WorksPhysics and Astronomy
From Symmetry to Stability: Structural and Electronic Transformation in CsKInI
Abstract
CsKInI is a promising lead-free halide double perovskite with a calculated direct band gap of 1.94 eV, ideal for solar cell applications. Our first-principles calculations reveal that its cubic phase (Fmm) is dynamically unstable. Using an accelerated machine learning approach, we identify 42 dynamically stable structures and further validate these findings using first-principles calculations on 11 of these. The most stable phase has Cmc symmetry with 20 atoms/unit cell. It lies 13 meV/atom above the convex hull but lacks octahedral cation coordination. The most stable perovskite-like structure has P symmetry with 10 atoms/unit cell and low octahedral connectivity. Structure-property trade-offs are highlighted, with calculated distortions generally widening the band gap, shifting it from direct to indirect, and flattening the band edges. This work showcases the synergy of genetic algorithms, machine-learned potentials, and first-principles validation for discovering stable, complex materials.
Halide perovskites have attracted significant attention for solar cell applications due to their tunable band gap [1], high absorption coefficient [2] and low-cost solution-based processability [3]. As a result, the power conversion efficiency of single-junction halide perovskite solar cells has increased rapidly, reaching 27% in a short period [4]. However, most high-performance halide perovskites, such as the organic–inorganic hybrid methylammonium (MA) lead halides [5, 6, 7] and inorganic cesium lead halides [8], contain toxic lead, which poses serious environmental and health concerns that hinder large-scale commercialization [9]. In addition, these materials often suffer from limited environmental stability [10, 11]. These challenges have prompted researchers to explore the broader chemical space in search of lead-free and more stable halide perovskite alternatives. In particular, lead-free halide double perovskites with the general formula have emerged as promising candidates for photovoltaic applications [12, 13, 14]. These structures are derived by doubling the formula unit of the conventional single perovskite ( being a halogen) and replacing the two lead cations with two different metal cations, and . These substituting cations may both adopt a oxidation state, or one may be and the other , so that their combined charge is equivalent to that of two ions. Ideally with a -type cation, which typically shows enhanced optical absorption, well-balanced carrier transport, good defect tolerance, and long carrier diffusion lengths [15]. This substitution scheme significantly expands the accessible chemical space compared to perovskites. Such perovskites have been reported to exhibit improved thermal and moisture stability compared to hybrid perovskites such as (MA)PbI₃. In particular, Cs₂AgBiBr₆ demonstrates improved durability and defect tolerance [12].
However, it has an indirect band gap [16], which limits its effectiveness in photovoltaic applications [14]. To find better candidates, Cai et al. [17] performed a high-throughput study and reported the Cs₂KInI₆ perovskite but filtered it out due to its high energy above the convex hull (56 meV/atom) [18]. In a more recent work, its cubic phase was identified as a promising candidate for photovoltaic applications among 1026 compounds [19] due to its direct band gap that matches optimal sunlight absorption. Nevertheless, its vibrational dynamical stability was not investigated. In this work, we study the dynamical stability of Cs₂KInI₆ in the cubic phase with space group (225). Our phonon analysis reveals that this high-symmetry phase is dynamically unstable, leading to 42 more stable phases with lower symmetry. In the cubic phase, both K and In cations occupy octahedral coordination environments. These octahedra are corner-sharing, forming a three-dimensional alternating network. The dynamically stable phases are analyzed on the basis of the possible modification of the local environments of the K and In cations and the polyhedra connectivity, which has a significant impact on transport properties [20]. Furthermore, the impact of these structural transformations on electronic properties is evaluated to estimate the suitability of this material for photovoltaic applications.
The initial crystal structure of cubic Cs₂KInI₆ was obtained from the Materials Project database [18]. Electronic structure calculations were performed using density functional theory (DFT) [21, 22] as implemented in the Quantum Espresso (QE) package [23]. The Perdew–Burke–Ernzerhof (PBE) [24] exchange-correlation functional was used with norm-conserving pseudopotentials from the PseudoDojo library [25]. For three structures (, and ), each containing 10 atoms per unit cell, the band structures were also computed using the hybrid Heyd–Scuseria–Ernzerhof (HSE06) [26] functional on a homogeneous grid and Wannier interpolated [27] to produce the band structure. After performing convergence tests on total energy per atom, lattice constants, and atomic positions, the plane-wave kinetic energy cutoff was set to 80 Ry. The convergence thresholds were chosen to be 0.1 mHa/atom for total energy and 0.1% of the converged value for lattice constants and atomic positions. The Brillouin zone was sampled using a Γ-centered Monkhorst–Pack [28] k-point mesh. Structural optimization was performed with a force convergence threshold of Ry/bohr. Phonon calculations were performed using density functional perturbation theory (DFPT) [29, 30] for high-accuracy validation, using q-point grids with a density below [31] and a k-point mesh denser or equal to the corresponding q-point grid. For high-throughput analysis, phonons were computed using the finite-displacement method with a machine-learned interatomic potential (MLIP) based on the message-passing atomic cluster expansion (MACE), specifically the MACE-OMAT-0 model [32, 33, 34], with phonon frequencies converged to within . The energies above the convex hull of the CsKInI phases were calculated with respect to their stable decomposition products identified from the Materials Project database, using the pymatgen [35] phase diagram module together with total energies recomputed with Quantum ESPRESSO. Structural analysis was performed using ChemEnv [36] to determine the coordination environments of K and In cations.
