betaFeedback

WorksPhysics and Astronomy

DeePKS Model for Halide Perovskites with the Accuracy of Hybrid Functional

Abstract

Accurate prediction for the electronic structure properties of halide perovskites plays a significant role in the design of highly efficient and stable solar cells. While density functional theory (DFT) within the generalized gradient approximation (GGA) offers reliable prediction in terms of lattice constants and potential energy surface for halide perovskites, it severely underestimates the band gap due to the lack of non-local exact exchange term, which exists in computationally expensive hybrid functionals. In this work, a universal Deep Kohn-Sham (DeePKS) model based on neural network is trained so as to enable electronic structure calculations with the accuracy of hybrid functional HSE06 and the efficiency comparable to GGA functional, for a plethora of halide perovskites, i.e., ABX3_3 (A=FA, MA, Cs; B=Sn, Pb; X=Cl, Br, I). Forces, band gaps, and density of states (DOS) predicted by our DeePKS model for all aforementioned perovskites are in good agreement with the HSE06 results, with significantly improved efficiency. In addition, even though the spin-orbit coupling (SOC) effect has not been taken into consideration during the training process, DeePKS+SOC offers highly consistent band gap and DOS as compared to HSE06+SOC for Pb-containing systems. We believe such DeePKS model can be readily applied for an accurate yet efficient prediction of various properties for the family of halide perovskites.

improved efficiency. In addition, even though the spin-orbit coupling (SOC) effect has not been taken into consideration during the training process, DeePKS+SOC offers highly consistent band gap and DOS as compared to HSE06+SOC for Pb-containing systems. We believe such DeePKS model can be readily applied for an accurate yet efficient prediction of various properties for the family of halide perovskites.

Introduction

Owing to the unsustainable nature of the fossil-based-energy sources, new forms of energy sources that are both renewable and environmentally friendly have been comprehensively investigated world-wide. [1, 2, 3, 4, 5] Solar cell, or photovoltaic, is regarded as one of the promising alternatives to traditional fossil fuel resources. [6, 7, 8, 9] While the silicon-based solar cells, i.e., the first- and second-generation solar cells, exhibit good performance in terms of stability and efficiency,[10, 11, 12, 13] the cost of ultra-high-pure metallic silicon somewhat hinders the application of such solar cells. Third-generation photovoltaic including organic solar cells and dye-sensitized solar cells has been developed later on, with significantly reduced processing cost.[14, 15, 16, 17] Nevertheless, the power conversion efficiency (PCE) of the third-generation photovoltaic remains less satisfactory compared with the silicon-based modules,[18, 19, 20] which limits their chances for commercialization.

First synthesized in 1890s,[21] halide perovskites (ABX3\mathrm{ABX}_3, X=halogen anion\mathrm{X}=\text{halogen anion}) have shown immense potential to become a low-cost alternative to the currently commercialized photovoltaic technologies owing to their desired semiconducting properties and the economic fabrication.[20, 22, 23, 24, 25] In addition to the relatively high PCE reached by halide perovskites, one of their favored traits that is rarely possessed by other conventional semiconductors is the tunable absorption edge wavelength (band gap),[26, 27, 28] which can be straightforwardly realized by varying the ratio of different halide ions. Accurate theoretical description of the electronic structure for halide perovskites is a long-standing goal for the realistic design of photovoltaic with higher efficiency and better stability. Density functional theory (DFT) with the generalized gradient approximation (GGA) combined with D3 dispersion correction has been extensively applied in previous studies, offering a balanced trade-off between accuracy and efficiency in predicting properties such as lattice constants and potential energy surfaces for various perovskites. [29, 30, 31, 32, 33, 34] Nevertheless, such methodology suffers a severe underestimation of the band gaps, which can only be correctly captured with either the hybrid functional that incorporates a portion of exact exchange from Hartree-Fock theory or the many-body perturbation theory such as the GW method. [34, 35, 36, 37, 38] Unfortunately, both of these two methods are computationally much more expensive compared to GGA and even become prohibitive for large systems.

With the development of massive computational resources and advanced machine learning (ML) algorithms, efficient yet accurate descriptions of the electronic structure for large systems have been made possible. Proposed in 2020, the Deep Kohn-Sham (DeePKS) approach introduces a general framework for generating highly accurate self-consistent energy functionals with remarkably reduced computational cost, [39, 40, 41] in which the energy difference between the expensive high-level method and the cheap low-level method is fitted by the neural network. While DeePKS has been comprehensively validated on molecular systems and specific condensed phase systems, [40, 41] its generalizability to multiple or a family of systems, though theoretically feasible, remains unexplored. Indeed, it would be essential to construct a general DeePKS model that can offer highly accurate electronic structure results with low computational cost for systems like halide perovskites, given their undeniable significance in photovoltaic and other fields.

In this work, our goal is to establish a general DeePKS model that can compute the electronic structure properties with the accuracy of the hybrid functional HSE06 and the efficiency comparable to GGA level for various halide perovskites, i.e., ABX3ABX_{3} (A=FA, MA, Cs; B=Sn, Pb; X=Cl, Br, I) as well as the organic-inorganic hybrid alternatives. Based on an iterative training process with 460 configurations over seven types of halide perovskites, we demonstrate that the resulting DeePKS model is capable to reproduce closely matched forces, stress, band gaps, and density of states (DOS) near the Fermi energy as compared to HSE06 for any constitution of the aforementioned ABX3\mathrm{ABX}_3 as well as the hybrid ones. In addition, even though the spin-orbit coupling (SOC) effect has not been taken into consideration during the training process, our computational results show that DeePKS+SOC offers highly consistent band gap and DOS as compared to HSE06+SOC for Pb-containing systems, in which the SOC effect is necessary for a quantitatively correct description of the band gap. [42] The equation of state (EoS) and corresponding band gap predictions with respect to varied sizes of the unit cell are also validated. We believe such DeePKS model can be readily applied for an accurate yet efficient prediction of various properties for the family of halide perovskites and potentially benefit the discovery and design of better photovoltaic based on such materials.

Methodologies and computational details

The DeePKS algorithm for periodic systems has been previously implemented in an open-source DFT package ABACUS in the basis of numerical atomic orbitals (NAO), [43, 44] of which the theoretical details can be found in Ref. 41. Extended labels for DeePKS training process implemented for this work are stress and band gap, which enable the converged DeePKS model to reproduce the stress and DOS near the Fermi energy given by the target method. The stress given by the DeePKS model can be calculated via:

σαβDeePKS[{ϕi∣ω}]=σαβbaseline[{ϕi∣ω}]−∂Eδ[{ϕi∣ω},ω]∂εαβ=σαβbaseline[{ϕi∣ω}]−∑nlmm′∂Eδ∂Dnlmm′I∑iddεαβ[fi⟨ϕi∣αnlmI⟩⟨αnlm′I∣ϕi⟩].(1)\begin{aligned} \sigma_{\alpha\beta}^{\mathrm{DeePKS}}[\{\phi_i\mid\omega\}] &= \sigma_{\alpha\beta}^{\mathrm{baseline}}[\{\phi_i\mid\omega\}] - \frac{\partial E^\delta[\{\phi_i\mid\omega\},\omega]}{\partial\varepsilon_{\alpha\beta}} \\ &= \sigma_{\alpha\beta}^{\mathrm{baseline}}[\{\phi_i\mid\omega\}] - \sum_{nlmm'} \frac{\partial E_\delta}{\partial D_{nlmm'}^I}\sum_i \frac{d}{d\varepsilon_{\alpha\beta}}\left[f_i\langle\phi_i|\alpha_{nlm}^I\rangle\langle\alpha_{nlm'}^I|\phi_i\rangle\right]. \tag*{(1)} \end{aligned}

where fif_i is the occupation number of orbital ϕi\phi_i. Note that the stress label for DeePKS training process is of the energy unit, i.e., without dividing by the volume of the unit cell, and only the upper triangle part of the stress tensor is served as labels. For band gap labels, the band gaps of each kk point, i.e., the energy differences between the highest valance band (HVB) and the lowest conducting band (LCB) at each kk point,

