Solid-state nuclear magnetic resonance spectroscopy of cements

Cement is the ubiquitous material upon which modern civilisation is built, providing long-term strength, impermeability and durability for housing and infrastructure. The fundamental chemical interactions which control the structure and performance of cements have been the subject of intense research for decades, but the complex, crystallographically disordered nature of the key phases which form in hardened cements has raised dif ﬁ culty in obtaining detailed information about local structure, reaction mechanisms and kinetics. Solid-state nuclear magnetic resonance (SS NMR) spectroscopy can resolve key atomic structural details within these materials and has emerged as a crucial tool in characterising cement structure and properties. This review provides a comprehensive overview of the application of multinuclear SS NMR spectroscopy to understand composition e structure e property relationships in cements. This includes anhydrous and hydrated phases in Portland cement, calcium aluminate cements, calcium sulfoaluminate cements, magnesia-based cements, alkali-activated and geopolymer cements and synthetic model systems. Advanced and multidimensional experiments probe 1 H, 13 C, 17 O, 19 F, 23 Na, 25 Mg, 27 Al, 29 Si, 31 P, 33 S, 35 Cl, 39 K and 43 Ca nuclei, to study atomic structure, phase evolution, nanostructural development, reaction mechanisms and kinetics. Thus, the mechanisms controlling the physical properties of cements can now be resolved and understood at an unprecedented and essential level of detail. © 2019 The Authors. 1. C 3 AH 6 ( d iso ¼ 12.3 ppm, C Q ¼ 0.705 MHz) and amorphous AH 3 ( d obs ¼ 8 ppm at B 0 ¼ 9.4 T) phases in CAC hydrated for 28 days; the isotropic chemical shift for C 3 AH 6 was determined via simulation of the spinning sidebands associated with this phase. Analysis of synthetic CAH 10 at multiple ﬁ elds exhibited a broad resonance in the octahedral Al region of the 27 Al MAS NMR spectrum and allowed determination of d iso ¼ 10.2 ppm and 1.9 MHz (cid:3) C Q (cid:3) 2.6 MHz for this site. These results are consistent with previous data acquired at lower ﬁ eld [275]. Subsequent work examined CAC hydrate phases using 27 Al MAS [50,276] and MQMAS [276] NMR, enabling determination of d iso ¼ 10.3 ppm and C Q ¼ 1.2 MHz for the octahedral Al site in the metastable C 2 AH 8 phase and d iso ¼ 11.9 ppm and C Q ¼ 4.3 MHz for the amorphous AH 3 phase. investigated by static 35 Cl NMR experiments and measurement of 35 Cl spin e lattice relaxation time constants (T 1 ) and spin e spin relaxation time constants (T 2 ) [218], showing that chloride is in rapid exchange between adsorbed surface sites and bulk solution sites in each phase and exists predominantly in a hydrated, solution-like chemical environment on the monocarbonate surface. This behaviour is similar to chloride sorption onto portlandite and onto jennite, as discussed previously.


Introduction
Modern society is heavily reliant on cementitious materials in constructing our built environment. Second only to water in terms of commodity use [1,2], the worldwide production of concrete exceeds 10 billion tonnes per annum. The most common cement used to produce concrete is Portland cement (PC) [3], a complex material made up of multiple calcium-rich mineral phases that react with water and harden to provide strength, impermeability and resistance to thermal, mechanical and chemical stresses over long timescales [4]. The excellent physical properties of PC have resulted in its incorporation into virtually all of the modern built environment.
However, PC production results in large associated CO 2 emissions, approximately 0.73e0.99 t CO 2 /t PC, which is~8% of current global anthropological CO 2 emissions when considering a PC production volume around 4 Gt per annum [5,6]. Consequently, low-CO 2 cementitious materials have emerged as attractive, more sustainable alternatives to PC [4]. Supplementary cementitious materials (SCMs) such as coal fly ash (FA), blast furnace slag, calcined clays, finely ground limestone and silica fume are widely used in concrete. By blending PC with SCMs, it is possible to enhance and control physical properties, e.g. strength, durability, phase formation and reaction kinetics, and also to improve sustainability by reducing associated CO 2 emissions and valorising industrial wastes [7]. PC-free binders such as calcium aluminate cements (CACs) and calcium sulfoaluminate (CSA) cements, magnesia-based cements, alkali-activated materials (AAMs) and geopolymers, among others, have also received significant attention from academia and industry due to the enhanced technical properties and/or sustainability that they can provide, when compared with PC [4].
The complex chemical nature of cement systems, which often involve crystallographically disordered phases and which continue to evolve as a function of time for many years after initial mixing, has caused significant difficulty in understanding their atomic structure. Nuclear magnetic resonance (NMR) spectroscopy has become a key tool which offers the opportunity to clarify many of these important structural details. NMR spectroscopy involves the detection of interactions between nuclei with intrinsic magnetic moments (i.e. non-zero spin) and applied magnetic fields and enables determination of the local chemical environments of NMRactive nuclei, regardless of the level of crystallographic order or disorder, and this is essential in its application to the study of cements. In cements, nuclei of interest include 1 H, 13 C, 17 O, 19 F, 23 Na, 25 Mg, 27 Al,29 Si, 31 P, 33 S, 35 Cl, 39 K and 43 Ca (Table 1). Solid-state (SS) NMR is ideally suited to study both anhydrous and hydrated solid phases in cementitious materials. Its application to cement samples of different ages is invaluable in resolving the reaction mechanisms, kinetics and structural evolution that dictate the properties and performance of cements (and thus also concretes made using these cements). NMR can also be applied as a powerful tool probing the dynamics of mobile ions and water within cementitious materials, in particular through the use of 1 H NMR relaxometry which provides information about water mobility and pore size distribution [8,9]. These various applications and nuclei will be discussed in detail in the sections which follow.

Solid-state NMR spectroscopy
An NMR experiment involves holding a sample in a magnetic field (denoted B 0 ) and applying pulses of radiofrequency radiation to induce precession of the nuclear spin and measuring the electromagnetic response produced as the nuclei relax back to their equilibrium states. The electromagnetic signal is measured as a free induction decay, which is then converted to an NMR spectrum by applying a Fourier transform. The NMR spectrum contains resonances characteristic of near-neighbour atomic environments [10,11]. Terms representing the magnetic dipolar interaction, chemical shielding and quadrupolar interactions between the nucleus and its environment are of particular interest in the study of cements.
NMR spectra of solids are broadened (compared to those of liquids) due to dipolar interactions, anisotropy of the chemical shielding and quadrupolar interactions between the intrinsic nuclear electric quadrupolar moment and the surrounding electric field gradient [10]. This necessitates the application of sample spinning methods to reduce broadening.
Magic angle spinning (MAS) requires spinning a sample at an angle of cos 2 (q MAS ) ¼ 1/3 (q MAS z 54.74 ) with respect to the static magnetic field (B 0 ) so that dipolar interactions are suppressed, and both chemical shielding anisotropy and first-order quadrupolar interactions are removed [10,11], narrowing spectral lineshapes. This means that the local chemical environments of spin S ¼ ½ nuclei (which do not experience second-order quadrupolar interactions) can be represented in terms of the isotropic chemical shift (d iso ) in MAS NMR spectra. Quadrupolar nuclei (S > ½) experience additional second-order quadrupolar interactions, so the quadrupolar interaction term of the Hamiltonian is non-zero under MAS conditions. This limits spectral resolution due to anisotropic broadening of the signals. However, these second-order quadrupolar interactions are inversely proportional to the strength of B 0 and can thus be reduced through the use of high magnetic fields [10,11]. Owing to the disordered structure of many cementitious phases, significant line broadening occurs at all practically achievable MAS rates, and so the spectra for relevant nuclei are often poorly resolved. This is exacerbated in quadrupolar nuclei such as 17 O(S ¼ 5/2), 23 Na (S ¼ 3/2), 25 Mg (S ¼ 5/2), 27 Al (S ¼ 5/2), 33 S(S ¼ 3/ 2), 39 K( S ¼ 3/2), and 43 Ca (S ¼ 7/2) [12].
The presence of paramagnetic species within cementitious materials can limit the applicability of SS NMR for their characterisation; these species influence the relaxation and chemical shift of nearby nuclear spins, which can severely dampen and shift the NMR signal of these nuclei [10]. This is particularly problematic in standard 'grey' PC and in ferrite-containing sulfoaluminate cements, which each contain tetracalcium aluminoferrite (4CaO$Al 2 O 3 $Fe 2 O 3 ,C 4 AF), as well as in binders containing coal FA (containing Fe 2 O 3 ) [13]. The use of Fe 2 O 3 -free cement systems for spectroscopic analysis [14e16] can, therefore, be advantageous.
Recent advances in this understanding of the chemistry and materials science of cements, in which the application of SS NMR has played a pivotal role, are discussed in the following section.
3. Terminology used in defining the local structure of silicates and aluminates Si sites are generally identified using notation of the type Q n (mAl) with 0 m n 4, where Si in tetrahedral coordination (represented by Q) is bonded to n other tetrahedral atoms (m of which are Al) via oxygen bridges. Q n (mAl) Si sites within cementitious materials typically resonate in the region from À60 ppm to À120 ppm relative to tetramethyl silane, with significant overlap between the broad resonances observable in disordered solid phases, but a more negative chemical shift induced by a higher connectivity (i.e. higher n) ( Fig. 1) [11,16e18]. Each additional tetrahedral Al atom which replaces an Si neighbouring a Q n (mAl) silicon site also increases the chemical shift by approximately 5 ppm [11], and this provides an important point of differentiation between Q n (mAl) sites as a function of n and m, although the effects of the two parameters cannot always straightforwardly be discriminated without application of multinuclear NMR techniques.
Aluminium sites are generally differentiated by the coordination number, with tetrahedral Al species identified using q n notation, where q represents Al in tetrahedral coordination linked to n tetrahedral silicon sites via oxygen bridges, and 0 n 4. Bridges between two tetrahedral Al sites are disfavoured (a crystallographic and thermodynamic observation known as 'Loewenstein's principle' [19]). Tetrahedral and octahedral Al sites typically resonate within the regions 80e50 ppm and 20 -0 ppm, respectively, relative to Al(H 2 O) 6 3þ (aq), while resonances assigned to fivecoordinated Al and highly distorted tetrahedral Al environments have been observed in the region 50e20 ppm, Fig. 2 [11,17,20e22].

MAS NMR of anhydrous cementitious materials
4.1.1. Portland cement PC primarily comprises the 'clinker' phases tricalcium silicate (3CaO$SiO 2 ,C 3 S, in numerous polymorphs), dicalcium silicate (2CaO$SiO 2 ,C 2 S; the a and b polymorphs of this composition are preferred in cements due to their hydraulic nature, whereas g-C 2 S is unreactive), tricalcium aluminate (3CaO$Al 2 O 3 ,C 3 A) and tetracalcium aluminoferrite. This assemblage of clinker minerals is produced from limestone and silicate minerals in a rotary kiln, then rapidly cooled and interground with calcium sulphate to form PC [23]. Among these clinker minerals, tricalcium silicate (or a slightly impure form of this phase, containing Al and/or Mg among other substituents, which is commonly described as 'alite' in the cements literature [24]) is the predominant constituent of modern PC, defining its hydraulic nature and high early strength development upon cement hydration. The expected values of d obs , d iso and C Q for different nuclei in various phases in PC are shown in Table 2 and discussed in the following text. (alite, hartrurite). C 3 S is known to exist in seven polymorphs: three triclinic, three monoclinic and one trigonal. The alite present in PC crystallises from the melt in a trigonal form and upon rapid cooling, transforms into metastable   [25]. 29 Si MAS NMR analysis of the M I polymorph of alite has shown that Si exists within nine crystallographically distinct sites, approximately within the range from À69 to À75 ppm [26,27], while 29 Si MAS NMR spectra of the M III polymorph exhibit overlapping resonances between À66 and À78 ppm corresponding to eighteen distinct SiO 4 tetrahedra (Fig. 3) 27 Al MAS NMR [30], then later confirmed unambiguously using 27 Al multiple quantum (MQ)MAS NMR. Reduction of the temperature at which alite forms, which is desirable for both economic and environmental reasons in cement production, is often achieved by adding 'mineralising' additives such as fluoride and phosphate, which substitute into the alite structure. Tran et al. [31] used 29 Si and 27 Al MAS, 19 F- 29 Si cross polarisation (CP) MAS and 19 [32]. Skibsted et al. [36] extended this work by applying 19 F- 27 Al CP MAS and 19 Fe 29 Si CP MAS NMR spectroscopy to a series of CaF 2 -modified PCs containing between 0.23 and 0.77 wt% fluorine, demonstrating that increased fluorine content drives increased fluorine substitution into alite via a double substitution mechanism and that F À and Al 3þ ions in alite are likely to be clustered together.

