Interdisciplinary Computational Science Town Hall Meeting
Interdisciplinary Computational Science
Town Hall Meeting
Tuesday, February 12, 2008
Meeting Summary
Background
Faculty from all schools and colleges on the University of Colorado, Boulder campus were invited to participate in a town hall meeting on interdisciplinary graduate education in computational science and engineering, high-performance computing and supercomputing research support needs, and other topics related to the formation and pursuit of the initiative. This town hall gathering followed a series of five open meetings which took place during the fall 2007 semester involving faculty members from very different disciplines. Prepared yet brief presentations were made by Mark Ablowitz, Juri Toomre, Tom Manteuffel, John Cary, Kamran Mohseni, and Dennis Maloney followed by an open discussion. The purpose of the meeting was to stimulate discussions and gather input from the campus community related to the many issues concerning the formation of an interdisciplinary graduate-level program on computational science and engineering at both the doctoral and masters levels, including professional masters and certificate programs, related curriculum, resource needs, research computing needs, and next steps.
Stein Sture, Vice Chancellor for Research and Dean of the Graduate School, gave a welcome and opening remarks that outlined the meeting agenda. Ten years ago, the Chancellor at the time, Richard Byyny, presented a similar initiative as one of the top five campus priorities. Unfortunately, at that time there was not enough support for the idea. Now, the initiative is being revisited because of campus wide interest and opportunity for resources to be earmarked for computational science endeavors. Before moving forward, Dr. Sture intends to accurately gauge the level of faculty interest. If there is enough interest, a steering committee, made up of faculty members, will be created who would develop a white paper by September 2008. Finally, Dr. Sture noted that colleagues from both NCAR and the Colorado School of Mines were in attendance at this town hall meeting and interested in exploring the opportunities presented.
Mark Ablowitz gave a short presentation regarding the need for structure in providing a program for interdisciplinary computational science and engineering (ICSE). He provided a summary of earlier meetings which took place monthly in the fall of 2007. His observations concluded that there is a significant interest in computational science and engineering on the Boulder campus, other CU campuses, and at other institutions along the front-range. He also found that large scale computational research/investigations are wide spread among units on the Boulder campus and that substantial collaboration crossing academic boundaries is needed. Additionally, there is value and a need to provide structure for a broad ICSE effort and there should be specific curricula for a campus-wide interdisciplinary computational science and engineering graduate program.
Ablowitz noted that there are new and emerging fields in ICSE driven by new concepts and new research directions. There is a wealth of data that is generated for many applications, i.e. data mining. There are well-known research areas that now can be significantly advanced with high performance computing/machine development. Software and hardware developments have been impressive. There are also regular conferences and journals solely dedicated to ICSE. Ablowitz asks about effective and efficient teaching of computational science and engineering. He states courses in ICSE might be taught by faculty as cross-listed courses or as new courses. The ICSE effort could provide PhD and masters degrees, professional masters and certificates, colloquium, and seminars as forums for new ideas and concepts. Assuming there is a broad and serious interest in ICSE, a faculty committee would need to be formed to establish ideas, concepts for what model(s) might be most appropriate and discuss general curriculum issues (courses and degrees).
Nationally, there are ICSE programs that cut across a variety of disciplines. A structured program can offer courses and training and possibly give potential to a professional masters program, certificate program, PhD and master degrees. Other universities have programs in ICSE include UT Austin, Florida State University, University of California-Davis, Stanford, University of New Mexico, University of California-Santa Barbara, Cal Tech, Chicago, McGill, and Purdue. Ablowitz ended his presentation with the question, is this the time for CU-Boulder to embark on a ICSE initiative?
