Effect of Cooling Rate on the Structure of CaO–SiO₂–CaF₂-based Glassy Mold Flux

Abstract

To evaluate the effect of cooling rate on the structure of CaO–SiO₂–CaF₂ based mold flux, the structure of glassy mold fluxes prepared at different cooling rates ranging from 20°C/s to 100°C/s were investigated by Raman spectroscopy, ²⁹Si MAS-NMR and XPS techniques. The results shown that the structural species Q² and Q³ slightly increased with the increase of cooling rate, while the ratio of Q⁰ slightly decreased. The change of Q¹ was negligible. The average number of bridging oxygen atoms increased with the increase of cooling rate. The polymerization degree of the silicate structural network of glassy CaO–SiO₂–CaF₂-based mold fluxes was found to increase with the increase of cooling rate. The change of the cooling rate obviously caused the structural change of the CaO–SiO₂–CaF₂ mold flux. The faster the cooling rate, the higher temperature structural feature was founded. The effect of the cooling rate on F-bonds was negligible, and F was found to be mainly in the form of F–Ca bonds.

1. Introduction

Mold flux has a significant impact on slab quality during the process of steelmaking due to its essential functions during casting, including providing thermal insulation, preventing the oxidation of liquid steel, absorbing and dissolving non-metallic inclusions, providing lubrication, and controlling heat transfer between the shell and the mold.¹⁾ During continuous casting, molten mold fluxes infiltrate the gap between the copper mold and the solidified shell. The molten fluxes then solidify to form solid layers and a liquid layer due to the cooling of the copper mold and heating of the molten steel, as shown in Fig. 1. The solid layers include glassy and crystalline mold fluxes, as deduced by the different cooling conditions in the two sides of the gap. This layered structure enables the mold flux to perform the two important functions of controlling lubrication and heat transfer. The distribution of the layers has significant effects on the lubrication control and heat transfer control of mold flux.

Fig. 1.

The interfacial slag layers in continuous casting process. (Online version in color.)

According to a study by Thomas et al.,²⁾ the temperature of liquid slag on the surface of the shell at the meniscus was found to be higher than 1400°C, while the temperature next to the mold wall was found to be generally no more than 300°C; this resulted in a huge difference in the cooling rates of liquid slag at different positions. The cooling rate of liquid slag near the water-cooled mold was found to be greater than 100°C/s, while that of liquid slag in contact with the shell was found to be less than 1°C/s. It can therefore be concluded that there is a wide range of cooling rates of mold fluxes. Different cooling rates will have great influences on the properties of mold fluxes. Many reports^3,4,5) have demonstrated that varying cooling rates lead to significant changes of the physiochemical properties of mold fluxes, which in turn have a great impact on the control of lubrication and heat transfer. Wen et al.³⁾ found that the crystallization temperature decreased as the cooling rate increased via the single hot thermocouple technique (SHTT). Zhang et al.⁴⁾ found that the precipitation of corresponding crystals of CaO–BaO–Al₂O₃-based mold fluxes was reduced with the increase of the cooling rate from 1 to 4°C/s. The present authors investigated the effect of the cooling rate on the crystallization behaviors of CaO–SiO₂–CaF₂-based mold fluxes, and found that different cooling rates have important effects on the change of phases and crystal morphologies.⁶⁾

