Machine Learning
Learning Scientific Exploration from Human Research Decision Trajectories
Xuchen Gong, Shane Gu, Haokun Liu et al. (+3 more)
A key challenge in building AI systems for scientific research is enabling $\textit{scientific exploration}$: the systematic process of investigating unknown phenomena or ideas to gain new knowledge through sequences of research decisions and actions. Yet this process is largely missing from existing scientific corpora; for example, research papers primarily record final outcomes rather than the trajectories that produced them. In this work, we introduce $\textbf{ResearchTrails}$, a
dataset of $\textbf{human research trajectories constructed from Git repositories}$, where $\textbf{commit histories}$ serve as proxies for research exploration. We develop an automated and scalable pipeline that extracts structured research trajectories from repository commits, capturing successive changes to methods, experiments, and ablations. We characterize the resulting
dataset and show that these trajectories contain meaningful signals about intermediate research decisions beyond what final papers reveal. We further demonstrate utilities of ResearchTrails in multiple use cases, including retrieving human research experience as external skills at test time and training models on research trajectories to improve generalization to new research decisions. Our results suggest a path toward AI systems that learn not only from the products of science, but from the evolving process of discovery itself.
Read the full abstract ▾
All authors Xuchen Gong, Shane Gu, Haokun Liu, Dixi Yao, Chenhao Tan, Tian Li
Astrophysics of Galaxies
Hierarchical Black Hole Mergers at High Redshift: Predictions for LISA and LGWA from SEEDZ
Daxal H. Mehta, Lewis R. Prole, John A. Regan et al. (+13 more)
We investigate the hierarchical assembly of black holes (BHs) in the early Universe using the next-generation SEEDZ cosmological zoom-in simulations. Our suite follows two independent regions (Rarepeak and Normal1) evolved with three BH feedback prescriptions: no feedback (NoFB), weak feedback (WeakFB), and standard thermal isotropic feedback (FidBH). Focusing on BH mergers at $z \geq 10$, we explore merger-driven BH growth and detectability with next-generation gravitational-wave observatories. We find that stronger AGN feedback models dramatically suppress hierarchical BH assembly compared to weaker feedback implementations. In the NoFB simulations, BHs frequently undergo repeated mergers, with some experiencing more than 50 merger events within a few hundred Myr and gaining over $10^8,\mathrm{M_{\odot}}$ through mergers alone. In contrast, BHs in the FidBH simulations rarely undergo more than three mergers. NoFB therefore represents an upper limit to BH merger-driven growth beyond what is expected in the Universe. The suppression of BH mergers in the FidBH case is not driven by differences in
galaxy assembly, BH occupation fractions, or BH seed abundances, which remain broadly similar across the simulation suites. Instead, feedback disrupts the dense gas surrounding newly formed BH seeds, suppressing their early
accretion-driven growth and preventing them from entering runaway hierarchical merger pathways. Our results demonstrate that early-Universe feedback prescriptions strongly influence hierarchical BH growth through both mergers and
accretion, producing distinct high-
redshift merger populations. Coupled with seeding choices and other BH subgrid prescriptions, this introduces multiple degeneracies that will need to be broken in advance of LISA and other next-generation GW facilities.
Read the full abstract ▾
All authors Daxal H. Mehta, Lewis R. Prole, John A. Regan, Rüdiger Pakmor, Sophie Koudmani, Ricarda S. Beckmann, Michael Tremmel, John H. Wise, Martin A. Bourne, Ralf S. Klessen, Simon C. O. Glover, Martin G. Haehnelt, John Brennan, Pelle Van De Bor, Debora Sijacki, Paul C. Clark
Astrophysics of Galaxies
Spectroscopic Confirmation of the Milky Way Satellites Boötes V and Leo Minor I and the LMC Satellite DELVE 2: Three Small Ultra-Faint Dwarf Galaxies
W. Cerny, M. Geha, J. D. Simon et al. (+22 more)
We present the first multi-object spectroscopy of Boötes V, Leo Minor I, and DELVE 2 – three small (20 pc $\le r_{1/2}\le$ 26 pc), ultra-faint ($-3.2\le M_V\le-2.1$)
stellar systems in the outer
Galactic halo (71 kpc $\le D_\odot\le$ 102 kpc) with heretofore ambiguous classifications. Leveraging at least two epochs of Keck/DEIMOS or Magellan/IMACS observations targeting each satellite, we identify 13, 7, and 7 spectroscopic members in the three systems, respectively. We tentatively resolve a $\sigma_v=2.1^{+1.5}_{-0.9}$ km s$^{-1}$ velocity dispersion for DELVE 2 but place upper limits of $\sigma_v<2.7$ km s$^{-1}$ and $\sigma_v<7.8$ km s$^{-1}$ for Boötes V and Leo Minor I. We find that all three systems are very metal-poor, with mean iron abundances $-2.85\le\overline{\rm [Fe/H]}\le-2.15$. Both Boötes V and Leo Minor I display unambiguous
metallicity spreads ($\sigma_{\rm [Fe/H]}=0.46^{+0.18}_{-0.12}$ dex and $\sigma_{\rm [Fe/H]}=0.57^{+0.23}_{-0.18}$ dex), while DELVE 2 displays a modestly significant spread ($\sigma_{\rm [Fe/H]}=0.52^{+0.29}_{-0.25}$ dex) across three
stars. Based primarily on these
metallicity measurements, we confidently classify Boötes V and Leo Minor I as
dwarf galaxies and determine that DELVE 2 is most likely a
dwarf galaxy as well. We explore the orbit of each satellite in a joint Milky Way + LMC potential, confirming for the first time that DELVE 2 is a long-term bound satellite of the LMC. Finally, we review the empirical Milky Way satellite classification landscape and find that nearly every spectroscopically characterized satellite with azimuthally averaged half-light radius $r_{1/2}>15$ pc and surface brightness $\mu_{V,1/2}>26.5$ mag arcsec$^{-2}$ is likely a
dwarf galaxy. This result can guide initial probabilistic classifications for satellites discovered by $\textit{Rubin}$, $\textit{Roman}$, and $\textit{Euclid}$.
