Influence of temperature and importance of sulfate content on the triethanolamine-enhanced aluminate reaction during Portland cement hydration Open Access

OPC is a key ingredient in civil infrastructure. To be more resource efficient and reduce carbon dioxide emissions the OPC reaction rate has to be optimised. Many studies on the hydration reactions of OPC are available in the literature. For more detailed information, the reader is advised to read several review articles (Bullard et al., 2011; Scrivener et al., 2015, 2019).

OPC clinker mainly consists of silicate phases (tricalcium silicate (C₃S) and dicalcium silicate (C₂S); ∼77 wt% for the OPC used in this study) and aluminate phases (tricalcium aluminate (C₃A) and tetracalcium aluminoferrite (C₄AF); ∼10 wt% in this study).

When cement containing C₃A is mixed with water, fast setting occurs. For better workability, a specific amount of sulfate carrier is added to the clinker to prevent flash setting. The reaction process of a neat OPC paste is shown in Figure 1. The overall reaction is generated by two different reactions: (a) the silicate reaction in which C₃S reacts with water to form a calcium silicate hydrate (C-S-H) phase and calcium hydroxide and (b) the aluminate reaction where C₃A reacts with water and sulfate to form ettringite (Taylor, 1997).

Upon first contact of water and cement, small amounts of C₃A and sulfate dissolve and ettringite precipitates. Myers et al. (2017) proposed that, after this initial event, C₃A dissolution is inhibited due to passivation by calcium–sulfur ion pairs on an aluminium-rich leached layer. The ettringite formation from dissolved ions in the pore solution continues slowly. During this phase, the silicate reaction occurs. The common understanding is that when sulfate ions in the pore solution are nearly fully consumed, sulfate depletion takes place and a further aluminate reaction is visible as a shoulder during the deceleration period of the C₃S reaction (Jansen et al., 2018).

The degree of hydration (DoH) of C₃A within the first 24 h of an average CEM I was calculated to be around 40–50% (Stabler, 2018). For reasons of sustainability, cost reduction and improved performance, it is desirable to enhance this reaction degree to produce more ettringite. For this reason, different admixtures have been investigated (Diamond, 1986; Dorn et al., 2022; Myrdal, 2007; Taylor, 1997), and triethanolamine (TEA) turned out to be one of the most promising. At appropriate dosages, TEA is able to change the sequence of the reaction process. At very high dosages, the aluminate reaction is strongly accelerated and is shifted before the silicate reaction. Furthermore, TEA enhances the DoH of C₃A and C₄AF (Lu et al., 2020). For higher ettringite formation, the sulfate content must be increased because the sulfate supplied by OPC is the limiting factor in the formation of ettringite. Otherwise, low-sulfate hydration products like monosulfate or sulfate-free phases (e.g. C₃AH₆, C₂AH_x) can be formed (Breval, 1976; Brown et al., 1984). (Kirchberger et al. (2023) first proposed the combination of TEA as an accelerator with tartaric acid (TA) as a retarder and increased sulfate content to form the maximum possible amount of ettringite from nearly fully reacted C₃A and C₄AF within the first few hours. With this innovative additive mixture, the hidden potential of OPC can be revealed, leading to a reduction in carbon dioxide emissions. Later, Li et al. (2024) reported on a fully developed commercial additive composed of several ingredients, including TEA and increased sulfate content, for the formation of high ettringite amounts at early ages.

