Seminar Details:

LANS Informal Seminar
"SASSy 2012"

DATE: August 2, 2012

TIME: 13:00:00 - 17:30:00
SPEAKER: Givens Fellows and Summer Interns, LANS
LOCATION: Building 240, Room 1416, Argonne National Laboratory


Talk: 01
Time: 13:10 to 13:25
Name: Xuan Zhou
Title: PETSc Interface for Elemental Package
Abstract: Elemental is a library written by Jack Poulson in C++
for distributed-memory dense linear algebra that strives to be both
fast and convenient. However Elemental's own interface is quite
different from PETSc's. The PETSc interface for Elemental is meant
to provide PETSc users with an interface they are familiar with to
access Elemental's functionality. This presentation introduces the
design of the interface and some intuitive examples.
Talk: 02
Time: 13:25 to 13:40
Name: Tom (Lihua) Li
Title: Parallel Solver for Toeplitz Linear System
Abstract: Solving a large Toeplitz linear system is a critical
step in various data analysis tasks, for a slow solver could
severely limit the overall performance. This work demonstrates
an novel approach to parallelize the solving process, resulting
an an efficient and scalable solver. Techniques such as global
communication reduction and grid data partitioning are implemented
to further enhance solver speed.
Talk: 03
Time: 13:40 to 13:55
Name: Hong Zhang
Title: Implementation of extrapolated IMEX methods for ordinary
differential equation in PETSc
Abstract: Implicit-explicit methods are intended for problems with
well-separated time scales. IMEX methods using extrapolation can
attain very high orders of accuracy and have great parallelization
potential. I developed an efficient implementation of extrapolated
IMEX ODE solver in PETSc. The solver provides various control
options and can be order adapative. Numerical results show that
the solver can achieve theoretical order for nonstiff problems and
order reduction is observed for stiff problems.
Talk: 04
Time: 13:55 to 14:10
Name: Michael Burkhart
Title: Gaussian Process Modeling
Abstract: Gaussian Processes provide a powerful tool for data
modeling and prediction. They prove more versatile and require fewer
assumptions than the standard linear model. Discussion will center
on the construction and use of Gaussian Processes. Examples will
illustrate the power and utility of this fascinating theoretical

Break: 14:10 to 14:25

Talk: 05
Time: 14:25 to 14:40
Name: Jay(Jie) Liu
Title: Brief introduction to LHS design and GP implementation GP in R
Abstract: Latin Hypercube Sampling(LHS) is a statistical method which
can be used to generate reasonable design for computation. It's
often applied in uncertainty analysis. I am going to make a
brief introduction about the LHS design and show some results for
GP(Gaussian Process) implementation in R which coincides with Mike's.
Talk: 06
Time: 14:40 to 14:55
Name: Ying He
Title: High-order solvers for Schrodinger and acoustic wave equations
Abstract: Targeting accurate prediction of charge carrier dynamics
in solar cells, to utilize the available energy from the sun, we have
developed Schrodinger solver, based on spectral element discontinuous
Galerkin approach, into the petascale solver NekCEM. I'll demonstrate
the numerical schemes, algorithms, and computational results. I'll
also discuss about non-reflecting boundary treatment, called
Drichlet to Neumann technique, that can be widely applied for
wave-dominated problems.
Talk: 07
Time: 14:55 to 15:10
Name: Shengyong Cai
Title: Mesh sweeping using a cage-based method for interior point
Abstract: The mesh sweeping is an algorithm for all-hexahedral mesh
generation by sweeping the all-quad mesh from the source surfaces
to the target surfaces. One of the difficult issues is to locate
the interior nodes inside the volume after all the source surfaces,
target surfaces and linking surfaces are meshed. In this project,
the cage-based method is used to locate the interior nodes. First,
a topological model is created and the interior nodes could be
placed by simple translation. Then the interior nodes are bound
with the cage, that is, the interior nodes can be computed as an
explicit function with the input argument by the cage. The binding
process is done on the topological model. Because all the surfaces
are meshed, the deformed cage can be obtained directly from the
real model. Finally, the interior nodes can be placed in real model
with the input from the deformed cage. The new interior nodes
locating method is able to guarantee that the interior nodes are
relocated accordingly when the bounding cage is deformed. Therefore,
any inverted element could be avoided during sweeping.
Talk: 08
Time: 15:10 to 15:25
Name: Drew Wicke
Title: Automatic Identification of Partial Separability
Abstract: Many optimization problems have a partially separable
(PS) formulation. ADIC2 exposed the sparsity in the Jacobian of
PS computations by expanding the elementals. The sparsity was then
exploited by computing the derivative of the smaller elementals. We
implemented a tool that automatically identifies PS computations
using activity and linearity analysis. Activity analysis is used
to identify statements that are possibly partially separable and
linearity analysis confirms partial separability and determines
the elemental variables.