Read the full abstract ▾
All authors W. Cerny, M. Geha, J. D. Simon, T. S. Li, A. B. Pace, R. P. van der Marel, N. Kallivayalil, A. P. Ji, A. Drlica-Wagner, D. Crnojević, G. Limberg, N. E. D. Noël, A. H. Riley, D. J. Sand, G. S. Stringfellow, Y. Choi, P. S. Ferguson, C. E. Martínez-Vázquez, P. Massana, G. E. Medina, B. Mutlu-Pakdil, J. D. Sakowska, C. Y. Tan, A. K. Vivas, J. Bruce
Earth and Planetary Astrophysics
Enhancing the Interpretability of Radial Velocity Analysis Using Leave-one-out Cross-validation
Matthew C. Nixon, Ritvik Basant, Rafael Luque et al. (+2 more)
Radial velocity (RV) observations are reaching increasingly high precision, enabling the targeting of low-amplitude signals. Assessing results obtained at the limits of our capabilities requires understanding the relationship between data, models and inference. Leave-one-out cross-validation (LOO-CV) has become an increasingly popular tool for interpretable model criticism in astronomical data analysis. In this work, we demonstrate how LOO-CV can provide insight into the RV analysis of exoplanetary systems through three case studies. We first investigate the anomalously high mass of HD 119130 b inferred from CARMENES RVs, later revised significantly downward after follow-up observations with other instruments. We show that this high mass was strongly driven by two individual measurements, whose removal significantly reduces the tension between data sets. Next, we use LOO-CV as a model selection tool for the GJ 4276 system. Here, LOO-CV prefers an eccentric single-
planet model over a two-
planet model, and identifies the orbital phases at which follow-up observations would best discriminate between them. Finally, we extend LOO-CV to models with correlated noise, applying it to the GJ 357 system, whose RVs are modeled with a Gaussian process. We find no predictive support for a fourth planetary signal that is moderately favored by the
Bayesian evidence, strengthening the interpretation of a three-
planet architecture previously reported in the literature. We provide practical guidance for implementing LOO-CV in RV analysis, together with an online tutorial. Such methods will become ever more important as future RV surveys push the limits of
planet detection and characterization.
Read the full abstract ▾
All authors Matthew C. Nixon, Ritvik Basant, Rafael Luque, Luis Welbanks, Pierrot Lamontagne
Number Theory
On the class number of real multiquadratic fields
Mohamed Mahmoud Chems-Eddin and Hamza El Mamry
In their 1988 monograph, Conner and Hurrelbrink classified all real quadratic and biquadratic fields with odd class number. They also investigated real quadratic fields whose $2$-class numbers are equal to $2$. Their results, together with some well-known results of Kučera, provide a complete classification of real quadratic number fields with $2$-class number $2$. The main aim of this paper is to provide a complete classification of all real biquadratic fields whose $2$-class numbers equal $2$; and moreover, we draw the complete list all real multiquadratic fields with odd class numbers. Finally, as an appendix we give explicit formulas for $\sqrt{\varepsilon_d}$, where $\varepsilon_d$ denotes the fundamental unit of the real quadratic field $\mathbb{Q}(\sqrt{d})$.
Read the full abstract ▾
Mesoscale and Nanoscale Physics
Adjoint-State Identifiability of Piezo-Tunable Valley Splitting in 2D Magnetic Heterostructures
Suhas Suresh Bharadwaj
Controlling valley degrees of freedom with mechanical strain is a promising approach for solid-state information processing. Current theoretical literature routinely predicts strain-tuned valley splitting at the microscopic level but rarely evaluates whether these
quantum predictions remain statistically recoverable in realistic macroscopic devices. This manuscript establishes a fully classical, partial differential equation-constrained multiscale inverse framework for quantifying the device-level identifiability of predicted strain-tunable valley effects in two-dimensional
magnetic heterostructures, demonstrated here for a molybdenum disulfide and chromium tribromide
heterostructure. First-principles structural relaxations confirm a
chiral $C_3$ point-group
symmetry which mathematically reduces the relevant exchange-strain coupling tensor to a single scalar. A partial differential equation-constrained adjoint-state architecture successfully bridges continuum elastodynamics to valley-resolved anomalous
Hall transport. Density functional theory yields a coupling estimate of $\eta \approx -0.07$ meV whose 95% confidence interval is consistent with zero. Evaluating this specific coupling magnitude against established thermal noise and velocity saturation limits defines a safe operating window bounded between 262.0 and 22,337.6 V/cm. Rather than asserting a confirmed nonzero material property this bounded operational window functions as a precise diagnostic threshold. Deploying this rigorous statistical identifiability framework provides a necessary mathematical filter to determine the true experimental viability of theoretically predicted two-dimensional materials before complex physical fabrication.
Read the full abstract ▾
Astrophysics of Galaxies
Molecular hydrogen formation on dust: The impact of gas-dust drift on formation efficiency
Stefan Reissl, Simon C. O. Glover, Ralf S. Klessen et al. (+2 more)
Molecular hydrogen is predominantly formed on
dust-grain surfaces in the interstellar medium, where relative gas-
dust motion can arise in dynamically active environments. While the dependence of H$_2$ formation on grain temperature and surface properties is well studied, the impact of gas-
dust drift has received little
attention. We investigate how gas-
dust drift modifies H$_2$ formation, focusing on the competition between the drift-enhanced H-atom collision rate and reduced sticking at higher impact energies. We use an event-driven kinetic Monte Carlo model that follows individual H atoms on spherical silicate and carbonaceous grains, including adsorption, surface migration, thermal desorption, and Langmuir-Hinshelwood (LH) and Eley-Rideal (ER) reactions. Drift is described by a shifted Maxwellian velocity distribution, and we compare constant and impact-energy-dependent sticking probabilities. Drift produces increasingly anisotropic distributions of adsorbed H and H$_2$ formation across the grain surface. Assuming constant sticking, increasing drift enhances H$_2$ formation through the higher collision rate, with efficiencies up to $\epsilon=0.3-0.4$. With energy-dependent sticking, strong drift instead suppresses formation on both materials, reducing efficiencies to $\epsilon=0.01-0.03$. Carbonaceous grains remain efficient to higher
dust temperatures than silicate grains. ER reactions dominate over most of the investigated parameter space and become increasingly important at strong drift as the reduced surface population suppresses LH reactions. Thus, enhanced collision rates under gas-
dust drift do not necessarily increase H$_2$ formation. Models of dynamically active environments should account for both relative gas-
dust velocities and their effects on sticking.
