Effect of silica particle size on the formation of calcium silicate hydrate

Calcium silicate hydrate (C-S-H) is made by mixing calcium hydroxide (portlandite), silica and water at two calcium oxide to silica (C/S) ratios, using nano-silica and silica fume. We investigate how silica particle size inluences the rate and extent of formation of C-S-H at room temperature by isothermal calorimetry, thermal analysis, Raman spectroscopy and X-ray difraction. Rate of reaction increases as silica particle size decreases, and is ive times greater with nano-silica than with silica fume. Final composition depends only on initial C/S ratio. Compositions at 28 days are estimated from thermogravimetric and X-ray difraction data. There is a weak maximum in the reaction rate of nanosilica pastes about 20–30h after mixing. The overall kinetics is well described by a simple exponential (irst-order) reaction model. The early-time reaction rate around the rate maximum is described by an Avrami model. calorimetry (DSC) data were collected using a NETSZCH STA 449 F1 Jupiter instrument. Measurements were made on 10–20 mg powder samples, under constant nitrogen low at a heating rate of 10 °C min −1 from 25 °C to 1000 °C. Results are plotted as weight loss (%) and heat low (W g −1 ) as a function of the temperature (°C). TG measurement results were also used to calculate the inal C/S molar ratio of the samples, according to the methodology of


Introduction
Increasing awareness of climate change and greenhouse gas emissions has led researchers to develop alternative materials to control the environmental impact of manufacturing industries. Cement production is responsible for about 7-10% of the total anthropogenic CO 2 emissions [1]. In recent years cement has been partially replaced by the addition of waste by-products such as ly ash and amorphous silica [2][3][4]1,5]. Fly ash is well established in the cement industry as an environmentally friendly replacement for cement at up to 35 wt% [6], without compromising the mechanical and physical properties of concrete [7]. Amorphous forms of silica, such as silica fume and nanosilica, have also been investigated. Silica fume is a by-product of the silicon and ferro-silicon industry, with a nominal particle size range of 100-1000 nm, while commercial nano-silica is ad-hoc synthesised from sodium silicate using a sol-gel technique. Recent investigations have also explored the production of 'green' nano-silica from waste dunite, which has a comparatively lower environmental footprint [8]. Silica fume has been shown to enhance the mechanical strength of Portland cement, reduce permeability and thermal conductivity [9,10,1], and accelerate the early hydration of C 2 S, C 3 A and C 4 AF [11], while nanosilica enhances strength from increased production of additional C-S-H [4], (Here we use the standard abbreviations C = CaO, S = SiO 2 , H = H 2 O generally employed in cement chemistry). The smaller particle size of nano-silica produces a correspondingly greater surface area to provide nucleation sites for the growth of C-S-H [12,13].
Thermal analytical techniques are extremely useful not only in understanding the hydration kinetics of cement, but also in providing reaction models to tailor cement paste for speciic applications. Kumar et al. [14] studied the efect of ly ash on the hydration of ordinary Portland cement by isothermal calorimetry and found that while the addition of 20 wt% ly ash retarded the hydration, this efect was negated by addition of 5 wt% of calcium hydroxide. Wei et al. [11] used TG/DSC to study hydrated cement modiied by silica fume. They found that silica fume accelerates the early hydration of cement and stabilizes ettringite, and therefore reduces damage to the set cement and concrete from delayed ettringite formation. However, when silica fume is combined with ly ash, the overall efect is a delayed hydration [15]. Oltulu and Sahin [16] studied the pore-structure development of cement mortars containing silica fume and other nano-powders. The addition of nano-silica resulted in an increased amount of C-S-H and other hydrated products. However, it has been shown that an increased content of silica fume in Portland cement produced in C-S-H with a lower C/S ratio [17]. Thermogravimetric analysis and calorimetry data have been used to determine the pozzolanic reaction of supplementary cementitious materials (such as ly ash) and to describe the cement or concrete hydration reaction, and validated results were useful in establishing models for the hydration of cement [18,19]. Although there are several experimental reports of the beneicial efects of amorphous silica added to cement, little research has been conducted on the efect of silica particle size on the formation of the C-S-H phases responsible for early strength and hardening [20]. This is partly due to the diiculty of accurately quantifying the C-S-H content when it is found among other hydrated products such as ettringite and calcium aluminate hydrates [20]. Nonetheless, the C-S-H formed in cement by hydration of alite (C 3 S) and belite (C 2 S) or in cement-alkali systems is morphologically diferent from synthetic C-S-H. In the irst case C-S-H assumes the typical ibrillar habit, whereas in synthetic formation it tends to form with a two-dimensional 'crumpled foil' habit in agreement with Richardson et al. [22]. A clearer understanding of C-S-H behaviour is obtained by isolating it from other accessory minerals, and the behaviour of pure C-S-H (made without the hydration of Portland cement clinker) is of signiicant relevance to the ields of nuclear waste and environmental contaminant trapping as C-S-H is the component which adsorbs.
In this work we prepared C-S-H in isolation from other mineral phases, and studied its formation by thermal analysis and isothermal calorimetry, supported by Raman spectroscopy and X-ray difraction. We synthesised C-S-H by reacting calcium hydroxide with silica and varying the calcium to silica (C/S) ratios. The efect of the silica particle size on the formation of C-S-H was investigated using two diferent silica sources: silica fume (particle size range 100-1000 nm) and nanosilica (particle size range 5-20 nm). We show how silica particle size and hydration kinetics afect the extent of reaction at 28 days and the inal composition.
C-S-H pastes were mixed manually in a nitrogen environment at room temperature. Calcium hydroxide, silica (either silica fume or nano-silica) and water were mixed at a liquid:solid weight ratio of 2.0. Two diferent target lime (CaO) to silica (C/S) ratios, 0.81 and 2.4 mol/ mol, were investigated ( Table 2). These mol ratios correspond to the ratios 1:1 and 3:1 portlandite:silica by weight. Specimens were cast in cube moulds, stored in a sealed environment at 20°C, 100% RH, and cured for 28 days. After curing, the samples were demoulded, immersed in an acetone bath for 24 h to arrest the hydration process [24,25], and then placed in a drying chamber under a nitrogen low. All samples were then ground by hand and sieved to obtain a homogeneous ine powder, and stored in a sealed environment.

