Seminar Details:

LANS Informal Seminar
"SASSy (Summer Argonne Students Symposium) Part 1"

DATE: August 1, 2013

TIME: 13:00:00 - 17:00:00
SPEAKER: LANS Summer Students
LOCATION: Building 240, 1406 - 1407 (TCS Conference Center), Argonne National Laboratory

Description:
+ Yueling Loh (Supervisor: Victor Zavala)
1:00 - 1:15 PM
Title: Warm-Start Strategies in Augmented Lagrangian Methods for Predictive Control
Abstract: One of the advantages of the augmented Lagrangian method is the ability to warm-start. This allows the algorithm to excel when we want to solve a sequence of related problems, such as in predictive control. We will explore various warm-start strategies, including the choice of the penalty parameter; we will examine effectiveness of these strategies, as measured through the number of iterations, the ability to identify the active set, and the tolerances reached.

+ Xiaocun Que (Supervisor: Sven Leyffer)
1:15 - 1:30 PM
Title: Perspective Reformulations of Mixed Integer Nonlinear Programs (MINLP) with Indicator Variables
Abstract: We examine various perspective reformulations of MINLP from an NLP perspective. We study the properties of the perspective reformulations, run numerical experiments, and propose our own perspective reformulation.

+ Will Boyd (Supervisor: Andrew Siegel)
1:30 - 1:45 PM
Title: OpenMOC: A Method of Characteristic Neutral Particle Transport Code for Shared Memory Parallel Architectures.
Abstract: A new deterministic neutron transport code, OpenMOC, is under development as part of efforts at the Center for Exascale Simulation of Advanced Reactors (CESAR) at ANL and MIT. OpenMOC has been used to investigate the method of characteristics solution technique for the neutron transport equation as applied to modern parallel hardware architectures. Solvers for shared memory systems using OpenMP and CUDA have been developed and weak and strong scaling results will be presented for a variety of hardware platforms, including Intel Xeon and IBM BGQ processors, and NVIDIA GPUs.

+ Matthew Otten (Supervisor: Lois McInnes)
1:45 - 2:00 PM
Title: SAWs: The Scientific Application Web Server
Abstract: The Scientific Application Web server (SAWs) turns any C, C++, or Fortran scientific or engineering application code into a webserver, allowing one to examine (and even modify) the state of the simulation with any browser from anywhere.

------------------------------
-- 15 Minute Break --
------------------------------

+ Kathleen Chalas (Supervisor: Boyana Norris)
2:15 - 2:30 PM
Title: Benchmarking the effective performance of hardware
Abstract: Users are often interested in achieving the best performance or understanding the underlying system on the machines they are running their programs. Typically, they need to run benchmarks and utilize performance tools in order to get an accurate portrayal of effective performance in terms of hardware and software. Effective performance differs from vendor-supplied peak performance metrics in that the peak is the ceiling for software capabilities run on a machine and is determined by the hardware. The typical user will have more complex programs that will not be able to achieve peak performance. Finding effective performance, however, is not a trivial task, despite its necessity as a preceding stage to the work the user is interested in doing. Gathering the desired metrics requires finding appropriate benchmarks and performance tools, installing them, running them with the correct input, and parsing and understanding the output. My task over the summer involved writing python scripts to call performance tools and benchmarks, as well as making system calls; parsing the output and compiling it into an user-understandable format.

+ Ying He (Supervisor: Misun Min)
2:30 - 2:45 PM
Title: A High-Order Spectral Element Method with Transparent Boundary Conditions for Periodic Layered Media
Abstract: The scattering of acoustic and electromagnetic waves by periodic structures plays an important role in a wide range of problems of scientific and technological interest. In this work we describe a High-Order Spectral Element Method (SEM) for the numerical simulation of scattering returns from an irregularly shaped, periodic, doubly layered medium and we have developed the Helmholtz solver, based on spectral element Galerkin approach, into the petascale solver NekCEM. I'll demonstrate the numerical schemes, algorithms, and computational results.

+ Steven Dalton (Supervisor: Jed Brown)
2:45 - 3:00 PM
Title: AMG Setup Phase Scalability Issues
Abstract: The trend towards algebraic multigrid (AMG) in computational science has placed increased interest in the scalability of the method during both the construction of the multilevel hierarchy and the iterative cycling during the solve phase. In contrast to the solve phase, which is composed of relatively simple and well understood primitive operations, AMG construction is composed of a diverse set of complex components which present many concerns to scalability of the method. In this work we present results concerning the scalability of the PETSc AMG implementation, GAMG, and specifically focus on the performance of the Sparse Matrix-Matrix Multiplication as the key performance limiting component.

+ Brett Alexander Rhodes (Supervisor: Tim Tautges)
3:00 - 3:15 PM
Title: Graph-based Mesh Generation with MeshKit
Abstract: In large simulation, it is often necessary to divide a large problem space into many smaller similar problem spaces using a process called mesh generation. In traditional mesh generation, meshes can be difficult to create and are often not flexible in design. MeshKit introduces a graph based approach to mesh generation, this graph based approach reduces the complexity of mesh generation from the user's view, and increases flexibility in mesh design. This talk will introduce mesh generation and MeshKit's approach to generating meshes.

+ (Vishwas) Hebbur Venkata Rao (Supervisor: Mihai Anitescu)
3:15 - 3:30 PM
Title: Optimal power flow using adjoints
Abstract: The focus of this summer research work is on ODE/DAE constrained optimization of the electrical power grid transmission network. An adjoint sensitivity analysis is used to solve the dynamics constrained optimization problem. We present this adjoint sensitivity formulation, and preliminary results, for a test ODE constrained optimization problem of maximizing the mechanical input power of an electrical generator subject to limiting its mechanical angle oscillation. In addition, we present a generalization of this formulation for optimizing the transmission network cost subject to DAE constraints.


 

Please send questions or suggestions to Debojyoti Ghosh: ghosh at mcs dot anl dot gov.