Hydrogen in Brownmillerite Perovskites: First-Principles Insights into Energetics and Induced Electronic-Magnetic Changes

Technical Summary: Hydrogen in Brownmillerite Perovskites

Problem Statement
Hydrogen uptake in oxygen-deficient brownmillerite perovskites ($A_2B_2O_5$) offers a chemically accessible route to tune functional properties for iono-electronic devices, sensors, and tunable spintronics. However, the microscopic mechanisms governing hydrogen incorporation, electron localization, and their coupled influence on electronic and magnetic structures remain insufficiently understood. Theoretical modeling of these systems is particularly challenging due to the vast configurational space of $H^+/e^-$ distributions, strong sensitivity to computational parameters (e.g., Hubbard $U$, exchange-correlation functionals), and the complex coupling between proton insertion, polaron formation, and magnetic order. Existing studies have largely focused on ionic transport, leaving a gap in the unified framework connecting hydrogen insertion to coupled electronic-magnetic transformations.

Methodology
The authors employ first-principles density functional theory (DFT) using the VASP package, primarily utilizing the meta-GGA SCAN(+U) functional, with benchmark comparisons at the PBEsol(+U) level. The study focuses on representative brownmillerites, specifically $Sr_2Fe_2O_5$ and $Sr_2Co_2O_5$, and extends to 14 experimentally reported $A_2B_2O_5$ compositions.

Key methodological features include:

• Configuration Sampling: Systematic evaluation of seven distinct interstitial sites ($H1-H7$) for proton placement and various transition-metal sites for electron localization (small polaron formation).
• Polaron Control: Explicit seeding of small-polaron formation via local bond elongation or site-selective Hubbard corrections to ensure the localized solution corresponds to a true energy minimum, avoiding artificial delocalization.
• Magnetic and Electronic Analysis: Calculation of magnetic exchange couplings ($J_1, J_2, J_3$) via mapping to a Heisenberg Hamiltonian, evaluation of magnetocrystalline anisotropy (MCA), and analysis of projected density of states (PDOS).
• Dynamics: Ab initio molecular dynamics (AIMD) simulations at 500 K to assess thermal effects on polaron hopping and band gap fluctuations.
• Machine Learning Benchmarking: Evaluation of universal machine-learning interatomic potentials (uMLIPs), specifically UMA (OMat24) and CHGNet, against DFT results for hydrogen absorption energies.

Key Contributions and Results

1. Coupled $H^+$–$e^-$ Mechanism and Localization:
   Hydrogenation introduces a localized electron that stabilizes near the $H^+$ as a small polaron. The preferred localization site is chemistry-dependent:

   • In $Sr_2Co_2O_5$, the polaron favors octahedral Co sites due to crystal-field splitting ($t_{2g}$ occupation).
   • In $Sr_2Fe_2O_5$, the polaron strongly localizes on tetrahedral Fe sites ($e_g$ preference).
   • The energy difference between localization sites can reach 0.31 eV, and the choice of localization site significantly impacts absorption energies.

2. Sensitivity to Computational Parameters:
   The study quantifies that hydrogen absorption energies are highly sensitive to:

   • Magnetic Order: Ferromagnetic (FM) states consistently yield lower absorption energies (by 0.2–0.3 eV) compared to antiferromagnetic (AFM) states for $Sr_2Fe_2O_5$, attributed to the metallic nature of the FM ground state favoring absorption.
   • Hubbard $U$ and Functionals: Variations in $U$ values and the choice between PBEsol and SCAN can shift absorption energies by up to 1 eV. SCAN+U is found to be less sensitive to $U$ variations than PBEsol+U.
   • Site Selection: Differences in proton site identification can introduce errors up to 0.6 eV.

3. Electronic and Magnetic Transformations:

   • Electronic Structure: Dilute hydrogen absorption creates localized in-gap states without inducing a full insulator-to-metal transition in $Sr_2Co_2O_5$ and $Sr_2Fe_2O_5$. However, AIMD simulations reveal that thermal fluctuations and polaron hopping reduce the effective band gap by ~0.5 eV.
   • Magnetic Properties: Hydrogenation weakens antiferromagnetic exchange couplings ($J_1, J_2$) and induces canting of the Néel vector (4.5° in $Sr_2Fe_2O_5$, 3.3° in $Sr_2Co_2O_5$), resulting in weak ferromagnetism. This is driven by the competition between superexchange and double-exchange mechanisms arising from mixed-valence configurations (e.g., $Fe^{3+}/Fe^{2+}$).

4. High-Throughput Screening and Descriptors:
   Screening 14 compositions reveals a strong negative correlation ($R = -0.88$) between the B-site d-electron count and hydrogen absorption energy; higher d-electron counts generally favor absorption. Favorable absorption is also linked to structural descriptors: larger O–O separations and lattice flexibility (soft phonon modes) that accommodate $H^+$-induced distortions. Candidates such as $Y_2Cu_2O_5$, $Sr_2Bi_2O_5$, and $Ba_2Tl_2O_5$ are identified as promising for experimental investigation.

5. Limitations of Machine Learning Potentials:
   Benchmarking of uMLIPs (UMA and CHGNet) against DFT shows mean absolute errors (MAE) of ~0.9–1.05 eV. The models fail to reliably predict site-resolved energetics or the correct ordering of stable sites. The authors attribute this to the lack of explicit training on proton-polaron pairs and magnetic states, indicating that current uMLIPs are not yet suitable for blind high-throughput screening of hydrogen intercalation in correlated oxides.

Significance and Claims
The paper establishes practical design rules and a consistent computational workflow for modeling hydrogen-responsive oxides. It emphasizes that accurate prediction requires the explicit treatment of coupled proton-electron-spin degrees of freedom, including the correct identification of magnetic ground states and polaron localization sites. The authors claim that ignoring these factors can lead to errors of up to 1 eV in absorption energies, rendering high-throughput studies unreliable without rigorous validation.

The work provides a framework for identifying materials with favorable hydrogen uptake based on B-site d-electron count and local structural flexibility (O–O separations). Furthermore, it highlights the risk of "hidden" hydrogen incorporation from ambient moisture in experimental settings, which could confound interpretations of catalytic, resistive switching, or superconducting properties. Finally, the study cautions against the uncritical application of current machine-learning potentials for this specific class of problems, advocating instead for descriptor-based surrogate models or targeted DFT validation.