Showing 1–12 of 12 results for author: Guidetti, V

Probing optimisation in physics-informed neural networks

Authors: Nayara Fonseca, Veronica Guidetti, Will Trojak

Abstract: A novel comparison is presented of the effect of optimiser choice on the accuracy of physics-informed neural networks (PINNs). To give insight into why some optimisers are better, a new approach is proposed that tracks the training trajectory curvature and can be evaluated on the fly at a low computational cost. The linear advection equation is studied for several advective velocities, and we show… ▽ More A novel comparison is presented of the effect of optimiser choice on the accuracy of physics-informed neural networks (PINNs). To give insight into why some optimisers are better, a new approach is proposed that tracks the training trajectory curvature and can be evaluated on the fly at a low computational cost. The linear advection equation is studied for several advective velocities, and we show that the optimiser choice substantially impacts PINNs model performance and accuracy. Furthermore, using the curvature measure, we found a negative correlation between the convergence error and the curvature in the optimiser local reference frame. It is concluded that, in this case, larger local curvature values result in better solutions. Consequently, optimisation of PINNs is made more difficult as minima are in highly curved regions. △ Less

Submitted 27 March, 2023; originally announced March 2023.

Comments: Accepted at the ICLR 2023 Workshop on Physics for Machine Learning

Axionic Festina Lente

Authors: Veronica Guidetti, Nicole Righi, Gerben Venken, Alexander Westphal

Abstract: The swampland conjecture known as Festina Lente (FL) imposes a lower bound on the mass of all charged particles in a quasi-de Sitter space. In this paper, we propose the aFL (axionic Festina Lente) bound, an extension of FL to axion-like particles arising from type II string theory. We find that the product of the instanton action and the axion decay constant is bounded from below by the vacuum en… ▽ More The swampland conjecture known as Festina Lente (FL) imposes a lower bound on the mass of all charged particles in a quasi-de Sitter space. In this paper, we propose the aFL (axionic Festina Lente) bound, an extension of FL to axion-like particles arising from type II string theory. We find that the product of the instanton action and the axion decay constant is bounded from below by the vacuum energy. This is achieved indirectly, using dimensional reduction on Calabi-Yau threefolds, and translating the FL result for dipoles into a purely geometric bound. We discuss axionic black holes evolution, and aFL constraints on Euclidean wormholes, showing that the gravitational arguments leading to the FL bound for U$(1)$ charged particles cannot be directly applied to axions. Moreover, we discuss phenomenological implications of the aFL bound, including constraints on string inflation models and the axion-photon coupling via kinetic mixing. △ Less

Submitted 23 January, 2023; v1 submitted 7 June, 2022; originally announced June 2022.

Comments: v3: minor revisions, version published in JHEP, 21 pages, 2 figures

Report number: DESY-22-098

Journal ref: JHEP 01 (2023) 114

Generalizing similarity in noisy setups: the DIBS phenomenon

Authors: Nayara Fonseca, Veronica Guidetti

Abstract: This work uncovers an interplay among data density, noise, and the generalization ability in similarity learning. We consider Siamese Neural Networks (SNNs), which are the basic form of contrastive learning, and explore two types of noise that can impact SNNs, Pair Label Noise (PLN) and Single Label Noise (SLN). Our investigation reveals that SNNs exhibit double descent behaviour regardless of the… ▽ More This work uncovers an interplay among data density, noise, and the generalization ability in similarity learning. We consider Siamese Neural Networks (SNNs), which are the basic form of contrastive learning, and explore two types of noise that can impact SNNs, Pair Label Noise (PLN) and Single Label Noise (SLN). Our investigation reveals that SNNs exhibit double descent behaviour regardless of the training setup and that it is further exacerbated by noise. We demonstrate that the density of data pairs is crucial for generalization. When SNNs are trained on sparse datasets with the same amount of PLN or SLN, they exhibit comparable generalization properties. However, when using dense datasets, PLN cases generalize worse than SLN ones in the overparametrized region, leading to a phenomenon we call Density-Induced Break of Similarity (DIBS). In this regime, PLN similarity violation becomes macroscopical, corrupting the dataset to the point where complete interpolation cannot be achieved, regardless of the number of model parameters. Our analysis also delves into the correspondence between online optimization and offline generalization in similarity learning. The results show that this equivalence fails in the presence of label noise in all the scenarios considered. △ Less

Submitted 24 July, 2023; v1 submitted 30 January, 2022; originally announced January 2022.