Figure 1 shows the phonon dispersion and phonon density of states (DOS) of cubic CsKInI using DFPT. The presence of unstable modes indicates that this high-symmetry phase lies at a saddle point on the potential energy surface [37], reflecting its dynamical instability. Although such an instability can sometimes be stabilized by including anharmonic effects at finite temperature [38], it is useful to identify the ground-state structure from DFT to describe perovskites at cryogenic temperatures or in situations where the high-symmetry phase
Figure 1. Phonon dispersion and phonon density of states (DOS) for cubic CsKInI (space group Fmm, 225), computed using density functional perturbation theory with a k-point grid and q-point grid.
cannot be stabilized. To achieve this, a MACE potential was used with our in-house code called VIBROMOL [39] to explore the complex potential energy surface of the CsKInI structure. To resolve its dynamical instabilities, we use a genetic algorithm for a global search aimed at identifying dynamically stable low-energy polymorphs in the energy landscape of the material. The procedure begins by identifying the soft phonon modes of the unstable cubic phase. An initial population of distorted candidate structures is then generated by displacing atoms along the eigenvectors of these unstable modes. This initial generation explores various displacement amplitudes and coupling with other soft modes to ensure a diverse search space. The algorithm then proceeds through an iterative evolutionary cycle. At each generation, the fitness of every candidate structure is evaluated on the basis of its total energy per atom after a structural relaxation using the MACE potential. The candidates with the lowest energies are preferentially selected as parents for the next generation. New offspring structures are created using genetic operators such as crossover, which combines features from parent structures, and mutation, which applies small, random perturbations. This process, inspired by natural selection, effectively navigates the energy landscape toward more stable configurations.
Crucially, the workflow incorporates a feedback loop: a phonon calculation is performed on the most promising low-energy candidates of each generation. If any instabilities persist, the corresponding soft modes are used to guide the creation of new distorted structures in subsequent iterations. This evolutionary approach allows for a broad and efficient search that systematically uncovers lower-energy, dynamically stable polymorphs originating from the unstable parent structure. This process yielded 42 new polymorphs of CsKInI, which were found to be dynamically stable based on frozen-phonon calculations using MACE (see Table S1 of the Supplemental Information, SI [40]). They can be classified according to the number of atoms in their primitive unit cell (10, 20, 40, and 80 atoms). However, confirmation using first-principles DFPT calculations is required to validate the dynamical stability predicted by MACE. Therefore, we selected all polymorphs with 10 atoms, two with 20 atoms (one perovskite structure and the most stable non-perovskite structure), one with 40 atoms, and the one for the 80 atoms count. This resulted in a total of 11 structures for which DFPT phonon calculations were performed. We found that in 9 cases the DFPT calculations confirmed the dynamic stability predicted by MACE, whereas in two cases MACE predicted stability but DFPT revealed dynamical instabilities. As a result, DFPT validation remains necessary for the remaining 31 structures. However, these structures contain 20 and 40 atoms in their primitive unit cells and therefore require substantial computational resources for DFPT phonon calculations, making a complete DFPT validation of all candidate polymorphs computationally demanding. In addition, some polymorphs of CsKInI were predicted to be dynamically unstable by both MACE and DFPT; these structures are listed in Table S2 [40]. Overall, the DFPT phonon dispersions and phonon densities of states show very good agreement with the corresponding MACE results (see Figure S1–Figure S3 and S5–S6 for phonon dispersions, and Figure S4 and S7 for phonon densities of states in the SI [40]). For the remainder of this work, we focus on four structures (IDs = 05, 07, 37, and 42), which are confirmed to be dynamically stable at both the MACE and DFPT levels. The P (147) structure (id=05) with 10 atoms per unit cell and a double perovskites structure; the I2m (121) structure (id=07) which is the most stable structure with 10 atoms and high symmetry; the Cmc2 (36) structure (id=37) as the most stable; and the P (2) structure (id=42) which is the largest structure with 80 atoms per unit cell. We report their lattice parameters, number of atoms in the primitive unit cell, and the energy above the convex hull () in Table I. The lattice parameters were recomputed using PBEsol exchange-correlation functional, which are reported in Table S3 of SI [40]. The lattice vectors obtained with PBEsol are on average 3.4% shorter than those obtained with PBE. According to their energies above the convex hull, the P (147), I2m (121), Cmc2 (36), and P (2) phases are thermodynamically metastable. However, their values lie within 25 meV/atom, which is the typical threshold for the experimentally observed iodides [41]. We also present their X-ray diffraction in Figure S8 of the SI [40] which show large differences that can easily be resolved experimentally. We also present their phonon dispersion and phonon DOS in Figure 2. Remarkably, the two perovskites structures, Figure 2(a) and (Figure 2b), show a continuous spectra with phonon frequencies below 150 cm while the two non-perovskite structures show multiple localized and high-energy vibrational modes up to 180 cm, see Figure 2(c) and (Figure 2d). In particular, both show the same iodine localized vibrational peak at 120 cm which is characteristic of these structures. These differences can be explained by studying their crystal structure, shown in Figure 3. The two struc-
Figure 2. DFPT phonon calculations with density of states (DOS) of four CsKInI phases. (a) P with -grid and -grid. (b) P with -grid and -grid. (c) I2m with -grid -grid. (d) Cmc2 with -grid and -grid.
Figure 3. Crystal structures of (a) , (b) , (c) , (d) and (e) , displaying the coordination environments of In and K cations and the way they are connected.
| id | Space group | Stability with DFPT | ||||
| 00 | 225 | 10 | 55 | 1.24 | unstable | |
| 01 | 10 | 10 | 46 | 1.76 | stable | |
| 02 | 2 | 10 | 33 | 1.95 | stable | |
| 03 | 162 | 10 | 33 | 1.95 | unstable | |
| 04 | 12 | 10 | 30 | 1.60 | stable | |
| 05 | 147 | 10 | 24 | 1.66 | stable | |
| 06 | 5 | 10 | 21 | 2.24 | stable | |
| 07 | 121 | 10 | 21 | 2.24 | stable | |
| 08 | 2 | 20 | 48 | 1.26 | - | |
| 09 | 14 | 20 | 41 | 1.58 | unstable | |
| 10 | 2 | 20 | 41 | 1.19 | - | |
| 11 | 2 | 20 | 38 | 1.78 | - | |
| 12 | 2 | 20 | 38 | 2.00 | - | |
| 13 | 2 | 20 | 37 | 1.82 | - | |
| 14 | 2 | 20 | 37 | 1.93 | - | |
| 15 | 2 | 20 | 37 | 1.78 | - | |
| 16 | 2 | 20 | 36 | 1.98 | - | |
| 17 | 2 | 20 | 35 | 1.85 | - | |
| 18 | 2 | 20 | 35 | 1.83 | - | |
| 19 | 2 | 20 | 33 | 1.83 | - | |
| 20 | 2 | 20 | 31 | 2.16 | - | |
| 21 | 2 | 20 | 31 | 2.00 | - | |
| 22 | 2 | 20 | 30 | 2.08 | - | |
| 23 | 2 | 20 | 30 | 1.94 | - | |
| 24 | 2 | 20 | 27 | 2.08 | - | |
| 25 | 2 | 20 | 27 | 1.87 | - | |
| 26 | 2 | 20 | 26 | 1.97 | - | |
| 27 | 2 | 20 | 25 | 2.37 | - | |
| 28 | 2 | 20 | 24 | 2.08 | - | |
| 29 | 2 | 20 | 24 | 1.67 | - | |
| 30 | 2 | 20 | 24 | 2.20 | - | |
| 31 | 2 | 20 | 23 | 1.68 | - | |
| 32 | 63 | 20 | 21 | 2.15 | - | |
| 33 | 2 | 20 | 21 | 1.89 | - | |
| 34 | 12 | 20 | 19 | 2.31 | - | |
| 35 | 2 | 20 | 19 | 2.31 | - | |
| 36 | 40 | 20 | 15 | 2.28 | - | |
| 37 | 36 | 20 | 13 | 2.26 | stable | |
| 38 | 2 | 40 | 31 | 1.93 | - | |
| 39 | 2 | 40 | 29 | 2.25 | - | |
| 40 | 2 | 40 | 27 | 1.36 | - | |
| 41 | 7 | 40 | 15 | 2.28 | stable | |
| 42 | 2 | 80 | 25 | 1.35 | stable |
Table S1. Polymorphs of which are stable according to MACE, with the energy above the convex hull ( in meV/atom), band gap ( in eV), and stability predicted with DFPT.
Table S2. Polymorphs of unstable according to MACE and DFPT, with energy above the convex hull ( in meV/atom), and band gap ( in eV).
| id | Space group | ||||
| 00 | Fmm | 225 | 10 | 55 | 1.24 |
| 43 | R | 148 | 10 | 38 | 1.53 |
| 44 | C2/m | 12 | 10 | 33 | 1.95 |
| 45 | C2/m | 12 | 10 | 26 | 1.48 |
| 46 | Fmm2 | 42 | 10 | 21 | 2.13 |
| 47 | Cm | 8 | 10 | 21 | 2.14 |
Table S2.
Figure S1. Phonon dispersions from DFPT (left) and finite displacement with MACE (right) for phases with id numbers from Table S1. All MACE calculations use a displacement of 0.01 Å. (a) id=01 with k-grid and q-grid (DFPT), and a supercell (MACE). (b) id=02 with k-grid and q-grid (DFPT), and supercell (MACE). (c) id=03 with k-grid and q-grid (DFPT), and a supercell (MACE).
Figure S3. Phonon dispersions from DFPT (left) and finite displacement with MACE (right) for phases with id numbers from Table S1. All MACE calculations use a displacement of 0.01 Å. (a) id=09 with k-grid and q-grid (DFPT), and a supercell (MACE). (b) id=37 with k-grid and q-grid (DFPT), and a supercell (MACE). (c) id=41 with k-grid and q-grid (DFPT), and a supercell (MACE). (d) id=42 with k-grid and q-grid (DFPT), and a supercell (MACE).
Figure S4. Phonon density of states (PDOS) obtained from DFPT and the finite-displacement method using MACE for phases with id numbers listed in Table S1. The corresponding phonon band dispersions are shown in Figure S1, Figure S2, and Figure S3. A q-point grid was used to compute the PDOS, except for the structures with 40 (id=41) and 80 (id=42) atoms, for which and grids were used, respectively. The PDOS were normalized by the area under the curve within a common frequency range shared by the DFPT and MACE results. (a) id=01, (b) id=02, (c) id=03, (d) id=04, (e) id=05, (f) id=06, (g) id=07, (h) id=09, (i) id=37, (j) id=41, (k) id=42.
