Blast furnace slag-Mg(OH)2 cements activated by sodium carbonate

The structural evolution of a sodium carbonate activated slag cement blended with varying quantities of Mg(OH)2 was assessed. The main reaction products of these blended cements were a calcium-sodium aluminosilicate hydrate type gel, an Mg-Al layered double hydroxide with a hydrotalcite type structure, calcite, and a hydrous calcium aluminate phase (tentatively identified as a carbonate-containing AFm structure), in proportions which varied with Na2O/slag ratios. Particles of Mg(OH)2 do not chemically react within these cements. Instead, Mg(OH)2 acts as a filler accelerating the hardening of sodium carbonate activated slags. Although increased Mg(OH)2 replacement reduced the compressive strength of these cements, pastes with 50 wt% Mg(OH)2 still reached strengths of ∼21 MPa. The chemical and mechanical characteristics of sodium carbonate activated slag/Mg(OH)2 cements makes them a potentially suitable matrix for encapsulation of high loadings of Mg(OH)2-bearing wastes such as Magnox sludge.


Introduction
Alkali-activated slag cements are Portland cement-free binders produced through the chemical reaction between an alkaline activator and ground granulated blast furnace slag (GGBS). The performance of these materials is strongly dependent on the mineralogy and composition of the slag, and the nature and concentration of the alkaline activator chosen, 1 among other factors. Sodium carbonate-activated slags have attracted less attention than other alkali-activated slag cements which use sodium hydroxide or sodium silicate as the main activators, despite proven successful usage in Eastern Europe and the CIS as alternative cements for production of mortars and concretes. 2 This is mainly because of the longer setting times identied in these binders (oen >5 days), 3,4 compared with cements produced using other alkali-activators.
The use of sodium carbonate as the activator for GGBS has advantages compared with sodium hydroxide or silicate solutions, 5 as it is easier and safer to handle due to its powdered and less hygroscopic nature, and when dissolved in water will reach a lower pH than that of sodium hydroxide or silicate solutions with an equivalent content of alkalis. 3 Recent studies have demonstrated 6,7 that the oen-observed delayed hardening of these cements is strongly dependent on the chemistry of the slag used, and therefore, the general assumption that setting time problems will occur when using this activator is misleading.
The UK has accumulated signicant quantities of radioactive Mg(OH) 2 sludges from its Magnox nuclear programme, which are classed as intermediate level waste (ILW) streams. 8 These wastes consist of up to 100% sludge, predominantly Mg(OH) 2 with some uranium metal or oxides, along with varying radionuclides from nuclear ssion. 9,10 Other ILW streams are processed by encapsulation/immobilisation in composite cements based on ordinary Portland cement, GGBS and pulverised fuel ash. 11 This Mg(OH) 2 -rich sludge is more challenging to treat using such binders, due to the high water content of the waste. 12 Equally, any alternative cementitious matrices which could make use of the abundant Mg within these wastes as part of the encapsulation process could potentially improve upon conventional wasteform properties. 13 GGBS rich-binders are benecial in the encapsulation of radioactive wastes, as the slag contains sulphide and potentially Fe(II) which act as reducing agents, controlling the solubility of key radionuclides. 14,15 The hydration of GGBS in cements also oen forms layered double hydroxides with a hydrotalcite-type structure (general formula M x 2+ M y where A nÀ is oen OH À , Cl À or CO 3

2À
, and x/y is generally between 2 and 3 (ref. [16][17][18]), which is of particular interest for immobilisation of nuclear wastes due to its ability to retain certain long-lived radionuclides such as 129 I and 36 Cl (ref. 19) that are present in Magnox sludges. 20 A high MgO content within GGBS favours hydrotalcite formation in sodium silicate-activated 21,22 and sodium carbonate activated slags. 6 The addition of either reactive MgO 23 or calcined Mg-Al layered double hydroxides 6 to alkali-activated slag binders can also favour the formation of larger amounts of hydrotalcite-group phases. Activated slag cements with increased contents of such phases appear to be more resistant to degradation by carbonation 22 and chloride penetration, 24 potentially opening the door to the development of more durable cementitious materials.
The effect of the addition of Mg(OH) 2 to alkali-activated binders has not been reported in the open literature, although Collier et al. 25 identied that the addition of Mg(OH) 2 to Portland cement/slag cement composites potentially promoted the formation of higher contents of hydrotalcite. If this were to additionally hold true for Mg(OH) 2 addition to alternative binders, then this could interact with the cement, enhancing hydration and forming part of the binding phase, increasing the achievable waste loadings in an immobilisation context.
In this study a sodium carbonate activated slag binder was blended with varying quantities of Mg(OH) 2 (10, 30 and 50 wt% of the total binder). Kinetics of reaction were assessed via isothermal calorimetry, and hardened samples were evaluated using X-ray diffraction, Fourier transform infrared spectroscopy, thermogravimetry, solid state nuclear magnetic resonance, and scanning electron microscopy, focussing on both early age and longer-term aged samples. Particular emphasis is laid on exploring and understanding the binding gels within this system, and the degree to which Mg(OH) 2 might have inuenced their chemistries, to assess the level of structural substitution of Mg-containing phases that is possible.

Sample preparation
All binders were formulated with 10 g of Na 2 CO 3 per 90 g of GGBS + Mg(OH) 2 binder, with the proportion of Mg(OH) 2 :GGBS varied (Table 2). Although the overall Na 2 CO 3 content remained constant, the equivalent ratio of Na 2 O to slag increased as more of the GGBS was replaced by Mg(OH) 2 .
Samples were prepared in a Kenwood benchtop mixer by rst combining the required quantity of water and solid sodium carbonate. Aer dissolution of the solid Na 2 CO 3 , the GGBS was added to the mix over 2 minutes, followed by the addition of the Mg(OH) 2 over a further 2 minutes of mixing. The amount of water added was increased commensurate with Mg(OH) 2 addition, to maintain roughly consistent uidity between samples (judged during mixing). The pastes for hardened-state analysis were cast in 50 mL centrifuge tubes and stored at 20 C and >90% relative humidity until analysis at ages of 28 days and 18 months.
