2026 Volume 35 Issue 5
Article Contents

G. Michael Bancroft, H. Wayne Nesbitt, John S. Tse, Grant S. Henderson, Ben J. A. Moulton. Free oxide ion abundances in Na, Ba, and K silicate glasses from O 1s XPS, 29Si NMR, Raman, and MD simulations[J]. Chinese Physics B, 2026, 35(5): 056102. doi: 10.1088/1674-1056/ae3606
Citation: G. Michael Bancroft, H. Wayne Nesbitt, John S. Tse, Grant S. Henderson, Ben J. A. Moulton. Free oxide ion abundances in Na, Ba, and K silicate glasses from O 1s XPS, 29Si NMR, Raman, and MD simulations[J]. Chinese Physics B, 2026, 35(5): 056102. doi: 10.1088/1674-1056/ae3606

Free oxide ion abundances in Na, Ba, and K silicate glasses from O 1s XPS, 29Si NMR, Raman, and MD simulations

  • Received Date: 23/10/2025
    Accepted Date: 06/01/2026
    Available Online: 01/05/2026
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Figures(13)  /  Tables(2)

Article Metrics

Article views(353) PDF downloads(0) Cited by(0)

Access History

Free oxide ion abundances in Na, Ba, and K silicate glasses from O 1s XPS, 29Si NMR, Raman, and MD simulations

Abstract: In alkali silicate glasses with ≤ 50 mol% M2O (M = Na, K, Rb, Cs), the existence of > 1 mol% reactive “free” oxide (FO, where O is not bonded to Si) has been a highly controversial topic over the past 15 years. Unlike their crystalline analogues, Raman and 29Si nuclear magnetic resonance (NMR) studies since 1980 have shown that two or more Qn (n = 0–4) species are present in silicate glasses over a wide range of compositions. For example, M2SiO3 crystals contain only Q2 species; however, glasses of the same composition exhibit Q1 and Q3 in addition to Q2. Previous Raman and NMR studies on alkali silicate glasses have related the abundances of these three species solely through disproportionation reactions (e.g. 2Q2 ⇔ Q1 + Q3). In doing so, polymerization reactions (e.g. 2Q2 ⇔ 2Q3 + FO) were completely neglected. By combining published O 1s x-ray photoelectron spectroscopy (XPS) spectra, 29Si NMR and Raman results for 40 mol% and 50 mol% Na2O, K2O, and BaO glasses, together with new molecular dynamics (MD) simulations of Na4SiO4 glass, we provide consistent and compelling evidence for the existence of > 1 mol% FO in these glasses and melts. In particular, for 50 mol% K2O silicate glass, all three experimental techniques estimate FO to be ≥ 7 mol%, while MD simulations of Na4SiO4 yield ∼5 mol% FO. Our analysis requires revised assignments (challenging decades of “conventional wisdom”) for 29Si NMR and Raman spectra, based on O mass balance, recognition of M—BO bonding effects first identified in O 1s XPS spectra, and quantitative analysis of Raman spectra for 40–50 mol% Na2O, K2O, and BaO glasses. These FO values are comparable to those now accepted for alkaline-earth silicate glasses. The importance of this reactive FO for chemical reactivity (e.g. with H2O and CO2), bioactivity, and physical properties (e.g. melting) of silicate glasses is discussed.

1.   Introduction
  • In contrast to the well-established atomic-level crystal structures of stoichiometric crystalline alkali and alkaline-earth silicates (e.g. Na2Si2O5,[1] Na2SiO3,[2] MgSiO3,[3] and Mg2SiO4[4]), the atomic-level structures of silicate glasses with similar compositions remain insufficiently characterized, even after ∼ 90 years since the pioneering x-ray diffraction studies of silicate glasses by Zachariasen[5] and Warren and Briscoe.[6,7] By combining these early x-ray studies with more recent Na and Si x-ray absorption fine structure (XAFS),[8,9] neutron diffraction,[10] and molecular dynamics (MD) simulations,[11–15] the Si–O and Na–O bond lengths, as well as the O–Si–O and O–Na–O bond angle distributions, are now well established.

    Despite the fundamental importance of SiO4 tetrahedra in both glasses and minerals, glasses are much more complex than their crystalline counterparts (see Refs. [16–18] for reviews). First, glasses lack long-range order, and the nature of Na–O bonding in the traditionally accepted modified random network (MRN) model,[18,19] including subsequent modifications for Na–BO bonding,[20,21] remains poorly understood in detail (even after numerous MD simulations).[11–15] Second, in contrast to crystalline silicates, which exhibit only a limited number of discrete compositions (e.g. disilicates, metasilicates, and orthosilicates such as Na2Si2O5, Na2SiO3, and Mg2SiO4, respectively), glasses (and melts) span a wide compositional range. Indeed, silicate glasses can be regarded as solid solutions of M2O (e.g. Na2O) or MO (e.g. MgO) with SiO2, with compositions ranging from < 1 mol% MO to > 66.7 mol% MO. This compositional range can be rationalized using the depolymerization equation[22]

    or

    where O2− represents the oxide ion or free oxygen (FO, an oxygen not bonded to Si, also known as non-network oxygen), BO denotes bridging oxygen (O in the Si–O–Si moiety), and NBO− denotes non-bridging oxygen (O in the Si–O–M moiety). The equilibrium constant for reaction (2) is

    Clearly, if FO = 0, then K3 = ∞, and each Na2O unit provides one O2− that attacks a BO (Si–O–Si), producing 2NBO (Na–O–Si); however, if K3 is finite, some FO (e.g. Na–O–Na) must remain. This simple yet critical thermodynamic equilibrium implies that some FO should exist at all Na2O concentrations in these solid solutions, even if it may not be experimentally detectable at low Na2O contents. Third, unlike minerals such as Na2Si2O5 and Na2SiO3, which contain only Q3 [Si(BO)3(NBO)] and Q2 [Si(BO)2(NBO)2] species, respectively (where the superscript in Qn denotes the number of BO units), significant progress was made in the 1970s and 1980s using Raman spectroscopy[23–28] and 29Si magic angle spinning (MAS) nuclear magnetic resonance (NMR)[29–32] to demonstrate the presence of additional Qn species (for example, Q2 and Q4, in addition to Q3, in Na2Si2O5 glass). For instance, Raman spectra of sodium silicate glasses from 1975–1984[23–28] showed peaks corresponding to Q1, Q2, and Q3 at ∼ 850 cm−1, ∼ 950 cm−1, and ∼ 1100 cm−1, respectively, assigned to the A1 symmetric stretching modes of these structural units. Shortly thereafter, 29Si NMR studies of Na silicate glasses[29–32] identified Q2 and Q3 peaks at ∼77 ppm and ∼88 ppm. These findings highlighted the opportunity to quantify the distribution of Q species in alkali and alkaline-earth silicate glasses. Such distributions, together with the quantification of NBO, BO, and FO, are critical for understanding the chemical and physical properties of melts. For example, Fine and Stolper[33,34] sought to identify the reactive species responsible for converting H2O to OH and CO2 to CO32−, and noted the potential importance of FO in these reactions. However, subsequent interpretations within the NMR and Raman communities largely dismissed the presence of FO, leading to the assumption that NBO is the reactive species. Later, we reinterpreted Stolper’s data and proposed that FO is, in fact, the reactive species.[35]

    How are these multiple Q species related? Can equilibrium constants such as K3 in Eq. (3) be calculated for the entire Na2O–SiO2 solid solution? How large is [FO] in these glasses? Answering these questions using O 1s XPS and Raman spectra has been a major goal of our research since 2007 and is the focus of this paper. In the literature prior to 1970 (before the observation of distinct Q species by 29Si NMR or Raman spectroscopy[23–32] or the resolution of BO and NBO in early O 1s XPS[36,37]), several studies[38–40] proposed polymerization of SiO4 units in melts, which is a potential mechanism for conversion between Q species

    Other similar polymerization reactions are possible (for n = 1, 2, 3)

    Given these earlier studies, it is perhaps surprising that early Raman[23–28] and NMR researchers[29–32] in the 1980s related the Q species exclusively through disproportionation reactions

    for example

    or

    In the 1980s, Schramm et al.[29] mentioned the possibility of FO in Li silicate glasses, and Maekawa et al.[41] in 1991 showed evidence of FO in Li silicates from 29Si NMR. Neither study, however, linked FO explicitly to the polymerization reactions in Eqs. (4) and (5). Despite the higher resolution of Raman spectra compared to 29Si NMR, Raman data could not be reliably assigned or quantified.[23–28] In contrast, the high-quality resolved 29Si NMR spectrum of a 41 mol% Na2O glass enriched in 29Si could be quantified, and Stebbins[32] obtained the first equilibrium constant K (0.011 ± 0.005) for reaction (7a) from the integrated areas of the Q2 and Q3 peaks in the 29Si MAS NMR spectrum, combined with the Q4 contribution from a static 29Si spectrum. Only the disproportionation framework was used to relate the three Q species. Ironically, as discussed below, this work[32] is the first to indicate the presence of a few mol% FO (and the importance of polymerization reactions) in an alkali or alkaline-earth silicate glass. As an aside, the disproportionation reaction (Eqs. (6) and (7)) does not change the number of Si–O–Si bonds or the NBO/Si ratio in a glass or melt, regardless of the reaction extent (e.g. Eq. (2) for Na2SiO3 and Eq. (4) for Na4SiO4).

    29Si (and 17O) NMR[41–48] and Raman[49–59] studies on alkali and alkaline-earth silicate glasses over the following four decades relied entirely on disproportionation reactions, with no consideration of possible polymerization reactions. A consensus and “conventional wisdom” emerging from “group thinking”[60] held that disproportionation was the only relevant mechanism linking Q species, and therefore FO was so small that it could be neglected. As stated in several reviews on FO in Na and K silicate glasses by Stebbins and coworkers,[61–63] any significant FO “seems to be highly unlikely”.[62] In 2020, Stebbins[63] further concluded from new 29Si and 17O NMR studies of ∼ 50 mol% Na2O and K2O glasses that any FO in Na or K glasses was below detection limits of approximately 0.5%–1.0% FO.

    When we attempted to publish results on Na silicates using a different technique (O 1s XPS) in 2011,[64] more than 25 years after many of the above NMR and Raman studies, the manuscript was rejected–rejection of the technique, rejection of the results, and rejection of the interpretation. The paper was ultimately published approximately 18 months after initial submission, following three rounds of revision, inclusion of new 29Si NMR spectra, and constructive editorial support that encouraged open discussion. This outcome was surprising, given that our first O 1s XPS study on Pb silicates in 2007[65] demonstrated that high-resolution XPS optimized for non-conducting samples under ultra-clean conditions yields well-resolved and readily interpretable O 1s spectra, in good agreement with 29Si NMR results,[66] and reveals substantial FO (19.5 mol% at 66.7 mol% PbO), corresponding to K3 ∼ 12. This work initiated a broader discussion on FO in Na and K silicates following our XPS study of K silicate glasses,[35] along with Malfait’s response[67] and our rebuttal.[68] Subsequent studies by Stebbins[61–63] acknowledged FO in Pb, Ca, and Mg silicates after our 2015 work[69] and additional NMR and Raman studies of alkaline-earth silicates near the orthosilicate composition, which reported several mol% FO.[70–73] However, those studies[61–63] continued to reject significant FO in Na and K glasses. The prevailing “conventional wisdom” therefore maintained that FO is negligible in Na and K silicates. Indeed, even recent Raman and 29Si NMR studies on alkali silicate glasses[74–78] continue to interpret the data solely in terms of disproportionation reactions.

    Five years after Stebbins’ last rebuttal,[63] it is timely to present our perspective, supported by both new results and a reanalysis of previously published data. We have not been able to publish further papers on the FO controversy in Na and K silicates since our 2015 contributions[35,68,69] noted above. However, over the past decade, we have shown that Raman spectra can be reliably assigned and quantified;[79–83] these assignments have been independently confirmed in several recent Raman studies.[77,84–86] Some of these studies have carefully considered the role of FO in Na and Ba silicate glasses.[83–85] For example, we have shown that 40 and 50 mol% Na2O glasses contain measurable amounts of FO based on both Raman and 29Si NMR spectra,[81,83] in agreement with earlier O 1s XPS results.[64] Notably, the FO content we obtain for the 40 mol% Na2O glass is identical to that calculated from the data in Stebbins’ 1987 study.[32]

    Perhaps the most compelling recent evidence for significant FO in K silicate glasses comes from newly published Raman spectra of 33–50 mol% K silicate glasses,[76] which enable direct comparison with earlier 29Si NMR spectra.[41,46,63] Reanalysis of these data yields substantial FO contents from the 29Si NMR spectra, consistent with FO values obtained from earlier O 1s XPS measurements.[35,68,87] In addition, our new MD simulations of Na4SiO4 glass show appreciable concentrations of Si–O–Si dimers and FO arising from polymerization reactions such as Eq. (5).