Comments: v3: version accepted at ECAI 2023 + Supplementary Material

Report number: DESY-22-016

Fuzzy Dark Matter Candidates from String Theory

Authors: Michele Cicoli, Veronica Guidetti, Nicole Righi, Alexander Westphal

Abstract: String theory has been claimed to give rise to natural fuzzy dark matter candidates in the form of ultralight axions. In this paper we revisit this claim by a detailed study of how moduli stabilisation affects the masses and decay constants of different axion fields which arise in type IIB flux compactifications. We find that obtaining a considerable contribution to the observed dark matter abunda… ▽ More String theory has been claimed to give rise to natural fuzzy dark matter candidates in the form of ultralight axions. In this paper we revisit this claim by a detailed study of how moduli stabilisation affects the masses and decay constants of different axion fields which arise in type IIB flux compactifications. We find that obtaining a considerable contribution to the observed dark matter abundance without tuning the axion initial misalignment angle is not a generic feature of 4D string models since it requires a mild violation of the $S f\lesssim M_P$ bound, where $S$ is the instanton action and $f$ the axion decay constant. Our analysis singles out $C_4$-axions, $C_2$-axions and thraxions as the best candidates to realise fuzzy dark matter in string theory. For all these ultralight axions we provide predictions which can be confronted with present and forthcoming observations. △ Less

Submitted 26 April, 2022; v1 submitted 6 October, 2021; originally announced October 2021.

Comments: 34 pages + 4 appendices, v2: minor changes, reviewed and extended results on C2 axions and thraxions, v3: improved final plot, additional background discussion, JHEP accepted version

Report number: DESY-21-153

A Fake Instability in String Inflation

Authors: Michele Cicoli, Veronica Guidetti, Francesco Muia, Francisco G. Pedro, Gian Paolo Vacca

Abstract: In type IIB Fibre Inflation models the inflaton is a Kaehler modulus which is kinetically coupled to the corresponding axion. In this setup the curvature of the field space induces tachyonic isocurvature perturbations normal to the background inflationary trajectory. However we argue that the associated instability is unphysical since it is due to the use of ill-defined entropy variables. In fact,… ▽ More In type IIB Fibre Inflation models the inflaton is a Kaehler modulus which is kinetically coupled to the corresponding axion. In this setup the curvature of the field space induces tachyonic isocurvature perturbations normal to the background inflationary trajectory. However we argue that the associated instability is unphysical since it is due to the use of ill-defined entropy variables. In fact, upon using the correct relative entropy perturbation, we show that in Fibre Inflation axionic isocurvature perturbations decay during inflation and the dynamics is essentially single-field. △ Less

Submitted 26 July, 2021; originally announced July 2021.

Comments: 5 pages + references, 3 figures

On the choice of entropy variables in multifield inflation

Authors: Michele Cicoli, Veronica Guidetti, Francesco Muia, Francisco G. Pedro, Gian Paolo Vacca

Abstract: We discuss the usefulness and theoretical consistency of different entropy variables used in the literature to describe isocurvature perturbations in multifield inflationary models with a generic curved field space. We clarify which is the proper entropy variable to be used to match the evolution of isocurvature modes during inflation to the one after the reheating epoch in order to compare with o… ▽ More We discuss the usefulness and theoretical consistency of different entropy variables used in the literature to describe isocurvature perturbations in multifield inflationary models with a generic curved field space. We clarify which is the proper entropy variable to be used to match the evolution of isocurvature modes during inflation to the one after the reheating epoch in order to compare with observational constraints. In particular, we find that commonly used variables, as the relative entropy perturbation or the one associated to the decomposition in tangent and normal perturbations with respect to the inflationary trajectory, even if more useful to perform numerical studies, can lead to results which are wrong by several orders of magnitude, or even to apparent destabilisation effects which are unphysical for cases with light kinetically coupled spectator fields. △ Less

Submitted 7 July, 2021; originally announced July 2021.