Depending on the dosage, TEA has different effects on the hydration of OPC and the dissolution of its main phases. Ramachandran (1973b) reported a comprehensive study on the influence of TEA on the hydration of OPC for TEA concentrations of 0.1–1.0%. The results showed an accelerating effect for C₃A and C₃A–gypsum (C$H₂) systems but a prolonged induction period for C₃S associated with the formation of TEA-complex layers on its surface. The dosage that is needed to have the aluminate reaction before the silicate reaction also depends on the type and amount of sulfate carrier (Hirsch et al., 2021a, 2021b; Kirchberger et al., 2023). This is due to the acting mechanism of TEA on the hydration of C₃A. (Ramachandran, 1973) proposed that TEA competes around the C₃A surfaces with an impermeable sulfoaluminate layer that normally retards the C₃A reaction. More recent models deal with a similar competitive effect between TEA and calcium–sulfur ion pairs that adsorb on C₃A surfaces, resulting in retardation (Myers et al., 2017 When these calcium–sulfur ion pairs cannot adsorb onto the surfaces of C₃A due to the addition of TEA and its ability to form complexes (with calcium and aluminium especially), acceleration of the aluminate reaction is the result. To completely override the retarding mechanism of calcium–sulfur ion pairs, an adequate amount of TEA must be added in relation to the calcium sulfate content, also depending on the kind of sulfate carrier. Due to the faster dissolution, a greater amount of TEA is needed when hemihydrate instead of anhydrite is chosen as the sulfate source (Kirchberger et al., 2023).

Temperature-dependent effects on the use of TEA are sparely described in the literature. Recently, Wang et al. (2023) studied the effect of TEA and triisopropanolamine at 23, 32 and 50°C. They reported a faster setting behaviour with increased temperature for both alkanolamines. To the authors’ knowledge no studies on the performance of TEA as an accelerator on OPC hydration at low temperatures have been published.

Fruit acids like TA are often used as retarders for cement hydration (Double, 1983; Wilding et al., 1984). TA is known to be an effective retarder for the hydration of C₃A, the formation of ettringite and the hydration of C₃S (Bishop and Barron, 2006). However, higher dosages of TA can lead to early onset of the aluminate reaction, comparable to TEA but with a reduced reaction rate. This effect was described by Stabler (2018) for TA concentrations >0.28 wt%. Often discussed retardation mechanisms are the formation of calcium tartrate, which adsorbs/precipitates onto clinker surfaces (Bishop and Barron, 2006; Rai et al., 2006). A recently published study by Costa et al. (2023) describes the effect of a commercial product consisting of hydroxycarboxylic acid salts/compound carbohydrates. It was found that the retarding effect decreased with an increase in temperature. Sun et al. (2022) studied the effect of TA on oil well cements at very high temperatures (up to 89°C) and noted the same effect – less retardation at increased temperatures. However, there appear to be no published works on the behaviour of TA at low temperatures.

The system studied in this work was a mixture of OPC and anhydrite, optimised with organic additives Kirchberger et al. (2023). Surplus sulfate is needed so that the system forms the maximum amount of ettringite when the aluminate phases are exhausted. If the sensitive balance between sulfate carrier addition and aluminate phases is somehow disturbed, it is possible that some sulfate remains. Therefore, potential risks like thaumasite formation have to be considered.

Thaumasite (CaSiO₃$.$CaCO₃$.$CaSO₄$.$15H₂O) is commonly a product of sulfate attack in mortars and concretes when a carbonate source and sulfate are available; it leads to expansion and other risks (Crammond, 2002). Barnett et al. (2002) reported on solid solutions between thaumasite and, which involves the replacement of silicon by aluminium and the replacement of sulfate/carbonate by sulfate/water.

Studies on the formation of thaumasite by gypsum addition and the addition of sulfate-rich solutions are available (Gaze, 1997; Hartshorn, Sharp and Swamy, 2002). Temperature is also known to influence formation of thaumasite. The stability and formation of thaumasite is often reported for temperatures below room temperature (Bensted, 1999; Kakali et al., 2003; Pipilikaki et al., 2008). In field studies, thaumasite has also been found in buildings exposed to higher temperatures (Collepardi, 1999).

