1st CMSI Poster Award Recipients
“Torrent” is a forum for CMSI interchange. On January 5-7 2011, Workshop: “Computational Materials Science: Challenges and Prospects” was held. This workshop was organized jointly by ISSP, CMSI and the Next-Generation Functional Nanomaterials
for Information Technology*. The three winners of the Poster Award at the workshop came together to discuss their individual research projects, their experiences using supercomputers, their interchange at CMSI, and the ties between young researchers.
* ISSP: the Institute for Solid State Physics, The University of Tokyo.
Next-Generation Functional Nanomaterials for Information Technology :
a group in Grand Challenges in Next-Generation Integrated Nanoscience Project.
|
■ Poster Award Recipients |
 |
You're all the type of people who love theory but also want visible output
Ohno : Dr. Misawa was awarded the Grand Prize Poster award for his poster entitled "Analysis of an Effective Model for Iron-based Superconductor Using the Multivariable Variational Monte Carlo Method."Mr. Suwa was awarded the Outstanding Poster award for his poster entitled "Markov Chain Monte Carlo Method without Detailed Balance."Dr. Shiba was awarded the Judge's Special Prize for his poster entitled "Biomembrane Molecular Simulation That Explicitly Includes Solvent Molecules in an Effort to Achieve Massively Parallel Computing."
By coincidence, all three of you are specializing in condensed matter physics, but I imagine that you each had different motivations for going into that field. Could you begin by telling us a little about yourselves?
Misawa : Since my undergraduate days, I enjoyed thinking and computing on my own, so I thought I would go into theory. It was after I entered graduate school that I began to study condensed matter physics. I liked the fact that it's easier to compare calculations and experimental results than in the case of elemental particles and atomic nuclei. Even the research I've done up to now in the area of condensed matter physics has been close to experimentation.
Suwa : I think condensed matter physics is a field in which you can enjoy "things that can be known." Elementary particles are not really familiar to us, and in chemistry and biology you need to establish a lot of hypotheses and you don't really feel that you've understood things. But in condensed matter physics you can consider problems that seem close to you, working from axioms and principles with only a few hypotheses. In my undergraduate days, I had to conduct experiments in the course of researching for my graduation thesis, so when I entered graduate school I wanted to study theory. In the laboratory, we use primarily the Monte Carlo method, and since I became affiliated with the laboratory I've been hooked on how interesting that method is. The theme of the poster for which I won the award was general modifications to the Monte Carlo method. I'd been thinking about that for a long time, and one day I suddenly got the idea.
Shiba : The other two are studying quantum systems, but I'm studying classical systems exclusively. I had a vague idea that I wanted to go into theoretical physics, but in the course of graduating from undergraduate school I was assigned to an experimental research laboratory. It was there that I encountered experiments of soft matter and linear non-equilibrium phenomena and other subjects. I thought the idea of "first just trying something" was interesting, and felt that there were aspects that were distinct from theorizing. At the laboratory I worked at in the master's program, we did full-on numerical calculation, so I began conducting simulations in which first I'd try something and then I'd try to figure out the results that were unexpected.
Ohno : So you all like theory and enjoy trying to understand things starting from axioms. But you also all want the output to be something that you can see. That's what you all have in common. Do you collaborate with people conducting experiments, such as by having them use the output from calculation in their experiments?
Misawa : I solve the effective models for certain materials, so I hold close discussions with the people conducting experiments on the interpretation of the experimental facts and whether or not they are consistent with the theory.
Ohno : I myself conduct calculations in the field of materials science. In collaborations with the people conducting experiments, I often find there's a wide gap between what they expect from calculations and what is actually possible through calculation, and this gap can't be bridged. What is it like in condensed matter physics?
Suwa : It's the same. If there's something in the model presented by the theoreticians that can't be explained, the experimenters are actually pleased. The theoreticians try to resolve it, but the experimenters use it as a jumping-off-point.
Ohno : That's true. Both parties would probably be in agreement in the case of a simple system, but with a real-world system it's hard to put together a simulation that will produce experimental results.