Read the full abstract ▾
All authors Stefan Reissl, Simon C. O. Glover, Ralf S. Klessen, Liam S. Morrissey, Mordecai-Mark Mac Low
Machine Learning
CLM-as-a-Judge: Evaluating an Open Contrastive Decision Model on Public Judge Benchmarks
Gowthamkumar Nandakishore
An open contrastive decision model is near chance as a judge on the hard public
benchmarks: Contrastive-LM/CLM-v0.1-8B scores between 0.351 (best- of-four, chance 0.250) and 0.593 (pairwise, chance 0.500), is statistically indistinguishable from coin flipping on RM-Bench and JudgeBench, and answers every HaluEval item with one constant label, matching the trivial always-first baseline at 0.581. Judges with the same parameter count score far higher everywhere: a reward model reaches 0.764 to 0.976 and a generative judge 0.611 to 0.778, and every gap to CLM is significant after Benjamini-Hochberg correction. Two properties do work. Raw confidences are overconfident by up to +0.401, yet one pooled temperature fit on held-out calibration items repairs expected calibration error to at most 0.062, and the repaired confidence ranks the model's own errors above chance on three of six
benchmarks. The decision order-flip rate is 0.0002 against 0.2188 for the generative judge, and the length-preference shift is -0.023 against -0.217. The confidence-gated cascade, however, escalates between 0.923 and 1.000 of items to the strong judge at the preregistered 0.97 retention bar: calibrated confidence about a near-chance judge has almost nothing to keep. The design: five public preference
benchmarks and one
hallucination benchmark with real labels, scored under a preregistration frozen before any test item was seen, against generative, reward-model, and trivial baselines, with per-item predictions released.
Read the full abstract ▾
High Energy Physics (Theory)
Generalised entanglement wedges in three dimensions
Abhisek Sahu
Generalized
entanglement wedges (GWs), proposed by Bousso and Penington for gravitating regions in general spacetimes, share many properties of ordinary
entanglement wedges, including nesting, monotonicity and strong subadditivity of their generalized
entropies. In AdS/CFT, the corresponding properties of
entanglement wedges are consequences of the structure of the dual CFT. In this paper we ask whether the properties of GWs have a similar origin. We explore this question in
vacuum AdS$_3$ at leading order in $G$. We first show that every GW on a static slice is a union of geodesically convex regions. Each such region is an intersection of ordinary
entanglement wedges, so every GW is built from unions and intersections of
entanglement wedges. We then use the Crofton formula of integral geometry to write the perimeter of a convex region as the measure of the set of geodesics that intersect it. In AdS$_3$ every geodesic is the Ryu–Takayanagi surface of a boundary interval, so this measure can be written in terms of the von Neumann
entropies of CFT intervals. Strong subadditivity of the CFT
entropy guarantees that the measure is positive. Monotonicity and strong subadditivity of GWs then follow from simple inequalities between sets of geodesics. The strong subadditivity of GWs is therefore a direct consequence of strong subadditivity on the boundary. Our results further suggest that GWs correspond to subsystems of the CFT that are more general than boundary subregions.
Read the full abstract ▾
Computer Vision
CALR: Continuous Anchored Latent Reasoning via Render-of-Thought Compression
Zhaoyang Wei, Bowen Jiang, Yanchao Hao et al. (+5 more)
Visual latent
reasoning compresses rendered derivations into compact intermediate states, reducing textual
reasoning overhead. Existing approaches differ in how they represent these states: continuous methods avoid vocabulary constraints, whereas discrete methods improve accuracy through quantization into a finite codebook. Our analysis of representative continuous and discrete systems identifies two functional requirements: answers must rely on latent states, and those states must carry valid, problem-specific
reasoning. Continuous latents influence answers despite collapsed
reasoning content, whereas discrete latents retain recoverable intermediate
reasoning that answer prediction largely bypasses. To address these challenges, we propose Continuous Anchored Latent
Reasoning (CALR), which connects latent formation with answer use through functional anchoring. With reference latents from information-balanced compression, CALR couples latent-mediated answer supervision with derivation-level semantic anchoring: the former routes answer supervision through intermediate states, while the latter grounds their decoded content in problem-specific derivations. A parallel-to-autoregressive curriculum develops sequential
reasoning by conditioning subsequent latent blocks on generated prefixes. Evaluations on five mathematical
reasoning benchmarks across model families show substantial accuracy gains. Under matched budgets, CALR gains 26.0 percentage points over a comparable continuous latent
reasoning method. Further analyses show that its latents support answer prediction and carry problem-specific intermediate
reasoning.
Read the full abstract ▾
All authors Zhaoyang Wei, Bowen Jiang, Yanchao Hao, Wenchao Ding, Zheng Wei, Shaocheng Wu, Zhenjun Han, Jianbin Jiao
Mesoscale and Nanoscale Physics
Composite Fermion Theory of Fractional Chern Insulators in Rhombohedral Graphene
Peize Ding, Andrew J. Millis and Yongxin Zeng
The experimental observation of the fractional
quantum anomalous
Hall effect in rhombohedral
graphene multilayers has raised questions about the nature of the fractional Chern
insulator (FCI) states and the role of the moiré potential. In this Letter we develop a composite
fermion (CF) description and discuss how moiré modulation and Coulomb interactions stabilize the FCI ground states. Under a flux-attachment transformation and mean-field approximation, the CF spectrum forms Landau levels that are further split by the moiré potential. At moiré filling fractions in the Jain sequence, an integer number of CF bands are occupied, resulting in an incompressible ground state with fractionally quantized
Hall response. Remarkably, when the attached flux opposes the Berry
curvature of the conduction band of rhombohedral
graphene, the CF Landau levels are widely spaced in energy and delocalized across layers, leading to a reduced density of states that helps protect the CF mean-field state and amplifying effects of the moiré potential acting on the layer away from the doped electrons. Interactions between CFs further enlarge the moiré-induced band gap and stabilize the FCI ground state. Our work sheds light on the nature of the FCI states in rhombohedral
graphene and provides a new theoretical framework for further studies.
Read the full abstract ▾
Quantum Physics
Connectivity Controls Variational Accessibility in Symmetry-Preserving Quantum Circuits
Sahinur Reja
Designing expressive
quantum circuits for the variational
quantum eigensolver (VQE) is a central challenge in simulating interacting many-body systems. Even
symmetry-preserving (SP)
circuits need not access the full many-body
symmetry sector. We demonstrate this explicitly for the
spin-$1/2$ XXZ and Hubbard models, where
Hamiltonian-inspired nearest-neighbor (NN) SP
circuits generate restricted dynamical Lie
algebras (DLAs). While these restricted
manifolds can accurately represent the ground state near the free-
fermion limits, $J_z=0$ for XXZ and $U=0$ for Hubbard, interactions expose inaccessible many-body directions and severely limit VQE performance even at large
circuit depth. Remarkably, a minimal modification of the generator connectivity can reconstruct the full orthogonal DLA within all relevant
symmetry sectors studied in this work. For XXZ model, the addition of only one SP next-to-NN single-excitation connection triggers this dramatic reconstruction, whereas the Hubbard case requires a SP correlated double-excitation;–showing that the relevant connectivity can extend from the spatial
lattice to many-body configuration space. This reconstruction enlarges the tangent space and accessibility of the energy-lowering
Hamiltonian direction, leading to near-unit ground-state fidelity in VQE simulations. Our results establish a direct link between generator connectivity and variational accessibility and suggest a broader
circuit-design principle: identify minimal
symmetry-preserving generators that enlarge the accessible many-body space rather than increasing
circuit depth alone. This perspective may have broader relevance for
quantum many-body simulation at the interface of condensed-matter physics and
quantum computing.
Read the full abstract ▾
Quantum Physics
Pauli Flat Quantum States Mimicking Maximal Magic
Paweł Cieśliński
Quantum magic is a necessary resource for fault-tolerant
quantum computing, capturing the departure of
quantum states from the classically simulable stabiliser set. Motivated by its role in
quantum information processing, we investigate the correlation structure of states that maximize the Rényi stabiliser
entropies, focusing on the T-type and Hoggar states. Inspired by their distinctive properties, we explore the notion of Pauli flat states, defined as N-
qubit states whose expectation values on all Pauli strings are equal in absolute value. These states reproduce the correlation structure of highly magical states, which could also be interpreted as a mixed variant of Pauli covariant N-
qubit symmetric informationally complete (SIC) Weyl-Heisenberg fiducial states, while not necessarily possessing nontrivial
quantum magic. We investigate their existence and possible constructions, including examples both inside and outside of the stabiliser polytope. To identify the latter, we examine a magic witness tailored to the introduced family of states. Finally, we discuss their
entanglement and Bell nonlocality, providing a broader characterization of their
quantum properties. For completeness, we find pure states with a relaxed condition on the Pauli expectations, which forces unequal elements to zero, and discuss their non-stabiliserness. Our results offer a new perspective on the relation between
quantum magic and Pauli correlation structures. They hold relevance to geometrical approaches to
quantum mechanics,
entanglement and magic detection, as well as estimation tasks.
Read the full abstract ▾
High Energy Physics (Phenomenology)
Compositeness and decay properties of the flavor partner of $D^{*}_{s0}(2317)$
J. Sánchez-Illana, R. Molina and Pan-Pan Shi
In view of the recent experimental finding of the heavy
quark flavor partner of the $D_{s0}(2317)$ in the bottom sector, the $B_{s0}(5700)$, we perform the $B^{(*)}\bar{K}^{(*)}$ coupled-channel calculation in the hidden-gauge formalism. By fixing the mass of the bound state generated by the $B\bar{K}$ interaction with $J^P=0^+$ to the observed state, we also predict five new states with $J^P=\lbrace 0,1,2\rbrace^+$ in this sector. In particular, three of them have $J^P=1^+$ and masses $5740$, $6005$ and $6065$ MeV, respectively. In addition, we evaluate the decay width of the $B_{s0}$ and $B_{s1}$ states through the isospin breaking decay mode $B_s^0\pi^0$ and $B_s^{*0}\pi^0$, respectively. We find that, because these states have large binding energies with respect to the $B^{(*)}\bar{K}$ threshold, the coupling to the bare component can reduce this decay width by more than half of the decay width, from $50$ keV in the purely molecular model to $10-20$ keV, when the $B_{s0}$ is coupled to the bare component. We also observe that, after considering heavy-
quark-
spin violating contributions, the decay width for the $B_{s1}$ becomes smaller than for the $B_{s0}$.
