Molecular dynamics simulations of LiCl ion pairs in high temperature aqueous solutions by deep learning potential
Introduction
Ion-pair association in high temperature supercritical fluid is of fundamental importance in hydrothermal technologies and hydrothermal geochemistry. The speciation of metal complex in hydrothermal fluids is essential to quantitative understanding its transport, enrichment, and mineralization [1], [2], [3]. Molecular dynamics (MD, both classical MD and ab initio MD), an efficient method to study the association behavior of metal complexes at the micro-level, provides a detailed molecular picture of the ion pairs [4], [5], [6], [7]. Ion association constants could be successfully calculated from the MD potential of mean force (PMF) [8], [9], [10], [11], [12]. The accuracy of classical molecular dynamics (CMD) simulations, is force fields sensitive, which counts on the precision of the applied empirical potential [13]. The force fields parameters obtained with only single ion hydration properties may not accurately enough to predict ion pair associate behavior [14]. In contrast, the ab initio MD (AIMD) simulations provide the accurate simulation results, but, the computational cost of AIMD is very high, and hence the system size and the time length of simulation are highly limited.
To solve the dilemma of accuracy versus efficiency in MD, machining learning (ML) method was adopted in many fields (ref. [15], and references therein). Deep Potential Molecular Dynamics (DPMD, generating ML interatomic potential based on artificial neural network deep learning) is one of the most popular method to obtain excellent interatomic potentials in recent years [16], [17]. Investigation of various systems using DPMD method have been reported [18], [19], [20], [21], [22]. For instance, DPMD potentials have been developed to successfully describe the reactive uptake of N2O5 by atmospheric aerosol [19], urea decomposition in water [18] and nitric and formic acid dissociations at air–water interfaces [23].
In this study, we focus on the lithium chloride ionic association in high temperature aqueous solutions. As a green energy metal, lithium is widely used in batteries, nuclear reactors, aerospace alloys, and other fields with its unique physical and chemical properties [24]. Meeting the increasing demand for lithium requires a fundamental understanding of the geological processes that enable Li concentration in the Earth’s crust [25]. In addition, lithium is an important element used as a geochemical tracer for mass transfer during geological processes [26], [27]. The speciation of lithium in geofluid is essential to quantitatively understand the ore forming processes and rock-fluid interaction occurred in subduction zone [28], [29], [30]. Research of LiCl association is critical in understanding the transport, enrichment and mineralization of Li as chloride is ubiquitous in geofluid [31]. The association behavior and speciation of LiCl under hydrothermal have been investigated by CMD method [7], [32], [33]. Zhang and Duan gave the association constants of LiCl ion-pair under four temperature-density conditions [32]. Wang et al. tabulated LiCl association constants at T = 298–1273 K and ρ = 0.1–1.0 g/cm3 [7]. They found the position and free energy of contact ion-pair (CIP) from CMD and AIMD were different. In addition to CMD, theoretical quantum-chemical studies have been adopted to characterize the hydration structure of Li+ in highly concentrated aqueous solution at room temperature [34]. However, few AIMD work has ever been conducted to investigate the speciation of LiCl under hydrothermal conditions, excepting for Wang’s work, in which AIMD is used to validate the result from CMD at a certain condition (1273 K-0.6 g/cm3).
In this paper, the ML approach in the DeePMD-kit package [16], [17] was used to develop the neural-network potential (NNP) for LiCl-H2O system. The association behavior of LiCl ion-pair was studied at T = 330–1273 K and ρ = 0.45–1.0 g/cm3 with this neural-network-potential and compared with the results from classical MD. The meta-eABF method [35] was used to calculate the potential of mean force (PMF), with which the stability constants were computed and compared with literature reported data. The computational methodology, the data preparation and the DPMD training were presented in Section 2. The simulated results were shown and discussed in Section 3. Section 4 summarized the main conclusions from the work.
Section snippets
Deep potential training
The smooth version of the deep potential (DP) model implemented in DeePMD-kit package (v2.0.3) [16], [17] was used for training interatomic potential energy and force field. The DP model assumed the total potential energy (Ei) could be expressed as a sum of atomic energies (Ei), . Each atomic energy Ei is determined by the local environment matrix of atom i within a smooth cutoff radius Rc. An embedding network is specially designed to map the local environment matrix to an embedded
Validation of DPMD
Fig. 1 compares the atomic forces predicted by DFT and DP for all the structures in the test database with insets showing the distributions of absolute errors. The comparison for atomic energies is shown in Fig. S1 in the supplementary information. We find an overall satisfactory agreement between DFT results and that from DP prediction with a root mean absolute error (RMSE) of 1.71 meV/atom for energy and 64.08 mev/ Å for force, indicating that the obtained DP model predicting the energy and
Conclusion
A neural-network potential model for LiCl ion-pair under hydrothermal fluid was developed via machine learning. The present DP potential document with sufficient accuracy of the microstructures of LiCl solution. Based on this model, the structures of LiCl solution, the 2D free energy surfaces of LiCl dissolution pathway and the PMF profiles were extensively investigated. And the effect of water dissociation on the PMFs was also evaluated. The results show the RDFs, and CNs of LiCl solution and
CRediT authorship contribution statement
Wei Zhang: Conceptualization, Methodology, Software, Supervision. Li Zhou: Resources, Writing – review & editing. Bin Yang: Resources, Writing – review & editing. Tinggui Yan: Conceptualization, Methodology, Validation, Writing – review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
We thank the anonymous reviewers for their careful reading of our manuscript and insightful comments and suggestions. This work was financially supported by the National Natural Science Foundation of China (92162219, 42063008).
References (80)
- et al.
Hydrothermal transport, deposition, and fractionation of the REE: Experimental data and thermodynamic calculations
Chem. Geol.
