Probing silicon and aluminium chemical environments in silicate and aluminosilicate glasses by solid state NMR spectroscopy and accurate first-principles calculations

Available online at www.sciencedirect.com

www.elsevier.com/locate/gca

ScienceDirect

Geochimica et Cosmochimica Acta 125 (2014) 170–185

Probing silicon and aluminium chemical environmentsin silicate and aluminosilicate glasses by solid state

NMR spectroscopy and accurate first-principles calculations

Elisa Gambuzzi a, Alfonso Pedone a,⇑, Maria Cristina Menziani a, Frederic Angeli b,Daniel Caurant c, Thibault Charpentier d,⇑

a Dipartimento di Scienze Chimiche e Geologiche, Universita degli Studi di Modena e Reggio Emilia, Via G. Campi 183, 41125 Modena, Italyb CEA, DEN, Laboratoire d’etude du Comportement a Long Terme, 30207 Bagnols-sur-Ceze, France

c Ecole Nationale Superieure de Chimie de Paris (Chimie ParisTech), Laboratoire de Chimie de la Matiere Condensee de Paris, (UMR

CNRS 7574), Paris, Franced CEA, IRAMIS, SIS2M, CEA–CNRS UMR 3299, 91191 Gif-sur-Yvette cedex, France

Received 23 January 2013; accepted in revised form 16 October 2013; Available online 28 October 2013

Abstract

Silicon and aluminium chemical environments in silicate and aluminosilicate glasses with compositions 60SiO2�20Na2O�20CaO (CSN), 60SiO2�20Al2O3�20CaO (CAS), 78SiO2�11Al2O3�11Na2O (NAS) and 60SiO2�10Al2O3�10Na2O�20CaO(CASN) have been investigated by 27Al and 29Si solid state magic angle spinning (MAS) and multiple quantum MAS(MQMAS) nuclear magnetic resonance (NMR) experiments. To interpret the NMR data, first-principles calculations usingdensity functional theory were performed on structural models of these glasses. These models were generated by Shell-modelmolecular dynamics (MD) simulations. The theoretical NMR parameters and spectra were computed using the gauge includ-ing projected augmented wave (GIPAW) method and spin-effective Hamiltonians, respectively. This synergetic computa-tional–experimental approach offers a clear structural characterization of these glasses, particularly in terms of networkpolymerization, chemical disorder (i.e. Si and Al distribution in second coordination sphere) and modifier cation distribu-tions. The relationships between the local structural environments and the 29Si and 27Al NMR parameters are highlighted,and show that: (i) the isotropic chemical shift of both 29Si and 27Al increases of about +5 ppm for each Al added in the secondsphere and (ii) both the 27Al and 29Si isotropic chemical shifts linearly decrease with the reduction of the average Si/Al–O–Tbond angle. Conversely, 27Al and 29Si NMR parameters are much less sensitive to the connectivity with triple bridging oxygenatoms, precluding their indirect detection from 27Al and 29Si NMR.� 2013 Elsevier Ltd. All rights reserved.

1. INTRODUCTION

Silicate and aluminosilicate glasses are materials of stra-tegic importance for technological applications: their prac-tical use ranges from optic fibre manufacturing to high-level

0016-7037/$ - see front matter � 2013 Elsevier Ltd. All rights reserved.

http://dx.doi.org/10.1016/j.gca.2013.10.025

⇑ Corresponding authors. Address: CEA Saclay, DSM/IRAMIS/SIS2M Bat 125, 91191 Gif-sur-Yvette Cedex, France.

E-mail addresses: [email protected] (A. Pedone),[email protected] (T. Charpentier).

nuclear waste confinement. Moreover, SiO2 and Al2O3 arethe main constituents of earth’s crust and mantle, wherethey are present together with many other oxides to formvery complex multicomponent materials. The study of ther-modynamics and transport properties, heat capacity ofmelts, silica activity and viscosity of silicate and aluminosil-icate glasses and melts is, thus, an issue of great interest ingeological and technological fields, and has inspired manyinvestigations (Cormier and Neuville, 2004; Duxson et al.,2005; McMillan et al., 1982; Neuville and Mysen, 1996;Stebbins and Xu, 1997; Toplis et al., 1997; Leko and

E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185 171

Mazurin, 2003; Xue et al., 2004; Provis et al., 2005;Kuryaeva, 2006; Neuville, 2006; Neuville et al., 2007;Henderson et al., 2009; Richet et al., 2009). Physical prop-erties of amorphous materials depend very strongly on theirstructure, which -in turn- depends on their chemical compo-sition. Briefly, the structure of aluminosilicate glasses hasbeen understood in terms of interconnected (SiO4/2) and(AlO4/2)� tetrahedra, and of non-framework cations. Thecations locally compensate the excess of negative charge of thenetwork, which is due to the presence of (AlO4/2)

� tetrahedraand of non-bridging oxygen atoms (NBOs) (i.e. oxygenatoms connected to only one network former cation), thatare introduced by adding network modifying oxides (suchas Na2O and CaO) to the glass composition.

On one hand, experimental techniques usually employedto investigate the structure of crystalline phases with an ex-treme accuracy, like X-ray diffraction and neutron scatter-ing, are much less effective with amorphous systems. Infact, the lack of periodicity of glasses hinders a deep inter-pretation of diffraction and scattering patterns: short rangeorder (SRO) as characterized by first-neighbours inter-atomic distances can be obtained, but it is generally difficultto obtain any direct information on intermediate range or-der (IRO), such as atomic speciations, chemical disorder, orbond angle distribution (BAD). On the other hand, solidstate nuclear magnetic resonance (NMR) spectroscopy,being very sensitive to the local environment (i.e., bond dis-tances and angles, coordination numbers) (Farnan et al.,1992; Zhang et al., 1996; Mauri et al., 2000; Ashbrooket al., 2001; Clark and Grandinetti, 2003; Clark et al.,2004) and to the chemical disorder in the second coordina-tion sphere (Merzbacher et al., 1990; Koller et al., 1997;Rocquefelte et al., 2007), has been firmly established as apowerful technique for glass structure investigation (Mass-iot et al., 2008; Ashbrook, 2009; Hanna and Smith, 2010;Eden, 2012). The extent of network chemical disorder inaluminosilicate glasses, i.e. Al and Si intermixing (see pic-ture in the Supplementary material), can be investigatedin terms of Si–O–Si, Al–O–Si and Al–O–Al species, whichare clearly detectable and quantifiable with 17O 3QMASNMR spectroscopy (Stebbins and Xu, 1997; Lee and Steb-bins, 2000; Lee, 2010). Moreover, (SiO4/2) tetrahedra sur-rounded by k (AlO4/2)� tetrahedra (k ranges from 1 to 4),indicated as Si(Q4)[kAl] species hereafter, are detectableby means of 29Si NMR MAS in one-dimensional andtwo-dimensional experiments (Engelhardt and Michel,1988; Duxson et al., 2005; Deschamps et al., 2008; Hietet al., 2009).

Topological disorder, described by the distribution of in-ter-tetrahedral Si/Al–O–T angles, can be quantified from27Al 3QMAS and 29Si MAS experimental spectra, exploit-ing well-known correlations between the 27Al isotropicchemical shift, diso, and Al–O–Si angles (Angeli et al.,2000), or the 29Si diso and Si–O–Si angles (Mauri et al.,2000; Mackenzie and Smith, 2002; Charpentier et al.,2009; Angeli et al., 2011).

In the past, short and medium range order in glassescould only be deduced by a peak assignment based onempirical knowledge of NMR parameters of active-nucleiin reference crystalline compounds (Engelhardt and Michel,

1988). However, NMR parameters are extremely sensitiveto structural variations and the topological disorder presentin glasses results in a continuous distribution of the NMRparameters, which causes an inhomogeneous broadeningof isotropic line, thus lowering the resolution achievablein crystalline solids. Therefore, the achievement of anunambiguous assignment is far from obvious and this isthe limit of this approach. The MD–GIPAW approach re-cently developed (Charpentier et al., 2013) has opened anew route for interpreting NMR parameter distributionsand refining the relationships between NMR parametersand local structural features. This approach is based onthe generation of structural models of the glass of interestby means of classical (or ab initio) molecular dynamics(MD) combined with density functional theory (DFT) cal-culations for obtaining the NMR parameters of each atompresent in the models. The accuracy of the structural mod-els obtained is, in turn, demonstrated by the consistency ofthe theoretical and experimental NMR parameters and theagreement between theoretical and experimental spectra(Ispas et al., 2010; Pedone et al., 2010a,b).

In this work, a synergetic computational/experimentalprocedure is exploited to investigate the structure of silicateand aluminosilicate multicomponent glasses with composi-tions 60SiO2�20Na2O�20CaO (CSN), 60SiO2�20Al2O3�20CaO(CAS), 78SiO2�11Al2O3�11Na2O (NAS), and 60SiO2�10Al2O3�10Na2O�20CaO (CASN). The aim is to gain new insights into(i) the extent of network chemical disorder, and (ii) the effectof non-framework cations on 29Si and 27Al NMR parameters,providing information regarding the correlation betweenthe NMR parameters of network former cations and theirlocal and intermediate environment. Such knowledge isvaluable for extracting information on the chemical andtopological disorder present in complex aluminosilicateglasses by using a structural NMR inversion approach,which allows transformation of NMR data into a distribu-tion of structural parameters (Farnan et al., 1992; Clarket al., 2004; Charpentier et al., 2009; Soleilhavoup et al.,2010; Angeli et al., 2011).

