Introduction
Granulated blast furnace slag (GBFS) has been widely applied in Marine cement, low-carbon cement, alkali-activated cementitious materials, and geopolymer cementitious materials due to its excellent potential hydration activity and hydration product optimization ability. The glass phase (SGP) is the predominant active mineral in GBFS, and its ionic dissolution ability in an alkaline environment directly influences the hydration activity of GBFS. Therefore, establishing the correlation between the structural characteristic and their hydration properties is imperative for optimizing the utilization of GBFS. Accurately describing the structural characteristics of SGP remains a pivotal challenge that has yet to be resolved. The hydration activity of GBFS can be partially elucidated by investigating the structural characteristics of alkali earth metal aluminosilicate glass using molecular dynamics (MD) simulation. Given the influence of potential functions on the accuracy of MD simulation results, it is imperative to undertake a comparative investigation into SGP structure simulation employing different potential functions. This paper compared the locally ordered structure, oxygen atom type, polymerization behavior, and clustering behavior using four different potentials: Buckingham potential, Broyden-Moré-Hénon (BMH) potential, Miyake potential, and SHIK potential. The accuracy and reference ability of the simulation is achieved by comparing the matched degree between the simulation results and experimental data, thereby providing a valuable reference for investigating the nanostructure and hydration characteristics of SGP in GBFS using MD simulations.
Methods
The SGP structure was simulated using molecular dynamics with the Buckingham potential, BMH potential (optimized based on mineral crystal structure), Miyake potential (combining multiple potential functions), and SHIK potential (optimized through first-principles molecular dynamics). The initial random configuration of the SGP structure, consisting of 3 000 Ca, Si, Al, Mg, and O ions, was generated using the Amorphous Cell module in Materials Studio software. The proportions of different ion types in the SGP structure were determined based on the chemical composition of industrial GBFS and encompassed variations within a specific range for the mass ratio of SiO2/Al2O3 and quaternary alkalinity. A Large-scale Atomic/Molecular Massively Parallel Simulator was employed to calculate the MD process with periodic boundaries in real units using an integral step size of 0.8 fs. The simulation process involved six stages of relaxation to simulate the quench granulation process observed in industrial GBFS.The laboratory synthesis of SGP utilizes chemically pure SiO2, Al2O3, CaO, and MgO powders with a chemical composition matching referenced in the MD simulation. The homogenized oxide powder is melted in a Muffle furnace using a graphite crucible. The melting process occurs at 1 500 ℃ for 2 h, followed by quench cooling using an ice water mixture at 0 ℃. After crushing and grinding the cooled SGP block with an agate grinder, the resulting fine powder (200-mesh) is selected for X-ray photoelectron spectroscopy (XPS), X-ray diffraction (XRD), and 29Si solid-state magic-angle spinning nuclear magnetic resonance (MAS NMR) experiments. These experimental results are used to evaluate the accuracy of the MD simulation findings.
Results and discussion
The MD simulation results of SGP structures under all potential functions are consistent with the glass random network model. The optimization of Buckingham potential, BMH potential, and Miyake potential parameters pertains to specific crystal structures, offering advantages over SHIK potential in simulating local ordered structure characteristics. Furthermore, a comparison between the MD simulation results and XPS experimental data reveals that Buckingham potential, BMH potential, and Miyake potential exhibit contrasting or localized fluctuation patterns in bridging oxygen, non-bridging oxygen, and free oxygen contents compared to the experimental findings. Notably, an increase in alkalinity leads to a decrease in non-bridging oxygen content, according to some simulation results; however, this contradicts the charge balance principle of glass structure. In comparison, the variation trend of non-bridging oxygen and bridging oxygen under SHIK potential aligns with the outcomes observed in XPS experiments.The experimental results from 29Si MAS NMR indicate that SGP primarily consists of Q0 and Q1 types [SiO4]. However, the simulation results using all potential functions exhibit a significantly higher degree of polymerization for [SiO4], with a large number of Q3 and Q4 types present. This discrepancy can be attributed to the short time scale of the simulation process, leading to an abundance of free oxygen. These findings highlight inherent technical limitations in MD simulations for accurately predicting the quantitative distribution of Qn. Notably, only the SHIK potential simulations demonstrate consistent trends with experimental observations, whereas Buckingham potential, BMH potential, and Miyake potential fail to capture the decreasing trend in polymerization degree as the SiO2/Al2O3 mass ratio decreases and alkalinity increases. Combined with the quantitative XRD data, the clustering behavior of Ca remained largely unaltered in the Miyake potential as the local enrichment of Ca increased; however, a transition from isolated Ca to clustered Ca was observed in both BMH and SHIK potentials.
Conclusions
The main conclusions of this paper are summarized as follows. Buckingham, BMH and Miyake potential exhibit suitability for simulating local ordered structure. Moreover, SHIK potential has higher stability in simulating the effect of chemical composition on the content variation trend of different types of oxygen and the Qn distribution. Lastly, SHIK potential and BMH potential, which define the interaction forces between metal ions, are more suitable for the clustering behavior as network modifiers. This study demonstrates that for the optimization of MD simulation of SGP in GBFS, it is a reasonable approach to add the potential parameters governing ion interactions and optimize the existing potential based on Ab initio molecular dynamics simulations of target structure characteristics.