diff --git a/source/_posts/ABACUS_15_04_2025.md b/source/_posts/ABACUS_15_04_2025.md
new file mode 100644
index 00000000..b7b96f6c
--- /dev/null
+++ b/source/_posts/ABACUS_15_04_2025.md
@@ -0,0 +1,111 @@
+---
+title : "Hybrid Gauge RT-TDDFT: An Algorithmic Innovation with Both Efficiency and Accuracy"
+date : 2025-04-15
+categories:
+- ABACUS
+---
+## Abstract
+
+Recently, Zhao Haotian, a PhD candidate at the University of Science and Technology of China, and Professor He Lixin proposed and implemented a Hybrid Gauge Real-Time Time-Dependent Density Functional Theory (Hybrid gauge rt-TDDFT) applicable to atomic orbital basis sets in the domestic open-source density functional theory software ABACUS. Based on the traditional velocity gauge, this method introduces a time-varying phase dependent on the vector potential, effectively overcoming the systematic errors caused by the incompleteness of the local basis set and providing consistent and reliable simulation results in both periodic and non-periodic systems. At the same time, this method significantly improves the calculation efficiency in periodic systems, offering a new solution that combines accuracy and efficiency for first-principles real-time dynamics simulations under the action of an electric field.
+## Introduction
+
+Real-time time-dependent density functional theory (rt-TDDFT) is an important theoretical method for studying the electronic dynamics of materials in excited states and under the drive of external fields. It is widely used in frontier fields such as nonlinear optical responses, ultrafast spectroscopy, and charge and energy transport. Compared with linear response methods, rt-TDDFT can directly track the quantum dynamics trajectory of electrons during time evolution, making it particularly suitable for strong-field excitation and non-linear processes far from equilibrium. Therefore, it has become a key means for studying phenomena such as light-matter interaction and carrier relaxation. However, conducting rt-TDDFT calculations in periodic systems still faces challenges. The direct application of the traditional velocity gauge in numerical atomic orbital (NAO) basis sets ignores the internal phase changes of orbitals caused by the vector potential, resulting in systematic errors that seriously affect the calculation accuracy of key physical quantities such as current responses and nonlinear optics. To solve this problem, Zhao Haotian and He Lixin proposed the hybrid gauge algorithm. By explicitly introducing the local phase correction induced by the vector potential, it effectively overcomes the limitations of the velocity gauge in local basis sets and constructs a rigorous and efficient theoretical framework, providing a new solution for the study of ultrafast electron dynamics in periodic systems.
+
+## Hybrid Gauge Theory
+
+In rt-TDDFT, we can simulate the dynamic response of a system under external excitations such as an electric field by adding a time-dependent external field term to the Hamiltonian. The most commonly used is the length gauge, and its Hamiltonian form is relatively simple, containing only an additional scalar potential term:
+ ```math
+H = H_0 + E(t) \cdot r
+```
+where $`H_0`$is the Hamiltonian of the system itself, and $`E(t)\cdot r`$ describes the interaction between the electric field and electrons. The corresponding time-dependent Kohn-Sham equation is:
+```math
+i\frac{\partial}{\partial t}\psi_{i}(r,t) = \left[-\frac{1}{2}\nabla^{2} + V_{KS}[\rho](r,t) + E(t) \cdot r\right]\psi_{i}(r,t)
+```
+where $`V_{KS}`$represents the Kohn-Sham effective potential, including electron-ion interaction, Hartree potential, exchange-correlation potential, etc. The length gauge is applicable to non-periodic systems. However, in periodic systems, this form destroys the periodic structure of the system, so it cannot be directly applied to calculations of crystal materials and other systems with periodic potential fields.To solve this problem, people usually use the method of gauge transformation to convert the Hamiltonian into the velocity gauge form. The electric field is derived from the time-dependent vector potential $`A(t)`$:
+```math
+E(t)=-\frac{dA(t)}{dt}
+```
+The Hamiltonian in the velocity gauge is:
+```math
+\hat{H} = \frac{1}{2} (p + A(t))^{2}+V_{l}^{ps}+V_{Hxc}+\tilde{V}_{nl}^{ps}
+```
+Its equation of motion is:
+
+```math
+\begin{equation} i\frac{\partial}{\partial t} \tilde{\psi}_i(\boldsymbol{r}, t) = \tilde{H} \tilde{\psi}_i(\boldsymbol{r}, t) \end{equation}
+```
+
+Theoretically, the length gauge and the velocity gauge are essentially equivalent, and they can be transformed into each other through the following gauge transformation:
+
+```math
+\tilde{\psi}(\boldsymbol{r},t)=e^{-iA(\boldsymbol{t})\cdot\boldsymbol{r}}\psi_i(\boldsymbol{r},t)
+```
+
+```math
+\tilde{H}=e^{-iA(\boldsymbol{t})\cdot\boldsymbol{r}}He^{iA\cdot\boldsymbol{r}}
+```
+
+However, in actual calculations, because we use finite basis sets (such as atomic orbitals), this phase relationship cannot be perfectly expressed, resulting in differences in calculation results under the two gauges. For example, in the atomic orbital basis set, the wave functions of both the length gauge and the velocity gauge are written in the form of a linear combination of atomic orbitals (LCAO):
+
+```math
+\begin{align} \psi_n(\boldsymbol{r},t) &= \sum_{\mu} c_{n\mu}(t)\phi_{\mu}(\boldsymbol{r}) \\ \tilde{\psi}_n(\boldsymbol{r},t) &= \sum_{\mu} \tilde{c}_{n\mu}(t)\phi_{\mu}(\boldsymbol{r}) \end{align}
+```
+
+Obviously, the wave function expression in the velocity gauge ignores the phase change inside the atomic orbitals caused by the vector potential, thus introducing systematic errors in the simulation, especially in the calculation of key physical quantities such as current and energy. Although the errors can be reduced by increasing the basis set, this will significantly increase the computational cost and severely limit the application efficiency of rt-TDDFT in periodic systems.
+To compensate for the phase effect introduced by the gauge transformation, we introduce a phase factor dependent on the vector potential in front of each atomic orbital:
+
+```math
+\bar{\phi}_{\mu}(\boldsymbol{r}-\boldsymbol{\tau}_{\mu}-\boldsymbol{R}_{i},t)=e^{-i\boldsymbol{A}(t)\cdot\Delta\boldsymbol{r}_{\mu i}}\phi_{\mu}(\boldsymbol{r}-\boldsymbol{\tau}_{\mu}-\boldsymbol{R}_{i})
+```
+
+where $`\boldsymbol{\tau}_{\mu}`$is the center position of the atomic orbital. The form of the wave function is:
+
+```math
+\begin{equation} \bar{\psi}_n(\boldsymbol{r},t)=\sum_{\mu} \bar{c}_{n\mu}(t)\bar{\phi}_{\mu}(\boldsymbol{r}) \end{equation}
+```
+
+Under this phase correction, the Hamiltonian matrix and the overlap matrix of the orbitals can finally be obtained as:
+
+```math
+\begin{align} \tilde{H}_{\mu\nu}(\boldsymbol{R}_{i}) &= e^{-i\boldsymbol{A}(t)\cdot\boldsymbol{\tau}_{\mu 0,\nu i}}\langle\phi_{\mu 0}|H_{0}+\boldsymbol{E}(t)\cdot\Delta\boldsymbol{r}_{\nu i}|\phi_{\nu i}\rangle\\ \tilde{S}_{\mu\nu}(\boldsymbol{R}_{i}) &= e^{-i\boldsymbol{A}(t)\cdot\boldsymbol{\tau}_{\mu 0,\nu i}}S_{\mu\nu}(\boldsymbol{R}_{i}) \end{align}
+```
+
+And the time-dependent Kohn-Sham equation is:
+
+```math
+i\sum_{\nu}\left(\frac{\partial}{\partial t}\bar{c}_{n\nu k}(t)\right)\bar{S}_{\mu\nu}(k,t)=\sum_{\nu} \bar{c}_{n\nu k}(t)\bar{H}_{\mu\nu}(k,t)
+```
+
+The above transformation can be regarded as a gauge transformation. Since this gauge contains both the vector potential $`A(t)`$ and the scalar field $`E(t)`$, we call it the hybrid gauge. The Hamiltonian form in the hybrid gauge is similar to that in the length gauge, but its external field term $`\boldsymbol{E}(t)\cdot\Delta\boldsymbol{r}_{\nu i}`$remains periodic, avoiding the defect of the length gauge that violates Bloch's theorem in periodic systems while retaining its advantage of simple calculation. Compared with the velocity gauge, the hybrid gauge also has significant advantages in computational efficiency, and it is more efficient in dealing with non-local potentials. Other physical quantities, such as charge density and current density, can also be processed in a similar way. Since there is a strict gauge transformation relationship between the hybrid gauge and the length gauge, the calculation results of the two are theoretically completely equivalent.