Tricalcium silicate
Recently, 17 Oe 29 Si CP heteronuclear correlation (CP-HETCOR) MAS NMR was used to resolve oxide ion sites bonded to one Si 4þ from those bonded to Ca 2þ in synthetic tricalcium silicate for the first time [37]. Oxide ion sites bonded to Ca 2þ play an important role in the structure and reactivity of C 3 S but have proven difficult to resolve. 43 Ca MAS NMR analysis of alite was first reported by Moudrakovski et al. [38], with the high-field (21.14 T) natural abundance 43 Ca MAS NMR spectrum of the monoclinic polymorph showing a broad, asymmetric resonance consistent with overlapping 43 Ca resonances from the 36 non-equivalent sites for Ca in this phase and deviations from a perfect lattice (Fig. 4) [28]. This contrasts with the 43 Ca MAS NMR spectrum of the triclinic polymorph, which exhibited well-separated 43 Ca resonances despite the 29 non-equivalent sites for Ca in this phase [39], suggesting that some of these non-equivalent sites are quite similar. This work demonstrates the power of SS 43 Ca MAS NMR to resolve polymorphism, which is valuable in identification, characterisation and quantification of metastable polymorphs within cement systems, although the time and instrumentation required to collect highresolution 43 Ca spectra may prove to be a limitation in a practical sense. Differences in 43 Ca MAS NMR spectra have also been exploited for spectroscopic analysis of the CaCO 3 polymorphs calcite, aragonite and vaterite [40,41], which are also present in many cement systems.   27 Al chemical shifts for different aluminium sites in aluminates and aluminosilicates [11]. Al and O atoms are represented by blue and red spheres, respectively.  [43]. Some PCs also contain the a polymorphs of C 2 S, but these do not appear to have been analysed specifically by high-resolution SS NMR in the available body of literature, so the focus here will be on the b form.
Despite the fact that there are four crystallographically distinct oxygen sites within b-C 2 S [42], a single broad resonance is observed in the 17 O MAS NMR spectrum [43], suggesting that there is some degree of disorder in this phase. As for the case of alite, substitution of Al 3þ for Si 4þ in C 2 S has been identified crystallographically since the 1950s [45,46]. In 1994, 27 Al MAS NMR was used to provide the first detailed spectroscopic understanding of Al substitution for Si in C 2 Si nP C [30], which has significant consequences for the dissolution and hydration reactions of this phase, given the thermodynamic preference for dissolution of Al from calcium aluminosilicates [47]. Skibsted   and (d). Adapted from Poulsen et al. [29]. Parts (e) and (f) show two different sections taken through the monoclinic superstructure of the M III polymorph of alite, viewed down the b axis, showing polyhedra located in (e) and above (f) the a,0,c plane. Adapted from Nishi et al. [28].  The 43 Ca MAS NMR spectrum of belite was first reported by Moudrakovski et al. [38], Fig. 4. The high-field (21.14 T) natural abundance 43 Ca MAS spectrum exhibited two distinct resonances consistent with the crystal structure of this phase and calculations from density functional theory (DFT) (  27 Al MAS NMR, is constrained by the nuclear-unpaired electron dipolar couplings between the 27 Al nucleus and paramagnetic Fe 3þ ions [53e55], which severely broaden the 27 Al MAS NMR spectra and dampen the signal [52,55] as demonstrated in a study by 27 Al MAS NMR (9.4T, 11. 5 n R 17.5 kHz) [55], examining Ca 2 Al x Fe 2-x O 5 with x ¼ 0.93, 1 and 1.33 (Fig. 6). Increasing Fe content dramatically dampened and broadened the observed resonances, but despite these difficulties, it was possible to observe that the central transitions showed a dominant resonance for Al IV species at 61 d obs 71 ppm, with increased shielding at higher Fe content, and Al VI species at 0 d obs 20 ppm.
As mentioned in section 2, the presence of C 4 AF in PC also has significant implications for acquisition of 29 Si MAS NMR data for C 3 S and C 2 S phases. A dramatic decrease in the longitudinal relaxation time, T 1 , in grey PC has been attributed to the close proximity of these diamagnetic phases to paramagnetic C 4 AF [56,57]. However, Poulsen et al. [29] attributed this effect primarily to Fe 3þ ions incorporated as guest ions within C 3 S and C 2 S phases, rather than intermixing with ferrite phases. These effects need to be considered carefully when attempting to quantify C 3 S and C 2 S phases in PC via deconvolution of 29 Si MAS NMR spectra as preferential incorporation of Fe 3þ ions within either of the silicate phases will result in a dampening of the 29 Si MAS NMR signal for that phase and consequently lead to incorrect quantification [57]. Further discussion of methods to mitigate this issue is provided in section 4.3.
4.1.1.5. Gypsum. Gypsum (CaSO 4 $2H 2 O, C$H 2 ), and/or its partially dehydrated forms, is interground with PC clinker in small quantities to control the setting rate of the hydrating cement. During PC hydration, C 3 A reacts with gypsum to form ettringite (AFt) which is subsequently converted (in part or in full) to monosulfoaluminate (AFm), starting a few hours after initial hydration [23]. In the absence of gypsum, C 3 A would hydrate rapidly and setting could occur within minutes (i.e. 'flash setting').
The first application of NMR to any of the cement phases discussed in this article was undertaken by Pake, who in 1948 applied 1 H NMR to single crystals and powdered hydrates of gypsum, demonstrating that the splitting observed in the spectrum (the Pake doublet) reflects the dipolar coupling of two 1 H nuclei and therefore the internuclear distance between H atoms within the water molecules of gypsum [58]. Natural abundance 43 Ca MAS NMR (8.6 T, n R ¼ 2 kHz) has revealed a single Ca site in gypsum resonating at approximately d obs ¼À28 ppm [40]. 33 S MAS NMR studies of gypsum (both natural abundance and enriched with 33 S) revealed a single S site in gypsum resonating at d iso ¼ 328 ppm (Fig. 7) [59]. However, it was noted that acquisition of natural abundance 33 S MAS NMR spectra for PC samples is not practical due to extremely long acquisition times, and consequently, this technique does not appear useful for field samples or those without isotopic enrichment.

Innovation in cements to reduce CO 2 emissions
Commercially produced cements contain a worldwide average of~20%, but up to 95% in some cases, replacement of PC clinker by SCMs, while maintaining similar performance to existing cements [61]. The SCMs used are primarily fine limestone, ground granulated blast-furnace slags (GGBFSs), coal FAs and silica fume, with recent developments also focussing on ternary blends, e.g. those containing both calcined clays and ground limestone [61]. The use of non-PC hydraulic clinkers such as CACs, CSA cements, beliteye'elimite-ferrite clinkers and magnesia-based cements, as well as non-clinkerebased cements such as AAMs, can also offer significant reductions in CO 2 emissions.
The materials used as SCMs are often crystallographically disordered, compositionally variable and multiphase in nature. SS NMR has consequently been instrumental in characterising the local structure of these materials, which are used both as SCMs in blends with PC and also as precursors for alternative cements. The   [60] and the NMR parameters reported in Ref. [59]. Adapted from d'Espinose de Lacaillerie et al. [59]. expected d obs , d iso and C Q for different nuclei in the main phases found in common SCMs are shown in Table 3 and discussed in the following text.  2 ). Calcite and aragonite, as well as the additional CaCO 3 polymorph vaterite, are also generated within cements in service as the calcium-rich cement hydrate phases react with atmospheric CO 2 in a process known as carbonation [62]. 43  GBFS GGBFS is generated as a secondary product of pig iron production in a blast furnace; the liquid silicate slag is tapped from the furnace, rapidly cooled by granulation and then ground to a particle size similar to that of PC. It can be blended with PC at replacement ratios as high as 95% under current European standards (EN 197-1) [66] and is also a primary precursor for low-CO 2 cements [4].
GGBFS consists primarily of depolymerised calcium silicate glasses, as well as small amounts of low-crystallinity phases within the melilite group [18,67e70]. The composition of GGBFS is dependent on the characteristics of the iron produced and ore used, as well as the operational details of the blast furnace [69], with CaO, Al 2 O 3 , SiO 2 and MgO contents typically varying between 35e42 wt %, 7e13 wt %, 34e36 wt % and 6e15 wt % respectively [67e69,71e73]. The Al in the depolymerised silicate glass is charge balanced mainly by Ca 2þ cations, with excess calcium cations contributing to the depolymerisation of the aluminosilicate framework [67e69]. As in most glasses, the reactivity of GGBFS is dependent on the level of depolymerisation of the aluminosilicate framework [67,68]. 29 Si and 27 Al MAS NMR studies of GGBFS were interpreted as indicating that silicon is present mainly as a dimeric tetrahedral species, and Al is present in tetrahedral coordination [74]. Shimoda et al. [75e78] examined local structures in an amorphous synthetic slag using isotopic enrichment and multinuclear MAS and MQMAS NMR, probing 27 Al,29 Si, 25 Mg, 17 O and 43 Ca nuclei (Fig. 8), and showed that the amorphous slag framework structure can be generally described as a depolymerised chain-like network of SiO 4 tetrahedra branched with AlO 4 tetrahedra, with oxygen atoms occupying structurally inequivalent sites (dependent on their bonding nature), and multisite occupancy of Mg and Ca ions. These observations are supported by molecular dynamics simulations [79]. The structure and chemical characteristics of GGBFS are analogous to those of numerous other calcium aluminosilicate glasses which have been examined in detail using MAS NMR [80e86].

Coal FA.
Coal FA, a by-product of thermoelectric coal combustion, comprises primarily Al 2 O 3 and SiO 2 within an assemblage of aluminosilicate glassy phases, with some additional minor crystalline constituents, which can include quartz, mullite, ferrite spinels, calcium aluminates and others [87e89]. The characteristics of coal FAs (composition, extent of heterogeneity in the glass and the local structure of each phase within the ash) vary widely between sources and as a function of time and are dependent on the type of coal used and the combustion process through which it passes [90e92]. Mineral constituents in coal (mostly silicates and aluminosilicates) are melted as the coal is burned, become entrained in the flue gas as small droplets and are then cooled and collected as a fine powder, by electrostatic precipitators in the chimney stacks of the power station. FA is blended with PC at replacement fractions up to~50% in many parts of the world as its reactive silicate fraction can react with calcium hydroxide through  27 Al and 29 Si MAS NMR resonances for slag typically spans between 60 and 80 ppm; for clarity, only the resonance in this distribution exhibiting the maximum intensity is provided.
the 'pozzolanic' reaction (Ca(OH) 2 þ reactive silicates / calcium silicate hydrate) to produce desirable binding phases, while reducing the overall environmental impact of the cement and also valorising this ash which would otherwise be a problematic highvolume waste requiring landfilling. Phase segregation upon cooling occurs over the majority of the relevant compositional range for aluminosilicate glasses, and so as the molten aluminosilicate materials are quenched rapidly as the flue gas cools, the various phases formed are finely interspersed within the FA particles. The result is a highly heterogeneous material, as different FA particles cool at different rates and are derived from different mineral matter entrained in the coal [93], resulting in both inter-and intra-particle variation in local phase composition [94]. As a result, FA reactivity when used as an SCM (i.e. blended with PC) or as a precursor for AAM varies significantly between ashes, and between phases within each particular ash [95e102].
The crystalline phases present in FA occur in small amounts; in the low-calcium ashes which are considered most desirable for blending with PC, these phases (predominantly quartz, mullite and ferrite spinels) are relatively unreactive when compared to the aluminosilicate glassy phase. Consequently, knowledge of the composition of the glassy phase is important for correct formulation when designing cements containing FA, and SS MAS NMR (used in conjunction with electron microscopy) is a key technique that can provide this information, as long as the ash is not too rich in iron. 29 Si and 27 Al MAS NMR studies of FA have shown that the vitreous aluminosilicate phase contains a distribution of Q 4 (mAl) Si species, with aluminium mainly present in poorly ordered tetrahedral coordination, and also revealed the presence of a small amount of Al in octahedral coordination in mullite-like crystallites and glasses ( Fig. 9) [20,88,103e105]. The quantity of each Q 4 (mAl) Si species, as well as the distribution of Al IV and Al VI sites, is dependent on the local phase chemical composition. 17 O MAS NMR studies have shown that alkali and alkali earth metals (mainly Na and K in FA) can act as network modifiers and form non-bridging oxygen sites if present in high enough concentrations in aluminosilicate glasses [106] which are analogous to the glassy phase in FA.

Metakaolin.
Metakaolin is a layered aluminosilicate material which is used in cements as a pozzolanic additive and is produced by dehydroxylation of kaolinite clay at temperatures of 500ee800 C [107,108]. It is frequently used as an SCM in blends with PC [61,109]. 29 Si and 27 Al MAS NMR studies of metakaolin have shown that it consists of alternating buckled silicate and aluminate layers [68], with silicon in tetrahedral coordination (Fig. 10a) [110] and aluminium in a distribution of tetrahedral, pentahedral and octahedral coordination (Fig. 10b and c) [107,110e113 ] , with each site exhibiting a significant degree of asymmetry in the local electric field gradient [113]. It has been suggested that metakaolin contains approximately equal amounts of tetrahedral (Al IV ), octahedral (Al VI ) and pentahedral Al (Al V ) [111,114,115]; however others have found through crystallographic analysis that some of the Al is tricoordinated [116].  29 Si, (b) 27 Al, (c) 17 O and (d) 43 Ca MAS NMR spectra (16.4 T, n R ¼ 18 kHz) of synthetic slag enriched in 17 O and 43 Ca. Adapted from Shimoda et al., 2008 [75]. * Spinning sidebands.

Silica fume.
Silica fume is an amorphous form of SiO 2 with extremely small (tens of nm) particle size, produced as a byproduct of semiconductor Si manufacture [118]. It is commonly blended with PC at relatively low levels (up to 10 wt%) to promote strength development and durability. Published work examining silica fume via NMR is focussed on 29 Si MAS NMR of PC-silica fume blends, which has shown that silica fume comprises a distribution of Si atoms in tetrahedral coordination (Q 4 units), connected by oxygen bridges [8,119,120]. Because silica fume is a pure silica source with a 29 Si resonance that is clearly distinct from that of either PC or any of its hydration products, it is possible to accurately determine the extent of reaction of silica fume within a blended cement through 29 Si MAS NMR; this is much more accurate for silica fume than for any other siliceous SCMs, as the other SCMs have resonances which at least partially overlap the signal due to the PC or its reaction products [121].  [23,122]. While considerably more expensive than PC, these cements are intended for use in applications where rapid hardening is useful and/or when resistance to thermal stresses, chemical (particularly acid) attack and impact or abrasion is desired [122], e.g. refractory concretes, cements and concretes for effluent treatment infrastructure, blended cements for marine applications and other demanding service environments. CACs are now restricted from bulk use in structural concrete due to a series of material failures in past decades [123], but this does not limit their use in specialist applications. The expected d obs , d iso and C Q for different nuclei in CAC are shown in Table 4 and discussed in the following text.