2. METHODOLOGY

2.1. Synthesis and characterization

The synthesis procedure employed to obtain CAS,CASN and CSN glasses was described in Angeli et al.(2007). The NAS glass sample was prepared by mixing99% 29Si enriched SiO2 (Cortecnet Cie), Al2O3 and Na2CO3

and melting them in a platinum crucible following a heatingrate of 1.7 �C/min up to 1350 �C and then keeping it for 3 h.To improve melt mixing and to decrease viscosity, the sam-ple underwent a second heating phase of +6 �C/min up to1450 �C and then was kept at this temperature for 1 h.The sample was then quenched in air at room temperature.Once cooled, the sample was ground with a mortar to in-crease glass homogeneity, melted again at 1450 �C for 3 h,and quenched at room temperature. It is worth to note that,since the CAS, CSN and CASN glasses were synthesizedfor a previous study (Angeli et al., 2007) with CaCO3 highlyenriched in 43Ca isotope (about 40%), samples were

172 E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185

produced in very small quantities that prevented us to ana-lyse the actual compositions of the glasses. However, thehomogeneity of the samples and the absence of crystalswere verified by XRD, NMR, and visual inspection (opticaltransparency).

The NMR data acquisition procedure for the 27Al activenucleus in CAS, CASN and CSN samples was described inAngeli et al. (2007). For the NAS glass, 27Al MAS spectrawere collected on a Bruker Avance II 500WB spectrometeroperating at a magnetic field of 11.72 T using 4-mm (outerdiameter) zirconia rotors spinning at 12.5 kHz. 29Si MASNMR spectra were collected on a Bruker 300WB spectrom-eter (7.02 T) with the sample spinning at 12.5 kHz. ACPMG pulse sequence (Larsen and Farnan, 2002) with�16–32 echoes was employed using a recycle delay of200 s (with typically 128–256 scans coadded).

2.2. Computational details

Three independent structural models containing about250 atoms were generated for each glass composition byclassical molecular dynamics simulations adopting themelt-quench method (Pedone, 2009). Glass compositionsand the number of atoms in the models are reported inTable 1.

The interatomic potential model employed here is basedon the core–shell approach, which allows a straightforwardtreatment of the polarizability of the oxygen ions (Dick andOverhauser, 1958). The total ionic charge Z is sharedamong a massless shell of charge �Y and a core of chargeZ + Y that are coupled by a harmonic spring potential. Theoxygen shell interacts with the Si, Al, Ca and Na cationsthrough a Buckingham term, and Coulombic forces act be-tween all species bearing a formal charge. A three-bodyscreened harmonic potential is necessary to control theO–Si–O and O–Al–O angles, for favouring tetrahedralcoordination. The full expression of the potential is givenby:

Uðrij; rc�s; hijkÞ ¼ Kqiqj

r2ijþ Aije

�rijqij

� �� Cij

r6ij

þ 1

2ks rcore�shellð Þ2

þ 1

2kb hijk � h0

ijk

� �2

exp � rij

q� rjk

q

� �ð1Þ

and the employed parameters (Tilocca et al., 2006; Pedoneet al., 2012b) are reported in Table S.1 of the Supplemen-tary material. The time step used for the integration ofthe equation of motion, run by the leap-frog algorithm asencoded in the DL_POLY package (Smith and Forester,

Table 1Glass compositions (mol%) and number of atoms in the models.

Glass SiO2 Al2O3 Na2O CaO

CSN 60% – 20% 20%NAS 78% 11% 11% –CAS 60% 20% – 20%CASN 60% 10% 10% 20%

1996), is 0.2 fs. This value is small enough to account forthe high frequency vibrations of the core–shell system(Pedone, 2009). The initial structural configuration wasgenerated by randomly placing the atoms in a cubic cell.The side length is calculated according to the experimentaldensities (Bansal and Doremus, 1986) (CAS: 2.624 g cm�3,CSN: 2.657 g cm�3, NAS: 2.369 g cm�3), and to a densitycalculated by the Priven empirical method (Priven, 2001)which is encoded in the SciGlass package (Leko and Mazu-rin, 2003) (CASN: 2.641 g cm�3). Glass structural modelswere generated in the NVT ensemble by holding the sys-tems at 3200 K for 100 ps, cooling at a nominal rate of�5 K/ps to 300 K, and finally keeping them at 300 K for200 ps.

The structural modelling employed here presents thewell-known problem that the melting temperature andrelaxation time scale during quenching are far above theexperimental ones. However, previous simulations havedemonstrated that even a faster cooling rate of around10 K/ps could also yield accurate structural properties (Vol-lmayer et al., 1996). Therefore a fast cooling rate is oftenused in MD studies of melt-derived glasses. The presentedmethod has already been successfully applied in generatinggood structural models for silicate (Tilocca et al., 2006;Pedone et al., 2010b), phosphosilicate (Tilocca et al.,2007; Pedone et al., 2012a) and aluminosilicate (Pedoneet al., 2012b) glasses.

In the case of the NAS glass, the first generated modelpoorly reproduced the 27Al quadrupolar coupling constant,CQ, with respect to the experimental counterparts. As inves-tigated in the Supplementary material, CQ was found to becritically dependent on Al–O–T angle distribution and an-gle mean value, as well as on the O–Al–O bond angle distri-bution. The models obtained with the original three-bodypotential parameters (kB[O–Al–O] = 100.0 eV/rad�2) wereexcessively disordered in terms of valence and inter-tetrahe-dral angles. This might be the consequence of a high cool-ing rate that does not allow complete relaxation of thestructure and leads to high temperature glass models, whichare obviously more disordered. The attainment of a struc-tural NAS glass model with theoretical NMR parametersin agreement with the experimental ones was achieved byincreasing the harmonic constant, kb, for O–Al–O valenceangles, and imposing a harmonic constraint on the inter-tetrahedral angle Al–O–Si. A complete description of anew MD method applied to NAS glass is reported in theSupplementary material.

MD-derived structural models were optimized (cellgeometry and atomic positions) by density functionaltheory (DFT) calculations and the NMR parameters wereobtained by using PAW (Blochl, 1994) and GIPAW

NSi NAl NNa NCa NO

162 – 108 54 432183 51 51 – 468141 90 – 51 468150 51 51 51 453

Table 2Pseudopotential details.

Atom Valence configurationused for ultra-softpseudopotentialgeneration

Coreradii[a.u.]

Angular momentumlocal channelprojectors

29Si 3s2 3p2 1.8 s, p and d27Al 3s2 3p1 1.8 s, p and d17O 2s2 2p4 1.3 s and p43Caa 3s2 3p6 4s2 2.0 s, p and d23Na 2s2 2p6 3s1 1.3 s

a 3d Level shifted as suggested by Profeta et al. (2004).

Fig. 1. Comparison between normalized experimental spectra(collected at a magnetic field of 11.7 T, dashed lines) and computed(black solid lines) 27Al MAS NMR spectra of NAS (a), CAS (b)and CASN (c) aluminosilicate glasses.

E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185 173

(Pickard and Mauri, 2001; Yates et al., 2007) formalisms,implemented in the CASTEP code (Segall et al., 2002).The generalized gradient approximation (GGA) PBE func-tional (Perdew et al., 1996) was employed. The core–valenceinteractions were described by ultrasoft pseudopotentials(generated on-the-fly in the CASTEP code) with parametersstated in Table 2. All the DFT calculations were performedat the C point by expanding the wave functions in planewaves with a kinetic energy cutoff of 700 eV. This valuehas been demonstrated to be suitable for total energy andNMR chemical shift convergence (Choi et al., 2009).

Outputs of DFT–GIPAW calculations were processedby the fpNMR package, a homemade code written byCharpentier (Charpentier et al., 2009; Pedone et al.,2010b), which allows to perform statistical and structuralanalysis, and to simulate NMR spectra (MAS, MQMAS)under various experimental conditions. The cutoff radii val-ues employed for the structural analysis, which is essentiallybased on the determination of Si–O and Al–O bonds and(Ca,Na)–O spatial neighbourhood, were obtained fromthe radial pair distribution functions averaged over the last201 configurations of the MD trajectories at 300 K (seeSupplementary material). This yielded the following cutoffradii values: 2.25 A for Si–O, 2.3 A for Al–O, 3.0 A forCa–O and 3.2 A for Na–O. Only the Si–O/Al–O bondinginformation was subsequently used for second coordinationsphere analysis of Si and Al.

Theoretical isotropic chemical shifts, dcalciso , were evalu-

ated from the calculated isotropic magnetic shielding,rcalc

iso , using dcalciso ¼ �ðrcalc

iso � rREFÞ, in which rREF for 29Siand 27Al were fixed to 322.1 (Pedone et al., 2010a) and555.2 ppm (Choi et al., 2009; Pedone et al., 2012b) by usinga-quartz and corundum, respectively.

27Al quadrupolar coupling constants CQ were calculatedwith a quadrupolar moment eQ of 140.4 mB. This value wasobtained by multiplying the experimental value of 146.6 mB(Kello et al., 1999) by a factor determined via a linear fittingbetween the experimental and theoretical 27Al CQ values ofcrystalline aluminate and aluminosilicates. This procedure isdescribed in the Supplementary material. In this study, wereport as theoretical NMR parameters those obtained bythe MD–GIPAW method and as experimental values thoseextracted from the experimental spectra through a fittingprocedure. Similarly, the theoretical and simulated spectraare obtained by MD–GIPAW and experimental (i.e., fitted)parameters, respectively.

