P573: Introduction to Scientific Computing

Introduction to Scientific Computing  Fall 2015

P573, Section number: 7387

8:00  9:15 AM, Monday and Wednesday

109 Ballantine Hall
Quick Links:
Not all of these exist yet, but will be filled in as the semester trudges
onwards.
General Information
Instructor:
 Randall Bramley
 301A Lindley Hall
 Office hours: Monday to Thursday: 1:30 PM  5:00 PM or by appointment
Associate Instructor:
 Anirudh RA
 301 Lindley Hall Suite
 Office hours: Friday: 1:00 PM  2:00 PM
 anramesh at indiana dot edu
Prerequisites:

Mathematics M343 and one of M303 or M301, and a working knowledge of
C or C++ or Fortran. The requirement in mathematics means
 At least the definition of a derivative from calculus
 Basic linear algebra, for example,
 how to add and multiply vectors and matrices
 how to compute an inner prodct, also called a dotproduct
 how to multiply a matrix times a vector, both mathematically and in
one of the languages C or C++ or Fortran
 what the null space and range space of a matrix are
 how to multiply two conformal matrices (and what
"conformal" means in matrix multiplication!)
 what an upper (or lower) triangular matrix is
The computational requirements include:
 The ability to write and run programs under a UNIX operating system, in
one of the languages C, C++, or Fortran
 The ability to write functions to a specified interface
 How to allocate memory, define and use arrays
 The ability to create executables involving multiple files
and libraries either by a script or a makefile
 Write programs that read and write formatted data from and to files
Try this short test to see if you have the kind of
math and coding abilities to succeed in P573. This is not a test to hand in; it is
strictly for your own edification.
Textbook:
No textbook will be required or followed. The following may be useful references,
but only for small parts of the course material. Math/CS books are expensive and CS
books are quickly outdated. Don't go out and buy any of the following until/unless
you are sure you would find them useful, and don't use them as a measure of what is
going to be in the course. As an example the first two references are mathematical
and emphasize the derivation and error analysis of numerical linear algebraic
algorithms  but all we will need and use are some of the algorithms themselves,
the implementation details, and occasionally the results of the analyses.
The rest is better addressed in a numerical analysis class.
In general all of the material you need will be presented in class, is available
via the Web, or (in the case of Matlab) through Matlab's
help, lookfor, and other commands.

Introduction to Matrix Computations by G.W. Stewart
is just what the name says; it gives the basic linear algebra ideas required.

Matrix Computations by Golub and van Loan is
a more definitive but higherlevel reference work, with numerical
linear algebra algorithms clearly stated in an implementable form.

High Performance Computing by Kevin Dowd
covers some performance optimization and measurement techniques.

Mastering Matlab by Hanselman and Littlefield is one of the
more definitive reference books for Matlab, and the one I use the most for my own
coding and debugging sessions. Matlab will be used for analyzing and plotting
results from some of the assignments. For the most part scripts will be provided
and they will also run using Octave, an open source Matlab wannabe.
Since this is not being used as a textbook, you can save money by getting an
earlier edition (but get one for Matlab version 6 or higher).
Computer Accounts:
If you have access or ownership of a Unixbased system in your lab or office, use it.
You'll want to develop the tools on the system you use in real life anyway.
Otherwise all CS students have accounts on the silo Linux system.
Students from outside the CS department need to give me their UITS
login to have accounts created. If you use UITS platforms, they
may be research computers and require some professor to sponsor the account,
unless you already have one.
Note that you don't have to physically sit at the console of those machines, and can
just login to them remotely via ssh, which provides for secure login and file
transfer. For Windows platforms ssh is available as the Putty app, downloadable from
UITS's IUWare and other places. ssh is on all IU Unix machines. Your UITS password
(from IU's ADS) will work on both CS and UITS systems.
If your lab or research group has other machines, use them as well 
in the long run, those are the machines to learn how to use effectively.
However, some assignments will require your codes to run
on a specified system so that we can have comparable results
across the class.
Course Description
Open to students from all scientific, engineering, and mathematical disciplines,
this course provides an overview of computer hardware, software, and numerical
methods that are useful on scientific workstations and supercomputers.
Topics include highperformance computer architectures, software tools
and packages, characteristics of commonly used numerical methods, graphical
presentation of results, and performance analysis and improvement.
The course is not the same as numerical analysis, which concentrates
on the study of convergence, stability, and error analysis in numerical
methods. For that, students should take Math 571572. Although numerical
analysis is an important component of scientific computing, it is only
a part of the field. Instead, this course concentrates on

practical implementation of solutions of scientific problems on computers

the efficient mapping of solution methods to modern architectures

software tools and methods useful in modern scientific computing.

the basic foundations of performance analysis, modeling,
and prediction (together, called performance engineering)
The last item is the most important one. The single most fundamental skill
you will need to master is loadstore analysis, and that is the
pass/fail criterion. Secondary tools and skills include how to recognize in practice
when problems in floating point arithmetic occur, how to write code that gives
scientifically reproducible results, how to efficiently implement
linear algebraic computational operations, and how to time and profile parts of codes.
P573 is not parallel computing.
Other courses including CSci
B673 concentrates on that aspect of scientific computing.
Some basic Matlab is coverd in the course,
a language that provides interactive graphing
capabilities and more importantly gives an easy way to recognize and use numerical
vector and matrix operations.
Matlab
is a rapid prototyping tool with easilyaccessed
graphics. Mostly scripts will be provided for you and
basic Matlab will be taught as needed in lectures.
Grading Policies
Grades are based on small projects, a midterm, and a final.
Each programming assignment will have questions
intended to begin your thinking process, not end it. Grading will also include the
questions you raise and answer in these projects. For example, if the question is
"Which of the methods A and B is faster," simply saying "A" is not sufficient. You
should also ask (and try to answer) the underlying question "What measure of fast
is appropriate here?", "Why is A faster?", etc.
Although some assignments can be done in teams, in all cases you must
follow my attribution of work policy.
Grading percentages:
 60% assignments
 10% midterm exam
 30% final exam, which is scheduled for 8:00  10:00 AM, Friday 18 December
2015.
All of the web pages posted should have a footer like this, giving a timestamped
log of any changes. Use that to know when something has changed or been added.
History of this page:
 Started: Thu 13 Aug 2015, 12:13 PM
 Modified: Mon 24 Aug 2015, 11:27 AM fixed links to requirements and
attribution or work pages
 Last Modified: Mon 31 Aug 2015, 11:38 AM