Canada

In this work, we present a method based on deep learning (DL) to predict the structure of amorphous silicon (a-Si). The accuracy of our approach is validated through training networks to estimate the potential energy. Two architectures, multilayer perceptron (MLP) and convolutional neural network (CNN), have been examined for this purpose.
The models have been trained on a dataset generated by molecular dynamics (MD) simulations. The dataset consists of 216-atom a-Si configurations.
The performances of the two architectures over the test set are compared while they have the same total number of parameters and layers. 
The results of training these two neural networks yield a
root mean square error of the order of 0.1 meV/atom. This error was reported in the range of 2-6 meV/atom in previous works [1-3], which confirms the accuracy of our models. Calculation of the average error shows that CNN performs better than MLP.
After validation of our approach, we trained a generative model, variational autoencoder (VAE), to generate a-Si configurations. The generated structures and their structural properties are comparable to the real ones.
In general, the performances of our models are quite satisfactory, and this suggests that they could be used to approximate the potential energy and generate a-Si structures, and thus be a viable model for investigating disordered systems.



References: 
1. Li, R., Lee, E. and Luo, T., 2020, A unified deep neural network potential capable of predicting the thermal conductivity of silicon in different phases, Materials Today Physics 12, 100181.
2. Comin, M. and Lewis, L.J., 2019, Deep-learning approach to the structure of amorphous silicon, Physical Review B 100, 094107.
3. Behler, J. and Parrinello, M., 2007, Generalized neural-network representation of high-dimensional potential-energy surfaces, Physical Review Letters 98, 146401.

CAP Virtual Congress 2021

Modeling Amorphous Silicon Using Neural Networks

amorphous silicon

deep generative models

neural networks

ICPR 2020 is the premier international conference on Pattern Recognition, gathering leading researchers from around the world to present their latest achievements in Pattern Analysis and Recognition, applied Artificial Intelligence and Machine Learning. The topic of the year is " Putting Artificial intelligence to work on Patterns". 
For the first time, ICPR has employed a two-round review process similar to journal submissions, with new papers allowed to be submitted in either the first or second round and papers submitted in the first round and not accepted allowed to be revised and re-submitted for second round review. 

In this edition of the **ICPR 2020** Conference, there are 1407 Oral/Poster presentations organized in five Tracks: 
- Artificial Intelligence, Machine Learning for Pattern Analysis; 
- Biometrics, Human Analysis and Behavior Understanding; Computer Vision, Robotics and Intelligent Systems; 
- Document and Media Analysis; and 
- Image and Signal Processing. 

The **ICPR 2020** Main Conference program also includes 7 Keynotes from prominent world-renowned scientists, 8 Tutorials, 28 Demos, 8 Challenges, 2 Panels, as well as an Industry/Sponsor Exhibition. In addition, 41 Workshops on a range of important current and emerging research topics complete the ICPR 2020 program. 


ICPR 2020

technical paper

**WELCOME TO THE 2021 CAP VIRTUAL CONGRESS UNDERLINE PLATFORM**

The CAP staff and our organizing and program committees are pleased to welcome you to our 2021 Congress filled with plenary talks, technical sessions/talks, student presentations and competition sessions, workshops, and several symposia offered by the CAP divisions and its partners.  

You will also have opportunities to network with other delegates through the Gathertown platform accessed by selecting the lounge or posters button on the left.

Registered delegates should use the schedule button to call up the details of the session/event they wish to attend.  Each sessions includes a link to access the online platform where the event is taking place, either within Underline or through other Congress-related Zoom meetings.  

Invited speaker talks are identified with an (I), and student competitors include a (G*) or (U*) to identify graduate or undergraduate presenters.

Please take the time to check out the information about our sponsors and exhibitors, request further information, or leave a note of thanks for their support.

General information about the Congress can be found on the CAP website Congress pages at https://www.cap.ca/congress/2021.

