Halide diffusion in mixed-halide perovskites and heterojunctions

Geert Brocks; Mike Pols; Shuxia Tao; Viren Tyagi

arxiv: 2601.11503 · v2 · submitted 2026-01-16 · ❄️ cond-mat.mtrl-sci

Halide diffusion in mixed-halide perovskites and heterojunctions

Viren Tyagi , Mike Pols , Geert Brocks , Shuxia Tao This is my paper

Pith reviewed 2026-05-16 13:10 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci

keywords halide diffusionmixed-halide perovskitesheterojunctionsvacanciesinterstitialsmolecular dynamicsinterface effects

0 comments

The pith

Mixed-halide perovskites show faster diffusion of halide vacancies and interstitials than pure compounds, with interface structure controlling cross-boundary migration.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper examines the movement of halide vacancies and interstitials in mixed iodide-bromide perovskites and at interfaces between pure iodide and bromide versions. Simulations demonstrate that mixing the two halides increases the diffusion rates of both types of defects relative to the unmixed materials. Bromine and iodine ions also move at different speeds inside the mixed compound. At heterojunctions the local arrangement of ions at the boundary decides whether defects cross, with bromine-rich layers blocking vacancies in particular while iodine-rich layers permit passage.

Core claim

Molecular dynamics simulations using neural network potentials trained on density functional theory calculations show enhanced diffusion of halide vacancies and interstitials in CsPb(I_x Br_{1-x})_3 compared with the pure CsPbI3 and CsPbBr3 compounds. In the mixed phase, Br and I ions exhibit different mobilities. Diffusion across CsPbI3/CsPbBr3 heterojunctions is governed by interface structure, where Br-rich interfaces block vacancy migration while I-rich interfaces remain permeable.

What carries the argument

Neural network potentials for molecular dynamics that model the energy barriers and hops of halide vacancies and interstitials, capturing composition dependence and interface effects.

Load-bearing premise

The neural network potentials accurately reproduce the energy barriers and dynamics of halide defect hops from the underlying density functional theory data without systematic bias between mixed and pure compositions or at interfaces.

What would settle it

Direct experimental measurements showing that halide diffusion coefficients in mixed CsPb(I_x Br_{1-x})_3 are not higher than those in the pure end members would contradict the enhancement claim.

read the original abstract

Migration of halide defects guides ion transport in metal halide perovskites and controls the kinetics of halide mixing and phase separation. We study the diffusion of halide vacancies and interstitials in \ce{CsPb(I_{x}Br_{1-x})_{3}} and \ce{CsPbI_{3}}/\ce{CsPbBr_{3}} heterojunctions by molecular dynamics simulations using neural network potentials trained on density functional theory calculations. We observe enhanced diffusion of both vacancies and interstitials in the mixed halide compounds compared to the single halide ones, as well as a difference in mobility between Br and I ions in the mixed compound. Diffusion across heterojunctions is governed by the interface structure, where a Br-rich interface blocks migration of vacancies in particular, but an I-rich interface is permeable.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

NN-MD simulations find faster halide diffusion in mixed CsPb(I,Br)3 and interface-dependent permeability, but the claims rest on unvalidated potentials.

read the letter

The core result is that neural network potentials trained on DFT let the authors run MD and see higher vacancy and interstitial diffusion rates in mixed-halide CsPb(IxBr1-x)3 than in the pure I or Br compounds, plus a mobility split between Br and I, and clear differences in how Br-rich versus I-rich interfaces let defects cross heterojunctions. The Br-rich interface blocks vacancies while the I-rich one does not. That combination of mixed-composition runs and explicit interface models is the concrete new piece; prior work stayed mostly with single-halide systems or static DFT barriers. The heterojunction part in particular gives device-relevant numbers on permeability that could guide interface engineering for stability. The approach stays close to the raw trajectories, which avoids some fitting circularity. The topic matters for perovskite solar cells and LEDs where halide mixing and phase separation limit lifetime, so the trends on composition and interface effects are worth knowing. The soft spot is the lack of reported validation for the potentials on the exact quantities that matter: migration barriers for vacancies and interstitials in the mixed compositions, formation energies at the interfaces, or direct comparison of hop rates against longer DFT runs or experiment. Without those checks, the reported enhancement factors and the selective blocking could shift if the potentials have composition-dependent errors. Finite MD lengths also leave open whether the sampling captures the full statistics. This is useful for people who model ion transport in perovskites or design stable mixed-halide devices; they can take the qualitative trends as a starting point and test them further. It is worth sending to referees because the simulation setup is standard, the claims are falsifiable with additional validation data, and the interface results address a practical gap even if the numbers need tightening.