Figure S2. Phonon dispersions from DFPT (left) and finite displacement with MACE (right) for CsKInI phases with id numbers from Table S1. All MACE calculations use a displacement of 0.01 Å. (a) id=04 with 4 × 4 × 4 k-grid and 2 × 2 × 2 q-grid (DFPT), and a 5 × 5 × 5 supercell (MACE). (b) id=05 with 4 × 4 × 4 k-grid and 2 × 2 × 2 q-grid, and a 5 × 5 × 5 supercell (MACE). (c) id=06 with 4 × 4 × 6 k-grid and 2 × 2 × 3 q-grid (DFPT), and a 3 × 3 × 5 supercell (MACE). (d) id=07 with 3 × 3 × 3 k-grid and 2 × 2 × 2 q-grid (DFPT), and a 5 × 5 × 5 supercell (MACE).
Figure S8. X-ray powder diffractogram of four dynamically stable CsKInI phases, obtained from VESTA with X-rays of wavelength 1.54 Å. (a) (147), (b) (2), (c) (121), (d) (36).
| Space group | id | |||||||
| Fmm | 225 | 00 | 12.332 | 12.332 | 12.332 | 90.0 | 90.0 | 90.0 |
| P | 147 | 05 | 8.267 | 8.267 | 8.381 | 90.0 | 90.0 | 120.0 |
| I2m | 121 | 07 | 7.575 | 7.575 | 18.206 | 90.0 | 90.0 | 90.0 |
| Cmc2 | 36 | 37 | 10.280 | 18.919 | 9.881 | 90.0 | 90.0 | 90.0 |
| P | 2 | 42 | 16.767 | 16.826 | 17.221 | 91.9 | 118.5 | 118.9 |
Table S3. Lattice parameters (, , and in Å; angles , , and in °) of the dynamically stable polymorphs of CsKInI, computed using PBEsol exchange-correlation functional.
| Space group | id | |||||||||
| Fmm | 225 | 00 | 12.669 | 12.669 | 12.669 | 90.0 | 90.0 | 90.0 | 10 | 55 |
| P | 147 | 05 | 8.649 | 8.649 | 8.649 | 90.0 | 90.0 | 120.0 | 10 | 24 |
| I2m | 121 | 07 | 7.756 | 7.756 | 19.018 | 90.0 | 90.0 | 90.0 | 10 | 21 |
| Cmc2 | 36 | 37 | 10.644 | 19.482 | 10.299 | 90.0 | 90.0 | 90.0 | 20 | 13 |
| P | 2 | 42 | 17.427 | 17.502 | 17.662 | 91.7 | 119.0 | 119.1 | 80 | 25 |
Table 1. Lattice parameters (, , and in Å; angles , , and in °) of the conventional unit cell, number of atoms in the primitive unit cell (), and energy above the convex hull () for the dynamically stable DFPT polymorphs of CsKInI in meV/atom.
tures shown in Figure 3(c) and (Figure 3d) cannot be classified as double perovskites as both cation sites (K and In in this case) are not in an octahedral coordination environments as the cubic phase of CsKInI. Indeed, the In cation is in a tetrahedral environment, while the geometry of the K cation is not well-defined; it is coordinated with six I atoms in I and seven I atoms in Cmc2. For the P structure, our ChemEnv analysis gives 100% octahedral similarity with a slightly distorted octahedron for the K cations. Finally, in the P structure, the K cations do not exhibit an octahedral environment. However, ChemEnv reports a combination of different coordinations for each K site, see Figure S9 of the SI [40]. From this analysis, the P (147) phase is classified as a double perovskite while the case of the P structure is more complex since the In cations are octahedrally coordinated while the K cations are not. Since the P structure is classified as a double perovskite, we compare it with the cubic phase. In the latter, the In and K octahedra share corners in all three directions, forming a three-dimensional alternating network, see Figure 3(a). In the trigonal phase, they share faces instead of corners, see Figure 3(b), producing one-dimensional strips in which the In and K octahedra alternate. This change from a three-dimensional corner-sharing network to one-dimensional face-sharing strips reflects a significant reduction in structural connectivity.
Figure S9. Analysis of the coordination environments of K cations for (2): T:5 = Trigonal bipyramid, PB:7 = Pentagonal bipyramid, T:6 = Trigonal prism, DD:8 = Dodecahedron with triangular faces, DDPN:8 = Dodecahedron with triangular faces – p2345 plane normalized, TT_1:9 = Tricapped triangular prism (three square-face caps), TT_2:9 = Tricapped triangular prism (two square-face caps and one triangular-face cap), SBT:8 = Square-face bicapped trigonal prism.