For fresh pastes, isothermal calorimetry measurements were carried out using a TA Instruments TAM Air isothermal calorimeter at 25 AE 0.02 C. The materials were mixed externally, then 20 g samples of paste were weighed into plastic ampoules and placed into the calorimeter. Heat output was measured for 14 days (336 hours).
Aer curing, the samples were crushed, ground in an agate mortar, and sieved to <63 mm prior to analysis. X-ray diffraction (XRD) was undertaken using a STOE STADI P diffractometer (Cu Ka radiation, 1.5418Å) with an imaging plate detector (IP-PSD) to collect data from 10 < 2q # 60 , and angle-corrected using an external silicon standard. Fourier transform infrared (FTIR) spectroscopy was undertaken on a Perkin Elmer Spectrum 2000 spectrometer in mid-infrared (MIR) mode, using 2 wt% of powdered sample in a pressed KBr disc. Solid-state 29 Si MAS nuclear magnetic resonance (NMR) spectra were collected on a Varian VNMRS 400 (9.4 T) at 79.438 MHz using a 6 mm zirconia rotor, at a 6 kHz spinning rate. Pulse duration was 4.5 ms (90 ), with a 10 s relaxation time for a minimum of 6000 repetitions, with chemical shis externally referenced to tetramethylsilane (TMS) at 0 ppm. 27 Al MAS NMR spectra were collected on a Varian VNMRS 400 (9.4 T) at 104.199 MHz using a 4 mm zirconia rotor, at a 14 kHz spin rate. Pulse duration was 1 ms (25 ), with a 0.2 s relaxation time for a minimum of 5200 repetitions. Chemicals shis were referenced externally to 1 M aqueous Al(NO 3 ) 3 . Scanning electron microscopy (SEM) was carried out on monolithic samples sliced using a diamond wafering blade; each specimen was mounted in epoxy resin, ground and polished using diamond paste to a 1 mm nish. Analysis was undertaken on a Hitachi TM3030 desktop SEM at 15 kV accelerating voltage, operating under a low vacuum, so coating of the samples was not needed. This was coupled with energy dispersive X-ray analysis (EDX) using a Bruker Quantax 70 Energy Dispersive X-ray Spectrophotometer for elemental mapping and spot analysis.
All samples analysed by thermogravimetry (TGA) were ushed with nitrogen for 30 minutes prior to analysis to remove surface water. Samples cured for 28 days were analysed using a Perkin Elmer Pyris 1 TGA, using 20 mg of material in an alumina crucible, heating at 10 C min À1 up to 1000 C in a nitrogen atmosphere. Aged (18 months) samples were analysed in a Perkin Elmer TGA 4000 instrument, using 20 mg of material in an alumina crucible, heating at 5 C min À1 up to 950 C in a nitrogen atmosphere. For these samples, the composition of the gases released during the TGA test was also determined using a Hiden Analytical mass spectrometer (MS) attached to the thermogravimeter.

Results and discussion
3.1. Early age evolution of a Mg(OH) 2 -modied sodium carbonate-activated slag cement The effects of GGBS replacement by Mg(OH) 2 on the reaction kinetics of a sodium carbonate-activated slag cement were determined by isothermal calorimetry, Fig. 1 and 2, along with XRD analysis (Fig. 3) of the pastes during the rst week of curing. Calorimetric data are normalised to both overall mass of sample ( Fig. 1), and to content of dry GGBS (Fig. 2) for further interpretation.
Three main regions are identied from the calorimetric curves: <1 h, <24 h and 3-7 days, that correspond to distinct  exotherms. Within the rst hour, initial particle wetting and dissolution, and disturbance from the insertion of vials into the calorimeter, dominate. Following this, each sample exhibited an exotherm between 5-24 h, which can be attributed to formation of gaylussite (Na 2 Ca(CO 3 ) 2 $5H 2 O, powder diffraction le (PDF) # 074-1235) and thermonatrite (Na 2 CO 3 $H 2 O, PDF #008-0448) identied by XRD in all the samples aer 1 day of curing (Fig. 3), independent of the content of Mg(OH) 2 . Additionally, calcite was identied in samples with higher contents of Mg(OH) 2 ( Fig. 3C and D). These results are consistent with what has been reported for sodium carbonate activated slag cements, where the Ca 2+ released from the dissolving slag reacts with CO 3 2À from the activator to form carbonate salts such as calcite and gaylussite, increasing the pH through the release of hydroxide ions. 4,6 The exothermic peak associated with the formation of carbonate phases was signicantly delayed in cements without Mg(OH) 2 , appearing $10 hours later than for samples with 30 wt% and 50 wt% Mg(OH) 2 . Aer this event, negligible heat evolution was observed for up to 3 days in all cases, although the XRD results (Fig. 3) demonstrated that the formation of gaylussite at the expense of thermonatrite continued during this time.
Between 3 and 7 days each sample exhibited a further large exotherm, consistent with an acceleration-deceleration period, associated with the nucleation, growth and precipitation of reaction products including a calcium silicate hydrate (C-S-H)- type gel, as evidenced by the appearance of a diffuse scattering feature at $29 2q in the X-ray diffractograms of all the cements assessed (Fig. 3). The formation of this gelwhich is likely to contain signicant substitution by alkalis and aluminium, and so is more accurately described as C-(N-)A-S-H 28-30was signicantly accelerated in each of the samples containing Mg(OH) 2 .
The overall heat output of 10 wt% and 30 wt% substitution of Mg(OH) 2 (M10 and M30 in Fig. 1) is similar to that without Mg(OH) 2 addition (M0). This demonstrates that Mg(OH) 2 can be safely added to an alkali-carbonate activated slag matrix without prompting an increased exotherm, which is highly desirable in the context of nuclear waste immobilisation. The heat of reaction is spread over a much wider timeframe than is typical in Portland cement blends, where the majority of heat is evolved within 40 h. 31,32 This brings the opportunity to lower the wasteform peak temperature and ensure safer processing. Further replacement of GGBS by Mg(OH) 2 , up to 50 wt% (M50), lowers the overall heat of reaction due to the reduced quantities of GGBS available to react, and the increased quantity of water within this sample diluting the heat output per gram of total material (as Fig. 1 is normalised to total sample mass).