The aim of this work was to examine an earlier studied mixture (OPC, 7.4% anhydrite, 2.2% TEA and 0.25% TA) at three different temperatures (10, 23 and 40°C). This special mixture was adjusted in such a way that the aluminate reaction takes place before the silicate reaction and, beyond this, the DoH of the aluminate phases is increased to over 90%. The addition of extra sulfate ensures that ettringite is formed.

The performance of this system at different temperatures needs to be analysed and evaluated in detail. This study of an exemplary formulation offers very valuable and new information for its application in different cementitious materials, such as tile adhesives or self-levelling floor compounds, which are installed at higher or lower temperatures depending on the time of application (summer or winter times) or the location of application (cold or hot regions).

In general, studies on admixtures at low temperatures are rare. Another question to be answered is whether the performance of a mix at different temperatures can be optimised by changing the retarder (TA) dosage.

An important issue, which will be clarified in this paper, is the sensitive aluminate–sulfate balance in such accelerated mixes and the potential risk of small changes in reactivity or content of phases leading to the potential formation of undesired phases such as thaumasite.

The experiments were conducted at 10°C, 23°C and 40°C using deionised water. For paste preparation, the TEA, TA solution, binder mix, spatula and stirring tool were stored separately in order to equilibrate. Before stirring, the TA solution was combined with the TEA. The paste was stirred using an IKA Eurostar 20 instrument with a tailor-made four-bladed propeller stirrer at 1350 rpm for 1 min, followed by a 30 s pause and another 30 s stirring. For all the hydration experiments at least two independent samples were prepared and evaluated. In the figures presented in this paper, the error is indicated as lighter areas behind the curves or as error bars. If no lighter area or error bar is shown, the error was smaller than the curve/symbol.

Gypsum with >99.9 wt% purity was used for the synthesis of anhydrite. The process involved gradually heating the gypsum from room temperature to 450°C at a rate of 180°C/h. Subsequently, the temperature was maintained at 450°C for 16 h. The Brunauer–Emmett–Teller surface area was determined to be 16.8 ± 0.1 m²/g. The synthesis conditions were selected to maximise the content of anhydrite II content and its reactivity. Anhydrite III occurs as a by-product, which converts to hemihydrate at ambient humidity (Seufert et al., 2009).

The OPC used in this study was a commercial CEM I 52.5R. The OPC was characterised by quantitative X-ray diffraction (QXRD) and X-ray fluorescence (XRF). To guarantee a precise phase analysis from XRD, selective phase dissolution experiments were executed (Gutteridge, 1979; Struble, 1985). In the following, the term C₃A always means the content of cubic and orthorhombic C₃A. Tables S1 and S2 in the online supplementary material provide detailed information about the XRD experiments and the structures used. The XRF and XRD results are provided in Table 1.

The mixture developed by Kirchberger et al. (2023) was used in the experiments. OPC and 7.4% anhydrite were mixed as powder, then mixed at a w/b ratio of 0.4063 with different TA concentrations in water, which was added to 2.2% TEA shortly before measurements. The binder in the w/b ratio is defined as the sum of hydraulic active substances (OPC and anhydrite), with each ingredient added based on the weight of cement (bwoc). The compositions of the different mixtures are detailed in Table 2. The TA dosage was adjusted for the tests at 10°C and 40°C to assess whether the timing of the early aluminate reaction was still adjustable independent of the temperature. In a previous study by the authors, it was found that, at 23°C, an increase in TA led to an increase in retardation of the aluminate reaction (Kirchberger et al., 2023).

To determine the phase content after defined points of time, stored samples were prepared and analysed by QXRD. After mixing, the pastes were transferred into plastic containers, and the lids were sealed using parafilm to prevent air ingress and reduce the evaporation of water. The freshly prepared samples were placed in a temperature-controlled environment at 10 ± 1°C, 23 ± 1°C or 40 ± 1°C. Even though the samples were sealed, evaporation might still have occurred especially at 40°C. Therefore, the containers were stored at 100% relative humidity. The plastic container was placed in a 60 ml box which was placed in a 120 ml box. The 120 ml box was filled with ∼1 cm of water and closed; the water and the sample were not in contact (Figure 2). After defined points of time the top layer of the sample was removed using sandpaper. Subsequently, the phase composition was analysed by QXRD.

