Three types of commercially available calcium sulfate dihydrate (DH) with different particle sizes fabricated by three companies were used in this study as raw material for DH-HH phase transition. DH with large particle size from JRS Pharma (USA) was named as DH-L; DH with medium particle size from BK Giulini PCG (Germany) was named as DH-M; DH with small particle size from Charles B. Chrystal Co. Inc. (USA) was named as DH-S. The α-HH prepared by using DH-L as raw material was nominated as α-HH-L. Similarly, the other two types of α-HH prepared by DH-M and DH-S were α-HH-M and α-HH-S, respectively. The same nomenclature principle works for DH cement prepared by α-HH. For instance, Cement-L means cement prepared by the α-HH-L that is synthesized by using DH-L as raw material. The nomenclature is summarized in Table 1. β-HH from Sigma (Sigma-Aldrich, St Louis, MO, USA) was also evaluated by SEM and XRD as a reference sample.

The experiments were performed in a 5 L double-walled cylindrical glass reactor (Globe, Syrris, UK) equipped with a glass condenser for vapor reflux, a thermometer for monitoring reaction temperature and a Teflon stirrer for homogenizing the slurry. The oil circulating through the double walls was regulated by a circulator (CF41 Cryo- Compact Circulator, Julabo, Germany) and the temperature was controlled within ± 0.5˚C. In each run, 500 ml 4 mol/L calcium chloride solution was added into the reactor and preheated to expected temperature. Then, 50 g calcium sulfate dihydrate (DH) was mixed in the reactor and the slurry was kept stirring at a constant stirring rate of 270 r/min. The powder obtained by vacuum filtration was quickly rinsed with 1 L boiling water to remove residual calcium chloride, and then with 200 ml isopropanol (Sigma- Aldrich, St Louis, MO, USA) to remove water in a suction flask, followed by drying at 60˚C for two hours.

X-ray diffraction (Diffractometer D5000, Siemens, Germany) was performed to invest-

tigate the crystalline phases by using Cu Kα X-ray at a diffraction angle from 10˚ to 80˚ with a step size of 0.02˚/step and an interval of 0.2 s/step. The particle size distribution was examined on a laser diffraction particle size analyser (Mastersizer X, Malvern Instrument Ltd., UK).

SEM (TM-1000, Hitachi, Japan) was used to observe the morphology of the DH at a 500 × magnification. Higher magnification morphology observation of DH, synthesized HH and DH cements were carried out on another SEM (LEO 1550, Zeiss, Germany).

Thermal gravity analysis (Q500, TA instrument, USA) was carried out for calcium sulfate phase identification and crystal water determination. Based on the weight loss percentage recorded by Thermal gravity analysis (TGA), the content of DH, HH and AH in the synthesized powder can be determined. According to literature [5] , the concentration (wt%) of crystallization water of HH is 6.2%, and that of DH is 20.9%. If the weight loss is less than 6.2% in TGA result, it means the synthesized powder is composed of HH and AH, whereas, weight loss over 6.2% indicates that the synthesized powder consists of HH and DH. In the first case, the following formula was used: 6.2% × X + 0% × (1 − X) = Y, while in the latter case, the equation was 6.2% × X + 20.9% × (1 − X) = Y, where X is the percentage of HH, Y is the weight loss recorded by TGA test.

The processes of making cements utilizing synthesized powders were as follows: Mix 1g of calcium sulfate powder with deionized water (about 10 - 15 s). Transfer the mixture to a syringe equipped with a 2 mm diameter tip. Inject the paste into a one side closed silicone mould with a size of 6 × 13 mm, and then vibrate the sample for 15 s. Let the paste set at room temperature (21˚C - 22˚C) for 24 h. Remove the sample from the mould, grind it with 800^# SiC grinding paper without any water and measure the compressive strength of the cement 24 hours after the cement was prepared. Compressive strength of cement was tested using an AGS-H universal material testing machine (Shimadzu, Kyoto, Japan) by applying a uniaxial compressive load at a cross-head displacement rate of 1 mm/min.