ϵg,k:=ϵLCB,k−ϵHVB,k,(2)\epsilon_{g,k} := \epsilon_{\mathrm{LCB},k} - \epsilon_{\mathrm{HVB},k}, \tag*{(2)}

are served as labels to the training process, aiming to provide an accurate description of the DOS near the Fermi energy. Total energies, forces, stresses, and band gaps of each kk points of the target method, are included in the loss function L(ω)L(\omega), and the optimization problem now becomes:

min⁡ωL(ω),L(ω)=∣Etarget−EDeePKS[{ϕi∣ω}]∣2+λ1∣Ftarget−FDeePKS[{ϕi∣ω}]∣2+λ2∣ϵgtarget−ϵgDeePKS[{ϕi∣ω}]∣2+λ3∣σtarget−σDeePKS[{ϕi∣ω}]∣2,(3)\min_{\omega} L(\omega),\quad L(\omega) = \left|E^{\mathrm{target}} - E^{\mathrm{DeePKS}[\{\phi_i|\omega\}]}\right|^2 + \lambda_1\left|\mathbf{F}^{\mathrm{target}} - \mathbf{F}^{\mathrm{DeePKS}[\{\phi_i|\omega\}]}\right|^2 + \lambda_2\left|\epsilon_g^{\mathrm{target}} - \epsilon_g^{\mathrm{DeePKS}[\{\phi_i|\omega\}]}\right|^2 + \lambda_3\left|\boldsymbol{\sigma}^{\mathrm{target}} - \boldsymbol{\sigma}^{\mathrm{DeePKS}[\{\phi_i|\omega\}]}\right|^2, \tag*{(3)}

where λ1\lambda_1, λ2\lambda_2, and λ3\lambda_3 correspond to the weighting factor of force, band gap, and stress in the loss function, are are set to be 1.0, 0.1, and 1.0 in this work, respectively.

It should be noted that the goal of this work is to contrive a general DeePKS model for various halide perovskites instead of a single system. This poses challenges to the energy fitting process by the neural network owing to the fact that the absolute energy difference between the base and the target method contributed from each element of multiple systems might be significantly variant. To circumvent this obstacle, we apply the method of linear least squares to calculate the element-specific contribution to the absolute energy difference between the base and the target method. Specifically, the sum of squares to be minimized in our case is

S=∑s[ΔEs−∑IϵINI,s]2(4)S = \sum_s \left[\Delta E_s - \sum_I \epsilon_I N_{I,s}\right]^2 \tag*{(4)}

where ΔEs\Delta E_s is the energy difference between the base and target method of system ss, and NI,sN_{I,s} is the number of atoms of element II contained in system ss. The minimization of SS gives ϵI\epsilon_I, i.e., the contribution of element II to the energy difference between the base and target method, and the absolute energy difference resulted from ϵI\epsilon_I for each system ss, i.e., ∑IϵINI,s\sum_I \epsilon_I N_{I,s}, is then deducted from the loss function.

The training sets of the DeePKS model consist of FAPbI3\mathrm{FAPbI_3}, MAPbI3\mathrm{MAPbI_3}, CsPbI3\mathrm{CsPbI_3}, MAPbBr3\mathrm{MAPbBr_3}, MASnCl3\mathrm{MASnCl_3}, CsPbCl3\mathrm{CsPbCl_3}, and CsSnBr3\mathrm{CsSnBr_3}, and the number of configurations of each system is given in Table S2. Configurations of the aforementioned XPbI3\mathrm{XPbI_3} family are randomly (subject to uniform distribution) picked from previously reported DeePMD training data, [45] while those of other systems are randomly picked from corresponding DP-GEN jobs. The test sets consist of four systems that are different from the training sets, i.e., MAPbCl3\mathrm{MAPbCl_3}, MASnBr3\mathrm{MASnBr_3}, CsSnI3\mathrm{CsSnI_3}, and FA0.125Cs0.875PbI3\mathrm{FA_{0.125}Cs_{0.875}PbI_3}. Configurations of FA0.125Cs0.875PbI3\mathrm{FA_{0.125}Cs_{0.875}PbI_3} are randomly picked from previously reported DeePMD training data, [45] while those of the other three are randomly picked from corresponding DP-GEN jobs.

Table S2: Training sets mean absolute errors (MAE) of energy, force, stress, and band gap given by DeePKS and PBE with respect to the HSE06 functional. DeePKS and PBE calculations are performed in ABACUS with numerical atomic orbital (NAO) basis set and SG15 ONCV pseudopotentials while HSE06 calculations are performed in VASP with plane-wave basis and PAW method.

MAE:Energy (meV/atom)Force (eV/A˚\mathring{\mathrm{A}})Stress (eV)Band gap (eV)
systems# of atoms# of framesDeePKSPBEDeePKSPBEDeePKSPBEDeePKSPBE
FAPbI3_348480.68051.62300.05510.11040.22973.11500.03550.7207
FAPbI3_324411.19042.25870.06510.11970.26771.60960.06720.8276
MAPbI3_348780.72001.31880.04600.09980.28092.26010.03390.7144
MAPbI3_396130.98511.01100.05570.10560.78974.53850.10490.7629
CsPbI3_31060.15901.16910.00620.00360.02910.40930.03690.6716
CsPbI3_320612.19793.38970.02570.04980.15360.87120.10090.7859
CsPbI3_34081.17912.27610.01290.02240.15691.64450.03870.6062
CsPbCl3_35452.16494.71580.01750.04610.06880.27600.05051.0011
CsSnBr3_35442.125811.65580.01620.06440.04770.30850.03940.8448
MAPbBr3_312771.80803.18120.04340.10110.09530.69940.05540.9077
MASnCl3_312391.69483.65680.04530.11470.09970.76240.06450.8668
Overall MAE:1.50253.54600.03900.08640.17701.36770.05640.8193

Table S2. Training sets mean absolute errors (MAE) of energy, force, stress, and band gap given by DeePKS and PBE with respect to the HSE06 functional. DeePKS and PBE calculations are performed in ABACUS with numerical atomic orbital (NAO) basis set and SG15 ONCV pseudopotentials while HSE06 calculations are performed in VASP with plane-wave basis and PAW method.

The target functional applied in this work for label generation is the hybrid GGA functional HSE06. Self-consistent field (SCF) jobs for all configurations in the training test sets are carried out with the HSE06 functional and projector-augmented wave (PAW) method in the Vienna Ab initio Simulation Package (VASP). [46, 47, 48] The energy cutoff is set to be 500 eV, and the allowed spacing between kk points (“kspacing”) is set to be 0.1 Bohr−1\mathrm{Bohr}^{-1}. Such energy cutoff has been previously validated in Refs. 45 and 49 for XPbI3\mathrm{XPbI_3} (X=Cs, FA, MA) systems, and the kspacing value for the HSE06 jobs is slightly larger compared to that applied in Ref. 45 for PBE calculation (i.e., 0.16 A˚−1\mathrm{\AA}^{-1}) due to the high computational cost of the HSE06 functional. We will demonstrate in next section that the DeePKS model trained with a relatively sparse kk-point label can successfully recover the energy, force, and DOS of the target method with a much denser kk-point grid. During the DeePKS iterative training process, the single-point SCF calculations are performed in ABACUS[43, 44] with double-zeta polarized (DZP) NAO basis, [50, 51] 100 Ry energy cutoff, 0.1 Bohr−1\mathrm{Bohr}^{-1} kspacing value, and the SG15 Optimized Norm-Conserving Vanderbilt (ONCV) [52] pseudopotentials. The aforementioned ϵI\epsilon_{I} of all elements involved in these systems is listed in Table S1. More details of the DeePKS training process can be found in the Supporting Information.

element IIϵI\epsilon_I
H4.3297
C2.5416
N13.4613
Cl18.9693
Br17.4021
Sn27.3945
I105.2054
Cs44.6507
Pb19.7978

Table S1. The contribution of each element to the absolute energy difference between the base and target method for all involved halide perovskites. Numbers are shown in the unit of a.u. Note that such large absolute energy difference is result from the fact that the SCF calculation of target functional HSE06 is performed with the PAW method while that of the base method PBE is performed with the SG15 ONCV pseudopotentials.