XRD analysis was performed on the dry samples (characterisation of cement and synthesised phases) using the front-loading method. However, the hydrated pastes were placed in a sample holder and covered with a Kapton polyimide film (7 µm) to prevent carbon dioxide ingress and water evaporation during analysis. The XRD measurements were conducted using a Bruker-AXS D8 diffractometer equipped with a LynxEye detector. Specific instrumental parameters and diffractometer settings used in the analysis can be found in Table S1 in the online supplementary material.

Rietveld refinement was used in combination with the G-factor method (Jansen et al., 2011) for quantification. The diffractograms were refined using Topas 5.0 software using items from the Inorganic Crystal Structure Database (FIZ Karlsruhe) shown in Table S2 in the online supplementary material. After hydration, an unknown phase was found, with its main peak located at ∼10.1° 2θ. This is not described in the literature but can be fitted with the structure file of hemicarbonate. Therefore, the c-lattice parameter needed to be refined.

Isothermal heat flow calorimetry was employed to follow the hydration process of all samples, using a TAM Air calorimeter at constant temperatures (10, 23 and 40°C). Before mixing, the powders and the liquid phases were individually equilibrated. After stirring, 1–2 g of the resulting paste was placed into a PET sample container and sealed with a Teflon lid. Data recording commenced immediately after the first contact between water and powder. For data acquisition, a frequency of 10 000 data points per 24 h was selected, resulting in intervals of 9 s between data points.

Thermodynamic modelling was performed to predict the stable hydration products based on the reacted cement phases determined by QXRD after 28 days. Therefore, Gems in combination with the cement database Cemdata18 was used (Kulik et al., 2012; Lothenbach et al., 2019; Wagner et al., 2012). For modelling the predicted phase composition after 28 days, only the amount of dissolved phases detected by QXRD were used as input to Gems.

For further thaumasite modelling, CEM I used was incorporated into Gems based on the XRF measurements. A complete reaction with different amounts of added anhydrite (6.6% bwoc (well balanced), 8.6% bwoc and 10.6% bwoc) was modelled to check which phases were calculated as stable if the sulfate balance was not well chosen. In addition, environmental exposure and/or the presence of reactive limestone was simulated by the addition of 0–5 wt% calcium carbonate (CaCO₃) and by providing sufficient water.

Isothermal heat flow calorimetry was performed at 10, 23 and 40°C. Figure 3 shows the heat flow curves of the mixes with 0.25% TA at the three different temperatures. On the left side the x-axis is shown from 0.5 to 10 h and on the right side from 10 to 120 h. For better visualisation the y-axes have different scales because much higher heat flow was recorded during the aluminate reaction (Figure 3(a)) than during the silicate reaction (Figure 3(b)). The aluminate reaction of the reference system (23°C) started after 2 h 5 min. When the temperature was lowered to 10°C, the reaction was delayed to ∼5 h 50 min. For 40°C the reaction is accelerated and started at 1 h 15 min. The same trend was visible for the silicate reaction: the reference system reached its maximum after 29 h, at 10°C it was delayed to 71 h and at 40°C it was accelerated to 19.5 h. Not only did the start of the reaction change with temperature, but also the peak shapes were altered. The lower the temperature, the later and the more prolonged the reactions.

The question then arises as to whether the retarder can be optimised for each temperature. Therefore, a higher retarder dosage of 0.35% TA for 40°C and a decreased retarder dosage of 0.10% TA for 10°C was investigated. The results are shown in Figure 4, and the expected effect was achieved. With 0.35% TA at 40°C the aluminate reaction (Figure 4(a)) was retarded by around 2 h and the silicate reaction (Figure 4(b)) was retarded by around 3 h. A decreased TA concentration at 10°C led to a ∼6 h earlier aluminate reaction whereas the silicate reaction maximum was accelerated by ∼15 h.