Juri Toomre, from JILA, gave a description of computational science with an example drawn from astrophysics and space astronomy and how the field is using computational science as a tool for data. For new researchers, the skill set in computational science will be very important. Rapid advances in hardware (driven by market forces) enable a vast variety of innovative computational modeling and simulations analysis and visualization that are now pervading our society. A quiet but transformative revolution is underway that is flattening the world (a concept developed by Thomas Friedman). Toomre asks how do we excel in such a world? Individuals can be anywhere and be a part of this transformation. How do we teach students to transform society? Clearly, computation is changing some of the mechanical means in our society. Some of the answers lie in tools like agile workstations (even laptops), fast networks and servers, and even massive supercomputers. We need investments to train the brain power to exploit and utilize simulation tools in various disciplines (business, art, etc.) we need to advance the use of such hardware that is evolving at a rapid pace. Protective agencies such as Microsoft try to control growth, which leads to confusion. We are not currently teaching elements involved in computations in any organized or cohesive fashion. Such teaching could be coordinated across departments, along with new faculty lines to bring in skills needed in fostering such teaching. Teach the art of devising new approaches and tools to use the computational revolution to our advantage. Do our students have access to creative things? A young researcher will need to know varied numerical analysis approaches used to represent 3-D partial differential equations (of compressible MHD), architecture and communication within massively parallel supercomputers that impact choice of numerical approaches, how to devise or adapt major simulation codes to address specific research goals, deal with effective analysis of vast data sets from simulations, across a teragrid of machines and visualization tools, and a firm analytic background in the complex physics studied under these highly nonlinear conditions.
Tom Manteuffel defined computational science and engineering as: “ICSE is a broad multidisciplinary applied mathematics area that encompasses applications in science, engineering, applied mathematics, numerical analysis and computer science, modeling physical phenomena in astrophysics, biology, climate modeling, device modeling, earth sciences, fusion stimulation (ITER), global weather, complex systems in operations research, network design, power grids and biological systems, data mining and data visualization, image processing, and pure math: theorem proving. The science of using a computer includes high performance computing, computational math, numerical analysis, applied math, software engineering, and computer science. The science of designing and building computers involves computer science and electrical engineering in the areas of processor design, inter-processor communication, and architecture design. All of the discussed components are important to the enterprise and any successful ICSE effort will foster interaction between components. The computational modeling process involves physical model, mathematical model, discrete model, solution algorithm, software implementation, verification, validation and visualization. No single discipline spans the ICSE process so interdisciplinary teams are required. The role of computational mathematics as described by the National Science Foundation High Performance Computing Center, Blue Book 1992, “our ability to model complex physical systems has benefited more from new mathematical algorithms than from the explosive growth in computer speed.” Manteuffel is a member of SIAM (Society for Industrial and Applied Mathematics) and has been involved in the federal labs. He feels that the people attending the meeting don’t need to be convinced of the usefulness of a ICSE program. He highlights the website www.siam.org/students/resources/report.php for ICSE information. The website describes ICSE and gives wonderful graphics which really show the intersection between applied math, computer science, ICSE and engineering/science. The report describes a variety of disciplines that use ICSE. Models for graduate degree programs are explained on the website as well. Manteuffel states the Boulder campus ICSE program needs to ultimately survive on its own. The University should support application scientists with hardware access to tera and peta scale computers, algorithm and modeling consulting support, training for students, new faculty, and postdocs, and new evaluations and rewards systems. The ICSE initiative requires financial support, new organizational structure, and facilitated interactions with industry and federal labs. What will not work at CU is throwing words without money, there needs to be an organizational structure that cuts across a variety of disciplines—i.e. cut across the stovepipes. Throwing money at the problem does not work, there needs to be buy in from the faculty and administration. To survive, the ICSE initiative needs external funding, the opportunity to foster research in modeling, ,algorithms and software design. To survive, it is imperative to train the next generations of computational scientists, serve the needs of application scientists, and foster interactions with industry and federal labs.
John Cary, Physics gave information on the future predictions of ICSE and highlighted some examples using laser beams. There is value in both theory and experiment but computational science is the 3rd leg of inquiry because of the incredible detail – in some instances more accurate detail than an actual experiment. Currently, we have lasers for medical and scientific research. We have accelerators for particles but they are not precisely accurate. Simulations provide accuracy, can handle complexity, and can explore so far unrealizable parameters (i.e. our simulation efforts with our laser beam provides three times better data than an actual accelerator or conventional means) – we tagged them “dream beams”. If you are going to publish, you will need corresponding computational simulations to back up your claims. Increasingly there is a well-defined discipline of these enabling technologies through the use of parallel computation, handling large terabyte and petabyte data sets. Access, analysis, and visualization—where do these domain folks fit? Domain folks include physicists, chemists, aerospace engineers and they don’t exist in a vacuum. Domain folks need to bring application-specific algorithms that may be generalizable. Don’t develop an algorithm in isolation.