We now turn to the analysis of the impact of structural differences on electronic properties. The cubic phase has a direct band gap at the point, with value of 1.24 eV at the PBE level (Figure S10), reduced to 1.06 eV upon inclusion of SOC (Figure S10), increased to 1.94 at the HSE06+SOC level, as shown in Figure S12 of the SI [40]. Due to structural distortions, most notably changes in the In and K coordination environment and the connectivity of the their octahedra, the band gaps of the four considered cases increase. This trend is evident from Figure 4, which presents the PBE+SOC electronic band structures of the four investigated phases of CsKInI. Moreover, the band gap changes from direct to indirect in two phases (P and I). The corresponding PBE band structures without SOC are provided in Figure S11 of the SI [40]. The double perovskite structure, P has an indirect band gap of 1.68 eV at the PBE level (Figure S11) [40], 1.49 eV with PBE+SOC (Figure 4(a)) and 2.29 eV with HSE06+SOC as presented in the Figure S12 [40]. This gap is only slightly larger than that of the cubic phase. The structure P, in which only the In cations form octahedrons has a direct band gap of 1.35 eV at the PBE level (Figure S11) [40] and 1.22 eV with PBE+SOC, see Figure 4(b). These band gap values are close to that of the cubic phase. Therefore, among the considered phases, P could be a suitable candidate for the photovoltaic applications if synthesized. The I and Cmc2 structures that are not classified as double perovskites have significantly larger band gaps than the cubic phase. I has band gap values of 2.24 eV at the PBE level (Figure S11) [40], 2.05 eV with PBE+SOC (Figure 4(c)), and 2.93 eV with HSE+SOC as reported in the Fig. S12(c) [40].
Figure 4. Electronic band structures and densities of states of dynamically stable phases of with PBE including spin-orbit coupling (SOC): (a) , (b) , (c) , and (d) . The energies are expressed with respect to the Fermi level (), located at the valence band maximum.

Figure S10. Electronic band structures and densities of states of CsKInI for the high-symmetry unstable Fmm (225) phase calculated using (a) PBE and (b) PBE+SOC. The energies are expressed with respect to the Fermi level (), located at the valence band maximum.
Figure S11. Electronic band structures and densities of states of dynamically stable phases of CsKInI calculated at the PBE level: (a) P, (b) P, (c) I2m, and (d) Cmc2. The energies are expressed with respect to the Fermi level (), located at the valence band maximum.
Figure S12. Electronic band structures of CsKInI phases calculated using HSE06+SOC and Wannier interpolation: (a) Fmm, (b) P, and (c) I2m. The energies are expressed with respect to the Fermi level (), located at the valence band maximum.
the Cmc2 has a PBE band gap of 2.26 eV (Figure S11) [40] and 2.04 eV with PBE+SOC, see Figure 4(d). P contains 80 atoms and Cmc2 contains 20 atoms. Therefore, their band structures could not be computed with HSE06+SOC level due to the high computational cost. However, based on the HSE06+SOC band gaps of Fmm, P, and I phases, the band gap increases by approximately 60% on average compared to the PBE+SOC values. Following this trend, the HSE06+SOC band gap of P should be approximately 1.95 eV, while that of Cmc2 approximately 3.26 eV. Overall, structural distortions increase the band gap and flatten the conduction and valence bands, thereby increasing the effective electron and hole masses, as reported in Table II.
| Space group | VBM | CBM | Direction | ||
| Fmm 225 | –L | 0.349 | 1.375 | ||
| –K | 0.349 | 1.648 | |||
| P 147 | A | –A | 1.402 | — | |
| –K | 0.548 | — | |||
| A– | — | 6.120 | |||
| A–L | — | 1.079 | |||
| P 2 | –Z | 0.517 | 2.418 | ||
| –L | 0.568 | 2.132 | |||
| I 121 | * | –Z | 1.419 | — | |
| –S | 0.615 | — | |||
| *– | — | 7.804 | |||
| *–S | — | 7.922 | |||
| Cmc2 36 | –Y | 0.724 | 0.607 | ||
| –Z | 1.111 | 2.233 |
Table II. Effective masses of electron and hole for the dynamically stable phases of CsKInI along with the high symmetry cubic phase. The valence band maximum (VBM) and the conduction band minimum (CBM) are also reported. For the I structure, the VBM does not lie at a high-symmetry point and is indicated with a star (*).
We investigate the structural stability and electronic properties of lead-free halide double perovskite CsKInI using first-principles calculations and the machine-learned interatomic potential MACE. Phonon calculations reveal that the high-symmetry cubic phase is dynamically unstable. To find lower-energy dynamically stable configurations, we applied a genetic algorithm along unstable phonon eigenmodes, yielding 42 dynamically stable candidates according to finite-displacement phonon calculations with MACE. A subset of structures was selected for further validation by DFPT to address possible MLIP inaccuracies. In two cases, structures predicted to be dynamically stable by the MLIP were found to remain dynamically unstable at the DFPT level, indicating that further DFPT validation is required for the remaining candidate structures. We focus on four dynamically stable phases: P, I, Cmc2, and P, classified by the coordination of In and K cations and their octahedral connectivity. These phases are thermodynamically metastable based on their calculated energy above the convex hull. However, they lie within the threshold limit of 25 meV/atom for the iodides. The most stable phases, and , are not double perovskites, as In is tetrahedrally coordinated and the K environment is ambiguous. is a double perovskite, whereas is less clearly so: In is octahedrally coordinated but K is not. Electronic-structure calculations show that distortions increase the band gap relative to the cubic phase and have changed it from direct to indirect in two cases.