Monocalcium aluminate, CaAl 2 O 4
Monocalcium aluminate, CA, is the principal reactive phase which controls the performance of CAC. The crystal structure of CA consists of AlO 4 tetrahedra forming six-membered rings in a tridymite-like structure [124]. 27 Al MAS NMR isotropic chemical shifts and quadrupolar coupling parameters for each of the six nonequivalent AlO 4 tetrahedra in CA have been determined by 27 Al MAS NMR at multiple fields   29 Si MAS NMR (9.4 T, n R ¼ 6 kHz, adapted from Dai et al. [117]), where * indicates a resonance from a quartz impurity), (b) 27 Al MAS NMR (11.7 T, n R ¼ 15 kHz, adapted from Duxson et al. [111]) and (c) 27 Al triple quantum (3Q) MAS NMR (11.7 T, adapted from Kobera et al. [113]) spectra of metakaolin.  27 Al MAS NMR spectra (14.1 T, n R ¼ 10 kHz) of fly ash. Adapted from Bernal et al. [20]. Note that the spectra are significantly broadened by the presence of iron (5.2 wt% represented as Fe 2 O 3 by X-ray fluorescence analysis) in the ash.
2.5 MHz C Q 4.3 MHz by deconvolution of both the resonance from the central transition and the spinning sidebands. Variations in the geometry of the AlO 4 tetrahedra (primarily differences in Al-O bond lengths [124]) are primarily reflected in the 27 Al MAS NMR quadrupolar parameters rather than in chemical shifts. This work contrasted earlier assertions that the resonance from the central transition could be simulated with a single set of quadrupolar parameters and two isotropic chemical shifts [51].
High-field (18.8 T) 17 O MAS NMR of CA has resolved eight resonances due to O atoms in Al IV -O-Al IV linkages [84] with chemical shifts between 50 and 90 ppm and C Q values less than 2 MHz, with the tricluster oxygen site exhibiting the largest C Q (Fig. 11). This work demonstrated the benefits of utilising high fields which can narrow resonances sufficiently to resolve additional sites; eight of the twelve crystallographically distinct 17  Calcium dialuminate (CA 2 ) is found in relatively alumina-rich CAC clinkers that are used as refractory cements and is a slowly reacting cement constituent. Its structure comprises two tetrahedrally coordinated Al atoms; one AlO 4 tetrahedron contains two tricoordinated oxygen atoms and the other contains a single tricoordinated oxygen atom, with the remaining oxygens each connected to two Al atoms (Fig. 12) [126,127]. 27 Al MAS NMR data collected at multiple fields (7.1 T, n R ¼ 18.0 kHz; 9.4 T, n R ¼ 18.0 kHz and 11.7 T, n R ¼ 15.2 kHz) resolved single resonances for each tetrahedrally coordinated Al atom, with d iso ¼ 75.5 ppm and 69.5 ppm [52], and significant C Q values (6.25 MHz and 9.55 MHz, respectively) due to electric field gradient (EFG) asymmetry caused by the nearby tricoordinated oxygens. Earlier work by Müller et al.
[51] determined similar d iso and C Q values for the site neighbouring two tricoordinated oxygens compared to those obtained by Skibsted et al. [52], but those for the AlO 4 tetrahedra containing a single tricoordinated oxygen atom exhibited larger discrepancies, likely because of the insufficient resolution and notable distortion of the central transition caused by the much lower spinning speed utilised in those early experiments. 17

Dodecacalcium hepta-aluminate (mayenite), C 12 A 7
Mayenite, (Ca 12 Al 14 O 33 ,C 12 A 7 ), is generally present as a minor phase in more calcium-rich CAC clinkers. It has also been discussed in the past as a potential constituent of PC clinkers but is now considered unlikely to form under realistic commercial PC production conditions [23]. 27 Al MAS NMR at high spinning speed (17.8 kHz) resolved two distinct Al IV sites within C 12 A 7 , with d iso ¼ 85.9 and 80.2 ppm, respectively, in close agreement with the known crystal structure [129] and other 27 Al MAS NMR investigations of this phase [50,51]. Differences between the d iso and quadrupolar parameters (3.8 MHz and 9.7 MHz) of these two Al IV sites arise due to distortions in one of the tetrahedra [50], which result in oxygen mobility [130]. Consequently (although maybe incidentally in the context of cement chemistry), mayenite has also received significant attention as a high-performance oxygen ion conductor [131].

CSA cements.
CSA cements comprise calcium sulfoaluminate (ye'elimite, Ca 4 (AlO 2 ) 6 SO 4 ,C 4 A 3 $) and other clinker phases such as belite, C 4 AF, ternesite (C 5 S 2 $) and/or calcium aluminates and are typically blended with calcium sulphate to regulate the rate of their setting [61,132]. 27 Al MAS NMR investigations of commercially produced CSA and synthetic ye'elimite showed nearly identical spectra, consisting of a single peak (with maximum intensity at approximately 68 ppm) with two high-field shoulders at approximately 60 and 52 ppm [34,133e135]. Evidence enabling discrimination of two of the eight non-equivalent Al sites present in the crystal structure of ye'elimite (which is an aluminate analogue of the sodalite framework) [136] was identified in this broad peak [133]. Recent work by Skibsted et al. [137] reported that the 27 Al MAS NMR spectral region for Al IV in ye'elimite (~80À50 ppm) is dominated by overlapping resonances from the eight different Al sites. Isotropic projections from 27 Al MQMAS NMR spectra obtained by the same authors revealed four distinct resonances with very similar isotropic chemical shifts and quadrupole coupling parameters [34,138].
Ternesite, sometimes also called sulfospurrite, is a common minor constituent in modern CSA cements that is now being understood to have some hydraulic activity but also forms within PC kilns as an undesirable deposit on the refractory that reduces kiln performance. Ternesite exhibits a single, well-defined 29 Si MAS NMR resonance at À73.0 ppm [139], consistent with its crystal structure [140] in which double layers of silicate tetrahedra alternate with single layers of sulphate tetrahedra, with calcium 7coordinated to oxygen atoms.
The application of NMR spectroscopy to the belite and tetracalcium aluminate phases present in CSA cements has been discussed in detail in sections 4.1.1.2 and 4.1.1.4, respectively.

Magnesia-based cements.
Magnesium silicate hydrate (M-S-H) cements are typically formed from a source of magnesium and a source of highly reactive silica, such as silica fume [141]. The magnesium source is typically light-burned MgO, and Mg(OH) 2 can also be used [142]. Although M-S-H was initially identified as a degradation product in PC exposed to chemically aggressive conditions [141,143], subsequent work has shown that M-S-H gels can also form a cementitious mass to generate high compressive strength [144e148]. These gels will be discussed in detail in section 4.2.6.
The 25 Mg MAS NMR spectrum of polycrystalline MgO exhibits a single sharp resonance at d iso ¼ 26 ppm [149,150], while the 17 O MAS NMR spectrum of this compound enriched with 17 O contains a single sharp resonance at 47 ppm [150,151], consistent with octahedral MgO 6 sites in the cubic crystal structure of this phase, and in good agreement with first-principles calculations using DFT [152]. The 25 Mg MAS NMR spectrum of Mg(OH) 2 (brucite) exhibits a typical quadrupolar resonance with d iso ¼ 14.1 ppm and [153]. Application of NMR spectroscopy to silica fume has been discussed in detail in section 4.1.2.1.5. Many other magnesia-based cements also exist, including those based on magnesium carbonates, phosphates and oxysalts (both chloride and sulphate) [141]; MgO is the primary Mg source used in the production of most of these cements.

MAS NMR of key cementitious binding phases
The vast majority of SS MAS NMR investigations of cementitious binding phases, which are often complex, disordered silicates [17,18,20,21,104,154,155], have probed 29 Si and 27 Al nuclei, yielding information about the coordination states of Al and the connectivity of Si (via oxygen bridges to Si or Al or to non-bridging oxygen sites). The low natural abundance of 29 Si (4.7%) results the need for long data acquisition times for non-enriched samples, but the wealth of information obtainable from this technique has meant that it is nonetheless widely used [22].

Calcium silicate hydrate gels
Calcium silicate hydrate (C-S-H) is the main product of PC hydration. It displays low crystallinity and variable composition and comprises (in general terms) Q 2 Si chains of varying lengths with Q 1 Si sites at chain termination points, as shown by 29 Si MAS NMR [17,54,156,157]. These silicate chains are flanked by an interlayer containing confined H 2 O molecules, aqueous cationic species (Ca 2þ and H þ in pure C-S-H, but with scope for substitution, particularly by alkali metals) and a calcium oxide sheet ( Fig. 13) [158,159]. The silicate chains are arranged in a 'dreierketten' structure, based on a repeating chain unit of three tetrahedral sites, two 'pairing' and one 'bridging', where vacancies in the bridging sites lead to characteristic chain lengths of (3nÀ1) for integer values of n [160]. The two pairing Si IV units (Q 2 (p)) are linked with the calcium oxide sheet [159]. The expected d obs , d iso and C Q for different nuclei in C-S-H, aluminium-substituted C-S-H (C-(A)-S-H) and alkali-and Illustration of the local environments in the crystal structure of CA 2 , based on information obtained from single-crystal diffraction data [127]. The two AlO 4 tetrahedra are denoted by Al1 (containing two tricoordinated oxygen atoms) and Al2 (containing a single tricoordinated oxygen atoms). Ca atoms are represented by cyan spheres, Al atoms are represented by dark and light blue spheres (denoting Al1 and Al2, respectively) and O atoms are represented by red spheres. Prepared using the VESTA software package [128].
aluminium-substituted C-S-H (C-(N,K)-A-S-H) are shown in Table 5 and discussed in the following text.
Numerous structural models for C-S-H have been proposed, including those based on jennite [161,162] and tobermorite [159,163], with recent SS NMR work probing both 29 Si [164e167] and 17 O [168] nuclei indicating strong structural similarity of C-S-H with a distorted tobermorite-like structure. 29 Si MAS NMR spectroscopy in combination with other techniques has shown that C-S-H forms a structural series which shows both continuity and diversity, with phase-pure C-S-H exhibiting Ca/Si ratios at least between 0. 55 and 2.0 [167,169,170]. Recent work combining dynamic nuclear polarisationeenhanced ( 29 Si) 29 Si double quantum coherence (the Incredible Natural Abundance Double Quantum Transfer Experiment, INADEQUATE) and ( 1 H) 29 Si HETCOR MAS NMR analysis approaches, with atomistic modelling, determined a series of atomic level structures for C-S-H based on defective tobermorite with Ca/Si ratios of 1.25e2.0 [170]. These models include a structurally important interlayer calcium site which bridges Q 1 silicate species and is associated with strong hydrogen bonding that stabilises the C-S-H, allowing high Ca/Si ratios to be reached in a phase-pure C-S-H.
Despite experimental and analytical difficulties arising from the low natural abundance and large quadrupole moment of 17 O, MAS NMR spectroscopy probing this nucleus has revealed valuable information regarding the role of oxygen species in C-S-H gels. Cong et al. [43,168]  As mentioned previously, natural abundance 43 Ca MAS NMR experiments face inherent difficulties due to the low natural abundance (0.145%), high quadrupole moment and small magnetogyric ratio of this nucleus, Table 1. Consequently, very few studies probing 43 Ca in C-S-H via MAS NMR have been performed. Moudrakovski et al. [38] used 43 Ca MAS NMR to study a series of C-S-H gels of differing Ca/Si ratios, as well as C-S-H produced by hydrating triclinic C 3 S, crystalline 11 Å tobermorite and portlandite (Fig. 14). The coordination state of Ca 2þ in C-S-H is complex and not yet fully resolved, but the work of Richardson [171] indicates the expectation that the most probable coordination numbers in plausible C-S-H structures are 6 or 7 depending on the site type and occupancy.
Paired SiO 4 chain sites (Q 2 (p)) and bridging SiO 4 chain sites (Q 2 (b)) in C-S-H in hydrated PC exhibit overlapping 29 Si MAS NMR resonances [16,172] and therefore cannot be resolved individually Fig. 13. Schematic representation of the structural features of calcium-silicate hydrate (C-S-H) gels, adapted from Provis and Bernal [4]. Tetrahedral Si sites and CaO layers are shown by blue triangles and green rectangles, respectively. Circles denote various interlayer species (water or cations). Table 5 Coordination states and expected d obs , d iso   using this technique; however, these sites can often be resolved by their differing 29 Si MAS NMR chemical shifts in synthetic C-S-H gels ( Fig. 15) [173e177]. 43 Ca MAS NMR spectra of the synthetic C-S-H gels show broad resonances at approximately 27e31 ppm, very similar to the spectrum of 11 Å tobermorite [178]. An increased Ca/Si ratio leads to decreased shielding of 43 Ca ions (i.e. resonances move to higher chemical shift), consistent with a decrease in the mean chain length (MCL) of the silicate, and also in good agreement with 29 Si MAS NMR data [167].
The 43 Ca MAS NMR spectrum of hydrated triclinic C 3 S contains a broad resonance across a chemical shift range that includes that of the main resonances from C-S-H gels with 0.8 Ca/Si 1.5, indicating that C-S-H in hydrated cement systems may contain various locally different chemical compositions [38]. Combining these observations with analysis of 43 Ca MAS NMR signal intensity vs. delay time, Moudrakovski et al. [38] proposed that Ca in the Ca-O sheets resonates at 10 ppm d obs 45 ppm and interlayer Ca 2þ ions resonate at 35 ppm d obs 85 ppm at a field B 0 of 21.14 T. Bowers and Kirkpatrick [179] used natural abundance 43 Ca MAS NMR to demonstrate that despite broad spectra, six-coordinate Ca (as present in jennite) and seven-coordinate Ca (as present in 11 Å tobermorite) may be resolved by differences in chemical shift (d obs ¼ 24 ppm vs. À9 ppm, at 21.1 T). By using this approach, it may be possible to resolve the various locally different chemical compositions (i.e. those with differing Ca/Si ratios) within C-S-H gels.  Despite severe line broadening in 1 H MAS NMR spectra which results from 1 H-1 H dipolar couplings, the high sensitivity of 1 H, and the development of multipulse decoupling sequences which reduce line broadening, have led to the application of this technique to study proton environments in C-S-H [157]. However, the information obtainable from 1 H MAS NMR spectra of C-S-H is generally limited, due to the distribution of protons across many chemical sites and the small chemical shift range (approximately 20 ppm) over which 1 H in solid phases can resonate [157]. 1  Recently, quantitative 29 Si MAS and 1 H- 29 Si CPMAS NMR experiments have been used to monitor hydration of 29 Si-enriched triclinic C 3 S in-situ over the first 24 h of reaction [183]. These results were correlated with isothermal calorimetry measurements to identify the previously unobserved phenomenon of partial passivation of the surface of C 3 S which drives deceleration of the reaction. 29 Si MAS NMR has revealed important information regarding incorporation of Al in CeSeH (i.e. forming C-(A)-S-H) in hydrated PC [16,57,167,172,184e186],P C eSCM blends [117,187,188] and 2000-year old Roman seawater harbour concrete [189]. Application of 29 Si MAS NMR has allowed observation of resonances from Q 1 ,Q 2 and Q 2 (1Al) sites which make up the silicate chains in the C-(A)-S-H gel [16,167,172]. The MCL can be calculated from the intensities of Q n (mAl) resonances obtained from 29 Si MAS NMR spectral deconvolutions by using equation (1), while the degree of Al-Si substitution in the tetrahedral chains can be obtained from the intensities of the Q 1 ,Q 2 and Q 2 (1Al) resonances (equation (2)), as demonstrated by Richardson et al. [184,185] and Andersen et al. [186].