Normalising the data to the mass of dry GGBS (Fig. 2B) highlights a clear progression towards higher heat of reaction per gram of GGBS as the Mg(OH) 2 content is increased up to 50 wt%, indicating either increased dissolution of the GGBS or a chemical reaction involving Mg(OH) 2 ; the latter possibility will be the particular focus of the following sections of this paper. Presentation of the heat ow data normalised in this way ( Fig. 2A) also shows the monotonic increase in the intensity of the rst heat release peak (up to 15 hours) with Mg(OH) 2 addition. Relative to the other samples, the composition containing 50 wt% Mg(OH) 2 produces more heat per gram of GGBS during the timeframe associated with C-(N)-A-S-H formation (3-7 days), suggesting that there may be more extensive dissolution of the slag in this sample.
The effect of Mg(OH) 2 addition in accelerating the precipitation of bulk reaction products (i.e. earlier onset of the second main reaction peak in all samples containing Mg(OH) 2 compared to the M0 sample in Fig. 1 and 2) could be a consequence of either a threshold effect of increasing the Na 2 O/GGBS ratio upon Mg(OH) 2 incorporation (see Section 2.2 and Table 2), or a ller effect from the Mg(OH) 2 itself. The higher proportion of activator to GGBS might have moved the composition beyond a point at which the formation of hydration products is accelerated, compared to the baseline mix without Mg(OH) 2 .
However, as the absolute quantity of Na 2 CO 3 is kept constant, and the higher water content will serve to dilute its alkalinity, the acceleration of reaction product precipitation might also be linked to a ller effect, whereby the added particles of Mg(OH) 2 act as sites around which reaction products can nucleate and precipitate. 33,34 According to the data in Fig. 2, the accelerating effect does not increase with higher additions of Mg(OH) 2 beyond 10 wt%, which tends to suggest that nucleation rather than a threshold chemical ratio is the controlling factor.
3.2. Long term structural evolution of a Mg(OH) 2 modied sodium carbonate activated slag cement 3.2.1. X-ray diffraction. Aer 28 days of curing, X-ray diffraction patterns (Fig. 4A) show the continued growth of a C-S-H type phase when compared with the 7 days data in   Fig. 3. High intensity reections of the crystalline phases brucite (Mg(OH) 2 ) and gaylussite were also identied. In all of the samples a reection at 11.6 2q was identied, and assigned to the formation of a hydrotalcite-like phase resembling Mg 6 Al 2 (CO 3 )(OH) 16 $4H 2 O (PDF # 041-1428) but potentially with differences in Mg/Al ratio and/or interlayer species; the effects of these two factors are convoluted in the XRD analysis of hydrotalcite-group phases relevant to alkaliactivated cements. 35 This phase was not observable aer 7 days of curing (Fig. 3), and so has grown in quantity and/or crystallinity during ongoing curing of the hardened samples.
The main reection assigned to this phase is especially prominent in samples with high contents of Mg(OH) 2  Magnesium is known to be incorporated into calcite, forming magnesian calcites. This has been well studied within marine calcites, with Mg incorporation varying depending on pCO 2 , temperature and pH. 36,37 However, any possible formation of magnesian calcite would be difficult to determine from the XRD data presented here; although magnesium incorporation affects the calcite lattice, resulting in a slight shi in the diffraction peaks, at low levels of incorporation this is difficult to discern from diffractograms as complex as those shown in Fig. 4. In the literature higher Mg : Ca ratios are observed to affect CaCO 3 polymorph formations and stabilities, stabilising aragonite and vaterite by retarding growth of calcite, and enhancing the formation of monohydrocalcite. [37][38][39][40] None of these are identied within any of the samples analysed here, suggesting that there is little involvement of the added Mg(OH) 2 in such processes, or that the quantities formed are too low (or of low crystallinity) to be easily detected in this XRD analysis.
Although all of the cements analysed here show similar phases aer 28 days of curing, their compressive strengths differ notably (Table 3); lower compressive strengths were obtained as more Mg(OH) 2 replaces GGBS. In the mix designs used here (Table 2), Mg(OH) 2 addition also necessitated an increased water to solids ratio to enable mixing and casting. In alkali activated materials this increased water content would typically reduce the concentration of the activator, however this does not occur here as activator dose was held constant with respect to the total solids content, meaning that the Na 2 O/GGBS ratio increases at higher Mg(OH) 2 contents. The fact that the sample without Mg(OH) 2 displays a lower strength than with 10 wt%, despite its lower water/solids ratio, suggests that the addition of Mg(OH) 2 does inuence the bulk properties of the cements in a substantive way.
The phase assemblage of each binder continued to develop from 28 days up to 18 months of curing (Fig. 4B), with only minor gaylussite reections identied in the samples at 18  months. In place of gaylussite, the continued formation of calcite was observed in all the samples, increasing in intensity with higher degrees of replacement of the slag by Mg(OH) 2 . A high intensity reection of hydrotalcite was observed in all samples, displaying especially strong reections in the sample with 30 wt% Mg(OH) 2 (Fig. 4B, M30). The reections assigned to Mg(OH) 2 were strong in all samples in which Mg(OH) 2 was added, suggesting that if there was any chemical reaction between Mg(OH) 2 and the other mix constituents, it was far from complete. Thermodynamic calculations of the stable phase assemblage for sodium carbonate-activated slag binders containing high Mg contents have predicted brucite as a stable constituent of the phase assemblage, 41 and the observation of its limited reaction here is consistent with those calculations. The increasing content of Mg(OH) 2 did not appear to inuence the observed phase assemblage, other than increasing the intensity of the reections assigned to calcite, C-(N-)A-S-H and hydrotalcite, which is commensurate with an increased degree of slag dissolution due to the increased activator/slag ratio in the higher-Mg(OH) 2 samples.
