Search arXivSearch

arXiv subjects

Fergus Simpson

Publications and source records attributed to Fergus Simpson.

At least 19 recordsLinked to original sources

From Evidence to Evident: Decisive Cosmological Evidence for the Normal Neutrino Mass Hierarchy

Cosmological data have reached the precision needed to turn the neutrino mass ordering from a weak Bayesian preference into a decisive model-selection test. We compute the evidence for the Normal and Inverted Hierarchies by combining DESI DR2 clustering with NuFIT oscillation data. In baseline $\Lambda$CDM, DESI DR2 plus Planck CamSpec gives $\Sigma m_\nu<0.0642\,{\rm eV}$ at 95\% confidence, close to the normal-ordering floor, $\Sigma m_\nu^{\rm NH}\simeq0.059\,{\rm eV}$, but well below the inverted-ordering minimum, $\Sigma m_\nu^{\rm IH}\simeq0.099\,{\rm eV}$. Thus the inverted hierarchy lies in the tail of the cosmological likelihood. The Bayes factor $K=P(D|{\rm NH})/P(D|{\rm IH})$ exceeds $460$ even for a conservative reference prior, and remains strong, $K>40$, in baseline-model extensions. We show that this result is robust to the choice between a reference prior and a physically motivated logarithmic hierarchical prior, marking the transition from {\em prior-sensitive evidence} to {\em likelihood-dominated exclusion} of the inverted hierarchy within standard cosmology. Embedding these priors in the two-dimensional design space of measure (logarithmic versus linear in mass) and structure (hierarchical versus non-hierarchical), we find that all four prior constructions give decisive evidence under DESI DR2, with residual prior dependence governed mainly by the measure -- a factor $\sim\!10$ in $K$ -- rather than by the hierarchy assumption. At the prior-family level, the evidence favors the SJPV prior predictive over HS by a Bayes factor above $4,700$ across each matched-support variation tested. The favored normal ordering pushes the effective Majorana mass to the few-meV regime, with median $m_{\beta\beta}=3.28\,{\rm meV}$ and 95\% credible interval $0.95<m_{\beta\beta}<11.55\,{\rm meV}$, below the inverted-ordering target for upcoming neutrinoless double-beta decay experiments.

astro-ph.CO

The Coherence Principle: A Falsifiable Prior for Model Selection from the Grammar of Theories

Bayesian model selection in cosmology and particle physics is often performed where posterior odds inherit a strong, often unacknowledged dependence on the prior assigned to competing models. Standard responses -- reference priors, hierarchical priors, or appeals to naturalness -- ignore relevant theoretical knowledge or rely on criteria hard to define operationally. We propose the \emph{Coherence Principle}: a reproducible prescription for assigning model priors according to compatibility with the validated structure of an existing theory. This structure, or \emph{grammar}, includes symmetries, conservation laws, locality, Lorentz invariance, and universality patterns. Unmotivated violations of these rules incur a coherence cost, converted into a prior weight through a maximum-entropy exponential form controlled by one calibratable parameter $\alpha$. The resulting prior is distinct from both the Bayesian Occam factor and naturalness: it penalizes not parameter volume or fine tuning, but departures from validated theoretical grammar. We illustrate the principle with examples from cosmology and fundamental physics: neutrino mass mechanisms, dark energy and modified gravity, inflation, beyond-Standard-Model sectors, and hierarchical astrophysical inference. We test it also on four historical cases -- general relativity, Pauli's neutrino, parity violation, and special relativity -- where evidential and theoretical contexts can be reconstructed. These examples show that it favors the historically successful choice when the proper grammar is defined in the correct domain and time. The Coherence Principle makes explicit a common but usually tacit part of physical reasoning: trust in validated structural rules. It turns this judgment into a transparent, testable, and overrulable component of Bayesian inference, leaving empirical likelihoods free to dominate when data are sufficiently constraining.

astro-ph.CO

Neural Diffusion Processes

Neural network approaches for meta-learning distributions over functions have desirable properties such as increased flexibility and a reduced complexity of inference. Building on the successes of denoising diffusion models for generative modelling, we propose Neural Diffusion Processes (NDPs), a novel approach that learns to sample from a rich distribution over functions through its finite marginals. By introducing a custom attention block we are able to incorporate properties of stochastic processes, such as exchangeability, directly into the NDP's architecture. We empirically show that NDPs can capture functional distributions close to the true Bayesian posterior, demonstrating that they can successfully emulate the behaviour of Gaussian processes and surpass the performance of neural processes. NDPs enable a variety of downstream tasks, including regression, implicit hyperparameter marginalisation, non-Gaussian posterior prediction and global optimisation.