Dr. Shiba, you work on classical molecular dynamics. Do you get the impression that using the new supercomputer will produce results that approach what is desired by the people doing experiments?
Shiba : With classical model simulations, it's easy to expand the scale of calculations to match the capacity of the computer being used, so it may be possible to get to the point desired by the experimenters to some degree. However, it's not possible to conduct many calculations on an unlimited basis, so it's a problem that depends on what you select. It's not that it's impossible technically, but I get the impression that we need to consider more carefully exactly what calculations should be done.
Ohno : Is there anything that you on the calculation side would like from the experimentation side?
Shiba : Up to now, the collaboration has been possible only when the experiments have been done for those things for which calculations could be carried out. Now, however, there's a strong sense that calculations are trying to keep up with experimentation, or that calculations are performed in order to further develop what's being done in experimentation. I think that these two may slowly be coming closer to one another due to the increased scale of computing. So collaboration is important.
Misawa : In quantum systems, optical lattices have recently become popular, and collaborations are underway between theoreticians who are working to solve the Hubbard model and people who are conducting experiments with optical lattices. It's even becoming possible to vary control parameters at one’s disposal in experiments as in computational studies.
"I'd like to come up with a
|
 |
|
■ Moderator |
What are the expectations for the K Computer, and what will we be able to do with it?
Ohno : I'd like to ask you about your involvement with supercomputers. Have you ever thought, "I'll bet I could dosuch-and-such if I had the K Computer?"
Suwa : In classical systems, if it were possible to use it for mesoscopic system simulations, it might be possible to study transport phenomena and so on more carefully.
Shiba : When it comes to soft matter, there's a gap between macroscopic description of hydrodynamics and the microscopic description of phenomena, and it's difficult to bridge this gap. I think efforts to bridge the gap are needed, but on the other hand I'd like to find a single method that would enable us to punch right through it.
Suwa : In quantum systems, there are few calculations that are well parallelized, and there is a fundamental difficulty with parallelizing the calculations that can be used with the broad-based models that I want to use. To master the K Computer, methods themselves will need to be developed.
Misawa : If the K Computer uses several tens of thousands of dedicated cores, it would be easy to search for many things systematically.
Ohno : Normally we are conducting researches within the machine power we have. So, if a new large-capacity machine is suddenly given to us, it's true that it would be a bit difficult to answer when someone asked us "OK, now what's your research topic?" To use the K Computer, we'll need to improve our knowledge of computer science apart from our direct research topics. It's getting so that you can't produce results without some knowledge of computers, even if you know a lot about condensed matter physics. On the other hand, just because you know a lot about computers doesn't mean you can produce results in condensed matter physics. How do you think that balance should be maintained?
Suwa : Rather than mastering the details of parallelization technique, I'd like to work on basic computing methods.
Shiba : For the past several months, I've been doing nothing but parallelization, and I've used supercomputers here and there. And I've realized that it's important to actually buckle down and get hands-on experience rather than focusing only on theory.
Ohno : What percentage of your time do you spend developing code and algorithms and conducting parallelization processing? And what about with regard to the research for which you won the award?
Misawa : I use partly resources created by other people, so the amount that I create myself from scratch comes to about 20%. It was actually the computing time that took longer. Computation time is proportional to N³, where N is system size, so if the system size doubles, the computation time will increase eight-fold.
Shiba : I began conducting parallel computing in earnest in September of last year, and I was practically a hermit for three months. So for this project it was 100%. (Laugh)
Suwa : My normal approach up to now has been to take something I know nothing about and tackle it starting with an analysis of methods, in order to try to understand it all at once. So my main focus is on program and algorithm development.
Physicists need to be flexible: crude in the micro scale but refined in the macro scale
Ohno : The advantage of CMSI is being able to work with people from different fields. What is your impression of the other fields? Does the philosophy that you have when you conduct calculations differ depending on the field?