    The main objective of this paper is to provide conclusive evidence for mol% FO in alkali silicate glasses using O 1s XPS, a reanalysis of published 29Si NMR and Raman spectra, and new MD simulations. To achieve this objective, we first evaluate the strengths and limitations of the three spectroscopic techniques. We show that O 1s XPS, when performed under stringent ultra-high-vacuum conditions, is a reliable method for determining the molar fractions of BO, NBO, and FO in Pb, Na, and K glasses. For Pb silicate glasses, the large FO values obtained from O 1s XPS over a wide compositional range are confirmed by 29Si and 17O NMR,[66,88] whereas the NMR community has argued against the validity of the XPS-derived FO values for Na and K silicates, based on poorly resolved NMR spectra.[61–63] Second, we summarize our efforts to resolve the long-standing challenge of assigning and quantifying Raman spectra of silicate glasses,[18,83–85,89] showing that quantitative determination of Q1, Q2, and Q3 species in 40–50 mol% alkali glasses is now achievable in many cases. Third, using our revised 29Si NMR and Raman assignments for 40 mol% Na2O and 40 mol% BaO glasses,[83–85] we demonstrate that the only consistent interpretation of recent Raman spectra of 50 mol% K2O samples contradicts previous NMR-based assignments[41,67,68] that assumed the absence of FO. In short, the 29Si NMR spectra of 50 mol% K2O glass have been misassigned since 1991,[41,46,63] and the FO content is substantial (≥ 7 mol%) in this composition, consistent with XPS results.[35,68] Fourth, we present new MD simulations of Na4SiO4 glass, which reveal significant populations of Q1 dimers from reaction (4), along with FO. Fifth, we highlight several oversights and post hoc adjustments made within the 29Si and 17O NMR and Raman literature after the community consensus formed that FO does not exist in alkali silicate glasses, a viewpoint that likely led Maekawa[41] to misassign his 29Si NMR spectrum of the 50 mol% K2O glass despite an initially more straightforward interpretation. Finally, we hope that this work also contributes to a broader discussion on the nature and application of the scientific method in the Earth sciences, as elegantly discussed by Richet.[60]

  • As discussed above, three unrelated spectroscopic techniques – NMR (29Si and 17O), Raman spectroscopy, and O 1s XPS – have been used to estimate the abundances of Qn species (n = 0–4) and O species (BO, NBO, and FO) in alkali and alkaline-earth silicate glasses. Spectral resolution is a critical property in any spectroscopic analysis. Without sufficient peak resolution, it is not possible to obtain reliable or objective estimates of Q species or of NBO, BO, and FO populations. As noted above, the resolved 29Si NMR spectrum of a 41 mol% Na2O glass yielded the first quantitative equilibrium constant K3 for relating Q2, Q3, and Q4 (Eq. (7a)).[32,42] However, 29Si NMR spectra are generally less well resolved than Raman spectra of the same glass compositions. The most important and influential of these NMR spectra is that of the 50 mol% Na2O glass shown in Fig. 1.[41] There is broad agreement on the assignment of the three features in this spectrum: a weak, narrow Q1 peak at 67 ppm, an intense Q2 peak at 76 ppm, and a broad, poorly resolved Q3 contribution at approximately 80–84 ppm. The question remains whether the Q1 and Q3 species can be reliably quantified from such a partially resolved spectrum. We emphasize again that, from Eq. (7b), Q3 = Q1 for pure disproportionation; whereas Q3 ≫ Q1 if polymerization reactions also occur, implying the presence of FO.

    The 1987/1991 spectra recorded at relatively low magnetic fields (e.g. 4.7 T by Maekawa,[41] shown here in Fig. 11(a)) are, in practice, as well resolved as spectra obtained by us at 9.4 T in 2011,[64] or those measured at 14 T by Stebbins in 2020.[63] Moreover, the 1991 spectrum is the best available for accurate digitization. It exhibits good signal-to-noise ratio, and the experimental points are reported individually rather than as smoothed curves. Consequently, the 1991 spectrum is the best available.

    Examining the spectral resolution in more detail for the 50 mol% Na2O glass shown in Fig. 1, the positions and linewidths of the Q1, Q2, and Q3 components are remarkably consistent across studies (all within approximately 1 ppm) between the 1991[41] and 2020[63] datasets (see Table 2 in both works). Specifically, the reported linewidths for Q1, Q2, and Q3 are 4.4 ppm vs. 5.0 ppm, 6.5 ppm vs. 7.2 ppm, and 10.2 ppm vs. 10.0 ppm, respectively, well within typical fitting uncertainties for such partially resolved spectra. It is also important to note that the 29Si NMR spectrum of the 50 mol% K2O glass measured at 4.7 T by Maekawa (our Fig. 11(a))[41] is, if anything, better resolved than the spectrum of the same composition measured at 14 T by Stebbins in 2020.[63] Indeed, Stebbins does not attempt a full deconvolution of the 14 T spectrum. There is no clear evidence that even higher magnetic fields would substantially improve the resolution for these systems. For example, Stebbins and coworkers have recently reported high-field (35 T) NMR measurements on silicate glasses for the quadrupolar nucleus 25Mg using the National High Magnetic Field Laboratory (NHMFL) in Florida.[90] However, no corresponding 29Si spectra were reported at 35 T in that study or in other work we are aware of.

    Experienced solid-state NMR spectroscopists are well aware that increasing the magnetic field does not necessarily improve the resolution of 29Si NMR spectra of glasses. For silicate glasses, the 29Si NMR linewidth is dominated by chemical shift anisotropy (CSA) and structural disorder, which produces a distribution of chemical shifts. These broadening mechanisms scale with the magnetic field strength; as a result, the linewidth in ppm remains approximately constant. Therefore, higher magnetic fields do not improve and may even degrade the apparent spectral resolution, as observed.

    Figures 1(a)–1(e) show immediately that the Q3 intensity varies strongly with both the assumed Q3 peak position and linewidth; consequently, neither the Q3 position nor the linewidth can be determined unambiguously from the data. For example, consider the Q3 linewidth estimates. Linewidths of approximately 10 ppm from fits to the 40 mol% Na2O glass have been used in previous studies[41,63] and also by us.[64] A significantly larger assumed Q3 linewidth shifts the fitted peak to lower ppm and increases the extracted Q3 area. As the linewidth decreases from Fig. 1(a) to 1(e), the Q3 peak position shifts and the Q3 area decreases substantially, approaching the Q1 area at the smallest linewidth (Fig. 1(e)). The fully unconstrained fit in Fig. 1(a) yields the lowest RMS value (61). However, this fit can be rejected because the resulting Q3 linewidth is unrealistically large and the peak position is also physically implausible. A low RMS value alone is therefore a misleading criterion for identifying the best fit, despite its widespread use in spectral fitting.

    In the remaining four fits (Figs. 1(b)–1(e)), the Q3 linewidth is fixed to more physically reasonable values (13 ppm, 11 ppm, 10 ppm, and 9 ppm). Under these constraints, four systematic trends are observed: (1) the RMS value increases systematically from 85 (Fig. 1(b)) to 98 (Fig. 1(e)); (2) the Q3 peak position shifts from 79 ppm to 82 ppm; (3) the Q3 fraction decreases from 36% to 19%, while the Q1 fraction remains approximately constant at ∼ 15%; and (4) the inferred FO content decreases from ≫ 1 mol% in Fig. 1(b) to ≪ 1 mol% in Fig. 1(e).

    These results demonstrate that the 29Si NMR spectrum in the Q2–Q3 region is too poorly resolved to allow an objective determination of the Q3 fraction (and hence FO content) without independent constraints on the Q3 linewidth or position. Further discussion of this spectrum, together with comments on Stebbins’ recent fitting results,[63] is provided in Section 2.5. This resolution limitation is common to most 29Si NMR spectra of silicate glasses, and as noted above, there is no evidence that higher magnetic fields significantly improve the resolution. Indeed, this spectrum is better resolved than those of many other alkali and alkaline-earth silicate glasses.[35,46,52]

  • XPS has become an important spectroscopic technique for chemical analysis of solids primarily because it is applicable to all elements except H and He, it generates high-quality spectra with good chemical resolution, and the peaks (including peak widths and shapes) can be readily assigned and quantified.[91–93] However, with the commonly used excitation energy of 1486 eV, the so-called escape depth (corresponding to 63% of the total intensity) is smaller than optimal for bulk analysis. For example, the escape depth has been measured for SiO2 as 27 Å.[92] The so-called analysis depth (corresponding to 95% of the total intensity) is approximately three times the escape depth — or 81 Å for SiO2.[91–93] As a result, XPS is usually considered to be primarily a surface-sensitive technique. However, XPS can also provide excellent bulk chemical information for minerals and glasses,[91–93] provided that there is little or no surface contamination or x-ray beam damage. These are the conditions routinely required by the surface science community, but not always by the routine analytical community. For example, in our XPS analysis, glasses are fractured in situ under high vacuum, and care is taken to ensure that little or no adventitious carbon or oxygen is present on the surface. About 5% of the total signal originates from the surface layer;[93] if this surface layer contains oxygen from hydrocarbon contamination, the relative NBO and BO O 1s intensities in the silicate glass will be affected. Our extensive studies show that accurate quantitative analysis of oxygen speciation in silicate glasses is not possible without these precautions. We also find that light Ar ion sputtering of the surface to remove C/O contaminants[94] does not yield satisfactory results. In addition, the relative NBO and BO intensities may be influenced by the nature of the freshly fractured surface. After presentation of the results (see Subsection 2.2), we show that such surface effects do not artificially enhance the measured BO abundances.[35,92]

    With the advent of the Kratos Axis Ultra XPS instrument in the late 1990s, featuring improved charge neutralization for non-conducting samples, XPS resolution has improved markedly for materials such as silicate glasses.[95] Equally importantly, the peak shapes are remarkably stable, reproducible, and predictable. Indeed, the O 1s XPS linewidth is 1.25 ± 0.1 eV for FO, NBO, and BO oxygen species in all minerals and glasses studied,[92,95,96] and the line shapes are consistently 30–50% Lorentzian.[92] To the best of our knowledge, these XPS linewidths and resulting resolution are unmatched in the literature for any modifier oxide–silicate glass system, and represent the intrinsic limit expected for oxygen in SiO4 tetrahedra in silicate minerals or glasses.[95,96] Constant linewidths and line shapes are also observed in Mössbauer spectra of silicate minerals,[97] but not in NMR or Raman spectra of silicate minerals or glasses.[23–32,41–59,61–64,98,99] As shown in Fig. 1, the 29Si NMR linewidths vary from 5 ppm to > 10 ppm in Na silicate glasses;[41,63,64] moreover, the exclusive use of Gaussian line shapes to fit spectra is not always justified. For example, the most recent 29Si NMR spectra of 50 mol% Na2O glass[63] have been fitted using unspecified Lorentzian contributions rather than purely Gaussian functions.[63] Similarly, although Raman resolution is higher than that of 29Si NMR, Raman peak widths used in fitting glass spectra vary widely (from < 40 cm−1 to > 70 cm−1), and the presence of unassigned “extra” peaks and asymmetries has prevented reliable quantification of Raman peak areas. In contrast, the constant XPS linewidths and line shapes, together with good spectral resolution, make XPS spectra much more straightforward to fit than corresponding NMR and Raman spectra. For transparency, we report peak positions, areas, linewidths, and line shapes without constraints for nearly all spectra, in contrast to common practice in much of the 29Si NMR and Raman literature.

2.   Results and discussion
  • Because several papers have questioned the validity of O 1s XPS spectra for obtaining quantitative BO and NBO information in silicate glasses,[61,62,67] we first present O 1s XPS results on Pb silicate glasses, together with 29Si and 17O NMR results,[65,66,88] to demonstrate that: (1) O 1s spectra provide accurate and transparent determinations of FO, BO, and NBO in these glasses; (2) these results are consistent with the Q-species distributions derived from 29Si NMR[66] and 17O NMR;[88] (3) the equilibrium description given by Eqs. (2) and (3) holds over a wide compositional range, yielding > 15 mol% FO at 66.7 mol% PbO, decreasing to approximately ∼ 1 mol% FO at 30 mol% PbO.