+
+## Results
+
+### Non - periodic Systems
+
+
+
+
+
+The figure above shows the response current of the Si unit cell under pulse excitation and its corresponding dielectric function. It can be seen from the results that the velocity gauge has obvious systematic errors, but through the correction scheme we proposed, these errors can be effectively alleviated. In contrast, the hybrid gauge can stably provide more accurate and physically consistent results.
+
+
+
+
+
+
+The figure above shows the response current of the Si unit cell under strong Gaussian pulse excitation and its corresponding high - order harmonic spectrum. As the maximum field strength increases from $`10^{10},W/cm^{2}`$ to $`10^{13},W/cm^{2}`$, the current signal gradually deviates from the fundamental frequency of the external field, exhibiting an increasingly enhanced nonlinear response characteristic. Meanwhile, the 1st, 3rd, 5th, and 7th harmonic signals appear successively in the high - order harmonic spectrum.
+The above results indicate that the hybrid gauge is applicable in different physical scenarios, including non - periodic and periodic systems, as well as strong - field and weak - field perturbations. It demonstrates good versatility and accuracy. Compared with the velocity gauge, it shows higher computational accuracy and efficiency in periodic systems. It can achieve stable and reliable real - time dynamics simulations under small basis set conditions, providing solid theoretical and technical support for the subsequent development of rt - TDDFT methods based on NAO.
+
+## Summary
+
+In this paper, a Hybrid Gauge Real-Time Time-Dependent Density Functional Theory (Hybrid gauge rt-TDDFT) based on atomic orbital basis sets is proposed and implemented. It effectively overcomes the systematic error problem of the velocity gauge in local basis sets, and demonstrates excellent accuracy and computational efficiency under various physical conditions, such as non-periodic and periodic systems, strong fields and weak fields. This method not only provides a reliable means for real-time electron dynamics simulations with small basis sets, but also lays a solid foundation for the further development and expansion of the domestic first-principles software ABACUS.
+The relevant results were recently published in Journal of Chemical Theory and Computation, and the link to the paper is as follows: https://doi.org/10.1021/acs.jctc.5c00111
\ No newline at end of file
diff --git a/source/_posts/ABACUS_16_04_2025.md b/source/_posts/ABACUS_16_04_2025.md
new file mode 100644
index 00000000..b2fae701
--- /dev/null
+++ b/source/_posts/ABACUS_16_04_2025.md
@@ -0,0 +1,59 @@
+---
+titile : ABACUS Can Also Do This? Helping to Understand the Layer-Dependence of Magnetic Anisotropy in 2D Magnetic Materials
+date : 2025-04-16
+categories :
+- ABACUS
+---
+
+Recently, Associate Professor Hu Zhixin from Tianjin University collaborated with the research groups of Professor Ji Wei and Associate Researcher Wang Cong from Renmin University of China. Based on first-principles calculations and using both VASP and the domestic first-principles software ABACUS, they revealed the microscopic mechanisms of the changes in the easy magnetization axis and topological properties with the number of layers in MnSe₂. The relevant research results were published in the journal Physical Review B under the title "Interlayer coupling driven rotation of the magnetic easy axis in MnSe₂ monolayers and bilayers" (DOI: 10.1103/PhysRevB.111.054422). The first authors of the paper are Zhang Zhongqin and Wang Cong.
+
+
+
+## Research Background
+
+Magnetic anisotropy plays a crucial role in maintaining the long-range magnetic order of 2D magnets at finite temperatures and is closely related to the magnetic coercivity (a core parameter determining the hard or soft magnetic behavior of materials). As an effective means of regulating magnetism, interlayer coupling has received extensive attention in 2D materials in recent years. However, research on how interlayer coupling affects magnetic anisotropy remains relatively limited, which has hindered to some extent the in-depth understanding of its physical mechanism and the expansion of its applications in spintronics.
+
+## Research Results
+
+1T-MnSe₂ is a 2D van der Waals ferromagnetic metal. In monolayers and bilayers, Mn atoms within the layer or between different layers are ferromagnetically coupled. The calculation results show that the easy magnetization axis of monolayer MnSe₂ is along an inclined direction deviated 67° from the z-axis; in the bilayer, the easy magnetization axis rotates to the z-axis direction.