Results and discussion

Accuracy crossing-checking of the DeePKS model

Mean absolute errors (MAE) of energies, forces, band gaps for multi-kk points, and stresses of the training set given by the converged DeePKS model are listed in Table S2, while those for the test set are listed in Table S3. Note that the absolute energies shift significantly between the HSE06 calculation and the DeePKS/PBE calculation due to the fact that PAW is employed for HSE06 jobs while ONCV pseudopotentials are applied in DeePKS/PBE. We exclude the influence of such “constant” shift by defining the mean absolute relative error (MARE) of the energy as

MAE:Energy (meV/atom)Force (eV/Å)Stress (eV)Band gap (eV)
systems# of atoms# of framesDeePKSPBEDeePKSPBEDeePKSPBEDeePKSPBE
CsSnI3_35814.04435.74400.03570.06090.12870.24420.05040.6690
MAPbCl3_3121311.84872.75680.05050.09940.12010.69130.06200.9848
MASnBr3_3121501.63533.82580.04520.10830.11450.72690.04880.7800
FA0.125_{0.125}Cs0.875_{0.875}PbI3_347110.03201.49940.03360.05220.24632.34150.02230.6965
Overall MAE:2.18613.79830.04470.09320.12340.65720.05300.8254

Table S3. Test sets mean absolute errors (MAE) of energy, force, stress, and band gap given by DeePKS and PBE with respect to the HSE06 functional. DeePKS and PBE calculations are performed in ABACUS with numerical atomic orbital (NAO) basis set and SG15 ONCV pseudopotentials while HSE06 calculations are performed in VASP with plane-wave basis and PAW method.

EMARE=1N∑i=1N∣EiHSE06−[EiDeePKS/PBE−E‾DeePKS/PBE+E‾HSE06]∣(5)E_{\mathrm{MARE}} = \frac{1}{N} \sum_{i=1}^{N} \left| E_{i}^{\mathrm{HSE06}} - \left[ E_{i}^{\mathrm{DeePKS/PBE}} - \overline{E}^{\mathrm{DeePKS/PBE}} + \overline{E}^{\mathrm{HSE06}} \right] \right| \tag*{(5)}

It can be seen that compared to PBE results, the MAE/MARE of four labels given by DeePKS is significantly reduced, especially that of band gaps and stresses, and similar accuracy is achieved via DeePKS for the test sets. As shown in Table S3, for four tested systems, pronounced decrease of the MAE is observed for the band gap and stress (0.053 eV and 0.123 eV, respectively) compared to the PBE results (0.825 eV and 0.657 eV, respectively), while the MAE/MARE for energy and force (2.2 meV/atom, 0.045 eV/Å, respectively) is also decreased by around a factor of two.

It should be noted that for halide perovskite systems, PBE has been validated as a reliable functional to describe the potential energy surfaces. [29, 30, 31, 32, 33, 34, 45] The key problem encountered by PBE is indeed the prediction of the band gap for these systems, which is essential for photovoltaic properties. HSE06, albeit computationally demanding, is able to offer significantly improved description of the band gap for perovskites. The successful recovery of the HSE06 band gap of DeePKS evinces its capability to offer a notably more accurate description of the band gap as compared to PBE. The reason that the stress MAE of DeePKS remains relatively large (>0.1> 0.1 eV) is attributed to the severe deviation between the PBE predicted stress and the HSE06 predicted one (with an overall MAE larger than 1.3 eV for the training set). That being said, the MAE of the stress predicted by the DeePKS model decreases remarkably as compared to that given by PBE, indicating the strength of the training process.

Property prediction on various halide perovskite systems

We extensively test our DeePKS model on 18 different types of organic or inorganic perovskites (mostly cubic phase unless otherwise specified), and the hybrid organic-inorganic perovskite FAaCs1.0−aPbI3\mathrm{FA}_{a}\mathrm{Cs}_{1.0-a}\mathrm{PbI}_{3} with varied aa values. Cell relaxation is performed for all non-hybrid perovskites, while ionic relaxation is performed for hybrid perovskites. All relaxation calculations are carried out with PBE functional and D3.0 van der Waals (vdW) dispersion correction in ABACUS. DZP NAO basis set with ONCV pseudopotentials is applied, and the energy cutoff is set to be 100 Ry. For all cubic phase systems, the Brillouin zone is sampled with a gamma-centered 9×9×99 \times9 \times9 kk-point grid, while for all other systems, a kk-spacing value is set to be 0.07 Bohr−10.07\ \mathrm{Bohr}^{-1}. DeePKS/PBE and HSE06 single-point SCF calculations are performed on these optimized structures, with the same basis sets, energy cutoff as those applied in the iterative training and label generation processes, while the kk-point settings are consistent with the aforementioned relaxation calculations. Detailed input parameters of HSE06 (carried out in VASP) and DeePKS/PBE (carried out in ABACUS) are provided in the Supporting Information.

Total CPU times of the SCF calculations performed via HSE06, DeePKS, and PBE for these tested perovskites are given in Table S4. It can been seen that the CPU time of DeePKS is of the same order of magnitude (though twice to three times as long) as compared to that of PBE for most tested systems, while it is around two orders of magnitude faster as compared to that of HSE06 based on our input parameter and parallelization settings. In addition to the omission of constructing the Hartree-Fock exact exchange part, such huge savings in computational time of DeePKS with respect to HSE06 also stem from the higher efficiency of NAO basis as compared to the plane-wave basis for large systems.

systemHSE06DeePKSPBE
CsPbBr3_3111.3901269.4545216875.736
CsPbCl3_3143.6753258.2207214457.284
CsPbI3_3195.4423394.7154513227.812
β\beta-CsPbI3_3255.41758107.91931192627.938
γ\gamma-CsPbI3_3331.70495182.74886250612.344
δ\delta-CsPbI3_31473.79506237.27452300858.344
CsSnBr3_397.7857857.6327415659.790
CsSnCl3_3155.8692358.7123614514.202
CsSnI3_3193.9680892.2665119305.125
FAPbBr3_3233.16509134.81837176083.594
FAPbCl3_3334.87374134.87047155415.516
FAPbI3_3310.39089174.69117345250.188
FASnBr3_3217.74080111.85347163583.922
FASnCl3_3269.62527128.92052172128.469
FASnI3_3316.82600156.07270189681.109
MAPbBr3_3254.16727119.34650136092.531
MAPbCl3_3323.73752125.7440497563.695
MAPbI3_3379.85375154.58523220930.016
MASnBr3_3288.36057105.26187115660.477
MASnCl3_3282.27638114.72428114202.844
MASnI3_3451.40572134.93733113742.766
FA0.125_{0.125}Cs0.875_{0.875}PbI3_31072.25264375.18319774865.938
FA0.25_{0.25}Cs0.75_{0.75}PbI3_3873.54869463.76868632852.938
FA0.375_{0.375}Cs0.625_{0.625}PbI3_3736.42967653.44151689350.938
FA0.5_{0.5}Cs0.5_{0.5}PbI3_31107.17502616.79290822730.000
FA0.625_{0.625}Cs0.375_{0.375}PbI3_32049.49370555.79434805206.375
FA0.75_{0.75}Cs0.25_{0.25}PbI3_31251.96995587.64888791275.375
FA0.875_{0.875}Cs0.125_{0.125}PbI3_31124.27960611.86466775134.500

Table S4. Total CPU times (in the unit of second) of the SCF calculation performed via HSE06, DeePKS, and PBE for all tested perovskites. All tested non-hybrid perovskites are cubic phase except for those indicated by Greek letters. The HSE06 calculations are carried out in VASP version 5.4 with plane-wave basis and PAW method on 16-core 256-mem CPU processor, starting from the converged PBE wavefunction. DeePKS and PBE calculations are carried out in ABACUS version 3.2 with DZP NAO basis set and SG15 ONCV pseudopotentials on 16-core 64-mem CPU processor. All calculations are ran with 16-core MPI parallelization. Input parameters of HSE06 (in VASP), DeePKS (in ABACUS), and PBE (in ABACUS) are provided below.