3. RESULTS AND DISCUSSION

3.1. Aluminium: local environment

Fig. 1 shows the theoretical and experimental 27Al MASNMR spectra of CAS, NAS and CASN glasses. The peakmaxima at about 60 ppm of the experimental spectra, typ-ical of four coordinated Al atoms, are well reproduced bythe DFT calculations. In fact, all Al atoms present in thestructural models are in Al(Q4) tetrahedral units, exceptin the CAS models that contain 5.32% of fivefold Al(VAl) and 2.66% of Al(Q3) species. The presence of VAl inglasses is dependent on the sample preparation during thecooling phase. Our calculations suggest that, if VAl atomsare present in the glass, they can be easily distinguishedfrom the dominant Al(Q4) units in a MAS spectrum. Infact, VAl diso and CQ lie at about 20 ppm and 11.7 MHz,respectively. Conversely, Al(Q3) species are not well spec-trally distinguishable from Al(Q4). However, Al(Q3) canbe detected with 17O MAS NMR, as the non-bridging oxy-gen NBO(–Al) is at least 20 ppm more deshielded thanNBO(–Si) and BOs (bridging oxygen atoms), as recentlydemonstrated in both experimental (Allwardt et al., 2003)and computational (Pedone et al., 2012b) investigations.

174 E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185

Some of the Al atoms are found to be bonded to threebridging oxygens (TBOs) (also referred to as tricluster oxy-gens, 4.9% and 1.7% of the total oxygen atoms in CAS andNAS structural models, respectively); however, as discussedbelow, the 27Al NMR parameters are not significantly af-fected by this connectivity, rendering the difficulty of detect-ing TBOs by 27Al NMR experiments. Regarding 27Alspectra line shapes, experimental and theoretical spectraare quite similar for NAS (Fig. 1a), fairly similar for CAS(Fig. 1b) and very similar for CASN glass (Fig. 1c). Thetheoretical line shape of the NAS glass is broader than thatof the experimental counterparts (Fig. 1a), denoting thatthe quadrupolar interactions are slightly overestimated inour DFT calculations. Part of the overestimation can be as-cribed to the narrow inter-tetrahedral Al–O–T angles(Lippmaa et al., 1986; Angeli et al., 2000) (see Supplemen-tary material) due to the presence of a small quantity ofthree-membered rings. The observed spectral dissimilarityfound in CAS glass (Fig. 1b) has been recently discussedby Pedone et al. (2012b), where they showed that the shoul-der at lower chemical shifts is a spectroscopic fingerprint ofAl atoms directly bonded to TBOs, consequence of the factthat Al–TBO bonds distort the AlO4 tetrahedral units.Moreover, the presence of Al atoms connected to TBOsseems to be overestimated in our CAS models. This is basedon the study by Iuga et al. (2005) where it was revealed thatthe Al(Q4) species bonded to TBOs are slightly moreshielded than Al species bonded to fourfold BOs.

The comparison of the theoretical and the experimentalNMR parameters (gQ, diso and CQ) is reported in Table 3.The experimental values have been obtained by fitting the27Al 3QMAS spectra of CAS (Fig. 2c) and CASN(Fig. 2f) glasses as described in Angeli et al. (2007). Topresent the computed 27Al NMR parameters, AlO4 unitshave also been classified with respect to the speciation ofthe first four oxygen atoms in NBO, BO and TBO, in orderto investigate the effects of the TBO-connectivity on the27Al NMR parameters (see Table 3). The theoretical diso

Table 3Theoretical and experimental values of 27Al NMR parameters (AlO4 unitsbridging oxygen; TBO: triple bridging oxygen).

Al (Q4) units MD Pop.% diso (ppm)

Th. Exp.

NAS

4BO 86.2 58.8(6.2) 60.1(4.83BO:1TBO 13.8 61.4(5.3)

CASa

4BO 52.2 62.4(6.6)3BO:1TBO 35.6 63.5(7.5)2BO:2TBO 5.6 59.5(10.4) 62.8(7.02BO:1TBO:1NBO 1.1 68.2b

3BO:1NBO 1.1 65.9b

CASN

4BO 100.0 69.3(5.9) 62.3(5.4

a CAS glasses contain a negligible amount of AlO6 (4NBO:2TBO ddiso = 22.5 ppm, CQ = 14.8 MHz gQ = 0.4 and 4BO:1TBO diso = 43.2 (1.

b The standard deviation value is not available: one site only.

are in good agreement with the experimental value, espe-cially for the Al(BO)4 units, which are only slightly affectedby the increasing number of bonded TBO. This is not thecase for CQ: a significant increase is found when TBOscoordinate Al. The similarity of diso will make it difficultto observe Al sites connected to TBOs, where peaks are hid-den in the dominant Al(BO)4 line, which is broadened bythe CaO content. In summary, our calculations suggest thatTBOs are rather difficult to detect, as predicted in a 17ONMR study (Pedone et al., 2012b).

3.2. Aluminium second coordination sphere: framework

disorder

Fig. 2 shows the 27Al experimental 3QMAS spectra ofthe glasses, together with the comparison between the com-puted NMR parameter distribution P(CQ,diso) (which arerepresented by symbols) and the experimentally determineddistribution (represented by dashed lines): NAS (Fig. 2b);CAS (Fig. 2d) and CASN (Fig. 2f). The figure shows thatthe agreement of the CQ distribution is satisfactory for bothCAS and CASN glasses, whereas large discrepancy isobserved for the diso distribution in CAS. This is expectedbecause of the overestimated presence of TBOs whichare characterized by lower diso shift. The overestimationof 27Al CQ, in the case of NAS glass, will be extensively dis-cussed later. The different symbols shown in the figure indi-cate the different species of Al(Q4) surrounded by j Siand 4 � j Al atoms. These species are named Al(Q4)[jSi,(4 � j)Al]. The relative distribution of the differentAl(Q4)[jSi,(4 � j)Al] species on the P(CQ,diso) plots is notuniform neither in the CQ, nor in the diso dimension; tothe contrary, they tend to occupy different, even thoughoverlapping, regions, thus revealing a certain sensitivity of27Al NMR parameters to network chemical disorder. It isinteresting to note that in NAS glass, most of the overesti-mation of the mean CQ value is due to Al(Q4)[3Si,1Al] spe-cies. This suggests that our structural models overestimate

) in NAS, CAS and CASN glasses (NBO: non-bridging oxygen; BO:

CQ (MHz) g

Th. Exp. Th. Exp.

) 6.4 (2.2) 5.0(1.6) 0.6 (0.3) 0.6(0.3)10.1 (3.5) 0.7 (0.2)

8.6 (2.8) 0.6 (0.2)9.4 (2.5) 0.6 (0.2)

) 13.1 (2.8) 8.1(2.6) 0.6 (0.2) 0.6(0.3)23.7b 0.1b

7.0b 0.8b

) 6.8(1.8) 6.5(1.8) 0.6(0.2) 0.6(0.3)

iso = 14.7 ppm, CQ = 6.5 MHz, gQ = 0.5) and AlO5 (2BO:3TBO4) ppm, CQ = 12.0 (2.7) MHz, gQ = 0.7 (0.2)) units.

Fig. 2. 3QMAS experimental spectra of NAS (a), CAS (c) andCASN (e) glasses (solid lines) and simulated spectra via fittingprocedure (dashed lines); P(CQ,diso) distribution of NAS (b), CAS(d) and CASN (f) glasses: fitted from experiment (dashed lines) andtheoretical counterparts derived from DFT calculations on MDderived models (symbols). Different symbols represent differentAl(Q4) sites, and refer to the speciation of Al(Q4) in Al(Q4)[-jSi,4 � jAl] species.

E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185 175

the presence of Al–O–Al bridges, which in fact are charac-terized by narrower averaged inter-tetrahedral angles. This,together with the vicinity of charge compensating Na, con-tributes to increase the average 27Al CQ value. The presenceof these Al–O–Al linkages explains most of the observeddiscrepancy between the theoretical and experimental 27AlMAS NMR spectra (Fig. 1a).

The theoretical mean values of CQ and diso NMRparameters for Al(Q4)[jSi,(4 � j)Al] species are reportedin Table 4. In all investigated glasses, 27Al diso decreases

by about 5 ppm for each (SiO4/2) tetrahedra linked to(AlO4/2)�. It ranges from 61.6 ppm for Al(Q4)[3Si,1Al] to57.7 ppm for Al(Q4)[4Si] in the NAS model, from70.4 ppm for Al(Q4)[2Si,2Al] to 58.7 ppm for Al(Q4)[4Si]in the CAS model, and from 72.0 ppm for Al(Q4)[1Si,3Al]to 62.9 ppm for Al(Q4)[4Si] in the CASN model. This trendnicely agrees with the recent measurements and GIPAWcomputations of crystalline gehlenite Ca2Al2SiO7 (Florianet al., 2012). In contrast, 27Al CQ decreases with the numberof Si atoms in the second coordination sphere.

