-
Classifying bulk-edge anomalies in the Dirac Hamiltonian
Authors:
Hansueli Jud,
Clément Tauber
Abstract:
We study the Dirac Hamiltonian in dimension two with a mass term and a large momentum regularization, and show that bulk-edge correspondence fails. Despite a well defined bulk topological index --the Chern number--, the number of edge modes depends on the boundary condition. The origin of this anomaly is rooted in the unbounded nature of the spectrum. It is detected with Levinson's theorem from sc…
▽ More
We study the Dirac Hamiltonian in dimension two with a mass term and a large momentum regularization, and show that bulk-edge correspondence fails. Despite a well defined bulk topological index --the Chern number--, the number of edge modes depends on the boundary condition. The origin of this anomaly is rooted in the unbounded nature of the spectrum. It is detected with Levinson's theorem from scattering theory and quantified via an anomalous winding number at infinite energy, dubbed ghost charge. First we classify, up to equivalence, all self-adjoint boundary conditions, using Schubert cell decomposition of a Grassmanian. Then, we investigate which ones are anomalous. We expand the scattering amplitude near infinite energy, for which a dominant scale captures the asymptotic winding number. Remarkably, this can be achieved for every self-adjoint boundary condition, leading to an exhaustive anomaly classification. It shows that anomalies are ubiquitous and stable. Boundary condition with a ghost charge of 2 is also revealed within the process.
△ Less
Submitted 7 March, 2024;
originally announced March 2024.
-
Counting and Algorithmic Generalization with Transformers
Authors:
Simon Ouellette,
Rolf Pfister,
Hansueli Jud
Abstract:
Algorithmic generalization in machine learning refers to the ability to learn the underlying algorithm that generates data in a way that generalizes out-of-distribution. This is generally considered a difficult task for most machine learning algorithms. Here, we analyze algorithmic generalization when counting is required, either implicitly or explicitly. We show that standard Transformers are bas…
▽ More
Algorithmic generalization in machine learning refers to the ability to learn the underlying algorithm that generates data in a way that generalizes out-of-distribution. This is generally considered a difficult task for most machine learning algorithms. Here, we analyze algorithmic generalization when counting is required, either implicitly or explicitly. We show that standard Transformers are based on architectural decisions that hinder out-of-distribution performance for such tasks. In particular, we discuss the consequences of using layer normalization and of normalizing the attention weights via softmax. With ablation of the problematic operations, we demonstrate that a modified transformer can exhibit a good algorithmic generalization performance on counting while using a very lightweight architecture.
△ Less
Submitted 12 January, 2024; v1 submitted 12 October, 2023;
originally announced October 2023.
-
Topology in shallow-water waves: a violation of bulk-edge correspondence
Authors:
Gian Michele Graf,
Hansueli Jud,
Clément Tauber
Abstract:
We study the two-dimensional rotating shallow-water model describing Earth's oceanic layers. It is formally analogue to a Schrödinger equation where the tools from topological insulators are relevant. Once regularized at small scale by an odd-viscous term, such a model has a well-defined bulk topological index. However, in presence of a sharp boundary, the number of edge modes depends on the bound…
▽ More
We study the two-dimensional rotating shallow-water model describing Earth's oceanic layers. It is formally analogue to a Schrödinger equation where the tools from topological insulators are relevant. Once regularized at small scale by an odd-viscous term, such a model has a well-defined bulk topological index. However, in presence of a sharp boundary, the number of edge modes depends on the boundary condition, showing an explicit violation of the bulk-edge correspondence. We study a continuous family of boundary conditions with a rich phase diagram, and explain the origin of this mismatch. Our approach relies on scattering theory and Levinson's theorem. The latter does not apply at infinite momentum because of the analytic structure of the scattering amplitude there, ultimately responsible for the violation.
△ Less
Submitted 2 January, 2020;
originally announced January 2020.
-
Radiative transfer modelling of parsec-scale dusty warped discs
Authors:
H. Jud,
M. Schartmann,
J. Mould,
L. Burtscher,
K. R. W. Tristram
Abstract:
Warped discs have been found on (sub-)parsec scale in some nearby Seyfert nuclei, identified by their maser emission. Using dust radiative transfer simulations we explore their observational signatures in the infrared in order to find out whether they can partly replace the molecular torus. Strong variations of the brightness distributions are found, depending on the orientation of the warp with r…
▽ More
Warped discs have been found on (sub-)parsec scale in some nearby Seyfert nuclei, identified by their maser emission. Using dust radiative transfer simulations we explore their observational signatures in the infrared in order to find out whether they can partly replace the molecular torus. Strong variations of the brightness distributions are found, depending on the orientation of the warp with respect to the line of sight. Whereas images at short wavelengths typically show a disc-like and a point source component, the warp itself only becomes visible at far-infrared wavelengths. A similar variety is visible in the shapes of the spectral energy distributions. Especially for close to edge-on views, the models show silicate feature strengths ranging from deep absorption to strong emission for variations of the lines of sight towards the warp. To test the applicability of our model, we use the case of the Circinus galaxy, where infrared interferometry has revealed a highly elongated emission component matching a warped maser disc in orientation and size. Our model is for the first time able to present a physical explanation for the observed dust morphology as coming from the AGN heated dust. As opposed to available torus models, a warped disc morphology produces a variety of silicate feature shapes for grazing lines of sight, close to an edge-on view. This could be an attractive alternative to a claimed change of the dust composition for the case of the nearby Seyfert 2 galaxy NGC 1068, which harbours a warped maser disc as well.
△ Less
Submitted 23 October, 2016;
originally announced October 2016.