Materials characterisation and thermal analysis
Powder XRD analyses were performed using a Bruker D8 Advance difractometer with CuKα radiation over the range 10-60°2θ, step size of 0.02°and scan rate of 0.5 s/step. The samples were also characterised by Raman spectroscopy, using a Renishaw InVia spectrometer equipped with Modulaser Stellar Pro ML/150 argon ion laser (785 nm). Raman shifts were recorded over a 9 point map (200 × 200 μm) and wavenumber range of 100-4000 cm −1 , using the 20× objective lens. The exposure time was 10 s, and each spectrum was accumulated three times. Speciic heat low and heat of reaction were measured using an isothermal calorimeter (I-Cal 4000 HPC, Calmetrix). Fresh paste (about 40 g) was cast into a cylindrical container and placed into the calibrated calorimeter, at a constant temperature of 21 ± 2°C. The heat low was recorded over 200 h. Samples were mixed externally, hence the irst 5-10 min of detected heat low were not taken into account for further calculations. Thermo-gravimetric (TG) and diferential scanning  Results are plotted as weight loss (%) and heat low (W g −1 ) as a function of the temperature (°C). TG measurement results were also used to calculate the inal C/S molar ratio of the samples, according to the methodology of Garbev et al. [26].

XRD characterisation and Raman spectroscopy
XRD patterns were obtained on specimens of all four mixes at 28 d (Fig. 2).
These were intended primarily to provide information on mineralogy and structure, but quantitative peak-area analysis also provides support for the TGA composition results. The XRD patterns show the main features expected of C-S-H(I) [26][27][28][29], namely a strong relection at ∼29.3°2θ, and other prominent relections at ∼34.1°, ∼49.7°. These correspond to the (020), (200) and (240) relections of tobermorite [27]. The (002) basal relection at ∼6.6°was outside the scan range and was not observed.
In NS24, SF81 and SF24, portlandite was present, with principal peaks at the positions shown in Fig. 2. In NS81, all of the portlandite in the original mix has been consumed, and we have an uncontaminated C-S-H pattern of low C/S ratio. We note that in all patterns the strong (020) peak at 29.3°is markedly asymmetric. The weak (101) relection at ∼16.1°2θ is visible only in the low C/S mixes NS81 and SF81, an observation consistent with the comment of Tajuelo Rodriguez et al. [28] who ind that this peak decreases in intensity as the C/S ratio increases. Likewise the broad relection at ∼44°is visible only in the NS81 pattern. A comparison of pseudo-Voigt peak areas allows us to estimate approximate compositions from the XRD patterns, using the NS081 pattern which contains no portlandite as a C-S-H reference. The estimated compositions from XRD are in good agreement with those from TGA ( Table 6).
The Raman spectra of the samples are reported in Fig. 3 with peak positions given in Table 3.