    In Fig. 2, we present O 1s XPS spectra[65,69] (acquired using a modern Kratos Axis Ultra spectrometer) for four PbO compositions, ranging from 50 mol% PbO in Fig. 2(a) to 76.7 mol% PbO in Fig. 2(d). It is important to emphasize that clean, freshly fractured surfaces and an effective charge compensation protocol are required to achieve narrow linewidths of approximately ∼ 1.2 eV and consistent symmetric line shapes for all oxygen species in our non-conducting mineral and glass systems.[92,95,96] The narrow O 1s peaks (linewidths ≤ 1.22 eV) in Figs. 2(a)–2(c) can be reliably fitted using two peaks of identical width and shape at approximately 530 eV and 531.5 eV. No fitting parameters are constrained in these O 1s spectral decompositions. These two components are assigned to NBO and BO oxygen species, respectively, since NBO possesses higher valence electron density than BO due to charge donation from Pb.[36,37,91]

    For the 76.7 mol% PbO spectrum in Fig. 2(d), two clear indications suggest an additional contribution within the NBO envelope: (1) the NBO peak maximum shifts by approximately 0.5 eV toward lower binding energy, and (2) the NBO linewidth increases from 1.22 eV to 1.67 eV. Under these conditions, the NBO region can be resolved into two peaks of identical width (1.22 eV), corresponding to NBO at 529.8 eV and FO at 529.0 eV, respectively. In the spectra shown in Figs. 2(a)–2(c), the NBO envelope contains both NBO and a much smaller FO contribution that is not resolved as a separate peak.

    From the areas of the two large peaks and assuming equal photoionization cross sections for the oxygen species,[91] the mole fraction of BO (XBO) (solid circles) can be determined as a function of XPbO in Fig. 2(e). Before discussing these XPS-derived values and the NMR results obtained from 29Si spectra[66] (open squares), it is useful to consider the short dashed line representing XBO for the limiting case K3 = ∞. This line corresponds to the absence of FO, with only BO and NBO present in the system. The expected XBO values for stoichiometric SiO2, PbSi2O5, PbSiO3, and Pb2SiO4 are 1.0, 0.60, 0.33, and 0, respectively, all of which lie on the dashed line in Fig. 2(e). For example, in PbSiO3 there is one BO and two NBO per Si, giving XBO = 1/3 ≈ 0.33. The 29Si NMR-derived XBO values[66] for glasses with XPbO < 0.4 lie very close to the K3 = ∞ line. However, for XPbO > 0.4, the XBO values from both techniques systematically lie above this limiting line. In particular, at XPbO = 0.67, the dashed-line model predicts XBO = 0, whereas the experimental O 1s XPS[65,69] and 29Si NMR[66] results give XBO = 0.19 and 0.15, respectively (see below for the derivation of XBO from the Q species obtained from the 29Si NMR spectra). The 17O NMR spectra show excellent resolution of peaks from BO, NBO, and FO;[88] however, no fits to the resolved spectra are performed, mainly due to the unusual peak shapes arising from the quadrupolar 17O nucleus.[67,88] Nevertheless, a large resolved FO peak is clearly observed, and the authors report XFO = XBO ≈ 0.135.[88] The two NMR and O 1s XPS results indicate a significant XFO, implying that reaction (2) does not proceed to completion and that K3 is finite. Most importantly, the substantial XFO contribution in all cases demonstrates that polymerization via Eq. (2) occurs and that FO is a critical species in the glass. Of course, disproportionation is not possible for a Q0 species.

    At XPbO = 0.67, the XPS results show that XBO ∼ 0.19 (Fig. 2(e), large black circle). What is K3 for this sample? From Eq. (2), and noting that XBO = XFO ≈ 0.19 for the 66.7 mol% PbO sample, while oxygen mass balance gives XBO + XNBO + XFO = 1, we obtain XNBO ≈ 0.62, and hence K3 = 10.7. The value of K3 is rather sensitive to XBO. For example, if XFO = 0.16, then K3 = 18.0. A value of K3 = 12 was used for the solid XBO curve, which provides a good visual fit within the uncertainties assigned to the XBO values from O 1s XPS and 29Si NMR (Fig. 2(e)). The accuracy of the K3 value cannot be precisely evaluated at this stage, but it likely lies between 10 and 20.

    The XFO values for K3 = 12 are also plotted as a solid curve at the bottom of Fig. 2(e). XFO decreases rapidly with decreasing XPbO; XFO ≈ 0.07 (7% FO) at 50 mol% PbO, and XFO ≈ 0.01 (1% FO) at 30 mol% PbO. The equilibrium treatment therefore requires appreciable (mol%) FO even at XPbO = 0.3 for K3 between 10 and 20. It is also important to note that the difference between the two BO curves (ΔXBO) directly yields XFO at all XPbO. This is most clearly seen for the 66.7 mol% PbO glass, where ΔXBO = 0.19, giving XFO = 0.19. At 50 mol% PbO, ΔXBO = 0.07.

    The Q-species distributions from 29Si NMR (and, as shown later, Raman) can be used to determine the abundances of NBO and BO in a glass using the following mass-balance equations (normalized to Ototal/Si)

    For example, for the 67 mol% PbO glass, the 29Si NMR[66] yields XQ3 = 0.05, XQ2 = 0.31, XQ1 = 0.52, and XQ0 = 0.12. Substituting into Eqs. (8) and (9) gives NBOt = 0.64 and NNBOt = 2.72. At this composition, NFO = NBOt = 0.64, and NOtotal/Si = 4 as expected for Pb2SiO4. Thus, XBO, XNBO, and XFO are obtained by dividing by four, yielding 0.16, 0.68, and 0.16, respectively, as shown in Fig. 2(e). For PbSiO3, the 29Si NMR[66] gives XQ4 = 0.06, XQ3 = 0.30, XQ2 = 0.51, and XQ1 = 0.13, yielding NBOt = 1.14 and NNBOt = 1.71, with a total of 2.85, so NFO = 0.15 (NOtotal = 3.0). The resulting XBO, XNBO, and XFO values are obtained by dividing by three, i.e., 0.38, 0.57, and 0.05, respectively. Note that this XBO value (0.38) is 0.05 larger than the K3 = ∞ value of 0.33, as expected. The FO value is slightly smaller than that obtained from K3 = 12 or from XPS. As shown above, this equilibrium treatment enforces a rigorous evaluation of oxygen mass balance,[35,83–85] which has often been overlooked in the NMR and Raman literature. Using oxygen mass balance makes it evident that the XBO values cannot lie below the dotted curve, because this would lead to negative FO values, which is physically impossible. Figure 2(e) should be essential for all spectroscopists studying silicate glasses, as it illustrates both oxygen mass balance and equilibrium among the three oxygen species.

    Our detailed analysis of FO contents in alkaline-earth silicates from literature 29Si and 17O NMR and Raman studies[70–73] on Mg, Ca, and Mg/Ca glasses, including orthosilicates (66.7 mol% MO) and metasilicates (50 mol% MO),[69] shows large FO concentrations: 5–10 mol% at 66.7 mol% and 1–2.6 mol% at 50 mol%, with associated NBO/Si values that are substantially below the stoichiometric crystalline values of 4 and 2, respectively.[100] MD calculations on these glasses also give large but variable FO values.[69] For example, MD calculations on Mg2SiO4[101–103] yield XFO values of 5%, 15%, and 16%, corresponding to a large range of K3 values (330, 21.8, and 18.0, respectively).

    For 66.7 mol% MO Pb, Ca, and Mg glasses–compositions that nominally suggest only Q0 species (as in orthosilicate crystals)–appreciable Q1 and Q2 species are nevertheless present. This can only arise via polymerization as in Eq. (4). Two additional results are worth noting. An 17O NMR[104] study on a 56 mol% CaO glass concluded that there was no appreciable FO based on measured BO and NBO fractions,[69] without considering oxygen mass balances. However, using the mass balance relations in Eqs. (10) and (11), the inferred FO would range from 1.7% to 3.2%, depending on the choice of BO values.[69,104] Conversely, a well-resolved 17O NMR spectrum of a 45 mol% BaO silicate glass yields unphysical negative FO values when using the reported BO assignments.[104] It has been difficult to quantify well-resolved 17O NMR spectra due to variable linewidths, irregular peak shapes arising from the quadrupolar nature of the 17O nucleus, and shifts in relative peak positions with changes in magnetic field.[21,89] As a result, Lee and Kim[88] did not fit their well-resolved 17O NMR spectra of Pb silicates, and Stebbins[63] did not attempt to fit or quantify the resolved 17O NMR spectrum of the 50 mol% K2O silicate glass.

  • Figures 3 and 4 show the O 1s XPS spectra and the XBO plots for our Na[20,64] and K silicate glasses[21,35,68,87] over a wide range of compositions, along with XBO values from the only other O 1s XPS study with reasonable resolution.[105] The counting times for these spectra are longer than for the Pb spectra in Fig. 2, and as a result the statistics in Figs. 3 and 4 are much better than in Fig. 2. Excellent fits are obtained in all cases to the two resolved O 1s peaks without constraints, and all fitting parameters are listed in the above six papers. The assignment of the two resolved peaks is the same as for the Pb spectra — BO peak at ∼ 532 eV and NBO+FO peak at ∼ 530 eV. As expected from Eq. (2), the relative intensity of the BO peak decreases as the mol% Na2O and K2O increases in Figs. 3 and 4. Note that the separation of the two peaks increases in the order Pb (∼ 1.5 eV) < Na (∼ 2.0 eV) < K (∼ 2.2 eV) — as the electronegativity (EN) of the cation decreases from 1.87 (Pb) to 0.93 (Na) to 0.82 (K).[106] This difference is mostly due to the decrease in O 1s BE of the NBO atoms due to greater charge transfer of electrons in the order Pb < Na < K, which results in a decrease in the NBO BE from 530.3 eV to 530.0 eV to 529.5 eV, respectively.

    The BO peak is substantially broader than the NBO peak by ∼ 0.2 eV for the Na spectra[21,64] and ∼ 0.3 eV for the K spectra,[87] suggesting strongly that the BO peak is composite, due to M–BO interactions discussed below. Fits with two and three BO peaks for the Na and K spectra,[20,21] respectively (Figs. 3(a)–3(d) and 4(a)–4(d)), constraining each linewidth to the NBO linewidth of ∼ 1.2 eV, are shown in Figs. 3(a)–3(d) and 4(a)–4(d). Because these BO peaks are very close together, the relative intensity of these BO components cannot be determined quantitatively; but most importantly, the XBO values are not significantly affected by the number of BO peaks included. Such fits of strongly overlapping peaks are only reasonable when there is exceptional reproducibility of the linewidths and lineshapes. Indeed, these O 1s spectra show exceptional reproducibility. Except for the 10 mol% K2O glass, the linewidths of the narrower NBO peak for all Na and K glasses are < 1.3 eV, and most are very close to 1.2 eV. These narrow linewidths and constant lineshapes are maintained over many hours of x-ray exposure, as shown in Table 1 of Ref. [21]. These widths are all considerably smaller than the somewhat variable NBO widths (1.4–1.6 eV in Matsumoto’s paper[104]) and the even larger NBO widths in earlier O 1s XPS studies[36,37] using older conventional charge compensation techniques. Moreover, the 1.2 eV widths are very close to the minimum linewidth expected for a SiO4 moiety due to vibrational broadening.[96]

    As stated above, the broadening of the BO peak is due to interactions of Na and K with BO, as also seen in crystalline Na and K silicates such as Na2Si2O5[107] and K2Si2O5.[108] For example, in the K glass spectra, the three BO peaks probably arise from three distinct BO environments associated with different numbers of K atoms bonded to BO (e.g., 0, 1, and 2). These peaks are separated by ∼ 0.5 eV (see Fig. 3(d)), compared to the separation between BO and NBO of 2 eV.[21] The interaction of each Na or K with a BO results in a more negatively charged BO and a lower O 1s binding energy (BE). In support of this interpretation, crystalline K2Si2O5[108] contains three types of BOs. It is important to emphasize that the calculated XBO values are not substantially affected by the number of BO peaks used in the fit. For example, Table 1 in Ref. [21] for the K glasses shows that the XBO values obtained from one-BO-peak and three-BO-peak fits differ by within 0.4% absolute for four glasses (e.g., 0.840 and 0.843 respectively for the 15 mol% K2O glass in Fig. 4(b)), and within 1% absolute for the other two glasses (e.g., 0.644 and 0.653 respectively for the 32 mol% K2O glass in Fig. 4(d)). Similarly, for the Na glasses,[21,64] the XBO values are also within 1% absolute for one- and two-peak BO fits: e.g., 0.737 and 0.731 for the 25 mol% Na2O glass, and 0.544 versus 0.536 for the 40 mol% Na2O glass.

    We also corrected for small observed increases in XBO with x-ray exposure time. The BO fraction was plotted against exposure time T, and the XBO values for each sample were extrapolated back to T = 0.[20,21] For most samples, this extrapolation changes XBO by much less than 1% absolute compared to the first recorded spectrum. For example, for the two-BO-peak fit to the 25 mol% and 40 mol% Na2O spectra, XBO decreases from 0.731 (first spectrum, see above) to 0.724 (extrapolated value); and for the 40 mol% spectrum, XBO decreases from 0.536 to 0.528. These “T = 0” values are those plotted in Figs. 3 and 4. The estimated errors in both composition and XBO are given by the error bars in Figs. 3(e) and 4(e).