The physiochemical properties of mold fluxes are closely related to their structural characteristics. According to previous studies,^7,8,9,10,11) the structures of mold fluxes are a good indicator of crystallization and viscosity properties. Zhang et al.¹⁰⁾ investigated the effects of basicity and the addition of B₂O₃ on the structure of a CaO–SiO₂-based mold flux, and the change of the viscosity was explained from the structural perspective. Yeo et al.¹²⁾ further studied the structural evolution of borates when CaF₂ replaced with B₂O₃ and Na₂O, and reported that the determinant of viscosity is the tetrahedral borate structures. Gao et al.¹³⁾ found that the addition of BaO tended to increase the viscosity of a CaO–Al₂O₃-based mold flux due to the polymerization of the network structure of melts. Seo et al.¹⁴⁾ analyzed the substitutional of Na₂O with K₂O on the viscosity and structure of CaO–SiO₂–CaF₂‐based fluxes and found that the NBO/Si decreased with higher K₂O/(Na₂O+K₂O), resulted greater polymerization of this structure. The present authors found a good relationship between the polymerization degree and the effective activation energy E_G in a traditional CaO–SiO₂–CaF₂-based mold flux.⁸⁾ Additionally, glass transition was found to be deeply related to the change of the melt structure; cooling at a given rate was found to become too slow to maintain equilibrium.^15,16) Further understanding of the cooling rate-dependent structural changes of glass or melts is conducive to better material processing.

The structures of mold fluxes have been extensively investigated. However, only few studies have been conducted on the effect of the cooling rate on the structure of silicates. According to previous studies,^17,18,19) Si–O⁻ tetrahedron structural units have been categorized into five types: Qⁱ (i = 0, 1, 2, 3, 4, where i represents the number of bridging oxygen atoms in each SiO₄ structural unit). Tan et al.²⁰⁾ investigated the effect of the cooling rate on the structure of sodium silicate glass using Raman spectroscopy, and found that the distribution of silicate species varied with the cooling rate, and that the structural species Q⁴ and Q² increased, while Q³ decreased, with the increase of cooling rate. Karakassides et al.²¹⁾ found that the concentration of silicon–oxygen units with two non-bridging oxygen atoms per tetrahedron is increased in potassium silicate glasses with faster quenching rates. To best of the authors’ knowledge, there has been no systematic study of the dependence of the mold flux structure on the cooling rate. Thus, the present work investigates the effect of the cooling rate on the traditional CaO–SiO₂–CaF₂-based glassy mold flux. Because a wide range of cooling rates can be achieved, confocal scanning laser microscopy (CSLM) equipment was employed to prepare CaO–SiO₂–CaF₂-based glassy mold fluxes quenched at cooling rates from 20 to 100°C/s. The structures of the CaO–SiO₂–CaF₂-based glassy mold fluxes were investigated via the combination of ²⁹Si MAS-NMR, Raman spectroscopy, and X-ray photoelectron spectroscopy (XPS).

2. Experimental

2.1. Glass Preparation

The chemical compositions of the CaO–SiO₂–CaF₂ mold fluxes are listed in Table 1. The reagent-grade powders of CaCO₃, Al₂O₃, SiO₂, Na₂CO₃, and CaF₂ were used as raw materials. CaCO₃ was calcined at 1100°C for 8 h in a muffle furnace (Nuobadi Corporation, Zhengzhou, China) to prepare CaO. All the raw materials were weighed precisely and well-mixed in an agate mortar for 0.5 h, after which the mixtures were placed in a platinum crucible and heated to 1350°C for 2 h in air in a high-temperature tube furnace. The melts were then quenched with water to obtain the basic glassy samples.

Table 1. Chemical composition of CaO–Al₂O₃–CaF₂ based mold flux.

Composition (Mass%)	CaO	Al2O3	SiO2	Na2O	CaF2
Nominal	28.13	5	46.87	8	12
Analyzed (20°C/s)	29.94	6.34	43.83	9.43	10.46
Analyzed (40°C/s)	28.75	6.99	43.56	9.53	11.17
Analyzed (60°C/s)	28.51	6.76	43.93	9.75	11.05
Analyzed (80°C/s)	28.53	6.89	43.86	9.81	11.23
Analyzed (100°C/s)	28.64	6.63	44.22	9.31	11.20

The basic glassy samples were respectively placed in a platinum crucible with a diameter of 7 mm and melted in a high-temperature furnace with CSLM equipment (VL2000DX, Lasertec Corporation, Yokohama, Japan). The platinum crucible containing the samples was rapidly heated by the light irradiated by a halogen lamp. Afterward, the samples reached thermal equilibrium at 1350°C for 5 min, and the molten fluxes were then cooled at a designated rate ranging from 20 to 100°C/s via helium cooling technology. The chemical composition of all quenched glassy samples prepared by different cooling rates was further identified by the scanning electron microscope equipped with energy dispersive X-ray spectroscopy (SEM-EDS, Zeiss, Sigma 300, Germany) and showed in Table 1. It can be seen that the cooling rates have no significant effects on the chemical composition. Only the concentration of fluorine showed a slight increase trend with the increase of cooling rate. This may be caused by the high temperature volatilization of F.

2.2. Nuclear Magnetic Resonance Analysis

²⁹Si MAS-NMR spectroscopy measurements were conducted using a 600M FT-NMR spectrometer (Bruker AVANCE III 600 MHz (14.09 T), Switzerland) with a MAS probe with a 4 mm ZrO₂ rotor and two pairs of Dupont Vespel caps, with resonance frequencies of 119.3 MHz. The spinning rate was 10 kHz for ²⁹Si MAS-NMR spectroscopy. Tetramethylsilane (TMS) was used as the reference material. The relaxation delay is 10 seconds. An exponential linebroadening factor of 120 Hz was used during spectral processing to smooth the spectra. The NMR measurements were taken twice for each sample. The peaks of the two measurements coincide perfectly due to the uniform glass sample. The MAS-NMR spectra were deconvoluted with Gaussian functions using Peak Fit software to obtain more specific structure information.

2.3. Raman Spectroscopy Analysis

A micro-Raman spectrometer coupled with a CCD detector (LabRAMHR Evolution, HORIBA Jobin Yvon, France) was used. Raman spectra were obtained at room temperature in the frequency range of 200–1600 cm⁻¹ using a 532 nm laser. The Raman measurements were performed at room temperature using an excitation wavelength of 532 nm. The light source was a semiconductor laser with power of 1 mW. The Raman measurements were taken twice for each sample. The peaks of the two measurements coincide perfectly due to the uniform glass sample. The Raman spectra were deconvoluted by assuming Gaussian line shapes for peaks of different structural units.

2.4. X-ray Photoelectron Spectroscopy

X-ray photoelectron spectroscopy (XPS) experiments were carried out by an X-ray photoelectron spectrometer (PHI-5300/ESCA, Perkin Elmer, USA) with an Al Kα X-ray source (1253.6 eV) to obtain the O1s and F1s XPS spectra. To eliminate surface contamination before XPS measurement, the surface of the glass sample was etched by an argon ion beam for 20 min. The C1s hydrocarbon peak was set at 284.8 eV to attain the correct binding energy.

3. Results and Discussion

3.1. Raman Spectra of CaO–SiO₂–CaF₂-based Glassy Mold Fluxes Cooled at Different Rates

The original Raman spectra of the CaO–SiO₂–CaF₂-based glassy mold fluxes cooled at different rates are presented in Fig. 2. According to a study conducted by Vidal et al.,²²⁾ the Raman frequency range between 340–360 cm⁻¹ corresponds to the network modifiers. In terms of the chemical composition of the CaO–Al₂O₃–SiO₂–Na₂O–CaF₂ system, the network modifiers may have been Na–O or Ca–O linkages. The middle-frequency region (550–750 cm⁻¹) is associated with the Si–O–Si bending vibrations within inter-tetrahedral linkages, while the high-frequency region (800–1200 cm⁻¹) is attributed to the Si–O⁻ stretching vibrations.^22,23) It was observed that the Raman shift slightly moved toward the high-frequency direction with the increase of cooling rate, demonstrating that the network of the investigated silicates gradually became more polymerized.

Fig. 2.

Original Raman spectra of the mold fluxes with different cooling rates. (Online version in color.)

In the present work, quantitative information was obtained via analysis of the results of the curve-fitting of the high-frequency envelope (800–1200 cm⁻¹) of the Raman spectra. The high-frequency range of all original Raman spectra were deconvoluted by the Gaussian fitting method (the minimum correlation coefficient r² ≥ 0.995),^18,24,25) as presented in Figs. 3(a)–3(e). The half-width and intensity were unconstrained variables when the curve-fitting was made. Based on previous studies,^26,27,28) the frequency center near 885 cm⁻¹ corresponded to Q⁰, and that at 950 cm⁻¹ corresponded to Q¹. The frequency centers near 1015 cm⁻¹ and 1065 cm⁻¹ were respectively assigned to Q² and Q³.

Fig. 3.

Deconvoluted results of Raman for the mold fluxes with different cooling rates. (Online version in color.)

The area percentages of each structural unit of the CaO–Al₂O₃–SiO₂–Na₂O–CaF₂ mold flux are listed in Table 2. The ratio of the chain (Q²) and sheet (Q³) slightly increased, while the ratio of the monomer (Q⁰) and dimer (Q¹) slightly decreased with the increase of the cooling rate. The average number of bridging oxygen atoms was employed to explain the change of the silicate structural network, which was equal to the number of bridging oxygen atoms of each structural unit multiplied by their area proportion.^7,8,28) Figure 4 presents the average number of bridging oxygen atoms of the CaO–Al₂O₃–SiO₂–Na₂O–CaF₂ mold fluxes at different cooling rates. The average number of bridging oxygen atoms slightly increased with the increase of the cooling rate, as indicated by the increase of the polymerization degree of the silicate structural network with the increase of the cooling rate of the glasses.

Table 2. Deconvolved results of Raman spectra and ²⁹Si MASNMR of CaO–Al₂O₃–CaF₂ based glassy mold flux at different cooling rates.

Cooling rates (°C/s)		Percentage of each unit (%)
		Q0	Q1	Q2	Q3
20	Raman	15.28	31.03	31.28	22.12
	NMR	9.06	39.54	37.51	13.89
40	Raman	13.56	31.29	32.00	23.14
	NMR	8.92	39.34	37.59	14.16
60	Raman	13.40	31.14	32.10	23.49
	NMR	4.71	39.84	39.43	16.02
80	Raman	12.87	31.48	32.21	23.65
	NMR	4.23	39.42	39.78	16.57
100	Raman	12.27	31.65	32.33	23.99
	NMR	3.05	39.76	40.12	17.07

Fig. 4.

The average number of bridging oxygen of the mold fluxes with different cooling rates. (Online version in color.)

3.2. ²⁹Si MAS-NMR Spectra of CaO–SiO₂–CaF₂-based Glassy Mold Fluxes Cooled at Different Rates

The ²⁹Si MAS-NMR spectra of the glassy CaO–Al₂O₃–SiO₂–Na₂O–CaF₂ mold fluxes at different cooling rates are plotted in Fig. 5. Only one peak was observed in all the spectra. Apparently, the chemical shift of the main peak moved toward the more negative direction. The center chemical shift of the main peak ranged from −80.7 to −82.1 ppm with the increase of the cooling rate of the glasses, revealing that the polymerization degree of the investigated silicates had increased.

Fig. 5.

²⁹Si NMR spectra of the mold fluxes with different cooling rates. (Online version in color.)

The quantitative information was obtained via the analysis of the result of the curve-fitting of the ²⁹Si MAS-NMR spectra using a similar method as that for Raman deconvolution (the minimum correlation coefficient r² ≥ 0.998). The fitting curves were shown in Figs. 6(a)–6(e). According to previous reports,^18,29) the center peak near 70 ppm was assigned to Q⁰ and the center peak near 78 ppm was assigned to Q¹. The center peaks near 84 ppm and 90 ppm corresponded to Q² and Q³, respectively. Q⁴ was negligible because no characteristic peak was identified in the range of 98–129 ppm in any of the ²⁹Si MAS-NMR spectra. The specific percentage of each unit is listed in Table 2. Q⁰ decreased as the cooling rate increased, whereas Q² and Q³ increased. The change of Q¹ was negligible. As presented in Fig. 4, the average number of bridging oxygen atoms increased with the increase of cooling rate, which is consistent with the Raman results.

Fig. 6.

Deconvolved results of ²⁹Si NMR for the studied mold fluxes with different cooling rates. (Online version in color.)

3.3. XPS Spectra of the Mold Fluxes with Different Cooling Rates

Figures 7 and 8(a) present the original F1s and O1s XPS spectra of the CaO–Al₂O₃–SiO₂–Na₂O–CaF₂ mold fluxes at different cooling rates. As shown in Fig. 7, the binding energy of the central peak was about 683.8 eV, and was kept constant with the variation of the cooling rate; this was recognized as Ca–F bonds, but may include a small amount of Na–F bonds.^30,31) This indicates that the changes of the F 1s XPS spectra with the increase of cooling rate were negligible, and that the increase of the cooling rate had no significant effect on F-bonds. Hill et al.,³²⁾ Li et al.,¹⁸⁾ and Stamboulis et al.³³⁾ reported similar results, namely that F⁻ primarily combines with Ca²⁺ to form Ca–F bonds in CaO–SiO₂-based mold flux systems.

Fig. 7.

F1s XPS spectra of the mold fluxes with different cooling rates. (Online version in color.)

Fig. 8.

O1s XPS spectra of the mold fluxes with different cooling rates. (a) the original spectral, (b) the fitted spectral of the cooling rate at 20°C/s. (Online version in color.)

The Gaussian deconvolution method was employed to deconvolute the O1s XPS spectra (the minimum correlation coefficient r² ≥ 0.998), and an example of the 20°C/s cooling rate is presented in Fig. 8(b). The mother O 1s XPS spectra was separated by two peaks, respectively representing non-bridging oxygen atom (NBO) and bridging oxygen atom (BO) linkages. The NBO linkages were located at approximately 530.4 eV, and the BO linkages were located at 531.4 eV.^31,34,35) The fractions of the relative peak areas of the NBO and BO linkages are plotted in Fig. 9. It was observed that the proportion of NBO decreased while that of BO increased with the increase of cooling rate. This demonstrates that the polymerization degree of the silicate structural network increased with the increase of cooling rate, which is consistent with the results of the Raman and ²⁹Si MAS-NMR analyses.

Fig. 9.

Fraction of BO and NBO for the mold fluxes with different cooling rates. (Online version in color.)

Therefore, it can be concluded that the polymerization degree of the CaO–SiO₂–CaF₂ system increased with the increase of the cooling rate. The effect of the cooling rate on F-bonds was negligible, and F was mainly in the form of F–Ca bonds. There is a corresponding equilibrium between the coexisting structural species in silicate melt at high temperature. When molten slags cool on a speed larger than the limiting value of relaxation speed, the equilibrium configuration is adopted. Oppositely, the melt structure is non-equilibrium, which will preserve part of high-temperature features. When molten slags are rapidly cooled, melts depart from the equilibrium at a higher temperature and consequently a higher temperature feature is frozen into the glass structure. The present structural analysis results clearly show that the faster the cooling rate, the higher temperature structural feature was founded.

The relationship between the structure and properties of mold fluxes was discussed in the authors’ previous publications,^8,36) in which it was determined that the effective activation energy increased and the crystallization ability decreased with the increase of the average number of bridging oxygen atoms in CaO–SiO₂–CaF₂-based mold fluxes. The increased polymerization degree of CaO–Al₂O₃-based mold fluxes was found to cause the increase of viscosity.³⁶⁾ For the present system, when the cooling rate increase from 20°C/s to 100°C/s, according to the result of the fraction of BO and NBO, we can calculated the NBO/T is from 1.6192 decrease 1.4312. Therefore, refer to the linear relation formula (ln viscosity = 2.94963−0.96248 NBO/T) between viscosities and NBO/T reported by Wu et al.,³⁷⁾ the increase in viscosity can be roughly calculated as 0.79649 Pa·s as the cooling rate increase from 20°C/s to 100°C/s. Additionally, the effects of the cooling rate on the crystallization of CaO-SiO₂-CaF₂-based mold fluxes was investigated in the authors’ previous work,⁶⁾ in which it was demonstrated that the crystallization temperature decreased and the morphology of the primary crystal changed with the increase of the cooling rate. Therefore, the change of the cooling rate during continuous casting should be further considered for the future design of more optimized mold fluxes.

4. Conclusion

The effect of cooling rate on the structure of CaO–Al₂O₃–SiO₂–Na₂O–CaF₂ mold fluxes was investigated. The structural species Q² and Q³ slightly increased with the increase of the cooling rate, while the ratio of Q⁰ slightly decreased. The change of Q¹ was negligible. As indicated by the results of Raman, ²⁹Si MAS-NMR, and XPS analyses, the average number of bridging oxygen atoms and the percentage of BO increased with the increase of cooling rate. The polymerization degree of the silicate structural network of glassy CaO–SiO₂–CaF₂-based mold fluxes was found to increase with the increase of cooling rate. The change of the cooling rate obviously caused the structural change of the CaO–SiO₂–CaF₂ mold flux. The faster the cooling rate, the higher temperature structural feature was founded. Finally, the effect of the cooling rate on F-bonds was negligible, and F was found to be mainly in the form of F–Ca bonds.

Acknowledgments

This work was supported by the Natural Science Foundation of China (51704050); China Postdoctoral Science Foundation (2018T110944, 2017M612905); and Fund of Sichuan Key Research & Development Projects (2018SZ0281).

References

• 1)   K. C.  Mills: ISIJ Int., 56 (2016), 1.
• 2)   Y.  Meng and  B. G.  Thomas: ISIJ Int., 46 (2006), 660.
• 3)   G.  Wen,  H.  Liu and  P.  Tang: J. Iron Steel Res. Int., 15 (2008), 32.
• 4)   S.  Zhang,  M.  Li,  L.  Zhu,  Q.  Wang,  S.  He and  Q.  Wang: Ceram. Int., 45 (2019), 11296.
• 5)   Y.  Kashiwaya,  C. E.  Cicutti,  A. W.  Cramb and  K.  Ishii: ISIJ Int., 38 (1998), 348.
• 6)   M.  Leng,  F.  Lai and  J.  Li: Materials, 12 (2019), 62.
• 7)   Y.  Gao,  M.  Leng,  Y.  Chen,  Z.  Chen and  J.  Li: Materials, 12 (2019), 206.
• 8)   J.  Li,  B.  Yan,  Q.  Shu and  K.  Chou: Metall. Mater. Trans. B, 46 (2015), 2458.
• 9)   H.  Park,  J. Y.  Park,  G. H.  Kim and  I.  Sohn: Steel Res. Int., 83 (2012), 150.
• 10)   L.  Zhang,  W.  Wang,  S.  Xie,  K.  Zhang and  I.  Sohn: J. Non-Cryst. Solids, 460 (2017), 113.
• 11)   W. J.  Huang,  Y. H.  Zhao,  S.  Yu,  L. X.  Zhang,  Z. C.  Ye,  N.  Wang and  M.  Chen: ISIJ Int., 56 (2016), 594.
• 12)   T. M.  Yeo,  J. W.  Cho,  M.  Alloni,  S.  Casagrande and  R.  Carli: J. Non-Cryst. Solids, 529 (2020), 119756.
• 13)   E.  Gao,  W.  Wang and  L.  Zhang: J. Non-Cryst. Solids, 473 (2017), 79.
• 14)   M. S.  Seo and  I.  Sohn: J. Am. Ceram. Soc., 102 (2019), 6275.
• 15)   D. B.  Dingwell and  S. L.  Webb: Eur. J. Mineral., 2 (1990), 427.
• 16)   M.  Wilding,  S.  Webb and  D.  Dingwell: Chem. Geol., 125 (1995), 137.
• 17)   B. O.  Mysen,  F. J.  Ryerson and  D.  Virgo: Am. Mineral., 65 (1980), 1150.
• 18)   J.  Li,  Q.  Shu and  K.  Chou: ISIJ Int., 54 (2014), 721.
• 19)   B. O.  Mysen,  D.  Virgo and  C. M.  Scarfe: Am. Mineral., 65 (1980), 690.
• 20)   J.  Tan,  S.  Zhao,  W.  Wang,  G.  Davies and  X.  Mo: Mater. Sci. Eng. B, 106 (2004), 295.
• 21)   M.  Karakassides,  G.  Kordas,  E.  Mylonas and  C.  Trapalis: Mater. Sci. Eng. B, 26 (1994), 35.
• 22)   L.  Vidal,  E.  Joussein,  M.  Colas,  J.  Cornette,  J.  Sanz,  I.  Sobrados,  J.  Gelet,  J.  Absi and  S.  Rossignol: Colloids Surf. A, 503 (2016), 101.
• 23)   J.  Gao,  G.  Wen,  T.  Huang,  B.  Bai,  P.  Tang and  Q.  Liu: J. Non-Cryst. Solids, 452 (2016), 119.
• 24)   B. O.  Mysen,  L. W.  Finger,  D.  Virgo and  F. A.  Seifert: Am. Mineral., 67 (1982), 686.
• 25)   Y.  Chen,  F.  Lai and  J.  Li: JOM, 72 (2020), 1414.
• 26)   J.  Chen,  F.  He,  Y.  Xiao,  M.  Xie,  J.  Xie,  W.  Zhang and  J.  Shi: Constr. Build. Mater., 216 (2019), 19.
• 27)   G. H.  Zhang,  Y. L.  Zhen and  K. C.  Chou: J. Iron Steel Res. Int., 23 (2016), 633.
• 28)   F.  Lai,  W.  Yao and  J.  Li: ISIJ Int., 60 (2020), 1596.
• 29)   E.  Lippmaa,  M.  Maegi,  A.  Samoson,  G.  Engelhardt and  A.  Grimmer: J. Am. Chem. Soc., 102 (1980), 4889.
• 30)   K.  Nagata and  H.  Fukuyama: Steel Res. Int., 74 (2003), 31.
• 31)   S.  Hayakawa,  A.  Nakao,  C.  Ohtsutki,  A.  Osaka,  S.  Matsumoto and  Y.  Miura: J. Mater. Res., 13 (1998), 739.
• 32)   R. G.  Hill,  N.  Da Costa and  R. V.  Law: J. Non-Cryst. Solids, 351 (2005), 69.
• 33)   A.  Stamboulis,  R. G.  Hill and  R. V.  Law: J. Non-Cryst. Solids, 333 (2004), 101.
• 34)   H.  Nesbitt,  G.  Henderson,  G.  Bancroft and  C.  O’Shaughnessy: Chem. Geol., 461 (2017), 65.
• 35)   K. N.  Dalby,  H. W.  Nesbitt,  V. P.  Zakaznova-Herzog and  P. L.  King: Geochim. Cosmochim. Acta, 71 (2007), 4297.
• 36)   J.  Li,  K.  Chou and  Q.  Shu: ISIJ Int., 60 (2020), 51.
• 37)   T.  Wu,  S.  He,  Y.  Liang and  Q.  Wang: J. Non-Cryst. Solids, 411 (2015), 145.