Force and stress

While the absolute energies between DeePKS and HSE06 are not directly comparable owing to the discrepancy between SG15 ONCV pseudopotential applied in DeePKS and PAW method applied in HSE06, we first investigate the accuracy of our DeePKS model in calculating the force of selected systems. Forces are tested on all systems but cubic phase inorganic perovskites, of which the atomic forces remain zero due to their ideal cubic symmetry (Pm3ˉmPm\bar{3}m space group). The MAE of the force predicted by DeePKS and PBE as compared to the HSE06 result for each tested system is listed in Table 1. Forces along three directions of hybird perovskite systems given by DeePKS and PBE are plotted with respect to the HSE06 values in Figure 1. It can be seen that while the PBE forces are already close to the HSE06 results, forces calculated by DeePKS are notably more concentrated on the diagonal. The overall MAE of the force given by DeePKS for all tested systems is around 0.030 eV/A˚\mathrm{eV}/\mathrm{\AA}, while that given by PBE is around 0.072 eV/A˚\mathrm{eV}/\mathrm{\AA}, indicating an improved accuracy in the description of the potential energy surface with a sharply higher efficiency as mentioned above.

systemDeePKSPBE
β\beta-CsPb3_30.00200.0054
γ\gamma-CsPb3_30.00250.0053
δ\delta-CsPbI3_30.00860.0268
MAPbCl3_30.02450.0765
MAPbBr3_30.02110.0779
MAPbI3_30.02100.0800
MASnCl3_30.02650.0888
MASnBr3_30.02190.0833
MASnI3_30.03170.0895
FAPbCl3_30.05350.0897
FAPbBr3_30.03810.0887
FAPbI3_30.04080.0927
FASnCl3_30.04530.0897
FASnBr3_30.03700.1053
FASnI3_30.04470.0942
FA0.125_{0.125}Cs0.875_{0.875}PbI3_30.02910.0445
FA0.25_{0.25}Cs0.75_{0.75}PbI3_30.02810.0540
FA0.375_{0.375}Cs0.625_{0.625}PbI3_30.03130.0620
FA0.5_{0.5}Cs0.5_{0.5}PbI3_30.03470.0782
FA0.625_{0.625}Cs0.375_{0.375}PbI3_30.03830.0817
FA0.75_{0.75}Cs0.25_{0.25}PbI3_30.03720.0909
FA0.875_{0.875}Cs0.125_{0.125}PbI3_30.04340.0905
Overall MAE0.03010.0725

Table 1. Force MAE for all tested halide perovskites given by DeePKS and PBE with respect to the HSE06 results. All tested non-hybrid organic perovskites are cubic phase, while phases of tested CsPbI3_3 are indicated by Greek letters. Numbers are shown in the unit of eV/A˚\mathring{\mathrm{A}}.

Forces along three directions of hybrid perovskite systems

Figure 1. Forces along three directions of hybrid perovskite systems given by DeePKS and PBE with respect to the HSE06 results.

Compared to the force label, the target stress given by HSE06 is more difficult to be reproduced by DeePKS, owing to the notable deviation between the PBE and HSE06 stresses as mentioned above. Here, we calculate the stress for those hybrid perovskites via DeePKS and PBE, and results are compared with those given by HSE06 as shown in Figure 2 (while explicit numbers can ben found in Table S5). It can be seen that the diagonal element of the stress calculated by PBE is significantly larger than the HSE06 counterparts, while the off-diagonal elements are severely underestimated. The stress calculated by DeePKS is qualitatively in line with the HSE06 result, with the remarkably reduced MAE for both diagonal and off-diagonal elements of the stress, i.e., 0.394 and 0.168 KBar, respectively (as compared to 5.435 and 1.556 KBar given by PBE).

Stresses of hybrid perovskite systems

Figure 2. Stresses (diagonal and off-diagonal elements) of hybrid perovskite systems given by DeePKS and PBE with respect to the HSE06 results.

Band gaps and density of states

Band gaps predicted by DeePKS for all tested systems are compared with the results of HSE06 in Table 2. Values computed via PBE functional are also listed for reference. As shown in Table 2, the DeePKS predicted band gaps for all tested perovskites (including the hybrid systems and the non-perovskite phase, i.e., δ\delta-CsPbI3_3) are in excellent agreement with the HSE06 results, with an overall MAE 0.0293 eV. The well-known underestimation of the band gap values suffered by PBE functional (indicated by the large overall MAE, i.e., 0.5142 eV) is completely surmounted by the DeePKS model. We visualize the band gaps results of DeePKS and PBE with respect to the HSE06 benchmarks in Figure 3. While the PBE predicted values lie significantly lower, the DeePKS results are concentrated on the diagonal line, demonstrating a remarkably improved accuracy for the band gap description obtained by the DeePKS model.

Table 2: Band gaps for all tested halide perovskites given by HSE06, DeePKS, and PBE. Absolute errors of the DeePKS and PBE results with respect to HSE06 results are listed, as well as the overall mean absolute error (MAE) for all tested systems. All tested non-hybrid perovskites are cubic phase except for those indicated by Greek letters. Numbers are shown in the unit of eV.

systemHSE06DeePKSDeePKS Abs. Err.PBEPBE Abs. Err.
CsPbCl3\mathrm{CsPbCl}_{3}3.14233.11350.02882.50790.6344
CsPbBr3\mathrm{CsPbBr}_{3}2.61742.65510.03772.05020.5672
CsPbI3\mathrm{CsPbI}_{3}2.22542.23310.00771.79060.4348
β-CsPbI3\beta\text{-}\mathrm{CsPbI}_{3}1.96501.97030.00531.47820.4868
γ-CsPbI3\gamma\text{-}\mathrm{CsPbI}_{3}2.18542.18610.00071.68730.4981
δ-CsPbI3\delta\text{-}\mathrm{CsPbI}_{3}3.33533.36030.02502.54540.7899
CsSnCl3\mathrm{CsSnCl}_{3}1.87271.72260.15011.42370.4490
CsSnBr3\mathrm{CsSnBr}_{3}1.45841.37500.08341.08560.3728
CsSnI3\mathrm{CsSnI}_{3}1.22881.19790.03090.97300.2558
MAPbCl3\mathrm{MAPbCl}_{3}3.29723.33180.03462.60550.6917
MAPbBr3\mathrm{MAPbBr}_{3}2.72112.74390.02282.08610.6350
MAPbI3\mathrm{MAPbI}_{3}2.17892.19810.01921.69660.4823
MASnCl3\mathrm{MASnCl}_{3}2.15582.12640.02941.62590.5299
MASnBr3\mathrm{MASnBr}_{3}1.60721.55570.05151.15420.4530
MASnI3\mathrm{MASnI}_{3}1.18581.14430.04150.88970.2961
FAPbCl3\mathrm{FAPbCl}_{3}3.29383.33190.03812.60240.6914
FAPbBr3\mathrm{FAPbBr}_{3}2.73252.74430.01182.10360.6289
FAPbI3\mathrm{FAPbI}_{3}2.23742.22840.00901.75210.4853
FASnCl3\mathrm{FASnCl}_{3}3.02263.04120.01862.36240.6602
FASnBr3\mathrm{FASnBr}_{3}1.86911.82730.04181.37950.4896
FASnI3\mathrm{FASnI}_{3}1.26221.25020.01200.95360.3086
FA0.125Cs0.875PbI3\mathrm{FA}_{0.125}\mathrm{Cs}_{0.875}\mathrm{PbI}_{3}2.39682.38050.01631.85560.5412
FA0.25Cs0.75PbI3\mathrm{FA}_{0.25}\mathrm{Cs}_{0.75}\mathrm{PbI}_{3}2.39292.39520.00231.84990.5430
FA0.375Cs0.625PbI3\mathrm{FA}_{0.375}\mathrm{Cs}_{0.625}\mathrm{PbI}_{3}2.00711.98750.01961.53130.4758
FA0.5Cs0.5PbI3\mathrm{FA}_{0.5}\mathrm{Cs}_{0.5}\mathrm{PbI}_{3}2.35472.34280.01191.80720.5475
FA0.625Cs0.375PbI3\mathrm{FA}_{0.625}\mathrm{Cs}_{0.375}\mathrm{PbI}_{3}2.02222.00610.01611.54990.4723
FA0.75Cs0.25PbI3\mathrm{FA}_{0.75}\mathrm{Cs}_{0.25}\mathrm{PbI}_{3}2.01451.99970.01481.54110.4734
FA0.875Cs0.125PbI3\mathrm{FA}_{0.875}\mathrm{Cs}_{0.125}\mathrm{PbI}_{3}2.08082.04290.03791.57600.5048
Overall MAE0.02930.5142

Table 2. Band gaps for all tested halide perovskites given by HSE06, DeePKS, and PBE. Absolute errors of the DeePKS and PBE results with respect to HSE06 results are listed, as well as the overall mean absolute error (MAE) for all tested systems. All tested non-hybrid perovskites are cubic phase except for those indicated by Greek letters. Numbers are shown in the unit of eV.