Looking for a specific talk?   A full searchable list is available on our Indico Congress site at https://indico.cern.ch/event/985448/contributions/

We look forward to sharing this exciting week with you!


We explore regions of parameter space that give rise to suppressed direct detection cross sections in a simple model of scalar dark matter with a scalar portal that mixes with the standard model Higgs. We found that even this simple model allows considerable room in the parameter space that has not been excluded by direct detection limits.
While a number of effects leading to this result have been previously noted, our main new result explores interference effects between different contributions to DM annihilation when the DM mass is larger than the scalar portal mass. New annihilation channels open up and the parameters of the model need to compensate to give the correct DM relic abundance, resulting in smaller direct detection cross sections.
We find that even in a very simple model of DM there are still sizeable regions of parameter space that are not ruled out by experiment.

Exploring Direct Detection Suppressed Regions in a Simple 2-Scalar Mediator Model of Scalar Dark Matter

A 3D topological insulator (TI) hosts an odd number of Dirac cones as its 2D surface states spectrum. The states are exponentially localized to the surface and their (pseudo)spin is locked to the surface momentum direction due to spin-orbit interaction. If the TI is also magnetic there are fixed magnetic moments on the surface, co-existing with the itinerant Dirac electrons. The magnetic moments interact with each other both directly (exchange) and indirectly (RKKY) and may form a ferromagnetically ordered state. The magnetic state couples to the Dirac electrons and serves as a Dirac mass which, when uniform, opens a gap in the spectrum. When the ordering of the magnetic moments is modified by excitation such as skyrmions, the Dirac electrons see a landscape of spatially dependent mass. A skyrmion texture binds an electronic state to it which we call a skyrmion-bound state.
In this talk we will see how the bound states can be detected by a surface conductivity measurement and how the skyrmion-skyrmion interaction is altered due to the presence of Dirac electrons. We will also discuss a skyrmion solid background to Dirac electrons.

Domain and Skyrmion bound states on the surface of magnetic topological insulators

Liquid xenon (LXe) is frequently employed to build detectors for rare event searches due to many of its advantageous properties including high stopping power, high ionization and scintillation yields, and relatively high cryogenic operating temperature. Time projection chambers (TPC) with LXe allow for 3D event topology reconstruction and identification which is important for reducing backgrounds. Due to the high drift fields TPCs are operated at, it is crucial to model breakdown properties in LXe. Often, a high voltage (HV) discharge is able to damage the detector instrumentation, e.g. the photo-sensors array. We study the breakdown formation in the context of the EXO-200 experiment, the first generation 0νββ search experiment from the Enriched Xenon Observatory (EXO) collaboration that will be followed by a much higher sensitivity experiment called nEXO. In the EXO-200 detector, instabilities have been observed on the TPC HV line, each accompanied by scintillation VUV light which was detected by the EXO-200 Avalanche Photo Diode (APD) arrays. Using a setup designed to study LXe breakdowns we investigate the origin of the HV instabilities as well as the correlation between those HV glitches and fully formed breakdowns.

High Voltage Breakdowns in Liquid Xenon

For 50 years, TRIUMF has stood at the frontier of scientific understanding as Canada’s particle accelerator centre. Driven by two made-in-Canada cutting edge accelerators - the world’s largest cyclotron, and our new high-power superconducting linear accelerator - we continue to ask the big questions about the origins of the universe and everything in it. 

With over five decades of experience in the production of accelerator-based secondary particles for science, TRIUMF also ensures that Canada remains on the leading edge of supplying radioisotopes, neutrons, photons, and muons enabling fundamental science in the fields of nuclear, particle and astrophysics, as well as solid state and medical sciences and applications. 