The irst low frequency bands in the range 200-305 cm −1 correspond to the Ca-O lattice vibrations in calcium carbonate. As demonstrated by Black et al. [30], these bands are only visible in samples which have begun to carbonate through exposure to air and are not present in fresh paste sealed from the environment. One is weakly present in NS81 at 279 cm −1 but it is poorly resolved compared to the bands at 270 and 302 cm −1 shown by Black et al. and our assignment is tentative. Carbonation is more obviously demonstrated in NS24, the sample with the highest C/S ratio and which contains 41 wt.% portlandite, by the doublet at 1075 and 1085 cm −1 . Black et al. attribute this to the ν 1 [CO 3 ] mode but it is signiicantly weaker than the very strong bands they ind in samples exposed for 1 month. We conclude that our samples are lightly carbonated which its with our sample preparation procedure. The presence of portlandite is clear from the intense peaks at c. 355 cm −1 in the spectra for samples NS24 and SF24, which are assigned to Ca(OH) 2 external translation mode, A 1g showing that the samples have not signiicantly carbonated as these bands are barely present after 40 hours of exposure in the carbonation study by Black et al. Ca-O lattice vibrations from C-S-H gel are described as occurring at 319-333 cm −1 in fresh samples of C/S ratio 0.5-1.5 [31] which broaden and move to lower frequencies upon carbonation, possibly due to weakening of the Ca-O bonds during C-S-H decalciication [30]. We see the band increase in frequency from 316-332 cm −1 as the C/S ratio increases from 0.79 (NS81) to 1.70 (NS24) which is in line with Garbev et al. The band at 445 cm −1 , attributed to O-Si-O bending vibrations in SiO 4 where the oxygen is non-bridging is present at 446 and 447 cm −1 in this work and is relatively weak, as shown in Fig. 3. The strongest peak present in our spectra at 666 cm −1 is produced by Si-O-Si symmetric bending vibrations in C-S-H which have been shown to broaden with increasing C/S ratio from the presence of silicate dimers [31]. This efect can be seen in Fig. 3 as the peak width increases (FWHM) from 28 (NS81, C/S = 0.79) to 35 (NS24, C/S = 1.70). A very weak peak at ∼885 cm −1 is visible in the higher C/S ratio samples (NS24 and SF24), which is assigned to the symmetric stretch of SiO 4 ) dimers (Q 1 ) by Garbev et al. who show it increases in intensity with   increasing C/S ratio and is barely visible when the C/S ratio is less than 0.83. The peak at ∼1004 cm −1 in Fig. 3 is strongest but also broadest in the spectra for NS81. The position is determined manually as the peak has a long asymmetric tail after the apex. This band is assigned to the symmetric stretch of SiO 4 chains (Q 2 ) and is described by Garbev et al. [31] as increasing in frequency with C/S ratio to a inal value of 1022 cm −1 in samples with C/S 1.33 and 1.55. We see an increase in frequency from c. 1004 cm −1 at C/S = 0.79 to 1018 cm −1 at C/ S = 1.70. The signiicant asymmetry of the peak in NS81 can be attributed to disorder in the silicate polymerisation and presence of different silicate species, such as long chains, octamers and pentamers [31]. The weak and comparatively narrow bands in NS24 and SF24 suggests only a few, uniform, Q 2 species are present [31]. As already discussed, CO 3 symmetric stretch (ν1) modes occur at ∼1080 cm −1 and are more clearly present in the samples with signiicant portlandite (SF24 and NS24) but are still weak by comparison with spectra from the 1 month old samples described by Black [30]. By comparison with NS spectra, the SF spectra, in particular SF81, are surprisingly featureless. Overall, the Raman spectra conirm the XRD and TGA evidence that there is only slight carbonation in the samples.