Band gaps predicted by DeePKS and PBE with respect to HSE06 results for all tested perovskites

Figure 3. Band gaps predicted by DeePKS and PBE with respect to the HSE06 results for all tested perovskites. All tested non-hybrid perovskites are cubic phase except for those indicated by Greek letters.

Next, we examine the DOS of various halide perovskites described by DeePKS. Even though the DOS does not directly serve as one of the training labels to be fitted by the neural network, we can see from Figure 4 to Figure 7 that for all tested halide perovskites, the overall DOSs near the Fermi energy predicted by DeePKS precisely match with those given by HSE06, while those of PBE exhibit significantly smaller band gaps. Such accurate description

Density of states for cubic phase inorganic halide perovskites

Figure 4. Density of states (DOS) for cubic phase inorganic halide perovskites, i.e., CsBX3\mathrm{CsBX}_{3} (B=Sn, Pb; X=Cl, Br, I).

DOS for hybrid organic-inorganic halide perovskites

Figure 7. DOS for hybrid organic-inorganic halide perovskites, i.e., FAaCs1−aPbI3\mathrm{FA}_{a}\mathrm{Cs}_{1-a}\mathrm{PbI}_{3}.

of the DOS given by DeePKS lies in the fact that bands predicted by PBE near the Fermi energy possess an overall reasonable structure (as speculated by the similar shape of the DOS predicted by HSE06 and PBE) but with a much narrower gap. By fitting the band gap for each kk point, such deficiency of PBE can be largely mitigated.

DOS for cubic phase organic halide perovskites

Figure 5. DOS for cubic phase organic halide perovskites, i.e., MABX3\mathrm{MABX}_{3} (B=Sn, Pb; X=Cl, Br, I).

DOS for non-cubic phase CsPbI3

Figure 6. DOS for non-cubic phase CsPbI3\mathrm{CsPbI}_{3}.

Spin-orbit coupling effect on Pb-containing systems

It is well-known that the SOC effect plays an important role to the magnitude of the band gaps in perovskite systems,[36, 53] especially for Pb-containing perovskite.[42, 54] The inclusion of the SOC effect may significantly narrow the band gap for APbX3\mathrm{APbX}_{3} and thus becomes essential for accurate property prediction in practice.[42, 55] Taking CsPbX3\mathrm{CsPbX}_{3} as examples, we investigate the performance of DeePKS model with the inclusion of the SOC effect. To do that, the full relativistic (FR) version of the SG15 ONCV pseudopotentials[56] is applied for all involved elements. As shown in Table 3 and Figure 8, while the SOC effect has not been taken into account during our iterative DeePKS training process, the DeePKS+SOC offers closely matched band gap values as well as the DOS for three tested systems as compared to HSE06+SOC. Indeed, the relativistic effect is mainly handled in the pseudopotential part, either via the PAW method (for HSE06 functional performed in VASP) or via the norm-conserving pseudopotentials (for DeePKS model performed in ABACUS). Even though the scalar relativistic approximation is applied in PAW for the valance electron [57] while the FR effect is employed in ONCV pseudopotentials, the influence of such discrepancy is subtle for our tested systems, as demonstrated by the almost identical DOS given by PBE+SOC calculations performed in VASP with PAW and that in ABACUS with NAO and the FR version of ONCV pseudopotentials (Figure S1). Therefore, the HSE06+SOC results can be successfully reproduced by DeePKS+SOC as long as the DeePKS model provides a close enough description of the electronic structure as compared to the HSE06 functional, which has been extensively evinced in former sections.

systemHSE06DeePKSDeePKS Abs. Err.PBEPBE Abs. Err.
CsPbCl3\mathrm{CsPbCl}_{3}2.11872.07770.04101.53820.5805
CsPbBr3\mathrm{CsPbBr}_{3}1.62981.62950.00031.12610.5037
CsPbI3\mathrm{CsPbI}_{3}1.23331.24560.01230.93530.2980

Table 3. Band gaps for cubic phase CsPbX3\mathrm{CsPbX}_{3} given by HSE06, DeePKS, and PBE with the SOC effect taken into account. Absolute errors of the DeePKS and PBE results with respect to HSE06 results are listed. Numbers are shown in the unit of eV.

Density of states for CsPbCl3, CsPbBr3, and CsPbI3

Figure 8. DOS for Pb-containing cubic phase inorganic halide perovskites, i.e., CsPbX3_3 (X=ClX=\mathrm{Cl}, Br, I) with the SOC effect taken into account.

DOS for Pb-containing cubic phase inorganic halide perovskites

Figure S1. DOS for Pb-containing cubic phase inorganic halide perovskites, i.e., CsPbX3\mathrm{CsPbX}_{3} (X=Cl\mathrm{X}=\mathrm{Cl}, Br\mathrm{Br}, I\mathrm{I}) given by PBE with PAW and PBE with SG15 ONCV pseudopotential, with the SOC effect taken into account.

Equation of states for cubic inorganic perovskites

Finally, we examine the capability of our DeePKS model in predicting mechanical properties by performing the EoS calculation for three cubic phase inorganic perovskites, i.e., CsPbCl3_3, CsPbBr3_3, and CsPbI3_3. A cell relaxation is first performed with DeePKS model and PBE functional (with the same input settings as those applied in former single-point calculation). Resulting optimized volumes are then scaled from 0.9 to 1.1 with a step of 0.01 to compute the total energy at different volumes. Due to the extremely high computational cost of HSE06 functional, we skip the relaxation step by applying the DeePKS-optimized structure at each volume and perform a single point SCF calculation via HSE06 functional. Since the absolute energies given by ONCV pseudopotentials and PAW method are not directly comparable, we plot the relative energies with respect to the lowest energy for each functional. As shown in the upper panel of Figure 9, the DeePKS computed EoS for these systems is in line with the HSE06 counterpart, while PBE predicts a larger volume of the equilibrium structure. Change of the band gap values with respect to the size of the unit cell is also explored as shown in the lower panel of Figure 9. Closely matched band gaps are obtained via DeePKS and HSE06, indicating that the band gap tuning by adjusting the pressure can be accurately predicted via DeePKS model.

Equation of states and band gap values for CsPbCl3, CsPbBr3, and CsPbI3

Figure 9. Equation of states (EoS) (upper panel) and band gap values (lower panel) with respect to the size of the unit cell of CsPbBr₃, CsPbCl₃, CsPbBr₃, and CsPbI₃ computed with HSE06, DeePKS, and PBE.

Conclusion and outlook

In this work, with 460 configurations over seven types of halide perovskites, we have established a general DeePKS model that is applicable to a variety of combinations, i.e., ABX3_3 (A=FA, MA, Cs; B=Sn, Pb; X=Cl, Br, I) as well as the organic-inorganic hybrid perovskites, with the accuracy of HSE06 in predicting electronic structure properties such as force, stress, band gap, and DOS, and the efficiency comparable to that of PBE. For all tested systems, especially hybrid perovskites with large unit cells, DeePKS offers orders of magnitude savings in computational cost as compared to HSE06. While the MAEs of the energy and force predicted by DeePKS with respect to HSE06 are reduced by a factor of two as compared to those of PBE, the MAEs for the stress and band gap are reduced at least one order of magnitude. Specifically, the MAE for the band gap given by DeePKS over 28 tested systems is less than 0.03 eV, demonstrating a promising alleviation of the underestimation issue suffered by PBE, of which the MAE is larger than 0.5 eV. The DOS predicted by DeePKS also closely matches with the target method in the vicinity of the Fermi energy.