    Are these errors realistic? For siliceous glasses with XNa2O < 0.3 and XK2O < 0.2, the XBO values from both our results and those of Matsumoto[104] lie directly on the solid and short-dashed lines required by Eq. (2). This provides strong evidence that the XPS method yields reliable XBO values. For XNa2O > 0.3 and XK2O > 0.2, both data sets from these two independent groups deviate systematically above the short-dashed line for K3 = ∞, strongly indicating that K is finite and that mol% FO is present in these samples. For example, for six different K glasses with 26–34 mol% K2O, the FO content varies from 1.5% to 3.5% (average 2.2% for all six, and 2.0% excluding the 3.5% outlier), with an uncertainty of ± 1% on each value.[35] This provides strong evidence that > 1 mol% FO is present at ∼ 30 mol% K2O. In contrast, the XBO data for comparable 25–30 mol% Na2O glasses does not show a significant deviation from the short-dashed line,[21,64] suggesting that FO is smaller in Na glasses of comparable composition than in K glasses.

    Previous publications,[61–63,67] as well as a referee of this paper, have argued that surface sensitivity in XPS could lead to an overestimation of BO, which we attribute instead to FO. We addressed this concern in earlier work,[35,92] but it is worth reiterating with additional evidence from Greaves et al.[109] He corroborates our previous conclusion that freshly fractured glass surfaces are, if anything, Na-rich (i.e., NBO-rich), because weak Na–O bonds (Na–O bond energy of 239 kcal/mol) are more easily broken than Si–O bonds (Si–O bond energy of 798 kcal/mol[106]). Greaves[109] states: “Certainly there is much evidence from a variety of studies for the large alkali content of freshly fractured glass surfaces”. See also Fig. 8 in that work. A similar M–O (M = Pb, Mg, Ca) enrichment would be expected for divalent cations. For an NBO-rich surface, the measured XNBO would, if anything, be slightly larger than the bulk value. Since XBO + XNBO = 1, an increase in XNBO necessarily implies a decrease in XBO, not an increase in XBO as claimed by critics.[61–63,67] If anything, this implies that we are underestimating the BO and FO amounts!

    There are three observations that show that such a NBO-rich surface layer does not affect our bulk XBO measurements within the quoted errors of 0.02. First, for the Pb glasses, there is good agreement between the O 1s XPS results and the 29Si and 17O NMR results: all techniques give large FO contents at ≥ 50 mol% PbO as shown above. Second, for the Na and K glasses at < 30 mol% Na2O or < 20 mol% K2O, the XBO results agree with the theoretical results using Eq. (3), assuming that K3 = infinity. Third, the O 1s XPS results for the Na and K glasses agree with the properly assigned and interpreted 29Si NMR and Raman results (see below). In summary, the intensity from a possible NBO surface signal does not affect our bulk analysis significantly.

    Some of the XBO values derived from 29Si NMR data[32,41,64] (e.g. the XQ values from Table 1) are also plotted on Figs. 3 and 4 after converting the Q species distributions to NBOt (Eq. (8)) and XBO (= NBOt/Ot). For example, for the 41 mol% Na2O sample (third and fourth glasses in Table 1), NBOt = 1.37 and Ot = 2.70 (Na1.40SiO2.70), giving XBO = 1.37/2.70 = 0.507, which is 0.026 (2.6 mol% FO) larger than XBO (and = XFO) for the short-dashed line for K3 = infinity (0.481). These values are within the round-off error of the 2.8 mol% FO given in Table 1. The solid lines give a good visual fit to the O 1s XPS data and most of the 29Si NMR data. For the 40 mol% Na2O glass, the four XBO results (two from XPS and two from 29Si NMR) given in Fig. 3 are in reasonable agreement and all show mol% FO. These solid lines are obtained using a K3 value of 30 and 8 for Na and K glasses, respectively. The K3 value for the Na glasses is larger than the K3 = 15 given in the initial publication.[64] As pointed out by Stebbins,[63] we should have used the actual quoted composition (47.6 mol% Na2O) rather than the nominal composition of 50 mol% Na2O, and this has been corrected in Fig. 3 for both the XPS[20] and 29Si NMR.[64] This change results in a smaller FO by ∼ 2 mol% from the earlier paper[64] with a larger K3; however, FO is still 1.8 mol% (XPS) (Table 3 in Ref. [64]) and 0.6 mol% (NMR). Similarly, accurate compositions for the K glasses were essential to obtain accurate FO and K3 values.[35] As indicated in the discussion of the Pb results in Subsection 2.1 above, the K3 values have large errors; but they do indicate strongly that: (1) in both Na and K glasses, mol% FO is present for 30–40 mol% glasses, and (2) that K3 for the K glasses is significantly smaller than for Na glasses, and is comparable to the K3 for the Pb glasses.

    The XBO and FO values from the 29Si NMR results for the ∼ 40 mol% Na2O glass strongly indicate mol% FO from several studies (Table 1[32,41,42,64,83]). Unlike other Na silicate glasses, the 40 mol% Na2O glass gives two intense peaks with reasonable resolution (Fig. 5(a)) making it possible to fit the large Q2 and Q3 peaks without applying constraints to Gaussian peaks and derive quantitative Q2 and Q3 (and Q4 from static 29Si NMR) and XBO from free fits without “adjustments”.[83] In contrast, the 33.3 mol% and 50 mol% Na2O glasses (e.g. the original spectrum and fit[41] shown in Fig. 5(b)) have only one intense peak with two small overlapping peaks.[41,63,64] As we discussed in Subsection 1.1 and later in Subsection 2.5, it is not possible to obtain an objective fit and quantitative Q species content from this spectrum.[63]

    The Q species abundances are listed in Table 1 from four 29Si NMR studies of the 40 mol% Na2O glass. There are three important observations: (1) the Q species distributions are in good agreement with each other; (2) they all result in between 1.0 mol% and 2.8 mol% FO; and (3) the BO values derived in Fig. 3e are in agreement with the two O 1s XPS results.[64,104] For example, the average FO for the four NMR studies in Table 1 is 1.9 mol%, and the XPS results give 2.0 mol% for the nominal composition (40%) and 3.0 mol% for the analyzed composition (41.1%).[64]

    The first, and best, 29Si NMR spectrum (Table 1) for a ∼ 40 mol% Na2O glass was published by Stebbins in 1987 (Fig. 5(a))[32] using an enriched 29Si glass. He quoted the Q4 abundance (from the static 29Si NMR spectrum), but he did not quote Q2 and Q3 abundances; and it is important to show how we obtained[83] the Q2 and Q3 values in Table 1. Although Stebbins did not show a fit to the spectrum, he obviously measured the Q2 and Q3 abundances because he quoted an equilibrium constant KD (=0.011) for the disproportionation reaction (Eq. (7a))

    With the mass balance equation

    and XQ4 = 0.02, we obtain two equations (Eqs. (12) and (13)) in two unknowns, XQ2 and XQ3. Solving the resulting quadratic equation gives XQ2 = 0.274, XQ3 = 0.706, and XQ4 = 0.02 (Table 1). To obtain the XFO for these 41 mol% Na2O glasses, we use Eq. (14), resulting from the O mass balance[36]

    As previously stated for Eqs. (8)–(11), NOtotal is the number of moles of O per mole of Si. Therefore, the composition of the glass is expressed as Na1.40SiO2.70, and NOtotal = 2.70. Then NQtotal is given by the sum of the NBO and BO contributions in Eqs. (8) and (9)

    Substituting the above XQ values into Eqs. (14) and (15) gives

    The mass balance shows that the total O from the Q species is 2.63, which is less than the total O in the sample (2.70). This discrepancy must be due to the presence of FO. Because Stebbins did not consider the O mass balance, he assumed that FO was 0. This significant value of 2.8 mol% FO, demonstrating the importance of polymerization, was already implicit in 1987! Note that the other FO values in Table 1 lie between 1.0 and 2.8 mol%.

    If one relies solely on disproportionation, the above O 1s XPS and 29Si NMR (and later Raman spectra) cannot be fully explained. Moreover, assuming FO = 0 mol% in silicate compositions with < 60 mol SiO2 is no longer a viable stance, and the depolymerization reaction must be considered in combination with the oxygen mass balance. Furthermore, recent papers have negated oxygen equilibria and undercut further development of the analysis of 29Si NMR spectra by reporting values for Qn and O species without explanation of the fitting procedure and assumptions used to obtain them.[62,63] This amounts to a step backwards in the quantitative data reduction for any technique and should cause concern for 29Si NMR users.

    The XBO results shown in Fig. 4(e) for the K2O glasses are more significant because of the small K3 of 8, with FO ≥ 1 mol% at 30 mol% K2O, projected to increase to ≥ 7 mol% at 50 mol% K2O (solid line). We have argued strongly in our 2015 papers[35,68] that the 29Si NMR spectrum of the 50 mol% glass should be re-assigned (see Subsection 2.4), and this reassignment yields the XBO values given in Fig. 4(e) (solid triangles[41] and shaded diamonds[46]) from the two major 29Si NMR studies. They lie right on the line for K = 8. With the publication of high-quality Raman spectra of the 50 mol% K2O glass in 2023,[76] combined with the firm assignment of Raman spectra of Na and Ba glasses,[83–85] we can now show strong new evidence that our proposed 29Si NMR reassignment of the 50 mol% K2O spectrum is justified. First, we address the new advances in Raman spectroscopy of silicate glasses which lead to firm assignments of the Na and Ba silicate spectra in the next section, before returning to the assignment of the 50 mol% K2O 29Si NMR spectrum in Subsection 2.4.

  • Franz and Mysen[52] and McMillan et al.[110] suggested that the Raman spectra of silicate glasses could be quantified by comparing the better resolved Raman spectra directly with the 29Si NMR spectrum of the same composition. Franz and Mysen[52] showed the two spectra of CaSiO3 glass (Figs. 6(a) and 6(b)) and proposed that the Raman cross sections for the three Q species could be obtained from the peak areas, assuming that the 29Si NMR cross sections were the same for all Q species. Unfortunately, there was no consensus on the assignment of the high-energy peak (labelled Q3 in Fig. 6(b)[52]) at ∼ 1050 cm−1; and the 29Si NMR spectrum was too poorly resolved (more poorly resolved than Fig. 1) to obtain quantitative areas for the three peaks. Although the two spectra (Fig. 6) are qualitatively similar (an intense Q2 peak with two weaker peaks on either side), the high-energy peak at ∼ 1050 cm−1 was assigned to something other than the obvious Q3 peak, mostly because “conventional wisdom” dictated that the Q3 intensity could not be larger than Q1 based only on disproportionation in a 50 mol% glass (see Eq. (7b)). The 1050 cm−1 peak in Fig. 6(b) (and many other Raman spectra) was assigned to a peak not related to a symmetric stretch of a Q species, for example, to an unspecified Si–BO peak sometimes labeled “Si–O”[51–54,57,74,75] or a second Q2 peak.[58,59,76] (see Ref. [111] for further discussion).

    There were five major questions that needed to be addressed before one could fit and quantify Raman spectra of silicate glasses/melts with confidence.[81] First, what line shapes should be used to fit these Raman spectra: Gaussian, Lorentzian, or mixed Gaussian–Lorentzian? Second, are the FWHM well defined, as observed for crystals, and are the FWHM for all Q species of silicate glasses and melts similar? Third, are all the peaks contributing to the 800–1200 cm−1 region due solely to A1 symmetric stretches from the SiO4 tetrahedra as in crystals, or are there “additional” peaks? Fourth, are these peaks located at well-defined frequencies? Fifth, and most importantly, if there are “additional” peaks in the 800–1200 cm−1 region, what are their origins? In 2015, we re-examined the previously published Raman spectra of alkali silicates[27] to address these questions and show that Raman spectroscopy could become a quantitative technique for obtaining Q species distributions.

    First, consider line shapes. The inherent Raman line shape for inorganic glasses is Lorentzian,[112,113] as clearly stated by Efimov[113] (page 100): “the band shape … unaffected by disorder is known to have a Lorentzian contour”. For crystalline silicate minerals, the Raman peaks are also Lorentzian.[55,114,115] As stated by Rossano and Mysen,[116] disorder in glasses should introduce a Gaussian component to line shapes; however, all binary alkali and alkaline-earth glass Raman spectra considered in the recent literature[79–85] yield > 50% Lorentzian peaks. It is likely that more complex glasses will exhibit a larger Gaussian component, but this remains to be tested.

    There were early attempts to consider peak shapes other than Gaussian;[26,50,117] however, most papers, such as Matson’s high-quality Raman spectra of alkali silicates,[27] did not fit their spectra or comment on line shapes. Mysen[26,51–54] did fit his spectra, but did not explore alternative line shapes, probably because Lorentzian and Gaussian–Lorentzian fitting programs were not available. Mysen states that “it is also assumed that the individual bands have Gaussian line shape”.[26] There was only one publication (in 1984) that strongly indicated that a Lorentzian line shape yields a better fit than a Gaussian line shape for a K silicate glass,[117] but this paper was largely ignored.