Isothermal calorimetry
The calorimetric heat low rate (Fig. 4) shows the following features: 1. An initial period of a few minutes in which the heat low rate is high but falls rapidly; the heat evolved over this short initial period no doubt includes contributions from wetting and dissolution as well as early-time reaction, and also non-isothermal efects in the calorimeter. 2. A period of some hours in which the heat low rate slowly decreases until a minimum is reached (the slow-rate or decelerating rate period). 3. After the minimum, the heat low rate increases (the accelerating rate period) for some hours until it reaches a maximum. 4. Beyond the maximum, the heat-low rate decreases slowly but in all cases is still measurable at 200 h.
We follow the usual practice of treating the calorimetric heat-low rate after the initial period as a direct measure of the rate of formation of C-S-H. The features we identify are similar in several ways to those observed in calorimetric data for the hydration of tricalcium silicate C 3 S [32,20,33,34]. As here, C 3 S hydration also shows a so-called slow-reaction period, followed by an accelerating period, which in turn is followed by a inal extended period of decreasing heat low rate. On the other hand, although the early-time minimum and maximum in the reaction rate are well deined, at least in the nano-silica systems, they are much less prominent than in C 3 S hydration where the minimum heat-low rate is close to zero [33]. Apart from these small variations in the hydration rate at early times, we ind the overall hydration kinetics are accurately described by a simple exponential model. This is readily seen by considering the cumulative heat evolved (the area under the heat low rate curve), which we can it to the equation where H(t) is the cumulative heat evolved at time t, H 0 is the heat of reaction, k is a rate constant, and b is a constant to take account of initial heat-low spike. Eq. (1) is the standard irst-order model of solidstate reaction kinetics [35]. Fig. 5 shows that using data from about 50 h onwards allows us to extract values of rate constant k and heat of reaction H 0 for the four mixes. The parameter values are given in Table 4. The efect of particle size is clear from our (admittedly limited) data. The rate of hydration is much faster in the NS mixes than in the SF mixes, with k values being about ive times smaller. This is approximately the same as the ratio of the speciic surface areas (see Table 1), which is ∼5.8. The duration of the reaction can be estimated as τ = 4.61/k, at which time the reaction is 99% complete. This shows that the NS mixes have fully reacted at 28 days. The SF mixes have only reached ∼60% of their inal composition at 28 days, and take 4-5 months to reach 99% completion. The parameter H 0 is also informative. The NS81 and SF81 mixes have the same silica content per unit weight of paste (Table 4), and we see that the ultimate heat of reaction which is the same for both does not depend on particle size. Likewise for the NS24 and SF24 systems. On the other hand, the NS81 mix contains exactly twice as much silica per unit weight of paste as the NS24 mix (see Table 4), but H 0 (per g of paste) of NS81 is signiicantly less than twice that of the NS24 mix. The observed ratio is 1.51. This suggests that there is unreacted residual silica in the NS81 systems, and that the true C/S ratio is greater than the nominal (molar) target value of 0.81. The same comparison can be made between the SF81 and SF24 mixes, with the same conclusion. Expressed per mol SiO 2 , H 0 is essentially the heat of reaction, which is smaller for the NS81 and SF81 mixes than for the NS24 and SF24 mixes. Once again, this allows the possibility that the NS81 and SF81 mixes contain unreacted silica at the end of the reaction, and that the true C/S ratio of the C-S-H formed is greater than the nominal value. The heats of reaction are comparable with those reported for other C-S-H forming systems, but further thermodynamic analysis is outside the scope of this study.

