AI for Science

Large language models (LLMs) have shown growing promise in biomedical research, particularly for knowledge-driven interpretation tasks. However, their ability to reliably reason from gene-level knowledge to functional understanding, a core requirement for knowledge-enhanced cell atlas interpretation, remains largely underexplored. To address this gap, we introduce SciHorizon-GENE, a large-scale gene-centric benchmark constructed from authoritative biological databases. The benchmark integrates curated knowledge for over 190K human genes and comprises more than 540K questions covering diverse gene-to-function reasoning scenarios relevant to cell type annotation, functional interpretation, and mechanism-oriented analysis. Motivated by behavioral patterns observed in preliminary examinations, SciHorizon-GENE evaluates LLMs along four biologically critical perspectives: research attention sensitivity, hallucination tendency, answer completeness, and literature influence, explicitly targeting failure modes that limit the safe adoption of LLMs in biological interpretation pipelines. We systematically evaluate a wide range of state-of-the-art general-purpose and biomedical LLMs, revealing substantial heterogeneity in gene-level reasoning capabilities and persistent challenges in generating faithful, complete, and literature-grounded functional interpretations. Our benchmark establishes a systematic foundation for analyzing LLM behavior at the gene scale and offers insights for model selection and development, with direct relevance to knowledge-enhanced biological interpretation.

Correct identification of fish species is highly significant for food security, economic development, and climate resilience in Bangladesh. Protein sequences directly reflect functional and evolutionary constraints which are important for species authentication and biodiversity monitoring. Yet there exists no benchmark for native Bangladeshi fish species identification from protein sequence. In this study, we addressed this gap by introducing the first curated dataset for nine native Bangladeshi fish species of 2845 high quality protein sequences. We also established the first protein sequence classification baseline for this domain through a systematic benchmarking of seven architectural paradigms. Moreover, we propose a realistic deployable novel hybrid architecture of MotifCNN and Transformer with Terminal-Aware Positional-Encoding (MotifCNN-Transformer+TA-PE). Our novel architecture achieves 79.80% accuracy with macro-F1 of 0.80. The highest 83.04% accuracy is achieved by finetuned protein language model ProtBERT that has 420M parameters and requires dual 16GB GPUs for inference. According to McNemar's test, ProtBERT's 3.24% accuracy gain over our MotifCNN-Transformer+TA-PE is statistically insignificant (p = 0.1120). Our novel architecture beats it among six of the nine classes in per class identification. Also our MotifCNN-Transformer+TA-PE is approximately 5x faster, 42x smaller, and supports 16x larger batch size than ProtBERT and has GPU free inference, making it more practical for deployment in resources constrained areas such as rural Bangladesh. Beyond this, our foundational work shows effects of phylogenetic relationships on sequence similarity and establishes pathways for fisheries management, food authentication and biodiversity conservation in South Asia's protein dependent economy.

Accurate predictions of exfoliation energies and lattice constants of layered materials hinge on a correct description of London dispersion physics. Modern a posteriori dispersion corrections in density-functional theory (DFT), such as the exchange-hole dipole moment (XDM) model, capture the proper asymptotic behaviour at long range while making use of damping functions to prevent unphysical divergence at short range. In the united-atom limit, the dispersion energy is damped to a finite, non-zero value by both the canonical Becke--Johnson (BJ) damping function and the new Z-damping function. XDM(BJ) has previously demonstrated exceptional accuracy for modelling layered materials, such as in the LM26 benchmark, which includes graphite, hexagonal boron nitride, lead(II) oxide, and transition-metal dichalcogenides. This work presents the first assessment of XDM(Z) on the same benchmark. We also show that inclusion of three-body interactions via the Axilrod--Teller--Muto (ATM) term further improves the computed exfoliation energies for both XDM(BJ) and XDM(Z), yielding the best performance achieved on LM26 using semi-local functionals to date, relative to reference data from the random-phase approximation.

