Molecular dynamics simulations of LiCl ion pairs in high temperature aqueous solutions by deep learning potential

https://doi.org/10.1016/j.molliq.2022.120500Get rights and content

Highlights

  • •

    A neural-network potential (NNP) model for LiCl ion-pair under hydrothermal fluid was developed.

  • •

    The NNP document with sufficient accuracy of the microstructures of LiCl solution.

  • •

    The Li-Cl ion pair association was investigated in a wide range of T-ρ (330–1273 K and 0.45–1.0 g/cm3) conditions.

  • •

    The association constants calculated from DPMD are satisfactory compared with the experimental data.

Abstract

Molecular dynamics simulation is an efficient method to study ion-pair association in high temperature supercritical fluid. Interatomic potentials based on neural-network machine learning shows outstanding ability of balancing the accuracy and the efficiency in molecular dynamics (MD) simulations. In present study, a neural-network potential (NNP) model for LiCl ion-pair in high temperature aqueous solutions was developed using database obtained by the first-principles density functional theory (DFT) calculations. With this NNP model, the structures of LiCl solution and the dissociation pathway of LiCl dissociation process were investigated. The results show that deep learning molecular dynamic (DPMD) simulations can accurately reproduce the radial distribution functions of ab initio MD simulations. And several metastable states were clearly identified from the dissociation 2D energy surfaces. In addition, the potential of mean force (PMF) profiles and corresponding association constants (Ka) were extensively investigated under a wide range of temperature-density (T = 330–1273 K and ρ = 0.45–1.0 g/cm3) conditions. The association constants calculated from DPMD are satisfactory compared with the experimental data. The study indicates that deep learning potential exhibits good capabilities to describe the association behavior of metal complex in high temperature aqueous solutions. This work also provides the microstructures and LiCl association constants in temperature aqueous solutions for which no experimental data exist.

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), E=∑iEi. 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)

  • S. Jahn et al.

    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

    (2009)
  • A.A. Iveson et al.

    Fluid-melt trace-element partitioning behaviour between evolved melts and aqueous fluids: Experimental constraints on the magmatic-hydrothermal transport of metals

    Chem. Geol.

    (2019)
  • Z. Zhang et al.

    Lithium chloride ionic association in dilute aqueous solution: a constrained molecular dynamics study

    Chem. Phys.

    (2004)
  • I. Pethes

    The structure of aqueous lithium chloride solutions at high concentrations as revealed by a comparison of classical interatomic potential models

    J. Mol. Liq.

    (2018)
  • Y. Zhang et al.

    DP-GEN: A concurrent learning platform for the generation of reliable deep learning based potential energy models

    Comput. Phys. Commun.

    (2020)
  • J. VandeVondele et al.

    Quickstep: Fast and accurate density functional calculations using a mixed Gaussian and plane waves approach

    Comput. Phys. Commun.

    (2005)
  • S. Plimpton

    Fast Parallel Algorithms for Short-Range Molecular Dynamics

    J. Comput. Phys.

    (1995)
  • W. Humphrey et al.

    VMD: visual molecular dynamics

    J. Mol. Graph.

    (1996)
  • M.J. Abraham et al.

    GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers

    SoftwareX

    (2015)
  • J.-P. Ryckaert et al.

    Numerical integration of the cartesian equations of motion of a system with constraints: molecular dynamics of n-alkanes

    J. Comput. Phys.

    (1977)
  • D.A. Sverjensky et al.

    Water in the deep Earth: The dielectric constant and the solubilities of quartz and corundum to 60 kb and 1200 °C

    Geochim. Cosmochim. Acta

    (2014)
  • A.A. Chialvo et al.

    H3O+/CI− ion pairing in hydrothermal solutions by simulation and electrical conductance. A review

    J. Mol. Liq.

    (2003)
  • K. Yui et al.

    Molecular dynamics simulation on ion-pair association of NaCl from ambient to supercritical water

    Fluid Phase Equilib.

    (2010)
  • Y. Zeng et al.

    Solvation structure and dynamics of Li+ ion in liquid water, methanol and ethanol: A comparison study

    Chem. Phys.

    (2014)
  • Y. Marcus et al.

    Ion Pairing

    Chem. Rev.

    (2006)
  • A.A. Chialvo et al.

    Na+–Cl− ion pair association in supercritical water

    J. Chem. Phys.

    (1995)
  • E. Cauët et al.

    Ion association in AlCl3 aqueous solutions from constrained first-principles molecular dynamics, lnorg

    Chem.

    (2012)
  • J. Stefanski et al.

    Yttrium speciation in subduction zone fluids from ab initio molecular dynamics simulations

    Solid Earth Discuss.

    (2020)
  • A.A. Chialvo et al.

    Aqueous Na+Cl− pair association from liquid like to steam like densities along near-critical isotherms

    J. Chem. Phys.

    (2003)
  • E. Pluhařová et al.

    Ab Initio Molecular Dynamics Approach to a Quantitative Description of Ion Pairing in Water

    J. Phys. Chem. Lett.

    (2015)
  • K. Newcomb et al.

    A Molecular Dynamics Investigation Of Actinyl-Ligand Speciation In Aqueous Solution

    Phys. Chem. Chem. Phys.

    (2018)
  • C.J. Fennell et al.

    Ion pairing in molecular simulations of aqueous alkali halide solutions

    J. Phys. Chem. B

    (2009)
  • J. Behler

    First Principles Neural Network Potentials for Reactive Simulations of Large Molecular and Condensed Systems

    Angew. Chem. Int. Ed.

    (2017)
  • L. Zhang et al.

    Deep potential molecular dynamics: a scalable model with the accuracy of quantum mechanics

    Phys. Rev. Lett.

    (2018)
  • M. Yang et al.

    Using metadynamics to build neural network potentials for reactive events: the case of urea decomposition in water

    Catal. Today

    (2021)
  • M. Galib et al.

    Reactive uptake of N2O5 by atmospheric aerosol is dominated by interfacial processes

    Science

    (2021)
  • D. Guo et al.

    Molecular dynamics simulation of molten strontium chloride based on deep potential

    J. Mol. Liq.

    (2021)
  • L. Cao et al.

    Ab Initio Neural Network MD Simulation of Thermal Decomposition of High Energy Material CL-20/TNT

    Phys. Chem. Chem. Phys.

    (2022)
  • M. de la Puente et al.

    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.

    (2022)
  • P.W. Gruber et al.

    Global Lithium Availability

    J. Ind. Ecol.

    (2011)

    • Cited by (6)

    View full text
    © 2022 Elsevier B.V. All rights reserved.