Particle size distribution test demonstrated that the median particle size of DH-L was 946.7 µm which was much larger than the other two types of DH. This is in good agreement with particle size distribution curve of DH-L which showed a sharp peak at approximately 900 µm (Figure 1). The median particle size of DH-M and DH-S were 162.4 µm and 62.4 µm, respectively. There were two rough peaks of the particle size distribution of DH-M and DH-S (Figure 1). One was at 10 µm and the other 100 µm, indicating that these two types of DH were composed of small particles and relatively large ones. However, DH-L showed more centralized particle size distribution, concentrating upon 900 µm. As can be seen from SEM results (Figure 2(a) & Figure 2(b)), DH-L had very large and irregular particles, which was in accordance with the result of particle size distribution test. DH-M and DH-S shared similar morphology characteristics, composing of plate-like particles. Both small and relatively large particles can be observed (Figures 2(c)-(f)). High magnification SEM images showed that DH-M pre

sented the most smooth particle surfaces (Figure 2(d)), while DH-S had relatively rougher particle surface with very small particles adhered on the plate-like particles (Figure 2(f)). DH-L also presented very rough surface (Figure 2(b)). All of the raw powders showed identical XRD peak positions. Only peaks associated with the DH structure were observed. No peaks related to impurities were detected, see Figure 3. It should be noted that the peak intensities of the three types of DH were different, presumably due to preferential orientation of certain lattice planes. As mentioned before, DH-L composed of large and irregular particles, without preferential orientation, while DH-M and DH-S presented plate-like morphology which probably contains preferential orientation of certain lattice planes.

One of the major objectives of this study is to optimize the reaction time and temperature to obtain high percentage α-HH. Only 13.7% DH transformed to HH within 75 min at 100˚C (Figure 4(a)). Reaction temperature exerts significant influence on the rate of the transition. As can be seen from Figure 4(b), 98.8% HH can be achieved after 90 min at 105˚C. For DH-M, 105˚C/30 min was the proper synthesis parameters which gave 96.7% HH (Figure 4(c)). In comparison, it took much shorter time for DH-S to transform to HH, achieving as high as 98.4% HH at 100˚C/45 min (Figure 4(d)). It was obvious that the reaction of DH-S and DH-M were much faster than that of DH-L. According to literatures [4] [16] , the dominant mechanism of transformation between gypsum (DH) and HH in aqueous solution is believed to be a through-solution reaction. Firstly, the dissolution of DH leads to a supersaturation with respect to HH, and then the HH particles precipitate from the solution after nucleation and growth of nuclei. Therefore, the dissolution rate of DH, nucleation and growth rate of HH affects the time needed for the complete transition of DH. Both particle size distribution (Figure 1) and SEM (Figure 2(a) & Figure 2(b)) results confirmed that particle size of DH-L

was much larger than the other two types of DH. Therefore, it took longer time to dissolve DH-L in the solution to form supersaturation. It was noteworthy that the synthesized powder of DH-S consisted of 98.4% HH and 1.6% DH at 100˚C/45 min (Figure 4(d)), indicating that the transformation of DH to HH has almost finished within 45 min at 100˚C which is 5˚C lower than the reaction temperature of the other two types of DH. The authors would still ascribe the high transition rate of DH-S to particle size effect. The median particle size of DH-S was 62.4 µm. Hence, DH-S can dissolve into the solution to form supersaturation within shorter time. What’s more, it is believed that nucleation can not only occur in the solution but also on the surface of the mother gypsum [5] [16] . As can be seen from Figure 2(f), there were a large number of small particles on the surface of DH-S particles. These small particles were washed away from the surface of the seeds and they were likely to be the nucleation sites for both heterogeneous and secondary nucleation, leading to higher transition rate.

It has been widely reported that the typical morphology of α-HH prepared under different conditions was a hexagon prism [3] [4] [6] . Therefore, the rod-like and prismatic particles clearly observed in Figures 5(a)-(c) were all α-HH. In contrast, the morphology of β-HH was quite different, consisting of irregular and flaky particles (Figure 5(d)). As can be seen from Figure 6, XRD patterns of the three types of α-HH were quite similar. β-HH shared great similarities in XRD patterns with α-HH, except that β-HH presented an overlap peak at 25˚, while α-HH had a single peak at the same degree. SEM observation and XRD results confirm that the synthesized HH were α-HH.

α-HH is a metastable phase, which can only maintain its form in a certain period in aqueous solution before further transit to a relatively stable phase. It is well accepted that at a given temperature solid phase that has the lowest solubility in aqueous solution is the stable phase [19] [20] [21] . Daniela et al. pointed that AH was the stable phase at temperatures higher than 70˚C in CaSO₄-based salts because of its low solubility [16] .