+
+
+
+
+
+ *Figure 1 : (a) Top and side views of monolayer MnSe₂; (b - c) Side and oblique views of AA-stacked bilayer MnSe₂; (d) Definition of polar angle θ and azimuthal angle φ in the spherical coordinate system; (e - f) Energies of magnetic moments of monolayer (e) and bilayer (f) MnSe₂ along different directions.*
+
+The calculation results of the interlayer differential charge density (Figure 2a) indicate that MnSe₂ has a strong interlayer coupling. The researchers further decomposed the contribution of the magnetic anisotropy energy (MAE) to atoms (Figure 2b) and orbitals (Figure 2c - d), and found that the interaction between the $`p_y`$ and $`p_z`$ orbitals of interface Se atoms plays a key role in the transformation of the easy magnetization axis.
+
+
+
+
+
+*Figure 2 : (a) Side view of the interlayer differential charge density of bilayer MnSe₂; (b) Contributions of Mn and Se atoms to MAE in monolayer and bilayer MnSe₂; (c - d) Contributions of orbitals of monolayer Se and bilayer Se-interface to MAE.*
+
+According to the second-order perturbation theory, the contribution of electron states to MAE can be expressed by the following formula:
+
+
+where o and u represent the occupied and unoccupied states, respectively. Since the energy difference $`E_{o}-E_{u}`$ between the occupied and unoccupied states appears in the denominator, the states closer to the Fermi level have a greater impact on MAE, while the states far from the Fermi level contribute relatively less.
+
+Combined with the electronic structure analysis, the researchers found that in monolayer MnSe₂, the p_z orbital of Se atoms is far from the Fermi level (Figure 3a, 3c), so the coupling between $`p_z`$ and $`p_y`$ is weak; in the bilayer structure, the interlayer coupling causes the $`p_z`$ orbitals of interface Se atoms to hybridize, forming bonding and antibonding states (Figure 3d). The antibonding states split and approach the Fermi level, thus enhancing the coupling between the $`p_y`$ and $`p_z`$ orbitals and making the easy magnetization axis of bilayer MnSe₂ out-of-plane.
+
+In addition, MnSe₂ also exhibits topological properties that change with the number of layers, including the evolution of the Chern number and surface states (Figure 3e - f). The layer evolution of the above electronic structure and topological properties was calculated and verified using the domestic first-principles software ABACUS.
+
+
+
+
+
+*Figure 3 here: (a - b) Spin-down band structures of monolayer and bilayer MnSe₂; (c) Projected density of states of $`p_y`$ and $`p_z`$ orbitals of (interface) Se at the Gamma point in monolayer and bilayer; (d) Charge densities of the marked states in (a - c); (e - f) Surface states of monolayer and bilayer MnSe₂.*
+
+Some external regulation methods can also affect the occupation state of the p orbitals of Se atoms, and thus are expected to achieve the regulation of the direction of the easy magnetization axis of the material. Based on this, the researchers systematically studied a variety of external regulation methods. The results show that by changing the interlayer stacking mode (Figure 4a - b), applying charge doping (Figure 4c), introducing biaxial strain (Figure 4d), and replacing non-metal atoms, the direction of the easy magnetization axis of MnSe₂ can be effectively regulated, providing new ideas for realizing the controllable regulation of magnetic anisotropy in 2D magnets.
+
+
+
+
+
+*Figure 4 here: (a) Top and side views of AB-stacked bilayer MnSe₂; (b) Atom-decomposed MAE of AA and AB stackings; (c - d) Contributions of different atoms to MAE in monolayer MnSe₂ and the changes of $`E_{X}-E_{ea}`$ with doping concentration and in-plane biaxial strain.*
+
+## Conclusion
+
+This study reveals the key role of interlayer coupling in regulating the direction of the easy magnetization axis of 2D magnetic materials. Through the MAE analysis of atomic and orbital decomposition, it is found that non-metal Se atoms, especially the occupation and coupling of their p orbitals, play a core role in the rotation of the easy magnetization axis. These findings not only deepen the understanding of the physical mechanism of magnetic anisotropy in 2D magnets but also provide a theoretical basis and material foundation for the design of future spintronic devices.