Even though the values of NMR parameters provided inTable 4 clearly show that different Al(Q4)[jSi,4 � jAl] spe-cies exhibit particular spectral fingerprints, they cannot betaken into account as constraints for the fitting proceduresof the experimental spectra. In fact, the populations of alu-minium atoms in our merged models (90 in CAS, 51 inCASN, 51 in NAS) is not statistically representative of alu-minium population in real samples of glasses having thesame overall composition. Table 4 also reports randompopulations of Al(Q4)[jSi,4 � jAl] species. These coincidewith the population of Al(Q4)[jSi,4 � jAl] obtained by ran-domly distributing network-former cations around alumin-ium tetrahedra. The procedure for calculating randompopulations is provided in the Supplementary material, to-gether with a detailed structural analysis of the Al environ-ment (in terms of first and second coordination spheres),and the corresponding NMR parameters (Table S.2). Inall MD-derived models, the presence of species associatedwith Al clustering (and formation of Al–O–Al bridges), likeAl(Q4)[4Al], Al(Q4)[1Si,3Al] and Al(Q4)[2Si,2Al], is gener-ally underestimated with respect to a randomly calculatedpopulation. MD-derived populations of (AlO4/2)� tetrahe-dra linked to 3 or 4 (SiO4/2) units are lower than randompopulations in CAS and NAS models, and higher than pop-ulations in CASN model. It is then possible to deduce thatfor aluminosilicate glasses, aluminium clustering is lowerthan expected for a random network chemical disorder.Nonetheless, Al–O–Al bonds have been found in amor-phous aluminosilicate. The degree of aluminium avoidancedepends on the composition, as demonstrated in severalprevious works (McMillan et al., 1982; Murdoch andStebbins, 1985; Tossell and Saghi-Szabo, 1997; Lee andStebbins, 1999, 2000; Stebbins et al., 1999). Moreover, thepresence of Ca with respect to Na, which has lower fieldstrength, favours the presence of Al–O–Al bonds (Pedoneet al., 2012c).

3.3. Aluminium second coordination sphere: non-framework

disorder

The investigation of the second coordination sphere ofAl can be also developed in terms of the neighbouring mod-ifier cations, with the aim to shed light on the respectiverole of Ca and Na where they are both present. This aspectcan be quantified for CASN glass by the MD-derived ratio(Eq. (2)) between the Al(Q4)–Ca coordination number(CNCa

Al ) and the Al(Q4)–Na coordination number (CNNaAl ),

normalized with respect to the ratio of number of Ca(NCa) and Na (NNa) in the model (Tilocca et al., 2007;Pedone, 2009).

Table 4Theoretical and experimental values of 27Al NMR parameters in NAS, CAS and CASN glasses of Al(Q4)[jSi.(4 � j)Al] units (j silicon atomsand 4 � j aluminium atoms in the second coordination sphere). MD-derived model and random percentage populations are also reported.

j Pop% MD diso (ppm) CQ (MHz) gQ (MHz) Pop % random

NASa

0 0.0 – – – 0.21 0.0 – – – 3.32 0.0 – – – 17.73 23.5 61.7(4.5) 7.5(2.5) 0.7 53.54 62.7 57.8(6.5) 6.0(1.9) 0.6 37.0

CASa

0 0.0 – – – 2.61 0.0 – – – 15.32 7.8 70.5(5.3) 9.7(3.4) 0.7 34.53 23.3 63.5(5.0) 8.7(2.7) 0.7 34.54 21.1 58.8(5.7) 7.8(2.8) 0.6 13.0

CASN

0 0.0 – – – 0.41 3.9 72.0(4.6) 10.3(2.3) 0.7 4.72 11.8 70.6(2.0) 6.1(2.0) 0.6 21.13 27.4 67.2(4.7) 7.2(1.6) 0.7 42.24 56.9 62.9(5.7) 6.6(1.6) 0.6 31.6

a Population sum is not 100% because Al connected to TBO has been neglected from the analysis and only Al(Q4) corresponding to Al(BO)4

has been considered.

176 E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185

RCa=NaAl ¼ CNCa

Al

CNNaAl

�NNa

NCa

ð2Þ

If RCa=NaAl is greater than 1, this indicates that Ca is pref-

erably located around Al(Q4) with respect to Na; if it issmaller than 1, the preferential charge compensator is Na.The obtained value of 0.74 demonstrates that Na tends tooccupy Al surroundings in Ca–Na aluminosilicate glasses;thus, it preferably behaves as the charge compensator oftetrahedral (AlO4/2)� .

3.4. Silicon: local environment

The theoretical and experimental 29Si MAS NMR spec-tra are compared in Fig. 3. Whereas the experimental peaksroughly show a Gaussian distribution shape, the theoreticalones exhibit more accentuated asymmetries, which might bedue to the limited number of silicon atoms used in the sim-ulation. In fact, ab initio calculations have huge computa-tional costs, and in order to keep acceptable calculationtime, it is necessary to perform DFT calculations on sys-tems constituted of no more than a few hundred atoms.Nonetheless, a good agreement in the overall peak shapeand position is observed. The MD-generated structuresare mainly characterized by four-coordinated silicon. Thevery small amount of fivefold coordinated silicon (1 sitein CASN at diso = �134.5 ppm and 1 site in CSN glassesat diso = �133.2 ppm) may be ascribed to the excessivelyhigh pressure and rapid cooling rate experienced by theatoms in the simulation boxes during the MD generationof glasses at a constant volume. Indeed, exiguous popula-tions of fivefold Si atoms (about 0.05%) have been charac-terized in silica-based glasses cooled both at high and 1 barpressures (Stebbins et al., 1992).

The theoretical populations of four coordinated Si(Qn)species present in the simulated structures are reported inTable 5. NAS glass exhibits the most polymerized network,where almost all (98.3%) silicon atoms belong to Q4 units,whereas the remaining 1.7% is in Q3 units. In the CAS glassSi atoms are organized in Q4, Q3 and Q2 units in the respec-tive percentages of 95.7%, 2.1% and 1.4%. Although NASand CAS glasses have a tectosilicate composition, i.e.(CaO + Na2O)/Al2O3 = 1, the Al atoms are not fullycharge-compensated by modifier cations and a smallfraction of TBO is present, resulting in a not completelypolymerized network. Table 5 also reports populationsand NMR parameters of the different SiO4 units as a func-tion of the number of BO, NBO and TBO in the first coor-dination sphere. Our calculations show that Si atomsbonded to TBO are not detectable on MAS spectra. Thispredicted structural arrangement is in agreement withexperimental evidence that denies the validity of compen-sated continuous random network for tectosilicate glasses(Stebbins and Xu, 1997; Toplis et al., 1997). Networks ofthe CASN and CSN MD model structures are less polymer-ized: in the former, the majority of silicon atoms are presentas Q4 (42.0%) and Q3 (47.3%) species, and Q2 units consti-tute the remaining 10%, whilst in the latter, also Q1 speciesare present (3.1%) in addition to Q4 (8.6%), Q3 (52.5%) andQ2 (35.2%) ones.

The theoretical NMR parameters (Table 5) have beenused to guide the fitting of the experimental spectra, whichhas allowed us to quantify Si(Qn) species populations. ForCAS and CASN glasses, MD and fitted Qn populationsare in good agreement, whereas CSN and NAS populationsdiffer up to 14.8%, which might be considered an acceptablevalue for 29Si MAS spectra. The theoretical and fitted diso

values range from about �100 to �70 ppm, consistent with

Fig. 3. Comparison between experimental and theoretical (blacksolid lines) 29Si MAS NMR spectra (normalized to the samemaximum height) of NAS (a), CAS (b) and CASN (c) and CSN (d)glasses. The individual contributions of Qn species to the totaltheoretical spectra are reported in colored solid lines. (Forinterpretation of the references to colour in this figure legend, thereader is referred to the web version of this article.)

E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185 177

previous experimental observations on glasses and crystals(Murdoch and Stebbins, 1985; Maekawa et al., 1991; Joneset al., 2001), and increase from Q4 to Q1 with an incrementbetween 3 and 10 ppm for each substitution of one BO forone NBO.

Another important structural parameter affecting the29Si diso is the average Si–O–T bond angle of the Qn species.Table 5 shows that a decrease of 0.1� to 9.7� in the <Si–O–T>angle (T@Si or Al) is observed in all the structural modelswith the number of NBO species connected to silicon. Q2

structural units of the CAS glass present an exception tothis rule: the <Si–O–T> angle increases from 135.3� forQ3 to 143.2� for Q2. In this case, the poor representativenessof the Q2 population (1.4% of total Si atoms) might be

invoked. Also Q3 species in NAS glass have a very smallpopulation (1.6%) and are not considered for this discus-sion. A general explanation for this behaviour is found inthe number of NBOs on a Qn unit; the NBOs species attractmodifier cations (Na+, Ca2+) in their surroundings, whichinduce a compression of the inter-tetrahedral angles pro-portional to the modifier field strength (Cormier et al.,2003). Summarizing, moving from Q4 to Q1 both an in-crease in isotropic chemical shift values and a decrease in<Si–O–T> angles is observed. This behaviour is consistentwith well-known correlations between <Si–O–Si>/<Al–O–Si>angles and 29Si/27Al diso, found in many silicate and alumi-nosilicate glasses (Lippmaa et al., 1986; Phillips and Kirk-patrick, 1994; Mauri et al., 2000; Angeli et al., 2011).

3.5. Silicon second coordination sphere: framework disorder

The quantification of Qn species in silicate, phosphosili-cate or aluminosilicate glass samples has often been ob-tained, as in this work (Table 5), by assuming a Gaussiandistribution contribution of each Qn species to the totalspectra of 29Si (Merzbacher et al., 1990; Maekawa et al.,1991; Pedone et al., 2010a). However, this assumption hasbeen debated (Mahler and Sebald, 1995) and a newapproach that uses of 2D techniques (Zhang et al., 1996;Florian et al., 2009; Davis et al., 2010) has been developed.In fact, because of the deshielding effect of Al on Si that isdiscussed further below, a single Gaussian function cannotreproduce the band shape of a Si(Qn) species (see Fig. 3).The interpretation of a 29Si NMR spectrum in terms ofSi(Q4)[mAl] species has already been elaborated in the pastafter extensive experimental studies on crystalline alumino-silicate (Engelhardt and Michel, 1988) and then extended togeopolymeric (Duxson et al., 2005) and glassy phases (Leeand Stebbins, 1999; Hiet et al., 2009; Moesgaard et al.,2010). Hiet et al. (2009) succeeded in their direct quantifica-tion for a CAS-type glass (25%CaO�25%Al2O8�50%SiO2)with the help of very promising advanced NMR techniques.The populations of the various Si(Q4)[mAl] species theydetermined are nevertheless very different from those pre-dicted by our CAS structural model (Table 6), even if diso

variations are within 6 ppm, which is the standard deviationof our theoretical values.

