NSC Seminars on High Performance Computing
Please click here for information on forthcoming
seminars.
List of previous seminars
Autumn 2000
Spring 2001
| Time |
Place |
Speaker |
Title (click to see abstract) |
| Tue., Jan 23, 2001, at 16.15 |
Room Schrödinger |
Prof. Matts Karlsson, Dept. of Biomechanical Engineering (IMT), Linköping
University |
Mechanics of the Heart |
| Tue., Jan. 30, 2001, at 16.15 |
Room Schrödinger |
Dr. Alexandr Malusek and Dr. Michael Sandborg, Dept. of Radiation Physics
(IMV), Linköping University |
Monte Carlo modelling of the imaging chain in x-ray
diagnostics |
| Tue., Feb 20, 2001, at 16.15 |
Room Schrödinger |
Prof Kenneth Runesson, Dept. Solid Mechnanics, Chalmers |
A Paradigm for Error Estimation and Adaptivity in
Computational Mechanics |
| Tue., Mar 13, 2001, at 16.15 |
Room Schrödinger |
Dr. Per Weinerfelt, Saab Aerospace |
Large Scale Aerodynamic Flow Computations on Parallel
Computers |
| Wed., Apr 18, 2001, at 16.15 |
Room Schrödinger |
Dr. H. E. Markus Meier (SMHI) and Torgny Faxén (NSC) |
Climate Modelling of the Baltic Sea |
Detailed Programme
Title: Atomic-Scale
Computational Materials Science
Speaker: Prof. Göran Wahnström, Chalmers University
of Technology and Göteborgs University
Time: Tuesday, Nov. 28, 2000,
at 13.15-14.00
Abstract
With the rapid development of advanced technologies materials science is
confronted with many challenging problems. Atomic-scale computational materials
science has become one important tool in improving our understanding and
in facilitating the design of new materials. Quantum mechanical calculations
provide a means to describe materials on a truly microscopic level. In
principle the laws governing the behaviour of the microscopic constituents
of a material, the electrons and atomic nuclei, are well known: it is sufficient
to solve the Schrödinger equation. However, besides in some trivial
cases this is a formidable numerical task.
I would like to present common approximations and numerical techniques
used in solving the Schrödinger equation in the field of real materials.
I will use vacancies in aluminium as an illustrative example. Aluminium
has often been used as a test case for developing the computational methodology
and, to some extent, it can be viewed as the "hydrogen atom" of computational
materials science.
Title: Mining
Microarray Data: Predicting Gene Function from Gene Expressions and Background
Knowledge -- State-of-the-art and Opportunities for High Performance Computing
Speaker: Prof. Jan Komorowski
Time: Tuesday, Dec. 12, 2000,
at 16.15-17.00
Abstract
The majority of states in health or disease is most likely controlled by
hundreds of genes. Until recently, molecular biomedicine could study
one or very few genes in parallel. With the advent of the so-called
high throughput micro-array technology it is now possible to observe the
levels of activity in literally thousands of genes. At the same time, genome
mapping projects such as, for instance, the Human Genome Initiative, which
is an international research program for the creation of detailed genetic
and physical maps of the human genome, generate enormous quantities of
data. For the human genome, there are approximately 100K genes out
of which ca 5K genes are known. The major goal of Functional Genomics
is an assignment of function to genes. State-of-the-art in Functional
Genomics is the use of unsupervised learning methods. Our multidisciplinary
team has developed a supervised learning method for inducing predictive
rule models for functional classification of gene expressions from microarray
hybridization experiments.
In this talk I show how we have resolved some of the thorny issues in
the functional classification of gene expressions. Then, I identify
potential research and application areas for High Performance Computing.
Nota bene. The talk is self-contained with respect to the knowledge
of molecular biology.
Title: Mechanics
of the Heart
Speaker: Prof. Matts Karlsson, IMT, Linköping University
Time: Tuesday, Jan. 23, 2001,
at 16.15-17.00
Place: Room Schrödinger (E324), F-building (Fysikhuset), Campus
Valla
Abstract
Non-invasive imaging is the clinical tool of the future for diagnosis of
heart diseases. A unique velocity measurement technique enables us to measure
4D (3D+time) velocity fields. By applying newly developed methods in engineering
mechanics and image processing we have pioneered the work on visualisation
and taken the newly developed
methods into clinical practice. The trend for the next generation in
diagnostic tools will be to go from qualitative to quantitative.
Thus, the work includes the possibility to extract established parameters,
such as pressure differences, completely non-invasively (without inserting
a pressure catheter) from the velocity data. Using a similar data acquisition
technique for the velocities of the heart muscle, a 4D (3D+time) mapping
of the strain and even stress can be used to describe the mechanical properties
of the heart muscle.
The data sets acquired are large (>512 MB for one heart beat). Gradient
estimation and computation of important parameters are difficult both due
to the size of the data sets but also due to the available time - if these
methods are to be used clinically, computational time is very important.
Therefore, both increased computer power and faster algorithms are important
for the future. In this presentation I will present the work in the group
and show some examples of the current research as well as point out a few
new directions for the future.