\ No newline at end of file
diff --git a/source/_posts/ABACUS_17_04_2025.md b/source/_posts/ABACUS_17_04_2025.md
new file mode 100644
index 00000000..7334b825
--- /dev/null
+++ b/source/_posts/ABACUS_17_04_2025.md
@@ -0,0 +1,50 @@
+---
+title : "Can DP + ABACUS Also Do This? Exploring Nuclear Quantum Effects in Electrochemical Reactions with Machine Learning and Developing Computational Methods"
+date : 2025-04-17
+categories :
+- ABACUS
+---
+
+Recently, the research group led by Researcher Xu Shenzhen from the School of Materials Science and Engineering at Peking University collaborated with the Beijing Academy of Intelligent Sciences (AISI) and Deep Potential Technology (DP Technology). They employed the deep potential method[1] to study the nuclear quantum effects in proton-coupled electron transfer during electrochemical reactions and develop computational methods. Notably, all first-principles calculations in this study were carried out using the domestic first-principles software ABACUS [2]. The relevant research findings were published in Nature Communications under the title "Probing Nuclear Quantum Effects in Electrocatalysis via a Machine-Learning Enhanced Grand Canonical Constant Potential Approach" [3]. Doctoral students Sun Menglin, Jin Bin, and Yang Xiaolong are the co-first authors, and Researcher Xu Shenzhen is the corresponding author.
+
+
+
+## Research Background
+
+Electrocatalytic systems can efficiently and cleanly convert electrical energy into chemical energy, making them one of the widely concerned fields in new energy technologies. They cover various scenarios such as water electrolysis, CO₂ reduction, and nitrogen reduction. The thermodynamic and kinetic properties of the elementary step of proton-coupled electron transfer (PCET) on the surface of electrode or catalyst materials, which is a crucial step in energy conversion, are key factors influencing performance indicators of electrocatalytic systems, such as product formation rate, energy conversion efficiency, and selectivity. They are also fundamental scientific issues that are the focus of experimental and theoretical electrochemical research.
+
+Microscopic simulation of electrocatalytic systems is an important tool for studying the PCET mechanism. However, the electrochemical open system formed jointly by the electrode surface and the solution poses great challenges to computational simulation. How to consider the constant potential condition, sampling of the complex electrode-solution interface structure, and nuclear quantum effects (NQEs) of protons in a rigorous and efficient manner in the simulation has become a key problem to be solved urgently in the field of electrochemical theoretical calculation. For a typical elementary step of PCET on the electrode surface, such as $\text{A}^* + \text{H}^+_{\text{sol}} + \text{e}^- \rightarrow {}^* \text{AH}$ (where "*" represents the surface adsorption site), the traditional simulation method is to optimize the structures of the initial and final states of the reaction and then use the transition state search method to calculate the reaction energy barrier. The defects of this set of methods are as follows: (1) Only specific initial and final state structures are considered, and the reaction path is single, lacking statistical averaging; (2) In the simulation of the reaction path, the number of electrons in the surface system remains constant, which does not conform to the experimental condition of constant potential. Some post-treatment constant potential correction methods also cannot handle this problem strictly; (3) The nuclear quantum effects of hydrogen cannot be considered, while NQEs have been proven to have a significant impact on proton transfer kinetics in many other application scenarios (also at room temperature).
+
+Based on the above problems, the research team developed a unified computational framework that can accurately handle NQEs under strict grand canonical constant potential conditions. Combined with the deep potential (DP) machine learning force field, it achieved sufficient sampling of electrochemical open systems. The DP force field can maintain an accuracy comparable to that of density functional theory (DFT) while greatly improving the computational efficiency, making it possible to conduct sufficient statistical sampling of the complex configurations of the electrode-solution interface in the reaction path at an affordable computational cost.
+
+## Research Results
+
+This work is based on a rigorous statistical physics model. It uses the grand canonical hybrid Monte Carlo (GC-HMC) algorithm and the path integral Monte Carlo (PIMC) method, combined with the DP machine learning force field, to simulate the nuclear quantum effects of protons under precisely constant potential conditions. The workflow of this GC-(PI)HMC is shown in Figure 1a. In addition, based on the traditional DP machine learning potential function that takes atomic positions as inputs, the research team introduced a new input parameter degree of freedom: the total number of electrons Ne in the interface model system. Figure 1b shows the construction framework and training workflow of the DP-Ne machine learning force field used in this study. In order to calculate the work function W of the instantaneously sampled configuration in the extended space [R, Ne], an additional output term $`\frac{\partial E(R,N_{e})}{\partial N_{e}}=-W(R,N_{E})`$ was introduced. For effective grand canonical ensemble sampling, it is necessary to satisfy: $`<\frac{\partial E\left(R, N_{e}\right)}{\partial N_{e}}>_{GC}=<-W\left(R, N_{e}\right)>_{GC}=\mu_{e}`$, where $`\mu_{e}`$ represents the external electrochemical potential in equilibrium with the system and is a thermodynamic macroscopic quantity. The updated DP-Ne force field is helpful for sampling grand canonical systems with variable electron numbers.