The cubic phase has a direct band gap of 1.94 eV. The double-perovskite phase has an indirect band gap of 2.29 eV. The phase could be an interesting candidate among the stable phases for photovoltaics, retaining a direct band gap which is lower than the discussed stable phases. Overall, exhibits multiple candidate dynamically stable phases identified from phonon calculations using the machine-learned interatomic potential MACE, several of which are further confirmed by DFPT. These phases range from double to non-double-perovskites, with varying octahedral connectivity in the double-perovskite polymorphs. These transformations significantly modify the electronic properties, clarifying structure–property relationships in halide double perovskites and demonstrating the effectiveness of combining machine-learned interatomic potentials with genetic algorithms to discover novel stable phases in complex materials.
Acknowledgments
G.-M.R. is Research Director of the Fonds de la Recherche Scientifique - FNRS. S. P. is a Research Associate of the Fonds de la Recherche Scientifique - FNRS. This work was supported by the Fonds de la Recherche Scientifique - FNRS under Grants number T.0183.23 (PDR) and T.W011.23 (PDR-WEAVE). This publication was supported by the Walloon Region in the strategic axe FRFS-WEL-T. V. T. acknowledges support from the FRS-FNRS through a FRIA Grant. Computational resources have been provided by the Consortium des Équipements de Calcul Intensif (CÉCI), funded by the Fonds de la Recherche Scientifique de Belgique (F.R.S.-FNRS) under Grant No. 2.5020.11 and by the Walloon Region. The present research benefited from computational resources made available on the Tier-1 supercomputer of the Fédération Wallonie-Bruxelles, infrastructure funded by the Walloon Region under grant agreement no1117545.
References
SI: From Symmetry to Stability: Structural and Electronic Transformation in
Mohammad Bakhsh; Victor Trinquet; Rogério Almeida Gouvêa; Gian-Marco Rignanese; Samuel Poncé
SI.pdf
SI: From Symmetry to Stability: Structural and Electronic Transformation in
Figure S7. Phonon density of states (PDOS) obtained from DFPT and the finite-displacement method using MACE for CsKInI phases with id numbers listed in Table S2. The corresponding phonon band dispersions are shown in Figure S5 and Figure S6. A 40 × 40 × 40 q-point grid was used to compute the PDOS. The PDOS were normalized by the area under the curve within a common frequency range shared by the DFPT and MACE results. (a) id=00, (b) id=43, (c) id=44, (d) id=45, (e) id=46, (f) id=47.
Figure S5. Phonon dispersions from DFPT (left) and finite displacement with MACE (right) for phases with id numbers from Table S2. All MACE calculations use a displacement of 0.01 Å. (a) id=00 with k-grid and q-grid (DFPT), and a supercell (MACE). (b) id=43 with k-grid and q-grid (DFPT), and a supercell (MACE). (c) id=44 with k-grid and q-grid (DFPT), and supercell (MACE). (d) id=45 with k-grid and q-grid (DFPT), and a supercell (MACE).
Figure S6. Phonon dispersions from DFPT (left) and finite displacement with MACE (right) for CsKInI phases with id numbers from Table S2. All MACE calculations use a displacement of 0.01 Å. (a) id=46 with 4 × 4 × 6 k-grid and 2 × 2 × 3 q-grid (DFPT), and a 5 × 5 × 7 supercell (MACE). (b) id=47 with 6 × 6 × 4 k-grid and 3 × 3 × 2 q-grid (DFPT), and a 5 × 5 × 3 supercell (MACE).
References
- [1]M. H. Miah, M. U. Khandaker, M. B. Rahman, M. Nur-E-Alam, and M. A. Islam, Band gap tuning of perovskite solar cells for enhancing the efficiency and stability: issues and prospects, RSC Advances 14, 15876–15906 (2024).DOI
- [2]S. De Wolf, J. Holovsky, S.-J. Moon, P. Löper, B. Niesen, M. Ledinsky, F.-J. Haug, J.-H. Yum, and C. Ballif, Organometallic halide perovskites: Sharp optical absorption edge and its relation to photovoltaic performance, The Journal of Physical Chemistry Letters 5, 1035–1039 (2014).DOI
- [3]A. Chilvery, S. Das, P. Guggilla, C. Brantley, and A. Sunda-Meya, A perspective on the recent progress in solution-processed methods for highly efficient perovskite solar cells, Science and Technology of Advanced Materials 17, 650–658 (2016).pubmed.ncbi.nlm.nih.gov/27877911
- [4]National Renewable Energy Laboratory, Best research-cell efficiency chart, https://www.nrel.gov/pv/cell-efficiency (2025), accessed: 2025-06-16.DOI
- [5]Y. Zhao and K. Zhu, Organic–inorganic hybrid lead halide perovskites for optoelectronic and electronic applications, Chemical Society Reviews 45, 655–689 (2016).DOI
- [6]S. Poncé, M. Schilf, and F. Giustino, Origin of low carrier mobilities in halide perovskites, ACS Energy Letters 4, 456–463 (2019).