Read the full abstract ▾
High Energy Physics (Theory)
Multi-Point Affine Bialgebras for AdS/CFT Integrability
Niklas Beisert, Egor Im and Hannes Malcha
In this work, we revisit the analytical structure of the classical affine Lie bialgebra underlying the integrable structures of AdS/CFT worldsheet scattering and the one-dimensional Hubbard model. Earlier work showed that the algebraic relations implied two singularity locations in the evaluation parameter plane besides the common expansion points infinity and zero for Laurent
polynomials forming the basis for loop and affine
algebras. Our aim is to formulate these singularities rigorously using meromorphic functions rather than by expansion into
geometric series. We thus carefully analyse the analytical structure of the affine bialgebra relations in the complex plane paying
attention to potential restrictions and extensions due to the additional singularities. As a first attempt with some shortcomings, we reformulate the affine bialgebra relations with the extra singularities serving as the expansion points for Laurent
polynomials. We finally overcome the remaining issues by proposing three-point and four-point affine
algebras for the rational and trigonometric cases. With a suitable choice of basis for meromorphic functions on the Riemann sphere with several punctures based on Krichever-Novikov
algebras, we show that the multi-point affine
algebras form consistent quasi-triangular Lie bialgebras.
Read the full abstract ▾
Mesoscale and Nanoscale Physics
Proposing one-sided tetrons as topological qubits for Majorana-based quantum computation
Tudor D. Stanescu and Sumanta Tewari
One-dimensional SM-SC hybrid nanowires with Rashba
spin-orbit coupling and applied Zeeman field can support Majorana zero modes (MZMs) at the ends of the wire. In finite-length systems, residual wave-function overlap between the end modes produces an energy splitting. In a two-sided tetron
qubit, composed of a pair of parallel Majorana wires coupled by a trivial SC backbone in the middle, this splitting limits the parity lifetime associated with Majorana pairs involving both wires and the coherence of the
qubit, adversely affecting measurement-based braiding and readout operations. Here, we propose a one-sided tetron made of four parallel Majorana wires connected at a common end by an SC backbone. In the topological phase, the free wire ends host four MZMs that define the topological
qubit, while the Majoranas localized near the opposite ends hybridize through the backbone and acquire finite energy. To study the basic hybridization mechanism, we consider a simpler two-wire subsystem, i.e., half of a one-sided tetron, which we call a doublon. In a doublon, the backbone, or bridge, strongly hybridizes the two Majorana modes at the connected end and shifts them away from zero energy, while the modes at the free ends form a near-zero energy
fermion. We show that the splitting of the surviving MZM pair can fall by about two orders of magnitude compared to the splitting energies in a two-sided tetron, thereby dramatically increasing the parity lifetimes. Thus, the one-sided tetron ensures a suppression of the Majorana overlap comparable to doubling the length of the corresponding two-sided tetron.
Read the full abstract ▾
Machine Learning
A theory of platonic representations in language models
Darshil Doshi, Wenjie Zhou, Corinna Elena Wegner et al. (+3 more)
Representations of translated sentences are similar in the inner layers of
multilingual language models – an observation connected to the platonic representation hypothesis, yet unexplained theoretically. We provide an explanation based on the assumption that data have a hidden hierarchical structure whose abstract levels are shared across
languages while surface levels are modality- or
language-specific. Concretely, we generate synthetic
languages from probabilistic context-free grammars sharing upper-level but not lower-level production rules. In this setting the Bayes-optimal next-
token predictor is belief propagation (BP); encoding its messages in successive layers yields analytical predictions that agree well with
transformers trained on the same data. The framework explains why cross-lingual similarity peaks in middle layers, coexists with
language-specific structure, and strengthens with
language proximity, model quality and data exposure. It distinguishes similarity (shared neighborhood geometry) from
alignment (shared coordinates), showing that the latter occurs when code-switched data, i.e. mixed-
language sentences, are abundant enough. It further predicts that subtracting from each layer the component linearly predictable from the preceding one increases cross-lingual similarity, which we confirm in pretrained LLMs.
Read the full abstract ▾
All authors Darshil Doshi, Wenjie Zhou, Corinna Elena Wegner, Daniel J. Korchinski, Santiago Acevedo, Matthieu Wyart
Systems and Control
Interleaved Projected Gradient Descent for Safe Imitation Learning
Shengfan Cao and Francesco Borrelli
We propose an imitation-learning design for neural-network control policies under state and input constraints. Training alternates a standard imitation gradient step with a block of $k$ safety steps that pull the network's actions toward their projection onto the safe set; at run time, the controller is the trained network alone, with no safety filter. We analyze this scheme as inexact projected gradient descent in the space of policy actions. When the projected actions are recomputed at every safety step and each step moves the actions consistently toward the safe set, letting $k$ grow logarithmically yields asymptotic constraint satisfaction on the training states and bounds the distance to the constrained optimum of the imitation loss; with the projected actions held fixed, the same holds only if they are exactly representable by the network. On a nonlinear autonomous racing task, we compare our method with adding a weighted constraint-violation penalty to the imitation loss. With a sufficiently large weight, our method matches the lap time of unconstrained imitation while reducing the fraction of violating episodes from $15\%$ to $1\%$, about six times fewer than the penalty approach at its best weight. Its lap times are less sensitive to the weight, which instead sets how quickly violations vanish during training. In racing, the safety corrections are sparse and the conditions of the analysis do not hold; the gain arises instead through the data collected during training. These gains come at the cost of additional training computation.
Read the full abstract ▾
General Relativity and Quantum Cosmology
Propagation and horizon formation in Einstein–Gauss–Bonnet scalar collapse. I. Characteristic conditions, bounds and formation
Cendikiawan Suryaatmadja
We study scalar collapse in five- and six-dimensional Einstein–Gauss–Bonnet
gravity with positive coupling, treated as an exact classical theory. On spherical scalar backgrounds, the characteristic conditions of the tensor, vector and scalar gravitational sectors reduce to inequalities in two local invariants: $K=1+\tilde\alpha F/R^2$, which compares the enclosed mass with the areal radius $R$, and a scalar-gradient invariant $j$. In five dimensions they imply $R_h^2>\tilde\alpha$ for the areal radius $R_h$ of a regular marginal sphere on the GR branch and the sharp bound $\tilde\alpha^2\mathcal R_{abcd}\mathcal R^{abcd}<28$ for nonnegative mass. For initially untrapped five-dimensional spherical data with a regular center, a bound on $K$ alone ensures smooth continuation on compact domains of dependence, so any first characteristic loss occurs on a smooth background, and a marginal sphere formed before any loss forces a later loss. Complete smooth data form marginal spheres at every allowed radius with mass arbitrarily close to its sharp lower bound; in six dimensions radius and mass can both be arbitrarily small. Without spherical
symmetry, complete initial data with weak
curvature contain, at sufficiently large scale, outermost strictly stable marginal surfaces with prescribed $S^3$ geometries satisfying a spectral condition; Berger surfaces among them have arbitrarily small area at fixed mass. Their formation from untrapped data is left open. Near a contracting homogeneous core, exact nonspherical developments of smooth constrained data converging with every derivative have bounded
curvature but, after the core's characteristic loss, a fixed discrepancy in a
geometric tidal record. Classical existence, local
geometric uniqueness and continuity of this pointwise record cannot all be guaranteed on a common observation region. A companion paper treats narrow incoming pulses.
Read the full abstract ▾
High Energy Physics (Theory)
Non-invertible SPT phases from fusion rings via Tannaka duality
Antonio Santaniello, Roberto Valandro and Massimo Zorzenon
In a (1+1)-dimensional QFT with non-invertible
symmetry, the fusion rules of topological lines are often the most accessible data, yet they do not fully determine whether the
symmetry is anomaly-free, i.e. whether it admits trivially gapped phases. Such phases are characterized by fiber functors of the corresponding fusion category. Building on Tannaka duality, we extract fiber functors from the fusion ring alone by solving simultaneously for monoidal structure maps and tensor
automorphisms. This allows us to find categorifications admitting a fiber functor without solving the pentagon equations directly. We test this procedure on the known Tambara–Yamagami ring of $\mathbb{Z}_2\times\mathbb{Z}_2$. Our main result concerns the rank-11 fusion ring shared by the representation categories of the two extraspecial groups of order 27, which has a multiplicity 3 channel and is relevant to self-triality. In both cases we find more than one fiber functor, i.e. several non-invertible SPT phases. We use the explicit data of these fiber functors to derive the $F$-symbols of the two representation categories in a compact form depending on a non-degenerate alternating bicharacter and an invertible object $\epsilon$. Using the reconstructed data, we compute the anomalous
algebra on the interface between two different SPT phases.
Read the full abstract ▾
Algebraic Geometry
Schubert compactifications of matrix pencils
Vincenzo Galgano, Fulvio Gesmundo and Hanieh Keneshlou
The Schubert compactification of a linear space of
matrices $L \subseteq \textrm{Mat}_{n\times m}$ is a natural compactification of $L$ in the Grassmannian $\textrm{Gr}(n, n+m)$. This construction generalises classical Schubert varieties and matroid Schubert varieties, and it turns invariants of linear
matrix spaces into projective and intersection-theoretic invariants of the associated variety. We develop this geometry completely in the case of
matrix pencils, that is $\dim L=2$. The resulting Schubert surfaces are singular analogues of rational surfaces obtained as blow-ups of $\mathbb{P}^2$. We show that they are normal with quotient singularities, compute their degree,
cohomology ring, and describe the distinguished rational curves and their intersection pairing in terms of the Kronecker invariants of the
matrix pencil. Remarkably, this dictionary can be reversed: the invariants of the pencil are read off from intersection numbers and other invariants of the variety. In particular, we prove that the Schubert surface determines the regular component of the pencil, up to the natural action of $\mathrm{GL}_2 \times \mathrm{GL}_n \times\mathrm{GL}_m$. Finally, we prove that the ideal of Schubert surfaces of
matrix pencils in the Plücker
embedding is generated in degree $2$ and satisfies Green's property $N_p$ in a certain range.
Read the full abstract ▾
Astrophysics of Galaxies
DIGging into HII regions: the role of ionizing photon leakage and spectral hardening in coupling HII regions and diffuse ionized gas
Dalya Baron, Kathryn Kreckel, Gwen C. Rudie et al. (+15 more)
The optical emission-line ratios of
star-forming
galaxies are broadly understood in the framework of photoionization, yet some systematic trends remain unexplained. Using 50–100 pc resolution PHANGS-MUSE observations of 19 nearby
galaxies, we investigate how escaping ionizing radiation couples HII regions to the diffuse ionized gas (DIG), analyzing thousands of HII regions and DIG sightlines. HII regions and the DIG form a continuous sequence across standard line-ratio diagnostic diagrams (e.g., BPT diagrams), where progressively fainter regions show elevated high- and low-ionization line ratios, and the DIG extends the same trends into the composite–LINER regime. Line-ratio variations with global specific
star formation rate (sSFR) suggest progressively harder ionizing radiation with increasing
star formation activity, contrary to expectations for emission increasingly dominated by older
stellar populations. Line-of-sight mixing explains part of the behavior but cannot reproduce all line-ratio trends. We therefore develop a model in which ionizing
photons escape HII regions and propagate through the diffuse ISM, undergoing
geometric dilution, attenuation, and spectral hardening. The model predicts that progressively fainter HII regions are increasingly ionized by leaked radiation from neighboring luminous regions, implying radiative coupling between HII regions. Joint modeling of HII regions and the DIG reproduces the HII region luminosity distribution and DIG surface brightness for escape fractions of $\sim$0.4$-$0.6 and ionizing-
photon mean free paths of $\sim$0.5$-$1 kpc, and qualitatively reproduces the observed line-ratio sequences and their dependence on sSFR. Our results suggest that ionizing
photon leakage may provide an important non-local feedback channel linking compact
star-forming sites to large-scale
galaxy disks.