Density-functional theory (DFT) has become the workhorse of modern computational chemistry, with dispersion corrections such as the exchange-hole dipole moment (XDM) model playing a key role in high-accuracy modelling of large-scale systems. All previous production implementations of XDM have used the two-parameter Becke--Johnson damping function based on atomic radii. Here, we introduce and implement a new XDM variant that uses a one-parameter damping function based on atomic numbers, recently proposed by Becke. Both this new Z damping and the canonical BJ-damping variants of XDM are benchmarked on the comprehensive GMTKN55 database using minimally empirical generalised-gradient-approximation, global hybrid, and range-separated hybrid functionals. This marks the first time that the XDM (and many-body dispersion, MBD) corrections have been tested on the GMTKN55 set. Using the new WTMAD-4 metric, an outlier analysis is performed for all new data, as well as for top-ranking functionals from the literature at each rung, providing insight into both performance and consistency across the dataset. We also extended our analysis to the DM21 and Skala machine-learned functionals that have garnered recent attention. To test Z damping's transferability to the solid state, four benchmarks involving molecular crystals are also considered. Across these molecular and solid-state benchmarks, the revPBE0 and B86bPBE0 hybrid functionals, paired with the Z-damped XDM variant, show excellent performance.

Developing intrinsic homojunctions without chemical heterogeneity remains a key challenge in future two - dimensional devices. Here, we report a geometry - defined metal--semiconductor--metal homojunction in bilayer blue phosphorus (BlueP) created by a localized bubble corrugation, without chemical doping or foreign - material interfaces. First - principles calculations show that enlarging the interlayer separation in the metallic A\(_1\)B\(_1\) - stacked BlueP bilayer opens a band gap, enabling a semiconducting barrier embedded between metallic segments. First - principles quantum - transport simulations reveal a crossover from ballistic to tunneling transport upon bubble formation. In the tunneling regime, transmission decreases exponentially with bubble width while remaining weakly sensitive to bubble height and bulging direction. The junction acts as an orientation - dependent \(k\) - space filter, producing transport anisotropy and momentum selectivity. Orbital - resolved scattering analysis shows that intralayer - bonding channels persist under deformation whereas interlayer - hybridized channels are quenched, and that σ- type bonding yields higher conductance than π- type bonding. These insights motivate two electromechanical device concepts: a mechanically switchable memory element with ON/OFF ratios up to 30 and a nanoscale sliding rheostat with reproducible exponential resistance tuning for Å- scale displacement sensing.

DNA nanostar (DNAns) hydrogels are promising materials for \textit{in vivo} applications, including tissue regeneration and drug and antibody delivery. However, a systematic and quantitative understanding of the design principles controlling their degradation is lacking. Here, we investigate hydrogels made of three-armed DNAns with varying flexible joints, arm lengths, and mesh sizes and use restriction enzymes (RE) to cut the DNAns structures while monitoring the gel's degradation. We discover that (i) removing flexible joints, (ii) increasing arm length, or (iii) relocating the RE site along a DNA linker markedly accelerates hydrogel degradation. In contrast, non-specific endonucleases, e.g. DNaseI, quickly degrade DNAns hydrogels regardless of design. Importantly, the release of antibodies from DNAns hydrogels can be modulated by the action of sequence-specific enzymes, confirming that programmable degradation can be leveraged for responsive drug-delivery systems. These findings provide a better understanding of the design principles for DNAns-based scaffolds with tunable degradation, cargo release, and responsive rheology.

Indium selenide (In2Se3), a layered chalcogenide with multiple polymorphs, is a promising material for optoelectronic and ferroelectric applications. However, achieving polymorph-pure thin films remains a major challenge due to the complex growth space. In this work, Bayesian optimization (BO) is successfully leveraged to guide the molecular beam epitaxy growth of In2Se3 on Al2O3 substrates. By training a predictive Gaussian process regressor with sequential learning, we efficiently explored substrate temperature, indium flux, selenium flux, and cracker temperature, reducing experimental trials required for successful synthesis. A γ-In2Se3 film with 91% phase purity was achieved in fewer than 10 BO run samples. Attempts to isolate α-In2Se3 were limited by amorphous film formation at low temperatures, indicating that single-step codeposition is unsuitable for crystalline α-In2Se3 on Al2O3. Overall, this study validates BO as a powerful approach for phase-selective growth in complex material systems.