- [7]C. Q. Xia, J. Peng, S. Poncé, J. B. Patel, A. D. Wright, T. W. Crothers, M. Uller Rothmann, J. Borchert, R. L. Milot, H. Kraus, Q. Lin, F. Giustino, L. M. Herz, and M. B. Johnston, Limits to electrical mobility in lead-halide perovskite semiconductors, The Journal of Physical Chemistry Letters 12, 3607–3617 (2021).DOI
- [8]G. E. Eperon, G. M. Paternò, R. J. Sutton, A. Zampetti, A. A. Haghighirad, F. Cacialli, and H. J. Snaith, Inorganic caesium lead iodide perovskite solar cells, Journal of Materials Chemistry A 3, 19688–19695 (2015).DOI
- [9]M. Lyu, J. Yun, P. Chen, M. Hao, and L. Wang, Addressing toxicity of lead: Progress and applications of low-toxic metal halide perovskites and their derivatives, Advanced Energy Materials 7, 1602512 (2017).DOI
- [10]Y.-H. Kye, C.-J. Yu, U.-G. Jong, Y. Chen, and A. Walsh, Critical role of water in defect aggregation and chemical degradation of perovskite solar cells, The Journal of Physical Chemistry Letters 9, 2196–2201 (2018).arxiv.org/abs/1708.07606
- [11]U.-G. Jong, C.-J. Yu, G.-C. Ri, A. P. McMahon, N. Harrison, P. R. F. Barnes, and A. Walsh, Influence of water intercalation and hydration on chemical decomposition and ion transport in methylammonium lead halide perovskites, Journal of Materials Chemistry A 6, 1067–1074 (2018).arxiv.org/abs/1708.07608
- [12]A. H. Slavney, T. Hu, A. M. Lindenberg, and H. I. Karunadasa, A bismuth-halide double perovskite with long carrier recombination lifetime for photovoltaic applications, Journal of the American Chemical Society 138, 2138–2141 (2016).DOI
- [13]G. Volonakis, M. R. Filip, A. A. Haghighirad, N. Sakai, B. Wenger, H. J. Snaith, and F. Giustino, Lead-free halide double perovskites via heterovalent substitution of noble metals, The Journal of Physical Chemistry Letters 7, 1254–1259 (2016).arxiv.org/abs/1603.01585
- [14]C. N. Savory, A. A. Walsh, and D. O. Scanlon, Can pb-free halide double perovskites support high-efficiency solar cells?, ACS Energy Letters 1, 949–955 (2016).DOI
- [15]T. Li, S. Luo, X. Wang, and L. Zhang, Alternative lone-pair ns2-cation-based semiconductors beyond lead halide perovskites for optoelectronic applications, Advanced Materials 33, 2008574 (2021).DOI
- [16]E. T. McClure, M. R. Ball, W. Windl, and P. M. Woodward, Cs2AgBiX6 (X = Br, Cl): New visible light absorbing, lead-free halide perovskite semiconductors, Chemistry of Materials 28, 1348–1354 (2016).DOI
- [17]Y. Cai, W. Xie, Y. T. Teng, P. C. Harikesh, B. Ghosh, P. Huck, K. A. Persson, N. Mathews, S. G. Mhaisalkar, M. Sherburne, and M. Asta, High-throughput computational study of halide double perovskite inorganic compounds, Chemistry of Materials 31, 5392–5401 (2019).DOI
- [18]A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, S. Dacek, S. Cholia, D. Gunter, D. Skinner, G. Ceder, and K. A. Persson, Commentary: The materials project: A materials genome approach to accelerating materials innovation, APL Materials 1, 1–11 (2013).DOI
- [19]F. Qi, X. Lv, J. Song, X. Fu, L. Meng, and C.-Z. Lu, Optoelectronic performance of perovskite Cs2KMn6 (M = Ga, In) based on high-throughput screening and first-principles calculations, Physical Chemistry Chemi- cal Physics 25, 11484–11492 (2023).
- [20]R. L. Z. Hoye, J. Hidalgo, R. A. Jagt, J. Correa-Baena, T. Fix, and J. L. MacManus-Driscoll, The role of dimensionality on the optoelectronic properties of oxide and halide perovskites, Advanced Energy Materials 12, 2100499 (2021).DOI
- [21]P. Hohenberg and W. Kohn, Inhomogeneous electron gas, Physical Review 136, B864–B871 (1964).
- [22]W. Kohn and L. J. Sham, Self-consistent equations including exchange and correlation effects, Physical Review 140, A1133–A1138 (1965).DOI
- [23]P. Giannozzi, O. Andreussi, T. Brumme, O. Bunau, M. Buongiorno Nardelli, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, M. Cococcioni, N. Colonna, I. Carnimeo, A. Dal Corso, S. de Gironcoli, P. Delugas, R. A. DiStasio, A. Ferretti, A. Floris, G. Fratesi, G. Fugallo, R. Gebauer, U. Gerstmann, F. Giustino, T. Gorni, J. Jia, M. Kawamura, H.-Y. Ko, A. Kokalj, E. Küçükbenli, M. Lazzeri, M. Marsili, N. Marzari, F. Mauri, N. L. Nguyen, H.-V. Nguyen, A. Otero-de-la-Roza, L. Paulatto, S. Poncé, D. Rocca, R. Sabatini, B. Santra, M. Schl ipf, A. P. Seitsonen, A. Smogunov, I. Timrov, T. Thonhauser, P. Umari, N. Vast, X. Wu, and S. Baroni, Advanced capabilities for materials modelling with quantum espresso, Journal of Physics: Condensed Matter 29, 465901 (2017).DOI
- [24]J. P. Perdew, K. Burke, and M. Ernzerhof, Generalized gradient approximation made simple, Physical Review Letters 77, 3865–3868 (1996).pubmed.ncbi.nlm.nih.gov/10062328
- [25]M. van Setten, M. Giantomassi, E. Bousquet, M. Verstraete, D. Hamann, X. Gonze, and G.-M. Rignanese, The pseudodojo: Training and grading a 85 element optimized norm-conserving pseudopotential table, Computer Physics Communications 226, 39–54 (2018).DOI
- [26]J. Heyd, G. E. Scuseria, and M. Ernzerhof, Hybrid functionals based on a screened coulomb potential, The Journal of Chemical Physics 118, 8207–8215 (2003).DOI
- [27]G. Pizzi, V. Vitale, R. Arita, S. Blügel, F. Freimuth, G. Géranton, M. Gibertini, D. Gresch, C. Johnson, T. Koretsune, J. Ibáñez-Azpiroz, H. Lee, J.-M. Lihm, D. Marchand, A. Marrazzo, Y. Mokrousov, J. I. Mustafa, Y. Nohara, Y. Nomura, L. Paulatto, S. Poncé, T. Ponweiser, J. Qiao, F. Thöle, S. S. Tsirkin, M. Wierzbowska, N. Marzari, D. Vanderbilt, I. Souza, A. A. Mostofi, and J. R. Yates, Wannier90 as a community code: new features and applications, Journal of Physics: Condensed Matter 32, 165902 (2020).DOI
- [28]H. J. Monkhorst and J. D. Pack, Special points for brillouin-zone integrations, Physical Review B 13, 5188–5192 (1976).