Kamran Mohseni, Aerospace Engineering has research interest in mobile sensor networking using unmanned vehicles, micro/nanofluidics, and turbulence. He directs two laboratories; multiphysics laboratory which is equipped with some fabrication capabilities for MAVs and UUVs, wet benching, spin coating, flow visualization capability (microscope, CCD camera, high speed camera 33000fps, etc), hot wire anemometry, water tank, low Reynolds number ‘1x1’ wind tunnel, and the computational laboratories equipped with a 112 processor PANTA system, a 48 processor SGI 03000, 8 PCs, and 2 printers. He stated, we are users of computers not developers of computer systems. We use computational tools and also develop computational tools if needed. Mohseni’s group consists of students (current or past) from many different disciplines. The 19th and 20th centuries have seen the significant patrician of science and technology in several disciplines. Such classifications, could be barriers to a truly multi-,inter-, or cross-disciplinary research needed in the 21st century. Dr. Mohseni also spoke about using the two labs and utilizing students from aerospace, mechanical, computer science, electrical engineering, applied math, and physics. The science grew so fast that we categorized ourselves into disciplines -- multidisciplinary (knowledge associated with more than one existing academic discipline or profession), interdisciplinary (knowledge extensions that exist between or beyond existing academic disciplines or professions), cross-disciplinary (knowledge that explains aspects of one discipline in terms of another) and trans-disciplinarity (knowledge that exists in every individual, thus eliminating the need for discipline boundaries). He gave the example of the visualization of an octopus and the locomotion of a jellyfish. After plotting their movements in a simulation, the lab was able to design an actuator, which led to designing a submarine with “parallel parking abilities”. Mohseni outlined some barriers and comments: first, discipline pride and willingness to fight with others rather than solve a problem. Nature doesn’t understand these barriers. Accept the other side – learn the language of other disciplines. Second issue: tenure and promotion – we have an inherent belief that what we do is better than what another discipline does. We need to accept new ways for tenure and promotion.
Third, funding: interdisciplinary funding needs to be more of an anchor point. Forth, publications and budgetary issues: we are competing with traditional disciplines and need to break new ground. There needs to be a balance between hardware and software. Finally, we need to do a good job at student recruiting.
Dennis Maloney, Executive Director, ITS spoke about the CU Boulder Cyberinfrastructure which is defined as coordinated aggregate of software, hardware and other technologies, as well as human expertise, required to support current and future discoveries in science and engineering. The challenge of Cyberinfrastructure (CI) is to integrate relevant and often disparate resources to provide a useful, usable, and enabling framework for research and discovery characterized by broad access and “end-to-end” coordination. We have network connectivity through Internet@ and the National Lambda Rail and there is localized CI support for the data center, HPC, software tools, collaboration tools, human expertise, networking connectivity, storage capabilities, and ability to get, or have cycles. Future CI support at UCB is to be determined. Maloney would like to interview researchers to solicit their point of view on how and where the campus should invest in CI support. If faculty members are interested, please contact stein.sture@colorado.edu or dennis.maloney@colorado.edu
Questions from faculty
John Cumalat, Physics
Resonated with the ideas Juri Toomre proposed, especially about programs and classes. Beyond that, he is not sure what is being proposed. Computation is important – where it would be located and housed. He didn’t see the added benefit. He pointed out that most of us are tied to international or national connections – beyond this campus. Where do we get the added benefit outside teaching?
Stein Sture, Vice Chancellor for Research and Dean of the Graduate School
Hopes the white paper will hammer out the answer
Jeff Vasil, from JILA and a PhD student in Toomre’s group
Responded to Cumalat’s concerns: stating, it would have been easier if everything was located in the right place and one spot, which will be beneficial if the ICSE initiative goes through. It was tough to learn the architecture, not getting it to crash, programming techniques, etc.
Tom Manteuffel, Applied Mathematics
Responding also to Cumalat: the forum at University of Texas at Austin brought in visiting scholars. Each had their own discipline, but were drawn to the multidisciplinary field. It’s the Texas Institute of Computational Science and has a $100M endowment.
John Cumalat, Physics
Stated University of California-Santa Barbara created a very successful ICSE center which was funded by NSF and then the funding was pulled and it died. He can see how there could be a master’s program but not necessarily a PhD.
Xiao-Chuan Cai, Computer Science
Made a case for the need on campus for such a center or institute. It would help with more understanding on campus of the discipline. Wants to understand the needs of multidisciplinary research. There should be both efficiency and effective teaching.