C-(A)-S-H gels.
Using 29 Si MAS NMR, Andersen et al. [186] showed that the Al IV / Si ratio in C-(A)-S-H in hydrated white PC is dependent on the availability of dissolved Al 3þ ions, but independent of hydration time. Richardson et al. [190] used 29 Si MAS NMR to determine that Al preferentially substitutes into Q 2 (b) sites and that the MCL of aluminosilicate chains in C-(A)-S-H is dependent on both the availability of Al 3þ ions in solution and hydration time, findings that were subsequently confirmed by Andersen et al. [184,186]. It has also been suggested that Al may substitute for Si in small amounts in pairing sites [188,191,192], although this is thermodynamically disfavoured compared to the bridging sites [191].
Al readily substitutes for Si in the bridging position in aluminosilicate chains in C-(A)-S-H up to a ratio of Al/Si 0.1, while at higher Al/Si ratios, katoite and/or str€ atlingite are also formed [15]. Analysis by 29 Si and 27 Al MAS NMR has shown that the Al IV /Si ratio in C-(A)-S-H in hydrated PC can be increased significantly by the presence of alkali cations (Na þ or K þ ) [16] via a mechanism in which the charge deficit introduced by Al-Si substitution is balanced by adsorption or bonding of alkali cations in the interlayer region. The role of alkalis in these structures will be discussed further in section 4.2.1.2. 27 Al MAS NMR experiments, particularly at high field (B 0 ! 14.1 T), have proven particularly useful in gaining detailed structural information regarding the role of Al in C-(A)-S-H, in both synthetic gels and hydrated PC. 27 Al MAS NMR experiments at multiple fields (e.g. Fig. 16) have shown in synthetic gels and hydrated PC that the Al IV bridging sites in C-(A)-S-H are charge balanced by interlayer Ca 2þ ,N a þ or H þ ions (d obs ¼ 66 ppm, B 0 ¼ 17.5 T) [190,193] or by interlayer or surface Al V or Al VI ions (d iso ¼ 74.6 ppm, C Q ¼ 4.5 MHz) [193].A l V sites in the C-(A)-S-H interlayer can potentially substitute for Ca 2þ ions (d iso ¼ 39.9 ppm, C Q ¼ 5.1 MHz) [192e194]. From these studies, and others, it was concluded that Al does not enter the Ca-O sheet, nor the pairing tetrahedral sites in the silicate chains of C-(A)-S-H [22,172,184,190,193,194].

C-(N,K)-A-S-H gels.
Synthesis of C-(A)-S-H gels in an environment containing high alkali metal concentrations, such as in the production of AAMs from GGBFS [154,160], yields a C-(N,K)-A-S-H gel. This is structurally similar to C-S-H, but with Al substitution in bridging tetrahedral sites and in the interlayer as discussed in the preceding section, and with charge balancing by alkali cations (Na þ and K þ ) in the gel interlayer, with additional alkalis able to sorb to the gel surfaces [74,154,185,195e197]. The structural characteristics of this complex gel have been elucidated largely by the application of SS NMR; representative spectra of alkali-activated GGBFS are shown in Fig. 17, while a schematic representation of the C-(N,K)-A-S-H gel structure is shown in Fig. 18. As with C-S-H, the C-(N,K)-A-S-H gel exhibits a disordered structure similar to that of crosslinked and/or non-crosslinked, structurally imperfect tobermorite [17,74,195]. As noted previously, there is a strong thermodynamic preference for Al substitution into bridging tetrahedra in the C-(N,K)-A-S-H gel [198]. 29 Si MAS NMR has provided extensive evidence of the presence of cross-linked (Q 3 and/or Q 3 (1Al)) sites within 2000-yearold Roman seawater harbour concrete [189], alkali-activated slag cements [20,188,199] and synthetic analogues [14,15,200e203]. The structure of these C-(N,K)-A-S-H gels was described by Myers et al. [195] as a mixture of cross-linked and non-crosselinked silicate chains within a defect-tobermorite structure, with crosslinking occurring through partially Al-substituted bridging sites [21,188,193,196]. Using this model and 29 Si MAS NMR spectral deconvolutions, it is possible to calculate the MCL and Al IV /Si ratio of the cross-linked C-(N)-A-S-H gel structures using the following equations (3) and (4) [195].
Crosslinking occurs predominantly in low-Ca C-(N,K)-A-S-H gels, and 29 Si and 27 Al MAS NMR results indicate that the extent of crosslinking is promoted by increased overall Al content up to a ratio of approximately 1 Al per 6 chain sites [21,204]. However, the extent of crosslinking has been observed to decrease over time as the reaction proceeds, attributed to the lower Al-binding capacity of crosslinked gel components [21]. Study of laboratory synthesised gels has shown that the amount of Al which is able to substitute into the C-(N,K)-A-S-H gel appears to be limited to approximately Al/Si 0.2, with C-(N,K)-A-S-H gel compositions typically within the range 0.5 < Ca/(Al þ Si) 1 and 0 < Al/Si 0.2 [15,192,193,205]. The MCL of the aluminosilicate chains is often 7e10 tetrahedra for C-(N,K)-A-S-H gels produced by sodium silicate or potassium silicate activation of slag [18,206]; however, this depends on the Ca/ (Al þ Si) ratio.
There are conflicting reports regarding the mechanism of alkali uptake and relationship with Al content of the C-(N,K)-A-S-H gel. 23 Na MAS NMR analysis has indicated that alkali cations exist in two distinct environments within C-(N,K)-A-S-H gels [164,207,208]; the spectra show a sharp resonance at d obs ¼À 7.6 ppm (7.05 T, n R ¼ 10 kHz) [164] attributed to either a   [195]. Paired and bridging Si tetrahedra are represented by blue and green triangles, respectively, and intralayer calcium, charge-balancing alkali cations and interlayer protons and/or calcium cations by the red diamonds, orange circles and purple squares, respectively. The yellow triangle indicates Al substituting for Si in a bridging tetrahedral site. highly symmetric bound alkali site or mobile alkali ions. An additional broad resonance at approximately d obs ¼À 3 ppm (14.1 T, n R ¼ 10 kHz) [14] to d obs ¼À 10 ppm (7.05 T, n R ¼ 10 kHz) [164] attributed to low-mobility Na within a disordered phase, possibly sorbed to the surface of the nanostructured gel.
Some observations suggest a direct correlation between Al content and alkali uptake [16,209], supported by an intuitive rationale of increased Al content driving an increase in alkali uptake to charge balance the AlO 4 À tetrahedra. However, other observations suggest an inverse proportionality between Al content and alkali uptake [210] or that Al content does not affect alkali uptake in the C-(N,K)-A-S-H gel [204,211,212]. Alkali uptake in C-A-S-H gels has been observed to be promoted by lower Ca/Si ratios and higher alkali concentration [213], presumably because lower levels of charge balancing Ca 2þ increase the requirement for charge balancing by alkali cations. It has been suggested that charge balancing alkali cations in the C-(N,K)-A-S-H gel interlayer induce additional structural disorder by reducing the periodicity of the interlayer and reducing the regularity of the stacking of the layers due to differences in ionic size, indicating that incorporation of alkalis into C-(N,K)-A-S-H has a significant effect on gel nanostructure [13,214]. However, when 29 Si MAS NMR was used to investigate compositionesolubilityestructure relationships in synthetic C-(N,K)-A-S-H gels [204], it appeared that the C-(N,K)-A-S-H gel structure was actually becoming more (rather than less) crystalline in the presence of alkalis, with both Na and K incorporated into its interlayer space [213]. The additional structural disorder was observed in C-(N,K)-A-S-H gel systems with lower pH (~13 [214]) than those in which greater ordering was observed (pH > 13.5 [204,213]); it is, therefore, likely to be the higher pH of the more alkali-rich systems which drives rearrangement to more crystalline structures, rather than the alkali cations themselves. Elevated temperature also drives increased polymerisation, ordering and phase purity of synthetic C-(A)-S-H gel structures [215].

Other guest ions in C-S-H gels (C-S-(X)-H).
In addition to Al 3þ and Na þ , other ionic species may be incorporated into the C-S-H gel structure. Owing to the 100% natural abundance and high magnetogyric ratio of 19 F and 31 P, SS NMR experiments probing these nuclei have been very useful in elucidating the structural sites in which these atoms can exist within the C-S-H gel. 19 F MAS NMR has shown that interlayer hydroxyl species may be substituted by F À [36,216]. Further analysis of F À speciation in C-S-H by 19 F- 29 Si-19 F forth and back (FB) CP MAS NMR (a modified double CP MAS NMR experiment which transfers polarisation forth and back between high-and low-g spin nuclei [217]) confirmed the aforementioned observation of F À for OH À substitution in the interlayer, as well as revealing a second F site attributed to F À ions associated with CaO 6 layers of a jennite-like part of the C-S-H gel [216,217]. PO 4 3À ions can also be incorporated in the interlayer of C-S-H gels. 31 35 Cl spinelattice relaxation time constants (T 1 ) and spinespin relaxation time constants (T 2 ) were used to study chloride sorption onto jennite (used as a model C-S-H phase) suspended in limesaturated NaCl solution [218]. This was intended to simulate hydrated PC in service in marine environments and showed that chloride undergoes rapid exchange between surface and bulk solution sites and exists predominantly in a hydrated, solution-like chemical environment.
High-resolution 29 Si, 23 Na and 133 Cs MAS NMR was used to study the interaction of sodium and caesium chloride salts with synthetic C-S-H [173]; this work is of particular relevance to the nuclear industry as some caesium isotopes are important fission product radionuclides. This work showed that both Na þ and Cs þ cluster in hydrated C-S-H, while in dry C-S-H Na þ , formed outersphere complexes (adsorbed with their hydration sphere) and Cs þ formed inner sphere complexes (bound directly to the surface with no intervening water molecules). 1 H, 13 C, 27 Al and 29 Si MAS NMR has been used to investigate the structure of the CaCO 3 polymorphs calcite, aragonite and vaterite [65,220],aswellasK 2 CO 3 [220], and their formation in hydrated PC (i.e. through carbonation) and interaction with PC hydrate phases [221]. This work showed that carbonation of hydrated PC occurs in two stages: gradual decalcification of C-S-H by removal of calcium from the interlayer and defect sites until Ca/Si ¼ 0.67 is reached (corresponding to infinite silicate chains in a tobermorite-like structure), followed by further decalcification by removal of calcium from the gel layers and formation of a layered or threedimensional silicate gel. The amount of carbonates formed is directly proportional to the Ca/Si ratio of the gel. The application of 13 C MAS and 1 H-13 CP MAS NMR also enabled carbonate anions in hydrous and anhydrous phases to be distinguished [65,220].
SS MAS NMR has been used to examine the interaction between a variety of organic additives and PC during hydration. Natural abundance 43 Ca and 13 C MAS NMR was used to examine the interaction between a poly(ethylene-vinyl acetate) (PEVAc) admixture, added to cement mortars to improve fracture toughness and impermeability, and hydrated white PC over the first 3 months of hydration. PEVAc hydrolysis occurs immediately as PC hydration commences, and induces minor structural changes to the hydrated PC, primarily formation of a small amount of amorphous, lowcoordinated calcium sites [222]. This demonstrates that PEVAc admixtures do not induce significant structural changes to hydrated white PC and are therefore unlikely to be detrimental to the performance of these cements. Conversely, 1 H-13 C CP MAS NMR showed that when mixed with 15 wt % poly (vinyl alcohol) and 10 wt % mineral antiblocking agents, PEVAc does not hydrolyse in the alkaline solution of hydrating PC and was adsorbed to the surface of the cement binder [223,224]. This is particularly important in the context of both chemical and physical durability as PEVAc and other similar latex dispersions are commonly used as waterproofing agents and sealants in cements.

Alkali aluminosilicate ((N,K)-A-S-H) gels
SS MAS NMR studies of hydrous alkali aluminosilicate in the context of cements are relatively recent compared to those examining C-S-H and related phases in hydrated PC, with the first application probing 29 Si speciation in alkali hydroxide/silicate activated metakaolin cements in 1988 [115,225,226]. Since these first experiments, SS MAS NMR analysis probing 29 Si and 27 Al has played a pivotal role in determining the structure of alkali-activated aluminosilicate cements, including those known as 'geopolymers' [68,227]. The main reaction product of alkali activation of aluminosilicate precursors has been revealed to be a three-dimensional hydrous alkali aluminosilicate gel, consisting of highly crosslinked (predominantly q 4 /Q 4 ) AlO 4 À and SiO 4 tetrahedra linked via bridging oxygen atoms, with a small number of terminal hydroxyl groups forming Q 3 sites. This structure is consistent across alkali aluminosilicate gels produced from reagent chemicals [228e231], calcined clays such as metakaolin [104,232e235], and aluminosilicate-rich waste materials such as coal FA [104,105,236e238]. This three-dimensional alkali aluminosilicate gel framework is often described as a highly crosslinked, disordered pseudo-zeolitic structure [13,239e243], and nanocrystalline zeolite phases have been observed as localised regions of ordering within the broadly less-ordered gel framework [239]. The short-range ordering has been observed by 29 Si and 27 Al MAS NMR spectroscopy [20,112,244,245] together with X-ray and neutron pair distribution function analysis [246e249] to extend to approximately 2e3 nearest neighbour shells. The expected d obs , d iso and C Q for different nuclei in (N,K)-A-S-H gels are shown in Table 6 and discussed in the following text. 27 Al and 29 Si MAS NMR have shown that Al and Si are present in tetrahedral coordination, with Si existing in Q 4 (mAl) environments where m is between 1 and 4 depending on the Al/Si ratio of the gel, and Al predominantly in q 4 (4Si) environments due to the energetic penalty associated with Al-O-Al bonding (Fig.  19) [236,237,250,251]. Reports of the observation of Al IV -O-Al IV bonds within alkali-activated metakaolin have also been made for materials with Si/Al close to 1.0 using 17 O triple quantum (3Q) MAS NMR [252], despite these bonds being significantly less thermodynamically favourable than Al IV -O-Si IV bonds, as entropic effects cause some violation of strict Si-Al ordering [250]. The negative charge associated with Al substitution for Si is balanced by hydrated alkali cations (Fig. 20) and is thought to be delocalised across all oxygen atoms, with the oxygen atom closest to the charge balancing alkali cation carrying the majority of this delocalised negative charge [114,252]. The (N,K)-A-S-H gel nanostructure is significantly influenced by kinetic limitations on silica and alumina release from solid precursors if these are used [253,254] and consequently evolves over time as the reaction process proceeds [237]. Schematic representations of the disordered N-A-S-H gel based on recent structural descriptions are shown in Fig. 20. 23 Na MAS NMR analysis of metakaolin-derived (N,K)-A-S-H gels (Fig. 21a) [111,257] identified a single disordered Na environment within the gel attributed to Na þ in a charge-balancing role, whose chemical shift (d obs ¼À3.5 to À6 ppm, B 0 ¼ 11.7 T) was independent of the Si/Al ratio. Aqueous Na þ (d obs ¼ 0 ppm, B 0 ¼ 11.7 T) was also observed within the pore solution, charge-balancing Al(OH) 4 À (aq) species, in samples with Si/Al 1.40. Charge balancing Na þ sites resonating at a similar frequency (d obs ¼À4 ppm, B 0 ¼ 14.1 T) have also been observed in alkali aluminosilicate gels produced from synthetic precursors [228]. These sites resonate at similar frequencies to those of hydrated Na þ ions in hydrothermally altered glass [258], providing opportunities to draw structural parallels to better understand the nature of the cementitious gels. K þ ions in alkali aluminosilicate gels balance the charge deficit from Al 3þ ⇔ Si 4þ substitution within the aluminosilicate framework and behave similarly to charge-balancing Na þ ions [259,260] (Fig. 21b), and the incorporation of K þ ions into the (N,K)-A-S-H gel is preferential compared to Na þ ions when both are present.
Related to this selectivity for larger alkali cations, SS NMR has also been applied to study the interaction of caesium with alkali aluminosilicate gels, to investigate their applicability in nuclear waste immobilisation applications. 133 Cs MAS NMR spectra of Csdoped alkali aluminosilicate gels produced from metakaolin and sodium silicate exhibit a single broad asymmetric resonance at approximately d obs ¼ 0 ppm (B 0 ¼ 9.4 T) [261] consistent with that of Cs þ bound tightly to an aluminosilicate surface (such as that in kaolinite and illite [208,262]), strengthening the appeal of alkali aluminosilicate gels in nuclear waste immobilisation applications.  29 Si MAS NMR spectra (7.05 T, n R ¼ 5 kHz) and (b) 27 Al MAS NMR spectra (11.7 T, n R ¼ 15 kHz) of N-A-S-H gels produced by reacting metakaolin with a sodium silicate solution (with Si/Al ratios as marked) and curing for 14 days. Adapted from Duxson et al. [111,112]. [20,263e265] (Fig. 22), and the gels appear to be stable in coexistence [14,20,231,266e268], as long as there is sufficient alkali content [117,266,269e271]. SS NMR has proven particularly useful in determining the structure and composition of these coexisting gel frameworks which are otherwise very difficult to discriminate. Owing to the structural similarity between these two gels, many resonances in their SS NMR spectra overlap; however, these can be distinguished and attributed appropriately with sound chemical reasoning and spectral deconvolution (discussed in section 4.3).

Multiphase C-(N,K)-A-S-H/(N,K)-A-S-H gels Coexistence of C-(N,K)-A-S-H and (N,K)-A-S-H gel frameworks occurs in many cement systems, including AAMs based on blends of high-calcium and low-calcium reactive precursors
Using 27 Al and 29 Si MAS NMR, the growth of an N-A-S-H gel within alkali silicate-PC-dehydroxylated halloysite blends was shown to preferentially consume available Si within the system through rapid reactions involving the initially dissolved silica [270]. 27 Al and 29 Si MAS NMR of AAMs based on GGBFS-metakaolin blends [266], combined with high-resolution X-ray diffractometry [272,273], has also shown significant phase coexistence in these binders, although the dissolution of Ca 2þ from the GGBFS, and subsequent reaction mechanisms, depends on both alkalinity and precursor chemistry.
Structural analysis via 29 Si, 1 H- 29 Si CP MAS, 27 Al, 23 Na and 1 H MAS NMR of multiphase C-(N)-A-S-H and N-A-S-H gel frameworks within synthetic AAMs has also revealed strong dependence of reaction kinetics, gel composition and structure on precursor composition [14]. This work demonstrated that increased Ca content of the solid precursor (or blend of precursors) promotes the formation of low-Al, high-Ca C-(N)-A-S-H with lower MCL, whereas increased Al content promotes Al inclusion and reduced crosslinking within C-(N)-A-S-H, formation of an additional N-A-S-H gel and increased presence of secondary alumina-rich hydrate phases.

Hydrated CACs
CACs comprising primarily CA (with smaller quantities of CA 2 and/or C 12 A 7 ) hydrate to form mixtures of the microcrystalline metastable phases CAH 10 and C 2 AH 8 , followed by conversion to the stable phases C 3 AH 6 and amorphous AH 3 [122,274]. The expected d obs , d iso and C Q for 27 Al in hydrated CACs are shown in Table 7 and discussed in the following text; information for other nuclei within these phases is not available in the open literature.
Skibsted et al. [52] used 27 Al MAS NMR to identify overlapping resonances attributed to octahedral Al environments in the stable   [20]. C 3 AH 6 (d iso ¼ 12.3 ppm, C Q ¼ 0.705 MHz) and amorphous AH 3 (d obs ¼ 8 ppm at B 0 ¼ 9.4 T) phases in CAC hydrated for 28 days; the isotropic chemical shift for C 3 AH 6 was determined via simulation of the spinning sidebands associated with this phase. Analysis of synthetic CAH 10 at multiple fields exhibited a broad resonance in the octahedral Al region of the 27  phase.
Because of the clear distinction between Al environments in hydrated (Al VI ) and anhydrous (Al VI ) CAC, the degree of hydration of this cement may be readily determined from 27 Al MAS NMR (Fig. 23) [44,50,51]. This, however, relies on the assumption that all Al resonances within the 27 Al MAS NMR spectra are observable and quantified correctly; this is not always the case due to the presence of 'NMR invisible' Al species (see section 4.3). 27 Al, 31 P and 1 H-31 P CP MAS NMR has also been used to show that modification of CAC by addition of phosphate prevents the loss of material performance which is caused by the conversion from the metastable calcium aluminate hydrates to stable C 3 AH 6 in hydrated CAC. Instead, phosphate modification yields an apparently stable amorphous calcium aluminophosphate gel comprising both Al IV and Al VI sites [277]. In these materials, phosphorus is present in hydrous environments with varying, but mostly low, degrees of crosslinking.

Hydrated CSA cements
Ye'elimite (C 4 A 3 $) hydrates in the presence of calcium sulphate (C$, added as gypsum or in dehydrated forms) to form ettringite ('AFt',C a 6 3 ), with the proportion of each phase dependent on the C 4 A 3 $/C$ ratio [278]. The presence of belite along with these phases in a CSA cement, which is common because of the prohibitive cost of Si-free Al sources for cement manufacture, results in the formation of str€ atlingite (2CaO$SiO 2 $Al 2 O 3 $8H 2 O; a silica-substituted AFm phase, see section 4.2.8.3) and/or C-S-H as additional reaction products. As mentioned in section 4.1.2.2.2, CSA clinkers can also contain C 5 S 2 $ (ternesite), which hydrates along with alumina-bearing clinker constituents to form str€ atlingite [279,280]. Most commercial CSA cements also contain C 4 AF (section 4.1.1.4) as the economically viable alumina sources used in cement manufacture also contain significant quantities of Fe; this will typically hydrate (in both CSA and PC cements) more slowly than the hydraulic aluminate phases and yields Fe-substituted forms of many of the same AFm and AFt hydrate phases [281e283].
SS NMR studies of hydrated CSA (synthetic and commercial) and CSA/PC blends have shown narrow 27 Al MAS NMR resonances for octahedral Al in ettringite and monosulphate at approximately d obs ¼ 13 ppm and 10.2 ppm (B 0 ¼ 14.1 T), respectively, as well as a broad low-intensity resonance at d obs ¼ 9 ppm at the same field which is attributed to Al(OH) 3 [133e135, 284,285], consistent with the resonances of pure phases in synthetic samples, section 4.2.4 [52,286,287]. As with CAC, the clear distinction between Al environments in hydrated (Al VI ) and anhydrous (Al VI ) CSA allows the degree of hydration of aluminium-containing phases to be readily determined from 27 Al MAS NMR. This has shown that aluminiumcontaining phases generally react to form hydrates within 24 h [133e135, 284]. Monitoring the degree of hydration in CSA/PC blends is significantly more difficult due to the presence of Al IV sites in C 3 A and C-(A)-S-H phases [133], as well as the presence of Fe which can severely dampen and shift the NMR signal of these nuclei [10]; however, it is made possible by spectral deconvolution using appropriate models and constraints (see section 4.3).

Magnesium silicate hydrate (M-S-H) gels
Magnesium silicate hydrate (M-S-H) cements are commonly produced from MgO and silica fume via hydrothermal or ambienttemperature processes [288e292], and generally comprise a talclike or serpentine-like phase [141,293], although there remains some debate around the most appropriate structural model for M-S-H. 29 Si MAS NMR spectra of M-S-H cements exhibit as many as five distinct 29 Si MAS NMR resonances, attributed to Q 1 ,Q 2 ,t w o Q 3 and Q 4 sites [142,146,292,294e298]. The major component of the 29 Si MAS NMR spectra is a Q 3 resonance at approximately d iso ¼À 93 ppm and the Q 1 and Q 2 resonances are typically broad [142,292], indicating the disordered nature of the gel. These environments have been attributed to Si sites within a disordered phase structurally similar to talc (Mg 3 Si 4 O 10 (OH) 2 ) [296] and/or serpentine-group minerals (polymorphs of Mg 3 (Si 2 O 5 )(OH) 4 , such as chrysotile, lizardite and antigorite) [142], with Mg/Si ratios ranging from 0.57 to 1.3 [291,293,298].

Magnesium potassium phosphate cements (MKPC)
Magnesium potassium phosphate cement (MKPC) is a clinkerfree acidebase cement which reacts to form struvite-K (MgKPO 4 $6H 2 O) as a crystalline main strength-giving phase; this phase is structurally analogous to the natural mineral struvite, NH 4 MgPO 4 $6H 2 O, but without requiring the presence of ammonium cations for its synthesis [141,300e302]. Application of SS NMR to investigate structural and phase evolution in MKPC is relatively recent, with the most extensive analysis probing 25 Mg, 27 Al, 29 Si, 31 P and 39 K nuclei in MKPC blended with FA and with GGBFS [303]. Struvite-K exhibits a 31 P resonance at d obs ¼ 6.2 ppm [303], which correlates well with the chemical shift of struvite, d obs ¼ 6.1 ppm [304]. A shoulder on this resonance at d obs ¼ 4 ppm was observed in MKPC/FA blends (but not MKPC/GGBFS blends) and was suggested to result from the presence of a disordered and/or partially substituted struvite-K phase. The predominant crystalline struvite-K phase exhibits 25 Mg and 39 K resonances at d iso ¼À 1.0 ppm (C Q ¼ 3.8 MHz) and d iso ¼À 73.1 ppm (C Q ¼ 2.2 MHz), consistent with previous observations for synthetic versions of this phase [305,306]. An amorphous orthophosphate environment was also identified, which has no intimate interaction with hydrogen and no measurable Al interactions [303].

Additional reaction products
Many additional reaction products are also formed along with the main silicate gels during hydration of the cements discussed previously and will be discussed in turn in this section. These additional reaction products are often intimately mixed with the dominant silicate gel frameworks [7,231] and can significantly influence the phase evolution and nanostructural development of these hydrate phases. The expected d obs , d iso and C Q for different nuclei in the additional reaction products observed in hydrated PC, CAC, CSA and magnesia-based cements, as well as cements based on alkali metal or alkali earth aluminosilicate chemistry, are shown in Table 8.

Portlandite.
Portlandite (CaOH 2 ), also called slaked lime, is a common reaction product in hydrated PC and is also present as a reaction product or remnant precursor in pozzolanic or slag-based binders activated with lime. In general, portlandite forms when the Ca/Si ratio is significantly higher than can be accommodated by C- S-H or related phases [307]. It consists of layers of octahedrally coordinated Ca bound to oxygen atoms in tetrahedral coordination [23] with hydrogen occupying a single crystallographic site. It is a key participant in longer term phase evolution of C-(A)-S-H gels in blended PC cements via the pozzolanic reaction [308]. High-field natural abundance 43 [180,182,310]. Chloride sorption onto portlandite in portlandite suspensions has been studied using static 35 Cl NMR experiments and measurement of 35 Cl spinelattice relaxation time constants (T 1 ) and spinespin relaxation time constants (T 2 ) [218]. Chloride on portlandite exists predominantly in a hydrated, solution-like chemical environment and is rapidly exchanged between the surface and bulk solution. Portlandite quantification in hydrated cements by NMR is not usually attempted, as neither 43 Ca nor 17 O nuclei are particularly straightforward to use in such studies, and other techniques such as thermogravimetry and X-ray diffraction (XRD) can give reliable quantitative results for this phase much faster and less expensively.