Therefore, it is understandable that AH can be detected in the synthesis experiments (Figures 4(b)-(d)). α-HH started to transform to AH after the completion of DH-HH transformation. Moreover, with the increase of reaction time more and more α-HH dehydrated to become AH. It was obvious that the dehydration rate of α-HH to AH was much slower than that of DH to α-HH for the synthesis experiments, see Figures 4(b)-(d). This phenomenon could be attributed to the differences on induction period of the two dehydration processes. Generally, the process of crystallization starts with a period of delay during which the solution concentration remains constant and crystal nuclei forms [21] [22] . This is so called induction period. It is influenced by several factors, such as supersaturation, temperature, component and impurities [18] [21] . Though no literature about comprehensive investigation on the induction period of crystallization processes in CaSO₄-based salts is available, based on the present results the authors tend to think that the induction period of AH crystallization is longer than that of α-HH crystallization. It has been reported that the crystallization of AH is the most difficult of all calcium sulfate phase due to its slow crystallization kinetics [16] .

Interestingly, particle size distribution test showed that the α-HH-S demonstrated the largest median particle size value, 44.1 µm. In contrast, the size of other two types of α-HH was much smaller, 10.7 µm and 3.8 µm for α-HH-M and α-HH-L, respectively. Figure 7 revealed that the particle size distribution curve of α-HH-S had two peaks. One peak was at 20 µm and the other 70 µm, which indicated that the α-HH consisted of small particles and relatively larger particles. The curves of the other two types of α-HH were similar, except that α-HH-L showed higher volume percentage at size of approximately 10 µm, whereas α-HH-M had higher volume percentage at around 50 µm. α-HH-M (Figure 5(b)) and α-HH-S (Figure 5(c)) composed of intact and homogenous hexagon-prism-sharped α-HH crystals, while α-HH-L (Figure 5(a)) had less intact particles. The nucleation and crystal growth processes are two competitive proc- esses in the beginning of crystallization process in terms of crystal size of the final crys- tallization products. Limited nuclei, low nucleation rate and high crystal growth rate results in larger crystal size, while numerous nuclei, high nucleation rate and low crystal growth rate results in smaller crystal size [6] [21] [22] . As mentioned before, the surface of the starting gypsum is able to serve as nucleation site, especially when the

surface is rough. As can be seen from Figure 2(b), the DH-L contained showed very rough surface which can act as sites of attachment. At the molecular level, Ca, O and S ions diffuse to the crystal surface and attach themselves to the attachment sites [21] . Numerous nucleation and growth sites in DH-L surfaces lead to the hindrance of the grain growth by impingement with neighbour grains, resulting in very small final α-HH-L particles. In contrast, the surfaces of DH-M were much smoother than those of DH-L, indicating that there would be less nucleation site and nuclei compared with DH-L. Once nuclei form, Ca, O and S ions in the solution tend to crystalize on the nuclei, which result in relatively larger crystal size of the final α-HH-M compared with α-HH-L. Regarding the particle size of α-HH-S, note that the transformation temperature of α-HH-S was 100˚C which is 5˚C lower than those of α-HH-L and α-HH-M. As mentioned above, transformation temperature exerts significant influence on the synthesis results, including the percentage, morphology and particle size of synthesized α-HH. Lowering transformation temperature would definitely reduce both nucleation and growth rates in the hydrothermal method by reducing the atom kinetics. Though no study has proven that temperature has more effects on nucleation rate than growth rate, the authors tend to believe that lowing temperature would exert more suppressing effects on nucleation rate than on growth rate. Hence, α-HH-S particles get more chance to grow, resulting larger particle size compare to α-HH-L and α-HH-M.

The liquid/powder ratio and strength of DH cements prepared by synthesized α-HH was shown in Table 2. Cement-L obtained the highest compressive strength (17.2 MPa) with a liquid/powder ratio of 0.6. In contrast, Cement-M achieved relatively lower compressive strength, 7.7 MPa, with a higher liquid/powder ratio of 0.8. These cement compressive strength values were quite close to those reported in previous literatures [23] [24] . α-HH-S paste can be achieved with a relatively lower liquid/powder ratio of 0.5, however, the paste was almost non-injectable. Only peaks of DH can be observed in XRD patterns (Figure 8), indicating that all α-HH transformed to DH during cement setting. SEM observations (Figure 9) showed that the three types of cements were compact materials without obvious pores. They were composed of micrometre-sized plate-like crystals. No significant difference on the morphology was observed, while Cement-L showed relatively larger crystal size.