Per Gunner Martinson, Applied Mathematics
In answering Cumalat’s concerns about having a Ph.D: program would be greatly facilitated by core structure, guest speakers, critical mass and ultimately an educational experience. TICAM (Texas Institute for Computational Analysis and Mathematics) provided an excellent example of a multidisciplinary program. The students could choose a specialization area and an application area.
Having ICSE would develop strong connections for students and the ability for them to choose ICSE as a specialization.
Carlos Felippa, Aerospace Engineering
What type of entity? Department? College? Center? Institute? Yes, Texas Institute did well because it had a champion, enterprise, fundraiser, donated building. Need TICAM collaboration – the effort at Standford died. Asked what is the ultimate goal for outcome involving several colleges and schools, creating an institute? Felippa felt the campus needs a champion for the cause.
James Syvitski, INSTAAR
Certain countries will not have access to supercomputing equipment that might be exported from USA – nothing above 20 teraflops – i.e. China. Export control remains an issue.
Charles Musgrave, Chemical and Biological Engineering
I transferred from Stanford and understand why it failed and I’m happy to share some information. One reason: promotion and tenure – there are often questions of who owns the intellectual problem. It’s critical to set up a value system about who receives rewards for their contributions. Consistency is needed in contributing to an institute and to their department. I have several reports and white papers available if interested in reviewing.
Transferred from Stanford to our campus and felt there were challenges for promotion and tenure in the field because computational scientists are looked at more as service providers rather than hard scientists. He brought information from Stanford and offers the information to Dr. Sture for using in the ICSE pursuit.
Rob Knight Chemistry and Biochemistry department
There are a dozen departments – model has been successful. Enthusiastic computational needs expanding rapidly.
Juri Toomre, JILA
A number of universities want visibility (i.e. Indiana) they would like to provide techniques across broad range to students – do require attention, but less costly. There needs to be more coordination and nurturing, but this should be predominately about how we teach students, not about conducting research at the institute level. We are not too far removed to being much better organized about how we teach the art of computing.
Paul Beale, Physics
Something needs to be useful to the researcher and assist in making a connection between faculty member and student in the area of computational sciences. Scientific computing is highly specialized. He asks how this would happen.
Charles Musgrave, Chemical and Bio Engineering
Does the teaching extend to the undergrad level? We have math at undergrad levels for engineering students, but students come from other institutions – service courses – people who aren’t quite as deeply invested and prepared for this?
Jim Curry, Applied Math
We, within the Applied Math department, have had the stewardship of teaching engineering students and done that. I agree that there should be a vertically integrated structure -- would be a success. Fighting that effectively and successfullyspecific battle at Stanford is resolved here.
John Cary, Physics
Undergrad education is needed. There is enough physics with software development – make enterprise work.
Jeremy Siek, ECE
Excited about opportunity – dissatisfaction with CSE. Wants to know how to build flexible software that runs really fast. Not all technology are mature on the software side, we need more resources and investments in the computer science side. Fortran was early programming language since then, general purpose stuff – i.e. JAVA, stuff for scientists that is easy to use that is fast.
John Cumalat, Physics
Regardless of what we call it: computing college/institute, etc. – the university community at large was against it, in part because resources would ultimately be siphoned away and tied to other agencies. Why have things changed to revisit this concept? I’d challenge the notion that they haven’t. He questions why things need to change at all.
Mark Ablowitz, Applied Mathematics
To respond to Cumalat: when we look at this, years ago there was a downturn in the economy and funding – we couldn’t put any money into it.
What happens to my center? Will it continue to be supported? Does everyone if uses the same platforms, specialized code, language? It isn’t clear that having a centralized facility would be the right approach. We currently have 17 SS clusters across campus. Chilling/cooling issues, power issues, upgrades issues need to be addressed
Texas is successful. Can we develop a ICSE that can reach out to units, and not harm successful efforts? What happens is that funding goes into initiatives and gets taken away from the existing units
Carlos Felippa, Aerospace Engineering
Five years ago the computer science department looked into forming their own college. Essentially the donors blocked this but now we have ATLAS – what is their mission? Where is the overlap? Parallel but separate way – what’s the large overview/strategy?
At this point, time ran out and Dr. Sture invited individual conversations to take place as people departed.