'Aluminate ferrite trisulphate'etype phases.
Ettringite is the archetypal 'aluminate ferrite trisulphate' (AFt) phase, with a columnar hydrous CSA structure and a very high water content (up to 32 water molecules per formula unit) [311]. Ettringite forms rapidly in the earliest stages of hydration of PC, as C 3 A reacts with gypsum, or equivalently through hydration of ye'elimite with excess gypsum in CSA cements. Early 27 Al MAS NMR experiments (9.4 T) examining ettringite displayed a resonance at d iso ¼ 13 ppm that appeared characteristic of a single octahedrally coordinated AlO 6 site [52], which contrasted with the then-proposed crystal structure of ettringite which contained two distinct octahedral AlO 6 sites [312]. Recent work using ultraehigh-field (22.3 T) 27 Al MAS and MQMAS NMR (Fig. 24) has in fact been able to resolve two distinct octahedrally coordinated AlO 6 sites in synthetic ettringite [137], consistent with the now well-accepted crystal structure solved by Moore and Taylor [312]. The two AlO 6 sites resonate at d iso ¼ 13.08 and 13.51 ppm and are highly symmetric, displaying very small quadrupolar coupling parameters of C Q ¼ 0.391 MHz and 0.337 MHz, respectively. The high symmetry and very similar isotropic chemical shifts of these resonances are due to very small variations in Al-O bond lengths between sites, consistent with high-resolution powder XRD [313], single-crystal XRD and DFT calculations [311]. As a consequence of the small distinction in isotropic chemical shift values, these resonances remained unresolved for more than two decades in 27 Al MAS NMR spectra acquired at lower fields [22,52,57,117,135,186,187,314,315]. This demonstrates the importance of acquiring high-field spectra for quadrupolar nuclei in complex phases, even where crystallinity is high (as in the case of ettringite) and for relatively 'accessible' nuclei such as 27 Al. Natural abundance 43 Ca MAS NMR analysis of ettringite also resolved only a single 43 Ca resonance (d obs ¼ 8 ppm) [40] despite the existence of two Ca sites in the crystal structure; this can be attributed to the structural similarity between the two Ca sites in ettringite and the poor signal-to-noise ratio of the spectrum resulting in poorly defined overlapping quadrupolar lineshapes for each site. High-field natural abundance 33  AFm structure, which is also given the mineral name str€ atlingite [154,324]. Substitution by multiple anions in ordered positions is also possible, for example, 1:1 hydroxide and carbonate to form hemicarbonate or 1:1 sulphate and chloride to form Kuzel's salt [322]. Friedel's salt (3CaO$Al 2 O 3 $CaCl 2 $10H 2 O) is also observed in hydrated PC, which has been exposed to a source of chloride, e.g. when serving in marine environments or cold climates with road salting. These phases each exhibit a single 27 Al MAS NMR resonance at approximately d obs ¼ 10e12 ppm, depending on field strength [52,117,133], in accordance with their crystal structures containing Al VI [327,332]. Owing to similarities in structure, each of these AFm phases exhibits very similar 27 Al MAS NMR resonances at standard field strengths, and so they are more often distinguished from each other via complementary techniques, e.g. XRD, where their different basal spacings can be observed. The monosulphate and monocarbonate 27 Al MAS NMR resonances are severely overlapping (d obs ¼ 12 ppm), but evidence for more than one peak in this region has been observed by ultraehigh-field 27 Al MAS NMR (23.3 T) [137], suggesting that it may be possible to resolve these resonances, and those of C 4 AH 13 , at ultrahigh field. The sorption of chloride onto the monocarbonate AFm phase C 4 ACH 11 was investigated by static 35 Cl NMR experiments and measurement of 35 Cl spinelattice relaxation time constants (T 1 ) and spinespin relaxation time constants (T 2 ) [218], showing that chloride is in rapid exchange between adsorbed surface sites and bulk solution sites in each phase and exists predominantly in a hydrated, solution-like chemical environment on the monocarbonate surface. This behaviour is similar to chloride sorption onto portlandite and onto jennite, as discussed previously. Str€ atlingite (Ca 4 Al 2 (OH) 12 [AlSi(OH) 8 ] 2 $2H 2 O) is an AFm phase which forms as a hydration product of aluminium-rich cements including PC-metakaolin blends and some non-Portland binders. 27 Al MAS NMR analysis of hydrated white PC-metakaolin blends [117] and hydrated CSA [137], as well as synthetic str€ atlingite [333], revealed two resonances at d obs ¼ 61 ppm and 11.3 ppm (14.1 T) attributed to Al IV and Al VI sites, respectively. These resonances are characteristic of the crystal structure of str€ atlingite [334], which contains Al IV in a double-tetrahedral environment as part of the aluminosilicate interlayer anion ([(T,,) 4 (OH,O) 8 $0.25H 2 O] À ,T¼ Si or Al, , ¼ vacancy), as well as the Al VI in the brucite-type layer ([Ca 2 Al(OH) 6 $2H 2 O] þ ) that defines the AFm structure. The Al VI site in str€ atlingite resonates at a similar frequency to Al VI sites in monosulphate, and consequently str€ atlingite is more conclusively identified from its Al IV resonance, unless working at extremely high fields. The 27 Al MAS NMR spectrum of hydrated CSA collected at B 0 ¼ 22.3 T appears to contain contributions from several individual overlapping Al VI resonances [137], suggesting that monosulphate, monocarbonate and str€ atlingite may be able to be resolved in 27 Al MAS or MQMAS NMR spectra acquired at ultra-high magnetic field.
The 29 Si MAS NMR spectrum of synthetic str€ atlingite exhibits a broad resonance at approximately d obs ¼ 86.5 ppm, containing contributions from resonances resulting from Q 3 (3Al), Q 2 (2Al), Q 2 (1Al) and Q 2 sites [333,335], consistent with the crystal structure [334] of str€ atlingite which contains double tetrahedral silicate ring structures. The presence of these resonances is often used to monitor hydration in PC-CSA blends [133]. As mentioned previously, Friedel's salt is an AFm phase with Cl À as its interlayer ion. Variable temperature static 35 Cl NMR has shown that interlayer Cl À in Friedel's salt exhibits a resonance at d iso ¼ 30 ppm with a quadrupolar coupling constant C Q ¼ 2.22e2.87 MHz, indicating uniaxial symmetry above 0 C resulting from dynamically averaged interlayer species [336]. Below 0 C, the Cl À in this site exhibits reduced (triaxial) symmetry resulting from a rigid crystal structure, as indicated by a resonance at d iso ¼ 26 ppm with a quadrupolar coupling constant

Third aluminate hydrate.
The 'third aluminate hydrate' (TAH) phase is an amorphous nanoscale aluminate hydrate phase which precipitates at the surface of the C-S-Hetype gels formed during PC hydration [172,194,337] and which has also been noted in sodium silicateeactivated GGBFS [18,20,155,324]. TAH is to date only identifiable by NMR, and consequently, the technique is key to the discovery of its existence. TAH is described as a poorly ordered Al(OH) 3 phase [194,337] which is intimately mixed with other hydrate phases. Stoichiometric arguments and thermodynamic modelling suggest that the presence of TAH is likely to be linked to high concentrations of available Ca and Al [154,338,339], as the opposing charges on CeSeH and AFm phases may produce strong mutual attractions and physically destroy the AFm crystals [22,340], which would then result in low-crystallinity aluminate (AFm-like) layers being dispersed throughout the CeSeH gel while being undetectable by XRD [22]. This hypothesis is supported by transmission electron microscopy observations [337]. TAH phases exhibit a 27 Al MAS NMR resonance at d obs ¼ 5 ppm (14.1 T) in the spectra of C-S-Hetype gels [172,194,337] and of sodium silicateeactivated GGBFS [18,20,155,324]. Broadening of the 29 Si MAS NMR resonance attributed to Q 1 Si species in sodium silicateeactivated slags [155] is attributed to the presence of Al VI -O-Si linkages between TAH and C-S-H.

Hydrotalcite-group phases.
Hydrotalcite (Mg 6 Al 2 CO 3 (OH) 16 [CO 3 ]$4H 2 O), Fig. 26) is an Mg-Al LDH mineral with a primary layer of positively charged brucite-like sheets (see section 4.1.2.2.3), comprising octahedrally coordinated Mg bound to oxygen atoms in tetrahedral coordination, with partial replacement of Mg 2þ by trivalent octahedrally coordinated cations, which are Al in the case of true hydrotalcite [341]. The Al atoms are randomly distributed within these layers at low concentrations [342] but take on a significantly more ordered configuration at the higher concentrations which are more characteristic of hydrotalcite-group phases observed in cements [23]. A secondary layer contains H 2 O molecules and anions, most commonly carbonate, CO 3 2À ,a n dh y d r o x y l ,O H À ,w h i c ha r e intercalated and charge balance the net positive charges induced in the sheets by incorporation of trivalent cations to replace Mg 2þ [341,343].M o r et h a n4 0i d e n t i fied mineral species conform to this general description (but with different cation and anion substituents and degrees of substitution) and are collectively denoted hydrotalcite-supergroup phases. Quintinite-group phases, which are a subset of the hydrotalcitesupergroup with Mg/Al ¼ 2, have been observed on a submicrometre scale in hydrated PC-GGBFS blends [99,220,344] and in cements produced from metallurgical slags activated with sodium silicate or sodium hydroxide [71,72,154,190,324,325,345e348], typically occurring when slags with Mg content ! 5% are used, consistent with predictions from thermodynamic modelling [349]. These phases generally contain higher Al content (Mg/Al ¼ 2) than true hydrotalcite (Mg/Al ¼ 3) [341]. Previous work identified a phase that was denoted 'M 4 AH 13 ' and proposed that it may be an analogue of the AFm-structured C 4 AH 13 , however, this phase has since been shown to be OH-quintinite (Mg 4 Al 2 (OH) 13 $4H 2 O)) [350,351]. 27 Al MAS NMR spectra of sodium silicateeactivated slag cements exhibit a resonance at d obs ¼ 9 ppm (9.4 T) attributed to a hydrotalcite-like phase containing Al VI sites [325,353], consistent with 27 Al MAS NMR analysis of synthetic hydrotalcites [342,353]. Simulation of 27 Al MAS NMR spectra of synthetic hydrotalcitegroup phases has determined the isotropic chemical shift of this site to be d iso ¼ 11.8 ppm, attributed to a single Al environment surrounded by Mg atoms in octahedral coordination [354], with broadening of this resonance indicating increased disorder as Al content is decreased. A broad shoulder on this resonance at d obs ¼ 3 ppm (8.45 T) is observed when the intercalated anion is CO 3 2À [354]. High-field (19.6 T) 25 Mg MQMAS NMR of these hydrotalcite-group phases resolved a single 25 (1-x) ) 2 (SiO 4 ) y (OH) 4(3-y) ;0 x 1 and 0 y 3) describes a group of garnet minerals where the (SiO 4 ) 4À tetrahedra are partially or completely replaced by hydroxyl ions. In PC hydrated at ambient temperatures, the formation of siliceous hydrogarnet is minimal [331,356]; however, formation of these phases has been observed in PC hydrated at higher temperatures [357e360] or in the presence of excess Fe species [357,361], and water-rich members of the grossular -katoite hydrogrossular series (Ca 3 Al 2 (SiO 4 ) y (OH) 4(3-y) ; 0 y 3) are major hydration products in CACs [52].
Katoite (Ca 3 Al 2 (SiO 4 ) y (OH) 4(3-y) ;0 y 1.5, often denoted C 3 AH 6 ) exhibits a 27 Al MAS NMR resonance with an isotropic chemical shift of d iso ¼ 12.4 ppm and a quadrupolar coupling [52,362] due to octahedral Al surrounded by six hydroxyl groups in the hydrogarnet structure [51]. Increased substitution of silica into this structure results in a 27 Al MAS NMR resonance at approximately d iso ¼ 4e6 ppm, with this resonance Fig. 26. Illustration of the local environments in the crystal structure of hydrotalcite (Mg 6 Al 2 CO 3 (OH) 16 [CO 3 ]$4H 2 O) obtained from single-crystal diffraction data [352]. Mg, Al, C, O and H atoms are represented by orange, blue, red and light pink spheres, respectively. Partial colouring of spheres indicates the average proportion of each respective atom occupying a specific site in the hydrotalcite crystal structure. Prepared using the VESTA software package [128].
shifting towards lower frequencies and exhibiting increased quadrupolar broadening with increasing silica content [50,362]. 27 Al MAS NMR resonances attributed to katoite have been observed in CACs [50,52] and alkali-activated slags [20,155,197]. 29 Si MAS NMR analysis of hydrated CAC attributed a broad resonance at approximately d iso ¼À 79.9 ppm to Si sites in [Si(OAl VI ) 4-x ] (OCa) x species (classified as Q 0 in the nomenclature of Fig. 1 as there are no bonds to tetrahedral species; all of the Al substituents are octahedral) and used spectral deconvolution to quantify this phase in the hydrated CAC [50].