    Because of the dominance of Gaussian line shapes in the literature, it was initially quite surprising that our new spectra of alkali silicates[79–82] gave consistent peak shapes of high Lorentzian character (e.g. Fig. 7). It is critical to emphasize that we do not “choose” a line shape or make any “assumptions” (as Mysen[26] did above) about the peak shape. A large Lorentzian character is immediately indicated by the “tails” in Fig. 7 on the narrow, intense Q3 peak. An unconstrained fit, which varies the Lorentzian/Gaussian character as well as the peak position and width, yields the best fit to the spectra. There are no constraints on the Q3 peak parameters (position, width, and %Lorentzian) in this fitting procedure. The Cs and Rb silicate glass spectra are nearly 100% Lorentzian, with 98% Lorentzian giving the best fit for the spectrum of the 5 mol% Cs2O glass in Fig. 7(c). The Lorentzian “tails” are obvious in these spectra; if a Gaussian were used (Fig. 7(a)), other small peaks would be required to fit the “tails” (e.g., see Fig. 3 of Ref. [76]). As noted above, the Q2 and Q3 line shapes in all other alkali and Ba silicate spectra[79–85] are 50–80% Lorentzian. The use of a strong Lorentzian component is especially important where a major central band is flanked by two much weaker bands, such as seen in the 29Si NMR spectrum in Fig. 5(b). The line shape of the large central Q2 peak can make a significant difference to the area of the small unresolved Q3 peak (see Subsection 2.5 below).

    Second, the linewidths (FWHM) for well-resolved peaks (e.g., the Q3 peak in Fig. 7(c) or the Q2 peak in Fig. 8) of many Raman spectra of binary alkaline and alkaline-earth glasses are similar at 40–55 cm−1.[79–83] The Q1 linewidths, however, are appreciably narrower (≤ 30 cm−1), as evident in spectra of Ba silicates.[84,85] Also, the linewidth of the overall composite Q3 band is broader than that of the Q2 band (Fig. 8). The linewidths are quite constant as a function of composition in a given binary system. For example, in the Cs spectra noted above, the Q3 linewidth varies from 40–55 cm−1 over 5–30 mol% Cs2O.[79–82] In melts, the linewidths are appreciably larger and often approach 100 cm−1 (see later in Subsection 2.6).[52,55]

    We now address the third problem–the assignment of the peaks. The early Raman studies[23–28] assigned peaks at ∼ 850 cm−1, ∼ 950 cm−1, and ∼ 1100 cm−1 to the Q1, Q2 and Q3 species, respectively (Figs. 6–8). As summarized by McMillan,[28,110] these major peaks can be readily assigned to the A1 symmetric stretches of SiO4 tetrahedra of those Q species, and recent calculations confirm these assignments.[111] The cross sections for these A1 stretches for different Q species should be similar.[25,83,118,119] Furukawa’s initial calculations[25] indicated that the Q2 cross section should be ∼ 30% larger than the Q3 cross section; however, two later papers[118,119] gave virtually equal cross sections for Q2 and Q3. Also, it was concluded[81] from a comparison of the 29Si NMR and Raman spectra of a ∼ 50 mol% Na2O glass that the Q1 cross section was similar to those of Q2 and Q3. The Q4 peak is unusual, because the Q4 tetrahedron in vitreous SiO2 does not have an A1 stretch in this region.[120] However, a detailed analysis of the Q3 and Q4 intensities in Cs glasses[80] shows that the Q4 cross section is not appreciably different from the Q3 cross section for ≥ 30 mol% Cs2O glasses. For the glasses in this paper (≥ 40 mol% M2O), the Q4 abundance is small, as shown by static 29Si NMR for the 40 mol% Na2O and BaO glasses (2% and 6.7%, respectively[32,84]).

    Fourth, the peak maxima for a given Q species are usually similar, as emphasized by McMillan.[28,110] For example, the Q3 peak maxima in all alkali and alkaline-earth glasses with ≤ 40 mol% M2O are at 1080–1105 cm−1;[23–28,79–85,110,111] and for a series of Cs glasses between 5 and 30 mol%,[80] the range is 1095 to 1099 cm−1. In contrast, the Q2 peak position decreases significantly as the electronegativity (EN) of the cation decreases from Li to Cs (e.g., from 954 cm−1 for Li to 928 cm−1 for Cs[27,106]). Also, for ≥ 45 mol% Na, K and Ba glasses,[76,81,84] the Q3 peak broadens and shifts to significantly lower energy, as discussed in the next section (Section 2.4) on K glasses.

    The fifth problem–the origin of “extra” peaks–is addressed using the 29Si NMR and Raman spectra in Figs. 9 and 10. Figure 9(a) shows the 29Si NMR spectrum of the 41 mol% Na2O glass again[32] (see also Fig. 5(a)), highlighting the similarity of this NMR spectrum with the three Raman spectra of the 40 mol% Na2O, 40 mol% K2O, and 40 mol% BaO glasses (Figs. 9(b)–9(d), respectively). All four spectra consist of two intense peaks, with the peak on the left being less intense and narrower than the peak on the right. Because of the poorer resolution of the 29Si NMR spectrum, it is likely that some of the broadening of the 88 ppm peak in Fig. 9(a) is due to the asymmetric broadening seen on the 1100 cm−1 peak in all three Raman spectra (Figs. 9(b)–9(d), see later). Perhaps surprisingly, in all four spectra, the left peak is always 30 ± 2% of the total intensity of the spectrum (Table 1).

    The assignment of the 29Si NMR spectrum in Figs. 5(a) and 9(a) is conclusive:[32,83] the smaller peak on the left arises from Q2 and the larger peak on the right arises from Q3. The O mass balance (Eq. (15)) demands that the Q3 intensity is about double the Q2 intensity, as seen from the average 29Si NMR values of XQ3 and XQ2 of 0.68 and 0.29 in Table 1. Note the excellent agreement with the observed Raman intensities of ∼ 70% and ∼ 30%. This agreement indicates that the assignment of the Raman spectrum must be analogous: Q2 at ∼ 950 cm−1 and the broader asymmetric Q3 at ∼ 1100 cm−1. This mass balance confirms that the broader Q3 peak in the 29Si NMR spectrum, and the asymmetric high-energy peak in the Raman spectrum, are due solely to the Q3 species. If the broad 1050 cm−1 peak (with about 30% of the area of the total spectrum[83]) is not due to Q3, e.g., from a “Si–O” peak[51–54] or an SiO2 peak,[75] then the ratio of Q3/Q2 becomes ∼ 1.4 instead of ∼ 2 required by the O mass balance. If the 1050 cm−1 peak arises from a second Q2 peak,[54,55,76] then the Q3/Q2 ratio is ∼ 0.7, which is clearly inconsistent with mass-balance considerations.

    These O mass balances, along with the similarity of the Q2 and Q3 intensities in all four spectra, show conclusively: (1) the asymmetry in the Q3 Raman peak arises solely from Q3 species; (2) the Raman cross sections for Q2 and Q3 are close to unity in all three Raman spectra; and (3) the Q species distributions in all three Raman spectra are remarkably similar. As initially suggested by McMillan,[110] the Q3 peak is due to two or more distinct Q3 environments. We have proposed that the “extra” Q3 peak(s), shifted by > 60 cm−1 from the main peak in Fig. 8, result from at least one Na atom bonded to BO as well as NBO.[81] Our observation of two or more BO peaks in the O 1s XPS spectra of Na and K glasses (Figs. 2 and 3) arises from these M–BO interactions;[20,21] and this result pointed us toward the important role of Na–BO bonding in Raman spectra. When the electropositive Na, K or Ba donates charge to a BO in the SiO4 tetrahedron, the Si–O Coulombic interaction is reduced, the Si–O force constant decreases, resulting in a lower Qn Raman frequency of the A1 symmetric stretch, and a decrease in the O 1s binding energy from BO to NBO.[121,122] The same charge-transfer mechanism is responsible for the decrease in Raman frequencies from Q3 to Q1 (∼ 250 cm−1) and in O 1s binding energies (∼ 2 eV) between NBO and BO when Na forms an NBO via Eq. (3).

    The best fits to these four spectra are shown in Fig. 10. There are a few important observations. First, the Q3 peak in the three Raman spectra (Figs. 10(b)–10(d)) has been fit using three peaks of identical shape (60% Lorentzian for all three spectra) and equal (but not constrained) widths of 55 cm−1, 52 cm−1, and 63 cm−1 for Figs. 10(b)–10(d), respectively. These three peaks represent two M–BO interactions on the three BOs in this Q3 species, e.g., Q3-0, Q3-1, and Q3-2, where 0, 1, and 2 indicate no, one, and two Na–BO bonds, respectively. No additional constraints are used other than peak shape and linewidth equality. Other fits with either two Q3 peaks[83] or four Q3 peaks[81] have also been applied successfully. A theoretical study also indicates that a range of different Q-species environments can be used to fit spectra of Na silicate glasses.[123] Second, and most importantly for quantification, the percentage of Q2 in all three spectra in Figs. 10(b)–10(d) is 30 ± 2%, as in Fig. 9, and the Q2 and Q3 fractions are not dependent on the number of Q3 peaks used in the fit. Third, each M–BO interaction decreases the SiO4 A1 symmetric stretch frequency by ∼ 40 cm−1[81,83] in all three spectra, consistent with similar electronegativities (EN) for Na, K, and Ba (0.93, 0.823, and 0.89, respectively). Indeed, the Q3-0, Q3-1, and Q3-2 peak positions in all three spectra are similar: 1090 cm−1, 1056 cm−1, and 1015 cm−1 for Na (Fig. 10(b)); 1084 cm−1, 1044 cm−1, and 995 cm−1 for K (Fig. 10(c)); and 1083 cm−1, 1048 cm−1, and 1013 cm−1 for Ba (Fig. 10(d)). The average of these values (1086 cm−1, 1049 cm−1, and 1007 cm−1) will be used in a subsequent discussion (and Fig. 11) of the 40–50 mol% samples when discussing the changes in Raman frequency and linewidth of the Q3 peak.

    The above assignment is now consistent across Na, K, and Ba Raman studies[81,84,85]: the entire asymmetric 1100 cm−1 peak arises from Q3 moieties. A recent independent paper supports this assignment.[78] However, resistance remains,[74,76,124] which is not surprising after 40 years of debate.

    The 29Si NMR results from Table 1 (with a new result from our fit to the spectrum in Fig. 5(a), reported in Table 2, second line) all yield FO values between 1 and 3 mol%, and the average of 1.9% likely has an uncertainty of ± 1 mol%. The Q2 and Q3 Raman cross sections were obtained by comparing the 29Si NMR abundances (assuming equal cross sections) with the Raman intensities. As shown in Table 1, the average Q species abundances from the 29Si NMR spectra for Q1, Q2, Q3, and Q4 for the 40 mol% Na2O glass are: 0.0%, 28.8%, 68.8%, and 2.4%, respectively. The Raman spectral intensities (taking 2.0% out of the Q3 intensity for Q4 based on the static 29Si NMR spectrum) are: 0.6%, 32.2%, 65.2%, and 2.0%, respectively (Table 2). Considering that the spectral intensity of a Qn species (IQn) is the product of its cross section (σQn) and mole fraction (XQn)

    The σQ3 is then 0.95 (65.2/68.8), and the σQ2 is 1.12 (32.2/28.8). Similar cross sections for Qn (n = 1–3) are expected from previous experimental and theoretical results.[81,115,116]

    To quantify other Raman spectra of silicate glasses containing different amounts of Na, or containing other elements such as K and Ba, these cross sections would need to apply universally. Is this possible? Considering that the Ba and K spectra are very similar to the Na spectrum (e.g., uncorrected % Q1, Q2, and Q3 for the 40 mol% BaO spectrum in Table 2 are 1.4%, 29.5%, and 69.1%, respectively[84]), it is worth exploring this possibility. Our independent fit of the 40 mol% Ba spectrum (Fig. 10(d)) gives Q species distributions (1.2%, 30.2%, and 68.6%, Table 2) in excellent agreement with Moulton’s values.[84] Moulton[84,85] derived cross sections from his own 29Si NMR spectra of Ba silicate glasses and compared them with the above values for the 40 mol% Na2O spectrum. He obtained Q3 and Q2 cross sections close to 1 for Ba glasses containing ∼ 35 mol% BaO, but the difference between σQ3 and σQ2 became larger for the 40 mol% (1.21 and 0.81) and 45 mol% (1.40 and 0.79) BaO glasses. Note that σQ3 > σQ2 for all Ba glasses, whereas σQ3 < σQ2 for the 40 mol% Na2O glass. Most of this difference is likely due to the poor resolution of the 29Si NMR spectra of Ba glasses. These 29Si NMR spectra, and indeed all 29Si NMR spectra of alkaline-earth silicates, have no defining features (e.g., see Fig. 6(a) for the spectrum of CaSiO3 glass). As emphasized in Subsection 1.1 and Ref. [68], it is not possible to obtain quantitative Q species distributions from an unconstrained three-peak fit to spectra with no defining spectral features. All such fits to the 29Si NMR spectra in the literature have effectively been constrained to Q species distributions that yield FO = 0 (e.g., Table 2). We suggest that it would be better to use equal Raman cross sections for Q1, Q2, and Q3 species and the Q4 abundance from static 29Si NMR spectra. With this approach, FO = 2.2% for the 40 mol% BaO glass (Fig. 10(d), Table 2). Using the cross sections from the 40 mol% Na2O spectrum, FO = 2.9% (Table 2). These FO values are consistent with other FO values obtained for alkaline-earth glasses noted earlier (1–2.6 mol%)[69] from other spectra and MD calculations. This promising result indicates that cross sections very close to unity are appropriate for alkali and alkaline-earth glasses.