The overall phase evolution of these slag cements was similar to the processes previously identied in sodium carbonate activated slags. 4,42 The major difference from the established literature is the presence of calcite, and the absence of the CaCO 3 polymorphs aragonite and vaterite, which is contrary to other results available in the literature for aged samples. 4,42 For sodium carbonate activation of a slag with a comparable composition to that which was studied here, Myers et al. 30,41 predicted the formation of a C-(N-)-A-S-H gel, calcite, the zeolitic phase natrolite (Na 2 Al 2 Si 3 O 10 $2H 2 O), a hydrotalcite-group Mg-Al LDH phase with OH À groups dominating its interlayer (MA-OH-LDH), and a calcium monocarboaluminate phase in the AFm family (C 4 AĈH 11 ). In this study, neither natrolite nor AFm phases were detected directly by XRD, however the presence of C 4 AĈH 11 remains possible as its key Bragg reections (PDF # 041-0219; 43 ) present signicant overlap with those of hydrotalcite-like phases, making condent identication difficult. The distinction between the MA-OH-LDH phase and carbonate-containing hydrotalcite-group phases is also challenging due to their similar diffraction patterns; this will be revisited below using thermogravimetry and mass spectroscopy. Gaylussite is not predicted to be a stable phase within these cements, which is in line with the decrease in its reections here from 28 days to 18 months, and in agreement with other studies. 4   , the most notable difference among the samples is related to the sharp peak at 3698 cm À1 , which is assigned to the O-H stretching vibration mode of brucite. 44 As expected, reduced contents of Mg(OH) 2 in the cement give a lower intensity in this band. Although signicant quantities of Mg(OH) 2 remain in the samples aer 18 months of curing, especially when adding high contents of Mg(OH) 2 (Fig. 5B, M50), a slight reduction in the intensity of this band is observed when comparing pastes containing lower Mg(OH) 2 contents cured for 18 months, against the data at 28 days. This tentatively suggests Mg(OH) 2 might not be completely inert, though there is potential for this to be an artifact of sample preparation or analysis, due to the 17 months sampling interval between data sets.
The broad band centred at 3458 cm À1 in each of the samples at 28 days, and 3455 cm À1 at 18 months, corresponds to the H-OH stretching mode of bound water, while the resonance at 1644 cm À1 is assigned to the bending mode of the H-OH bond. 45,46 These resonances are typical of bound and interlayer water within C-A-S-H type gels, 47,48 along with crystal water within a hydrotalcite-type phase and in gaylussite, 46 consistent with the phases observed by X-ray diffraction (Fig. 4). The broad band at 1420 cm À1 is attributed to the n 3 carbonate stretching band typical of carbonate phases, consistent with the identication of calcite, gaylussite and the hydrotalcite-like phase in the XRD data ( Fig. 3 and 4) for these samples. 46,49 Carbonate vibrations from calcite were more clearly evident in a sharp peak (n 2 band) at 874-875 cm À1 , which increased in intensity in the samples with higher contents of Mg(OH) 2 and also with curing age. This correlates well with the trends in the XRD reections assigned to calcite (Fig. 4).
Both the OH stretching (3800-3600 cm À1 ) and carbonate vibration (900-850 cm À1 ) regions of the FTIR spectra of aged samples are shown in an expanded view in Fig. 6, as these regions provide evidence for the formation (or absence) of other carbonate or hydroxy-carbonate phases.
Due to the large quantity of Mg(OH) 2 in these samples, there might exist opportunities for the formation of magnesium carbonate phases. Magnesite (MgCO 3 ) typically exhibits a sharp vibration at $885 cm À1 , 46 while other hydrated magnesium carbonates display sharp vibrations between 3700-3500 cm À1 (e.g. nesquehonite (MgCO 3 $3H 2 O) at 3570 cm À1 , 46 hydromagnesite (Mg 5 (CO 3 ) 4 (OH) 2 $4H 2 O) at 3645, 3505 and 3445 cm À1 , 46 and dypingite (Mg 5 (CO 3 ) 4 (OH) 2 $5H 2 O) at 3648 cm À1 50 ). The absence of these vibration modes, combined with the lack of reections characteristic of magnesium carbonates from the XRD data as shown above, rules out the formation of pure magnesium carbonates other than the hydrotalcite-like phase. This analysis also rules out the formation of the CaCO 3 polymorphs aragonite and vaterite, as they display characteristic infrared vibrations in the region 900-850 cm À1 . 46,51 The possibility of magnesium incorporation within Cacontaining carbonate phases is somewhat more difficult to determine within these samples. Huntite (CaMg 3 (CO 3 ) 4 ) can be excluded, as this mineral exhibits strong vibrational modes at 892 and 869 cm À1 , 46 which are not observed in Fig. 6. Dolomite (CaMg(CO 3 ) 2 ) is known to have slow crystallisation kinetics so is rarely formed in cements in any case, 52 and typically displays a sharp vibration at 881 cm À1 . 53 Incorporation of Mg into calcite to form a magnesian calcite is much more difficult to ascertain, as was the case for the XRD analysis in the preceding section. FTIR analysis of calcites of different Mg contents has shown that the wavenumber of the carbonate n 2 vibrational mode increases with the proportion of magnesium in the calcite structure, though this is typically only up to $876-7 cm À1 in naturally occuring magnesian calcites (e.g. Ca 0.84 Mg 0.16 CO 3 analysed in 54 ). The carbonate n 2 vibration within these samples does not alter in position from 0 to 50% Mg(OH) 2 addition, suggesting that there is little to no Mg incorporation within the calcite that is formed.