Stored samples were analysed for a better understanding of the hydration process and analysis of the long-term stability. The start values in Figure 5 were calculated from the powder/water ratio. After the aluminate reaction (marked ‘a. alum.’ in the figure) means a sample was measured as soon as the aluminate reaction was nearly completed (e.g. for 23_025, 3 h). The exact points of time are listed in Table 3. To include the main peak of the silicate reaction from all systems, the next point of time was 7 days. The last analysed point was 28 days to obtain data if the formed phases were stable over a longer period. Figure 5(a) shows the phase contents at 10°C for 0.25% TA (solid line) and 0.10% TA (dotted line). Figure 5(b) shows the phase contents at 23°C for 0.25% TA. Figure 5(c) shows the phase contents at 40°C for 0.25% TA (solid line) and 0.35% TA (dotted line).

For all the systems, the C₃A content decreased strongly during the aluminate reaction, together with calcium sulfate (anhydrite) with the isochronous formation of ettringite. For the all systems, an ettringite content of 20–22.2 wt% was reached. The expected maximum after the complete aluminate reaction was 27 wt% ettringite. After 7 days, the anhydrite was fully consumed and C₃A was strongly decreased. For the samples at 10°C and 23°C, an increase up to 25.6 wt% ettringite was determined, it was slightly decreased in the 40°C samples. After the aluminate reaction, a small amount of OH–SO₄–CO₃ phase of not exactly known composition formed. From 7 days to 28 days, the ettringite content further decreased in the 40°C samples and, for the 10°C and 23°C samples, a smaller amount of ettringite (compared with the after aluminate reaction) was determined. C₄AF was not noted here because the content in the cement used was always below 1.5 wt% and it dissolved together with C₃A. The reaction degree of C₃A ranged from 77% (10_025), to 91% (23_025), up to 96% (40_035).

As shown by the heat flow calorimetry results, the silicate reaction started a few hours after completion of the aluminate reaction. Nevertheless, around 3–5 wt% of initial C₃S dissolution was noticeable in all the mixes, but no hydration products were observed. This changed up to 7 days, when a calcium (alumino) silicate hydrate (C-(A)-S-H) phase and calcium hydroxide were detected. The lower than expected content of calcium hydroxide is a well-known fact when TEA is used as an accelerator (Kirchberger et al., 2023; Zhang et al., 2016). The DoH of C₃S after 28 days was in the range 75–79%.

Gems modelling was applied to check if the phase composition determined by QXRD after 28 days had already reached thermodynamic equilibrium or which phases were calculated to be stable. Therefore, the amounts of reacted phases were determined by QXRD and provided as input for the software for thermodynamic calculation. Organic components (TA and TEA) were not included in the Gems calculation because no data was available. TEA is a complexing agent for ions like calcium, aluminium and iron but, as described by Lu et al. (2020), these complexes seem to make their complexed ions available during the rapid formation of ettringite. Lu et al. (2020) showed that the TEA concentration in the pore solution decreases with the strong aluminate reaction. They also described the consumption of TEA from the pore solution by the formation of hydration products/surfaces at later ages, when the silicate reaction proceeds (Lu et al., 2020). Therefore, it was assumed that no significant amount TEA remained in the pore solution after 28 days.