Comments: 20 pages + appendices, 3 figures

dNNsolve: an efficient NN-based PDE solver

Authors: Veronica Guidetti, Francesco Muia, Yvette Welling, Alexander Westphal

Abstract: Neural Networks (NNs) can be used to solve Ordinary and Partial Differential Equations (ODEs and PDEs) by redefining the question as an optimization problem. The objective function to be optimized is the sum of the squares of the PDE to be solved and of the initial/boundary conditions. A feed forward NN is trained to minimise this loss function evaluated on a set of collocation points sampled from… ▽ More Neural Networks (NNs) can be used to solve Ordinary and Partial Differential Equations (ODEs and PDEs) by redefining the question as an optimization problem. The objective function to be optimized is the sum of the squares of the PDE to be solved and of the initial/boundary conditions. A feed forward NN is trained to minimise this loss function evaluated on a set of collocation points sampled from the domain where the problem is defined. A compact and smooth solution, that only depends on the weights of the trained NN, is then obtained. This approach is often referred to as PINN, from Physics Informed Neural Network~\cite{raissi2017physics_1, raissi2017physics_2}. Despite the success of the PINN approach in solving various classes of PDEs, an implementation of this idea that is capable of solving a large class of ODEs and PDEs with good accuracy and without the need to finely tune the hyperparameters of the network, is not available yet. In this paper, we introduce a new implementation of this concept - called dNNsolve - that makes use of dual Neural Networks to solve ODEs/PDEs. These include: i) sine and sigmoidal activation functions, that provide a more efficient basis to capture both secular and periodic patterns in the solutions; ii) a newly designed architecture, that makes it easy for the the NN to approximate the solution using the basis functions mentioned above. We show that dNNsolve is capable of solving a broad range of ODEs/PDEs in 1, 2 and 3 spacetime dimensions, without the need of hyperparameter fine-tuning. △ Less

Submitted 15 March, 2021; originally announced March 2021.

Comments: 16 pages + 9 pages of appendices, 7 figures, LaTeX, code to be released soon

Report number: DESY-21-037

Resonant backreaction in axion inflation

Authors: Valerie Domcke, Veronica Guidetti, Yvette Welling, Alexander Westphal

Abstract: Axion inflation entails a coupling of the inflaton field to gauge fields through the Chern-Simons term. This results in a strong gauge field production during inflation, which backreacts on the inflaton equation of motion. Here we show that this strongly non-linear system generically experiences a resonant enhancement of the gauge field production, resulting in oscillatory features in the inflaton… ▽ More Axion inflation entails a coupling of the inflaton field to gauge fields through the Chern-Simons term. This results in a strong gauge field production during inflation, which backreacts on the inflaton equation of motion. Here we show that this strongly non-linear system generically experiences a resonant enhancement of the gauge field production, resulting in oscillatory features in the inflaton velocity as well as in the gauge field spectrum. The gauge fields source a strongly enhanced scalar power spectrum at small scales, exceeding previous estimates. For appropriate parameter choices, the collapse of these over-dense regions can lead to a large population of (light) primordial black holes with remarkable phenomenological consequences. △ Less

Submitted 12 August, 2020; v1 submitted 7 February, 2020; originally announced February 2020.

Comments: LaTeX, 2+32 pages, 11 figures; appendix C added, version accepted by JCAP

Report number: DESY-20-017

Geometrical Destabilisation of Ultra-Light Axions in String Inflation

Authors: Michele Cicoli, Veronica Guidetti, Francisco G. Pedro

Abstract: We perform a detailed analytical and numerical analysis of the multi-field evolution of Fibre Inflation and show that, regardless of the microscopic realisation of the model, the mass-squared of one of the two ultra-light axions becomes always negative during inflation. This implies that the corresponding isocurvature perturbations experience a potential geometrical destabilisation which seems to… ▽ More We perform a detailed analytical and numerical analysis of the multi-field evolution of Fibre Inflation and show that, regardless of the microscopic realisation of the model, the mass-squared of one of the two ultra-light axions becomes always negative during inflation. This implies that the corresponding isocurvature perturbations experience a potential geometrical destabilisation which seems to bring the system away from the perturbative regime. Therefore we conclude that a full understanding of the inflationary evolution of Fibre Inflation can be achieved only via a non-perturbative analysis where the backreaction of tachyonic isocurvature perturbations is properly taken into account. △ Less

Submitted 4 March, 2019; originally announced March 2019.

Comments: 22 pages + bibliography, 9 figures

A geometrical instability for ultra-light fields during inflation?

