Search

In this paper we prove that there exists a compact perturbation of the Ricci flat Taub-Bolt metric that evolves under the Ricci flow into a finite time singularity modelled on the shrinking solition FIK [5]. Moreover, this perturbation can be made arbitrarily L^2-small with respect to the Taub-Bolt metric. The method of proof closely follows the strategy adopted in [14], where the author constructs, via a Wa\.zewski box argument, Ricci flows on compact manifolds which encounter finite singularities modelled on a given asymptotically conical shrinking soliton., Comment: arXiv admin note: text overlap with arXiv:2202.03386 by other authors

The integration of new modalities into frontier AI systems offers exciting capabilities, but also increases the possibility such systems can be adversarially manipulated in undesirable ways. In this work, we focus on a popular class of vision-language models (VLMs) that generate text outputs conditioned on visual and textual inputs. We conducted a large-scale empirical study to assess the transferability of gradient-based universal image "jailbreaks" using a diverse set of over 40 open-parameter VLMs, including 18 new VLMs that we publicly release. Overall, we find that transferable gradient-based image jailbreaks are extremely difficult to obtain. When an image jailbreak is optimized against a single VLM or against an ensemble of VLMs, the jailbreak successfully jailbreaks the attacked VLM(s), but exhibits little-to-no transfer to any other VLMs; transfer is not affected by whether the attacked and target VLMs possess matching vision backbones or language models, whether the language model underwent instruction-following and/or safety-alignment training, or many other factors. Only two settings display partially successful transfer: between identically-pretrained and identically-initialized VLMs with slightly different VLM training data, and between different training checkpoints of a single VLM. Leveraging these results, we then demonstrate that transfer can be significantly improved against a specific target VLM by attacking larger ensembles of "highly-similar" VLMs. These results stand in stark contrast to existing evidence of universal and transferable text jailbreaks against language models and transferable adversarial attacks against image classifiers, suggesting that VLMs may be more robust to gradient-based transfer attacks.

The proliferation of generative models, combined with pretraining on web-scale data, raises a timely question: what happens when these models are trained on their own generated outputs? Recent investigations into model-data feedback loops proposed that such loops would lead to a phenomenon termed model collapse, under which performance progressively degrades with each model-data feedback iteration until fitted models become useless. However, those studies largely assumed that new data replace old data over time, where an arguably more realistic assumption is that data accumulate over time. In this paper, we ask: what effect does accumulating data have on model collapse? We empirically study this question by pretraining sequences of language models on text corpora. We confirm that replacing the original real data by each generation's synthetic data does indeed tend towards model collapse, then demonstrate that accumulating the successive generations of synthetic data alongside the original real data avoids model collapse; these results hold across a range of model sizes, architectures, and hyperparameters. We obtain similar results for deep generative models on other types of real data: diffusion models for molecule conformation generation and variational autoencoders for image generation. To understand why accumulating data can avoid model collapse, we use an analytically tractable framework introduced by prior work in which a sequence of linear models are fit to the previous models' outputs. Previous work used this framework to show that if data are replaced, the test error increases with the number of model-fitting iterations; we extend this argument to prove that if data instead accumulate, the test error has a finite upper bound independent of the number of iterations, meaning model collapse no longer occurs.

Simulations of pre-supernova evolution suggest that some intense shell burning can be so active that, in extreme cases, it can merge with the outer shell, changing the initial conditions for the supernova explosion. However, such violent activity in the interior of stars has been difficult to confirm from observations of stars. Here we propose that the elemental composition of O-rich ejecta in supernova remnants can be a tool to test for this kind of intense shell burning activity in the final stages of progenitor evolution. As an example, we discuss the origin of "Mg-rich" ejecta in the supernova remnant N49B. A high Mg/Ne mass ratio $\gtrsim 1$ suggests that the Ne- or O-burning shell has broken into or merged with the outer shell before the collapse. Such Mg-rich (or Ne-poor) ejecta has been identified in some other supernova remnants, supporting the idea that some destratification process, such as a shell merger, does indeed occur in the interiors of some massive stars, although they may not be the majority. Our results suggest that X-ray observations of O-rich ejecta in core-collapse supernova remnants will be a unique tool to probe the shell burning activity during the final stage of a massive star's interior., Comment: Submitted to ApJ and received positive peer review. This is the version in the first submission. 11 pages, 7 figures, 2 tables

