arXiv subjects

Barry T. Chiang

Publications and source records attributed to Barry T. Chiang.

11 recordsLinked to original sources

arXiv2026

How Good Are Frontier Models at Physics? Expert Re-Grading Reveals Broken Evaluations and Near-Saturation of Leading Benchmarks

Low reported scores on leading physics benchmarks, including those featured in the Artificial Analysis Intelligence Index (2026), suggest that frontier language models still struggle with advanced physics, a demanding test of their scientific reasoning and quantitative problem-solving abilities. Yet this impression does not always align with domain experts' experiences using these models in their work. We revisit these reported findings by evaluating frontier models on six widely used physics benchmarks and auditing them with experts, focusing on text-only problems with verifiable final answers. For each subfield of physics, faculty and graduate researchers with relevant expertise carefully review problem statements, reference solutions, and model responses to distinguish genuine model errors from grader errors, incorrect reference solutions, and ambiguous or underspecified questions. Most audited cases initially evaluated as incorrect reflect these benchmarking issues rather than errors in the models' physics reasoning. We then ask experts to address these benchmarking issues by correcting erroneous reference solutions and repairing or excluding flawed questions. We find that GPT-5.6-Sol's measured mean@4 rises from 47.3% to 78.7% on HLE-Physics and from 61.0% to 87.2% on CMT-Benchmark, while its corrected pass@4 reaches 94.4% on the 54 retained CritPt challenges. Corrected scores are computed on the retained evaluation subsets following expert review. Scores on the audited subsets of UGPhysics, PRISM-Physics, and PHYBench also rise substantially after correction. These findings suggest that current benchmarks substantially understate frontier models' ability to solve well-posed physics problems. Near-saturation on these closed-ended tasks highlights the need for more demanding, expert-validated evaluations.

cs.AI↗

arXiv2026

Scatter, bias, and chaos of satellite orbits in triaxial dark matter haloes

The pericentric distances of satellite galaxies govern their tidal stripping, quenching, and survival, yet their orbits are almost universally computed in spherical host potentials, whereas dark matter haloes are generically triaxial. We quantify, orbit by orbit, the error this simplification incurs. We integrate $10^6$ satellites, drawn from a cosmological infall distribution, in static NFW hosts of systematically varying triaxiality at fixed mass profile. Triaxiality leaves the population medians of pericentre, apocentre, and orbital period essentially unchanged. Instead, successive pericentres of an individual orbit scatter by 5-23% depending on host shape, irrespective of orbital energy; the minimum pericentre distance reached within a Hubble time shrinks systematically by up to 21%, enhancing the peak tidal mass loss. This triaxiality-driven orbital dephasing is overwhelmingly regular rather than chaotic; chaotic satellite orbits are common and dominate the near-centre pericentric passages in strongly flattened hosts, but diverge on time-scales far exceeding the Hubble time. Satellite orbits spherically reconstructed via direct backward integration in a truly triaxial host diverge by a tenth of the virial radius within 2-4 Gyr, with comparable uncertainties sourced separately by the unconstrained shape of the host and by its unknown absolute orientation. For the Milky Way dwarf Triangulum II, the unknown halo shape and orientation alone spread the inferred pericentre by ~50%, five times its reported uncertainty and more than ten times the LMC-induced shift, and bias it by 6-70%. This host-shape uncertainty can dominate satellite orbital error budgets and should be incorporated in future inferences.

