SIAM ALA 2009 Panel: Linear Algebra in Industry
Panelists: Iain Duff, Roger Grimes,
Volker Mehrmann, Rosemary Renaut
Chairs: Mark Embree and Esmond Ng
Report by Chen Greif
November 15th, 2009
A 75 minutes long forward looking session entitled ‘Role of Linear Algebra in Industrial Applications’ took place on Wednesday, October 28th, 2009, as part of the SIAM Applied Linear Algebra Conference in Monterey, California. Mark Embree was the moderator. In his opening remarks he aptly defined the main focus and goal of the discussion to be: how to keep the ‘i’ in SIAM. Mark presented three topics for discussion:
- the role linear algebra plays in industry;
- the role of industry in education;
- ways to form collaboration between industry and academia.
In the discussion that followed the first two topics were discussed, but the lively exchange occasionally touched also upon the third topic.
The Role of Linear Algebra in Industry
The discussion started by
addressing the question what role linear algebra plays in industry. Methodologies
that were identified were: extensive use of iterative solvers; structured
eigenvalue problems; parameter dependent linear systems and eigenvalue
problems; direct solvers; uncertainty (though a comment was made
that it is not covered properly by the techniques we currently have).
On the question of what computing environments are used, the response of the panelists was that industry is quite inhomogeneous, and it is reflected accordingly. There are multiple platforms and operating systems, and the languages used are anything from Fortran to C++ to Java. There is not always correspondence between the language taught at a computer science department, and languages used by the scientific computing community. Fortran is a prominent example here, of a language extensively used but not taught; Java is often on the other extreme. It was suggested that Matlab cannot solve everything (and one page algorithms cannot either), and students need to be more open to learning and using other languages.
In the oil industry, for example, what dominates are very large scale computations, and hence large parallel architectures. Smaller companies, on the other hand, deal with applications such as CAD/CAM, done on large desktops (8/16 core). Multicore is now becoming a type of architecture common to small desktops and super computers, and perhaps this will change the way research and development is done. Direct solvers, for example, take this into consideration.
Another issue is the use of software packages. Trilinos and other national lab packages are used, but industrial companies often have their own legacy code, and want code tightly tailored to their needs.
How to Educate Linear Algebra People and Prepare them for
Industry
The general sentiment was that PhDs are valuable for industry.
But it was pointed out that there is a gap between what the success of
a supervisor is measured by, and the definition of success in industry. A
professor’s measure of success often goes by where in academia his or her
students end up; preparation for industry is driven by other parameters. The
issue of keeping deadlines also came up. There was agreement that there are
fundamental difficulties in keeping deadlines in an academic environment.
The type of preparation vastly varies between different places. In Germany, 60-80% of students go to industry, and German programs are strongly geared towards industry. This is in contrast to the typical situation in North America.
The general feeling among the panelists was that there is a true need to think beyond narrow linear algebra issues, and the curriculum needs to be broad. Students need to know statistics, experimental design, scientific computing. Of course, we generate applied mathematicians, not engineers, so cannot cover all of applications.
Communication skills are important, and summer jobs are useful, although limited. Typically, students spend 4-6 weeks in a company. But the company’s people often do not have time, and students end up doing lower level things.
There are special programs where the company pays also the adviser of the student. It works well, and makes the university commit more. The panelists expressed support for this idea, and for getting faculty more involved in collaboration with industry. There are several technical difficulties in making this happen, though, not only on the side of professors but also in terms of making it work for international students.
There are many industrial workshops, e.g., IMA workshops, and other opportunities provided by NSF. Typical students that attend those workshops come from mathematics, computer science and engineering disciplines. There is a DOE program, CSGF (Computational Science Graduate Fellowship), which is currently very small, and we should make an effort to make it more known. There are also industrial workshops, NSF Graduate Research Fellowship, and again there is a strong need to make people more aware of those opportunities.