For Pb-containing perovskites, the SOC effect needs to be taken into account so as to offer a quantitatively correct description for the band gap and DOS. We have shown that DeePKS+SOC is able to offer highly consistent band gap and DOS as compared to HSE06+SOC for Pb-containing systems even though the SOC effect is not considered during the iterative training process. In addition, mechanical properties given by DeePKS have also been validated for inorganic halide perovskites by performing the EoS calculation, with highly consistent band gaps with respect to the size of the unit cell as compared to HSE06. In summary, with the closely-matched electronic structure description with respect to the HSE06 functional and remarkably reduced computational cost, our DeePKS model can be extensively applied to precisely predict a variety of properties and/or generate highly accurate labels for machine learning based potential energy models for the family of halide perovskites, facilitating the discovery and design of better photovoltaics based on such systems.

Acknowledgement

The authors thank Dr. Han Wang and Dr. Mohan Chen for their helpful conversations and computational guidance. The authors gratefully acknowledge the funding support from the AI for Science Institute, Beijing (AISI) and DP Technology Corporation. The computing resource of this work was provided by the Bohrium Cloud Platform (https://bohrium.dp.tech), which is supported by DP Technology. The work of Y.C. was supported by the Computational Chemical Sciences Center: Chemistry in Solution and at Interfaces (CSI) funded by DOE Award DE-SC0019394 and a gift from iFlytek to Princeton University.

Supporting Information Available

Details of the SCF calculation (including input parameters and run time statistics) and the DeePKS iterative training procedure. MAEs of various electronic structure properties for the training and test sets given by the DeePKS model.

Supporting Information for “DeePKS Model for Halide Perovskites with the Accuracy of Hybrid Functional”

Qi Ou; Ping Tuo; Wenfei Li; Xiaoxu Wang; Yixiao Chen; Linfeng Zhang

Input parameters of HSE06, DeePKS, and PBE SCF calculations.

HSE06 jobs in VASP

SYSTEM = perovskite
ISTART = 1 //starts from a converged PBE wavefunction
ICHARG=0
ENCUT = 500 // in the unit of eV
PREC = Accurate
ISMEAR = 0 // Gaussian smearing
SIGMA = 0.02 // in the unit of eV
EDIFF = 1E-05
GGA = PE
LREAL= Auto
LORBIT = 11
ISIF = 2
LHFCALC = .TRUE.
ALGO = Damped
TIME = 0.4
HFSCREEN = 0.2
PRECFOCK = Fast
AEXX = 0.25
NELM = 200

DeePKS/PBE jobs in ABACUS

INPUT_PARAMETERS

calculation scf

ecutwfc 100.000000    // energy cutoff in the unit of Rydberg

scf_thr 1.000000e-07

scf_nmax 100

basis_type lcao

dft_functional pbe

gamma_only 0

mixing_type pulay

mixing_beta 0.400000

out_dos 1

nspin 1

smearing_method gaussian

smearing_sigma 0.00147    //in the unit of Rydberg

cal_force 1

cal_stress 1

out_bandgap 1

deepks_scf 1    // 1 for DeePKS; 0 for PBE

deepks_model ./model.ptg

Table S5: Stress elements predicted by HSE06, DeePKS, and PBE for hybrid perovskites. Absolute errors of the DeePKS and PBE results with respect to HSE06 results are listed. Numbers are shown in the unit of KBar.

systemstress elementHSE06DeePKSDeePKSPBEPBE
Abs. Err.Abs. Err.
σxx\sigma_{xx}1.96992.23480.26495.40213.4322
σyy\sigma_{yy}2.08912.89390.80494.72892.6398
FA0.125_{0.125}Cs0.875_{0.875}PbI3_3σzz\sigma_{zz}1.74962.41780.66825.04063.2909
σxy\sigma_{xy}-0.5075-0.63430.12680.08480.5923
σxz\sigma_{xz}-0.1183-0.27820.1598-0.00870.1096
σyz\sigma_{yz}0.29390.36030.0664-0.07310.3670
σxx\sigma_{xx}0.49480.86360.36885.97665.4818
σyy\sigma_{yy}2.83903.36340.52446.59143.7525
FA0.25_{0.25}Cs0.75_{0.75}PbI3_3σzz\sigma_{zz}1.89122.35460.46355.64413.7529
σxy\sigma_{xy}0.35370.34570.00800.13040.2233
σxz\sigma_{xz}0.12250.02970.0928-0.05240.1750
σyz\sigma_{yz}0.08570.10560.0199-0.11870.2044
σxx\sigma_{xx}1.98722.21200.22486.32854.3413
σyy\sigma_{yy}2.69013.13810.44816.89354.2034
FA0.375_{0.375}Cs0.625_{0.625}PbI3_3σzz\sigma_{zz}2.46842.92130.45296.79894.3305
σxy\sigma_{xy}1.41721.21930.1979-0.11191.5292
σxz\sigma_{xz}1.20611.09100.1151-0.14551.3517
σyz\sigma_{yz}1.50971.69050.18090.57570.9340

Table S5. Table S5: Stress elements predicted by HSE06, DeePKS, and PBE for hybrid perovskites. Absolute errors of the DeePKS and PBE results with respect to HSE06 results are listed. Numbers are shown in the unit of KBar.

systemstress elementHSE06DeePKSDeePKS Abs. Err.PBEPBE Abs. Err.
FA0.5_{0.5}Cs0.5_{0.5}PbI3_3σxx\sigma_{xx}0.90481.10910.20437.35496.4501
FA0.5_{0.5}Cs0.5_{0.5}PbI3_3σyy\sigma_{yy}0.37620.95930.58316.94356.5673
FA0.5_{0.5}Cs0.5_{0.5}PbI3_3σzz\sigma_{zz}2.49292.90880.41597.41924.9262
FA0.5_{0.5}Cs0.5_{0.5}PbI3_3σxy\sigma_{xy}1.24021.11410.12610.12681.1135
FA0.5_{0.5}Cs0.5_{0.5}PbI3_3σxz\sigma_{xz}2.17861.91820.26040.11992.0587
FA0.5_{0.5}Cs0.5_{0.5}PbI3_3σyz\sigma_{yz}0.73530.81470.07930.03400.7013
FA0.625_{0.625}Cs0.375_{0.375}PbI3_3σxx\sigma_{xx}1.07421.34530.27117.12706.0528
FA0.625_{0.625}Cs0.375_{0.375}PbI3_3σyy\sigma_{yy}1.59542.02330.42787.46095.8655
FA0.625_{0.625}Cs0.375_{0.375}PbI3_3σzz\sigma_{zz}1.37221.69500.32287.34465.9724
FA0.625_{0.625}Cs0.375_{0.375}PbI3_3σxy\sigma_{xy}2.57972.34060.23900.10392.4758
FA0.625_{0.625}Cs0.375_{0.375}PbI3_3σxz\sigma_{xz}2.42072.26810.15270.02412.3966
FA0.625_{0.625}Cs0.375_{0.375}PbI3_3σyz\sigma_{yz}2.08502.30390.21890.54641.5386
FA0.75_{0.75}Cs0.25_{0.25}PbI3_3σxx\sigma_{xx}0.67250.95210.27967.52906.8565
FA0.75_{0.75}Cs0.25_{0.25}PbI3_3σyy\sigma_{yy}1.03901.34730.30837.66366.6246
FA0.75_{0.75}Cs0.25_{0.25}PbI3_3σzz\sigma_{zz}0.68700.88340.19647.58156.8944
FA0.75_{0.75}Cs0.25_{0.25}PbI3_3σxy\sigma_{xy}3.18802.92200.26600.16303.0251
FA0.75_{0.75}Cs0.25_{0.25}PbI3_3σxz\sigma_{xz}3.07592.93920.13670.14752.9284
FA0.75_{0.75}Cs0.25_{0.25}PbI3_3σyz\sigma_{yz}2.28512.61750.33240.49011.7950
FA0.875_{0.875}Cs0.125_{0.125}PbI3_3σxx\sigma_{xx}-0.00640.12780.13427.79517.8014
FA0.875_{0.875}Cs0.125_{0.125}PbI3_3σyy\sigma_{yy}0.64321.12860.48537.97537.3321
FA0.875_{0.875}Cs0.125_{0.125}PbI3_3σzz\sigma_{zz}0.32530.74460.41937.89757.5722
FA0.875_{0.875}Cs0.125_{0.125}PbI3_3σxy\sigma_{xy}3.97243.62950.34290.38163.5907
FA0.875_{0.875}Cs0.125_{0.125}PbI3_3σxz\sigma_{xz}3.77663.50360.27290.32933.4473
FA0.875_{0.875}Cs0.125_{0.125}PbI3_3σyz\sigma_{yz}2.67412.80360.12950.51962.1545