astro-ph.GA↗

arXiv2026

New Cold Dark Matter Crisis Revealed by Multiscale Cluster Lensing

The properties of substructure in galaxy clusters, exquisitely probed by gravitational lensing, offer a stringent test of dark matter (DM) models. Combining strong- and weak-lensing data for massive clusters, we map their total mass -- dominated by DM -- over the dynamic range needed to confront small-scale predictions for collisionless cold DM (CDM). Using state-of-the-art lens models, we extract four key subhalo properties: the mass function, projected radial distribution, internal density profile, and tidal truncation radius. We find that the subhalo mass functions and truncation radii are consistent with CDM expectations. In contrast, the inner density profiles and radial distributions of subhalos are strongly discrepant with CDM. The incidence of galaxy-galaxy strong lensing from subhalo cores exceeds CDM predictions by nearly an order of magnitude, requiring inner density slopes as steep as $γ\gtrsim 2.5$ within $r \lesssim 0.01R_{200}$ consistent with core-collapsed self-interacting DM (SIDM), while the same subhalos behave as collisionless in their outskirts. Additionally, the observed radial distribution of subhalos hosting bright cluster member galaxies, explicitly modeled in the lens reconstructions, remains incompatible with CDM. Taken together, these small-scale stress tests reveal an intriguing paradox and challenge the DM microphysics of purely collisionless CDM, motivating hybrid scenarios -- such as a dual-component model with both CDM and SIDM or entirely new classes of DM theories.

astro-ph.CO↗

arXiv2026

Statistical Properties of Cold Streams In Massive Star-Forming Halos in TNG50

Cold, dense streams of gas are predicted to penetrate deeply into massive halos (> 10^12 Msun) at cosmic noon (z=4-2), fueling galaxies to sustain high star formation rates. We investigate the prevalence of such cold streams in TNG50 over the range z=4-0, using a novel algorithm to automatically detect cold streams in simulated halos. We qualitatively and quantitatively characterize the geometric and physical properties of the detected streams over cosmic time. We find that cold streams are ubiquitous in massive halos at cosmic noon, occurring in more than 80 percent of such systems down to z=1, before becoming rare by z=0. At their peak prevalence (z=2-1), streams are often found in roughly co-planar, three-stream configurations. These streams generally exhibit a dense and cool core, surrounded by a diffuse and warmer envelope. However, we find that in TNG50, these streams typically disrupt in the outer halo and do not penetrate efficiently to the central galaxy, with the total mass inflow from streams peaking at z=2. Our results underscore the importance of cold streams in fueling galaxies at early times, but they highlight the need for higher-resolution simulations to fully capture their survival and impact at later epochs. Future cosmological zoom-in simulations, with better resolution in the CGM, will be essential to resolve turbulent mixing layers and feedback-inflow interactions that determine whether cold streams can reach the galactic disk.

astro-ph.GA↗

arXiv2026

Universal numerical convergence criteria for subhalo tidal evolution

Dark matter subhalos and satellite galaxies in state-of-the-art cosmological simulations still suffer from the ``overmerging'' problem, where inadequate force and/or mass resolution cause artificially enhanced tidal mass loss and premature disruption. Previous idealized simulations addressing this issue have been restricted to a small subset of the subhalo orbital parameter space, and all assumed subhalos to be isotropic. Here, we present the first extensive simulation suite that quantifies numerical convergence in the tidal evolution of anisotropic subhalos under varying numerical resolutions and orbits. We report a universal force resolution criterion: the subhalo's instantaneous tidal radius must always be resolved by at least 20 cells in adaptive mesh refinement (AMR)-based simulations, or by 20 softening lengths (Plummer equivalent) in tree-based simulations, regardless of refinement details or subhalo physical properties such as concentration or velocity anisotropy. We also report a universal expression for the discreteness-noise-driven scatter in the bound-mass fraction of subhalos that depends only on the subhalo mass resolution at infall and the instantaneous bound mass fraction, agnostic of any further subhalo properties. Such stochastic discreteness noise causes both premature disruption and, notably, spurious survival of poorly mass-resolved subhalos. We demonstrate that as many as 50 percent of all subhalos in state-of-the-art cosmological simulations are likely to be either force and/or mass unresolved. Our findings advocate for adaptive softening or grid refinement based on the instantaneous tidal radius of the subhalo.

astro-ph.CO↗

arXiv2025

Constraining the Nature of Dark Matter from Tidal Radii of Cluster Galaxy Subhalos

Gravitational lensing by galaxy clusters provides a powerful probe of the spatial distribution of dark matter and its microphysical properties. Strong and weak lensing constraints on the density profiles of subhalos and their truncation radii offer key diagnostics for distinguishing between collisionless cold dark matter (CDM) and self-interacting dark matter (SIDM). Notably, in the strongly collisional SIDM regime, subhalo core collapse and enhanced mass loss from ram-pressure stripping predict steeper central density slopes and more compact truncation radii--features that are directly testable with current lensing data. We analyze subhalo truncation in eight lensing clusters (Abell 2218, 383, 963, 209, 2390, and MACS J0416.1, J1206.2, J1149.6) that span the redshift range <$z_\text{spec}$>$ \simeq 0.17$-$0.54$ with virial masses $M_{200} \simeq0.41$-$2.2\times 10^{15}$ M$_\odot$ to constrain SIDM versus CDM. Our results indicate that the outer spatial extents of subhalos are statistically consistent with CDM, corroborated by redshift- and mass-matched analogs from the Illustris-TNG simulations. We conclude that the tidal radii of cluster galaxy subhalos serve as an important and complementary diagnostic of the nature of dark matter in these violent, dense environments.

astro-ph.CO↗

arXiv2025

The tidal evolution of anisotropic subhaloes: A new pathway to creating isotropic and cored satellites

It is common practice, both in dynamical modelling and in idealised numerical simulations, to assume that galaxies and/or dark matter haloes are spherical and have isotropic velocity distributions, such that their distribution functions are ergodic. However, there is no good reason to assume that this assumption is accurate. In this paper we use idealised $N$-body simulations to study the tidal evolution of subhaloes that are anisotropic at infall. We show that the detailed velocity anisotropy has a large impact on the subhalo's mass loss rate. In particular, subhaloes that are radially anisotropic experience much more mass loss than their tangentially anisotropic counterparts. In fact, in the former case, the stripping of highly radial orbits can cause a rapid cusp-to-core transformation, without having to resort to any baryonic feedback processes. Once the tidal radius becomes comparable to the radius of the core thus formed, the subhalo is tidally disrupted. Subhaloes that at infall are tangentially anisotropic are far more resilient to tidal stripping, and are never disrupted when simulated with sufficient resolution. We show that the preferential stripping of more radial orbits, combined with re-virialisation post stripping, causes an isotropisation of the subhalo's velocity distributions. This implies that subhaloes that have experienced significant mass loss are expected to be close to isotropic, which may alleviate the mass-anisotropy degeneracies that hamper the dynamical modelling of Milky Way satellites.

astro-ph.GA↗

arXiv2024

Galactic disc heating by density granulation in fuzzy dark matter simulations

Fuzzy dark matter (FDM), an attractive dark matter candidate comprising ultralight bosons (axions) with a particle mass $m_a\sim10^{-22}$ eV, is motivated by the small-scale challenges of cold dark matter and features a kpc-size de Broglie wavelength. Quantum wave interference inside an FDM halo gives rise to stochastically fluctuating density granulation; the resulting gravitational perturbations could drive significant disc thickening, providing a natural explanation for galactic thick discs. Here we present the first self-consistent simulations of FDM haloes and stellar discs, exploring $m_a=0.2-1.2\times10^{-22}$ eV and halo masses $M_\text{h} = 0.7-2.8\times10^{11}$ M$_\odot$. Disc thickening is observed in all simulated systems. The disc heating rates are approximately constant in time and increase substantially with decreasing $m_a$, reaching $dh/dt \simeq 0.04$ ($0.4$) kpc Gyr$^{-1}$ and $dσ_z^2/dt \simeq4$ ($150$) km$^2$s$^{-2}$Gyr$^{-1}$ for $m_a=1.2$ ($0.2$) $\times10^{-22}$ eV and $M_\text{h} =7\times10^{10} \text{M}_\odot$, where $h$ is the disc scale height and $σ_z$ is the vertical velocity dispersion. These simulated heating rates agree within a factor of two with the theoretical estimates of Chiang et al., confirming that the rough estimate of Church et al. overpredicts the granulation-driven disc heating rate by two orders of magnitude. However, the simulation-inferred heating rates scale less steeply than the theoretically predicted relation $dσ^2_z/dt \propto m_a^{-3}$. Finally, we examine the applicability of the Fokker-Planck approximation in FDM granulation modelling and the robustness of the $m_a$ exclusion bound derived from the Galactic disc kinematics.