Deconvolution strategies and pitfalls related to spectra of cementitious materials
Mathematical deconvolution of MAS NMR spectra allows resolution of individual resonances which contribute to the overall spectra and can enable quantification of the resonating species [60]. The simplest case in which deconvolution can be performed is for non-quadrupolar (spin S ¼ ½) nuclei, the spectra of which can be deconvoluted using a series of Gaussian, Lorentzian or mixed (Voigt or pseudo-Voigt) peaks to simulate individual resonances using a least squares fitting method [363]. This is performed routinely in the study of anhydrous and hydrated cements and related phases.
The most appropriate lineshape to simulate the individual resonances for spin S ¼ ½ nuclei depends on the ordering and dynamics of the material under investigation [364,365]; for ordered and rigid SS systems, a Gaussian (statistical) distribution best represents the resonance lineshape as interatomic distances are generally constant. For highly dynamic systems (e.g. solutions or gases), a Lorentzian distribution (based on collision theory) best represents the resonance lineshape as mobility is high [365]. Consequently, resonance lineshapes for spin S ¼ ½ nuclei in dynamic SS systems may be best represented by Voigt or pseudo-Voigt distributions weighted heavily towards a Gaussian profile. In reality, dynamics of solids in cementitious systems are extremely slow relative to those in solution, and a Gaussian distribution provides a suitable approximation for the resonance lineshapes.
Deconvolution of the spectra of quadrupolar nuclei (S > ½)i s also possible; however, quadrupolar interactions must be taken into account when simulating the spectra (Fig. 27). Quadrupolar parameters can be estimated [21]; however, owing to the large variation in NMR parameters across samples and instruments, this approach is prone to error must be appropriately justified. It is often more appropriate to determine the quadrupolar parameters directly.
Quadrupolar interactions can be determined by acquiring single-pulse MAS NMR spectra at multiple fields and simulating the NMR resonances observed for the central transition at each field with a consistent set of NMR parameters (d iso ,C Q and the asymmetry parameter h) [30,172]; this is often the most readily implementable approach if instruments at multiple fields are available as it does not require complex experiments or fitting approaches.
Quadrupolar parameters can also be determined (along with d iso and h) at a single field by simulation of manifolds of spinning sidebands observed for satellite transitions in MAS NMR spectra [367e369], typically across spectral widths on the order of 2 MHz. This method can be computationally expensive relative to simulation of single-pulse MAS NMR spectra.
Two-dimensional experiments such as MQMAS NMR techniques [242,255,370] are also used to determine d iso ,C Q and h, and thus aid in deconvolution of NMR spectra for quadrupolar nuclei. MQMAS achieves high resolution by conversion of symmetric MQ coherence (pQ) to the single quantum (1Q) detectable central transition and plotting pQ and 1Q correlations in two dimensions [371]. Consequently, chemical shift anisotropy is removed from the MQ dimension, and d iso can readily be determined from the two dimensional MQMAS spectrum. C Q and h can then be determined by simulating the two-dimensional MQMAS spectrum or the pQ dimension of this spectrum using readily available software (e.g. DMFit [60] or Quadfit [372]), information which can in turn be used to simulate the single-pulse MAS NMR spectrum. A suitable model which incorporates these interactions should be used, e.g. the Czjzek model [373]. When performing any spectral deconvolution, the minimum possible number of constituent subpeaks should be used to enable an accurate and meaningful interpretation of the spectra [363]. Subpeak widths should be set in a rational manner, considering the expected disorder within each site type, and their positions must be consistent with literature data for specific site types in pure materials. The relative deconvoluted subpeak intensities should be consistent with a mass balance for all elements, the structural constraints dictated by data from additional analytical techniques (e.g. quantitative XRD) and those described by relevant structural models, e.g. the 'substituted general model' for C-S-H gels [159], the 'crosslinked substituted tobermorite model' for C-(N)-A-S-H gels [195] and the thermodynamics of a statistical distribution of Si and Al sites within a Q 4 aluminosilicate network for N-A-S-H products [250].
To provide an accurate representation of binder chemistry, and considering that the conversion of solid cementitious phases to hydrate products is almost never fully complete in a practical cementing binder (even after decades or more), the contribution of the remnant unreacted precursor to the spectra must be taken into account [13]. This can be achieved either through selective isotopic labelling of the binder [371],b y subtraction of a scaled component spectrum for the remnant unreacted precursor during deconvolution if congruent dissolution is assumed [18,20,345] (Fig. 28)o rb yu s i n gt h es p e ctrum of a leached residue (obtained via selective dissolution) to represent a residual precursor component if congruency is uncertain [315,345]. For quantitative analysis, the congruency and kinetics of dissolution and reaction processes involving cementitious precursor materials must be carefully considered and modelled appropriately [14,20] as the underlying assumptions can dramatically affect the final deconvolution. This can be achieved by a variety of methods, including those used to measure the degree of reaction of SCMs in PCeSCM blends [374]. Methods include selective dissolution [375],s c a n n i n g electron microscopy [323,376], thermogravimetry [374] and XRD with iterative Rietveld refinement (e.g. the 'partial or no known crystal structure',P O N K C S ,m e t h o d [377])a p p r o a c h e s . In a practical sense, GGBFS is a highly depolymerised glass that is likely to dissolve close to congruently [325], whereas FA is a much more complex mineral assemblage that dissolves incongruently [101,102].
Quantification of individual resonances can often be complex, even with seemingly simple spectra of non-quadrupolar nuclei such as 29 Si (see the following section). Quantification of 27 Al resonances can be particularly difficult due to the presence of Al species which exhibit resonances so broad that they are often undetectable. 'NMR invisible' Al species have been described, largely in older literature, in zeolites, minerals and gels; up to 30% of all Al species in some zeolites could not be observed at 9.4 T [378]. However, these resonances were detectable at high field (18.8 T) [378], highlighting the importance of acquiring SS NMR spectra at the highest field possible if quantification is desired.

Complexities in quantification of tricalcium and dicalcium silicate in PC
The 29 Si spinelattice relaxation time constants (T 0 1 )f o r belite are approximately 30 times those of alite at the same magnetic field [29], and consequently, it is crucial to ensure that sufficient recycle delays are used during NMR data acquisition to allow for complete relaxation of both alite and belite. This can be achieved via inversion-recovery 29 Si spinelattice relaxation MAS NMR [27,29] (e.g. Fig. 29)o rT 0 1relaxation-time-filtered 29 Si MAS NMR [57]. Deconvolution of such data allows independent quantification, and hence validation, of the relative proportions of alite and belite within PC [27,29]. This work has shown that the determination of alite and belite content in PC using traditional Bogue calculations [379] strongly overestimates the belite content at the expense of alite, consistent with results from XRD and other techniques; the modified Taylor-Bogue calculation [380] provides a much better match to the alite/belite ratio determined by 29 Si MAS NMR spectral deconvolution [27].

DFT computation of NMR spectra and parameters
Isotropic chemical shifts, coupling constants and electric field gradients of molecules can also be predicted ab initio from calculations using DFT [381,382] and used to simulate NMR spectra. Computation of NMR parameters from DFT is widely used to determine the crystal structure and ordering in ceramics, zeolites and other crystalline materials; however, there has been little application of this approach to the phases in cementitious materials. Rejmak et al. [165,166] performed simulations of various C-S-H gel structures based on 14 Å tobermorite and jennite models and calculated the resultant isotropic 29 Si chemical shifts. Calculated chemical shifts for Q 2 sites were found to be largely independent of the structural model used, and Q 1 and Q 2 sites each showed a large dispersion of calculated chemical shifts. The calculated 29 Si MAS NMR chemical shifts were found to be in good agreement with those of 14 Å tobermorite. This work also showed that terminal silicate sites (Q 1 ), and paired silicate tetrahedra with an AlO 4 unit within the first coordination sphere (Q 2 (1Al)), which exhibit similar isotropic chemical shifts, may be distinguished easily by large differences between their chemical shift anisotropies [166]. Ab initio prediction of NMR parameters from calculations using DFT is yet to be applied extensively to cementitious materials and represents an important opportunity for further advancement. In particular, combining DFT predictions with experimental data acquired at high field of synthetic model systems will reveal important new structural insights which may then be used to untangle many remaining questions in more complex, heterogeneous systems.

Spin-echo MAS NMR
MAS NMR spectra containing broad resonances, such as those of disordered cementitious systems, can experience artefacts and distortion introduced in the first few microseconds of instrumental dead time at the beginning of the free induction decay [10]. Spinecho pulse sequences [383] can be used to overcome this by refocussing the spin system. The spin system is given an initial nonselective (90 ) pulse, after which the spins are allowed to precess and dephase (due to the varying precession rates of individual nuclear moments within the system). Applying a 180 pulse after a time t flips the spin system and restores phase coherence to the precession at a time 2t after the first 90 pulse, refocussing the spin system [10,383]. A small number of authors have used spin-echo pulse sequences analysing cementitious materials to accurately quantify the sites present in these systems [8,260,299,384].

Cross polarisation MAS NMR
CP MAS NMR involves transfer of magnetisation from a nucleus with more abundant spins to a nucleus with less abundant spins, allowing acquisition of the NMR spectrum of the less abundant nucleus with a much higher signal-to-noise ratio than would be obtained by MAS NMR [10]. This is achieved by irradiation of the two nuclei at their correct Larmor frequencies in fulfilment of the HartmanneHahn condition [385]. The signal from the less abundant nucleus in closest proximity to the more abundant nucleus is preferentially enhanced, and so additional structural information can be gained by comparison of CP MAS and MAS spectra of the same sample. 1 H is the most common high-abundance nucleus in cementitious materials, and consequently, this technique can be particularly advantageous for probing local environments in cementitious materials where the structural role of water is of interest. 29 Si CP MAS was used to examine hydration of b-C 2 S and show that the initial hydration products contain monomeric silicate hydrates, the amount of which determines the initial hydration rate [386], while 31 P MAS and 31 29 Si CP MAS NMR (Fig. 30) has also been used to identify Si species closely associated with water (i.e. Si-OH linkages) that are likely to exist as part of the C-S-H gel [43,167,185,386e392], C-(N)-A-S-H gel or AFm phases in alkali-activated slag [154,196], metakaolin [242] and synthetic C-(N)-A-S-H gels [14], while 1 H- 27 Al CP MAS NMR has also been used to identify Al-OH linkages in synthetic analogues of PC hydration products [393]. 13 C MAS and 1 H-13 C CP MAS NMR have also been combined to study carbonation of synthetic C-S-H and C-A-S-H gels [221] and to determine the NMR parameters of inorganic carbonates relevant to cement chemistry [220]. During early studies of TAH, 1 H- 27 Al CP MAS NMR revealed Al(OH) 6 3À units closely associated with C-S-H, leading to the identification of the TAH as an amorphous aluminate hydroxide or calcium aluminate hydrate produced either as a separate phase or a nanostructured precipitate on the surface of the C-S-H gel [194]. 1 H-29 Si and 1 H- 27 Al CP MAS NMR experiments are also particularly useful for constraining 29 Si and 27 Al MAS NMR spectral deconvolutions of the C-S-H model system tobermorite and C-A-S-H gels [14,393].

X-Y CP MAS.
Studies using X-Y CP MAS NMR experiments (that is, those CP experiments which do not probe protons) to study cementitious materials are fewer; however, they have still revealed new insight into the interactions in these materials. 17 O- 27 Al CP MAS NMR has been used to support a claim that Al atoms within GGBFS are connected only to bridging oxygen atoms and not nonbridging oxygen atoms [75]. 19

Multiple quantum MAS NMR
Recently, MQMAS spectroscopy [394,395], probing half integer quadrupolar nuclei including 17 O, 23 Na, and 27 Al, has been used to obtain high-resolution NMR spectra of synthetic C-S-H [188],C A C [276], AAM [113,242,252,256,276,370,371,396], related reaction products [137,276,354,355,397] and precursors [34,36,75,138]. MQMAS achieves high resolution by conversion of symmetric MQ coherence (pQ) to the single quantum (1Q) detectable central transition, and plotting pQ and 1Q correlations in two dimensions [371]. Consequently, chemical species with similar local structures and coordination environments, which would otherwise result in overlapping resonances in one-dimensional MAS NMR spectra, may be resolved. 27 Al 3Q MAS NMR has been used to examine symmetry of Al sites within synthetic calcium aluminate hydrates and to obtain the isotropic chemical shift, quadrupolar coupling constant and asymmetry parameter of these sites [276]. 27 Al 3QMAS NMR spectra of these materials resolved additional Al sites to those that could not be obtained by 27 Al MAS NMR alone, including two Al sites within C 4 AH 13 [276]. 27 Al 3Q MAS NMR has also been used to probe Al substitution in C-S-H gels [276,393], identifying two distinct tetrahedral Al sites that were attributed to Al substitution for Si into both bridging and non-bridging positions [276], although other studies (as discussed in section 4.2.1.1) have shown that bridging positions are a strongly preferred environment. 27 Al 3QMAS NMR has provided additional insight into coordination and symmetry of Al environments within N-A-S-Hetype gels ( Fig. 31a  and b), including determination of the isotropic chemical shift (d iso ), quadrupolar coupling constant (C Q ) and asymmetry (h) parameter of the electric field gradient tensor H- 29 Si CP (black) MAS NMR spectra of a synthetic calcium aluminosilicate precursor and of a synthetic alkali-activated cement produced from it (containing C-(N)-A-S-H), cured for 3 days. All spectra are normalised to constant total intensity. Adapted from Walkley et al. [14]. (c) 1 H- 27 Al CP-MAS NMR of tobermorite. Adapted from Houston et al. [393]. [113,242,255,256,370,371,396] and also Al environments in C-A-S-H/N-A-S-H gel blends [396]. Both 27 Al 3QMAS NMR [276] and 25 Mg 3QMAS NMR [354] have also been used to resolve individual Al and Mg sites within Mg-containing LDHs, while recent work used 27 Al MQMAS NMR to resolve for the first time 8 distinct Al sites in ye'elimite [34,138] and high-field 27 Al MQMAS NMR to resolve for the first time two distinct octahedrally coordinated AlO 6 sites in synthetic ettringite [137] (see section 4.2.8.2 'Aluminate ferrite trisulphate'etype phases). MQMAS NMR, therefore, appears to offer significant scope for future advances in the understanding the atomic structure of cements. Processing of MQMAS spectra using standard single-axial isotropic shearing (iso-shearing) in the indirect dimension removes the second-order quadrupolar term and leaves the projection of the spectra onto the indirect dimension axis as a pure function of isotropic variables such as the isotropic chemical shift (d iso ) and the quadrupolar-induced shift (d QIS ), allowing chemically distinct sites to be resolved in the indirect (pQ) dimension [242,371,398,399]. The more recently introduced method of biaxial Q-shearing [371] in both the indirect and direct dimensions allows orthogonal separation of d iso from the quadrupolar parameters d QIS and the quadrupolar coupling constant (C Q ) and is particularly useful for disordered materials which typically give MQMAS spectra displaying a distribution of chemical shift and quadrupolar parameters, that can be difficult to interpret for identification of chemically distinct sites [113,371] (Fig. 32) to identify bridging hydroxyl groups connecting Al IV and Si IV tetrahedra in a series of alkali-activated metakaolin cements, supporting the identification (see section 4.2.2) of charge balancing extra-framework Al species in these systems. 29 Si{ 1 H} REDOR 3Q MAS NMR has also been applied to a series of aluminosilicate geopolymers [370] to show that these materials contain a heterogeneous distribution of charge-balancing Na ions and associated water molecules, with the Na ions and water molecules preferentially clustered around fully polymerised Q 4 (4Al) Si species.
Recently, Greiser et al. [409] used 29 Si{ 27 Al} TRAPDOR NMR, 27 Al { 29 Si} and 27 Al{ 1 H} REDOR NMR experiments to reveal new information regarding the phase assemblage and nanostructure of cementitious sodium aluminosilicate gels produced via the onepart alkali-activation route using variety of silica sources. This work showed the extent of formation of an amorphous sodium   27 Al MQMAS NMR spectrum (red contours) of a sodium silicateeactivated metakaolin cement. Adapted from Brus et al. [242]. The main resonance results predominantly from AlO 4 À tetrahedra charge balanced by Na þ ions, and the shaded orange and green areas indicate regions where AlO 4 À tetrahedra connected to SiO 4 tetrahedra via bridging hydroxyl groups (Al IV e OH þ e Si IV ) and AlO 4 À tetrahedra charge balanced by extra-framework Al species are expected to resonate, respectively. aluminosilicate gel, coprecipitation of an amorphous hydrous alumina gel (Al(OH) 3 ) and formation of zeolitic phases are dependent on the silica source, despite very similar Si/Al of the starting reaction mixtures. The identification of amorphous Al(OH) 3 has not been previously reported and demonstrates the importance of using double-resonance NMR techniques to resolve multiple overlapping 27 Al resonances that are often present in single-pulse 27 Al MAS NMR spectra (Fig. 33).
Tran et al. [396] used a 29 Si{ 27 Al} REAPDOR NMR technique to distinguish between Si-O-Al connectivity environments in alkaliactivated metakaolin and to identify the number of Al atoms substituted in the second coordination sphere of each probed Si site. The same authors also used 29 Si and 27 Al MAS, 19 27 Al REAPDOR NMR has also been used to demonstrate the existence of aluminiumephosphorus interactions within phosphate-modified CACs [277], where no other analytical technique was able to provide direct evidence about the existence (or otherwise) of these bonds within the disordered gel systems of interest. The universal REAPDOR curve [410] which is commonly used to fit REAPDOR NMR data and determine internuclear distances is, however, specifically designed for an isolated spin pair (i.e. monodentate structures) which limits its applicability (in a quantitative sense) to the polydentate structures which comprise most cements.

2D homonuclear correlation MAS NMR spectroscopy.
2D homonuclear correlation MAS NMR spectroscopy (COSY) is a multiple resonance MAS NMR experiment involving transfer of magnetisation between two nuclei of the same isotope and is useful for determining connectivity between different sites when many resonances are present. Double quantum 29 Si- 29 Si COSY experiments on synthetic C-S-H isotopically enriched with 29 Si has resolved both dimeric (Q 1 -Q 1 ) and chain end group (Q 1 -Q 2 ) Si sites as well as (Q 3 -Q 3 ) and (Q 3 -Q 2 ) bridging linkages [176], confirming the presence of these sites in C-S-H as discussed in section 4.2.1.

2D heteronuclear correlation MAS NMR. 2D HETCOR MAS
NMR is a multiple resonance MAS NMR experiment which involves transfer of magnetisation between two heteronuclei and is useful for determining connectivity between different nuclei when many resonances are present. This experiment is similar to the X-Y CP MAS NMR experiment; however, in the X-Y 2D HETCOR experiment, the magnetisation of nucleus Y is allowed to evolve for a (variable) time t 1 prior to being transferred to nucleus X, whose response is measured directly during the detection period t 2 [10]. Significant sensitivity enhancement can be gained by combining MQMAS with the HETCOR experiment (HMQC) [411]. Rawal et al. [57] used 1 H-29 Si and 1 H- 27 Al HETCOR MAS NMR with different 1 H spin-diffusion times to probe correlations between different molecular moieties in hydrated white PC: both short (10 ms, probing strongly coupled species with internuclear distance < 1 nm) and long (30 ms for 29 Si and 20 ms for 27 Al, probing more weakly coupled species over several nm), Fig. 33. 1 H- 29 Si HETCOR MAS NMR experiments were able to resolve chemically bonded -OH groups, adsorbed water and strongly hydrogen bonded -OH groups associated with Q 1 ,Q 2 and Q 2 (1Al) Si sites, with chemically bonded -OH groups and adsorbed water primarily associated with Q 1 and Q 2 (1Al) sites, as well as water strongly adsorbed to six-coordinate Si sites within a thaumasite-like phase that was claimed to be present in the hydrated PC [57], although the identification of this mineral in cements that have not been subjected to chemical damage is extremely uncommon.
Resonances from protons in H 2 O, Ca-OH and Si-OH groups have also been identified by 1 H- 29 Si HETCOR MAS NMR analysis of synthetic C-S-H [176], with both types of hydroxyl sites correlating with all Si sites in C-S-H and water molecules correlating with Q 1 and both bridging and pairing Q 2 sites. 1 H- 27 Al HETCOR MAS NMR experiments resolved hydroxyl groups chemically bound to Al VI sites within ettringite and TAH and an absence of correlation between 1 H species and four-coordinate 27 Al IV moieties. Enhanced resolution provided by the 1 H-29 Si and 1 H- 27 Al HETCOR MAS NMR experiments allowed resolution of 29 Si-OH and 27 Al-OH bonds which were not resolved in single-pulse MAS NMR experiments.   27 Al{ 1 H} REDOR NMR spectra (S, dashed line) overlaid with 27 Al MAS NMR spectra (S 0 , black line) and difference spectra (DS ¼ S 0 e S, grey or blue lines) of a cementitious sodium aluminosilicate gel produced via one-part alkaliactivation of rice husk ash. Adapted from Greiser et al. [409]. The AlO 6 resonance (d obs ¼ 8 ppm) does not show a discernible 27 AlÀ( 29 Si) REDOR NMR effect and shows a strong 27 AlÀ( 1 H) REDOR interaction, demonstrating that this resonance is due to AlO 6 within a hydrous alumina gel (Al(OH) 3 ) with little or no Si in proximity.
interpretation of SS MAS NMR spectra. Determination of these parameters from single-pulse MAS NMR spectroscopy observing the central transition (m S ¼ ± ½) between nuclear spin states, however, is hindered due to the second-order quadrupolar shift. Satellite transition spectroscopy involves observing the satellite transitions in MAS NMR spectra of quadrupolar nuclei, where residual second-order quadrupolar effects under MAS (quadrupolar broadening and the second-order quadrupolar shift) are reduced relative to the central transition [367]. The dependence of the second-order quadrupolar shift for a particular spin S on the nuclear spin state transition (m) results in a different net isotropic chemical shift for each m; consequently, by simulating all observed transitions, it is possible to determine the isotropic chemical shift from a single spectrum [367e369]. Jakobsen et al. [412] and Skibsted et al. [368] demonstrated simulation of all transitions in MAS NMR spectra of crystalline powders to accurately determine C Q and h for 17 O, 23 Na and 27 Al in systems with large C Q values (C Q > 2 MHz), as well as applying this method to accurately determine the 27 Al C Q value of 1.9 MHz for Al VI in hydrated white PC [44]. Skibsted et al. [52] also applied this method to a series of calcium aluminate phases in CAC and in PC (Fig. 34), determining 27 Al d iso ,C Q and h values and distributions. These were then used to describe linear relationships between C Q and the mean deviation of bond angle from ideal tetrahedral symmetry, as well as between C Q and a calculated estimate of the geometrical dependency of the EFG tensor. This method has also been used to identify the two crystallographically distinct octahedrally coordinated Al sites in kaolinite [413], as specified by its crystal structure [414]. 1 H is the most sensitive nucleus in cementitious materials which may be probed using NMR; however, 1 H-1 H dipolar couplings cause severe line broadening which limits the resolution of 1 H MAS NMR spectra, which is further compounded by the small chemical shift range (approximately 20 ppm) over which 1 H nuclei in solid phases resonate. These limitations have impeded the widespread application of single-pulse MAS NMR experiments probing 1 Hi n cementitious materials. However, 1 H relaxation rates (spinelattice, 1/T 1 , and spinespin, 1/T 2 ) are sensitive to proton mobility in hydration products and water in the vicinity of solideliquid interfaces and can therefore provide extensive information regarding porosity, pore size distributions and tortuosity in these cementitious materials.

1 H relaxation NMR
Early work used proton NMR relaxometry to monitor hydration of C 3 S and PC by measurement of T 1 and T 2 relaxation times [415e421], identifying an induction period (~3 h) [417] as well as water in various hydrate phases [416], mobile water in micropores and water with restricted mobility in a solid gel component [420,421].
Spinespin relaxation rates are modulated by the motion of the species on which the spins reside and vary by several orders of magnitude for protons in rigid (e.g. solid hydrates) or mobile (e.g. molecular water) species. Rapid measurements of spinespin relaxation rates are, therefore, particularly attractive to monitor reaction in-situ [22]. This approach has been used to monitor hydration of white PC, and synthetic C 3 S and C 3 A [422e424], and led to identification of five components with distinct T 2 values corresponding to capillary pore water, solid-like crystalline water and OH À groups (i.e. in portlandite, gypsum, and ettringite), mobile water molecules incorporated in CeSeH, water molecules with restricted mobility in the CeSeH interlayer and secondary hydration water released by the decomposition of ettringite (Fig. 35).
Spinespin relaxation times have also been used to observe the retardation of PC hydration in the presence of various organic additives [425]. 1 H relaxation rates can also be exploited to link the structure and kinetics at the atomic level of cementitious materials to microscale and macroscale engineering properties such as porosity, pore size distributions and tortuosity in hydrated cement pastes, mortars and concretes. Measurement of spinelattice relaxation rates has been used to develop relaxation models for the dependency of T 1 on the pore surface to the volume ratio and the presence of nearby paramagnetic ions [426e429], while T 1 -T 2 [427,428] and T 2 -T 2 [430,431] correlation NMR experiments have been used to identify discrete pores (e.g. large capillary pores or smaller gel pores, Fig. 36) in hydrated white PC and synthetic C-S-H and measure exchange of water between these locations, by exploitation of the fact that increased pore size results in increased relaxation times. Variable temperature proton NMR relaxometry experiments have also been used to monitor freezeethaw phenomena in hydrated PC and C-S-H [432,433].
More recently, 1 H NMR relaxometry has been used to quantify all the water present, as well as the size and volume of discrete population of pores containing this water, in hydrated white PC both with [8] and without [434,435] silica fume additions. This work showed that addition of silica fume alters the chemical composition of the C-S-H formed, but does not alter the density, and identified four discrete populations of liquid water: water in the C-S-H interlayer in pores approximately 1 nm in diameter, water in the C-S-H gel in pores approximately 3 nm in diameter and water in small (~10 nm in diameter) and large ([ 10 nm in diameter) capillary pores. Correlation of 1 H NMR relaxometry with results from mercury intrusion porosimetry (MIP) has shown that MIP can accurately determine the porosity of cement pastes [436].

Conclusions and perspectives
SS NMR spectroscopy has played a pivotal role understanding the complex compositionestructureeproperty relationships in cements and related materials. In particular, SS NMR spectroscopy has afforded a level of insight into their atomic structure, reaction mechanisms, kinetics of formation and structural evolution which has been previously unattainable and has provided the basis for the current state of the art in understanding, modelling and predicting compositionestructureeproperty relationships, reaction mechanism and kinetics in PC, CAC, CSA, magnesia-based, AAM and geopolymer-based cements.
Application of advanced MAS NMR experiments probing 1 H, 13 C, 17 O, 19 F, 23 Na, 25 Mg, 27 Al,29 Si, 31 P, 33 S, 35 Cl, 39 K and 43 Ca nuclei have been used to study the atomic structure, phase evolution, nanostructural development, reaction mechanisms and kinetics occurring in these cementitious systems, including multidimensional, multiresonance and in-situ experiments. Single-pulse MAS NMR experiments probing 27 Al and 29 Si nuclei are by far the most routinely applied NMR experiments used to study cements and have been instrumental in building our current understanding. However, owing to the extensive structural information these experiments have provided to date, their role in continued advancement of the understanding of compositionestructureeproperty relationships in cementitious materials is limited.
Performing NMR experiments at high field (B 0 ! 20.0 T), probing less commonly investigated nuclei such as 17 O and 43 Ca in isotopically enriched samples and application of less commonly used techniques including CP MAS, MQMAS and multiresonance experiments (e.g. SEDOR, REDOR, TEDOR, REAPDOR and TRAPDOR) presents the best opportunity for elucidation of information regarding the mechanisms which control the structure and physical properties of cementitious materials, that has been previously unattainable. The difficulty in conducting these experiments and interpretation of the spectra obtained, however, is likely to be the largest obstacle to their routine application. This presents a key challenge which must be overcome and will require a Fig. 36. Proton spinespin relaxation times (T 2 ) vs. hydration time of white PC hydrated at 20 C. Proton signals come from water molecules in the pore solution (triangles), crystalline water and OH À groups (filled circles) in C-S-H and portlandite, more mobile water molecules in the interlayer of C-S-H (squares) and secondary hydration water released by the decomposition of ettringite (diamonds). Adapted from Holly et al. [422]. concerted, collaborative effort from both the SS NMR spectroscopy and cement materials chemistry communities.