Title: Monte
Carlo modelling of the imaging chain in x-ray diagnostics
Speaker: Dr. Alexandr Malusek and Dr. Michael Sandborg,
Dept. of Radiation Physics, Linköping University
Time: Tuesday, Jan. 30, 2001,
at 16.15-17.00
Place: Room Schrödinger (E324), F-building (Fysikhuset), Campus
Valla
Abstract
X-rays are of great importance in medical imaging. Their ability to penetrate
through matter makes them invaluable in examining internal structures.
In the simplest approach, they behave like independent particles that are
either absorbed by the patient's body or penetrate it, possibly contributing
to the image detector signal. To achieve better predictions in modalities
like planar X-ray imaging or computed tomography (CT) imaging, more complicated
models including photon scatter, generation of secondary particles, etc.
have to be considered. Knowing the details of photon interactions with
matter, Monte Carlo-method models of photon-transport inside a given geometry
can be created. Fortunately, photon Monte Carlo histories can be considered
independent and thus a high level of parallelization of the computational
task can be achieved. Thousands of CPUs can be used and cheap PC clusters
available today permits a whole range of new tasks can be solved.
In this talk, two examples of an application of the Monte Carlo method
in X-ray medical imaging will be presented: a model for calculating the
optimal imaging parameters in planar imaging and the development of a model
of a medical CT scanner. Both models have been developed at the Radiation
Physics Department of the Linköping University in collaboration with
scientist in London. A short overview of X-ray physics will be given and
examples of its application. No special knowledge is required. Issues related
to vectorization and parallelization of the code will also be discussed.
Title: A Paradigm for Error Estimation
and Adaptivity in Computational Mechanics
Speaker: Prof. Kenneth Runesson, Dept. Solid Mechnanics,
Chalmers University of Technology, Göteborg
Time: Tuesday, Feb. 20, 2001, at 16.15-17.00
Place: Room Schrödinger (E324), F-building (Fysikhuset), Campus
Valla
Abstract
In this talk we give an overview of the basic principles behind a paradigm
for a posteriori error computation in finite element analysis. An important
feature is the possibility to select "goal-oriented" error measures that
are of interest to the engineer (and not only to the analyst)! The chosen
error measure is used as a part of the associated adaptive refinement in
space or space-time in order to meet a predefined stopping criterion (tolerance).
A key feature of the error computation is the identification and solution
of an auxiliary problem, that is the dual of the actual (primal) problem
whose solution is sought. In mechanics the dual solution is known as "influence
function". In fact, the dual problem is linear(ized) even when the primal
problem is nonlinear, which seems to be an attractive feature for large
scale problems in computational mechanics where material and geometric
nonlinearities are commonplace. Nevertheless, to compute the dual problem
as efficiently as possible is a challenging task. The talk is concluded
by some preliminary results concerning the application to nonlinear elasticity
(error in space) and viscoplasticity (error in space-time) for simple model
problems in 2D. It is conjectured how to use the suggested strategies to
large scale problems.
Title: Large Scale Aerodynamic Flow
Computations on Parallel Computers
Speaker: Dr. Per Weinerfelt, Saab Aerospace
Time: Tuesday, March 13, 2001, at 16.15-17.00
Place: Room Schrödinger (E324), F-building (Fysikhuset), Campus
Valla
Abstract
The development of high speed parallel computers has made it possible
today to solve realistic industrial flow problems around complex 3D geometries.
The talk will discuss a number of aerodynamic flow problems, their difficulties
and show how these can be numerically solved using parallel computers.
We will also mention the importance of large scale flow calculations to
aerodynamic shape optimization and design.
Title: Climate Modelling of the Baltic
Sea
Speakers: Dr. H.E. Markus Meier (SMHI) and Torgny Faxén
(NSC)
Time: Wednesday, April 18,
2001, at 16.15-17.00
Place: Room Schrödinger (E324), F-building (Fysikhuset), Campus
Valla
Abstract
Within the Swedish Regional Climate Modelling Programme, SWECLIM,
a 3D coupled ice-ocean model for the Baltic Sea - the Rossby Centre Ocean
model (RCO) - has been developed to simulate physical processes on timescales
of hours to decades. The code has been developed based on the massively
parallel OCCAM (Ocean Circulation Climate Advanced Modelling project in
Southampton, UK) version of the widely used Bryan-Cox-Semtner model. Significant
changes of the code have been done with respect to surface fluxes, open
boundary conditions, mixing parameterization and sea ice modeling. An elastic-viscous-plastic
ice rheology is employed resulting in a fully explicit numerical scheme
that improves computational efficiency. An improved two-equation turbulence
model has been embedded to simulate the seasonal cycle of surface mixed
layer depths as well as deep water mixing on decadal timescale. The model
has open boundaries in the northern Kattegat and is forced with realistic
atmospheric fields and river runoff. Optimized computational performance
and advanced algorithms to calculate processor maps make the code fast
and suitable for multi-year simulations.
Selected results from key processes using two different horizontal resolutions
are presented and compared to observations. The agreement between model
results and observations is regarded as good. On the short time scale,
the existing picture of the major inflow event in January 1993 has been
completed. On the long timescale, the important question whether 3D models
of the Baltic can perform correctly over decades, has been addressed. Especially,
the results of the 13-year hindcast period from 1980 until 1993 are impressive
as erosion of the halocline in the Baltic proper could be avoided. Hence,
a tool to simulate long-term climate change and natural variability of
the Baltic Sea is now available.
Computational single and parallel processor performance issues will
be discussed.