astro-ph.GA↗

arXiv2022

Can ultralight dark matter explain the age-velocity dispersion relation of the Milky Way disc: A revised and improved treatment

Ultralight axion-like particles $m_a \sim 10^{-22}$ eV, or Fuzzy Dark Matter (FDM), behave comparably to cold dark matter (CDM) on cosmological scales and exhibit a kpc-size de Broglie wavelength capable of alleviating established (sub-)galactic-scale problems of CDM. Substructures inside an FDM halo incur gravitational potential perturbations, resulting in stellar heating sufficient to account for the Galactic disc thickening over a Hubble time, as first demonstrated by Church et al. We present a more sophisticated treatment that incorporates the full baryon and dark matter distributions of the Milky Way and adopts stellar disc kinematics inferred from recent Gaia, APOGEE, and LAMOST surveys. Ubiquitous density granulation and subhalo passages respectively drive inner disc thickening and flaring of the outer disc, resulting in an observationally consistent `U-shaped' disc vertical velocity dispersion profile with the global minimum located near the solar radius. The observed age-velocity dispersion relation in the solar vicinity can be explained by the FDM-substructure-induced heating and places an exclusion bound $m_a \gtrsim 0.4\times10^{-22}$ eV. We assess non-trivial uncertainties in the empirical core-halo relation, FDM subhalo mass function and tidal stripping, and stellar heating estimate. The mass range $m_a\simeq 0.5-0.7\times10^{-22}$ eV favoured by the observed thick disc kinematics is in tension with several exclusion bounds inferred from dwarf density profiles, stellar streams, and Milky Way satellite populations, which could be significantly relaxed due to the aforesaid uncertainties. Additionally, strongly anisotropic heating could help explain the formation of ultra-thin disc galaxies.

astro-ph.GA↗

arXiv2021

Soliton Oscillations and Revised Constraints from Eridanus II of Fuzzy Dark Matter

Fuzzy dark matter (FDM) has been a promising alternative to standard cold dark matter. The model consists of ultralight bosons with mass $m_b \sim 10^{-22}$ eV and features a quantum-pressure-supported solitonic core that oscillates. In this work, we show that the soliton density oscillations persist even after significant tidal stripping of the outer halo. We report two intrinsic yet distinct timescales associated, respectively, with the ground-state soliton wavefunction $τ_{00}$ and the soliton density oscillations $τ_\text{soliton}$, obeying $τ_\text{soliton} /τ_{00} \simeq 2.3$. The central star cluster (SC) in Eridanus II has a characteristic timescale $τ_\text{soliton} / τ_\text{SC} \sim 2$ to $3$ that deviates substantially from unity. As a result, we demonstrate, both analytically and numerically with three-dimensional self-consistent FDM simulations, that the gravitational heating of the SC owing to soliton density oscillations is negligible irrespective of $m_b$. We also show that the subhalo mass function to form Eridanus II does not place a strong constraint on $m_b$. These results are contrary to the previous findings by Marsh & Niemeyer (2019).

astro-ph.CO↗

arXiv2020

Faint Dark Matter Annihilation Signals and the Milky Way's Supermassive Black Hole

A wide range of mechanisms predict present-day s-wave dark matter (DM) annihilation cross-sections that are orders of magnitude below current experimental sensitivity. We explore the capability of DM density spikes around the Milky Way's supermassive black hole to probe such faint signals of DM annihilations, considering a range of possible spike and halo distributions. As an exemplar of a theory with a suppressed s-wave annihilation cross-section, we consider a hidden sector axion portal model of DM. In this model, the leading contribution to the annihilation cross-section in the early universe is p-wave, while s-wave annihilations occur at higher order in the coupling constant. We provide a unified treatment of DM freezeout in this model including both s- and p-wave annihilations and analytically determine the photon spectrum for the dominant DM annihilation process in the universe today. We find that Fermi and H.E.S.S. observations of the Galactic Center offer excellent sensitivity to this model over a wide range of parameter space, with prospects depending sensitively on the properties of the DM spike as well as the central halo.

hep-ph↗