Our MD–GIPAW data offer the opportunity to investi-gate theoretically network former intermixing in glasses andto provide a glass-tailored investigation of NMR parame-ters, for Q3 structural units. Q2 and Q1 species are omittedfrom the investigation because of their small populations inthe structures. Fig. 4 shows computed contributions ofSi(Qn)[mAl] to theoretical spectra of Qn species for NAS,CAS and CASN glasses. As expected, the most populatedsites show signals whose shapes are more similar to thatof a Gaussian. A trivial justification for the deviation fromGaussian shapes of the computed contribution with respectto the one reputedly obtained by a real sample is the smallamount of species considered in the calculation. For exam-ple, Si(Q4)[2Al] species in CAS glass amounts to barely 42atoms, and the most populated species among the three alu-minosilicate glasses, Si(Q4)[1Al] in NAS glass, counts 88atoms.

Table 5Theoretical and experimental 29Si NMR parameters in SiO4 units for NAS, CAS, CASN and CSN glasses.

Qn speciation Connectivity speciation Pop % diso (ppm) Si(Qn)–O–T

MD Exp. Th. Exp(dexpm )a Th.

NAS

Si 100.0 100.0 �101.1(8.0) �98.7 142.5(6.9)Q4 4BO 97.2 89.9% �101.4 (7.9) �99.6 (6.6) 142.8(6.7)

3BO:1TBO 1.1 �92.9 (10)Q3 3BO:1NBO 1.7 10.1% �89.8 (5.4) �90.9(5.6) 130.7(5.0)

CAS

Si 100.0 100.0 �92.5(7.6) �91.9 137.1(7.6)Q4 4BO 71.6 93.6% �93.8 (7.5) �92.2 (8.0) 137.7(12.7)

3BO:1TBO 9.2 �89.2 (7.5)2BO:2TBO 14.9 �81.3

Q3 3BO:1NBO 0.7 4.9% �90.4 (6.7) �89.8 (3.9) 135.3(13.2)2NBO:1TBO:1NBO 1.4 �84.2 (9.8)

Q2 2BO:2NBO 1.4 1.5% �84.1 (7.1) �83.1 (1.0) 143.2(13.6)

CASNb

Si 100.0 100.0 �89.9(8.3) �91.2 137.0(6.9)Q4 4BO 42.0 44.7% �93.9 (7.2) �96.6 (6.4) 137.7(12.7)Q3 3BO:1NBO 47.3 44.7% �88.0 (6.4) �87.1 (5.9) 137.3(13.5)Q2 2BO:2NBO 10.0 53.7% �79.1 (6.4) �82.7 (1.7) 132.8(11.7)

CSNb

Si 100.0 100.0 �88.7(9.0) �87.1 138.8(6.4)Q4 4BO 8.6 4.0% �102.2 (6.8) �102.2 (5.9) 141.5(13.0)Q3 3BO:1NBO 52.5 46.0% �91.5 (4.9) �90.4 (4.6) 138.6(12.1)Q2 2BO:2NBO 35.2 50.0% �82.0 (3.9) �77.3 (5.0) 138.5(11.9)Q1 1BO:3NBO 3.1 0.0% �69.7 (4.7) – 128.8(3.2)

a (dexpcg is the center of gravity of the experimental 29Si spectrum).

b CASN and CSN glasses contain one SiO5 unit with diso = �134 ppm, Dcs = 61.2 ppm, gcs = 0.9 and diso = �133 ppm, withDcs = 109.7 ppm, gcs = 0.7, respectively.

Table 6Theoretical NMR parameters (mean values and in parentheses standard deviation values) of Si(Qn)[mAl] species, where n = 3, 4, surroundedby m aluminium atoms in the second coordination sphere (m = 0–4). Si bonded to TBO (2 sites in NAS, 15 sites in CAS) and fivefoldcoordinated Si (1 site in CASN) have been neglected from the analysis.

M Q4 Pop %a Q4 diso(ppm) Q4 |DCS|(ppm) Q3 Pop% Q3 diso(ppm) Q3 |DCS| (ppm)

NAS

0 23.5 �105.8(6.2) 14.9(8.0) 1.6 �89.8(5.4) 86.3(11.2)1 48.1 �102.7(6.6) 36.0(10.0) 0.0 – –2 22.4 �95.3(6.8) 40.0(10.4) 0.0 – –3 2.2 �84.9(4.5) 35.5(6.2) 0.0 – –4 0.0 – – 0.0 – –

CAS

0 2.1 �99.3(4.6) 18.5(2.1) 2.1 �93.1(4.6) 66.9(5.7)1 17.7 �99.6(7.6) 30.8(10.4) 7.8 �92.4(6.7) 66.6(16.1)2 33.3 �93.2(6.5) 25.0(14.7) 3.5 �88.7(5.0) 58.2(15.5)3 17.7 �89.1(4.8) 28.5(11.3) 1.4 �79.9(1.3) 31.2(3.4)4 0.8 79.2 22.4 0.0 – –

CASN

0 5.3 �100,4(7.0) 19.4(5.8) 13.3 �93.3(5.6) 64.2(15.8)1 15.3 �95.9(6.9) 27.8(9.9) 22.0 �88.3(5.1) 58.6(15.3)2 18.0 �92.0(5.7) 38.5(9.8) 10.0 �82.7(2.6) 52.7(14.6)3 3.3 �84.7(3.4) 28.9(12.8) 2.0 �76.9(1.8) 43.6(23.4)4 0.0 – – 0.0 – –

a Population referred to total amount of Si per structural model.

178 E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185

Fig. 4. Computed contribution of Si(Qn)[mAl] (n is 3 or 4; m from 0 to 4) to Qn species theoretical spectra for all aluminosilicate glasses NAS(a,d) CAS (b,e) and CASN (c, f); Q2 species are not considered because of non-representative populations.

Table 7Quantification of chemical disorder of NAS network: fittingexperimental spectra with Si(Qn)[mAl] (n = 3,4; m = 0–3) speciesconstraints.

Species Pop %, EXPa diso, EXPa (ppm)

Si 100.0 103.0(5.9)Si(Q4)[0Al] 68.8 �105.6(5.9)Si(Q4)[1Al] 21.3 �102.8(5.9)Si(Q4)[2Al] 0.0 –Si(Q4)[3Al] 9.9 �85.8(5.9)Si(Q3)[0Al] 0.0 –

a The errorP

i yexp;i � ysim;i

� �2=P

iy2exp;i associated with the

fitting procedure is 0.27%.

E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185 179

Focusing on the more populated Si(Qn)[mAl] species (atleast 5% of total Si atoms), the NMR parameters listed inTable 6, diso, and chemical shift anisotropy expressed inabsolute value, |DCS|, show interesting trends. They both in-crease with the number of aluminium cations in the siliconsecond coordination sphere and the extent of increase variesfrom +2.9 to +6.6 ppm per aluminium atom in the case ofdiso, and from +1.3 to +8.8 ppm per aluminium atom in thecase of |DCS|. The results obtained are in agreement with theempirical assumption that diso increases almost constantlyby +5 ppm per Al as proposed from previous experimentalwork (Engelhardt and Michel, 1988; Lee and Stebbins,1999; Moesgaard et al., 2010). The results also reveal a cer-tain sensitivity of |DCS| to network chemical disorder andprovide a further instrument for accurately quantifying net-work chemical disorder. Indeed, a deconvolution of NMRspectra using specific constraints, like those provided inTable 6, will help to quantify the species that representnetwork disorder.

This has been done for the NAS glass, where reliableconstraints are available for the Si(Qn)[mAl] species. Table 7reports the relative populations and isotropic chemicalshifts extracted from the fitting of the spectra. To fit Sili-con-29 MAS experimental spectra (Fig 3a), the theoreticaldiso values reported in Table 6 have been used as constraints(for Si(Q3)[2Al] and Si(Q4)[0/1/2/3Al] species). In this case,the error is 0.27% and is lower with respect to that (0.37%)obtained by fitting the spectra considering the presence ofQ4 and Q3 species and without decomposing the Qn speciesin the different Si(Qn)[mAl] species. The obtained isotropicchemical shifts are in excellent agreement with those com-puted at the MD–GIPAW level reported in Table 6,whereas the very high percentage of Si(Q4)[0Al] species(68.8%) and the low amount of Si(Q4)[2Al] and Si(Q4)[3Al]species found in our structural model demonstrate that the

MD simulations overestimate the clustering of aluminiumaround silicon. Finally, the Si(Q4)[mAl] experimental popu-lations sum to 100% of the Si(Qn) units; this is much morereasonable than the results obtained by fitting the spectrawith only the 2 constraints of theoretical 29Si Q3 and Q4

diso, as reported in Table 5.Another interesting point on network cation intermixing

is to verify whether there is a particular Si(Qn) unit thatexhibits a preference to build Si–O–Al bridges. To shedlight on this, for each Qn species, a count of Si(Qn)–O–Albridges has been carried out. The number has been dividedby the total number of Si(Qn)–O–T bridges and then nor-malized with respect to the CAS Al/Si ratio, i.e.: multipliedby the (Al/Si)glass/(Al/Si)CAS ratio. The populations of Al–O–Si, Si–O–Si and Al–O–Al bridges are reported inTable S.3 of the Supplementary material, whereas theresults of the normalized Si(Qn)–O–T bridges are reportedin Table 8. From this a preference for Al atoms to formSi–O–Al bridges with Si(Q4) species is observed.

Table 8Ratio of Si(Qn)–O–Al (n = 2, 3, 4) bridges with respect to totalSi(Qn)–O–T bridges and normalized Al/Si ratio in CAS glass.

SiðQnÞ �O�Al

SiðQnÞ �O� T�Al=Siglass

Al=SiCAS

NAS CAS CASN

Si(Q2) –* – 0.449Si(Q3) 0.000 0.486 0.632Si(Q4) 0. 594 0.701 0.683

* The respective Si(Qn) species is not present in the structuralmodel of the considered glass (NAS) or its population is notsignificant, i.e.: less than 5% of total Si atoms (CAS).

180 E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185

3.6. Silicon second coordination sphere: non-framework

disorder

Further information about the silicon environment canbe obtained by analysing the presence of calcium and/or so-dium in its second coordination sphere, similarly to theinvestigation performed for Al, and then evaluating theireffect on the 29Si NMR parameters. The Si coordinationnumber with respect to each modifier, M that is Ca orNa, and (CNM

Si ) and the ratio between both modifiers,(CNCa=Na

Si ), has been quantified (Table 9). The calcium andsodium concentration around silicon varies with Qn specia-tion: CNCa

Si decreases by �1.2–1.3 units in CAS and CASNand by �1.0 in CSN when increasing the number of BO inthe (SiO4/2) unit, whereas CNNa

Si decreases by �0.9 units inCASN glass and �1.8 units in NAS glass. On the otherhand, CNNa

Si in CSN glass has an almost constant value ofabout 3.8–4. In both CASN and CSN glasses, total siliconexhibits no preference for Ca or Na, as RNa=Ca

Si �1, whereasan in-depth examination of the Qn species reveals that cal-cium cations are mainly located near Q1 and Q2 species, andsodium is more localized around Q3 and Q4 units. Theseoutcomes can be explained by considering that Ca, whosefield strength is higher than that of Na, prefers electron richenvironments, whereas Na prefers to lie close to the mostpolymerized regions of the network.

The effect of the presence of modifier cations (Ca and/orNa) around silicon atoms on 29Si NMR parameters has alsobeen investigated for all glasses. The second coordinationshell of silicon expressed in terms of modifiers defines theSi[kM] and Si(Qn)[kM] species, where k is the total amountof modifier cations (NCa + NNa atoms) and M isindistinguishable Ca or Na. Detailed populations of these

Table 9Average coordination number, CNM

Si , of silicon atoms and Qn species, wit(M = Ca and/or Na), and the normalized ratio of CNCa

Si and CNNaSi with

CNCaSi CN

Glass CAS CASN CSN NA

Si total 1.77 1.93 1.90 1.6Si(Q4) 1.54 1.08 0.46 1.5Si(Q3) 2.71 2.28 1.51 3.3Si(Q2) 4.00 3.40 2.56 –Si(Q1) –* – 3.8 –

* The respective Si(Qn) species is not present in the structural model of t

species for all structural models are reported in Table S.4of the Supplementary material. Contrary to DCS and gCS,

the computed isotropic chemical shift, diso, shows interest-ing trends with respect to the presence of modifier cations.These are shown in Fig. 5 for total Si and its more popu-lated (>5.0%) Si(Qn) species. In aluminosilicate glassesNAS (Fig. 5a), CAS (Fig. 5b) and CASN (Fig. 5c) it is pos-sible to observe the deshielding effect of calcium and/or so-dium on silicon-29 nuclei for both total silicon and Si(Qn)species (Jones et al., 2001). Nevertheless, the Qn species inthe CSN glass do not follow this trend; therefore, the desh-ielding effect on total silicon is only apparent. It is actuallyascribable to the fact that, although diso for each Si(Qn)[kM]unit does not exhibit a dependence on k, Si(Qn)[kM] unitsvariably populate Si(Qn) sites. The deshielding effect ofCa and Na modifiers is then an “apparent effect”, relatedto the second coordination shell in terms of aluminium,rather than in terms of modifiers.

3.7. Relationships between 27Al and 29Si diso and inter-

tetrahedral angles

Accurate relationships between NMR parameters andstructural features are extremely useful for the interpreta-tion of experimental data, as they make a reverse approachpossible (Charpentier et al., 2009; Angeli et al., 2011; Clarket al., 2004; Mauri et al., 2000; Soleilhavoup et al., 2010). Inthis way, structural features of a glass sample could, in prin-ciple, be directly obtained from the experimental data dis-tribution. Two of the most investigated relationships arethose between 27Al and 29Si diso and inter-tetrahedral angles(Lippmaa et al., 1986; Engelhardt and Michel, 1988; Angeliet al., 2000; Mauri et al., 2000; Mackenzie and Smith, 2002;Charpentier et al., 2009). Figs. 6 and 7 display, for theglasses under study, the plots of 27Al and 29Si diso as a func-tion of the <Al–O–T> and <Si–O–T> angles, respectively.At first glance, the distribution of 27Al diso values does notcorrelate very well with the <Al–O–T> angle, and the good-ness-of-fit of the linear correlations seems to vary as a func-tion of glass composition. However, the study of the samelinear correlation, for Si and Al, taking into account theirconnectivity to different oxygen species (BOs, NBOs,TBOs), leads to reasonable correlation coefficients and theslope and intercept values are rather independent on chem-ical composition of the glasses. It is worth noting that thoseglasses containing calcium, i.e.: CAS, CASN and CSN,

h regards to modifier cations, M, in the second coordination sphererespect to the composition.

NaSi R

Ca=NaSi

S CASN CSN CASN CSN

1 1.91 3.92 1.01 0.978 1.79 4.07 0.60 0.238 1.97 3.82 1.16 0.79

2.06 3.98 1.65 1.29– 4.4 – 1.73

he considered glass.

Fig. 5. Correlations between the isotropic chemical shift, diso, and the amount of modifier cations, k = n�M = (n�Ca + n�Na), in the siliconsecond coordination sphere (black circles). The correlations obtained by means of the contribution of modifier cations, in the secondcoordination sphere of each Si(Qn) species are also reported (colored dots). The population of Qn species, in the MD-derived structuralmodels, is reported in each panel. (For interpretation of the references to colour in this figure legend, the reader is referred to the web versionof this article.)

Fig. 6. Plot of 27Al diso vs <Al–O–T> reported for different connectivity environments of Al in NAS, CAS and CASN. Al(BO)4 are red circles,Al(BO)3(TBO)1 are green squares and Al(BO)2(TBO)2 are blue triangles. Linear regression fitting lines are also reported with coherent colorsas visual guides. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185 181

showed the poorest correlations, as previously observedexperimentally (Florian et al., 2009) and theoretically (Ped-one et al., 2010b) on CaO–SiO2 phases and glasses.

As stated in the previous section, Ca2+ in mixed Ca–Naglasses, like CASN and CSN, is not uniformly allocatedaround the different Si(Qn) species, but its concentrationis maximal around Si(Q2) and minimal around Si(Q4). Itis then reasonable to expect that 29Si diso differently corre-

lates with <Si–O–T> depending on the considered Qn site.Fig. 7 shows these correlations and, together with the anal-ysis of regression coefficients, reveals a decrease in the linearcorrelations with n: from 0.72 for Si(Q4) to 0.60 for Si(Q3)in CASN glass (Fig. 7b), and from 0.96 for Si(Q4) to 0.52for Si(Q2) in CSN glass (Fig. 7c). Analogous behaviourhad already been observed in bioglasses (Pedone et al.,2010a). It is also worth noting that correlation coefficients

Fig. 7. Plot of 29Si diso vs <Si–O–T> reported for different connectivity environments of Si in NAS, CAS, CASN and CSN. Si(BO)4 are greendots, Si(BO)3(NBO)1 are red dots, Si(BO)2(NBO)2 are blue dots and Si(BO)3(TBO)1 are black circles. Linear regression fitting lines are alsoreported with coherent colors as visual guides. (For interpretation of the references to colour in this figure legend, the reader is referred to theweb version of this article.)

182 E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185

in CASN glass are lower than those in CSN glass, suggest-ing that aluminium, as well as calcium, may contributeto weakening of the correlations between 29Si diso and<Si–O–T> inter-tetrahedral angles.

4. CONCLUSIONS

The structures of a silicate, CSN, and three aluminosil-icate glasses, CAS, NAS, CASN, have been investigatedusing a simultaneous experimental–computational ap-proach that considers the employment of MD-derivedstructural models to support the interpretation of experi-mental solid state nuclear magnetic resonance data.

The agreement between computed and experimental 29Siand 27Al NMR parameters and MAS spectra is good andproves the reliability of the computational method for theinvestigation of amorphous system structures. Few discrep-ancies, such as the 27Al CQ and MAS spectra in NASglass, are interpreted as consequences of simulation andcomputational conditions and may inspire further effortsto improve the calculation of 27Al NMR parameters whenNa is present and plays the role of a charge compensatorcation.

The aluminium environment for the compositions stud-ied is fully characterized by Q4 species; thus its role is thatof a pure network former. Structural analysis of the modelsin terms of Al(Q4)[jSi], where j is the amount of (SiO4/2) tet-rahedra linked to (AlO4/2)� units, reveals that the amount offramework cation intermixing into the network is higherthan predicted by randomly distributed Al and Si, thoughLoewenstein’s rule (Loewenstein, 1954) is still not respected.