    Now we turn to the quantification of the Raman spectrum of 40 mol% K2O glass (Fig. 10(c), Table 2). The fit in Fig. 10(c) is remarkably similar to the fit to the Na spectrum in Fig. 10(b), including the three-peak fit to the Q3 envelope. The 29 mol% Q2 species is substantially lower than the Q2 value for the 40 mol% Na2O fit (31 mol%), indicating greater polymerization (i.e., more FO) in K than in Na glasses. This result is consistent with the O 1s XPS results in Figs. 2 and 3. Using equal cross sections, FO for the K glass (1.2%) is larger than for the Na glass (0.6%). With σQ3 = 0.95 and σQ2 = 1.12, the FO of the 40 mol% K2O glass increases to 1.8 mol%. Note that we also list the abundances from 29Si NMR, but Maekawa does not show his fit for this glass,[41] and Malfait’s spectra of 38.5 and 42.9 mol% K2O glasses[46] have poor resolution and cannot be used to standardize the Raman spectra. Note again that the 29Si NMR Q values yield an FO of 0. Malfait[46] demonstrates one of the “adjustment” techniques that had to be applied to such unresolved spectra.[46,67] Their initial fits yielded negative FO values, which they recognized as physically impossible. They then “adjusted” the numbers in unspecified ways to enforce FO = 0. Some of these adjustments were large.[68]

    Several important conclusions follow directly from our analysis of Figs. 9 and 10, along with Tables 1 and 2. First, the best-resolved 29Si MAS NMR spectrum for the 41 mol% Na2O glass, and three other such studies on 40 mol% Na2O glasses (Table 1),[32] provide strong evidence for ∼ 2 mol% FO in these glasses.[32] As correctly stated by Stebbins,[63] these FO values are quite sensitive to chemical composition (e.g., a decrease in Na2O content of 1% causes a decrease in FO of ∼ 1%) and to small amounts of Q1 and Q4 species; however, the overall uncertainty is less than ± 1 mol%, especially considering that there are four independent measurements quoted in Table 1[32,41,42,64] along with our independent fit in Fig. 9(a). Second, the quantitative comparison of the 29Si NMR and Raman spectra of the 40/41 mol% Na2O samples shows that the Raman spectrum can be assigned to a narrow Q2 peak and a broader asymmetric Q3 peak. Third, the Raman spectra can be fitted without constraining intensities, resulting in Raman cross sections that are close to unity. Fourth, for the three 40 mol% samples, the Q3 peak is composed of three components, each separated by about 40 cm−1, which are attributed to M–BO interactions that decrease the Q3 Raman frequency. Fifth, the remarkable similarity of the Na, Ba, and K spectra shows that there is a comparable amount of FO (∼ 2 mol%) in the three glasses. Sixth, the similar amounts of FO in these alkali and alkaline-earth glasses indicate that the thermodynamic treatment of Eq. (1) based on crystalline phases[61–63] cannot be applied to homogeneous solid solutions, as pointed out by Fraser[125,126] and Nesbitt.[64] Moreover, the cation field-strength argument used by Stebbins[61–63] does not predict the degree of polymerization: low-field alkali cations produce FO comparable to that in alkaline-earth glasses.

  • There is strong evidence for FO in the 40 mol% Na2O glass as indicated by both 29Si NMR and Raman spectra. In both cases, the Q3 abundance is greater (i.e., the glass is more polymerized) than expected if only disproportionation reactions occurred. Moreover, the similarity of the Raman spectra of 40 mol% Na2O, K2O, and BaO glasses indicates ∼ 2 mol% FO in all three systems. Even stronger evidence for FO in potassic glasses is provided by re-analysis of the classic 29Si NMR spectrum of 50 mol% K2O glass (Fig. 11(a)) from 1991.[41] The original spectrum was misassigned,[41] it was misassigned again in 2007,[46] and it remained misassigned in 2020.[63] As now shown, with reassignment, the 50% K2O glass contains 7–9 mol% FO, which is consistent with the O 1s XPS results in Fig. 3.

    Maekawa et al.[41] noted that the 50% K2O spectrum (Fig. 11(a)) was different from other metasilicate 29Si NMR spectra in that it contained two intense bands. The 50% Na2O glass spectrum (Fig. 5(b)), as well as all previously published 29Si NMR spectra of alkali and alkaline-earth silicates, showed no indication of two resolved peaks arising from a single Q species, and the assignment of the 50 mol% K2O spectrum was therefore not straightforward. By Ockham’s razor, the simplest approach would be to assign the four spectral bands to four different Q species: Q1, Q2, Q3, and Q4 at 70.8, 76.2, 81.7, and 89.0 ppm, respectively, as had been done for earlier NMR spectra. Maekawa et al.[41] considered this possibility, but it leads to Q3 ≫ Q1, with NBO/Si = 1.44, which was inconsistent with the disproportionation reaction (i.e., 2Q2 ⇔ Q1 + Q3). By “conventional wisdom,” the assignment was therefore regarded as incorrect (and not publishable). The prevailing view required Q1 = Q3 and NBO/Si = 2.0 for all metasilicate glasses. Maekawa et al.[41] thus adopted a more complex interpretation, stating that: “If we assume instead that two types of the Q2 species are present in the potassium silicate glasses and the two (intense) peaks at ∼ 82 ppm and ∼ 76 ppm (Fig. 11(a)) are both assigned to Q2 species, the NBO/Si ratio satisfies the stoichiometric value” (2.0). He acknowledged that there was no direct evidence for two Q2 species in alkali glasses but suggested that Q2 species in chain and ring configurations observed in potassium silicate aqueous solutions could give rise to two Q2 signals. There was, however, no evidence for chains in silicate glasses, as noted in Henderson’s 2005 review (Ref. [17] in 1942), and there is still no evidence in 2025. Despite this lack of evidence, subsequent 29Si NMR and Raman studies[46,58,59] adopted the “two Q2 peak hypothesis,” including Stebbins’ review in 2020[63] and Koroleva’s work in 2023.[76] Because of the high reactivity of this glass, it is important to note that later 29Si NMR spectra of 50 mol% K2O glass[46,63] are very similar to the Maekawa spectrum[41] shown in Fig. 4(b), strongly suggesting that the composition is close to the nominal 50 mol%, and that no significant decomposition occurred in these studies.

    There is now strong Raman evidence supporting a simple “one peak per Q species” assignment. It comes from the latest Raman spectra of 30–50 mol% K2O glasses shown in Fig. 11(b).[76] Each Q species exhibits a single resolved band in these spectra. Specifically, Q1 appears at ∼ 850 cm−1, Q2 at ∼ 950 cm−1, and Q3 at 1070–1100 cm−1. The Q3 band is broadest for the 50% K2O glass (Fig. 11(b)), but importantly, only one Q2 band is observed. The assignments for the 50 mol% K2O glass spectrum (Fig. 11(b)) correlate closely with the 29Si NMR spectrum (Fig. 11(a)): both exhibit a weak Q1 feature at low ppm/frequency, an intense, relatively narrow Q2 peak at intermediate ppm/frequency, and a broad, intense Q3 peak at higher ppm/frequency. If two distinct Q2 contributions were present in potassic glasses, they would have been resolved in the higher-resolution Raman spectrum of the 50 mol% K2O glass. The Malfait et al.[58] spectra for 39.1–49.4 mol% K2O glasses (their Fig. 6), as well as the 40 mol% Na, K, and Ba spectra (Figs. 10(b)–10(d)), show the same sequence of bands. The correlation of peak assignments between Raman and 29Si NMR spectra strongly indicates that the broad intense peak in the 29Si NMR spectrum (centered at ∼ 82 ppm) corresponds to a Q3 band, while the weak feature centered at ∼ 90 ppm corresponds to a Q4 band. Ockham’s razor therefore leads to a simpler and more consistent interpretation of the K silicate 29Si NMR spectrum.

    In both the 29Si NMR and Raman spectra, the increase in broadening and the shifting to lower energy of the Q3 band from 30–50 mol% K2O is readily explained using the expected increase in K-BO interactions with increasing K content. Consider this K-BO effect on the Raman spectrum first. As outlined earlier, each K-BO interaction decreases the SiO4 A1 symmetric stretch frequency. As already shown for the three 40 mol% Raman spectra in Figs. 9 and 10, the three Q3 peaks (Q3-0, Q3-1 and Q3-2) are at ∼ 1090 cm−1, ∼ 1050 cm−1 and ∼ 1010 cm−1 respectively (also see Figs. 11(b) and 11(c)). The Q3-1 and Q3-2 peaks should become more intense with addition of K from 40 mol% to 50 mol% (Fig. 11(b)). The increase in Q3-1 and Q3-2 peak intensities has to: (1) shift the Q3 peak maximum to lower energy and (2) broaden the overall peak, as observed. This effect is even more noticeable in the Raman spectrum of the 50 mol% Na2O glass (Fig. 11(c)), published by several groups,[23,52,55,127] where the overall Q3 maximum is close to 1000 cm−1, and Q3-2 is the most intense Q3 peak.

    Now consider, apparently for the first time, the K-BO effect on the chemical shifts and peak widths in the 29Si NMR spectra. Nearly all the 29Si NMR studies of Na and K glasses (e.g. Refs. [32,41,63,64]) show that the full-width at half-maximum (FWHM) increases markedly from Q1 (≤ 5 ppm) to Q2 (6–7 ppm) to Q3 (10–11 ppm), and these linewidths are independent of the magnetic field strength (4.7 T to 14 T) used to obtain these spectra (as noted earlier). Note that the Raman linewidths also increase similarly,[81–85] although it is more difficult to obtain a quantitative overall Q3 linewidth from the asymmetric Q3 peaks in Figs. 9 and 10. For example, the Q1 and Q3 Raman linewidths for the 40 and 50 mol% Na2O and K2O glasses are ∼ 40 cm−1 and > 70 cm−1 respectively (Figs. 10 and 11[81]). Only Maekawa[41] has mentioned this large variation in the Q species NMR linewidths, and he attributed the large difference between Q1 and Q3 to a larger distribution in ‘bridging angle, bond length or third neighbor environments’ as n increases from Q1 to Q3. Our K-BO mechanism mentioned above for Raman and O 1s spectra can now explain the broadening and shifting of the Q3 peak as K increases.