The progenitor of the W49B supernova remnant is still under debate. One of the candidates is a jet-driven core-collapse supernova. In such a highly asymmetric explosion, a strong $\alpha$-rich freezeout is expected in local high entropy regions, which should enrich elements synthesized by the capture of $\alpha$-particles such as $^{44}$Ti and $^{48}$Cr (decaying to $^{44}$Ca and $^{48}$Ti, respectively). In the present work, in order to infer the progenitor of the W49B remnant, we constrain the amount of stable Ti ($^{48}$Ti) synthesized, using the {\it Suzaku} observation. We found no firm evidence for the Ti line and set the upper limit of $M_{\rm Ti}/M_{\rm Fe} < 8.2 \times$ 10$^{-4}$ (99\% limit using Xspec) and $M_{\rm Ti}/M_{\rm Fe} < 1.9 \times$ 10$^{-3}$ (99\% limit using SPEX), and thus excluded almost all hypernova/jet-driven supernova models. Our results, as complemented by some previous studies, suggest that a Type Ia supernova from a near-$M_{\rm Ch}$ (Chandrasekhar mass) white dwarf is the most favorable candidate for the origin of W49B. Future observations with X-ray calorimeter missions, such as XRISM, will give us a stronger constraint on the progenitor., Comment: Submitted to ApJ, 7 pages, 3 figures, 1 table

Common methods for aligning large language models (LLMs) with desired behaviour heavily rely on human-labelled data. However, as models grow increasingly sophisticated, they will surpass human expertise, and the role of human evaluation will evolve into non-experts overseeing experts. In anticipation of this, we ask: can weaker models assess the correctness of stronger models? We investigate this question in an analogous setting, where stronger models (experts) possess the necessary information to answer questions and weaker models (non-experts) lack this information. The method we evaluate is debate, where two LLM experts each argue for a different answer, and a non-expert selects the answer. We find that debate consistently helps both non-expert models and humans answer questions, achieving 76% and 88% accuracy respectively (naive baselines obtain 48% and 60%). Furthermore, optimising expert debaters for persuasiveness in an unsupervised manner improves non-expert ability to identify the truth in debates. Our results provide encouraging empirical evidence for the viability of aligning models with debate in the absence of ground truth., Comment: For code please check: https://github.com/ucl-dark/llm_debate

Understanding the behaviour of a system's API can be hard. Giving users access to relevant examples of how an API behaves has been shown to make this easier for them. In addition, such examples can be used to verify expected behaviour or identify unwanted behaviours. Methods for automatically generating examples have existed for a long time. However, state-of-the-art methods rely on either white-box information, such as source code, or on formal specifications of the system behaviour. But what if you do not have access to either? e.g., when interacting with a third-party API. In this paper, we present an approach to automatically generate relevant examples of behaviours of an API, without requiring either source code or a formal specification of behaviour. Evaluation on an industry-grade REST API shows that our method can produce small and relevant examples that can help engineers to understand the system under exploration.

Nebular-phase observations of peculiar Type Ia supernovae (SNe Ia) provide important constraints on progenitor scenarios and explosion dynamics for both these rare SNe and the more common, cosmologically useful SNe Ia. We present observations from an extensive ground-based and space-based follow-up campaign to characterize SN 2022pul, a "super-Chandrasekhar" mass SN Ia (alternatively "03fg-like" SN), from before peak brightness to well into the nebular phase across optical to mid-infrared (MIR) wavelengths. The early rise of the light curve is atypical, exhibiting two distinct components, consistent with SN Ia ejecta interacting with dense carbon-oxygen rich circumstellar material (CSM). In the optical, SN 2022pul is most similar to SN 2012dn, having a low estimated peak luminosity ($M_{B}=-18.9$ mag) and high photospheric velocity relative to other 03fg-like SNe. In the nebular phase, SN 2022pul adds to the increasing diversity of the 03fg-like subclass. From 168 to 336 days after peak $B$-band brightness, SN 2022pul exhibits asymmetric and narrow emission from [O I] $\lambda\lambda 6300,\ 6364$ (${\rm FWHM} \approx 2{,}000$ km s$^{-1}$), strong, broad emission from [Ca II] $\lambda\lambda 7291,\ 7323$ (${\rm FWHM} \approx 7{,}300$ km s$^{-1}$), and a rapid Fe III to Fe II ionization change. Finally, we present the first-ever optical-to-mid-infrared (MIR) nebular spectrum of an 03fg-like SN Ia using data from JWST. In the MIR, strong lines of neon and argon, weak emission from stable nickel, and strong thermal dust emission (with $T \approx 500$ K), combined with prominent [O I] in the optical, suggest that SN 2022pul was produced by a white dwarf merger within carbon/oxygen-rich CSM., Comment: 23 pages, 11 figures, submitted to ApJ