stat.ML

Neutrino Masses and Mass Hierarchy: Evidence for the Normal Hierarchy

The latest cosmological constraints on the sum of neutrino masses, in combination with the latest laboratory measurements on oscillations, provide ``decisive" Bayesian evidence for the normal neutrino mass hierarchy. We show that this result holds across very different prior alternatives by exploring two extremes on the range of prior choices. In fact, while the specific numerical value for the Evidence depends on the choice of prior, the Bayesian odds remain greater than 140:1 across very different prior choices. For Majorana neutrinos this has important implications for the upper limit of the neutrino-less double beta decay half life and thus for the technology and resources needed for future double beta decay experiments.

hep-ph

Validating Gaussian Process Models with Simulation-Based Calibration

Gaussian process priors are a popular choice for Bayesian analysis of regression problems. However, the implementation of these models can be complex, and ensuring that the implementation is correct can be challenging. In this paper we introduce Gaussian process simulation-based calibration, a procedure for validating the implementation of Gaussian process models and demonstrate the efficacy of this procedure in identifying a bug in existing code. We also present a novel application of this procedure to identify when marginalisation of the model hyperparameters is necessary.

cs.LG

Kernel Identification Through Transformers

Kernel selection plays a central role in determining the performance of Gaussian Process (GP) models, as the chosen kernel determines both the inductive biases and prior support of functions under the GP prior. This work addresses the challenge of constructing custom kernel functions for high-dimensional GP regression models. Drawing inspiration from recent progress in deep learning, we introduce a novel approach named KITT: Kernel Identification Through Transformers. KITT exploits a transformer-based architecture to generate kernel recommendations in under 0.1 seconds, which is several orders of magnitude faster than conventional kernel search algorithms. We train our model using synthetic data generated from priors over a vocabulary of known kernels. By exploiting the nature of the self-attention mechanism, KITT is able to process datasets with inputs of arbitrary dimension. We demonstrate that kernels chosen by KITT yield strong performance over a diverse collection of regression benchmarks.

stat.ML

The Minecraft Kernel: Modelling correlated Gaussian Processes in the Fourier domain

In the univariate setting, using the kernel spectral representation is an appealing approach for generating stationary covariance functions. However, performing the same task for multiple-output Gaussian processes is substantially more challenging. We demonstrate that current approaches to modelling cross-covariances with a spectral mixture kernel possess a critical blind spot. For a given pair of processes, the cross-covariance is not reproducible across the full range of permitted correlations, aside from the special case where their spectral densities are of identical shape. We present a solution to this issue by replacing the conventional Gaussian components of a spectral mixture with block components of finite bandwidth (i.e. rectangular step functions). The proposed family of kernel represents the first multi-output generalisation of the spectral mixture kernel that can approximate any stationary multi-output kernel to arbitrary precision.

stat.ML

Marginalised Gaussian Processes with Nested Sampling

Gaussian Process (GPs) models are a rich distribution over functions with inductive biases controlled by a kernel function. Learning occurs through the optimisation of kernel hyperparameters using the marginal likelihood as the objective. This classical approach known as Type-II maximum likelihood (ML-II) yields point estimates of the hyperparameters, and continues to be the default method for training GPs. However, this approach risks underestimating predictive uncertainty and is prone to overfitting especially when there are many hyperparameters. Furthermore, gradient based optimisation makes ML-II point estimates highly susceptible to the presence of local minima. This work presents an alternative learning procedure where the hyperparameters of the kernel function are marginalised using Nested Sampling (NS), a technique that is well suited to sample from complex, multi-modal distributions. We focus on regression tasks with the spectral mixture (SM) class of kernels and find that a principled approach to quantifying model uncertainty leads to substantial gains in predictive performance across a range of synthetic and benchmark data sets. In this context, nested sampling is also found to offer a speed advantage over Hamiltonian Monte Carlo (HMC), widely considered to be the gold-standard in MCMC based inference.