Read the full abstract ▾
All authors Dalya Baron, Kathryn Kreckel, Gwen C. Rudie, Karin M. Sandstrom, Alberto D. Bolatto, Guillermo A. Blanc, Oleg V. Egorov, Eva Schinnerer, Ashley T. Barnes, Enrico Congiu, Daniel A. Dale, Simon C. O. Glover, Jaeyeon Kim, Ralf S. Klessen, José Eduardo Méndez-Delgado, Justus Neumann, Ismael Pessa, Thomas G. Williams
Machine Learning
Adversarial Training for Deep Hedging in Nonstationary Markets
Philipp J. Schneider, Lukas Looser, Antoine Garin et al. (+2 more)
Deep hedging learns trading policies from historical or simulated
market trajectories, yet under nonstationarity these training paths may not represent future
market conditions. We propose WRAP (Wasserstein-Reweighting
Adversarial Perturbation), a drift-aware
adversarial training framework derived from a two-budget distributionally robust optimization (DRO) formulation. The formulation is anchored to a weighted empirical reference distribution whose fixed baseline weights are chosen to balance sampling uncertainty against temporal drift. Around this reference distribution, the ambiguity set addresses two complementary forms of distributional misspecification by allowing an adversary to reweight the observed trajectories subject to a $\phi$-divergence constraint and perturb their paths subject to an optimal-transport (OT) constraint. We derive a joint first-order expansion in which the leading-order increase over the nominal expected loss decomposes into a reweighting contribution determined by the dispersion of hedging losses across trajectories and a transport contribution determined by the sensitivity of the loss to path perturbations. This expansion yields an explicit finite-dimensional
adversarial attack that replaces the distributional inner supremum with a tractable first-order approximation. Across stationary and nonstationary Heston dynamics and a generalized affine
diffusion (GAD), the experiments show complementary benefits from reweighting and transport, with joint
adversarial training providing the largest gains under nonstationarity.
Read the full abstract ▾
All authors Philipp J. Schneider, Lukas Looser, Antoine Garin, Shuhan Liu, Daniel Kuhn
Earth and Planetary Astrophysics
Discovery of a second-generation planet candidate accreting onto a white dwarf
Jamie T. Williams, Boris T. Gänsicke, Nicholas C. Stone et al. (+10 more)
Many white
dwarfs accrete the debris of disrupted planetary bodies, detected in the form of metal-enrichment in their atmospheres, dusty or gaseous circumstellar discs, and photometric transits from debris. The composition of accreted debris has been shown to be diverse yet overall closely resembles
Solar System bodies. We report the discovery of a white
dwarf accreting material that is unlike any
Solar System object. The atmosphere of HS 0209+0832 is strongly enriched in trans-iron elements including zinc, copper, and niobium but depleted in the canonical rock-forming elements silicon and iron. The composition of the accreted object is consistent with a candidate second-generation
planet, formed from the
stellar material ejected during the
giant phase. The key abundance signature which differentiates it from a first-generation
planet is the high enrichment of s-process elements. Photospheric helium and the absence of terrestrial rock-forming elements implies the white
dwarf is accreting from the escaped atmosphere of a photo-evaporating
giant planet candidate. This is further supported by the detection of a sinusoidal photometric period of 4.399+/-0.026 days at an amplitude of 0.120+/-0.018 %, which we attribute to thermal emission phase variability from a planetary day-night cycle or a transiting cometary tail of an evaporating gas
giant.
Read the full abstract ▾
All authors Jamie T. Williams, Boris T. Gänsicke, Nicholas C. Stone, Detlev Koester, Benjamin D. R. Davies, Christopher Tong, David J. Wilson, Snehalata Sahu, Andrew Swan, Thomas G. Beatty, Sergio H. Ramírez, Tim Cunningham, Knox S. Long
High Energy Astrophysical Phenomena
Power-Law Injection Spectrum from an Ensemble of Reconnecting Current Sheets
Omar French
A collisionless, low-$\beta$
plasma undergoing strong sub-Alfvénic
turbulence is expected to form a broad hierarchy of reconnecting current sheets immersed in a strong guide
magnetic field. If the reconnecting field strength of a sheet scales with its length as a power law of index $\kappa$, and the number of sheets per unit interval of sheet length decreases as a power law of index $x$, the current-sheet ensemble injects nonthermal particles with a power-law energy spectrum $\propto \gamma^{-s}$, where $s = 1 + (x-2)/(1 + \kappa)$. The spectrum is hard ($s < 2$) when the longest sheets deliver most of the injected energy, which requires $x < 3 + \kappa$. Past numerical studies have measured $2 \lesssim x \lesssim 3.3$, and the scalings of intense current sheets support $\kappa \gtrsim 0$, resulting in $1 \lesssim s \lesssim 2.3$. This injection may yield flat or mildly steep
radio spectra in optically thin sources and hard ion spectra in the coronae of active
galactic nuclei.
Read the full abstract ▾
High Energy Physics (Phenomenology)
Magnus: neutrino oscillation probabilities for any Hermitian Hamiltonian, any number of flavors, and any matter profile
Mauricio Bustamante
Interpreting
neutrino oscillation measurements requires computing the probability that a
neutrino born with one
flavor is detected with another. In
vacuum and in matter of constant density, this probability has exact formulas. Where the density varies along the path, as in the Earth, the Sun, or a
supernova, it does not. There, the usual approaches either replace the profile by constant-density steps, whose error falls only as the square of their width, or integrate the
evolution equation with a general-purpose solver, which must shorten its steps to meet a tighter tolerance. Both become slow at high accuracy, yet an analysis needs the probability at many energies, directions, and parameter values. Each new experiment or model often requires writing and validating a new solver. We present Magnus, an open-source Python code that avoids these limitations. It uses the Magnus expansion, whose error falls as a high power of its step width and whose cost is set by how fast the density varies, not by how many oscillations the path holds. Magnus accepts any density profile, any number of
flavors, and any Hermitian
Hamiltonian, such as those of non-standard interactions, sterile
neutrinos, Lorentz-invariance violation, and pseudo-Dirac
neutrinos. Where a cheaper route exists, it takes it automatically. For example, where the density varies slowly, as in the Sun, it follows the instantaneous eigenstates and applies the expansion only at resonances. Besides oscillating probabilities, Magnus returns phase-averaged ones, which describe
solar and astrophysical
neutrinos. A typical probability takes a few milliseconds, and batched scans are one to two orders of magnitude cheaper per point. Among the public codes we compared, Magnus is the fastest at high accuracy on a varying profile. Magnus shifts the effort of studying a new experiment or model from writing a solver to specifying its physics.