Table S5 (PDF p. 34). Stress elements predicted by HSE06, DeePKS, and PBE for hybrid perovskites. Absolute errors of the DeePKS and PBE results with respect to HSE06 results are listed. Numbers are shown in the unit of KBar. (cont.).

References

  1. [1]Chu, S.; Majumdar, A. Opportunities and Challenges for a Sustainable Energy Future. Nature 2012, 488, 294–303.DOI
  2. [2]Deng, Y.; Xu, J.; Liu, Y.; Mancl, K. Biogas as a Sustainable Energy Source in China: Regional Development Strategy Application and Decision Making. Renewable Sustainable Energy Rev. 2014, 35, 294–303.DOI
  3. [3]Chu, S.; Cui, Y.; Liu, N. The Path towards Sustainable Energy. Nat. Mater. 2016, 16, 16–22.DOI
  4. [4]Parida, B.; Iniyan, S.; Goic, R. A Review of Solar Photovoltaic Technologies. Renewable Sustainable Energy Rev. 2011, 15, 1625–1636.
  5. [5]Riede, M.; Spoltore, D.; Leo, K. Organic Solar Cells—The Path to Commercial Success. Adv. Energy Mater. 2021, 11, 2002653.DOI
  6. [6]Nayak, P. K.; Mahesh, S.; Snaith, H. J.; Cahen, D. Photovoltaic Solar Cell Technologies: Analysing the State of the Art. Nat. Rev. Mater. 2019, 4, 269–285.DOI
  7. [7]Gong, J.; Liang, J.; Sumathy, K. Review on dye-sensitized solar cells (DSSCs): Fundamental concepts and novel materials. Renewable Sustainable Energy Rev. 2012, 16, 5848–5860.DOI
  8. [8]Shah, A.; Torres, P.; Tscharner, R.; Wyrsch, N.; Keppner, H. Photovoltaic Technology: The Case for Thin-Film Solar Cells. Science 1999, 285, 692–698.pubmed.ncbi.nlm.nih.gov/10426984
  9. [9]Polman, A.; Knight, M.; Garnett, E. C.; Ehrler, B.; Sinke, W. C. Photovoltaic materials: Present efficiencies and future challenges. Science 2016, 352, aad4424.DOI
  10. [10]Andreani, L. C.; Bozzola, A.; Kowalczewski, P.; Liscidini, M.; Redorici, L. Silicon solar cells: toward the efficiency limits. Adv. Phys.: X 2019, 4, 1548305.DOI
  11. [11]Sun, Z.; Chen, X.; He, Y.; Li, J.; Wang, J.; Yan, H.; Zhang, Y. Toward Efficiency Limits of Crystalline Silicon Solar Cells: Recent Progress in High-Efficiency Silicon Heterojunction Solar Cells. Adv. Energy Mater. 2022, 12, 2200015.DOI
  12. [12]Battaglia, C.; Cuevas, A.; De Wolf, S. High-efficiency crystalline silicon solar cells: status and perspectives. Energy Environ. Sci. 2016, 9, 1552–1576.
  13. [13]Peng, K.-Q.; Lee, S.-T. Silicon Nanowires for Photovoltaic Solar Energy Conversion. Adv. Mater. 2011, 23, 198–215.DOI
  14. [14]Jin, J.; Wang, Q.; Ma, K.; Shen, W.; Belfiore, L. A.; Bao, X.; Tang, J. Recent Developments of Polymer Solar Cells with Photovoltaic Performance over 17%. Adv. Funct. Mater. 2023, 33, 2213324.DOI
  15. [15]Wang, X.; Zhao, B.; Kan, W.; Xie, Y.; Pan, K. Review on Low-Cost Counter Electrode Materials for Dye-Sensitized Solar Cells: Effective Strategy to Improve Photovoltaic Performance. Adv. Mater. Interfaces 2022, 9, 2101229.DOI
  16. [16]Datt, R.; Bishnoi, S.; Lee, H. K. H.; Arya, S.; Gupta, S.; Gupta, V.; Tsoi, W. C. Down-conversion materials for organic solar cells: Progress, challenges, and perspectives. Aggregate 2022, 3, e185.DOI
  17. [17]Zhang, G.; Lin, F. R.; Qi, F.; Heumüller, T.; Distler, A.; Egelhaaf, H.-J.; Li, N.; Chow, P. C. Y.; Brabec, C. J.; Jen, A. K.-Y.; Yip, H.-L. Renewed Prospects for Organic Photovoltaics. Chem. Rev. 2022, 122, 14180–14274, PMID: 35929847.DOI
  18. [18]Lin, Y.; Zhan, X. Oligomer Molecules for Efficient Organic Photovoltaics. Acc. Chem. Res. 2016, 49, 175–183.DOI
  19. [19]Bernardo, G.; Lopes, T.; Lidzey, D. G.; Mendes, A. Progress in Upscaling Organic Photovoltaic Devices. Adv. Energy Mater. 2021, 11, 2100342.DOI
  20. [20]Jena, A. K.; Kulkarni, A.; Miyasaka, T. Halide Perovskite Photovoltaics: Background, Status, and Future Prospects. Chem. Rev. 2019, 119, 3036–3103.DOI
  21. [21]Wells, H. L. Über die Cäsium- und Kalium-Bleihalogenide. Z. Anorg. Allg. Chem. 1893, 3, 195–210.DOI
  22. [22]Xu, F.; Zhang, M.; Li, Z.; Yang, X.; Zhu, R. Challenges and Perspectives toward Future Wide-Bandgap Mixed-Halide Perovskite Photovoltaics. Adv. Energy Mater. 2023, 13, 2203911.DOI
  23. [23]Wang, R.; Huang, T.; Xue, J.; Tong, J.; Zhu, K.; Yang, Y. Prospects for Metal Halide Perovskite-Based Tandem Solar Cells. Nat. Photon. 2021, 15, 411–425.DOI
  24. [24]Chen, Y.; Liu, X.; Wang, T.; Zhao, Y. Highly Stable Inorganic Lead Halide Perovskite toward Efficient Photovoltaics. Acc. Chem. Res. 2021, 54, 3452–3461, PMID: 34428021.DOI
  25. [25]Yang, W.-F.; Igbari, F.; Lou, Y.-H.; Wang, Z.-K.; Liao, L.-S. Tin Halide Perovskites: Progress and Challenges. Adv. Energy Mater. 2020, 10, 1902584.DOI
  26. [26]Tong, Y.; Najar, A.; Wang, L.; Liu, L.; Du, M.; Yang, J.; Li, J.; Wang, K.; Liu, S. F. Wide-Bandgap Organic–Inorganic Lead Halide Perovskite Solar Cells. Adv. Sci. 2022, 9, 2105085.DOI
  27. [27]Gu, S.; Lin, R.; Han, Q.; Gao, Y.; Tan, H.; Zhu, J. Tin and Mixed Lead–Tin Halide Perovskite Solar Cells: Progress and their Application in Tandem Solar Cells. Adv. Mater. 2020, 32, 1907392.DOI
  28. [28]Rigter, S. A.; Quinn, X. L.; Kumar, R. E.; Fenning, D. P.; Massonnet, P.; Ellis, S. R.; Heeren, R. M. A.; Svane, K. L.; Walsh, A.; Garnett, E. C. Passivation Properties and Formation Mechanism of Amorphous Halide Perovskite Thin Films. Adv. Funct. Mater. 2021, 31, 2010330.DOI
  29. [29]He, R.; Wu, H.; Zhang, L.; Wang, X.; Fu, F.; Liu, S.; Zhong, Z. Structural phase transitions in SrTiO₃ from deep potential molecular dynamics. Phys. Rev. B 2022, 105, 064104.DOI