+
+
+
+
+
+*Figure 1. a GC-(PI)HMC method workflow. The trial steps of different types of degrees of freedom are selected from PIMC($`\mathbf{R}_\Delta^{(k)}`$), centroid atomic coordinates (R), and the number of electrons Ne using the random variable ξ according to a preset ratio. b The construction framework and training workflow of the DP-Ne force field used in this study. The added new degree of freedom - the total number of electrons Ne in the interface system is used as an input parameter for the fitting network.*
+
+The research team applied the new method to the hydrogen evolution reaction (HER) on the surface of a Pt electrode - a classic electrocatalytic science problem scenario, focusing on the impact of nuclear quantum effects on the thermodynamics and kinetics of the reaction. Electrocatalytic HER involves two basic PCET processes: Volmer ($`\text{H}^+_{\text{sol}} + \text{e}^- + {}^* \rightarrow \text{H}^*`$), Heyrovsky ($`\text{H}^+_{\text{sol}} + \text{e}^- + \text{H}^* \rightarrow \text{H}_2`$), and a non-electrochemical Tafel step (H* + H* ➜ H₂). After considering NQEs in the free energy calculations of the Volmer and Heyrovsky PCET steps, the obtained activation free energy results decreased by 0.1 - 0.15 eV (see Figure 2d), indicating that traditional electrocatalytic calculation models that do not consider NQEs may underestimate the PCET reaction rate at room temperature by dozens of times. The research results also provide a clear physical picture of proton tunneling when overcoming the energy barrier in the PCET path. The characteristic is that the position of the quantized proton has obvious uncertainty, reflecting the inherent quantum nature of proton transfer (see Figure 2e, 2f).
+
+
+
+
+
+*Figure 2. a Work function and b number of electrons fluctuations with the number of MC steps in the Volmer reaction. c The change process of the total number of electrons in the interface system during the reaction paths of Volmer and Heyrovsky reactions under classical conditions. d Free energy diagrams of Volmer and Heyrovsky steps under classical and quantum conditions. e Reaction coordinate (RC) distributions of ring-polymer beads in the path integral simulations of Volmer and Heyrovsky reactions. f Schematic diagrams of the structures of the beads when the transferred proton or hydrogen atom is considered with NQEs.*
+
+## Summary
+
+In this work, the researchers developed a GC-(PI)HMC computational framework that can explicitly handle NQEs under precisely constant potential conditions. With the assistance of a machine learning force field suitable for electrochemical conditions, it can accurately describe the PCET mechanism in electrochemistry. The work also reveals that NQEs have a non-negligible impact on HER on the Pt surface at room temperature. The quantum properties of the transferred protons are conducive to particles tunneling through classical potential barriers in the PCET path, thus significantly reducing the activation free energy compared with classical simulations. This discovery provides new physical insights into how protons overcome kinetic barriers during transfer.
+
+This study not only provides new tools and methods for the theoretical research of electrocatalytic HER but also offers new research means for PCET reactions in a wider range of energy conversion processes. The computational framework developed by the research team is expected to become an important tool for future electrochemical simulations and provide theoretical support for the development of green energy technologies.
+
+## References
+
+[1] Zhang, L., Han, J., Wang, H., Car, R. & E, W. Deep Potential Molecular Dynamics: A Scalable Model with the Accuracy of Quantum Mechanics. Phys. Rev. Lett. 120, 143001 (2018).
+
+[2] Chen, M., Guo, G.-C. & He, L. Systematically Improvable Optimized Atomic Basis Sets for Ab Initio Calculations. J. Phys.: Condens. Matter 22, 445501 (2010).
+
+[3] Sun, M., Jin, B., Yang, X. & Xu, S. Probing Nuclear Quantum Effects in Electrocatalysis via a Machine-Learning Enhanced Grand Canonical Constant Potential Approach. Nature Communications 16, 3600 (2025)
\ No newline at end of file