Early-time behaviour and the rate maximum
The similarities between the hydration of the silica/portlandite system described here and that of the C 3 S system are noteworthy. The same maximum in reaction rate can be seen in the calorimetric data on nano-silica/portlandite systems reported by Lin et al. [36], although it is not discussed. The existence of an early slow-reaction period is generally regarded as a particular feature of the hydration of calcium silicate systems. Yet it is observed here in a system where the Ca concentration in solution is bufered by the presence of excess solid portlandite, and where the C-S-H can only form by transport of Ca through solution to the silica surface. This is in sharp contrast to the case of C 3 S hydration where both Ca and Si are co-located, and only water is required for reaction; furthermore, because the C/S ratio of the C-S-H product is < 3, there must be a net outward difusion of Ca into solution, rather than an inward transport as here. In the C 3 S case, the accelerating-rate period following the minimum is generally described by means of a nucleation and growth [NG] mechanism, often using the Avrami or Cahn models, or variants. NG models ascribe the hydration kinetics to the interplay of nucleation rates and crystal-growth rates, and assume that simple difusive transport is not rate-controlling. It is however recognised that at long times the reaction rate is likely to be controlled by difusion of reactant species through the C-S-H to the unhydrated surface beneath. Thus the long tail in the C 3 S heat-low rate at extended times is not represented by NG models. We ind that the early-time heat low data for the nano-silica mixes NS81 and NS24 can be represented by a simple stretched-exponential Avrami model for the extent of reaction X at time t where k 2 is a rate constant, and m is a constant. The heat low rate R is proportional to dX/dt, so that for this model where t 0 is a time-ofset and a is a constant. We show results in Fig. 5b, with it parameters given in Table 5.
In the data analysis, we have calculated the diference between the measured cumulative heat low and the simple exponential model, and diferentiated this numerically to obtain the non-exponential contribution. That we can it the data with such a model does not have any clear mechanistic implications, but the existence of the rate minimum and maximum is unexpected, and future work may provide an explanation. The values of the constant m are rather lower than the typical Avrami values of 2-3. We note that of course the Avrami model (in its site-saturation form) reduces to our simple exponential form when m = 1.

Thermal analysis
TG/DSC curves of C-S-H samples hydrated for 28 days are shown in Fig. 6. Thermogravimetric curves are typical of C-S-H with a total weight loss of 30-35%. The irst thermal step is observed in the range 25-200°C and corresponds to the evaporation of pore and capillary water and dehydration of C-S-H [26,4]. The second weight loss occurs between 350-550°C and is associated with the dehydroxylation of portlandite. The thermal step in the range 600-700°C is attributed to the loss of CO 2 from calcium carbonate, due to the surface carbonation of samples during instrument operations [26]. As noted in [28], there is a small continuous weight loss over much of the temperature range, and this is attributed to continuing dehydration of C-S-H. TG results are summarised in Table 6. The weight loss values for portlandite and calcite were found by a simple tangent method [37]. The weight loss was obtained as diference in weight at the intersections of the tangent drawn at the point of maximum slope in the weight loss curve and straight lines itted to the data above and below the decomposition.
At 28 days, samples made with nano-silica contain more C-S-H than samples made with silica fume. In agreement with Raman and XRD analysis, at the same initial stoichiometric C/S ratio, nano-silica promotes the formation of C-S-H and no Ca(OH) 2 is detected in the mix NS81. Additionally, the portlandite content, proportional to the weight loss, in %, in the range 350-550°C, is higher in samples made with silica-fume, where it is up to 40 wt%. TG data were used to calculate the C/S ratio of hydrated samples (inal C/S ratio at 28 days of hydration), according to the methodology described in Garbev et al. [38]. Results are reported in Table 7. Using nano-silica instead of silica fume as the silica source, results in an overall higher inal C/S ratio. This relects the higher reactivity of nano-silica and its ability to form additional C-S-H. The calculated C/S ratio values should be taken more as a guide than absolute values. In sample NS81, while the reaction went to completion at 28d, the calorimetry data suggests some unreacted silica, which indicates the C/S ratio is higher than the nominal value. Sample NS24, the reaction was also complete by 28 days and the system had an excess of calcium, expressed as wt% portlandite, so it is unlikely there is any unreacted silica. The C/S value of 1.7 is higher than anticipated from a synthetic C-S-H, where an upper value of 1.6 [39] is expected, this could be explained by the slow decomposition of some nano-crystalline   Time-ofset t 0 = 12 h.

