Machine learning (ML) models for electronic structure typically rely on large datasets generated by computationally expensive Kohn-Sham density functional theory calculations, as it is not known a priori which portions of the data are essential for accurate learning. Here, we reveal significant redundancies in electronic structure datasets across diverse material systems and attribute them to the low intrinsic dimensionality of the underlying data. We show that even random pruning can substantially reduce dataset size with minimal degradation in predictive accuracy. Moreover, a state-of-the-art coverage-based pruning strategy that samples data across all learning difficulties almost always preserves chemical accuracy and maintains model generalizability while using up to two orders of magnitude less data and reducing training time by a factor of three or more. We further demonstrate that the essential electronic structure information lies on a low-dimensional, non-linear manifold, providing a potential geometric explanation for the observed prunability. These observations are consistent with the predominance of local atomic environments in determining electronic properties, as suggested by nearsightedness arguments, and indicate that large-scale datasets may contain highly overlapping information. Our findings challenge the prevailing assumption that such extensive datasets are necessary for accurate ML-based electronic structure predictions and open a path toward identifying minimal, representative datasets for each material class.
Disordered porous materials offer broad mechanical-response spaces, but their non-periodic three-dimensional architectures make inverse design ill-posed: structurally distinct configurations can produce similar stress-strain curves, while scalar descriptors such as porosity cannot resolve the underlying topology. Here, we introduce a response-function-conditioned inverse design framework that learns directly from XCT-derived closed-cell aluminium-foam architectures and targets complete nonlinear stress-strain responses rather than individual properties. A point-cloud encoder and forward surrogate establish the structure-response relation, while conditional generation and gradient-based refinement search the learned structural space for manufacturable candidates. Trained on 844 XCT-derived subvolumes spanning porosities of 0.4034–0.8935, the forward model achieves a pointwise R² of 0.9918 and a median curve-level NRMSE of 3.015% on a parent-specimen-separated test set. For targets admitting multiple accepted solutions, inverse-designed candidates retain a median response NRMSE of 0.0358 while exhibiting a median pairwise structural distance of 0.2328, demonstrating non-trivial one-to-many design. Finite-element analyses and compression tests on additively manufactured architectures across three aluminium-alloy systems further confirm physical realizability. These results establish full-response conditioning as a practical route for designing real disordered porous materials beyond predefined unit cells.
Altermagnetism (AM) and Luttinger-compensated ferrimagnetism (LC-FiM) both combine zero net magnetization with spin-split electronic bands, but realizing both states and their interconversion within a single platform remains a significant challenge. Using first-principles calculations and symmetry analysis, we show that a graphene-MnPS3 van der Waals (vdW) heterostructure provides exactly such a platform. The graphene layer breaks the $${{\mathcal{P}}}{{\mathcal{T}}}$$ symmetry of Néel-antiferromagnetic MnPS3, and the interfacial stacking dictates the symmetry relation between the two spin sublattices: hollow stacking preserves joint rotation and mirror operations, yielding an i-wave AM state with momentum-dependent spin splitting, while top stacking removes all sublattice-connecting symmetries and stabilizes an LC-FiM state with globally spin-split bands. The two phases are separated by an ultralow sliding barrier of 28 meV/f.u., allowing reversible switching by interlayer translation alone. Out-of-plane (OOP) pressure further amplifies the exchange spin splitting to 72.2 meV (AM) and 80.2 meV (LC-FiM), arising from strengthened interfacial coupling accompanied by enhanced orbital hybridization and asymmetric built-in polarization. Beyond the ideal-stacking limit, the intrinsic lattice mismatch generates a moiré superlattice of alternating AM and LC-FiM domains with distinct magneto-optical Kerr signatures. These results establish interfacial symmetry engineering as a general route to programmable compensated magnetism in vdW materials.
The pursuit of high operating voltage and energy density of battery packs imposes stringent requirements on the electronic insulation of solid-electrolyte interphases (SEI) in rechargeable lithium-ion batteries. Owing to complicated kinetic processes, direct experimental characterization of electronic insulation poses challenges, whereas first-principles computations provide valuable access to underlying microscopic parameters. However, existing methods often yield conflicting evaluations by focusing solely on electronic transmission at specific parts of SEI/electrolyte interfaces, ignoring electrolyte decomposition kinetics. Herein, we develop a computational framework combining the Marcus electron transfer model and constrained density functional theory to predict the resistance to electrolyte decomposition (RED), which quantifies how effectively the SEI suppresses decomposition. RED unifies two key physical factors: electronic transmission across interfaces and reorganization energy associated with electrolyte decomposition, thereby enabling component‑resolved, electrolyte‑dependent evaluation of SEI electronic insulation. Applying the framework to idealized crystalline graphitic C(0001)|SEI | electrolyte interfaces, we find that electrolyte-dependent reorganization can substantially alter, and in some cases reverse, the insulation ranking inferred from electronic coupling alone, identifying reorganization as a key contributor to the comparatively high electronic insulation of Li2CO3 in the systems examined. The present work establishes a proof-of-concept, component-resolved framework and a basis for future studies of structurally heterogeneous SEIs.
Abstract Space-charge layers (SCLs) at heterogeneous interfaces generate electrostatic potential drops that can enhance or impede ionic and electronic transport near the interface. In solid-state Li-ion batteries, previous models typically only consider Li-ions as the mobile defect, even though cations, anions, and electrons can all participate in SCL formation. Here, we introduce a general first-principles defect-based framework to quantify these multi-species contributions and apply it to LiPON/Li x V2O5 interfaces (x = 0, 1, 2 for the alpha, delta, and gamma phases). A systematic screening of charged point defect formation energies identified Li-ion vacancies, Li-ion interstitials, and O-ion vacancies as dominant species. The model reveals spontaneous Li-ion transfer from LiPON into Li x V2O5 and O-species transfer in the opposite direction, behaviors missed by Li-only treatments. Band alignments across the interfaces show electrons initially transferring from LiPON into V2O5, however, when the lithium content increases in Li x V2O5, the direction of electron transfer reverses. Consequently, the predicted band-bending switches from downward bending in V2O5 to upward bending in the LiV2O5 and Li2V2O5 phases. The implications of these SCL-induced potentials for Li-ion and electron transport are discussed for solid-state batteries and electrochemical neuromorphic devices, highlighting the importance of explicitly accounting for multi-ion defect transfer.
