eV between Li2CO3 and the glasses. The area beneath peak A in the LS glasses has been calculated and is presented in Fig. 5.4a. There appears to be a Chapter 5. The Li K -edge of lithium silicate glasses 101 positive correlation between the Li content of the glasses and the area of peak A. Using these areas to describe the exciton process in their respective glasses we propose that with increasing lithium content the glasses have a stronger exciton. Thus, LS4 would have the weakest Li−O ionic bond and LS1.5 would have the strongest. That being said, the ionicity of the bonds in the glasses are clearly weaker than that observed in crystalline lithium metasilicate.

a) b) Ba LS1.5 Sr

Ca LS2

Mg LS4

Figure 5.4: a) Normalized area of peak A of the lithium silicate glasses as a function of Li content. b) Normalized area of peak A of the lithium alkaline-earth silicate glasses as a function of their charge density (Z/r2) as calculated using the atomic radii of Shannon and Prewitt(1969).

The presence of peak A in the LMS glasses also seems to indicate that the weak exciton process is present, though the morphology of the peak varies drastically depending on the alkaline-earth atom present. The area of peak A calculated for the LMS glasses is shown in Fig. 5.4b as a function of charge density (Z/r2) as calculated using the ionic radii of Shannon and Prewitt(1969). With increasing charge density (from barium to magnesium) we note a decrease in the area of peak A. Following the same aforementioned reasoning, we conclude that the ionicity of the Li−O bond is weakest in the magnesium- bearing glass and strongest in the barium-bearing glass. It is worth mentioning that the areas of the LMS glasses are all less than LS4 (which has the smallest area of the LS series). It follows that, with the replacement of lithium for an alkaline-earth, the strength of the Li−O ionic bond decreases. Interestingly, the spectrum of LiBa resembles that of LS1.5, we speculate that this is due to barium having a charge density similar to that of lithium and perhaps these elements aﬀect the network in a similar way despite the size diﬀerence of barium with respect to lithium.

5.5 Conclusions

The Li K -edge XANES spectra of standard compounds and two series of lithium-bearing silicate glasses (LS and LMS) were reported. The spectra of all the glasses differ dramatically from crystalline lithium metasilicate (Li2SiO3) confirming a substantially different bonding environment in the glasses compared to the crystalline counterpart.

The spectra of both the LS and LMS glasses appear to resemble that of lithium carbonate (Li2CO3) and other compounds, such as, Li metal or Li2O. The existence of a weak absorption edge feature (peak p for Li2CO3 and peak A of the LS and LMS glasses) has been shown to be a product of a weak exciton Chapter 5. The Li K -edge of lithium silicate glasses 102 process which is characterized by having an occupied 2p orbital (Tsuji et al., 2002). We believe that the area of this absorption edge peak is important for understanding the occupancy of these 2p orbitals. The calculated areas for the LS series demonstrate a positive correlation with the lithium content of these glasses whereas the areas of the LMS series demonstrate a negative correlation with the charge density of the alkaline-earth cation present in the glass. The change in the area of peak A points to a difference in the ionicity of the bond depending on both abundance of lithium and the presence of additional cations. The Li K -edge XANES measurements are useful to study of the bonding environment of lithium in a variety of different chemical structures and can help decipher the type of bonding present. In order to verify the findings additional work on both the Si L-edge and O K -edge should be conducted. Combining multiple XANES measurements at different elemental edges would allow for more robust assessments of the bonding of all elements in the glasses.

5.6 Acknowledgments

Conclusions

This thesis attempts to address many questions regarding the structure of alkali silicate glasses and melts. These glass samples are simplistic analogues to natural glasses and melts which can contain many additional types of elements. Applying these techniques to increasingly more complex glassy systems will lead to a much better understanding of both natural glasses and melts. The importance of understanding the atomic structure of glasses and melts cannot be understated as it controls almost every aspect of the macroscopic behaviour of these systems (e.g. viscosity, conductivity, etc.). These highlights reﬂect the order presented in the preceding chapters starting with the Raman spectroscopy (Chapters 2 & 3) followed by the Li K -edge XANES (Chapters 4 & 5) results:

6.1 Highlights of the Raman spectroscopy investigation on alkali silicate glasses

(i) 3- and 4-membered SiO4 rings: The abundance of 3- and 4-membered SiO4 rings varies with composition and this is reﬂected in the relative intensities of the D1 and D2 bands. Increase in

the proportion of 3-membered SiO4 rings is a known densiﬁcation mechanism and the ratio of ID1/ID2 tracks extremely well with glass density and molar volume for all alkali silicate glasses (O’Shaughnessy et al., 2017).

(ii) M−BO interactions: The presence of modifier bound bridging oxygen (M−BO) has implications for the structure of silicate glasses as they represent an additional oxygen environment (along with BO and NBO). These M−BO have been identified with the use of XPS (Nesbitt et al., 2015) and MD simulations (Du and Cormack, 2004) have also indicated their presence in sodium silicate glasses. This second type of BO must be taken into consideration when conducting spectroscopic studies on glasses as their effects can be significant.

(iii) The importance of lineshapes: The high frequency area (800-1300 cm−1) attributed to the n vibrations of the SiO4 symmetric stretch of the different Q species was fit using both Gaussian (Q4) and pseudo-Voigt (Q3,2,1) line shapes following Bancroft et al.(2018) and Nesbitt et al.(2019). This new method of fitting the high frequency resulted in more realistic fits with fewer unattributed bands which were a consequence of using solely Gaussian lineshapes. We believe that a similar

103 Chapter 6. Conclusions 104

methodology should be adopted in the future when ﬁtting the Qn species envelope of all silicate glasses and melts.

(iv) Qn species distributions: The presence of alkali in the vicinity of Q4 species have a direct influence on this increase in band intensity and these species are referred to as primed Q4 species.The total area of the Q4 bands decreases with increasing alkali content with the exception of the lithium silicate glasses. This is indicative of network depolymerization in which the addition of modifier cations ruptures Si−O−Si bonds and creates NBO. Similarly, multiple Q3 species have been identified leading to the asymmetry of the Q3 band. These species are created due to the interaction of modifier cations with BO which weakens Si−O force constants leading to a shift to lower wavenumbers for Q3-2 (one M−BO bond) and Q3-3 (two M−BO bonds) with respect to conventional Q3 (no M−BO bond) species. The area of the Q2 peak across all alkalis is due to the depolymerization of the silicate network with 2 increasing M2O content. The Q peak appears at increasingly higher concentrations of M2O with increasing cation size (from Li to Cs). We believe that cations with a higher charge density (Z/r2) cluster more readily than cations with a lower charge density. It is then probable that caesium silicate glasses have a more random distribution of modifier atoms than lithium silicate glasses.

(v) High temperature spectra and the melting mechanism of silicates: The three diﬀerent compositions (Cs5, Cs10, and Cs15) displayed a negative shift in the high frequency envelope as shown by both the barycentre trend and the Qn species relative abundances. Lower Qn species appear to be created, lending credence to the possibility that the melting reaction taking place is similar in nature to the depolymerization reaction: O2– + BO −−→ 2 NBO– . These ﬁndings agree with recent studies using the Toop-Samis model (Ottonello and Moretti, 2004) and have been further elaborated upon in Nesbitt et al.(2017a).

6.2 Highlights of the Li K -edge XANES investigation on lithium silicate minerals and glasses

(i) Peak intensity related ionicity of the bond: The spectrum of lithium metasilicate (Li2SiO3)

has a very sharp Li K -edge with features resembling those present in Li2SO4. This is indicative of a bond with a more ionic character due to the absence of a 2p electron (Tsuji et al., 2002; Pascal et al., 2014). Additionally, we see that varying amounts of aluminum produce diﬀerent spectral features in the lithium aluminosilicate minerals. We believe this is linked to the increased bond strength and site distortion of Li. Lithium acts as a charge compensator due to the proximity of Al rather than Si in these minerals. Montebrasite (P-bearing) and eucryptite (Si-bearing) edge positions are shifted by 0.6 eV which we believe is caused by the electronegativity of phosphorous being higher than that of silicon. The edge position shifts to lower energy with decreasing electronegativity of the network former. This behaviour seems to follow the same trend as lithium halides (Handa et al., 2005) but is due to the next-nearest neighbor instead of lithium’s ﬁrst coordination shell. Chapter 6. Conclusions 105

(ii) Lithium environment in silicate glasses: The spectra of all the glasses diﬀer dramatically from

crystalline lithium metasilicate (Li2SiO3) conﬁrming a substantially diﬀerent bonding environment in the glasses compared to the crystalline counterpart.

The spectra of both the LS and LMS glasses appear to resemble that of lithium carbonate (Li2CO3)

and other compounds, such as, Li metal or Li2O. The existence of a weak absorption edge feature

(peak p for Li2CO3 and peak A of the LS and LMS glasses) has been shown to be a product of a weak exciton process which is characterized by having an occupied 2p orbital (Tsuji et al., 2002). The calculated areas for the LS series demonstrate a positive correlation with the lithium content of these glasses whereas the areas of the LMS series demonstrate a negative correlation with the charge density of the alkaline-earth cation present in the glass. The LS glasses become more ionic with increasing lithium content whereas the LMS series become less ionic with increasing charge density.

6.3 Implications for the structural model of alkali silicate glasses

The current accepted model for the structure of alkali silicate glasses, the Modiﬁed Random Network (MRN), proposes that percolation channels form areas with a higher concentration of alkali (Greaves et al., 1981). These structures would then yield areas which are more organized than others. We propose that the broader peaks with a higher Gaussian component may be a result of variations in the local structure within these glasses as alkali is added to the network. This would then imply that the variation in the % Lorentzian lineshapes would be an indicator of the degree of ordering in the glasses. The presence of M−BO interactions is not a part of the MRN model, which only includes BO and NBO. However, the M−BO could reside within the covalent network but along the perimeter separating the percolation channels from the network (Nesbitt et al., 2015). Thus, the network is zoned with M−BO near the perimeter and BO are concentrated in the interior (zone of higher SiO2 content). There also appears to be an important transition of the glass network which occurs at ∼20 mol. %

Abraitis, P., Livens, F., Monteith, J., Small, J., Trivedi, D., Vaughan, D., Wogelius, R. (2000) The kinetics and mechanisms of simulated British Magnox waste glass dissolution as a function of pH, silicic acid activity and time in low temperature aqueous systems. Applied Geochemistry 15, 1399– 1416.

Albright, J. G. (1933) The crystal structure of lithium sulphate. Zeitschrift fur Kristallographie 84, 1158–1167.

Angell, C. A., Ngai, K. L., McKenna, G. B., McMillan, P. F., Martin, S. W. (2000) Relaxation in glassforming liquids and amorphous solids. Journal of Applied Physics 88, 3113–3157.

Arlt, T., Angel, R. J. (2000) Displacive phase transitions in c-centred clinopyroxenes: spodumene,

LiScSi2O6 and ZnSiO3. Physics and Chemistry of Minerals 27, 719–731.

Atkins, P. W. (1998) Physical chemistry. Oxford University Press 6.

Bancroft, G. M., Nesbitt, H. W., Henderson, G. S., O’Shaughnessy, C., Withers, A. C., Neuville, D. R. (2018) Lorentzian dominated lineshapes and linewidths for Raman symmetric stretch peaks (800-1200 cm−1) in Qn (n=1-3) species of alkali silicate glasses/melts. Journal of Non-Crystalline Solids 484, 72–83.

Bancroft, G. M., Nesbitt, H. W., Ho, R., Shaw, D. M., Tse, J. S., Biesinger, M. C. (2009) Toward a comprehensive understanding of solid-state core-level XPS linewidths: Experimental and theoretical studies on the Si 2p and O 1s linewidths in silicates. Physical Review B 80.

Barrio, R. A., Galeener, F. L., Martinez, E., Elliot, R. J. (1993) Regular ring dynamics in AX2 tetrahedral glasses. Physical Review B 48, 15672.

Bauchy, M., Micoulaut, M. (2011) From pockets to channels: Density-controlled diﬀusion in sodium silicates. Physical Review B 83, 184118.

Baur, W. H. (1959) Die Kristallstruktur des Edelamblygonits LiAlPO4(OH,F). Acta Crystallographica 12, 988–994.

Bell, R. J., Bird, N. F., Dean, P. (1968) The vibrational spectra of vitreous silica, germania and beryllium ﬂouride. Journal of Physics C: Solid State Physics 1, 299–303.

Bell, R. J., Dean, P. (1970) Atomic vibrations in vitreous silica. Discussions of the Faraday Society 50, 55–61.

106 BIBLIOGRAPHY 107

Born, M. (1939) Thermodynamics of crystals and melting. Journal of Chemical Physics 7, 591.

Brawer, S. A., White, W. B. (1975) Raman spectroscopic investigation of the structure of silicate glasses: I. The binary alkali silicates. Journal of Chemical Physics 63, 2421.

Buchenau, U., Nucker, N., Dianoux, A. J. (1984) Neutron scattering study of the low-frequency vibrations in vitreous silica. Physical Review Letters 53, 2316–2319.

Buchenau, U., Prager, M., Nucker, N., Dianoux, A. J., Ahmad, N., Phillips, W. A. (1986) Low-frequency modes in vitreous silica. Physical Review B 34, 5665–5673.

Cabaret, D., Bordage, A., Juhin, A., Arfaoui, M., Gaudry, E. (2010) First-principles calculations of X-ray absorption spectra at the K -edge of 3d transition metals: an electronic structure analysis of the pre-edge. Physical Chemistry Chemical Physics 12, 5619–5633.

Cabaret, D., Mauri, F., Henderson, G. S. (2007) Oxygen K -edge XANES of germanates investigated using ﬁrst-principles calculations. Physical Review B 75, 184205.

Cannillo, E., Mazzi, F., Rossi, G. (1966) The crystal structure of neptunite. Acta Crystallographica 21, 200–208.

Ching, W. Y., Murray, R. A., Lam, D. J., Veal, B. W. (1983) Comparative studies of electronic structures of sodium metasilicate and α and β phases of sodium disilicate. Physical Review B 28, 4724–4735.

Clark, J. R., Appleman, D. E., Papike, J. J. (1968) Bonding in eight ordered clinopyroxenes isostructural with diopside. Contributions to Mineralalogy and Petrology 20, 81–85.

Daniels, P., Fyfe, C. A. (2001) Al, Si order in the crystal structure of α-eucryptite (LiAlSiO4). American Mineralogist 86, 279–283. de Groot, F. (2001) High-Resolution X-ray Emission and X-ray Absorption Spectroscopy. Chemical Reviews 101, 1779–1808. de Groot, F. (2005) Multiplet eﬀects in X-ray spectroscopy. Coordination Chemistry Reviews 249, 31–63.

Deschamps, T., Kassir-Bodon, A., Sonneville, C., Margueritat, J., Martinet, C., de Ligny, D., Mermet, A., Champagnon, B. (2013) Permanent densiﬁcation of compressed silica glass: a Raman-density calibration curve. Journal of Physics: Condensed Matter 25, 025402.

Dingwell, D. B. (1989) Shear viscosities of ferrosilicate liquids. American Mineralogist 74, 1038–1044.

Donnay, G., Barton, R., J. (1972) Reﬁnement of the crystal structure of elbaite and the mechanism of tourmaline solid solution. TMPM Tschermaks Mineralogische und Petrographische Mitteilungen 18, 273–286.

Doweidar, H., Feller, S., Aﬀatigato, M., Tischendorf, B., Ma, C., Hammersten, E. (1999) Density and molar volume of extremely modiﬁed alkali silicate glasses. Physics and Chemistry of Glasses 40, 339– 344.

Du, J., Cormack, A. (2004) The medium range structure of sodium silicate glasses: a molecular dynamics simulation. Journal of Non-Crystalline Solids 349, 66–79. BIBLIOGRAPHY 108

Du, J., Corrales, L. R. (2006) Compositional dependence of the ﬁrst sharp diﬀraction peaks in alkali silicate glasses: A molecular dynamics study. Journal of Non-Crystalline Solids 352, 3255–3269.

Dupree, R., Holland, D., Williams, D. (1986) The structure of binary alkali silicate glasses. Journal of Non-Crystalline Solids 81, 185–200.

Duval, E., Mermet, A., Parc, R. L., Champagnon, B. (2004) About the origin of the boson peak in vitreous silica. Philosophical Magazine 84, 1433–1436.

Eﬁmov, A. M. (1999) Vibrational spectra, related properties, and structure of inorganic glasses. Journal of Non-Crystalline Solids 253, 95–118.

El-Shennawi, A., Omar, A., El-Ghannam, A. (1990) Crystallization of spodumene-lithium magnesium orthosilicate glasses. Ceramics International 16, 47–52.

Elliott, S. R. (1992) A uniﬁed model for the low-energy vibrational behaviour of amorphous solids. Europhysics Letters 19, 201–206.

Eyring, H., Ree, T., Hirai, N. (1958) Signiﬁcant structures in the liquid state I. Proceedings of the National Academy of Sciences 44, 683–691.

Farges, F., Brown, Gordon E., J., Rehr, J. J. (1996) Coordination chemistry of Ti(IV) in silicate glasses and melts: I. XAFS study of titanium coordination in oxide model compounds. Geochimica et Cos- mochimica Acta 60, 3023–3038.

Fister, T. T., Schmidt, M., Fenter, P., Johnson, C. S., Slater, M. D., Chan, M. K. Y., Shirley, E. L. (2011) Electronic structure of lithium battery interphase compounds: Comparison between inelastic X-ray scattering measurements and theory. Journal of Chemical Physics 135, 224513.

Fokin, V. M., Zanotto, E. D., Yuritsyn, N. S., Schmelzer, J. W. (2006) Homogeneous crystal nucleation in silicate glasses: A 40 years perspective. Journal of Non-Crystalline Solids 352, 2681–2714.

Frantz, J. D., Mysen, B. O. (1995) Raman spectra and structure of BaO−SiO2, SrO−SiO2 and

CaO−SiO2 melts to 1600 C. Chemical Geology 121, 155–176.

Galeener, F. (1982) Planar rings in glasses. Solid State Communications 44, 1037–1040.

Galeener, F. L. (1979) Band limits and the vibrational spectra of tetrahedral glasses. Physical Review B 19, 4292–4297.

Galeener, F. L., Lucovsky, G. (1976) Longitudinal optical vibrations in glasses: GeO2 and SiO2. Physical Review Letters 37, 1474–1478.

18 Galeener, F. L., Mikkelsen, J. C. (1981) Vibrational dynamics in O-substituted vitreous SiO2. Physical Review B 23, 5527–5530.

Galeener, F. L., Thorpe, M. F. (1983) Rings in central-force network dynamics. Physical Review B 28, 5802–5813.

Garvie, L. A., Buseck, P. R. (2004) Unoccupied states of pyrite probed by electron energy-loss spectroscopy (EELS). American Mineralogist 89, 485–491. BIBLIOGRAPHY 109

Gatta, G. D., Danisi, R. M., Adamo, I., Meven, M., Diella, V. (2012) A single-crystal neutron and X-ray diﬀraction study of elbaite. Physics and Chemistry of Minerals 39, 577–588.

Gedeon, O., Machacek, J., Charvatova, S. (2010) Changes of structure in potassium depleted binary silicate glass studied by MD. Journal of Non-Crystalline Solids 356, 2473–2479.

Geller, S., Durand, J. L. (1960) Reﬁnement of the structure of LiMnPO4. Acta Crystallographica 13, 325–331.

Giacomazzi, L., Umari, P., Pasquarello, A. (2009) Medium-range structure of vitreous SiO2 obtained through ﬁrst-principles investigation of vibrational spectra. Physical Review B 79, 064202.

Glock, K., Hirsch, O., Rehak, P., Thomas, B., Jager, C. (1998) Novel opportunities for studying the short and medium range order of glasses by MAS NMR, 29Si double quantum NMR and IR spectroscopies. Journal of Non-Crystalline Solids 232–234, 113–118.

Gouadec, G., Colomban, P. (2007) Raman Spectroscopy of nanomaterials: How spectra relate to disorder, particle size and mechanical properties. Progress in Crystal Growth and Characterization of Materials 53, 1–56.

Greaves, G. (1985) EXAFS and the structure of glass. Journal of Non-Crystalline Solids 71, 203–217.

Greaves, G. N., Fontaine, A., Lagarde, P., Raoux, D., Gurman, S. J. (1981) Local structure of silicate glasses. Nature 293, 611–616.

Groat, L. A., Chakoumakos, B. C., Brouwer, D. H., Hoﬀman, C. M., Fyfe, C. A., Morell, H., Schultz,

A. J. (2003) The amblygonite (LiAlPO4(F))-montebrasite (LiAlPO4(OH)) solid solution: A combined powder and single-crystal neutron diﬀraction and solid-state 6Li MAS, CP MAS, and REDOR NMR study. American Mineralogist 88, 195–210.

Guggenheim, S. (1981) Cation ordering in lepidolite. American Mineralogist 66, 1221–1232.

Guggenheim, S., Bailey, S. W. (1977) The reﬁnement of zinnwaldite-1M subgroup symmetry. American Mineralogist 62, 1158–1167.

Gurevich, V. L., Parshin, D. A., Schober, H. R. (2003) Anharmonicity, vibrational instability, and the Boson peak in glasses. Physical Review B 67, 094203.

Handa, K., Kojima, K., Taniguchi, K., Ozutsumi, K., Ikeda, S. (2005) Studies on the extremely soft X-ray absorption spectrometry at BL2. Memoires of the SR Center Ritsumeikan University 7, 3–6.

Hatert, F., Ottolini, L., Wouters, J., Fontan, F. (2012) A structural study of the lithiophilite-sicklerite series. Canadian Mineralogist 50, 843–854.

Hawthorne, F. C., Henry, D. J. (1999) Classiﬁcation of minerals of the tourmaline group. European Journal of Mineralogy 11, 201–215.

Hehlen, B., Courtens, E., Vacher, R., Yamanaka, A., Kataoka, M., Inoue, K. (2000) Hyper-Raman Scattering Observation of the Boson Peak in Vitreous Silica. Physical Review Letters 84, 5355–5358.

Hehlen, B., Neuville, D. R. (2015) Raman response of network modiﬁer cations in alumino-silicate glasses. Journal of Physical Chemistry B 119, 4093–4098. BIBLIOGRAPHY 110

Hehlen, B., Simon, G. (2012) The vibrations of vitreous silica observed in hyper-Raman scattering. Journal of Raman Spectroscopy 43, 1941–1950.

Hemley, R. J., Mao, H. K., Bell, P. M., Mysen, B. O. (1986) Raman spectroscopy of SiO2 glass at high pressure. Physical Review Letters 57, 747–750.

Henderson, G. (1995) A Si K -edge EXAFS/XANES study of sodium silicate glasses. Journal of Non- Crystalline Solids 183, 43–50.

Henderson, G. S. (2005) The structure of silicate melts: a glass perspective. The Canadian Mineralogist 43, 1921–1958.

Henderson, G. S., Calas, G., Stebbins, J. F. (2006) The structure of silicate glasses and melts. Elements 2, 269–273.

Henderson, G. S., de Groot, F. M. F., Moulton, B. J. A. (2014) X-ray Absorption Near-Edge Structure (XANES) Spectroscopy. Reviews in Mineralogy and Geochemistry 78, 75–138.

Henderson, G. S., Neuville, D. R., Cochain, B., Cormier, L. (2009) The structure of GeO2−SiO2 glasses and melts: A Raman spectroscopy study. Journal of Non-Crystalline Solids 355, 468–474.

Hesse, K. (1985) Crystal structures of natural and synthetic α-eucryptite, LiAlSiO4. Zeitschrift fur Kristallographie 172, 147–151.

Hesse, K. F. (1977) Reﬁnement of the crystal structure of lithium polysilicate. Acta Crystallographica Section B Structural Crystallography and Crystal Chemistry 33, 901–902.

Hodroj, A., Simon, P., Florian, P., Chopinet, M.-H., Vaills, Y. (2013) Phase separation and spatial morphology in sodium silicate glasses by afm, light scattering and nmr. Journal of the American Ceramic Society 96, 2454–2460.

Hu, Y. F., Zuin, L., Wright, G., Igarashi, R., McKibben, M., Wilson, T., Chen, S. Y., Johnson, T., Maxwell, D., Yates, B. W., et al. (2007) Commissioning and performance of the variable line spacing plane grating monochromator beamline at the Canadian Light Source. Review of Scientiﬁc Instruments 78, 083109.

Huang, C., Cormack, A. N. (1991) Structural diﬀerences and phase separation in alkali silicate glasses. Journal of Chemical Physics 95, 3634–3642.

Ida, T., Ando, M., Toraya, H. (2000) Extended pseudo-Voigt function for approximating the Voigt proﬁle. Journal of Applied Crystallography 33, 1311–1316.

Ingram, M. D. (1989) Ionic conductivity and glass structure. Philosophical Magazine B 60, 729–740.

Ispas, S., Charpentier, T., Mauri, F., Neuville, D. (2010) Structural properties of lithium and sodium tetrasilicate glasses: Molecular dynamics simulations versus NMR experimental and ﬁrst-principles data. Solid State Sciences 12, 183–192.

Ispas, S., Zotov, N., De Wispelaere, S., Kob, W. (2005) Vibrational properties of a sodium tetrasilicate glass: Ab initio versus classical force ﬁelds. Journal of Non-Crystalline Solids 351, 1144–1150. BIBLIOGRAPHY 111

Jacobs, D. J., Thorpe, M. F. (1996) Generic rigidity percolation in two dimensions. Physical Review E 53, 3682–3693.

Jambon, A., Semet, M. P. (1978) Lithium diﬀusion in silicate glasses of albite, orthoclase, and obsidian composition: An ion-microprobe determination. Earth and Planetary Science Letters 37, 445–450.

Jiang, N., Spence, J. C. (2006) Interpretation of Oxygen K pre-edge peak in complex oxides. Ultrami- croscopy 106, 215–219.

Jiang, N., Spence, J. C. H. (2004) Core-hole eﬀects on electron energy-loss spectroscopy of Li2O. Physical Review B 69, 115112.

Kalampounias, A. G., Nasikas, N. K., Papatheodorou, G. N. (2009) Glass formation and structure in

the MgSiO3−Mg2SiO4 pseudobinary system: From degraded networks to ioniclike glasses. Journal of Chemical Physics 131, 114513.

Kalampounias, A. G., Yannopoulos, S. N., Papatheodorou, G. N. (2006) Temperature-induced structural changes in glassy, supercooled, and molten silica from 77 to 2150 K. Journal of Chemical Physics 124, 014504.

Karki, B. B. (2010) First-principles molecular dynamics simulations of silicate melts: Structural and dynamical properties. Reviews in Mineralogy and Geochemistry 71, 355–389.

Kasrai, M., Fleet, M. E., Bancroft, G. M., Tan, K. H., Chen, J. M. (1991) X-ray-absorption near-edge structure of alkali halides: The interatomic-distance correlation. Physical Review B 43, 1763–1772.

Kichkailo, O. V., Levitskii, I. A. (2005) Lithium-bearing heat-resistant ceramics (a review). Glass and Ceramics 62, 178–183.

Kikas, A., N˜ommiste,E., Ruus, R., Saar, A. (2001) Core excitons in na k photoabsorption of naf: resonant auger spectroscopy. Physical Review B 64, 235120.

Kikkawa, J., Terada, S., Gunji, A., Haruta, M., Nagai, T., Kurashima, K., Kimoto, K. (2014) Un-

derstanding Li K -edge structure and interband transitions in LixCoO2 by electron energy-loss spectroscopy. Applied Physics Letters 104, 114105.

Kim, S. S., Sanders, T. H. (1991) Thermodynamic modeling of phase diagrams in binary alkali silicate systems. Journal of the American Ceramic Society 74, 1833–1840.

Kirkpatrick, S. (1973) Percolation and conduction. Reviews of Modern Physics 45, 574–588.

Kobayashi, H., Emura, S., Arachi, Y., Tatsumi, K. (2007) Investigation of inorganic compounds on the surface of cathode materials using Li and O K -edge XANES. Journal of Power Sources 174, 774–778.

Kogure, T., Bunno, M. (2004) Investigation of polytypes in lepidolite using electron back-scattered diﬀraction. American Mineralogist 89, 1680–1684.

Koroleva, O. N., Anﬁlogov, V. N., Shatskiy, A., Litasov, K. D. (2013) Structure of Na2O - SiO2 melt as a function of composition: in-situ Raman spectroscopic study. Journal of Non-Crystalline Solids 375, 62–68. BIBLIOGRAPHY 112

Kr¨uger,S., Deubener, J., Ritzberger, C., H¨oland,W. (2013) Nucleation kinetics of lithium metasilicate

in ZrO2-bearing lithium disilicate glasses for dental application. International Journal of Applied Glass Science 4, 9–19.

Le Losq, C., Neuville, D. R. (2013) Eﬀect of the Na/K mixing on the structure and the rheology of tectosilicate silica-rich melts. Chemical Geology 346, 57–71.

Le Losq, C., Neuville, D. R., Chen, W., Florian, P., Massiot, D., Zhou, Z., Greaves, G. N. (2017) Percolation channels: a universal idea to describe the atomic structure and dynamics of glasses and melts. Scientiﬁc Reports 7, 16490.

Le Losq, C., Neuville, D. R., Florian, P., Henderson, G. S., Massiot, D. (2014) The role of Al3+ on rheology and structural changes in sodium silicate and aluminosilicate glasses and melts. Geochimica et Cosmochimica Acta 126, 495–517.

Le Losq, C., Neuville, D. R., Moretti, R., Roux, J. (2012) Determination of water content in silicate glasses using raman spectrometry: Implications for the study of explosive volcanism. American Min- eralogist 97, 779–790.

Lee, S. K., Eng, P. J., Mao, H. (2014) Probing of pressure-induced bonding transitions in crystalline and amorphous earth materials: Insights from X-ray Raman scattering at high pressure. Reviews in Mineralogy and Geochemistry 78, 139–174.

Lennard-Jones, J. E., Devonshire, A. F. (1939) Critical and co-operative phenomena IV. a theory of disorder in solids and liquids and the process of melting. Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences 170, 464–484.

Li, C.-T., Peacor, D. R. (1968) The crystal structure of LiAlSi2O6-II (β spodumene). Zeitschrift fur Kristallographie 126, 46–65.

Liebau, F. (1961) Untersuchungen an schichtsilikaten des formeltyps (Am(Si2O5)n) III. zur kristallstruk-

tur von petalit LiAlSi4O10. Acta Crystallographica 14, 399–406.

London, D. (2005) Granitic pegmatites: an assessment of current concepts and directions for the future. Lithos 80, 281–303.

Loudon, R. (1963) Theory of stimulated Raman scattering from lattice vibrations. Proceedings of the Physical Society 82, 393–400.

Lu, Y.-C., Kwabi, D. G., Yao, K. P. C., Harding, J. R., Zhou, J., Zuin, L., Shao-Horn, Y. (2011) The

discharge rate capability of rechargeable Li−O2 batteries. Energy Environmental Sciences 4, 2999.

Maehara, T., Yano, T., Shibata, S. (2005) Structural rules of phase separation in alkali silicate melts analyzed by high-temperature Raman spectroscopy. Journal of Non-Crystalline Solids 351, 3685–3692.

Maekawa, H., Maekawa, T., Kawamura, K., Yokokawa, T. (1991) The structural groups of alkali silicate glasses determined from 29Si MAS-NMR. Journal of Non-Crystalline Solids 127, 53–64.

Malfait, W. J. (2009) Quantitative Raman spectroscopy: speciation of cesium silicate glasses. Journal of Raman Spectroscopy 40, 1895–1901. BIBLIOGRAPHY 113

Marchand, P., Marmet, L. (1983) Binomial smoothing ﬁlter: A way to avoid some pitfalls of least-squares polynomial smoothing. Review of Scientiﬁc Instruments 54, 1034–1041.

Matson, D. W., Sharma, S. K., Philpotts, J. A. (1983) The structure of high-silica alkali-silicate glasses. a raman spectroscopic investigation. Journal of Non-Crystalline Solids 58, 323–352.

Mauchamp, V., Boucher, F., Ouvrard, G., Moreau, P. (2006) Ab initio simulation of the electron energy-

loss near-edge structures at the Li K edge in Li, Li2O , and LiMn2O4. Physical Review B 74, 115106.

Mauchamp, V., Moreau, P., Ouvrard, G., Boucher, F. (2008) Local ﬁeld eﬀects at Li K -edges in electron

energy-loss spectra of Li, Li2O and LiF. Physical Review B 77, 045117.

McMillan, P., Poe, B., Gillet, P., Reynard, B. (1994) A study of SiO2 glass and supercooled liquid to 1950 K via high-temperature Raman spectroscopy. Geochimica et Cosmochimica Acta 58, 3653–3664.

McMillan, P. F. (1984) Structural studies of silicate glasses and melts - applications and limitations of Raman spectroscopy. American Mineralogist 69, 622–644.

McMillan, P. F. (1989) Raman spectroscopy in mineralogy and geochemistry. Annual Review of Earth and Planetary Sciences 17, 255–279.

McMillan, P. F., Hofmeister, A. M. (1988) Infrared and raman spectroscopy. Reviews in Mineralogy and Geology 18, 99–159.

McMillan, P. F., Wolf, G. H. (1995) Vibrational spectra of silicate liquids. Reviews in Mineralogy 32, 247–315.

McMillan, P. F., Wolf, G. H., Poe, B. T. (1992) Vibrational spectroscopy of silicate liquids and glasses. Chemical Geology 96, 351–366.

Meyer, A., Horbach, J., Kob, W., Kargl, F., Schober, H. (2004) Channel formation and intermediate range order in sodium silicate melts and glasses. Physical Review Letters 93, 027801.

Micoulaut, M. (2008) Amorphous materials: Properties, structure, and durability: Constrained interactions, rigidity, adaptative networks, and their role for the description of silicates. American Mineralo- gist 93, 1732–1748.

Micoulaut, M., Phillips, J. C. (2003) Rings and rigidity transitions in network glasses. Physical Review B 67, 104204.

Moulton, B. J., Henderson, G. S., Sonneville, C., O’Shaughnessy, C., Zuin, L., Regier, T., de Ligny, D. (2016) The structure of haplobasaltic glasses investigated using X-ray absorption near edge structure

(XANES) spectroscopy at the Si, Al, Mg, and O K -edges and Ca, Si, and Al L2,3-edges. Chemical Geology 420, 213–230.

Mozzi, R. L., Warren, B. E. (1969) The structure of vitreous silica. Journal of Applied Crystallography 2, 164–172.

Mysen, B. O. (1995) Experimental in-situ high-temperature studies of properties and structure of silicate melts relevant to magmatic processes. European Journal of Mineralogy 7, 745–766. BIBLIOGRAPHY 114

Mysen, B. O., Finger, L. W., Virgo, D., Seifert, F. A. (1982a) Curve-ﬁtting of Raman spectra of silicate glasses. American Mineralogist 67, 686–695.

Mysen, B. O., Frantz, J. D. (1992) Raman spectroscopy of silicate melts at magmatic temperatures:

Na2O−SiO2,K2O−SiO2 and Li2O−SiO2 binary compositions in the temperature range 25-1475 c. Chemical Geology 96, 321–332.

Mysen, B. O., Frantz, J. D. (1994) Silicate melts at magmatic temperatures: in-situ structure determination to 1651 c and eﬀect of temperature and bulk composition on the mixing behavior of structural units. Contributions to Mineralogy and Petrology 117, 1–14.

Mysen, B. O., Richet, P. (2005) Silicate Glasses and Melts: Properties and Structure. Elservier, Ams- terdam.

Mysen, B. O., Virgo, D., Seifert, F. A. (1982b) The structure of silicate melts: Implications for chemical and physical properties of natural magma. Reviews of Geophysics 20, 353–383.

Nascimento, M. L. F., Fokin, V. M., Zanotto, E. D., Abyzov, A. S. (2011) Dynamic processes in a silicate liquid from above melting to below the glass transition. Journal of Chemical Physics 135, 194703.

Nesbitt, H. W., Bancroft, G. M., Henderson, G., Ho, R., Dalby, K., Huang, Y., Yan, Z. (2011) Bridging, 2− non-bridging and free (O ) oxygen in Na2O - SiO2 glasses: An X-ray photoelectron spectroscopic (XPS) and nuclear magnetic resonance (NMR) study. Journal of Non-Crystalline Solids 357, 170–180.

Nesbitt, H. W., Bancroft, G. M., Henderson, G. S. (2018) Temperature dependence of Raman shifts and line widths for Q0 and Q2 crystals of silicates, phosphates, and sulfates. American Mineralogist 103, 966–976.

Nesbitt, H. W., Bancroft, G. M., Henderson, G. S., Richet, P., O’Shaughnessy, C. (2017a) Melting, crystallization, and the glass transition: Toward a uniﬁed description for silicate phase transitions. American Mineralogist 102, 412–420.

Nesbitt, H. W., Henderson, G., Bancroft, G. M., O’Shaughnessy, C. (2017b) Electron densities over Si and O atoms of tetrahedra and their impact on Raman stretching frequencies and Si-NBO force constants. Chemical Geology 461, 65–74.

Nesbitt, H. W., Henderson, G. S., Bancroft, G. M., Ho, R. (2015) Experimental evidence for Na coordination to bridging oxygen in Na-silicate glasses: Implications for spectroscopic studies and for the modiﬁed random network model. Journal of Non-Crystalline Solids 409, 139–148.

Nesbitt, H. W., O’Shaughnessy, C., Henderson, G. S., Michael Bancroft, G., Neuville, D. R. (2019) Factors aﬀecting line shapes and intensities of Q3 and Q4 Raman bands of Cs silicate glasses. Chemical Geology 505, 1–11.

Neuville, D. R., Cormier, L., Flank, A.-M., Prado, R. J., Lagarde, P. (2004) Na K -edge XANES spectra of minerals and glasses. European Journal of Mineralogy 16, 809–816.

Neuville, D. R., de Ligny, D., Henderson, G. S. (2014) Advances in Raman Spectroscopy Applied to Earth and Material Sciences. Reviews in Mineralogy and Geochemistry 78, 509–541. BIBLIOGRAPHY 115

Neuville, D. R., Ligny, D. d., Cormier, L., Henderson, G. S., Roux, J., Flank, A.-M., Lagarde, P. (2009) The crystal and melt structure of spinel and alumina at high temperature: An in-situ XANES study at the Al and Mg K -edge. Geochimica et Cosmochimica Acta 73, 3410–3422.

Newville, M., Stensitzki, T., Allen, D. B., Ingargiola, A. (2014) LMFIT: Non-Linear Least-Square Min- imization and Curve-Fitting for Python. Zenodo.

Ogasawara, Y., Hibino, M., Kobayashi, H., Kudo, T., Asakura, D., Nanba, Y., Hosono, E., Nagamura,

N., Kitada, Y., Honma, I., et al. (2015) Charge/discharge mechanism of a new co-doped Li2O cathode material for a rechargeable sealed lithium-peroxide battery analyzed by X-ray absorption spectroscopy. Journal of Power Sources 287, 220–225.

Olovsson, W., Tanaka, I., Mizoguchi, T., Puschnig, P., Ambrosch-Draxl, C. (2009a) All-electron Bethe- Salpeter calculations for shallow-core X-ray absorption near-edge structures. Physical Review B 79, 041102.

Olovsson, W., Tanaka, I., Puschnig, P., Ambrosch-Draxl, C. (2009b) Near-edge structures from ﬁrst principles all-electron Bethe-Salpeter equation calculations. Journal of Physics: Condensed Matter 21, 104205.

O’Shaughnessy, C., Henderson, G. S., Moulton, B. J., Zuin, L., Neuville, D. R. (2018a) The eﬀect of alkaline-earth substitution on the Li K -edge of lithium silicate glasses. Journal of Non-Crystalline Solids 500, 417–421.

O’Shaughnessy, C., Henderson, G. S., Moulton, B. J. A., Zuin, L., Neuville, D. R. (2018b) A Li K -edge XANES study of salts and minerals. Journal of Synchrotron Radiation 25, 543–551.

O’Shaughnessy, C., Henderson, G. S., Nesbitt, H. W., Bancroft, G. M., Neuville, D. R. (2017) Structure- property relations of caesium silicate glasses from room temperature to 1400 K: Implications from density and Raman spectroscopy. Chemical Geology 461, 82–95.

Ottonello, G., Moretti, R. (2004) Lux-ﬂood basicity of binary silicate melts. Journal of Physics and Chemistry of Solids 65, 1609–1614.

Pascal, T. A., Boesenberg, U., Kostecki, R., Richardson, T. J., Weng, T.-C., Sokaras, D., Nordlund, D., McDermott, E., Moewes, A., Cabana, J., et al. (2014) Finite temperature eﬀects on the X-ray absorption spectra of lithium compounds: First-principles interpretation of X-ray Raman measurements. Journal of Chemical Physics 140, 034107.

Pasquarello, A., Sarnthein, J., Car, R. (1998) Dynamic structure factor of vitreous silica from ﬁrst principles: Comparison to neutron-inelastic-scattering experiments. Physical Review B 57, 14133– 14140.

Pedesseau, L., Ispas, S., Kob, W. (2015a) First-principles study of a sodium borosilicate glass-former. I. the liquid state. Physical Review B 91.

Pedesseau, L., Ispas, S., Kob, W. (2015b) First-principles study of a sodium borosilicate glass-former. II. the glass state. Physical Review B 91. BIBLIOGRAPHY 116

Phillips, J. (1984) Microscopic origin of anomalously narrow Raman lines in network glasses. Journal of Non-Crystalline Solids 63, 347–355.

Prado, R. J., Flank, A.-M. (2005) Sodium K -edge XANES calculations in ”NaCl” type structures. Physica Scripta B T115, 165–167.

Rehr, J. J., Albers, R. C. (2000) Theoretical approaches to X-ray absorption ﬁne structure. Reviews of Modern Physics 72, 621–654.

Richet, N. F. (2009) Heat capacity and low-frequency vibrational density of states. inferences for the boson peak of silica and alkali silicate glasses. Physica B: Condensed Matter 404, 3799–3806.

Richet, P., Mysen, B. O., Andrault, D. (1996) Melting and premelting of silicates: Raman spectroscopy

and X-ray diﬀraction of Li2SiO3 and Na2SiO3. Physics and Chemistry of Minerals 23, 157–172.

Rieder, M. (1968) Zinnwaldite: Octahedral ordering in lithium-iron micas. Science 160, 1338–1340.

Ross, N. L., Zhao, J., Slebodnick, C., Spencer, E. C., Chakoumakos, B. C. (2015) Petalite under pressure: Elastic behavior and phase stability. American Mineralogist 100, 714–721.

Rossano, S., Mysen, B. (2012) Raman Spectroscopy of Silicate Glasses and Melts in Geological Systems. Raman spectroscopy applied to Earth sciences and cultural heritage 12, 321–366.

Roy, R., Osborn, E. F. (1949) The system lithium metasilicate-spodumene-silica. Journal of the American Chemical Society 71, 2086–2095.

Sanchez-Bajo, F., Cumbrera, F. L. (1997) The use of the pseudo-Voigt function in the variance method of X-ray line-broadening analysis. Journal of Applied Crystallography 30, 427–430.

Sarnthein, J. (1997) Origin of the high-frequency doublet in the vibrational spectrum of vitreous SiO2. Science 275, 1925–1927.

Sartori, F., Franzini, M., Merlino, S. (1973) Crystal structure of a 2m2 lepidolite. Acta Crystallographica Section B: Structural Crystallography and Crystal Chemistry 29, 573–578.

Sawyer, R., Nesbitt, H. W., Bancroft, G. M., Thibault, Y., Secco, R. A. (2015) Spectroscopic studies of oxygen speciation in potassium silicate glasses and melts. Canadian Journal of Chemistry 93, 60–73.

Sawyer, R., Nesbitt, H. W., Secco, R. (2012) High resolution X-ray photoelectron spectroscopy (XPS)

study of K2O - SiO2 glasses: Evidence for three types of O and at least two types of Si. Journal of Non-Crystalline Solids 358, 290–302.

Sen, P. N., Thorpe, M. F. (1977) Phonons in AX2 glasses: From molecular to band-like modes. Physical Review B 15, 4030–4038.

Sen, S., Youngman, R. (2003) NMR study of Q-speciation and connectivity in K2O−SiO2 glasses with high silica content. Journal of Non-Crystalline Solids 331, 100–107.

Shannon, R. D., Prewitt, C. T. (1969) Eﬀective ionic radii in oxides and ﬂuorides. Acta Crystallographica Section B: Structural Crystallography and Crystal Chemistry 25, 925–946. BIBLIOGRAPHY 117

Sharma, S. K., Mammone, J. F., Nicol, M. F. (1981) Raman investigation of ring conﬁgurations in vitreous silica. Nature 292, 140–141.

Soares, P.C, J., Zanotto, E., Fokin, V., Jain, H. (2003) TEM and XRD study of early crystallization of lithium disilicate glasses. Journal of Non-Crystalline Solids 331, 217–227.