Read the full abstract ▾
Strongly Correlated Electrons
Local-moment magnets on all close-packed lattices are equivalent
Sam Studdy, Alwyn Jose Raja and R. Ganesh
The Heisenberg antiferromagnet on the fcc
lattice is a canonical example of frustration. The fcc
lattice is but one of an infinite number of close-packed structures, all of which have the same frustration-inducing local geometry. In this letter, we demonstrate a hidden
symmetry that relates classical magnets on all close-packed
lattices. This
symmetry derives from a similar property in tight binding models that may explain the disordered low-temperature structures of lithium and sodium metals. In magnets, the
symmetry holds when couplings are restricted to nearest ($J_1$) and next-nearest neighbours ($J_2$). It guarantees that all close-packed
lattices are equivalent at the Luttinger-Tisza level. For any given values of $J_1$ and $J_2$, they have the same ground-state energy. Their ground state momenta are identical when projected onto the stacking plane. We construct a unified phase diagram as a function of $J_1$ and $J_2$ couplings for Heisenberg, XY and Ising magnets on any close-packed
lattice. In each phase, the ground-state degeneracy is the same on all close-packed
lattices, even after accounting for multi-Q spirals. This phase diagram can be easily adapted to any magnet composed of stacked triangular layers with similar couplings. Fluctuations above the classical ground state(s) are not related by any
symmetry. This suggests that close-packed structures can show interesting differences in order-by-disorder selection.
Read the full abstract ▾
High Energy Physics (Theory)
Interacting multicomponent $SU(N)$ magnets
Alexios P. Polychronakos and Konstantinos Sfetsos
We study magnets composed of molecules with atoms interacting via $SU(N)$-invariant two-body interactions. Atoms carry states transforming in generically different irreducible representations of $SU(N)$, and the interaction strength between atoms depends on the specific atoms involved. We derive the thermodynamics of these magnets in the mean-field approach, and find that they can exhibit mixed ferro-
antiferromagnetic behavior. We present a full analysis of the simplest nontrivial case of molecules consisting of two atoms in the
spin-$j$ representation of $SU(2)$, which generalizes conventional
magnetic systems, and derive its phase structure and transitions.
Read the full abstract ▾
Strongly Correlated Electrons
Chiral Parafermions in Floquet $\mathbb{Z}_N$ Topological Order
Melissa Will, Harald Schmid, Jan Egger et al. (+3 more)
Floquet topological order (FTO) extends
topology beyond
equilibrium, giving rise to phenomena, such as the dynamical transmutation of anyons. Here, we realize $\mathbb{Z}_N$ FTO in a generalized Kitaev honeycomb model with clock degrees of freedom using a three-step periodic drive. The driven system admits an emergent description in terms of constrained parafermions coupled to conserved $\mathbb{Z}_N$ gauge fields. At the boundary, the drive generates
chiral parafermionic Floquet edge modes. We resolve their propagation through correlation functions, characterize their fractional
quantum-information transport through the radical
chiral unitary index, and use their dynamics to probe exchange statistics. In the bulk, we explore the dynamical anyon transmutation with a non-
equilibrium loop order parameter. At a Floquet sweet spot, where the drive becomes Clifford, these phenomena can be accessed using stabilizer simulations. Short-time
matrix-product-state simulations away from this point show that the
chiral edge response persists at accessible times. Our results establish $\mathbb{Z}_N$ Kitaev models as a setting for exploring intrinsically non-
equilibrium topological phenomena on qudit-based
quantum processors.
Read the full abstract ▾
All authors Melissa Will, Harald Schmid, Jan Egger, Adam Gammon-Smith, Michael Knap, Frank Pollmann
High Energy Physics (Phenomenology)
Coherent versus incoherent amplification: from parametric resonance to the kinetic regime in stimulated axion decay
Clemente Smarra and Elisa Todarello
Although the spontaneous decay of
axions into two
photons is extremely slow on cosmological timescales, it can be strongly enhanced in environments with large bosonic occupation numbers. From a classical field perspective, stimulated
axion decay is described by modified Maxwell equations. For a coherent
axion background, these equations exhibit a parametric resonance characterized by a
photon growth exponent $s\propto g_{a\gamma}$. By contrast, a standard kinetic description, for instance based on Boltzmann equations, involves the squared
matrix element of the decay process and therefore would naturally predict a characteristic rate scaling as $s\propto g_{a\gamma}^2$. In this work, we show that these apparently different behaviors originate from the coherence properties of the
axion field. Starting from an
axion field with a finite momentum dispersion, we show how dephasing among its momentum modes modifies the
photon instability and derive an effective
evolution equation in which this effect is encoded in a kernel. For a continuous Maxwell-Boltzmann distribution of
axion velocities, we obtain an implicit expression for the
photon growth exponent that reproduces the two asymptotic regimes and provides an approximate description of the crossover. When amplification develops on scales shorter than the
axion coherence length, the parametric resonance scaling $s\propto g_{a\gamma}$ is recovered, whereas in the opposite limit an effectively kinetic behavior with $s\propto g_{a\gamma}^2$ emerges. Although the detailed form of the crossover depends on the
axion momentum distribution, we expect the asymptotic scaling with $g_{a\gamma}$ to be largely insensitive to the specific distribution assumed.
Read the full abstract ▾