ISAC-TRIUMF is the only ISOL facility worldwide that is routinely producing radioisotope beams (RIB) from secondary particle production targets under irradiation in the high-power regime in excess of 10 kW. TRIUMF’s current flagship project ARIEL, Advanced Rare IsotopE Laboratory, will add two new target stations providing isotopes to the existing experimental stations in ISAC I and ISAC II at keV and MeV energies, respectively. In addition to the operating 500 MeV, 50 kW proton driver from TRIUMF’s cyclotron, ARIEL will make use of a 30 MeV, 100 kW electron beam from a newly installed superconducting linear accelerator. Together with additional 200 m of RIB beamlines within the radioisotope distribution complex, this will put TRIUMF in the unprecedented capability of delivering three RIBs to different experiments, while producing radioisotopes for medical applications simultaneously – enhancing the scientific output of the laboratory significantly.

Secondary Particle Production for Fundamental Science at TRIUMF

The basic structure of quantum field theory that is used to describe the Standard Model of fundamental interactions of nature is usually formulated for zero temperature. However, the effects of temperature are extremely important for understanding a number of physical processes such as the electro-weak phase transition and quark-gluon plasma. Thermal field theory is the extension of quantum field theory to a non-zero temperature environment and is achieved by modifying the propagators in loop integrations represented by Feynman diagrams. The Python program pySecDec numerically calculates dimensionally-regularized loop integrals in quantum field theory using the sector decomposition approach. It is shown how pySecDec can be applied to thermal field theory numerical calculations using modifications within the Matsubara formalism. Using the formulated algorithm, a 2-point correlation function (such as those occurring in QCD correlation functions) at finite temperature can be numerically calculated for a variety of spacetime dimensions.

Numerical Loop-Integration Methods for Thermal Field Theory

It is an ongoing challenge to engineer setups in which Majorana zero modes at the ends of one-dimensional topological superconductor are well isolated which is the essence of topological protection. Recent developments have indicated that periodic deriving of a system can dynamically induce symmetries that its static counterpart does not possess [1]. We further develop the original protocol [2] where this idea [1] is applied to a system of quantum dot (QD) coupled to a Kitaev chain hosting an imperfect (overlapping) Majorana zero modes. We numerically simulate a protocol in which an electron periodically hops back and forth from the QD and the wire. We demonstrate that the current protocol reduces a non-zero hybridization energy that manifests from imperfect Majoranas by orders of magnitude. Furthermore, we examine the efficiency of the suppression and how robust it is to imperfections.

[1] K. Agarwal and I. Martin, Phys. Rev. Lett. **125**, 080602 (2020).

[2] I. Martin and K. Agarwal, PRX Quantum **1**, 020324 (2020).

Reducing Majorana Hybridization via Periodic Driving

Flat band systems are becoming popular due to special properties. For instance, the strong correlation of electrons leads to realization of unconventional superconductivity [1]. Typically, such bands are only approximately flat and are engineered by fine tuning Vanderwaal’s structures. Here we consider Kagome and Lieb lattices with perfectly flat bands. However, at some points in the Brillouin zone the bands superimpose leading to degeneracies. It has been shown that the degeneracy can be lifted when time-reversal symmetry (TRS) is broken [2]. In this presentation, we explore further means to lift the degeneracy while preserving the flat band. We show that the flatness is robust under certain changes to the lattice and that breaking TRS is not sufficient to isolate the flat band. Instead, we show that modulating the flux based on Chern-Simons field as outlined in [3] successfully gaps out the band.
[1] Cao, Y. et al. (2018). Unconventional superconductivity in magic-angle graphene superlattices. Nature, 556(7699), 43-50
[2] Green, D., Santos, L., & Chamon, C. (2010). Isolated flat bands and spin-1 conical bands in two-dimensional lattices. Physical Review B, 82(7).
[3] Maiti, S., & Sedrakyan, T. (2019). Fermionization of bosons in a flat band. Physical Review B, 99(17).