(2016) - et al.
A molecular dynamics investigation of La3+ and Lu3+-ligand speciation in aqueous solution
J. Mol. Liq.
(2022) - et al.
A molecular dynamics study of Li speciation in hydrothermal fluids and silicate melts
Chem. Geol.
(2021) - et al.
Ion association in binary mixtures of water-CO2 in supercritical conditions through classical molecular dynamics simulations
J. Mol. Liq.
(2018) - et al.
The dissociation mechanism and thermodynamic properties of HCl(aq) in hydrothermal fluids (to 700 °C, 60 kbar) by ab initio molecular dynamics simulations
Geochim. Cosmochim. Acta
(2018) - et al.
Molecular dynamics simulation on Na+−F− ion-pair association from ambient to supercritical water
Fluid Phase Equilib.
(2020) - et al.
DeePMD-kit: A deep learning package for many-body potential energy representation and molecular dynamics
Comput. Phys. Commun.
(2018) - et al.
Dissolving salt is not equivalent to applying a pressure on water
Nat. Commun.
(2022) - et al.
Lithium isotope systematics of the Sumdo Eclogite, Tibet: Tracing fluid/rock interaction of subducted low-T altered oceanic crust
Geochim. Cosmochim. Acta
(2019) - et al.
Lithium and Li-isotopes in young altered upper oceanic crust from the East Pacific Rise
Geochim. Cosmochim. Acta
(2012)
Lithium speciation in aqueous fluids at high P and T studied by ab initio molecular dynamics and consequences for Li-isotope fractionation between minerals and fluids
Geochim. Cosmochim. Acta
Fluid-melt trace-element partitioning behaviour between evolved melts and aqueous fluids: Experimental constraints on the magmatic-hydrothermal transport of metals
Chem. Geol.
Lithium chloride ionic association in dilute aqueous solution: a constrained molecular dynamics study
Chem. Phys.
The structure of aqueous lithium chloride solutions at high concentrations as revealed by a comparison of classical interatomic potential models
J. Mol. Liq.
DP-GEN: A concurrent learning platform for the generation of reliable deep learning based potential energy models
Comput. Phys. Commun.
Quickstep: Fast and accurate density functional calculations using a mixed Gaussian and plane waves approach
Comput. Phys. Commun.
Fast Parallel Algorithms for Short-Range Molecular Dynamics
J. Comput. Phys.
VMD: visual molecular dynamics
J. Mol. Graph.
GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers
SoftwareX
Numerical integration of the cartesian equations of motion of a system with constraints: molecular dynamics of n-alkanes
J. Comput. Phys.
Water in the deep Earth: The dielectric constant and the solubilities of quartz and corundum to 60 kb and 1200 °C
Geochim. Cosmochim. Acta
H3O+/CI− ion pairing in hydrothermal solutions by simulation and electrical conductance. A review
J. Mol. Liq.
Molecular dynamics simulation on ion-pair association of NaCl from ambient to supercritical water
Fluid Phase Equilib.
Solvation structure and dynamics of Li+ ion in liquid water, methanol and ethanol: A comparison study
Chem. Phys.
Ion Pairing
Chem. Rev.
Na+–Cl− ion pair association in supercritical water
J. Chem. Phys.
Ion association in AlCl3 aqueous solutions from constrained first-principles molecular dynamics, lnorg
Chem.
Yttrium speciation in subduction zone fluids from ab initio molecular dynamics simulations
Solid Earth Discuss.
Aqueous Na+Cl− pair association from liquid like to steam like densities along near-critical isotherms
J. Chem. Phys.
Ab Initio Molecular Dynamics Approach to a Quantitative Description of Ion Pairing in Water
J. Phys. Chem. Lett.
A Molecular Dynamics Investigation Of Actinyl-Ligand Speciation In Aqueous Solution
Phys. Chem. Chem. Phys.
Ion pairing in molecular simulations of aqueous alkali halide solutions
J. Phys. Chem. B
First Principles Neural Network Potentials for Reactive Simulations of Large Molecular and Condensed Systems
Angew. Chem. Int. Ed.
Deep potential molecular dynamics: a scalable model with the accuracy of quantum mechanics
Phys. Rev. Lett.
Using metadynamics to build neural network potentials for reactive events: the case of urea decomposition in water
Catal. Today
Reactive uptake of N2O5 by atmospheric aerosol is dominated by interfacial processes
Science
Molecular dynamics simulation of molten strontium chloride based on deep potential
J. Mol. Liq.
Ab Initio Neural Network MD Simulation of Thermal Decomposition of High Energy Material CL-20/TNT
Phys. Chem. Chem. Phys.
Acids at the Edge: Why Nitric and Formic Acid Dissociations at Air-Water Interfaces Depend on Depth and on Interface Specific Area
J. Am. Chem. Soc.
Global Lithium Availability
J. Ind. Ecol.
Cited by (6)
Experimental and molecular dynamics simulation study on solubility characteristics of chloride and sulfate salts in supercritical water
2024, Journal of Supercritical FluidsApplications of machine learning in supercritical fluids research
2023, Journal of Supercritical FluidsAccurate modeling of aqueous chemistry at the nano-/micro-scale
2024, Scientia Sinica InformationisHigh-Throughput Aqueous Electrolyte Structure Prediction Using IonSolvR and Equivariant Graph Neural Network Potentials
2023, Journal of Physical Chemistry LettersSpeciation of La<sup>3+</sup>-Cl<sup>-</sup> Complexes in Hydrothermal Fluids from Deep Potential Molecular Dynamics
2023, Journal of Physical Chemistry BFirst-principles based deep neural network force field for molecular dynamics simulation of N-Ga-Al semiconductors
2022, Physical Chemistry Chemical Physics