Both 29Si and 27Al diso are affected by the nature of theirsecond coordination sphere in terms of former cations,whilst an effect from modifier cations, Ca2+ and Na+, is

not evident. In general, aluminium atoms of the secondcoordination sphere deshield both 27Al and 29Si nuclei. Thiseffect should be taken into account for fitting NMR spectrawith the aim of quantifying network chemical disorder andQn species. The quantification of the former is possible byfitting 29Si MAS spectra by using theoretical diso range val-ues obtained from sufficiently populated Si(Qn)[mAl](n = 1–4, m = 0–4) species. The employment of theseconstraints also allows to obtain lower fitting errors withrespect to the fitting of the structures with Si(Qn) species.Moreover, the quantification of Qn species that reflect net-work polymerization gives better results when this fittingprocedure is followed, as demonstrated by the case ofNAS glass. A future development of this work could in-volve the building of larger structural models for CAS,CASN and CASN with the aim to obtain reliable con-straints for Si(Qn)[mAl] in these systems.

Simple and effective correlations between 29Si and 27Aldiso and inter-tetrahedral angle can be found for these mul-ticomponent systems if the connectivity of former cations todifferent oxygen species (BOs, NBOs, TBOs) is properly ta-ken into account. Nonetheless, it is recognized that thepresence of calcium cations tends to critically compromisethe correlations. This effect is evident for different composi-tions with different percentages of CaO, but also for Si(Qn)species surrounded by different numbers of Ca2+. An anal-ogous but weaker effect is observed around Al cations.Then, in order to obtain very accurate correlation betweeninter-tetrahedral angles and the isotropic chemical shift ofnetwork former cations, it would be appropriate to elabo-rate multivariate correlations that include, beyond the in-ter-tetrahedral angle, structural features concerning Caand Al positions with respect to the investigated activenucleus.

E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185 183

ACKNOWLEDGMENTS

The authors thank the Italian Ministry of University andResearch for funding (Project COFIN2008, prot. 2008J9RNB3“Integrazione Temporale per l’Evoluzione Molecolare”). Thiswork was partially granted access to HPC resources of CCRTunder the allocation 2012-t2012096303 made by GENCI (GrandEquipement National de Calcul Intensif). The authors also thankAlan Wong for reviewing this paper before submission.

APPENDIX A. SUPPLEMENTARY DATA

Supplementary data associated with this article can befound, in the online version, at http://dx.doi.org/10.1016/j.gca.2013.10.025.

REFERENCES

Allwardt J. R., Lee S. K. and Stebbins J. F. (2003) Bondingpreferences of non-bridging O atoms: evidence from O-17 MASand 3QMAS NMR on calcium aluminate and low-silica Ca–aluminosilicate glasses. Am. Mineral 88, 949–954.

Angeli F., Delaye J. M., Charpentier T., Petit J. C., Ghaleb D. andFaucon P. (2000) Investigation of Al–O–Si bond angle in glassby 27Al 3Q-MAS NMR and molecular dynamics. Chem. Phys.

Lett. 320, 681–687.Angeli F., Gaillard M., Jollivet P. and Charpentier T. (2007)

Contribution of 43Ca MAS NMR for probing the structuralconfiguration of calcium in glass. Chem. Phys. Lett. 440, 324–328.

Angeli F., Villain O., Schuller S., Ispas S. and Charpentier T.(2011) Insight into sodium silicate glass structural organizationby multinuclear NMR combined with first-principles calcula-tions. Geochim. Cosmochim. Acta 75, 2453–2469.

Ashbrook S. E. (2009) Recent advances in solid-state NMRspectroscopy of quadrupolar nuclei. Phys. Chem. Chem. Phys.

11, 6892–6905.Ashbrook S. E., Berry A. J. and Wimperis S. (2001) O-17

multiquantum MAS NMR study of high-pressure hydrousmagnesium silicate. J. Am. Chem. Soc. 123, 6360–6366.

Bansal N. P. and Doremus R. H. (1986) Handbook of Glass

Properties. Academic Press, Inc., Orlando, FL.Blochl P. E. (1994) Projector augmented-wave method. Phys. Rev.

B 50, 17953–17979.Charpentier T., Kroll P. and Mauri F. (2009) First-principles

nuclear magnetic resonance structural analysis of vitreous silica.J. Phys. Chem. C 113, 7917.

Charpentier T., Menziani M. C. and Pedone A. (2013) Computa-tional simulations of solid state NMR spectra: a new era instructure determination of oxide glasses. RSC Adv. 3, 10550–10578.

Choi M., Matsunaga K., Oba F. and Tanaka I. (2009) 27Al NMRchemical shifts in oxide crystals: a first-principles study. J. Phys.

Chem. C 113, 3869–3873.Clark T. and Grandinetti P. J. (2003) Dependence of bridging

oxygen 17O quadrupolar coupling parameters on Si–O distanceand Si–O–Si angle. J. Phys.: Condens. Matter 15, S2387–S2395.

Clark T. M., Grandinetti P. J., Florian P. and Stebbins J. F. (2004)Correlated structural distributions in silica glass. Phys. Rev. B

70, 202.Cormier L. and Neuville D. R. (2004) Ca and Na environments in

Na2O–CaO–Al2O3–SiO2 glasses: influence of cation mixing andcation–network interactions. Chem. Geol. 213, 103–113.

Cormier L., Ghaleb D., Neuville D. R., Delaye J. M. and Calas G.(2003) Chemical dependence of network topology of calciumaluminosilicate glasses: a computer simulation study. J. Non-

Cryst. Solids 332, 255–270.Davis M. C., Kaseman D. C., Parvani S. M., Sanders K. J.,

Grandinetti P. J., Massiot D. and Florian P. (2010) Q(n) speciesdistribution in K2O�2SiO2 glass by 29Si magic angle flippingNMR. J. Phys. Chem. A 114, 5503–5508.

Deschamps M., Fayon F., Hiet J., Ferru G., Derieppe M., PellerinN. and Massiot D. (2008) Spin-counting NMR experiments forthe spectral editing of structural motifs in solids. Phys. Chem.

Chem. Phys. 10, 1298–1303.Dick G. and Overhauser A. W. (1958) Theory of the dielectric

constant of alkali halide crystals. Phys. Rev. 112, 90–103.Duxson P., Provis J. L., Lukey G. C., Separovic F. and van

Deventer J. S. J. (2005) 29Si NMR study of structural orderingin aluminosilicate geopolymer gels. Langmuir 21, 3028–3036.

Eden M. (2012) NMR studies of oxide-based glasses. Annu. Rep.

Prog. Chem. Sect. C 108, 177–221.Engelhardt G. and Michel D. (1988) High-Resolution Solid-State

Nmr of Silicates and Zeolites. John Wiley & Sons Inc..Farnan I., Grandinetti P. J., Baltisberger J. H., Stebbins J. F.,

Werner U., Eastman M. A. and Pines A. (1992). Nature 358, 31.Florian P., Fayon F. and Massiot D. (2009) 2J Si�O�Si scalar

spin–spin coupling in the solid state: crystalline and glassywollastonite CaSiO3. J. Phys. Chem. C 113, 2562–2572.

Florian P., Veron E., Green T. F. G., Yates J. R. and Massiot D.(2012) Elucidation of the Al/Si ordering in gehlenite Ca2Al2-

SiO7 by combined 29Si and 27Al NMR spectroscopy/quantumchemical calculations. Chem. Mater. 24, 4068–4079.

Hanna J. V. and Smith M. E. (2010) Recent technique develop-ments and applications of solid state NMR in characterisinginorganic materials. Solid State NMR 38, 1–18.

Henderson G. S., Neuville D. R. and Cormier L. (2009) An O K-edge XANES study of glasses and crystals in the CaO–Al2O3–SiO2 (CAS) system. Chem. Geol. 259, 54–62.

Hiet J., Deschampes M., Pellegrin N., Fayon F. and Massiot D.(2009) Probing chemical disorder in glasses using silicon-29NMR spectral editing. Phys. Chem. Chem. Phys. 11,6935–6940.

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

Iuga D., Morais C., Gan Z., Neuville D. R., Cormier L. andMassiot D. (2005) NMR heteronuclear correlation betweenquadrupolar nucleai in solids. J. Am. Chem. Soc. 127, 11540–11541.

Jones A. R., Winter R., Greaves G. N. and Smith I. H. (2001) MASNMR study of soda-lime-silicate glasses with variable degree ofpolymerization. J. Non-Cryst. Solids 293–295, 87–92.

Kello V., Sadlej A. J., Pyykko P., Sundholm D. and Tokman M.(1999) Electric quadrupole moment of the 27Al nucleus:converging results from the AlF and AlCl molecules and theAl atom. Chem. Phys. Lett. 304, 414–422.

Koller H., Meijer E. L. and van Santen R. A. (1997) 27Alquadrupole interaction in zeolites loaded with probe molecules– a quantum-chemical study of trends in lectric field gradientsand chemical bond clusters. Solid State NMR 9, 165–175.

Kuryaeva R. G. (2006) Degree of polymerization of the CaAl2Si2-

O8 aluminosilicate glass. Glass Phys. Chem. 32, 505–510.Larsen F. H. and Farnan I. (2002) 29Si and 17O (Q)CPMG-MAS

solid-state NMR experiments as an optimum approach for half-integer nuclei having long T1 relaxation times. Chem. Phys.

Lett. 357, 403–408.

184 E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185

Lee S. K. (2010) Recent technique developments and applicationsof solid state NMR in characterising inorganic materials. Solid

State NMR 38, 45–57.Lee S. K. and Stebbins J. F. (1999) The degree of aluminium

avoindance in aluminosilicate glasses. Am. Mineral. 84, 937–945.