For each temperature, the data measured for the mix with 0.25% TA were chosen. Figure 6 shows two columns per temperature. The left-hand columns show the phase compositions measured by QXRD after 28 days. The right-hand columns (shaded with lines) show the phase compositions modelled by Gems. The upper part of the diagram shows the region up to 100 wt% whereas the lower part zooms in to show the region from 0–3 wt%. Comparing both columns, the calculated reaction products of the silicate reaction were higher than those measured. The hkl model calculated for the quantification of C-S-H was developed for neat C₃S pastes (Bergold et al., 2013). Modification of the Ca/Si ratio caused by TEA cannot be taken into account. The changed C/S ratio is also responsible for the under-quantification of calcium hydroxide by QXRD because, with the addition of TEA, calcium hydroxide grows in a less crystalline way (Zhang et al., 2016). With respect to the aluminate reaction, there were several differences between the measured and modelled results. The major part is the unknown phase formed during hydration. The composition was calculated approximately using a mass balance approach in the previous study (Kirchberger et al., 2023). This phase takes up ions which are then not available to form ettringite and cannot be found in the Gems columns since it is not defined in the database. The measured C₃A/C₄AF contents were low after 28 days but there were minor differences between the samples at different temperatures. C₄AF was only visible in 10_025 in very low amounts, and the remaining C₃A decreased with increasing temperature. This resulted in different calculated phase compositions. For 10_025, small amounts of thaumasite were calculated, while monocarbonate was predicted for 23_025 and 40_025. For 10_025, the available amount of aluminium from C₃A/C₄AF was not enough that all the sulfate could be transformed into ettringite.

As shown in Figure 3, the desired effect of pushing the aluminate reaction before the silicate reaction worked at all temperatures but differences in relation to the reaction rate were evident. The DoH of C₃A was influenced by the temperature, as shown in Figure 7(a). At 10°C, the DoH was lower for the aluminate reaction but increased between 7 and 28 days. It is well known that the reaction of OPC at low temperatures is slower than that at room and higher temperatures (Escalante-Garcia and Sharp, 2000; Lothenbach et al., 2007).

Comparing the calculated and measured amounts of ettringite (Figures 7(b) and 7(c)), for all temperatures, the measured ettringite decreased between 7 and 28 days. Overall, the amount formed at 40°C was lower than expected. This destabilising effect was also observed by Noor et al. (2024), who studied the hygrothermal stability of ettringite in OPC-rich ternary systems. In general, the destabilisation of ettringite over time in ettringite-forming systems is noted in the literature, even at room temperature. The formation of a ternary AFm phase was described by Nehring et al. (2018), who studied a ternary OPC–CAC–Anhydrite (C$) system. With a decrease in ettringite, an increase in hemicarbonate was found but no sulfate-bearing phase was identified. Nehring et al. (2018) argued that the hemicarbonate peak was shifted to lower 2θ values, which could mean that sulfate was incorporated in the structure. For all the systems studied in this work, the unknown phase was fitted with a modified hemicarbonate structure with an extended c-lattice parameter because its basal peak occurred between the monosulfate phases and hemicarbonate. Nevertheless, the phase was poorly crystalline with a broad basal reflex at 10.1° 2θ. A solid solution between monosulfate and hemicarbonate has been described in the literature (Matschei et al., 2007; Pöllmann, 2005; Seufert, 2011). Additionally, Xu et al. (2017) described the faster conversion from ettringite to plate-like monosulfate at 40°C.

After 7 days, the calcium sulfate was fully consumed, which seems to have led to a destabilising effect on the formed ettringite at 10 and 23°C. Figure 8 shows XRD plots of stored samples at 10, 23 and 40°C with 0.25% TA at 7 days and 28 days. The peak of the unknown phase did not change position between 7 days and 28 days. As a result of the aluminate reaction, ettringite was the dominant hydration product at elevated temperatures. The existence of ettringite at higher temperatures (>30°C) has also been reported in the literature (Hesse et al., 2008; Jakob et al., 2023; Scrivener et al., 2016; Zhang et al., 2018). It is possible that a less crystalline phase was formed that was not detectable by XRD. Further research needs to be done to check if the dissolution of ettringite continues with time and if other phases can be found after longer storage.

The thermodynamic modelling in Figure 6 shows that the addition of extra sulfate based on the assumption of the complete hydration of aluminate phases can lead to undesired phase development if complete hydration of the aluminate phases does not happen. In Figure 6, this is seen for the 10°C samples. The addition of TEA did not lead to a high enough DoH of the aluminate phases, which meant that the calcium sulfate was not completely consumed by the ettringite. An unknown phase was noted, which probably consumed sulfates. The thermodynamically stable phase was then ettringite, along with thaumasite. As already mentioned, sulfate attack comes with potential risks for hardened cement pastes. This sulfate attack can originate from sulfate-rich solutions or surplus sulfate carriers (Gaze, 1997; Hartshorn et al., 2002).

Figure 9 shows Gems modelling of the cement used in this study at 10°C assuming full reaction of all the included phases. The phase compositions of the dry powder and paste are shown in Table 1. When the sulfate content in the system was well balanced with the aluminate phase content (6.6 wt% anhydrite in total), thaumasite was not predicted to form. The balance of aluminium and sulfate leads to the maximum possible amount of ettringite. If the amount of sulfate is increased (e.g. 8.8 wt% or 10.8 wt% anhydrite in total), the excess sulfate is predicted to form thaumasite with available carbonate because of a lack of aluminium for ettringite formation. This increased amount of sulfate could possibly be a result of different factors.

• Too much sulfate was added because the aluminate content of the OPC was not known and a standard recipe developed for another cement was followed.

• The amount of C₃A/C₄AF that reacted was unpredictable low (lower temperature).

• Exposure to sulfate-containing solutions that penetrated the cement stone.

For all anhydrite additions, Figure 9 shows the results for increases in the amount of added calcium carbonate, which could be the case for CEM II or even limestone calcined clay cements where up to 15 wt% limestone are added. Calcite does not normally dissolve completely, therefore only up to 5 wt% of calcite availability was modelled. The more sulfate and carbonate that is available, the more likely thaumasite could form.

Both effects combined played a key role in the possible formation of thaumasite. On the one hand, the comparably lower DoH of C₃A that disrupted the sulfate balance at 10°C and, on the other, the increased possibility that thaumasite forms preferentially at low temperatures (Pipilikaki et al., 2008; Schmidt et al., 2008).

Based on the data recorded in this study a few final conclusions can be made.

• Increasing the reaction rate of aluminates from OPC with TEA–TA–C$ was found to work in over a wide range of temperature. For lower temperatures, the retarder dosage needs to be decreased because low temperatures result in slowing of the cement reaction. Likewise, for higher temperatures the retarder dosage has to be increased.

• Both aluminate and silicate reactions started later in time. below room temperature and earlier for higher temperatures

• The DoH of C₃A was lowered for the 10°C samples. This had an important impact on the sulfate–C₃A balance of the system.

• The amount of sulfate added to the system always needs to be adjusted depending on the amount of available aluminium, otherwise delayed ettringite or (in the presence of carbonate) thaumasite can be formed. Therefore, the cement and the behaviour of the cement paste under different conditions needs to be well known. Applying a standardised recipe to cements from different manufacturers or even different types of cement is not recommended under any circumstances considering the results of the present study. This is shown in more detail in Figure 10. In case 1, if the cement used in this study is taken with the additive mix developed based on its phase composition and hydrated at 23°C, the maximum amount of ettringite is formed. Nevertheless, if this mix is hydrated at 10°C, the DoH of the aluminate phases is not as expected and thaumasite may form. For case 2, a different cement composition is assumed, hydrated with the additive mix of the cement from this study. For example, the cement for case 2 has less aluminate phases and sulfate carrier is left over, which could lead to the formation of secondary hydration products.

I. Kirchberger: data curation, formal analysis, investigation, methodology, validation, visualisation, writing – original draft, writing – review & editing. F. Götz-Neunhoeffer: resources, supervision, review & editing. J. Neubauer: funding acquisition, resources, supervision, review & editing