3.2.3. Thermogravimetric analysis. Thermal analysis of the samples aer 28 days (Fig. 7A) and 18 months (Fig. 7B and 8) shows a mass loss at temperatures below 200 C, which can be attributed to removal of loosely bonded water, and the initial loss of the interlayer and structural water present in the C-(N-)A-S-H. 55,56 The sharp peak at 110 C is assigned to gaylussite dehydration, 57 in agreement with the identication of this phase by XRD analysis, where notably reduced gaylussite intensities were observed aer 18 months of curing.
Dehydroxylation of Mg(OH) 2 was prominent at $400 C, where mass loss increased commensurately with the addition of Mg(OH) 2 . At $630 C, calcite decomposition was observed, followed by gaylussite decarbonation. 58 The decomposition of the hydrotalcite-like phase was more difficult to determine as it typically exhibits a multi-step process of dehydration, dehydroxylation and/or decarbonation up to $600 C, depending on the structure and interlayer anions, 59-61 and potentially overlapping with the mass loss due to Mg(OH) 2 decomposition. This is highlighted within the differential thermogram of the 18  Fig. 7B), which shows a distinct shoulder at $200 C and further mass loss at $350 C. This is characteristic of a hydrotalcite-like phase 62 and can be noted in the other 18 months samples in Fig. 7B, where there is an apparent shi in the Mg(OH) 2 decomposition peak from $380 C towards a lower temperature, $350 C.
The TG-MS data (Fig. 8) for samples with different contents of Mg(OH) 2 addition aer 18 months might provide further insight into the nature of this Mg-LDH phase, as the XRD analysis presented above was not able to unequivocally dene whether it is a solely hydroxide-containing product, or whether it contains carbonate. In these samples, the mass loss at $350 C only registered in the mass spectroscopy data as H 2 O, not as CO 2 . Typically a carbonated hydrotalcite-like phase would evolve interlayer CO 2 in this temperature range. 63 Plotting data for CO 2 alone (Fig. 8B) does reveal slight evolution of CO 2 here, however this is part of a broader decarbonation culminating in calcite decomposition at $630 C rather than a discrete event which could be attributed to a carbonatecontaining LDH. Unfortunately Mg-Al-CO 3 LDH phases have been shown to release CO 2 at >400 C, 61 which overlaps with CaCO 3 decomposition. Although evidence from thermodynamic modelling predictions of Myers et al. 30,41 suggest that the hydrotalcite-group phase formed in these systems is in fact an MA-OH-LDH phase rather than a carbonate containing phase, TG-MS is unable to denitely determine this within these samples.
3.2.4. Microstructural features of aged Mg(OH) 2 -containing sodium carbonate-activated cements. Backscattered electron (BSE) micrographs of the sample without Mg(OH) 2 (M0) are displayed in Fig. 9, with corresponding elemental maps. The slag grains (labelled 's' in the micrographs) have undergone varying degrees of dissolution, resulting in the formation of a dense matrix composed of a C-(N-)A-S-H type gel, as evidenced by the widespread distribution of Al, Si, Ca and Na, in which all the other features are embedded. The remnant slag grains are surrounded by thin darker reaction rims (labelled 'r s ' in Fig. 9A), clearly observed at a higher magnication (Fig. 9B). This 'rim'like region is particularly evident for slag nes, where the slag grains have entirely reacted, leaving regions with darker greyscale values than the bulk matrix (these are labeled 'r f ' in Fig. 9A and particularly evident in the Mg map in Fig. 9B), without a bright remnant slag core. This region is richer in Mg than the bulk matrix, indicating the presence of the hydrotalcite-like phase identied above, and consistent with observations by other authors in slag cements activated by sodium hydroxide or silicate, 64,65 and in aged sodium carbonate-activated slag cements. 6 Large darker regions (labelled as 'g' in Fig. 9A) are assigned to gaylussite, identied by enrichment in Na and C compared to the bulk matrix, and the absence of Al and Si. Similar features have been observed in sodium carbonate/silicate-activated slag cements, 4 where gaylussite is reported as secondary reaction product. Formation of calcite (labelled 'c' in Fig. 9A), corresponding to the regions observed in the EDX maps to be rich in Ca and C but lacking in Al and Si, was observed around gaylussite. The large darker patches (labelled 'ca' in Fig. 9A) are particularly low in Si but still contain appreciable quantities of Ca, C and Al, suggesting the formation of a hydrous calcium carboaluminate phase. This may potentially be the monocarbonate AFm (C 4 AĈH 11 , or 'CO 2 -AFm') phase predicted by Myers et al. 30,41 in sodium carbonate-activated slag cements via thermodynamic modelling, and recently identied by Ke et al. 6 in aged cements of that composition.
SEM results for specimens with 10 wt% Mg(OH) 2 (sample M10) are shown in Fig. 10. Again, a signicant quantity of unreacted slag (marked s) remained, around which reaction rims (r s ) have developed. As in Fig. 9, remnant reaction rims from dissolution of slag nes (r f ) are observed, along with the presence of gaylussite (g), and with calcite (c) forming on the edges of gaylussite patches. Small Mg(OH) 2 particles (b) are observed throughout the sample, and do not appear to have undergone any reaction but instead remain embedded within the C-(N-)A-S-H rich bulk matrix. Fig. 10B presents a higher magnication view of two large Mg(OH) 2 particles, which are This journal is © The Royal Society of Chemistry 2018 embedded in the bulk gel. The EDX maps do not reveal any reaction region surrounding the particles, nor is there evidence of decalcication or Al enrichment in the bulk surrounding these particles which would indicate MA-OH-LDH formation around the Mg(OH) 2 particles. This suggests that the Mg(OH) 2 is not participating in any reaction with the slag, and is merely acting as a ller or site for nucleation for a C-(N-)A-S-H gel.