We present an analysis of ground-based and JWST observations of SN~2022pul, a peculiar "03fg-like" (or "super-Chandrasekhar") Type Ia supernova (SN Ia), in the nebular phase at 338d post explosion. Our combined spectrum continuously covers 0.4--14 $\mu$m and includes the first mid-infrared spectrum of an 03fg-like SN Ia. Compared to normal SN Ia 2021aefx, SN 2022pul exhibits a lower mean ionization state, asymmetric emission-line profiles, stronger emission from the intermediate-mass elements (IMEs) argon and calcium, weaker emission from iron-group elements (IGEs), and the first unambiguous detection of neon in a SN Ia. Strong, broad, centrally peaked [Ne II] line at 12.81 $\mu$m was previously predicted as a hallmark of "violent merger'' SN Ia models, where dynamical interaction between two sub-$M_{ch}$ white dwarfs (WDs) causes disruption of the lower mass WD and detonation of the other. The violent merger scenario was already a leading hypothesis for 03fg-like SNe Ia; in SN 2022pul it can explain the large-scale ejecta asymmetries seen between the IMEs and IGEs and the central location of narrow oxygen and broad neon. We modify extant models to add clumping of the ejecta to better reproduce the optical iron emission, and add mass in the innermost region ($< 2000$ km s$^{-1}$) to account for the observed narrow [O I]~$\lambda\lambda6300$, 6364 emission. A violent WD-WD merger explains many of the observations of SN 2022pul, and our results favor this model interpretation for the subclass of 03fg-like SN Ia., Comment: 20 pages, 10 figures, published in ApJ

Neutrino interactions with stellar material are widely believed to be fundamental to the explosion of massive stars. However, this important process has remained difficult to confirm observationally. We propose a new method to verify it using X-ray observations of the supernova remnant Cassiopeia A. The elemental composition in its Fe-rich ejecta that could have been produced at the innermost region of the supernova, where neutrinos are expected to interact, allows us to examine the presence of neutrino interactions. Here we demonstrate that the amount of Mn produced without neutrino nucleosynthesis processes (i.e., the $\nu$- and $\nu$p-process) is too small to explain the Mn/Fe mass ratio we measure (0.14--0.67\%). This result supports the operation of significant neutrino interactions in the Cassiopeia A supernova. If the observed Mn/Fe mass ratio purely reflects the production at the innermost region of the supernova, this would be the first robust confirmation of neutrino-matter interactions in an individual supernova. We further show that the Mn/Fe mass ratio has the potential to constrain supernova neutrino parameters (i.e., total neutrino luminosity, neutrino temperature). Future spatially-resolved, high-resolution X-ray spectroscopy will allow us to investigate the details of neutrino-supernova astrophysics through its signatures in elemental composition not only in Cassiopeia A but also in other remnants., Comment: 17 pages, 10 figures. Accepted for publication in ApJ

Observations of the millimeter sky contain valuable information on a number of signals, including the blackbody cosmic microwave background (CMB), Galactic emissions, and the Compton-$y$ distortion due to the thermal Sunyaev-Zel'dovich (tSZ) effect. Extracting new insight into cosmological and astrophysical questions often requires combining multi-wavelength observations to spectrally isolate one component. In this work, we present a new arcminute-resolution Compton-$y$ map, which traces out the line-of-sight-integrated electron pressure, as well as maps of the CMB in intensity and E-mode polarization, across a third of the sky (around 13,000 sq.~deg.). We produce these through a joint analysis of data from the Atacama Cosmology Telescope (ACT) Data Release 4 and 6 at frequencies of roughly 93, 148, and 225 GHz, together with data from the \textit{Planck} satellite at frequencies between 30 GHz and 545 GHz. We present detailed verification of an internal linear combination pipeline implemented in a needlet frame that allows us to efficiently suppress Galactic contamination and account for spatial variations in the ACT instrument noise. These maps provide a significant advance, in noise levels and resolution, over the existing \textit{Planck} component-separated maps and will enable a host of science goals including studies of cluster and galaxy astrophysics, inferences of the cosmic velocity field, primordial non-Gaussianity searches, and gravitational lensing reconstruction of the CMB., Comment: The Compton-y map and associated products will be made publicly available upon publication of the paper. The CMB T and E mode maps will be made available when the DR6 maps are made public

The community agrees that Type Ia supernovae arise from Carbon/Oxygen white dwarfs undergoing thermonuclear runaway. However, the full progenitor system and the process that prompts the white dwarf to explode remain unknown. Most current models suggest that the white dwarf explodes because of interaction with a binary companion which may survive the process and remain within the resulting remnant of the exploded star. Furthermore, both the pre-supernova interaction process and the explosion of the primary are expected to imprint a significant departure from ordinary stellar radii and temperatures onto the secondary, making the star identifiable against the unrelated stellar population. Identification of a surviving companion inside an SN Ia remnant might confirm a specific corresponding SN Ia progenitor channel based on the identity of the companion. We conducted a surviving companion search of the Type Ia remnant SNR 0509-67.5 based in the Large Magellanic Cloud. The well-constrained distance to and foreground extinction of the Large Magellanic Cloud allow for Bayesian inference of stellar parameters with low correlation and uncertainties. We present a deep catalog of fully characterized stars interior to SNR 0509-67.5 with radii, effective temperatures, and metallicities inferred using combined Hubble Space Telescope photometric observations across multiple visits. We then compile a list of surviving companion models appropriate for the age of the remnant (roughly 400 years after the explosion). We compare these predictions with the inferred stellar parameters and conclude that none of the stars are consistent with the predicted signatures of a surviving companion., Comment: 16 pages, 8 figures, 4 tables