Gallium oxide is an emerging material of interest due to its unique combination of functional properties and the existence of multiple polymorphs - α, β, γ, δ, and κ - each exhibiting distinct characteristics arising from their different lattice symmetries. Optical properties are particularly important, as they determine potential device applications and enable phase identification. However, direct comparison of optical signatures, including key parameters such as bandgaps, is hindered by inconsistent, sparse, or even missing data in the literature. To address this issue, in the present work we systematically cross-correlate optical emission and absorption features of α, β, γ, δ, and κ thin films, as well as differently oriented β-phase bulk crystals and γ/β double polymorph structures. We demonstrate that optical bandgaps and emission features scale consistently across the polymorphs when methodological uncertainties are minimized by applying identical experimental conditions and unified analysis procedures to a structurally similar set of thin film samples. In addition, we extend conventional far field optical phase identification to the nanoscale by reporting near field optical signatures of Ga2O3 polymorphs via nano FTIR. Overall, the present dataset provides a comprehensive reference of near- and far-field optical polymorph signatures to support ongoing multidisciplinary research on Ga2O3.

The compaction mechanisms of SiO2 glass under pressure include under certain conditions a specific reduction of the elastic moduli and a complex inelastic behavior whose nature is not yet fully understood. In our work we establish a variational framework describing the evolution of SiO2 glass under hydrostatic pressure. Based on a previous work that presents a model for multi-phase transformations in inelastic materials, we assume isothermal conditions during a compaction process and interpret the typical sigmoidal stress response as indicator of a binary phase transformation. During the process, two volume fractions coexist macroscopically and microstructures such as shear bands or disclination pattern develop in between. We restrict our approach to resolve only the volume fractions, not the corresponding microstructures. Nevertheless, the resulting model is shown to match experimental findings very well. Numerical examples successfully illustrate the relationship between the changes in the elastic moduli and the corresponding change in volume with respect to pressure.

We present PEPSKit$.$jl, a Julia package for simulating two-dimensional quantum many-body systems with infinite projected entangled-pair states (iPEPS). PEPSKit$.$jl builds on the TensorKit$.$jl package for tensor computations and provides high-level algorithms for iPEPS simulations that support both Abelian and non-Abelian symmetries, as well as fermionic systems. This work gives an overview of the main package features, which include support for ground-state, time-evolution, and finite-temperature simulations in systems with different physical symmetries and lattice geometries. These capabilities are illustrated through various examples and technical benchmarks.

Physics-informed neural networks (PINNs) often become difficult to optimize for incompressible flow problems with large spatial domains, multiscale stresses, or long-time invariant dynamics. We propose an invariant-guided PINN (IG-PINN) framework that uses partitioned training as a conservative preconditioning stage rather than as the final piecewise representation. A globally defined architecture is trained successively on spatial subdomains or temporal slabs; selected field traces, structural information, and conservative diagnostics are then transferred to a final global correction, yielding a single neural field on the full spatial or space-time domain. The framework is tested on two incompressible flow problems: steady Oldroyd--B flow past a confined cylinder and a rotational Newtonian flow with helicity diagnostics. In the Oldroyd--B case, IG-PINN transfers velocity, polymeric stress, and mass-flux information while avoiding pressure traces at artificial interfaces. In the helicity case, endpoint velocity is transferred through a hard temporal constraint and kinetic energy is controlled during slab training and residual global correction. The experiments demonstrate improved optimization robustness, reduced conservation errors for the cylinder wake, and controlled energy and helicity diagnostics for the transient rotational flow.

We propose an algorithm for the computational homogenization of locally periodic hyperelastic structures undergoing large deformations due to external quasi-static loading. The algorithm performs clustering of macroscopic deformations into subsets called "centroids", and, as a new ingredient, approximates the homogenized coefficients using sensitivity analysis of micro-configurations with respect to the macroscopic deformation. The novel "model-order reduction" approach significantly reduces the number of microscopic problems that must be solved in nonlinear simulations, thereby accelerating the overall computational process. The degree of reduction can be controlled by a user-defined error tolerance parameter. The algorithm is implemented in the finite element framework SfePy, and its performance effectiveness is demonstrated using two-dimensional test examples, when compared with solutions obtained by the proper orthogonal decomposition method, and by the full "FE-square" simulations. Extensions beyond the present implementations and the scope of tractable problems are discussed.

Nonreciprocal interactions that violate Newton's law 'actio=reactio' are ubiquitous in nature and are currently intensively investigated in active matter, chemical reaction networks, population dynamics, and many other fields. An outstanding challenge is the thermodynamically consistent formulation of the underlying stochastic dynamics that obeys local detailed balance and allows for a rigorous analysis of the stochastic thermodynamics of non-reciprocally interacting particles. Here, we present such a framework for a broad class of active systems and derive by systematic coarse-graining exact expressions for the macroscopic entropy production. Four independent contributions to the thermodynamic dissipation can be identified, among which the energy flux sustaining vorticity currents manifests the presence of non-reciprocal interactions. Then, Onsager's non-reciprocal relations, the fluctuation-response relation, the fluctuation relation and the thermodynamic uncertainty relations for non-reciprocal systems are derived. Finally, we demonstrate that our general framework is applicable to a plethora of active matter systems and chemical reaction networks and opens new paths to understand the stochastic thermodynamics of non-reciprocally interacting many-body systems.

We introduce SoliDualSPHysics, a novel open-source and GPU-accelerated software that extends DualSPHysics to enable the numerical simulation of hyperelastic, finite-strain plastic, and brittle fracture behavior in deformable solids within a unified smoothed particle hydrodynamics (SPH) software framework. The software implements a total Lagrangian formulation for solid mechanics that allows direct application of external loads and boundary conditions, enabling independent solid mechanics simulations. Brittle fracture is modeled through a phase-field approach coupled with SPH, allowing crack initiation, propagation, and branching under dynamic loading without explicit crack tracking, ad hoc crack-path criteria, or local refinement. The framework also supports user-defined mathematical expressions to prescribe time- and space-dependent quantities, complementing the solid and fracture extensions and enhancing flexibility across existing and future DualSPHysics applications. Leveraging DualSPHysics' native CPU/GPU parallel architecture, the software achieves substantial computational acceleration for large-scale simulations, and the implementation is verified and validated against benchmark numerical problems and experimental data, demonstrating accuracy, robustness, and favorable scaling performance. Comprehensive implementation details and user documentation are provided to ensure reproducibility and to support further development by the community. The framework and source code are freely available through a public GitHub repository.

In the present paper, we construct a theoretical description of a three-dimensional photonic structure in the form of a tetragonal lattice of bianisotropic resonators applying a dyadic Green's function approach. By representing the resonators as point electric and magnetic dipoles, we obtain the Bloch Hamiltonians for the approximations considering the interactions between the nearest, next-nearest, and next-to-next-nearest resonators, and construct the corresponding real-space tight-binding models. We analyze the band diagrams, spatial structure of the eigenmodes, and their localization, revealing quadratic degeneracies in the vicinity of high-symmetry points in the absence of bianisotropy and the emergence of in-gap states localized at a domain wall upon the introduction of bianisotropy. Finally, we compare the theoretical results with full-wave numerical simulations for an array of bianisotropic resonators.

Using mean-field theory, we determine the electronic phase diagram of undoped AB-stacked bilayer graphene in the presence of a transverse electric field. In addition to multiple competing electronic instabilities characterized by excitonic order parameters, our framework incorporates the long-range Coulomb energy associated with interlayer polarization. This long-range interaction plays a crucial role, as it significantly influences both the structure and the relative energies of the competing ordered states. We derive a set of self-consistency equations and solve them both numerically and analytically. Our findings reveal that, as the bias field is varied, the bilayer undergoes a cascade of first-order transitions between several ordered insulating phases for which order-parameter structures are explicitly identified. Some of these phases are characterized by two inequivalent single-particle gaps, whose magnitudes depend on the valley and spin quantum numbers. Field-driven transitions are accompanied by discontinuous and non-monotonic variations of the single-electron gap. We relate our results to Hartree-Fock numerical calculations and to experimental research, including observations of fractional metallic phases that emerge upon doping the bilayer system.

In this work, we employ a field-theoretic renormalization group approach to study a paradigmatic model of directed percolation. We focus on the perturbative calculation of the equation of state, extending the analysis to the three-loop order in the expansion parameter $\varepsilon = 4-d$. We show that a large group of the necessary three-loop Feynman diagrams can be mapped onto already existing three-loop results, and develop a technique for the calculation of the remaining -- truly novel -- ones. The described semi-analytic procedure is further used to verify existing two-loop results. The main aim of this study is to provide an update on this ongoing work, as full three-loop calculations utilizing the described procedure are in progress.

We study the quantum field theory of zero temperature perfect fluids. Such systems are defined by quantizing a classical field theory of scalar fields $ϕ^I$ that act as Lagrange coordinates on an internal spatial manifold of fluid configurations. Invariance under volume preserving diffeomorphisms acting on these scalars implies that the long-wavelength spectrum contains vortex (transverse modes) with an exact $ω_T=0$ dispersion relation. As a consequence, physically interpreting the results obtained via perturbative quantization of this theory has proven to be challenging. In this paper, we show that correlators evaluated in a class of semi-classical (Gaussian) initial states prepared at $t=0$ are well-defined and accessible via perturbation theory. The width of the initial state effectively acts as an infrared regulator without explicitly breaking diffeomorphism invariance of the classical action. As an application, we compute the stress tensor two-point correlators and show that vortex modes give a non-trivial contribution to the response function, non-local in both space and time.

In this paper we present a newly developed GRPIC code framework called FPIC, providing a detailed description of the Maxwell-equations solver, of the particle ``pushers'', and of the other algorithms that are needed in this approach. We describe in detail the code, which is written in Fortran and exploits parallel architectures using MPI directives both for the fields and particles. FPIC adopts spherical Kerr-Schild coordinates, which encode the overall spherical topology of the problem while remaining regular at the event horizon. The Maxwell equations are evolved using a finite-difference time-domain solver with a leapfrog scheme, while multiple particle ``pushers'' are implemented for the evolution of the particles. In addition to well-known algorithms, we introduce a novel hybrid method that dynamically switches between the most appropriate scheme based on the violation of the Hamiltonian energy. We first present results for neutral particles orbiting around black holes, both in the Schwarzschild and Kerr metrics, monitoring the evolution of the Hamiltonian error across different integration schemes. We apply our hybrid approach, showing that it is capable of achieving improved energy conservation at reduced computational cost. We apply FPIC to investigate the Wald solution, first in electrovacuum and subsequently in plasma-filled configurations. In the latter case, particles with negative energy at infinity are present inside the ergosphere, indicating that the Penrose process is active. Finally, we present the split-monopole solution in a plasma-filled environment and successfully reproduce the Blandford-Znajek luminosity, finding very good agreement with analytical predictions.

This paper investigates density driven flow in porous media, focusing on the roles of viscosity contrast, density contrast, and linear adsorption. In this setup, the fluid on top is heavier and more viscous than the fluid below. Under the effect of gravity, this system becomes unstable, and finger-like structures appear. The phenomenon is described mathematically by coupling Darcy's law with a convection-diffusion reaction equation. The nonlinearity in this model arises mainly from the concentration dependence of viscosity and the convective transport term. The existence of a unique pair of weak solutions is shown using the Galerkin approximation method and truncation technique. Moreover, an application of the maximum principle shows non-negativity of the concentration. Additionally, we analyze the long-time behavior of the solution and prove that the concentration converges exponentially to zero in the $L^p$-norm for all $1 \le p \le \infty$ as $t \to \infty.$ To complement the theoretical analysis, we perform numerical simulations based on a pressure formulation. By tracking total kinetic energy and mixing measures over time, we discuss the instability and the mixing efficiency, respectively. The present study reveals that although increasing the density contrast amplifies the total kinetic energy, the marginal impact diminishes with successive increments of density contrast. Similarly, while adsorption acts to suppress mixing, its efficiency in doing so tends to saturate with further increases. These behavior are consistent with the numerical simulations.

We present an attention-based foundation model architecture for learning and predicting quantum states across Hamiltonian parameters, system sizes, and physical systems. Using only basis configurations and physical parameters as inputs, our trained neural network is able to produce highly accurate ground state wavefunctions. For example, we build the phase diagram for the 2D square-lattice $t-V$ model with $N$ particles, from only 18 parameters $(V/t,N)$. Thus, our architecture provides a basis for building a universal foundation model for quantum matter.

For a gas-solid interfacial system where chemical species undergo reversible adsorption, we develop a mesoscopic stochastic modeling method that simulates both gas-phase hydrodynamics and surface coverage dynamics by coupling the Langmuir adsorption model with compressible fluctuating hydrodynamics. To this end, we derive a thermodynamically consistent mass-energy update scheme that accounts for how the mass and energy variables in the gas and surface subsystems should be updated according to the changes in the number of molecules of each species in each subsystem due to adsorption and desorption events. By performing a stochastic analysis for the ideal Langmuir model and the full hydrodynamic system, we analytically confirm that our mass-energy update scheme captures thermodynamic equilibrium predicted by equilibrium statistical mechanics. We find that an internal energy correction term is needed, which is attributed to the difference in the mean kinetic energy of gas molecules colliding with the surface from that computed from the Maxwell-Boltzmann distribution. By performing an equilibrium simulation study for an ideal gas mixture of CO and Ar with CO undergoing reversible adsorption, we validate our overall simulation method and implementation.