Further increasing the level of replacement of GGBS by Mg(OH) 2 (30 wt%, sample M30) resulted in a substantial change in the microstructure of the hardened cement paste, and an increased tendency to cracking and desiccation during sample preparation for electron microscopy. In M30 (Fig. 11), little unreacted slag remains, and that which is le is surrounded by large reaction rims (r s ), which are higher in Mg and Al than the bulk matrix. Signicant amounts of nely dispersed Mg(OH) 2 particles (b) are observed embedded throughout the matrix. As in Fig. 10, these particles appear to be unreacted, with no transition or visible reaction rims that could indicate participation in the chemical reaction of the cementing system. The XRD analysis (Fig. 4B) suggested the presence of increased amounts of calcite in this sample compared with those with lower Mg(OH) 2 contents. The observations here from SEM are consistent with this, as large clusters of calcite (c) are identied, instead of the gaylussite observed in the specimens with lower Mg(OH) 2 contents. Like the gaylussite, calcite is largely conned to large clusters rather than being uniformly distributed throughout the sample. It is, however, also present as minor clusters associated with the regions identied above as potentially being the CO 2 -AFm phase, in Fig. 11. This micrograph shows the extensive formation of this hydrous calcium carboaluminate phase in three locations, highlighted by dashed regions and identiable by a very strong content of Ca and Al within the EDX maps.
The replacement of 50 wt% of the GGBS by Mg(OH) 2 (M50) further changes the observed microstructure of the cement. No remnant slag grains are identied in Fig. 12A and B, having undergone full dissolution and leaving large reaction regions (r f ) enriched in Mg and Al. These appear to have also developed a secondary reaction rim, decient in Si and Ca relative to both the bulk and interior of the reaction regions. This was in some instances further encircled by a thin layer of calcite (c). Mg(OH) 2 (b) was dispersed throughout the sample, both as large clumps and as smaller particles, commensurate with the very high replacement level in this sample. A distinct CO 2 -AFm growth (ca) can be observed in Fig. 12A, which appeared to be growing around one of the Mg(OH) 2 particles, although no reaction region is identied within the particle itself, and Mg substitution into the AFm structure is not likely. The AFm phase appeared lighter in this micrograph than in previous samples, but this was likely a contrast effect due to the high Mg(OH) 2 content and lack of unreacted slag, which typically appeared lighter in the micrographs. Looking at a wider section of the sample in Fig. 12B, the morphology was quite different from that of samples containing less Mg(OH) 2 . Very large clusters of calcite (c) were present, along with ne speckles of calcite distributed throughout the binder.
Looking more closely at one of the reacted slag particles in this sample, Fig. 13, distinct multiple reaction rims (r o ) are observable. The inner reaction rim (r f ) is similar in Si, Ca and Na concentrations to the bulk matrix, although enriched in Mg and Al. This is identied as formation of a C-(N-)-A-S-H gel intermixed with an MA-OH-LDH phase, as the degree of incorporation of Mg into C-S-H type gels is known to be very limited. 66 The next reaction rim has a lower density darkened ring decient in most elements (i.e. is a porous region), followed by another ring (r 0 ) enriched in only Al and Mg, indicating the presence of the MA-OH-LDH phase alone. This was also observed in several reacted slag regions throughout the sample. Both sites highlighted also exhibit thin calcite (c) deposits around the outer rings; similar Liesegang-type ring formations have been observed in aged silicate-activated slag concrete, 67 and their presence suggests that the reaction of these cements at advanced age, independent of the activator used, is likely to follow an Oswald supersaturation-nucleation-depletion cycle.
Further analysing the large Mg(OH) 2 particle marked (b) in Fig. 13, the distributions of Ca, Si, Al and Na all reach up to the particle edge according to the EDX maps, indicating that this particle is embedded directly within a C-(N-)A-S-H gel. There is no preferential accumulation of Al along the particle edge which would indicate the formation of a hydrotalcite-like phase. Any formation of a magnesium silicate hydrate (M-S-H) as a reaction product would lead to competition between Ca-rich and Mg-rich phases for the Si, and this would leading to regions rich in Mg and Si only, to the exclusion of Ca. However, no such regions are identied along the edge of this Mg(OH) 2 particle, nor any others in the analysis presented here, so it can be concluded that the bulk Mg(OH) 2 did not react to form M-S-H to a signicant extent within this system. This also indicates that the source of Mg for formation of the hydrotalcite-like phase identied by XRD analysis (Fig. 3) was the slag alone, which had an MgO content of 8.4 wt%, Table 1.
3.2.5. Chemical composition of the main reaction products. Spot EDX measurements were taken throughout the bulk matrix, the reaction rims and the AFm phase. Table 4 details the average atomic ratios measured for these distinct features (rim, bulk, inner rim and outer rim) throughout all the samples aer 18 months of curing. The Ca/Si ratio of the bulk gel remained fairly similar between samples M0 -M50, ranging from 1.09 to 1.22. This is a typical Ca/Si ratio for sodium carbonate activated slags, 4 though higher than typically identied in sodium silicate activated binders. 68 The same Ca/Si ratio was measured in the reaction rims featured in all of the samples studied, supporting the identication of these regions being intermixed MA-OH-LDH and C-(N-)A-S-H, while the bulk was largely a C-(N-)A-S-H gel with some MA-OH-LDH present.
The Al/Si values of the rims were consistently higher than those of the bulk (up to 1.74 for sample M50), which also correlated with a higher Mg/Al ratio in the rims. The exception to this is the measured Mg/Al ratio of 7.69 in M50, where it is postulated that the bulk spot maps included Mg(OH) 2 particles due to their ne dispersal throughout the microstructure of this sample. Fig. 14 correlates Ca/Si vs. Al/Si atomic ratios of the bulk matrix in all of the samples. There is considerable scatter between the individual points, however there is no discernible difference between any of the samples, suggesting that the overall bulk composition of calcium, aluminium and silicon within the gel does not change as Mg(OH) 2 is added into the system. This does not, however, mean that no changes are occurring; as changes in Al-Si ordering, or interactions with excess Na, are not easily observed through bulk EDX. There will be an increase in Na/Si ratios when 30 wt% or 50 wt% Mg(OH) 2 replaces GGBS, though Na is easily lost in sample preparation and so its analysis by EDX is oen unreliable. A higher Na content raises the possibility of increased charge compensation within a C-(N-)A-S-H gel, leading to structural changes, or the potential formation of a co-precipitated N-A-S-H gel. Neither of these would be easily determined via bulk EDX analysis, but the following section will address these points by NMR analysis. The Al/Si values of the bulk gel for samples M0 -M50 were very similar, at 0.27-0.31. Although this is high for a C-(N-)A-S-H gel, the true Al/Si ratio of the gel will be lower once the Al present in MA-OH-LDH is removed from the analysis (as determined below). We postulate that the nature of this gel is likely to change as a result of the increased Na/Si ratio associated with higher Mg(OH) 2 replacement levels, as the Na 2 O/ GGBS ratio also increased. Due to the signicant quantity of Na 2 CO 3 used relative to the GGBS in formulations M30 and M50 (9.75 and 14.62% Na 2 O eq. respectively), there is the potential for the additional formation of an N-A-S-H gel, as will be assessed in more detail in the following sections.