R. Maddalena et al.
T h e r m o c h im ic a A c t a 6 7 2 ( 2 0 1 9 ) 1 4 2 -1 4 9 portlandite trapped within the C-S-H pores. Calorimetry data shows that at 28 days, both silica fume samples had not inished reacting and therefore the C/S ratio is likely to be higher than calculated due to unreacted silica. The C/S ratio plays a key role in the hydration kinetics of Portland cement and its strength development [40]. In hydrated Portland cement, it varies in the range 0.6-2.0 mol/mol, and many authors have found that mortar and concrete pastes modiied by the addition of pozzolanic nano-particles produce a denser C-S-H with a lower C/S ratio [41][42][43]. Results presented here show that at 28 days, which is the standard curing time for most applications, the C/S ratio is increased by smaller particles. This is due to the enhanced reactivity of nano-silica compared to that of silica fume. C/S ratio is a key parameter for speciic application and not only in cement industry: C-S-H can be crystallographically tailored to incorporate radionuclides or heavy metals into its structure [44,45], or could be expanded to increase porosity and be used in waste-water treatment as an adsorbent material [46,47]. Hence, understanding the efect of ine additives on the C/S ratio and quantity is valuable.
Results of the diferential scanning calorimetry of samples hydrated for 28 days (curves shown in Fig. 6) are summarised in Table 8. DSC curves show the irst endothermic event at about 80°C, corresponding to the formation of C-S-H and evaporation of pore water. The integrated peak area is greater in sample NS81 (603 J g −1 ) than sample SF81 (586 J g −1 ). Correspondingly, the endothermic event at about 430°C, associated with the decomposition of Ca(OH) 2 , is greater in samples with higher nominal C/S ratio and/or larger silica particle size range (sample NS24, 195 J g −1 ; sample SF24, 533 J g −1 ). Furthermore, the higher C/S ratio causes a shift of the DSC peak, which represents portlandite breakdown, towards a higher temperature, and close to the peak centre of pure portlandite, at 460°C. The DSC response of sample NS81 exhibits a narrow exothermic peak at around 800°C, attributed to the crystallisation of wollastonite [48,49,44,28].

Conclusions
We report the results of a study of the formation of C-S-H by reaction of amorphous silica and portlandite at low and high C/S ratio (0.81 and 2.4 mol/mol). Two diferent sources of silica were used: silica fume (particle size range of 100-1000 nm) and nano-silica (particle size range of 5-20 nm). Our analysis of experimental data lead to the following conclusions.   Notes: Calcite was not determined by XRD analysis.

Table 7
Mix compositions at 28 d and initial (nominal) C/S molar ratio of mix, and calculated C/S molar ratio of C-S-H, from thermogravimetric analysis.  Note: NS81, no portlandite detected.

R. Maddalena et al.
T h e r m o c h im ic a A c t a 6 7 2 ( 2 0 1 9 ) 1 4 2 -1 4 9 reaction rate on the particle size. 5. Calorimetric data show that the overall kinetics is well described by a simple exponential model. This allows the timescale for the completion of the reaction to be estimated. 6. Heat low rate data show the existence of a weak minimum and maximum in the rate of reaction at 20-30 h after mixing. This feature is similar to the well known induction period in the hydration tricalcium silicate C 3 S, although the latter is more pronounced. Heat low rate data around the rate maximum can be described by a sitesaturation Avrami model, with m ∼ 1.5. The rate maximum is unexpected and deserves further study. 7. Mass balance calculations using TG and XRD composition data provide estimates of the C/S ratio of the C-S-H formed, although in the absence of any direct information on possible unreacted silica, these C/S ratios are nominal lower limits. Nonetheless, there is strong indication that the C/S of the C-S-H product increases as the C/S of the initial mix increases.