It is essential that mesoscopic simulations of reactive systems reproduce the correct statistical distributions at thermodynamic equilibrium. By considering a compressible fluctuating hydrodynamics (FHD) simulation method of ideal gas mixtures undergoing reversible reactions described by the chemical Langevin equations, we show that thermodynamic consistency in reaction rates and the use of instantaneous temperatures for the evaluation of reaction rates is required for fluctuations for the overall system to be correct. We then formulate the required properties of a thermodynamically-consistent reaction (TCR) model. As noted in the literature, while reactions are often discussed in terms of forward and reverse rates, these rates should not be modeled independently because they must be compatible with thermodynamic equilibrium for the system. Using a simple TCR model where each chemical species has constant heat capacity, we derive the explicit condition that the forward and reverse reaction rate constants must satisfy in order for the system to be thermodynamically consistent. We perform equilibrium and non-equilibrium simulations of ideal gas mixtures undergoing a reversible dimerization reaction to measure the fluctuational behavior of the system numerically. We confirm that FHD simulations with the TCR model give the correct static structure factor of equilibrium fluctuations. For the statistically steady simulation of a gas mixture between two isothermal walls with different temperatures, we show using the TCR model that the temperature variance agrees with the corresponding thermodynamic-equilibrium temperature variance in the interior of the system, whereas noticeable deviations are present in regions near walls, where chemistry is far from equilibrium.

Violation of local realism via Bell inequality - a profound and counterintuitive manifestation of quantum theory that conflicts with the prediction of local realism - is viewed to be intimately linked with quantum entanglement. Experimental demonstrations of such a phenomenon using quantum entangled states are among the landmark experiments of modern physics and paved the way for quantum technology. Here we report the violation of the Bell inequality that cannot be described by quantum entanglement in the system but arises from quantum indistinguishability by path identity, shown by the multi-photon frustrated interference. By analyzing the measurement of four-photon frustrated interference within the standard Bell-test formalism, we find a violation of Bell inequality by more than four standard deviations. Our work establishes a connection between quantum correlation and quantum indistinguishability, providing insights into the fundamental origin of the counterintuitive characteristics observed in quantum physics.

Resonances associated with complex-frequency poles are ubiquitous across physics and can arise in any open system, ranging from subwavelength particles and cavities to biological structures. When two such resonances coalesce, they form exceptional points (EPs), non-Hermitian singularities known to produce unusual spectral and dynamical behavior. However, the dynamics of the response of such resonances and exceptional points to complex frequency drive remained largely unexplored. Here, we experimentally observe the temporal response of complex-frequency resonances and theoretically study this for exceptional points. We unveil a universal transient phenomenon of open cavities driven at complex frequencies: the system's initial response grows linearly, with enhanced growth at exceptional points (EPs), even though the system is passive and the excitation decays. Closed-form theory for general resonators, extended to higher-order modes, predicts efficient power transfer with $t$ and $t^2$ scaling for complex single poles and exceptional points (EPs), respectively, at all times. We demonstrate these effects in subwavelength optical scatterers and experimentally in an electrical circuit analogue, with excellent agreement, and explore configurations that capture EP-enhanced growth.

Equivariant network architectures are a well-established tool for predicting invariant or equivariant quantities. However, almost all learning problems considered in this context feature a global symmetry, i.e. each point of the underlying space is transformed with the same group element, as opposed to a local ``gauge'' symmetry, where each point is transformed with a different group element, exponentially enlarging the size of the symmetry group. Gauge equivariant networks have so far mainly been applied to problems in quantum chromodynamics. Here, we introduce a novel application domain for gauge-equivariant networks in the theory of topological condensed matter physics. We use gauge equivariant networks to predict topological invariants (Chern numbers) of multiband topological insulators. The gauge symmetry of the network guarantees that the predicted quantity is a topological invariant. We introduce a novel gauge equivariant normalization layer to stabilize the training and prove a universal approximation theorem for our setup. We train on samples with trivial Chern number only but show that our models generalize to samples with non-trivial Chern number. We provide various ablations of our setup. Our code is available at https://github.com/sitronsea/GENet/tree/main.

