This is a past event.
Thursday, February 18 @ 4pm – Zoom
Neerav Kaushal [Advisor: Dr. Elena Giusarma] will present:
Simulating the Universe with Convolutional Neural Networks
The upcoming galaxy surveys such as EUCLID, DESI and WFIRST are expected to improve the constraints on cosmological parameters. These surveys contain the information of the matter distribution in the Universe in the form of galaxy clusters, laments and voids, whose properties contain information about the Dark Matter, Dark Energy and the laws of gravity. In order to maximize the information that can be retrieved from those observations, accurate theoretical predictions are needed. A powerful tool to obtain rigorous theoretical predictions in cosmology is performing a suite of cosmological simulations. However, generating these simulations require large computational resources and require many hours to complete. This creates a strong need for the development of new computational methods to speed up the process of generating simulations. In this talk, I present a deep learning-based model to quickly generate fast cosmological simulations of the Universe. This model maps fast but moderately accurate Lagrangian Perturbation Theory simulations to accurate but slow N-body simulations and generate fast as well as accurate simulations. The use of such an approach will be particularly useful to compare theory and predictions, to study the impact of neutrino masses on clustering in fully non-linear scale in real-space, generating mock galaxy catalogs and optimizing observational strategies. Preliminary results show that deep neural networks are very efficient in generating cosmological simulations than the traditional methods and are computationally orders of magnitude faster relative to the N-body simulations.
Geeta Sachdeva [Advisor: Prof. Ravindra Pandey] will present:
Interfacial Characteristics of Composite Materials for Spacecraft Applications: A DFT Study
For transportation of humanity to Mars and beyond, our space vehicles will require a large store of fuel, food, and water, all of which add tremendous weight. Polymer composites possess an integrated combination of structures and properties associated with the host matrix and the fiber material and thus hold the potential of being light-weighted high-strength materials. In general, the load transfer from the matrix to the fiber depends upon the strength of bonding at the interface, which characterizes the mechanical strength. In this work, first-principles calculations based on density functional theory are employed to provide the molecular-level description of the interface formed by Epoxy, BMI, and cyanate ester monomers with graphene. The results show that the interaction strength between a monomer and graphene is mainly governed by the nature of bonding at the interface, and subsequently, the mechanical response follows the hierarchical order of the interaction strength at the interface. Moreover, the change in the polarity of the surface from graphene to the BN monolayer improves the interfacial strength, and thereby a higher transverse stiffness is obtained at the molecular level. These results emphasize the need to use computational modeling to efficiently and accurately determine molecular-level polymer/surface combinations that yield optimal mechanical performance of composite materials which helps drive lightweight, high strength materials for aerospace structural applications.
Please register in advance for this event:
*If you have registered in the past for the Physics Colloquium series, please use your orignal zoom link from Dr. Nakamura (email@example.com), or request him to send the registration confirmation email to you again.
0 people added