Soltay, L. G., Henderson, G. S. (2005) The structure of Lithium-containing silicate and germanate glasses. Canadian Mineralogist 43, 1643–1651.

Sonneville, C., Mermet, A., Champagnon, B., Martinet, C., Margueritat, J., de Ligny, D., Deschamps, T., Balima, F. (2012) Progressive transformations of silica glass upon densiﬁcation. Journal of Chemical Physics 137, 124505.

Stebbins, J. F. (1987) Identiﬁcation of multiple structural species in silicate glasses by 29Si NMR. Nature 330, 465–467.

Stebbins, J. F., McMillan, P. F., Dingwell, D. B. (1995) Structure, dynamics and properties of silicate melts. Reviews in Mineralogy 32.

Surovtsev, N. V., Sokolov, A. P. (2002) Frequency behavior of Raman coupling coeﬃcient in glasses. Physical Review B 66, 054205.

Taguchi, N., Kitta, M., Sakaebe, H., Kohyama, M., Akita, T. (2015) Lithium analysis using reﬂection EELS for lithium compounds. Journal of Electron Spectroscopy and Related Phenomena 203, 40–44.

Thorpe, M. (1985) Rigidity percolation in glassy structures. Journal of Non-Crystalline Solids 76, 109– 116.

Tischendorf, B., Ma, C., Hammersten, E., Venhuizen, P., Peters, M., Aﬀatigato, M., Feller, S. (1998) The density of alkali silicate glasses over wide compositional ranges. Journal of Non-Crystalline Solids 239, 197–202.

Tomascak, P. B. (2004) Developments in the understanding and application of lithium isotopes in the earth and planetary sciences. Reviews in Mineralogy and Geochemistry 55, 153–195.

Toop, G. W., Samis, C. S. (1962a) Activities of ions in silicate melts. Transactions of the Metallurgical Society of AIME 224, 878–887.

Toop, G. W., Samis, C. S. (1962b) Some new ionic concepts of silicate slags. Canadian Metallurgical Quarterly 1, 129–152.

Tribaudino, M., Mantovani, L., Bersani, D., Lottici, P. P. (2012) Raman spectroscopy of (CaMg)MgSi2O6 clinopyroxenes. American Mineralogist 97, 1339–1347.

Tsuji, J., Nakamatsu, H., Mukoyama, T., Kojima, K., Ikeda, S., Taniguchi, K. (2002) Lithium K -edge XANES spectra for lithium compounds. X-Ray Spectrometry 31, 319–326.

Uchino, T., Yoko, T. (1999) Sodium and lithium environments in single- and mixed-alkali silicate glasses. an ab initio molecular orbital study. Journal of Physical Chemistry B 103, 1854–1858. BIBLIOGRAPHY 118

Uhlig, H., Hoﬀmann, M. J., Lamparter, H.-P., Aldinger, F., Bellissent, R., Steeb, S. (1996) Short- range and medium-range order in lithium silicate glasses, Part I: Diﬀraction Experiments and Results. Journal of the American Ceramic Society 79, 2833–2838. van der Laan, G. (1991) M 2,3 absorption spectroscopy of 3d transition-metal compounds. Journal of Physics: Condensed Matter 3, 7443–7454.

Vlastélic,I., Staudacher, T., Bachèlery, P., Télouk,P., Neuville, D., Benbakkar, M. (2011) Lithium isotope fractionation during magma degassing: Constraints from silicic differentiates and natural gas condensates from Piton de la Fournaise volcano (RéunionIsland). Chemical Geology 284, 26–34.

Voigt, U., Lammert, H., Eckert, H., Heuer, A. (2005) Cation clustering in lithium silicate glasses: quantitative description by solid-state NMR and molecular dynamics simulations. Physical Review B 72, 064207.

Wang, A., Jolliﬀ, B. L., Haskin, L. A., Kuebler, K. E., Viskupic, K. M. (2001) Characterization and comparison of structural and compositional features of planetary quadrilateral pyroxenes by Raman spectroscopy. American Mineralogist 86, 790–806.

Wang, D., Zuin, L. (2017) Li K -edge X-ray absorption near edge structure spectra for a library of lithium compounds applied in lithium batteries. Journal of Power Sources 337, 100–109.

Warren, B. E. (1934) The Diﬀraction of X-Rays in Glass. Physical Review 45, 657–661.

Warren, B. E., Biscce, J. (1938) The structure of silica glass by X-ray diﬀraction studies. Journal of the American Ceramic Society 21, 49–54.

Warren, B. E., Biscob, J. (1938) Fourier analysis of X-ray patterns of soda-silica glass. Journal of the American Ceramic Society 21, 259–265.

Waychunas, G. A. (1987) Synchrotron radiation XANES spectroscopy of Ti in minerals: Eﬀects of Ti bonding distances, Ti valence, and site geometry on absorption edge structure. American Mineralogist 72, 89–101.

Wenger, M., Armbruster, T. (1991) Crystal chemistry of lithium: oxygen coordination and bonding. European Journal of Mineralogy 3, 387–399.

Wojdyr, M. (2010) Fityk: a general-purpose peak ﬁtting program. Journal of Applied Crystallography 43, 1126–1128.

Wright, A. C. (1990) Diﬀraction studies of glass structure. Journal of Non-Crystalline Solids 123, 129– 148.

Wyckoﬀ, R. W. G. (1963). Crystal Structures 1, 85–237.

Yiu, Y. M., Yang, S., Wang, D., Sun, X., Sham, T. K. (2013) Ab-initio calculation of the XANES of

lithium phosphates and LiFePO4. Journal of Physics: Conference Series 430, 012014.

You, J., Jiang, G., Xu, K. (2001) High temperature Raman spectra of sodium disilicate crystal, glass and its liquid. Journal of Non-Crystalline Solids 282, 125–131. BIBLIOGRAPHY 119

Zachariasen, W. H. (1932) The atomic arrangement in glass. Journal of the American Chemical Society 54, 3841–3851.

Zanotto, E. D., Leite, M. L. G. (1996) The nucleation mechanism of lithium disilicate glass revisited. Journal of Non-Crystalline Solids 202, 145–152.

Zanotto, E. D., Tsuchida, J. E., Schneider, J. F., Eckert, H. (2015) Thirty-year quest for structure–nucleation relationships in oxide glasses. International Materials Reviews 60, 376–391.

Zemann, J. (1957) Die kristallstruktur von Li2CO3. Acta Crystallographica 10, 664–666.

Zhang, M., Boolchand, P. (1994) The central role of broken bond-bending constraints in promoting glass formation in the oxides. Science 266, 1355–1357.