Referee Report

2 major / 2 minor

Summary. The manuscript uses molecular dynamics simulations driven by neural network potentials trained on DFT to investigate the diffusion of halide vacancies and interstitials in CsPb(IxBr1-x)3 mixed-halide perovskites and CsPbI3/CsPbBr3 heterojunctions. It reports enhanced diffusion rates for both defect types in the mixed compositions relative to the pure end-members, a mobility difference between Br and I ions within the mixed phase, and interface-structure-dependent permeability, with Br-rich interfaces blocking vacancy migration while I-rich interfaces remain permeable.

Significance. If the underlying potentials are shown to reproduce DFT barriers without composition-dependent bias, the work would provide mechanistic insight into the kinetics of halide mixing and phase separation that limit perovskite device stability. The use of ML potentials to reach microsecond-scale trajectories is a clear methodological strength for accessing rare defect hops.

major comments (2)

[Methods] Methods section: No quantitative validation metrics (e.g., MAE on migration barriers, formation energies, or phonon spectra) are reported for the neural network potentials on mixed-halide defect configurations or heterojunction models; the abstract states only that the potentials were 'trained on DFT calculations,' leaving open the possibility of systematic error in the relative rates claimed for mixed vs. pure compounds.
[Results] Results, diffusion-coefficient tables/figures: Reported diffusion constants lack error bars derived from multiple independent trajectories or block-averaging, and no convergence tests versus simulation length are shown; this weakens the quantitative claim of 'enhanced diffusion' in the mixed phase.

minor comments (2)

[Figures] Figure captions for the heterojunction trajectories should explicitly state the simulation temperature and total MD length used to generate the permeability observations.
[Abstract] The range of x values examined in CsPb(IxBr1-x)3 is not stated in the abstract or early results; this should be added for clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading of our manuscript and the constructive comments on validation and statistical robustness. We address each major comment below and will revise the manuscript to incorporate the suggested improvements.

read point-by-point responses

Referee: [Methods] Methods section: No quantitative validation metrics (e.g., MAE on migration barriers, formation energies, or phonon spectra) are reported for the neural network potentials on mixed-halide defect configurations or heterojunction models; the abstract states only that the potentials were 'trained on DFT calculations,' leaving open the possibility of systematic error in the relative rates claimed for mixed vs. pure compounds.

Authors: We agree that explicit quantitative validation metrics for the neural network potentials on mixed-halide defect configurations and heterojunction models were not reported in the original manuscript. Although the training dataset included diverse DFT configurations spanning pure and mixed compositions as well as defect-containing structures, we did not provide MAE values or similar metrics specifically for migration barriers and formation energies in those regimes. In the revised manuscript we will add a dedicated validation section reporting mean absolute errors on held-out test sets for migration barriers, formation energies, and phonon spectra, including mixed-halide and interface models. These metrics will be used to confirm the absence of composition-dependent bias in the relative diffusion rates. revision: yes
Referee: [Results] Results, diffusion-coefficient tables/figures: Reported diffusion constants lack error bars derived from multiple independent trajectories or block-averaging, and no convergence tests versus simulation length are shown; this weakens the quantitative claim of 'enhanced diffusion' in the mixed phase.

Authors: We accept that the absence of error bars and convergence tests limits the strength of the quantitative claims. The diffusion coefficients were extracted from long microsecond-scale trajectories, yet the original submission omitted statistical uncertainties and explicit checks against simulation length. In the revised manuscript we will include error bars obtained from multiple independent trajectories (using block averaging where appropriate) and add supplementary figures demonstrating convergence of the diffusion coefficients with increasing simulation time. These additions will support the reported enhancement in the mixed phase. revision: yes

Circularity Check

0 steps flagged

No significant circularity; results are direct MD outputs from trained NN potentials

full rationale

The paper's claims rest on molecular dynamics trajectories produced by neural network potentials trained on external DFT calculations. These trajectories generate the reported diffusion enhancements, mobility differences, and interface effects as emergent simulation outcomes. No self-definitional reductions, fitted inputs renamed as predictions, or load-bearing self-citations appear in the derivation chain. The methodology is self-contained against the DFT training data as an independent benchmark, yielding a low circularity score consistent with direct computational results rather than algebraic collapse to inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claims rest on the transferability of DFT-trained neural network potentials to mixed-halide compositions and heterojunctions, plus the assumption that MD sampling captures representative diffusion events without rare-event bias.

axioms (1)

domain assumption Neural network potentials trained on DFT reproduce the relevant potential energy surfaces for halide defect migration with sufficient accuracy for diffusion coefficient ordering.
Invoked implicitly in the choice of simulation method; no explicit validation metrics given in abstract.

pith-pipeline@v0.9.0 · 5438 in / 1363 out tokens · 26450 ms · 2026-05-16T13:10:46.653045+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We study the diffusion of halide vacancies and interstitials in CsPb(IxBr1-x)3 ... by molecular dynamics simulations using neural network potentials trained on density functional theory calculations.
IndisputableMonolith/Foundation/ArithmeticFromLogic.lean LogicNat_equivNat unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

The diffusion coefficients can be fitted with an Arrhenius expression D = D0 exp(-Ea/kBT).

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.