    In Fig. 11(d), we show the strong correlation between the Raman shift and 29Si NMR shifts for Q1, Q2 and Q3. The 29Si NMR Q2 and Q3 chemical shifts were obtained by averaging values for the 33.3 mol% and 36.4 mol% Na2O and K2O glasses;[41] and the Q1 chemical shift is the average of the chemical shift from the 50 mol% glasses.[41] Results from other NMR studies are in good agreement.[63,64] The Raman frequencies (Q3-0, Q2 and Q1) are derived from averaging the frequencies from the three fits in Figs. 10(b)–10(d). The range of Raman and 29Si NMR shifts are indicated by the bars. Clearly, when the NMR shift becomes more negative, the Raman shift increases; and the trend is close to linear between all five Q species. From the average Q3-1 and Q3-2 Raman shifts (1049 cm−1 and 1008 cm−1), we can then predict that the 29Si NMR chemical shift decreases substantially from 89.5 ppm (Q3-0) to ∼ 86.5 ppm (Q3-1) to ∼ 83.5 ppm (Q3-2) as shown in Fig. 11(d). When Q3-2 is the most intense peak at ∼ 1000 cm−1 in the Raman spectrum (as indicated in Fig. 11(c) for the 50 mol% Na2O spectrum), the NMR peak should shift by ∼ 6 ppm to ∼ 84 ppm as shown more clearly by the resolved peak in the K2O NMR spectrum at 82 ppm (Fig. 11(a)). Thus, a shift in the Raman Q3 peak (from Q3-0 to Q3-2) by ∼ 80 cm−1 leads to an NMR shift of ∼ 6 ppm. This shift to lower frequency in the Raman spectra (Fig. 11(b)), and the broadening from larger amounts of Q3-1 and Q3-2, should have similar effects on the 29Si NMR spectra. This expected shift by ∼ 6 ppm in the Q3 peak is critically important for the assignment of both the 50 mol% K2O spectrum and the 50 mol% Na2O spectrum in section 2.6 below. There is no longer any need for a new Q2 peak as postulated by Maekawa[41] and Malfait[67]: their new Q2 peak is indeed our Q3-2 peak broadened by Q3-1 and Q3-0. This confirms our previous assignment in 2015[68] and strongly suggests that our plot of 29Si NMR chemical shifts for the 50 mol% alkali silicate glasses (Fig. 3 in Ref. [68]) is the meaningful chemical shift trend. The trend is no longer “inconsistent with known chemical shift trends”.[63,67]

    A few theoretical comments on the trends of 29Si NMR chemical shifts, and the reasons for the correlation in Fig. 11(d), are warranted. Obviously, the 29Si peak shifts to smaller negative ppm with increase in electron density on the Si from the increase in Na/K content[121] on going from Q3 to Q2. For the 40 mol% Na2O glass (Figs. 4(a), 9(a) and 10(a)), the Q3 and Q2 positions are −88 ppm and −77 ppm respectively. As outlined by Maekawa[41] and Tossell,[128] the 29Si chemical shift shielding constant (σ) consists of three terms, with the local paramagnetic term, σpara, dominating. It can be expressed as

    where C is a constant, Q is the electron density on the Si, 〈r−3〉 is the expectation value of r−3 for valence Si 3p electrons, and ΔE is the average excitation energy from highest occupied molecular orbital (HOMO) to lowest unoccupied molecular orbital (LUMO). The Q and ΔE terms dominate. With the addition of electron density to the SiO4 tetrahedra from additional Na or K via NBO and/or BO bonds, the Q and ΔE terms reinforce each other; Q increases while ΔE decreases.[121,129,130] Changes in the charge and electron density on the Si from Na and K bonding to the SiO4 tetrahedra should have a similar effect on the 29Si NMR as has already been discussed for O 1s XPS and Raman spectra. We propose that this change in electron density is the major factor for the NMR chemical shift trend, and a strong contributor to the increase in NMR linewidths from Q1 to Q3. Additional theoretical and experimental studies are now required.

    From the above assignment of the Raman spectrum to Q3, and the similarities between the Raman and 29Si NMR spectra, the 29Si NMR spectrum of the 50 mol% K2O glass must be assigned as suggested by Ockham’s razor: one peak per Q species. Without calculating the FO content, it is immediately obvious that Q3 ≫ Q1 and there is a large FO present. This FO has been calculated in four different ways in Table 2. The first FO content from the quoted 29Si NMR intensities[41] is 9.7%. To show the sensitivity of the FO to a change in the assignment, by including the small broad Q4 peak as Q3, we obtain 8.2 mol%. From our preliminary fit to the Raman spectrum using equal cross sections, one obtains 7.4%; and by using the Q cross sections from the 40 mol% Na2O spectrum,[83] 8.5 mol% is obtained. The FO is 7–10 mol%. These values are in good agreement with those predicted by the O 1s results (see Fig. 4). Polymerization is important as reflected in the NBO/Si of ∼ 1.5 (1.44 given initially[41]). Further studies of 40–50 mol% K glasses are required to better quantify the FO content. Unfortunately, the literature 29Si NMR spectra for the 35–50 mol% K2O samples are very poorly resolved.[67,68]

    Obviously, the above results (and the O 1s XPS results in Figs. 3 and 4) show that FO is larger for K glasses than Na glasses of the same composition because of greater polymerization of K glasses. There is additional spectroscopic evidence which supports this conclusion. In the very first 1975 Raman spectra of K and Na silicate glasses (Figs. 2, 3 and 4 in Ref. [23]), it is obvious that the three glasses between 33.3 mol% and 40 mol% M2O display a Q2 peak in the K spectrum that is substantially smaller than Q2 in the Na spectrum. Even for 20 and 25 mol% (Kx, Na1−x)2O (x = 0–1) silicate glasses, recent Raman spectra[131] showed that K glasses are more polymerized than the comparable Na glasses, as indicated by the O 1s XPS spectra in Figs. 3 and 4. All these spectra show small peaks due to Q2, but the K glasses have very noticeable smaller Q2 than the comparable Na glasses.

  • In 2020, Stebbins published the 29Si NMR spectra of a number of glasses with ∼ 50 mol% Na2O and K2O,[63] all of which were similar to the original spectra published by Maekawa in 1991[41] (Figs. 5(a) and 11(a) respectively, with Maekawa’s fits). He concluded that the FO content in both cases was “below detection limits of about 0.5 to 1 mol% FO”. There are three serious problems with his study.

    First, and most importantly, the resolution of the weak Q3 shoulder in the 29Si NMR spectrum of 50 mol% Na2O (Figs. 1 and 5(b)) is not sufficient to obtain an objective estimate of the Q3 intensity and FO content (see Subsection 1.2). Stebbins emphasizes the problems by using adjustments such as unreported Lorentzian character for the central Q2 peak to account for the Q2 “tails”,[63] and an extra Q2 peak (of “no structural significance”[63]). Both of these “adjustments” decrease the Q3 intensity to give the desired result of FO = 0. Maekawa’s fit in Fig. 5(b) has similar problems. The Q2 peak maximum in his fit reaches the top of the experimental spectrum, and this is not possible with an overlapping Q3 peak as shown in Fig. 1. Moreover, the three-peak fits from both Stebbins and Maekawa are very similar to the fit in Fig. 1(e), which has the highest RMS value.

    Second, his previous 29Si NMR study of the 41 mol% Na2O sample (see above) yielded 2.8 mol% FO (Table 1) from his own data! The equilibrium considerations from Eq. (3) show that the 50 mol% sample should have about twice the FO as the 40 mol% sample (Fig. 3). Either one study or the other is incorrect! The resolution of Q2 and Q3 peaks in the 50 mol% spectrum is poorer than in the 40 mol% spectrum and, as shown above in Subsection 1.1, the 50 mol% spectrum cannot be fit objectively to obtain a reliable Q3 amount. In contrast, we have already shown that the 40 mol% spectrum can be fit without constraints to give a reliable Q3 content. The 40 mol% spectrum is far more likely to be correct.

    Third, the Raman spectrum of 50 mol% Na2O glass (Fig. 11(c)) was ignored. This spectrum has been duplicated by at least five groups,[23,52,55,81,127] with the first published in 1975[23] and the latest in 2013.[127] These Raman spectra all show the same major features as the 29Si NMR spectrum — a weak narrow Q1 peak, a narrow intense Q2 peak and a very broad unresolved Q3 envelope. As always, the Raman spectrum is better resolved than the 29Si NMR spectrum and fits to the Raman spectrum (Fig. 11(c))[81] gives FO = 4 mol% (Table 2) assuming equal cross sections. If the chemical composition is only 48 mol% Na2O (unlikely for all five studies), FO would be ∼ 2 mol%. Note also that the Q3 envelope in the 50 mol% Na2O Raman spectrum (Fig. 11(c)) is very broad and shifted to lower frequency, as for the 50 mol% K2O spectrum in Fig. 11(b). Three peaks of equal width (with no position or intensity constraints) give positions of 1079, 1049 and 1015 cm−1 as shown by dashed lines on Fig. 11(b). It is perhaps surprising that these two peak positions are so close to those in the 40 mol% spectrum Fig. 10(b), namely 1080, 1050 and 1010 cm−1. This agreement gives more credence to the concept that each of these three Q3-0, -1 and -2 species can be represented by a well-defined peak.

  • Several excellent Raman spectra of silicate melts have been published, and three of these spectra of 50 mol% Na2O, Li2O and CaO are shown in Figs. 12(a)–12(c).[52,55] Both the spectra and their reasonable fits are taken directly from the two publications.[52,55] All of these spectra yield much broader lines and poorer resolution than for the glasses, as expected from the systematic increase in linewidths as the temperature increases (e.g. see Fig. 7 in Ref. [81]). The three spectra are qualitatively similar with two shoulders (at ∼ 850 cm−1 and ∼ 1050 cm−1) on either side of a strong central peak at ∼ 950 cm−1. Using our previous analysis of glass spectra, the peaks/shoulders can be readily assigned to Q1 (weak shoulder at ∼ 850 cm−1), Q2 (strong peak at ∼ 950 cm−1) and Q3 (strong shoulder at ∼ 1050 cm−1). Obviously, Q3 ≫ Q1 with this assignment, indicating significant polymerization and non-zero mol% FO for melts with alkali and alkaline earth cations. In the previous papers,[52,55] the ∼1050 cm−1 peak was not assigned to the expected Q3 peak,[52,55] mostly because “conventional wisdom” demanded that Q3 had to be equal to Q1 to obtain FO = 0 (Eq. (7b)). For example, the 1050 cm−1 peak in Fig. 12(c) was assigned to a Si-O0 bridging oxygen stretching frequency from all Q species.[52] This assignment is now known to be incorrect.[111,120]

    Quantification of these poorly resolved spectra is difficult, as emphasized in Subsection 1.1. Different linewidths and lineshapes will substantially affect the Q species estimates. But in this case, it is obvious (much more obvious than for Fig. 1) that Q3 ≫ Q1, indicating non-zero mol% FO in all cases. The spectra were fit in the original publications[52,55] to three or four peaks as shown in Fig. 12. The “extra” smaller peak at ∼ 1000 cm−1 can be readily assigned to our Q3-1, as in Figs. 9 and 10 for the silicate glasses. The Q species abundances (Fig. 12, Table 2) are taken from Table 1 in Ref. [52] for CaSiO3 (Fig. 12(c)), and from our equivalent fits to Figs. 12(a) and 12(b) for Na2SiO3 and Li2SiO3. For a preliminary semi-quantitative estimate of FO in these three melts, we use the Q abundances in Fig. 12 and Table 2. In all cases, Q3 ≥ 3Q1. FO contents of 3.3–5.0 mol% are obtained (Table 2), comparable to the FO of 4 mol% for the 50 mol% Na2O glass in Fig. 11(c) (Table 2). Note again that the FO abundances for the alkali cations Na and Li are comparable to, or larger than, the FO for the Ca melt. This is one more indication that both Stebbins’ thermodynamic treatment based on crystals, and the field strengths of cations,[61–63] are not relevant to the amount of FO in silicate solid solutions.[125,126]

  • Previous simulations of Na-silicate and Mg-silicate glasses and melts[11,101,132,133] indicate the presence of FO. For example, Na2SiO3 glass at 300 K[11] yielded 0.47 mol% FO, whereas MgSiO3 and Mg2SiO4 melt simulations at 3000 K yielded 12.5 mol% and 15 mol%, respectively.[101,132,133] Here, we report ab initio MD simulations on orthosilicate Na4SiO4. The orthosilicate composition was chosen because BO and O2− disappear simultaneously from an orthosilicate glass or melt if Rxn. (1) goes to completion, yielding only NBO: at this composition the presence of either reactant, BO or O2−, necessarily implies the presence of the other. No previous MD calculations have been performed on the Na4SiO4 composition.

    Micro-canonical ensemble (constant-energy, constant-volume) first-principles molecular dynamics (MD) calculations were performed on Na4SiO4 and Mg2SiO4 using the Vienna ab Initio simulation package (VASP[134,135]). The core electrons of each atom were replaced by ultrasoft pseudopotentials.[136,137] The generalized gradient approximation[138] with the Perdew–Burke–Ernzerhof parametrization[139] was used for the exchange-correlation functional. The valence electron orbitals were expanded in a plane-wave basis set, and Brillouin zone sampling was performed at the Γ point. To verify the results, calculations were also performed on Na4SiO4 using the more accurate projector augmented-wave (PAW33) potential.[135] A 396 eV energy cutoff was used in all calculations.

    Following de Koker et al.[101] and Tse,[134] the model systems consist of 16 formula units (Na4SiO4 = 144 atoms and Mg2SiO4 = 112 atoms). The melted glasses were obtained by heating crystalline Na4SiO4[140] to between 4000 K and 5000 K. Such heating is required to rapidly disorder the crystalline structure.[141] Our heating, cooling, and annealing protocol is similar to previous simulations on Mg-silicates.[132,133] The systems were melted and annealed for at least 3 ps at high temperature, then cooled for several ps per temperature step, with the final structure in all runs being annealed at 300 K for an additional 3 ps. The full heating and cooling cycle took over 20 ps.

    In one run, Na4SiO4 was heated from 300 K to 4000 K, then cooled in steps to 3000 K, 2000 K, 1000 K, 500 K, and 300 K. In a second run, Na4SiO4 was heated from 300 K in steps to 3000 K, 4000 K, and 4500 K, then cooled in steps to 2500 K, 1000 K, and 300 K. A third run used the more accurate PAW33 potential. It melted at 4500 K and was then cooled to 4000 K, 2500 K, 1000 K, and 300 K. With this cooling rate (necessary for reasonable computational cost), it is not possible to guarantee that the resulting low-temperature structure is the true equilibrium glass structure rather than a frozen melt; nevertheless, we refer to it as a glass. The three runs yield similar Si–O bond lengths and O–Si–O angles, as well as similar numbers of BO atoms at 1000 K and 300 K, indicating that the results are not particularly sensitive to the pseudopotential used or the cooling rate.