Analysis of compositional changes within the bulk matrix is to some degree complicated by the formation of the MA-OH-LDH phase. Although this has been noted to mainly precipitate in rims surrounding GGBS particles (or in the remnants of dissolved particles), analysis of the Mg/Si vs. Al/Si spot map data (as plotted in Fig. 15 for M0) shows evidence for its formation within the bulk as well.
There is a clear positive linear correlation between Mg and Al contents, indicating LDH formation in this sample, in both bulk and rim regions. Linear regression of these data to zero Mg content (the intercept with the horizontal axis) shows that the Al/Si ratio in the C-(N-)A-S-H gel is 0.21, and the slope of the trendline indicates an Mg/Al ratio within the MA-OH-LDH phase of 1.68. This is a lower Mg/Al ratio than is typically found in hydrotalcite-group minerals, which are more commonly $2-3. 17 This suggests either that an unusually Alrich LDH phase has formed here, or that there is an additional Al-rich phase present intermixed with the bulk matrix. It may be possible that some of the AFm-structured material is nely dispersed through the bulk gel in addition to its presence in discrete crystallites as identied by SEM. However, a more likely explanation is the formation and intermixing of a disordered sodium aluminosilicate hydrate (N-A-S-H) type gel, as identied by Myers et al. 29 in sodium silicate-activated slag cements, and this will be further addressed by NMR in the following section.
It was not possible to perform this type of quantitative analysis of Mg/Si vs. Al/Si ratios to determine phase compositions in samples with Mg(OH) 2 additions, as the ne dispersion of Mg(OH) 2 throughout the samples meant that not all of the Mg detected could be assigned to a hydrotalcite-like phase. Nonetheless, Fig. 16 shows the Mg/Si vs. Al/Si ratios for the bulk and rim EDX points in samples M0 -M50, along with the data points for the phase in M0 identied as the CO 2 -AFm as a comparison. Overall, the Mg content of both the bulk and the reaction rims in Mg(OH) 2 -containing samples appeared to increase, due to interference from dispersed Mg(OH) 2 particles, which caused the bulk to appear rich in Mg, and also caused the reaction rim appear to have a higher Mg/Al ratio than its true value. Interestingly, the outer rim in the M50 sample showed a much higher Mg/Al ratio than the reacted slag rims in that sample. This mirrors the observations from the EDX maps, suggesting that this is almost completely a Mg-Al LDH phase, rather than a mix of C-(N-)A-S-H and MA-OH-LDH as was the case for the inner reaction rims, and for the slag rims observed in the other samples.

Solid state nuclear magnetic resonance spectroscopy
The 29 Si and 27 Al MAS NMR spectra of 18 months-cured samples with different contents of Mg(OH) 2 are displayed in Fig. 17, along with the spectrum of unreacted GGBS for reference. All spectra are normalised to integrated area for comparison. The incomplete, and potentially incongruent, dissolution of slag glass in sodium carbonate activated slag pastes hinders detailed deconvolution of the 29 Si MAS NMR spectra. However, no residual slag was detected by SEM in the M50 sample ( Fig. 12  and 13), and so this spectrum was deconvoluted (Fig. 18) and the resultant peak positions were used as the basis for interpretation of the other spectra.
The spectral deconvolution for sample M50 (Fig. 18) was conducted following, where possible, the peak assignments previously determined by Myers et al. 29 for alkali activated slag binders, who interpreted this structure as a mixture of C-(N-)A-S-H and N-A-S-H type gels. The data presented show strong similarities to those for slag cements activated with both with sodium carbonate and sodium silicate activators. 6,22,69 Four distinct peaks were readily identiable within the 29 Si MAS NMR spectra (Fig. 17A), along with a broad resonance centred at $À74 ppm associated with residual slag. Other peaks and broad shoulders were inferred from the deconvolution presented in Fig. 18. The Q 1 region is split into two sites (À77.0 and À79.2 ppm), following the reasoning of Myers et al. 29 who highlighted the potential existence of multiple Q 1 environments due to differences in shielding by charge balancing Ca 2+ , H + and Na + cations. In general, as GGBS is replaced by Mg(OH) 2 , the broad resonance associated with the slag decreases in intensity. The proportion of Q 2 (1Al) to Q 2 sites appears to remain constant, as both component peaks increase in intensity, consistent with the formation of longer chains in a cross-linked bulk C-(N-)A-S-H gel and more extensive dissolution of the slag at higher Mg(OH) 2 replacement levels.
The deconvoluted 29 Si MAS NMR spectrum for the paste with 50 wt% Mg(OH) 2 (Fig. 18) had a lower proportion of Q 1 (II) and  29 Si, (B) 27 Al. Q 2 (1Al), and a larger Q 2 contribution, than is typically identied for alkali-activated slag systems. 22,28,69 Although all the slag glass had reacted in sample M50, a minor Q 0 peak at À73.5 ppm remained, which is tentatively attributed to residual ordered akermanite-like regions 70 from the slag which were less reactive than the glass (although not sufficiently crystalline to be iden-tied by XRD), as also seen in all other 29 Si MAS NMR spectra (Fig. 17A). The reduced intensity for the Q 2 (1Al) site indicates that the C-(N-)A-S-H gel contains a lower degree of Al substitution than silicate-activated slag binders, 22,29 consistent with the Al being incorporated into hydrotalcite-like and C 4 AĈH 11 phases previously identied via XRD (Fig. 4) and SEM analysis (Fig. 12).
