LANS SASSy - Summer Argonne Students' Symposium 2016

Hoffman, Kentaro
Narayanamurthi, Mahesh
Hart, Joey
Yu, Jing
Offermans, Nicolas
Rao, Shruti-Dwarkanath
Wang, Wendy
Seminar

Summer students in LANS give talks on their research. The following students will be presenting:

Hoffman, Kentaro 1:00PM
Title: Finite Volume Formulation of Water Pipeline Systems using PETSc
Abstract: This project seeks to create a numerical simulation of 1-D water flow through a pipe along with strict integration with the High Performance capabilities provided by the PETSc library. The simulations have so far produced high-fidelity physical models along with expected performance improvements. Particular attention will be paid to the PDE formulation and what modifications are necessary to ensure accurate and fast simulations.

Narayanamurthi, Mahesh 1:15PM
Title: Center of the Universe
Abstract: Algorithmic differentiation (AD) is a technique by which a program can be transformed into another equivalent program that computes the sensitivities of the output of the original program with respect to its inputs. Tools that perform this transformation are complex and are constantly evolving with changing computing paradigms. Last summer, I started working on a test-suite to test AD software, which was left partly unfinished but is slowly being adopted by the AD community. In this talk I will present the advances that I have made to the test-suite - in terms of new examples, changes in the organization of the examples and other tools that were developed to make AD software testing a little bit less painful. If time permits, I will discuss some interesting problems in the test-suite, challenges for AD and some techniques to handle these challenges.

Hart, Joey 1:30PM
Title: Global Sensitivity Analysis for Stochastic Wind Simulation
Abstract: Global sensitivity analysis (GSA) is frequently used to analyze high dimensional mathematical models. In this talk, we consider GSA on the hyperparameters of a Gaussian process wind speed model. Having an appropriate joint probability distribution on the hyperparameters is necessary for GSA, but poses a significant challenge when dealing with hyperparameters in a statistical model. A new method for approximating the joint distribution and computing sensitivity indices is proposed; preliminary numerical results are presented.

Yu, Jing (Alice) 1:45PM
Title: Design of Experiments Framework for Sensor Placement in Gas
Pipeline System
Abstract: When monitoring spatial phenomena, choosing sensor locations is
a fundamental task. In this talk, we present a design of experiments
framework for sensor placement in the natural gas pipeline network, with
the aim of minimizing the uncertainty in the parameters from a Bayesian
linear inverse problem governed by hyperbolic partial differential
equations. We investigate two objectives: the variance of total flow and
A-optimal design criterion, and approach the resulting integer nonlinear
program by a heuristic sum-up rounding strategy. We give numerical results
within this framework and demonstrate sensor locations in finite cases.

BREAK 2-2:30PM

Offermans, Nicolas 2:30PM
Title: Error based mesh refinement in a spectral element code (Nek5000)
Abstract: Direct Numerical Simulation (DNS) is the only mean to accurately capture the smallest flow structures.
However, the mesh has to be designed in such fashion that regions which are subject to high flow fluctuations are refined enough while preserving a level of coarseness in regions of less interest.
To this end we employ an iterative conformal mesh refinement strategy based on a posteriori error estimators.
The implementation is performed in the code Nek5000, while the mesh refinement is conducted in the Nek5000 native mesh generator and also in CUBIT. The problem studied here is the flow past a cylinder in 2D.

Rao, Shruti-Dwarkanath 2:45PM
Title: Transient security constrained optimal power flow problem
Abstract: The current practices in the energy markets to obtain the individual dispatch for the committed generators in the network ensure that in the event of loss of any component of the network, the post-disturbance steady state is within limits. However, for any fault it is important to ensure that the system can ride through the transients caused by the fault to be able to reach the post-contingency steady state in the first place. This is accomplished using transient security constrained OPF. I will present an overview of the current practices, the proposed method, and my involvement in the project.

Wang, Wendy 3:00 PM
Title: Deployment Mechanisms for AD Tools
Abstract: Automatic differentiation (AD) is useful for evaluating derivatives of arbitrary order for simulations and models. The AD tools most frequently used for these purposes tend to rely on complex compiler infrastructures, or only exist as binaries targeting archaic distributions, rendering them challenging to employ. Explored solutions to increase the availability of AD tools include virtualization, containerization, web server interfaces, and command-line interfaces.