    Figures 13(a) and 13(b) show the silicate network in Na4SiO4 at 1000 K and 300 K, where Na atoms, Si atoms, and O atoms bonded to Si are shown to emphasize the abundance of bridging oxygen atoms present. The snapshots demonstrate that, upon heating, the structure becomes fully disordered, with no vestige of the original crystalline Na4SiO4 structure. Heating to 5000 K results in many dissociated Si–O bonds, producing SiO4, SiO3, SiO2, and Si2O7 units; there is one BO in the system. Decreasing the temperature to 3000 K results in the loss of all SiO2 units and all but one SiO3 unit. One Si2O7 and one Si3O10 unit are present, indicating three BO atoms. Below the melting point (1360 K), all Si atoms are coordinated to four O atoms, with three Si2O7 units (ringed) present at 1000 K (Fig. 13(a)) and two at 300 K (Fig. 13(b)), indicating 3 and 2 BO atoms respectively. One FO is evident in Fig. 13(b). The PAW potential run yielded 4 BO and 4 Si2O7 units at 300 K. These three Na-silicate simulations together provide results consistent with MD simulations using larger system sizes, as demonstrated by Van Ginhoven et al.[142]

    It can be difficult to count the number of FO, as indicated by a reviewer concerned with Na–O and Si–O cutoff distances. One FO was identified at the top right of Fig. 13(b). For this orthosilicate glass, it is easier to estimate the FO content from the number of dimer Si2O7 groups using Eq. (4), because each dimer gives rise to one FO (Eq. (4))

    or

    From this equilibrium equation, the number of dimers is equal to the number of FO (and, as an aside, the number of BO, one per dimer, is also equal to the number of FO). We observe between 2 and 4 dimers in our calculations, which yields an average of three FO – about 5% of the 64 O atoms in our 144-atom simulation.

    The calculated Si–O bond lengths are within the range observed by Baur et al.[140] for crystalline Na4SiO4 (1.59–1.68 Å) and are similar to bond lengths obtained from neutron scattering, EXAFS, and MD calculations using empirical potentials for Na silicate glasses (1.61–1.67 Å[8–15]). Our two other Na4SiO4 simulations yield Si–O average bond lengths within 0.01 Å of the above values. The O–Si–O bond angles are also similar to those obtained for crystalline Na4SiO4 (103.10°–116.90°[140]). The simulated Na–O bond lengths (av. = 2.305 ± 0.108 Å) are within the range observed for crystalline Na4SiO4 (2.28–2.55 Å[140]) and for Na silicate glasses (2.31–2.4 Å[11]).

    In Figs. 13(a) and 13(b) there are two obvious Si2O7 dimers out of sixteen initial SiO4 units. One other Si2O7 unit is less obvious. As previously noted, the FO content of the Na4SiO4 glass simulations averages about 5 mol% of total oxygen (3 out of 64). Its abundance is somewhat greater in simulated high-temperature Mg2SiO4 and MgSiO3 melts.[45,101,103,132,133] The presence of FO indicates that Rxn. (1) does not go to completion in alkali or alkaline-earth silicate glasses, as discussed above.

    Finally, we have performed a similar MD simulation on a larger 324-atom cluster of Na4SiO4.[143] This simulation also revealed a number of Si2O7 dimers with associated FO. A 34-atom cluster from this calculation shows the structure of one of these dimers in detail (Fig. 13(c)).

  • There are many examples from the literature which show that FO is critically important for the physical and chemical properties of these glasses. Indeed, as mentioned in a previous paper,[69] FO should be treated as a master variable in silicate melts, just as H+ (pH) is in aqueous solutions. Using both polymerization and disproportionation reactions at the atomic level,[144,145] we have shown that FO is directly involved in fundamental physical properties via nucleophilic substitution (SN) reactions when Lewis bases such as FO and NBO attack SiO4 tetrahedra to form five-coordinate SiV. FO must be a stronger nucleophile than NBO or BO, and FO should be the dominant nucleophile in basic melts with ≤ 40–50 mol% SiO2. Melting of crystalline alkali and alkaline-earth orthosilicates and metasilicates produces more polymerized melts than their precursor crystals, and FO is directly involved. We have shown, for example, that melting of crystalline Na2SiO3[55,99,144] results in conversion of Q2 species of the crystal to Q3 and FO. We have also shown that oxidation of Fe2+ to Fe3+ in mid-ocean ridge basaltic melts and glasses (MORBs) with mol% SiO2 ≤ 50 mol% is probably driven by FO present in all alkali and alkaline-earth melts. In addition, we show that SN reactions involving FO are critical for other physical properties such as diffusion, conductivity, and viscosity.[145]

    Regarding the effect of K on other physical properties, it is well known that replacement of Li, Na, or Mg by K can enhance the strength of glasses by 2–4 times, improve hardness, and decrease fragment size.[146,147] For example, the Raman spectrum of an initial 30 mol% Li2O glass shows a strong Q3 envelope along with a weaker Q2 peak of about 15% of the total spectral intensity.[82,146] After ion exchange of Li for K, the Q2 intensity decreases from ∼ 15% to about 9% of the total intensity (Fig. 6 in Ref. [146]). This shows that the K-rich surface is more polymerized than the Li bulk due to conversion of Q2 to Q3 via the polymerization reaction, which produces more FO. Such polymerization will increase the strength of the glass surface.

    We briefly consider the importance of FO for the chemistry of silicate glasses. Stolper[33,34] initially showed from IR spectra that there were two species in his melts after reaction with both CO2 and H2O: CO2 and CO32−, and H2O and OH− respectively, and it was simple to write

    and

    where Or represents a reactive O species (BO, NBO, or FO). Stolper noticed that at low Na2O glass compositions, mostly molecular CO2 and H2O were present after reaction with CO2 and H2O respectively; but for higher Na2O mol% glasses, only CO32− and OH− were present. Both NBO and FO increase with Na2O (Figs. 2 and 3), but FO increases much faster than NBO (Figs. 2 and 3), and FO is likely the reactive O species.[35]

    The Raman and NMR communities did not accept the presence of FO in these glasses, and, through conventional wisdom, the community adopted the convention that NBO was the reactive O species. It is much more likely that FO is the reactive species based on expected reactivity and steric effects.[35] In support, Eggler and Rosenhauer[148] in 1979 recognized that FO is the favorable species because the reaction “involves no rearrangement of the silicate network and will be strongly favored if such oxides are present”. Also, there is good evidence for “free-ionic clusters” (FIC, with CO32− bonded only to a cation) in low SiO2 glasses reacted with CO2.[149] FICs would be expected if FO is the reactive O species in Eq. (20).

    There is also evidence that the reactivity of K glasses with gases such as H2O and CO2 is significantly enhanced compared with the reactivity of Na glasses of the same composition. For example, we have suggested[35] that FO is likely to have a large influence on the reactivity of K silicate glasses and may explain their exceptional reactivity with H2O. Also, in relatively low-SiO2 glasses with about 45 mol% SiO2 (with 7 mol% Al2O3, 15 mol% (Mg,Ca)O and 20 mol% (Nax, K1–x)2O (x = 0 to 1)), the uptake of CO2 (as shown by the CO32− IR band at 1070 cm−1) is significantly enhanced when Na is replaced by K, as expected from the larger FO content in K glasses. For example, for glasses equilibrated at high pressure (50 MPa) and high temperature (1250 °C), the CO2 solubility is ∼ 1.75 wt% in the K-free glass (x = 1) and increases to ∼ 3.0 wt% in the Na-free glass (x = 0).[150]

    Finally, it seems appropriate to comment on the possible effect of FO on the bioactivity of soda-lime silicate glasses. Although commercial bioactive glasses contain several mol% P2O5, low-silica glasses with ∼ 50 mol% SiO2 and no P2O5 have also been shown to promote in vivo bone formation or osteogenesis. Two recent 29Si NMR and Raman studies on soda-lime silicate glasses with ∼ 70–50 mol% SiO2, ∼ 15–25 mol% CaO, and ∼ 15–25 mol% Na2O have attempted to correlate the observed Q species distributions with chemical reactivity.[77,78] For example, the 29Si NMR spectrum of a metasilicate glass with 50 mol% SiO2, 25 mol% CaO, and 25 mol% Na2O shows a poorly resolved spectrum indicating a weak Q1 peak (4.5%), a strong Q2 peak (71.8%), and a moderate Q3 peak (23.7%). Network connectivity (NC)[151,152] was used to discuss structural differences, and BO and NBO can be readily calculated from Eqs. (8) and (9). Note that NC = BO/Si = 4 – (NBO/Si), where 4 is the number of Si–O bonds. As discussed above, Q3 ≫ Q1 indicates polymerization and mol% FO (ignored in Ref. [77,78]). Calculation of FO using Eqs. (14) and (15) yields 2 mol% FO, as is also suggested by their higher-than-expected NC (BO/Si = 2.19 and NBO/Si = 1.81) versus the analyzed NC (BO) value of 2.09 (NBO/Si = 1.91). The decrease in NBO/T by 0.1 (1.91 to 1.81) also indicates ∼ 2 mol% FO, as noted earlier in this paper and in Table 1 of Ref. [145]. We suggest that this FO is the most reactive species in these glasses. Obviously, more research is needed to confirm this suggestion.

3.   Conclusion
  • We have realized our most important objective. We have presented overwhelming evidence for mol% FO in Na and K glasses from all three spectroscopies (O 1s XPS, 29Si NMR, and Raman), with 1.9 (±1) mol% in 40 mol% Na2O glass and ≥7 mol% in 50 mol% K2O glass. The evidence is too strong to ignore. The essential solid-solution equilibrium treatment (Eqs. (1)–(3)) shows that significant FO must also be present at much lower mol% M2O. For example, for K2O glasses, there should still be ≥ 1 mol% FO in K glasses with ∼ 30 mol% K2O, as also demonstrated by our O 1s XPS results. MD simulations on Na4SiO4 are also consistent with substantial FO, showing ∼ 5 mol% FO in a Na4SiO4 glass. Also, the 29Si NMR spectra of 50 mol% Rb2O and Cs2O silicates[67,68] also show FO contents comparable to K glasses. We conclude that alkali silicate glasses show comparable (or greater) FO% than alkaline-earth silicate glasses of comparable composition.

    The justification of FO = 0 for alkali glasses using the thermodynamics of crystalline phases and the low crystal-field strengths of alkali cations is not appropriate. The 29Si NMR spectra of the well-resolved 40 mol% Na2O and 50 mol% K2O glasses are critical to the present analysis; however, the poor resolution of nearly all other 29Si NMR spectra of alkali and alkaline-earth silicate glasses makes it difficult to justify the use of 29Si NMR alone to quantify Q species distributions in most silicate glasses.

    After reassignment of the Raman spectra (using Lorentzian peak shapes and M–BO interactions which broaden and distort the Q3 peak), we have shown that the Raman spectra and some 29Si NMR spectra are semi-quantitatively consistent, as initially envisaged by Franz and Mysen[52] and McMillan.[110] For example, in both spectroscopies there is one resolved peak per Q species, there is a good correlation of peak positions for all Q species, and even the relative widths of the Q species peaks are similar. Most importantly, the Raman cross sections are close to unity, and a reasonable estimate of FO can be obtained from the raw Raman spectra. For ≫ 40 mol% samples, these spectra can now be fit, assigned, and quantified as well as other spectroscopies such as Mössbauer, XPS, or NMR.

    We have also shown the importance of FO for determining physical and chemical properties such as melting and strength of glasses, reactions with gases such as H2O and CO2, and bioactivity. There is a need for more accurate measurements of FO (and more accurate Q species amounts and more accurate K3 values), which should enable better predictions of these physical and chemical properties.

    Following Richet’s article on the development of scientific truth[60] in the earth sciences, there was a scientific “consensus” in the 1980s, 1990s, and 2000s that disproportionation alone was responsible for Q species distributions. This led to rejection of papers that contradicted the accepted interpretation, along with “adjusting” results[51] to fit the prevailing model. There are some analogies with the controversy over the age of the Earth from the 1860s to the 1930s.[60] The geological community had a consensus from 1860–1905 that the age of the Earth was 20–100 M years. With the discovery of radioactivity in 1895, a new scientific community (physicists) using a new technique (isotopic dating) was rejected for decades by the geological community, which had to “adjust” their results. Richet concludes that “if truth can only be approached little by little, the intrusion of outsiders into the debate is certainly not to be rejected but rather encouraged to break with any group thinking”.

Figure (13)  Table (2) Reference (152)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return