  30. [30]Jain, D.; Chaube, S.; Khullar, P.; Goverapet Srinivasan, S.; Rai, B. Bulk and surface DFT investigations of inorganic halide perovskites screened using machine learning and materials property databases. Phys. Chem. Chem. Phys. 2019, 21, 19423–19436.DOI
  31. [31]Grote, C.; Berger, R. F. Strain Tuning of Tin–Halide and Lead–Halide Perovskites: A First-Principles Atomic and Electronic Structure Study. J. Phys. Chem. C 2015, 119, 22832–22837.DOI
  32. [32]ten Brinck, S.; Infante, I. Surface Termination, Morphology, and Bright Photoluminescence of Cesium Lead Halide Perovskite Nanocrystals. ACS Energy Lett. 2016, 1, 1266–1272.DOI
  33. [33]Kashtiban, R. J.; Patrick, C. E.; Ramasse, Q.; Walton, R. I.; Sloan, J. Picoperovskites: The Smallest Conceivable Isolated Halide Perovskite Structures Formed within Carbon Nanotubes. Adv. Mater. 2023, 35, 2208575.DOI
  34. [34]Xiao, Z.; Yan, Y. Progress in Theoretical Study of Metal Halide Perovskite Solar Cell Materials. Adv. Energy Mater. 2017, 7, 1701136.DOI
  35. [35]Brivio, F.; Butler, K. T.; Walsh, A.; van Schilfgaarde, M. Relativistic quasiparticle self-consistent electronic structure of hybrid halide perovskite photovoltaic absorbers. Phys. Rev. B 2014, 89, 155204.DOI
  36. [36]Even, J.; Pedesseau, L.; Jancu, J.-M.; Katan, C. Importance of Spin–Orbit Coupling in Hybrid Organic/Inorganic Perovskites for Photovoltaic Applications. J. Phys. Chem. Lett. 2013, 4, 2999–3005.DOI
  37. [37]Wiktor, J.; Rothlisberger, U.; Pasquarello, A. Predictive Determination of Band Gaps of Inorganic Halide Perovskites. J. Phys. Chem. Lett. 2017, 8, 5507–5512, PMID: 29077408.DOI
  38. [38]Kaiser, W.; Carignano, M.; Alothman, A. A.; Mosconi, E.; Kachmar, A.; Goddard, W. A. I.; De Angelis, F. First-Principles Molecular Dynamics in Metal-Halide Perovskites: Contrasting Generalized Gradient Approximation and Hybrid Functionals. J. Phys. Chem. Lett. 2021, 12, 11886–11893, PMID: 34875174.DOI
  39. [39]Chen, Y.; Zhang, L.; Wang, H.; E, W. Ground State Energy Functional with Hartree–Fock Efficiency and Chemical Accuracy. J. Phys. Chem. A 2020, 124, 7155–7165.DOI
  40. [40]Chen, Y.; Zhang, L.; Wang, H.; E, W. DeePKS: a Comprehensive Data-Driven Approach towards Chemically Accurate Density Functional Theory. J. Chem. Theory Comput. 2020, 17, 170–181.DOI
  41. [41]Li, W.; Ou, Q.; Chen, Y.; Cao, Y.; Liu, R.; Zhang, C.; Zheng, D.; Cai, C.; Wu, X.; Wang, H.; Chen, M.; Zhang, L. DeePKS + ABACUS as a Bridge between Expensive Quantum Mechanical Models and Machine Learning Potentials. J. Phys. Chem. A 2022, 126, 9154–9164, PMID: 36455227.DOI
  42. [42]Das, T.; Di Liberto, G.; Pacchioni, G. Density Functional Theory Estimate of Halide Perovskite Band Gap: When Spin Orbit Coupling Helps. J. Phys. Chem. C 2022, 126, 2184–2198.DOI
  43. [43]Li, P.; Liu, X.; Chen, M.; Lin, P.; Ren, X.; Lin, L.; Yang, C.; He, L. Large-Scale Ab Initio Simulations Based on Systematically Improvable Atomic Basis. Comput. Mat. Sci. 2016, 112, 503–517.arxiv.org/abs/1503.00097
  44. [44]Chen, M.; Guo, G.-C.; He, L. Systematically Improvable Optimized Atomic Basis Sets for Ab Initio Calculations. J. Phys.: Condens. Matt. 2010, 22, 445501.DOI
  45. [45]Tuo, P.; Li, L.; Wang, X.; Chen, J.; Zhong, Z.; Xu, B.; Dai, F.-Z. Spontaneous Hybrid Nano-Domain Behavior of the Organic–Inorganic Hybrid Perovskites. Adv. Funct. Mater. n/a, 2301663.DOI
  46. [46]Kresse, G.; Furthmüller, J. Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set. Comput. Mater. Sci. 1996, 6, 15–50.DOI
  47. [47]Kresse, G.; Furthmüller, J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 1996, 54, 11169–11186.pubmed.ncbi.nlm.nih.gov/9984901
  48. [48]Kresse, G.; Hafner, J. Ab initio molecular dynamics for open-shell transition metals. Phys. Rev. B 1993, 48, 13115–13118.pubmed.ncbi.nlm.nih.gov/10007687
  49. [50]Jin, G.; Zheng, D.; He, L. Calculation of Berry Curvature Using Non-Orthogonal Atomic Orbitals. J. Phys. Condens. Matter. 2021, 33, 325503.DOI
  50. [51]Lin, P.; Ren, X.; He, L. Strategy for Constructing Compact Numerical Atomic Orbital Basis Sets by Incorporating the Gradients of Reference Wavefunctions. Phys. Rev. B 2021, 103, 235131.DOI
  51. [52]Hamann, D. R. Optimized Norm-Conserving Vanderbilt Pseudopotentials. Phys. Rev. B 2013, 88, 085117.DOI
  52. [53]Zhang, L.; Zhang, X.; Lu, G. Intramolecular Band Alignment and Spin–Orbit Coupling in Two-Dimensional Halide Perovskites. J. Phys. Chem. Lett. 2020, 11, 6982–6989, PMID: 32787199.DOI
  53. [54]Yumoto, G.; Hirori, H.; Sekiguchi, F.; Sato, R.; Saruyama, M.; Teranishi, T.; Kanemitsu, Y. Strong Spin-Orbit Coupling Inducing Autler-Townes Effect in Lead Halide Perovskite Nanocrystals. Nat. Commun. 2021, 12.DOI
  54. [55]Chatterjee, S.; Payne, J.; Irvine, J. T. S.; Pal, A. Bandgap bowing in a zero-dimensional hybrid halide perovskite derivative: spin–orbit coupling versus lattice strain. J. Mater. Chem. A 2020, 8, 4416–4427.DOI
  55. [56]Scherpelz, P.; Govoni, M.; Hamada, I.; Galli, G. Implementation and Validation of Fully Relativistic GW Calculations: Spin–Orbit Coupling in Molecules, Nanocrystals, and Solids. J. Chem. Theory Comput. 2016, 12, 3523–3544, PMID: 27331614.DOI
  56. [57]Hafner, J. Ab-initio simulations of materials using VASP: Density-functional theory and beyond. J. Comput. Chem. 2008, 29, 2044–2078.DOI

Paper details

vlnc:W_37j64vwf
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents