Composition of the three starting steel slags was measured by ICP-AES and -MS analysis. Average results from three replicas are presented in Table 1. Accounting for nearly half of the sample mass, calcium oxide is the main constituent of the steel slag samples followed by iron oxide and silica. Average chemical compositions are in good agreement with literature values (Proctor et al., 2000; Malvoisin et al., 2013; Piatak et al., 2015). Only minor variations in chemical composition were observed between the three samples, this chemical homogeneity among samples reflects the Basic Oxygen Furnace process mastering. Considering that CO₂ sequestration occurs through calcium carbonate formation, an overall carbonation potential of 387 kg CO₂ can be deduced from the bulk CaO content. SMC was determined to be 63 kg CO₂/t steel slag on the freshest sample (weeks, Table 1) by TGA (Figure 2). This SMC figure corresponds to 5.9 wt.% CO₂ in the slag (Table 1).

Total CO₂, LOI, and iron oxidation are the chemical parameters that significantly vary with storage time (Table 1). The longer the steel slag was stored, the higher the total CO₂ content. Iron oxidation due to weathering is evidenced by a decrease of the Fe2+∑Fe ratio as storage time increases. This ratio drifted from 45% after weeks and months to 33% after years of outdoor storage.

Mineralogical content of the BOF slag sample aged for several weeks on the storage site (Figure 3) was determined combining bulk composition (Table 1), phase identification by XRPD, FE-SEM imaging, and EDS analysis (Table 2; Figure 4). Three calcium carbonates were identified by XRPD: calcite, aragonite, and vaterite (Table 2). Based on TGA measurement which yielded 5.9 wt.% CO₂, these calcium carbonates represent around 15 wt.% of the initial steel slag (Figure 2). Larnite, Ca₂SiO₄ (C₂S), and brownmillerite are the main mineral phases in the starting slag which account together for more than 50% of the slag mass. Brownmillerite and magnetite host the ferric iron contained in the starting steel slag. Ferrous iron is hosted in magnesian wüstite (called wüstite for simplification hereafter).

Carbonation was conducted at RT on the freshest slag immerged in water. CO₂ pressure drop and pH were first monitored over 24 h. The carbonation experiment started effectively with the CO₂ injection (t = 0 min) after pressure and pH stabilization, at 10 bar and a pH value of 10.6, respectively. Temperature rose from 25 to 30°C and the CO₂ pressure dropped exponentially before it stabilized at around 4 bar after ca. 85 min (Figure 5A). Similarly, pH decreased exponentially and reached 3.7 in less than 15 min and 3.3 after 40 min and for the rest of the experiment (Figure 5B). The initial steel slag which was already partly carbonated was found to store an equivalent of 1.5 bar CO₂.

In parallel to CO₂-pressure and pH monitoring, time-resolved phase changes in the aqueous suspension were monitored by in situ Raman spectroscopy (Figures 5C,D). Before CO₂ injection, portlandite and calcite were the two main components identified in the suspension. After CO₂ was injected, portlandite consumption started and lasted until its total disappearance after 20 min of reaction. Calcite signal started to grow after 10 min of reaction. It kept growing even after portlandite consumption until 720 min of reaction and the observation of signal instabilities. In addition to calcite growth, a minor peak belonging to aragonite was transiently detected at 210 cm⁻¹, which disappeared after 420 min of reaction.

Figure 5D shows spectra in the 1,200–600 cm⁻¹ wavenumber range for five different run durations. Calcite is easily identified with two characteristic bands, one at 1,095 cm⁻¹ and another at 715 cm⁻¹. As reaction proceeds, a narrowing of the Raman band in the 710–680 cm⁻¹ range is observed after more than 30 min of reaction. This signal is attributed to the respective contributions of portlandite and a calcium silicate, possibly larnite. At t₀, another small contribution is also detected at 960 cm⁻¹ which quickly disappeared after 20 min of carbonation. This contribution may be related to the partial hydration product of calcium silicate such as afwillite or hibschite. Signal instabilities were observed after 720 min of reaction. As the sapphire peak from the window remains constant, these instabilities are not related to the Raman light source. After opening the autoclave, most of the solid sample was found to have formed a hardened paste, stuck at the bottom of the autoclave, i.e., away from the volume sampled by the Raman probe.

X-ray powder diffraction patterns confirmed the full consumption of portlandite and calcium silicates (larnite) as well as the crystallization of calcite (Table 2). Starting aragonite and vaterite all converted into calcite. TGA data indicate a single weight loss event (ca. 23 wt.%) consistent with calcite as the sole Ca-carbonate in the system (Figure 2) and which is characterized by a DTA peak centered around 765°C. An OC of 306 kg CO₂/t is deduced which corresponds to an experimental carbonation (EC) of 243 kg CO₂/t steel slag. It is calculated that about 79% of the CaO mass in the slag is now combined to CO₂ in the form of carbonates. In the starting steel slag (Figure 3), CaCO₃, larnite, and portlandite hosted about 70% of the CaO mass contained in the sample. Therefore, the derived OC is broadly consistent with the carbonation of portlandite and larnite although partial carbonation of brownmillerite cannot be excluded. No major mineralogical changes were observed among iron-bearing compounds. Iron was mostly recovered as oxides in the form of brownmillerite, magnetite, and wüstite (Table 2).

Steel slag hydrothermal oxidation was performed at 300°C according to two experiments, i.e., with and without CO₂ (Figure 6). Hydrogen was produced in both cases following a similar kinetic model which is satisfactorily fitted to a simple squared root function of time: (2)H2(t)=Ant0.5.

In the absence of added CO₂, hydrogen production rate was found to be roughly constant during the 30 first hours before it progressively slowed down. A total of 2.13 g H₂/kg FeO were produced after 72 h and A₁ was estimated to 0.276 g H₂/h^1/2/kg FeO. In the presence of CO₂ (initial P_CO2 = 2 bar), a total of 1.55 g H₂/kg FeO was produced after 72 h; A₂ in Eq. 2 is estimated to 0.185 g H₂/h^1/2/kg FeO. CO₂ gas concentration was found roughly constant over the whole reaction. It corresponded to an average CO₂ partial pressure of 16 bar at 300°C as calculated assuming CO₂ as an ideal gas. No CH₄ was detected in the analyzed gas with a CH₄ detection limit estimated at 100 ppm.

As for RT carbonation, calcite is the dominant mineral phase in the sample carbonated at 300°C (Table 2). XRPD (Figure 7A) and SEM characterization (Figure 7B) revealed another carbonate with a composition (Fe_0.56Mg_0.36Mn_0.06Ca_0.02)CO₃ as inferred from EDS data. The diffraction signal of this carbonate could be refined using a siderite structure assuming however lattice parameters that significantly exceed those expected along the siderite-magnesite solid solution. XRPD pattern of HT-carbonated products also revealed the crystallization of saponite, a trioctahedral clay mineral which belongs to the smectite group. The retrieved lattice parameter, b, of around 9.22 Å, suggests that the newly formed saponite has octahedral sites dominated by iron (Brigatti, 1983), Fe²⁺ (Baldermann et al., 2014), and/or Fe³⁺ (e.g., Treiman et al., 2014). Iron mass balance from Rietveld data along with the constraint that Fe²⁺/Fe_tot in the slag must decrease in order to account for measured H₂ production, suggesting that Fe in saponite is dominantly ferric.

Thermal gravimetric analysis data indicate a two-stage mass loss (Figure 2) with two DTA peaks located at 530 and 789°C. From 120 to 1,000°C, the sample has lost about 30% of its initial mass, whereas Rietveld refinement predicts a (H₂O + CO₂) content in the HT-carbonated slag of 35.5 wt.%. The main mass loss event (24.2 wt.%) corresponds to the 789°C DTA peak and is very likely correlated with calcite decomposition since calcite is the dominant mineral according to XRPD. The lower temperature mass loss (ca. 4 wt.%) may be correlated with either Fe-rich saponite dehydroxylation (Brigatti, 1983; Földvári, 2011) or Mg-siderite decomposition since 530°C is comprised between magnesite and siderite TGA decomposition temperatures (Földvári, 2011). Based on the amount of Fe-saponite (11 wt.%) obtained from Rietveld refinement, the contribution of saponite dehydroxylation to the sample weight loss should not exceed half a weight percent and therefore the lower temperature TGA peak is more likely related to Mg-siderite decomposition. A 3.5–4% mass loss would correspond to the decomposition of 8–9 wt.% siderite which compares relatively well with the 10 wt.% inferred from Rietveld refinement. EC is estimated to 327 kg CO₂/t from TGA. Moreover, no more wüstite nor brownmillerite was found in the XRPD pattern. Iron is now partitioned between magnetite, hematite, saponite and siderite. Carbonates were observed in the form of rhombohedral micrometric particles (Figure 7C). Mineralogical analysis of the sample oxidized at 300°C without added CO₂ yielded, by order of abundance, hydrated calcium silicates (katoite–hibschite–jaffeite), calcite, hydroxylapatite, magnetite, wüstite, and portlandite (Crouzet et al., in press).

Further FE-SEM characterization was performed on iron oxides extracted magnetically with a permanent magnet in a water-filled ultrasonic bath. This magnetic separation allowed to emphasize a multitude of iron oxide nanoparticles (Figure 7D).

In order to investigate the potential of the CO₂ injection at high temperature in lowering the pH and enhancing oxidation rate (H₂ production), hydrothermal steel slag oxidation was performed in deionized water at 300°C to 150 bar under argon atmosphere with and without added CO₂ partial pressure. Hydrogen production was observed in both cases, following similar kinetic models. So apparently, the addition of CO₂ does not fundamentally modify the steel slag oxidation process. In absence of a plateau in H₂ production, a square root function of time was found to yield a better fit to the H₂ production data than a first-order kinetic model (Crouzet et al., 2017). These data indicate a H₂ yield by 30% lower when CO₂ gas is added. We also show that the addition of CO₂ gas leads to the disappearance of wüstite which is otherwise observed as residual phase.

There is a clear discrepancy between recovered H₂ and wüstite consumption in the HT carbonation experiments. This apparent H₂ deficit can be explained by the formation of an (Fe,Mg)-carbonate, (Fe_0.56Mg_0.36Mn_0.06Ca_0.02)CO₃, of siderite structure (Rietveld refinement) which actually incorporated ferrous iron and prevented it from further oxidation. The Fe/Mg ratio of this carbonate as well as the presence of manganese traces suggest that it directly formed from wüstite with little fractionation of the divalent cations (Fe, Mg, Mn). Another remarkable feature is the presence of hematite. Using the concentration of aqueous H₂ at the magnetite–hematite equilibrium and the Henry constant of H₂ as provided by SUPCRT92 (Johnson et al., 1992), it is calculated that, at 300°C and 190 bar, hematite should no longer be stable over 1% reaction progress (Appendix). Hematite might therefore be an intermediate compound of brownmillerite breakdown which persists as a metastable phase indicating a relatively low reduction kinetics into magnetite.

Total H₂ produced in the carbonation experiment was found to only represent 37 mol.% of maximum H₂ that can be produced based on the initial Fe²⁺ content of the starting slag. On the other hand, about 50 wt.% of the initial FeO content is sequestered in iron carbonates. Furthermore, the reduction of hematite into magnetite might consume H₂ and thereby decrease the overall H₂ yield.

Assuming that the pH in the HT carbonation experiments is controlled by the calcite–CO₂ equilibrium, an estimated pH of ca. 5.6 is calculated for the investigated experimental conditions. Crouzet et al. (2017) noted a significant kinetic improvement in FeO oxidation in the presence of acetic acid solution at a pH of ca. 4.7. Unfortunately, in the present HT carbonation experiments, slag wüstite is involved in two competing reactions, oxidation on one hand and carbonation on the other. It is therefore not possible to precisely assess the effect of shifting the pH below 7 on wüstite oxidation kinetics alone. However, in the 72-h carbonation/oxidation experiment, 1.55 g H₂/kg FeO has been produced. If now, we consider that about half of the FeO has been entrapped into carbonates (additional Fe²⁺ might also have been incorporated into saponite), it follows that more than 3 g H₂/kg of available FeO has been effectively produced in the time span, i.e., 40% more than in the hydrothermal experiment where no CO₂ was added. It remains that HT carbonation lowered the H₂ yield in comparison to wüstite oxidation in water without added CO₂. Moreover, injection of CO₂ gas for the sake of kinetic improvements led to the alteration of hydrogen purity. As far as H₂ production is concerned slag oxidation under P_CO2 has obviously no added value.

In parallel to H₂ production, hydrothermal oxidation/carbonation of steel slag led to the formation of magnetite. FE-SEM investigations were conducted on magnetically separated samples. Magnetite was observed as nanoparticle agglomerates. Crouzet et al. (2017) already reported the formation of magnetite nanoparticles as the result of the hydrothermal oxidation of FeO in the presence of acetic acid solution. Magnetite nanoparticles and its economical relevance are discussed in the companion article (Crouzet et al., in press). It would however be interesting to study whether HT carbonation could influence the properties of the produced nanomagnetite (size, morphology, and composition).

Other H₂ production routes using CO₂ still need to be explored. For example, recently, Michiels et al. (2015) showed that the in situ formation of aqueous CO32− has a catalytic effect on the mild hydrothermal oxidation of iron metal powder. The stability of aqueous CO32− required these experiments to be conducted at pH = 11 with potassium hydroxide. This catalytic effect was found to be dependent of the CO₂ partial pressure. For an initial CO₂ overpressure of 15 bar in the reactor at 25°C, the authors reported that siderite formation limits H₂/magnetite yield in the 140–260°C range. If this result on Fe metal can be transposed to FeO, slag oxidation under P_CO2 at pH > 12 may be worth exploring.

Although HT carbonation occurred to be detrimental to the H₂ yield, it led to higher carbonation degree compared to RT carbonation. Indeed, mineral sequestration (OC) is, respectively, estimated to 306 and 390 kg CO₂/t of fresh steel slag at 30 and 300°C by TGA (Table 2). Note that the carbonation degree obtained at HT exceeds 100% as defined on the basis of the Ca-content of the steel slag used in this study. This CO₂ sequestration gain at 300°C is explained by (1) the partial carbonation of magnesio-wüstite and (2) the carbonation of the calcium contained in brownmillerite at high temperature. Indeed, no more brownmillerite was detected in the sample recovered from HT carbonation, whereas brownmillerite is still present in the RT-carbonated one. Brownmillerite carbonation liberates Al and Fe³⁺ which are likely responsible for the formation of saponite and hematite, respectively, as secondary products.