Authors: Michele Cicoli, Veronica Guidetti, Francisco G. Pedro, Gian Paolo Vacca

Abstract: Westudythebehaviourofisocurvatureperturbationsinnon-linearsigmamodels which naturally emerge in supergravity and string inflationary scenarios. We focus on the case of negatively curved field manifolds which can potentially lead to a geometrical destabilisation of isocurvature modes. We find however that heavy fields are stable when their effective mass is computed on the attractor background solu… ▽ More Westudythebehaviourofisocurvatureperturbationsinnon-linearsigmamodels which naturally emerge in supergravity and string inflationary scenarios. We focus on the case of negatively curved field manifolds which can potentially lead to a geometrical destabilisation of isocurvature modes. We find however that heavy fields are stable when their effective mass is computed on the attractor background solution. On the other hand, we show that ultra-light fields can potentially suffer from a geometrical instability when the background trajectory is a geodesic in field space. In this case a full understanding of the system is likely to require the use of non-perturbative methods. △ Less

Submitted 4 March, 2019; v1 submitted 10 July, 2018; originally announced July 2018.

Comments: Matches published version, 5 pages, 2 figures

Chiral Global Embedding of Fibre Inflation Models

Authors: Michele Cicoli, David Ciupke, Victor A. Diaz, Veronica Guidetti, Francesco Muia, Pramod Shukla

Abstract: We construct explicit examples of fibre inflation models which are globally embedded in type IIB orientifolds with chiral matter on D7-branes and full closed string moduli stabilisation. The minimal setup involves a Calabi-Yau threefold with h^{1,1}=4 Kaehler moduli which features multiple K3 fibrations and a del Pezzo divisor supporting non-perturbative effects. We perform a consistent choice of… ▽ More We construct explicit examples of fibre inflation models which are globally embedded in type IIB orientifolds with chiral matter on D7-branes and full closed string moduli stabilisation. The minimal setup involves a Calabi-Yau threefold with h^{1,1}=4 Kaehler moduli which features multiple K3 fibrations and a del Pezzo divisor supporting non-perturbative effects. We perform a consistent choice of orientifold involution, brane setup and gauge fluxes which leads to chiral matter and a moduli-dependent Fayet-Iliopoulos term. After D-term stabilisation, the number of Kaehler moduli is effectively reduced to 3 and the internal volume reduces to the one of fibre inflation models. The inflationary potential is generated by suitable string loop corrections in combination with higher derivative effects. We analyse the inflationary dynamics both in the single-field approximation and by numerically deriving the full multi-field evolution in detail. Interestingly, we find that the Kaehler cone conditions set strong constraints on the allowed inflaton field range. △ Less

Submitted 1 February, 2018; v1 submitted 5 September, 2017; originally announced September 2017.

Comments: 34 pages + appendix; reference added and typos corrected

Journal ref: JHEP 1711 (2017) 207

The 3.5 keV Line from Stringy Axions

Authors: Michele Cicoli, Victor A. Diaz, Veronica Guidetti, Markus Rummel

Abstract: An interesting result in particle astrophysics is the recent detection of an unexplained 3.5 keV line from galaxy clusters. A promising model, which can explain the morphology of the signal and its non-observation in dwarf spheroidal galaxies, involves a 7 keV dark matter particle decaying into a pair of ultra-light axions that convert into photons in the magnetic field of the clusters. Given that… ▽ More An interesting result in particle astrophysics is the recent detection of an unexplained 3.5 keV line from galaxy clusters. A promising model, which can explain the morphology of the signal and its non-observation in dwarf spheroidal galaxies, involves a 7 keV dark matter particle decaying into a pair of ultra-light axions that convert into photons in the magnetic field of the clusters. Given that light axions emerge naturally in 4D string vacua, in this paper we present a microscopic realisation of this model within the framework of type IIB flux compactifications. Dark matter is a local closed string axion which develops a tiny mass due to subdominant poly-instanton corrections to the superpotential and couples via kinetic mixing to an almost massless open string axion living on a D3-brane at a singularity. The interaction of this ultra-light axion with photons is induced by U(1) kinetic mixing. After describing the Calabi-Yau geometry and the brane set-up, we discuss in depth moduli stabilisation, the resulting mass spectrum and the strength of all relevant couplings. △ Less

Submitted 16 October, 2017; v1 submitted 10 July, 2017; originally announced July 2017.

Comments: 27 pages + appendices, 1 figure; typos corrected, references added, additional comments on the cosmological history and DM production in the conclusions