The presence of a N-A-S-H-type gel, as postulated above, is supported by the existence of resonances upeld of that of the Q 2 site. The resonance at À88.0 ppm has been assigned to both Q 3 (1Al) from cross linking within the C-(N-)A-S-H gel, and to a Q 4 (4Al) site from a N-A-S-H type gel. This cannot arise from a C-(N-)A-S-H gel alone, as the total resonance intensity at this frequency is much larger than is possible for a C-(N-)A-S-H gel (Table 5), which structurally can accommodate a maximum of 1 Q 3 (1Al) per 4 Q 2 (0,1Al) sites. 71 The presence of a further resonance at À94.3 ppm, which is assigned to a Q 4 (3Al) site, provides additional evidence for the contribution from a N-A-S-H gel.
Although this sample (M50) contained a substantial proportion of Mg(OH) 2 , there was no evidence of the reaction of this Mg(OH) 2 to form additional magnesium-containing phases according to FTIR, XRD or SEM as presented above. If an M-S-H type phase was formed, this would be expected to show distinct resonances between À86 and À89 ppm and between À93 and À98 ppm, depending on the M-S-H composition. 72,73 Although these fall in the regions assigned to Q 3 (1Al)/Q 4 (4Al) and Q 4 (3Al) here, there was no evidence for the formation of these from any other techniques, and the fact that the atomic ratio correlation plots indicate that the additional disordered phase present is Alrich rather than Mg-rich (i.e. N-A-S-H not M-S-H) supports this argument.
The resonances in the 27 Al MAS NMR spectra (Fig. 17B)  region, different q n sites have been tentatively designated, denoting connectivity comparable to the Q n notation for Si. Considering the results of Myers et al. 29 for sodium silicate activated slag cements, two q 2 sites are observed, corresponding to Al sites in a C-(N-)A-S-H gel. A resonance at 74 ppm has been reported as being due to the Al in bridging positions linked to Q 2 (1Al) silicon environments within a C-(N-)A-S-H gel, and would typically exhibit a sharp peak, 28 but is not so prominent within these spectra. This is in agreement with the reduced prominence of the Q 2 (1Al) Si site in these samples (Fig. 17A) compared to the existing literature. The q 3 site, which is the bridging crosslink site in C-(N-)A-S-H, is particularly prominent within these samples, and q 4 sites are characteristic of N-A-S-H. This ts well with the identication of Q 4 (4Al) and Q 4 (3Al) sites within the 29 Si MAS NMR spectra.
The Al[V] region shows only a minor Al environment in the range $45-20 ppm, present in the glassy slag fraction, which correspondingly disappeared in samples with higher contents of Mg(OH) 2 replacement (M30 and M50) as the slag underwent reaction. The C-(N-)A-S-H here does now show a signicant presence of Al(V) in its interlayer.
The Al[VI] region only displays a single broad resonance at 8-9 ppm, associated with a combination of MA-OH-LDH hydrotalcite-like structures and the hydrated calcium aluminate (postulated C 4 AĈH 11 ) phase, 74,75 which have similar chemical shis and so overlap in the spectra. This made further elucidation of the signals in this region difficult, but the presence of both contributions is supported by the identication of both these phases in the XRD and SEM/EDX analysis.
The NMR analysis here demonstrates the presence of a highly cross-linked C-(N-)A-S-H gel, an N-A-S-H gel, and MA-OH-LDH and CO 2 -AFm phases. This is a complex system, with many phases which have the potential to immobilise or retain radionuclides if used in a waste immobilisation application. None of the key binding phases appear to have undergone any reaction or ionic substitution with Mg from the Mg(OH) 2 which was added into these cements. There is ultimately no evidence presented here for the formation of a M-S-H or M-A-S-H gel, from either NMR, XRD or SEM, and this indicates encapsulation of Mg(OH) 2 rather than its incorporation through chemical reaction.

Conclusions
This study has highlighted the hydration mechanisms and products of sodium carbonate activated slags, both with and without the addition of Mg(OH) 2

. A C-(N-)A-S-H/N-A-S-H assemblage, a hydrotalcite-like compound and a hydrous
AFm-structured calcium carboaluminate phase were the major reaction products identied in all samples, with calcite increasingly formed as the slag was replaced with Mg(OH) 2 and the Na 2 CO 3 /slag ratio thus increased. The hydrotalcite-like compound was identied as a carbonate-free layered double hydroxide, containing OH À interlayer anions, in agreement with recent thermodynamic modelling for similar systems. Carbonate appears to be sequestered initially within gaylussite, which progressively dissolves and converts to calcite over time.
The added Mg(OH) 2 appeared only to act as a ller and nucleation seed, reducing the time taken for major strength forming phases to develop, while undergoing little to no further reaction with any other components within the system. No magnesium bearing phases were identied as rims around the Mg(OH) 2 particles, suggesting that the nucleation seeding primarily inuenced C-(N-)A-S-H (and maybe N-A-S-H) gels.
Overall, a sodium carbonate activated slag matrix would be suitable for the encapsulation of Mg(OH) 2 -bearing wastes such as Magnox sludge. Little if any reaction would be expected to occur between the Mg(OH) 2 -bearing waste material and the cementitious matrix. The release of heat of reaction is accelerated when adding Mg(OH) 2 , though this does not increase the cumulative heat of reaction above that which is observed in Mg(OH) 2 -free samples. The slow hydration of these samples means they can be safely processed without risk of thermal runaway or heat-induced cracking in larger samples.

Conflicts of interest
There are no conicts of interest to declare.