stat.ML

KiDS-450: Enhancing cosmic shear with clipping transformations

We present the first "clipped" cosmic shear measurement using data from the Kilo-Degree Survey (KiDS-450). "Clipping" transformations suppress the signal from the highest density, non-linear regions of cosmological fields. We demonstrate that these transformations improve constraints on $S_8=\sigma_8(\Omega_{\rm{m}}/0.3)^{0.5}$ when used in combination with conventional two-point statistics. For the KiDS-450 data, we find that the combined measurements improve the constraints on $S_8$ by 17%, compared to shear correlation functions alone. We determine the expectation value of the clipped shear correlation function using a suite of numerical simulations, and develop methodology to mitigate the impact of masking and shot noise. Future improvements in numerical simulations and mass reconstruction methodology will permit the precise calibration of clipped cosmic shear statistics such that clipping can become a standard tool in weak lensing analyses.

astro-ph.CO

Strong Bayesian Evidence for the Normal Neutrino Hierarchy

The configuration of the three neutrino masses can take two forms, known as the normal and inverted hierarchies. We compute the Bayesian evidence associated with these two hierarchies. Previous studies found a mild preference for the normal hierarchy, and this was driven by the asymmetric manner in which cosmological data has confined the available parameter space. Here we identify the presence of a second asymmetry, which is imposed by data from neutrino oscillations. By combining constraints on the squared-mass splittings with the limit on the sum of neutrino masses of $\Sigma m_\nu < 0.13$ eV, and using a minimally informative prior on the masses, we infer odds of 42:1 in favour of the normal hierarchy, which is classified as "strong" in the Jeffreys' scale. We explore how these odds may evolve in light of higher precision cosmological data, and discuss the implications of this finding with regards to the nature of neutrinos. Finally the individual masses are inferred to be $m_1 = 3.80^{+26.2}_{-3.73} \, \text{meV}, m_2 = 8.8^{+18}_{-1.2} \, \text{meV}, m_3 = 50.4^{+5.8}_{-1.2} \, \text{meV}$ ($95\%$ credible intervals).

astro-ph.CO

Apocalypse Now? Reviving the Doomsday Argument

Whether the fate of our species can be forecast from its past has been the topic of considerable controversy. One refutation of the so-called Doomsday Argument is based on the premise that we are more likely to exist in a universe containing a greater number of observers. Here we present a Bayesian reformulation of the Doomsday Argument which is immune to this effect. By marginalising over the spatial configuration of observers, we find that any preference for a larger total number of observers has no impact on the inferred local number. Our results remain unchanged when we adopt either the Self-Indexing Assumption (SIA) or the Self-Sampling Assumption (SSA). Furthermore the median value of our posterior distribution is found to be in agreement with the frequentist forecast. Humanity's prognosis for the coming century is well approximated by a global catastrophic risk of 0.2% per year.

stat.OT

Bayesian evidence for the prevalence of waterworlds

Should we expect most habitable planets to share the Earth's marbled appearance? For a planetary surface to boast extensive areas of both land and water, a delicate balance must be struck between the volume of water it retains and the capacity of its perturbations. These two quantities may show substantial variability across the full spectrum of water-bearing worlds. This would suggest that, barring strong feedback effects, most surfaces are heavily dominated by either water or land. Why is the Earth so finely poised? To address this question we construct a simple model for the selection bias that would arise within an ensemble of surface conditions. Based on the Earth's ocean coverage of 71%, we find substantial evidence (Bayes factor K ~ 6) supporting the hypothesis that anthropic selection effects are at work. Furthermore, due to the Earth's proximity to the waterworld limit, this model predicts that most habitable planets are dominated by oceans spanning over 90% of their surface area (95% credible interval). This scenario, in which the Earth has a much greater land area than most habitable planets, is consistent with results from numerical simulations and could help explain the apparently low-mass transition in the mass-radius relation.

astro-ph.EP

Dark energy from the motions of neutrinos

We demonstrate that a scalar field is unable to reverse its direction of motion while continuously exchanging energy with another fluid. If the rate of transfer is modulated by the scalar's acceleration, the field can undergo a rapid process of freezing, despite being displaced from the local minimum of its potential. This enables dark energy to form from any potential, regardless of its shape. The field's equation of state mimicks that of a cosmological constant. We present a physically motivated realisation in the form of a derivative neutrino-majoron coupling. Coherent motions, which form only once the neutrinos become non-relativistic, could be responsible for instigating the freezing process. This would provide a natural resolution to the dark energy coincidence problem, while avoiding the dynamical instabilities associated with mass-varying neutrino models. Finally we discuss possible means by which this model could be experimentally verified.

astro-ph.CO

Cosmology and Fundamental Physics with the Euclid Satellite

Euclid is a European Space Agency medium class mission selected for launch in 2020 within the Cosmic Vision 2015 2025 program. The main goal of Euclid is to understand the origin of the accelerated expansion of the universe. Euclid will explore the expansion history of the universe and the evolution of cosmic structures by measuring shapes and redshifts of galaxies as well as the distribution of clusters of galaxies over a large fraction of the sky. Although the main driver for Euclid is the nature of dark energy, Euclid science covers a vast range of topics, from cosmology to galaxy evolution to planetary research. In this review we focus on cosmology and fundamental physics, with a strong emphasis on science beyond the current standard models. We discuss five broad topics: dark energy and modified gravity, dark matter, initial conditions, basic assumptions and questions of methodology in the data analysis. This review has been planned and carried out within Euclid's Theory Working Group and is meant to provide a guide to the scientific themes that will underlie the activity of the group during the preparation of the Euclid mission.

astro-ph.CO

Structure formation simulations with momentum exchange: alleviating tensions between high-redshift and low-redshift cosmological probes

Persisting tensions between the cosmological constraints derived from low-redshift probes and the ones obtained from temperature and polarisation anisotropies of the Cosmic Microwave Background -- although not yet providing compelling evidence against the $\Lambda $CDM model -- seem to consistently indicate a slower growth of density perturbations as compared to the predictions of the standard cosmological scenario. Such behavior is not easily accommodated by the simplest extensions of General Relativity, such as f(R) models, which generically predict an enhanced growth rate. In the present work we present the outcomes of a suite of large N-body simulations carried out in the context of a cosmological model featuring a non-vanishing scattering cross section between the dark matter and the dark energy fields, for two different parameterisations of the dark energy equation of state. Our results indicate that these Dark Scattering models have very mild effects on many observables related to large-scale structures formation and evolution, while providing a significant suppression of the amplitude of linear density perturbations and the abundance of massive clusters. Our simulations therefore confirm that these models offer a promising route to alleviate existing tensions between low-redshift measurements and those of the CMB.

astro-ph.CO

The longevity of habitable planets and the development of intelligent life

Why did the emergence of our species require a timescale similar to the entire habitable period of our planet? Our late appearance has previously been interpreted by Carter (2008) as evidence that observers typically require a very long development time, implying that intelligent life is a rare occurrence. Here we present an alternative explanation, which simply asserts that many planets possess brief periods of habitability. We also propose that the rate-limiting step for the formation of observers is the enlargement of species from an initially microbial state. In this scenario the development of intelligent life is a slow but almost inevitable process, greatly enhancing the prospects of future SETI experiments such as the Breakthrough Listen project.

astro-ph.EP

Cosmic explosions, life in the Universe and the Cosmological Constant

Galactic Gamma-Ray Bursts (GRBs) are copious sources of gamma-rays that can pose a threat to complex life. Using recent determinations of their rate and the probability of GRBs causing massive extinction, we explore what type of universes are most likely to harbour advanced forms of life. For this purpose we use cosmological N-body simulations to determine at what time and for what value of the cosmological constant ($\Lambda$) the chances of life being unaffected by cosmic explosions are maximised. We find that $\Lambda-$dominated universes favour the survival of life against GRBs. Within a $\Lambda$CDM cosmology, the parameters that govern the likelihood of life survival to GRBs are dictated by the value of $\Lambda$ and the age of the Universe. We find that we seem to live in a favorable point in this parameter phase space which minimises the exposure to cosmic explosions, yet maximises the number of main sequence (hydrogen-burning) stars around which advanced life forms can exist.

astro-ph.CO

Enhancing the Cosmic Shear Power Spectrum

Applying a transformation to a non-Gaussian field can enhance the information content of the resulting power spectrum, by reducing the correlations between Fourier modes. In the context of weak gravitational lensing, it has been shown that this gain in information content is significantly compromised by the presence of shape noise. We apply clipping to mock convergence fields, a technique which is known to be robust in the presence of noise and has been successfully applied to galaxy number density fields. When analysed in isolation the resulting convergence power spectrum returns degraded constraints on cosmological parameters. However substantial gains can be achieved by performing a combined analysis of the power spectra derived from both the original and transformed fields. Even in the presence of realistic levels of shape noise, we demonstrate that this approach is capable of reducing the area of likelihood contours within the $\Omega_m - \sigma_8$ plane by more than a factor of three.

astro-ph.CO