- [29]X. Gonze and C. Lee, Dynamical matrices, born effective charges, dielectric permittivity tensors, and interatomic force constants from density-functional perturbation theory, Phys. Rev. B 55, 10355 (1997).DOI
- [30]S. Baroni, S. de Gironcoli, A. Dal Corso, and P. Giannozzi, Phonons and related crystal properties from density-functional perturbation theory, Reviews of Modern Physics 73, 515–562 (2001).DOI
- [31]M. Bercx, S. Poncé, Y. Zhang, G. Trezza, A. G. Ghezel-jehmeidan, L. Bastonero, J. Qiao, F. O. von Rohr, G. Pizzi, E. Chiavazza, and N. Marzari, Charting the landscape of bardeen-cooper-schrieffer superconductors in experimentally known compounds, PRX Energy 4, 033012 (2025).
- [32]I. Batatia, D. P. Kovács, G. N. C. Simm, C. Ortner, and G. Csányi, Mace: Higher order equivariant message passing neural networks for fast and accurate force fields (2022).arxiv.org/abs/2206.07697
- [33]I. Batatia, S. Batzner, D. P. Kovács, A. Musaelian, G. N. C. Simm, R. Drautz, C. Ortner, B. Kozinsky, and G. Csányi, The design space of e(3)-equivariant atom-centred interatomic potentials, Nature Machine Intelligence 7, 56–67 (2025).DOI
- [34]A. Loew, D. Sun, H.-C. Wang, S. Botti, and M. A. L. Marques, Universal machine learning interatomic potentials are ready for phonons, npj Computational Materials 11, 10.1038/s41524-025-01650-1 (2025).arxiv.org/abs/2412.16551
- [35]S. P. Ong, W. D. Richards, A. Jain, G. Hautier, M. Kocher, S. Cholia, D. Gunter, V. L. Chevrier, K. A. Persson, and G. Ceder, Python materials genomics (pymatgen): A robust, open-source python library for materials analysis, Computational Materials Science 68, 314–319 (2013).DOI
- [36]D. Waroquiers, J. George, M. Horton, S. Schenk, K. A. Persson, G.-M. Rignanese, X. Gonze, and G. Hautier, ChemEnv: a fast and robust coordination environment identification tool, Acta Crystallographica Section B Structural Science, Crystal Engineering and Materials 76, 683–695 (2020).DOI
- [37]U.-G. Jong, C.-J. Yu, Y.-H. Kye, S.-H. Choe, J.-S. Kim, and Y.-G. Choe, Anharmonic phonons and phase transitions in the vacancy-ordered double perovskites Cs2SnI6 from first-principles predictions, Physical Review B 99, 184105 (2019).DOI
- [38]J. Yin, O. Hellman, and S. Poncé, Impact of anharmonicity on the carrier mobility of the pb-free CsSnBr3 perovskite, Phys. Rev. B 112, L140303 (2025).arxiv.org/abs/2505.20092
- [39]R. A. Gouvêa and G.-M. Rignanese, Vibroml: an automated toolkit for high-throughput vibrational analysis and dynamic instability remediation of crystalline materials using machine-learned potentials (2026).arxiv.org/abs/2604.27685
- [40]See Supplemental Material at [URL will be inserted by publisher] for additional structural, vibrational, and electronic properties of Cs2KInI6 polymorphs, including lists of dynamically stable and unstable structures, phonon dispersions and phonon densities of states from DFPT and MACE calculations, X-ray powder diffractograms, coordination environment analysis, lattice parameters with PBEsol, and electronic band structures and densities of states calculated using PBE, PBE+SOC, and HSE+SOC.
- [41]W. Sun, S. T. Dacek, S. P. Ong, G. Hautier, A. Jain, W. D. Richards, A. C. Gamst, K. A. Persson, and G. Ceder, The thermodynamic scale of inorganic crystalline metastability, Science Advances 2, 10.1126/sciadv.1600225 (2016).DOI