Lee S. K. and Stebbins J. F. (2000) Al–O–Al and Si–O–Si sites inframework aluminosilicate glasses with Al/Si = 1: quantifica-tion of framework disorder. J. Non-Cryst. Solids 270, 260–264.

Leko V. K. and Mazurin O. V. (2003) Analysis of regularities incomposition dependence of the viscosity for glass-formingoxide melts: II. Viscosity of ternary alkali aluminosilicate melts.Glass Phys. Chem. 29, 216–227.

Lippmaa E., Samoson A. and Magi M. (1986) High-resolution 27AlNMR of aluminosilicates. J. Am. Chem. Soc. 108, 1730–1735.

Loewenstein W. (1954) The distribution of aluminum in thetetrahedra of silicates and aluminates. Am. Miner. 39, 92–96.

Mackenzie K. and Smith M. E. (2002) Multinuclear Solid-state

NMR of Inorganic Materials. Pergamon.Maekawa H., Maekawa T., Kawamura K. and Yokokawa T.

(1991) The structural groups of alkali silicate glasses determinedfrom 29Si MAS-NMR. Solid State NMR 127, 53–64.

Mahler J. and Sebald A. (1995) Deconvolution of 29Si magic-anglespinning nuclear magnetic resonance spectra of silicate glassesrevisited – some critical comments. J. Non-Cryst. Solids 5.

Massiot D., Fayon F., Montouillout V., Pellerin N., Hiet J.,Roiland C., Florian P., Coutures J.-P., Cormier L. and NeuvilleD. R. (2008) Structure and dynamics of oxide melts and glasses:a view from multinuclear and high temperature NMR. J. Non-

Cryst. Solids 354, 249–254.Mauri F., Pasquarello A., Pfrommer B. G., Yoon Y. G. and Louie

S. G. (2000) Si–O–Si bond-angle distribution in vitreous silicafrom first-principles 29Si NMR analysis. Phys. Rev. B 62,R4786–R4789.

McMillan P. F., Piriou B. and Navrotsky A. (1982) A Ramanspectroscopy study of glasses along the joins silica.calciumaluminate, silica-sodium, aluminate, ans silica-potassium alu-minate. Geochim. Cosmochim. Acta 46, 2021–2037.

Merzbacher C. I., Sherriff B. L., Hartman J. S. and White W. B.(1990) A high-resolution 29Si and 27Al NMR study of alkalineearth aluminosilicate glasses. J. Non-Cryst. Solids 124, 194–206.

Moesgaard M., Keding R., Skibsted J. and Yue Y. (2010) Evidenceof intermediate-range order heterogeneity in calcium alumino-silicate glasses. Chem. Mater. 22, 4471–4483.

Murdoch J. B. and Stebbins J. F. (1985) High-resolution 29Si NMRstudy of silicate and aluminosilicate glasses: the effect ofnetwork-modifyng cation. Am. Mineral. 70, 332–343.

Neuville D. R. (2006) Viscosity, structure and mixing in (Ca, Na)silicate melts. Chem. Geol. 229, 28–41.

Neuville D. R. and Mysen B. O. (1996) Role of aluminium in thesilicate network: In situ, high-temperature study of glasses andmelts on the join SiO2–NaAlO2. Geochim. Cosmochim. Acta 60.

Neuville D. R., Cormier L., Montouillout V. and Massiot D.(2007) Local Al site distribution in aluminosilicate glasses by27Al MQMAS NMR. J. Non-Cryst. Solids 353, 180–184.

Pedone A. (2009) Properties calculations of silica-based glasses byatomistic simulations techniques: a review. J. Phys. Chem. C

113, 20773–20784.Pedone A., Charpentier T., Malavasi G. and Menziani M. C.

(2010a) New insights into the atomic structure of 45S5 bioglassby means of solid-state NMR spectroscopy and accurate first-principles simulations. Chem. Mater. 22, 5644–5652.

Pedone A., Charpentier T. and Menziani M. C. (2010b) Multinu-clear NMR of CaSiO3 glass: simulation from first-principles.Phys. Chem. Chem. Phys. 12, 6054–6066.

Pedone A., Charpentier T. and Menziani M. C. (2012a) Fluorineenvironment in bioactive glasses: new insights from first-principles calculations and solid State NMR spectroscopy. J.

Mater. Chem. 22, 12599–12608.Pedone A., Gambuzzi E., Malavasi G. and Menziani M. C. (2012b)

First-principles simulations of the 27Al and 17O Solid StateNMR spectra of a calcium alumino-silicate glass. Theor. Chem.

Acc. 131, 1147–1157.Pedone A., Gambuzzi E. and Menziani M. C. (2012c) Unambig-

uous description of the oxygen environment in multicomponentaluminosilicate glasses from 17O solid state NMR computa-tional spectroscopy. J. Phys. Chem. C 116, 14599–14609.

Perdew J. P., Burke K. and Ernzerhof M. (1996) Generalizedgradient approximation made simple. Phys. Rev. Lett. 77, 3865–3868.

Phillips B. L. and Kirkpatrick R. J. (1994) Short-range Si–Al orderin leucite and analcime: determination of the configurationalentropy from 27Al and variable-temperature 29Si NMR spec-troscopy of leucite, its Cs- and Rb-exchanged derivatives, andanalcime. Am. Mineral. 79, 1025–1031.

Pickard C. J. and Mauri F. (2001) All-electron magnetic responsewith pseudopotentials: NMR chemical shifts. Phys. Rev. B, 63.

Priven A. I. (2001) Calculation of the viscosity of glass-formingmelts: V. Binary borate systems. Glass Phys. Chem. 26, 541–558.

Profeta M., Benoit M., Mauri F. and Pickard C. J. (2004) First-principle calculation of the 17O NMR parameters in Ca oxideand Ca aluminosislicates: the partially covalent nature of theCa–O bond, a challenge dor density functional theory. J. Am.

Chem. Soc. 126, 12628–12635.Provis J. L., Duxson P., Lukey G. C. and van Deventer J. S. J.

(2005) Statistical thermodynamic model for Si/Al ordering inamorphous aluminosilicates. Chem. Mater. 17, 2976–2986.

Richet P., Nidaira A., Neuville D. R. and Atake T. (2009)Aluminium speciation, vibrational entropy and short-rangeorder in calcium aluminosilicate glasses. Geochim. Cosmochim.

Acta 73, 3894–3904.Rocquefelte X., Clabau F., Paris M., Deniard P., Le Mercier T.,

Jobic S. and Whangbo M. H. (2007) Resolving the aluminumordering in aluminosilicates by a combined experimental/theoretical study of 27Al electric field gradients. Inorg. Chem.

46, 5456–5458.Segall M. D., Lindan P. J. D., Probert M. J., Pickard C. J., Hasnip

P. J., Clarck S. J. and Payne M. C. (2002) First-principlessimulation: ideas, illustrations and the CASTEP code. J. Phys.:

Condens. Matter 14, 2717–2744.Smith W. and Forester T. R. (1996) DL_POLY_2.0: a general-

purpose parallel molecular dynamics simulation package. J.

Mol. Graphics 14, 136–141.Soleilhavoup A., Delaye J. M., Angeli F., Cuarant D. and

Charpentier T. (2010) Contribution of first-principles calcula-tions to multinuclear NMR analysis of borosilicate glasses.Magn. Res. Chem. 48, S159–S170.

Stebbins J. F. and Xu Z. (1997) NMR evidence for excess non-bridging oxygen in an aluminosilicate glass. Nature 390, 60–62.

Stebbins J. F., Farnan I. and Xue X. (1992) The structure anddynamics of alkali silicate liquids: a view from NMR spectros-copy. Chem. Geol. 96, 371–385.

Stebbins J. F., Lee S. K. and Oglesby J. V. (1999) Al–O–Al oxygensites in crystalline aluminosilicates and aluminosilicate glasses:high-resolution oxygen-17 NMR results. Am. Mineral. 84, 983–986.

Tilocca A., de Leeuw N. and Cormack A. N. (2006) Shell-modelmolecular dynamics calculations of modified silicate glasses.Phys. Rev. B 73, 104209–104223.

E. Gambuzzi et al. / Geochimica et Cosmochimica Acta 125 (2014) 170–185 185

Tilocca A., Cormack A. N. and de Leeuw N. H. (2007) Thestructure of bioactive silicate glasses: new insight from molec-ular dynamics simulations. Chem. Mater. 19, 95–103.

Toplis M. J., Dingwell D. B., Hess K. U. and Lenci T. (1997)Viscosity, fragility, and configurational entropy of melts alongthe join SiO2–NaAlSiO4. Am. Mineral. 82, 979–990.

Tossell J. A. and Saghi-Szabo G. (1997) Aluminosilicate andborosilicate single 4-rings: effect of counterions and water onstructure, stability and spectra. Geochim. Cosmochim. Acta 61,1171–1179.

Vollmayer K., Kob W. and Binder K. (1996) Cooling rate effects inamorphous silica: a computer simulation. Phys. Rev. B 54,15808–15827.

Xue X., Kanzaki M. and Neuville D. R. (2004) Structure andproperties of silicate melts and fluids. Geochim. Cosmochim.

Acta 61, 1171–1179.Yates J. R., Pickard C. J. and Mauri F. (2007) Calculation of

NMR chemical shifts for extended systems using ultrasoftpseudopotentials. Phys. Rev. B 76, 024401.

Zhang P., Dunlap C., Florian P., Grandinetti P. J., Farnan I. andStebbins J. F. (1996) Silicon site distribution in an alkali silicateglass derived two-dimensional 29Si nuclear magnetic resonance.J. Non-Cryst. Solids 204, 294–300.

Associate editor: Wolf Uwe Reimold