Break 15:25 to 15:40

Talk: 09
Time: 15:40 to 15:55
Name: Xu Zhang
Title: Uncertainty Quantification of Fluid Flow Simulation using
Model Reduction and Kriging
Abstract: We use Galerkin POD model reduction technique
in uncertainty quantification of the fluid flow
simulation. Gaussian-processes based Kriging is applied to
calibrate the outputs of the reduced model. Our experiments
indicate that statistical quantities such as mean, variance, and
cumulative distribution can be well predicted. We explore several
approaches to improve the prediction of the full model outputs
in our experiments. Numerical results are presented to indicate
features of these approaches.
Talk: 10
Time: 15:55 to 16:10
Name: Carmeline Dsilva
Title: Molecular dynamics simulations to extract phase field model
Abstract: Phase field models allow us to simulate material systems on
much longer time and length scales than is feasible with traditional
atomistic simulation methods. However, such models require various
thermodynamic and kinetic parameters. I will discuss using molecular
dynamics simulations to extract the relevant parameters for phase
field simulations of uranium dioxide grains.
Talk: 11
Time: 16:10 to 16:25
Name: Lulu Liu
Title: The quarter-five spot problem with ASPIN
Abstract: The quarter-five spot problem is a well-known model in
reservoir simulation. As we all know, the advantage of implicit
time integration methods is to take a larger time step, however,
we have to solve the nonlinear system. It is shown that the Additive
Schwartz preconditioned inexact Newton(ASPIN) methods work well for
several computational fluid dynamics problems, but few applications
of ASPIN to porous media problems have been reported. Hence, we
try to apply ASPIN to our model equations obtained from a fully
implicit discretization of immiscible two-phase flow in heterogeneous
porous media.
Talk: 12
Time: 16:25 to 16:40
Name: Yunfei Song
Title: Solving Mixed Integer Nonlinear Programming by Diving
Abstract: We present the performance comparisons of 40
diving-heuristics in solving mixed-integer nonlinear programming
(MINLP). These diving-heuristics combine fractional, vector-length,
lexicographic and reduced-cost heuristics. Moreover, we present a
novel heuristic by incorporating Special-Order-Set (SOS) with the
traditional diving-heuristics. Rather than only considering variables
in the traditional diving-heuristics, our approach concerns both
variables and SOS constraints simultaneously. The numerical result
shows the new approach can obtain better feasible solutions than
the traditional diving-heuristics for some instances.
Talk: 13
Time: 16:40 to 16:55
Name: Jing Fu
Title: Parallel I/O Optimizations for a Massively Parallel
Electromagnetic System
Abstract: In this talk, we examine a few parallel I/O approaches
for the checkpointing of a massively parallel electromagnetics
solver system called NekCEM. We discuss a MPI-IO collective
approach (coIO), an application-level I/O staging approach,
called reduced-blocking I/O (rbIO) and threaded version of
rbIO. We demonstrate their respective performance advantages over
the traditional 1 POSIX file per processor approach and also
show their impact on production performance improvement on up to
32K processors on BG/P and Cray.
Talk: 14
Time: 16:55 to 17:10
Name: Ryan Lewis
Title: Improved Point Location for Solving Weakly Coupled PDEs
Abstract: We present the point location problem and briefly survey a
few algorithms for solving it.
Talk: 15
Time: 17:10 to 17:25
Name: Tom Hayes
Title: Efficient and accurate climate data representation in MOAB
Abstract: Managing climate data in an efficient and scalable manner
poses an interesting challenge to computational scientists. The
performance bottleneck in some simulations comes from reading large,
high-resolution input data files. This study develops a parallel file
reader for the unstructured High Order Multi-scale Modeling Enviroment
(HOMME) data format in the Mesh Oriented datABase (MOAB) software.
MOAB’s capabilities for general mesh manipulation combined with an
efficient HOMME data reader would enable scientists to perform more
targeted climate data analysis. The scalability of the parallel reader
is demonstrated.


Please send questions or suggestions to Jeffrey Larson: jmlarson at anl dot gov.