Identifying and Isolating Flat Bands in 2D Systems

Separating and analyzing DNA, RNA or proteins is often done through gel electrophoresis. This simple technique is conducted millions of times in labs every day world-wide. Analysis of the gels is typically done visually on a photo by comparing measured intensities with a well-defined reference ladder. Using digital photography and image analysis, we have developed a technique to quantitatively analyze standard gel images to determine exact sized distributions. This information can then be used to compare experimental data to theoretical models and predictions. The software consists of a smartphone-based application with an easy-to-use GUI to analyze images of any quality, dimension, or type. Within seconds, the user can sign in, choose ladder specifications, and select the image for analysis. They can then tap on the ladders of interest and send the image for analysis. The cloud-based analysis uses an exponential model to characterize the propagation of molecules within the gel and automatically measures the size and relative quantity of detectable solutes. I will provide direct evidence that the software allows measurement of RNA molecule lengths with resolution down to a single monomer. The entire analysis process takes under 10 seconds, after which the results are automatically emailed to the user. The novel way in which this data is processed and presented to the end-user in a seamlessly integrated workflow with possibilities for additional integration makes this a major step forward over existing solutions. Enabling, supporting, and promoting this research by creating easy-to-use and powerful tools has the potential to promote change in physical and biological research.

The quantitative analysis of gel electrophoresis data using a cloud-based smartphone application

We describe the **offset logarithm function** and illustrate its applications to physics. The offset log can be considered an extension of the logarithm function, and it can also be considered as a generalization of the Lambert W function.
The offset log function shows up in a variety of physical and biological models, such as the mean field model (Weiss model) of ferromagnetism, and models of carbon nanoribbons. It provides a convenient framework for describing phenomena in which the behaviour of a physical variable at one location or time, has an exponential or logarithmic relationship to the value of attribute at a displaced location or time.

The offset logarithm function of order $k$ is the multivalued function $w = L_k(z)$ of the complex variable $z$, which satisfies the equation
\begin{equation} \frac{ w-k }{w+k}e^w = z \qquad(1) \end{equation}
Each of $k$, $z$, and $w$ can be in general a complex number. However, for this, we consider $k$, $z$ and $w$ as real numbers. Approximate solutions to Equation (1) can be calculated numerically, but a study of the analytic properties of that equation can provide insight into the system being described. 

The offset log function is multi-branch. Each branch of $L_k(z)$ is an analytic function for certain regions of $ k$ and $z$ values. We present the main characteristics of the $L_k(z)$ function. For each domain of the $k$ parameter, the properties of $L_k$ in terms of the branches are given. We show that the number of solutions of Equation (1) depends upon the domains of $k$ and $z$. For these domains we give the number of solutions and we give the interval where each solution lies. We identify the cases where a solution is a multiple solution. We give a symbolic formula for the multiple solutions.

The offset logarithm function and some applications in physics: A generalization of the Lambert W function

Transitions between fractionalized and conventional quantum phases of matter in 2+1 dimensions are conceptually best understood within the framework of parton gauge theories, whereby the confinement of fractionalized excitations and spontaneous breaking of global symmetries in conventional phases is argued to result from the proliferation of gauge monopoles/instantons. To complement recent studies of the quantum numbers and scaling dimensions of monopole operators in parton gauge theories of frustrated quantum antiferromagnets, we provide an explicit semiclassical derivation of instanton contributions to the low-energy effective parton gauge theory for hardcore bosons on the honeycomb lattice using methods originally developed by 't Hooft in the solution of the U(1) problem in quantum chromodynamics. While the symmetry-breaking effect of instantons is typically associated with massless fermions and the Atiyah-Singer index theorem on compact manifolds, we show that the spontaneous breaking of the U(1) boson number conservation symmetry in the superfluid phase results here from Euclidean zero modes of massive Dirac fermions bound to instantons in noncompact spacetime.

Premium content

Modeling Amorphous Silicon Using Neural Networks

Downloads

Next from CAP Virtual Congress 2021

Exploring Direct Detection Suppressed Regions in a Simple 2-Scalar Mediator Model of Scalar Dark Matter

Similar lecture

ISP4ML: The Role of Image Signal Processing in Efficient Deep Learning Vision Systems