Uncovering the true informational architecture of real-world complex systems requires disentangling how their components uniquely store, redundantly share, and synergistically integrate information over time. Integrated Information Decomposition ($Φ$ID) is a framework for decomposing the information dynamics of multivariate systems into sixteen non-overlapping atoms that characterize redundant, unique, and synergistic modes of information storage, transfer, and integration. Existing methods to compute $Φ$ID are restricted to Gaussian or discrete systems, preventing its application to continuous non-Gaussian dynamical systems. We address this limitation by proposing DIPHINE (Diffusion-based $Φ$-ID Neural Estimator), the first neural estimator that leverages score-based diffusion models to jointly estimate all the mutual information terms required by $Φ$ID from a single amortized network, recovering the sixteen atoms through Möbius inversion. We provide a theoretical analysis of error propagation through the inversion, showing that the Jacobian of the mapping from mutual informations to atoms is integer-valued and that the synergy-to-synergy atom is provably the hardest to estimate. We demonstrate accurate recovery of ground-truth atoms on synthetic benchmarks, superior performance compared to established mutual information estimators, and the ability to extract physiologically interpretable information-dynamic structure on an application involving real data without any distributional assumptions.

Manually curated biomedical repositories -- spanning bioactivity, genomics, and chemistry -- are expensive to maintain, lag behind primary literature, and discard experimental context, obscuring nuances needed to assess data correctness and coverage. We show that PubMed itself can be autonomously and cost-effectively turned into structured datasets that are larger, more nuanced, and more accurate than the curated databases they replace. We present three coupled contributions: (1) an LLM-based entity-tagging pipeline, grounded in nine biomedical ontologies, that tags 4.5B entities across 19 categories in a 22.5M-paper, 2.5T-token PubMed corpus; (2) hybrid sparse-dense retrieval supporting entity-filtered semantic queries over the tagged corpus; and (3) Starling, a multi-agent deep research system that, given only a natural-language task description, designs precision- and recall-targeted retrieval filters, induces an extraction schema, and emits structured records with nuance-rich fields and supporting passages. Across six tasks -- blood-brain barrier permeability, oral bioavailability, acute toxicity (LD50), gene-disease associations, protein subcellular localization, and chemical reactions -- Starling produces ~6.3M records (91K-3M per task); several are, to our knowledge, the largest public datasets for their property. Frontier-model rejection of our extractions is 0.6-7.7% across tasks, far below error rates we measure on widely used curated counterparts (e.g., 16.5% on BBB_Martins, 7.3% on Bioavailability_Ma). Beyond scale and accuracy, the supporting passages carry nuance tabular databases discard -- e.g., oral bioavailability may depend on fed vs. fasted state. Together, the corpus, retrieval, and agent establish a foundation for AI-driven therapeutic design. Code and datasets: https://github.com/starling-labs/starling.

Water-based Liquid Scintillator (WbLS) was proposed over a decade ago as a novel detector medium that might allow the separation and tuning of the relative ratio of the Cherenkov and Scintillation signals. A detector deploying this technology could combine GeV-scale and MeV-scale neutrino detection at scale. Furthermore, the metal-loading capability of such a material enables neutron tagging and allows the effective particle containment to be tuned. WbLS is attractive both for the potential to use it in large detectors and the ability to modify the configuration in situ. At Brookhaven National Laboratory (BNL), two prototypes have been built for understanding WbLS properties and stability, with masses of 1-ton and 30-ton, respectively. We present here the 30-ton prototype detector design, installation, and operation. Results from the analysis of data collected in the two detectors will follow in future publications.

The development of Time-Series Forecasting (TSF) models is often constrained by the lack of comprehensive datasets, especially in Global Station Weather Forecasting (GSWF), where existing datasets are small, temporally short, and spatially sparse. To address this, we introduce WEATHER-5K, a large-scale observational weather dataset that better reflects real-world conditions, supporting improved model training and evaluation. While recent TSF methods perform well on benchmarks, they lag behind operational Numerical Weather Prediction systems in capturing complex weather dynamics and extreme events. We propose PhysicsFormer, a physics-informed forecasting model combining a dynamic core with a Transformer residual to predict future weather states. Physical consistency is enforced via pressure-wind alignment and energy-aware smoothness losses, ensuring plausible dynamics while capturing complex temporal patterns. We benchmark PhysicsFormer and other TSF models against operational systems across several weather variables, extreme event prediction, and model complexity, providing a comprehensive assessment of the gap between academic TSF models and operational forecasting. The dataset and benchmark implementation are available at: https://github.com/taohan10200/WEATHER-5K.