Shiba : When researchers studying the soft matter in physical systems look at molecular dynamics, they construct bold models. But I get the impression that molecular science researchers construct rigorous models and bring them to a supercomputer.
Ohno : I see. So the term "first principle" is used differently in molecular science and materials science. Apparently chemical science researchers use rigorous calculations that take into account exact electron correlation, and I understand that they refer to that as "first principles."
Suwa : In a sense, physicists are sometimes on the casual side. (Laugh)
Ohno : In physics, you often don't try to be any more rigorous than is necessary. As a result, it's possible that we may be broad in the micro
realm and exacting in the macro realm. Do you have a desire for expanded interchange with other fields? In materials a n d c o n d e n s e d m a t t e r p h y s i c s , w e sometimes study similar materials, but the two fields also have different academic societies, so you couldn't say there's much interchange. But it's interesting once you start working together, as you get a lot of different perspectives. Wouldn't you agree?
Suwa : The Monte Carlo method that I'm working on is employed in a broad range of fields, and there's a great need for it. I think listening to the views of a lot of different people would give you new ideas about the method. The Monte Carlo method originated in the field of physics in the 1950s, and it spread from there. Since the 1990s, it's been used in statistics and other fields as well. But it took 40 years for that to happen. I
think that having as much interchange as possible with other fields and having a shared awareness of issues would have a positive effect in enabling methods to spread more quickly. I'm personally making an effort to check out the methods used in a variety of fields.
Ohno : Mr. Suwa's work has a great ripple effect, so it'll be interesting if it attracts the attention of people in other fields.
Suwa : What's difficult about interchange between different fields is the fact that the terminology is different. I think that's the reason that interchange hasn't made much progress.
Shiba : In physics, the models are sometimes changed completely, but in chemistry the models of each atom are put into a library and used over and over. Molecular dynamics in physics is conducted from a stance of flexibility. I think constructing a framework f o r m o l e c u l a r d y n a m i c s t h a t c o u l d accommodate massively parallel processing would be useful for both sides.
Suwa : Physicists are interested in mechanisms, while chemists are interested in the materials themselves. When modeling as well, physicists create abstractions that omit the details, while chemists use the properties of atoms and molecules as is.
Misawa : Don't you find that many quantum chemistry researchers are concerned with quantitative attributes? In physics, many people adopt the stance of trying to think of the minimum model that will enable you to understand the essence, so their approach is qualitative. The research for which I won the award is close to quantum chemistry, and its goal is to create an actual model based on the first principles and solve it completely. But even in that effort, I have a strong awareness that I need to identify the essence that is concealed within that model. In that sense, I guess it's a physics model. I’m greedy. (Laugh)
Enabling interchange with different fields, information sharing and links between young researchers at CMSI
Ohno : What do you hope to get from the interchange provided by CMSI, regardless of whether or not you use supercomputers?
Misawa : I'd love to get large-scale computing expertise from other fields. Although it would be difficult to turn it into an academic paper, I'd like to be able to exchange information. At the first CMSI research seminar, I got to give a presentation to people I would not normally meet, and that was stimulating.
Ohno : It's been six months since CMSI was founded, and there is starting to be a demand for greater knowledge about the techniques of computer use. What do you think would prove useful in this regard?
Suwa : Open programs, for example.
Ohno : If we created a map, not for individual research topics but for a method-based one, for example one based around the Monte Carlo method as a way of showing that CMSI includes these kinds of research topics, I think that would attract people based on the method. It's interesting to exchange information with people who use different methods, so I think it would be good if there were a map that could serve as a catalyst.
Suwa : I'd like to see small-scale meetings where the discussion could really get down to the nitty-gritty.
Shiba : I agree. I wonder if it would be possible to have vendors to participate in that kind of meeting as well.
Ohno : That's an interesting idea.
Shiba : It would be great if we could share expertise for use when consulting vendors and topics for consultation. Normally things end with no one being able to see what was discussed or the results of the discussion.
Ohno : In the case of experiments, in some cases the techniques and so on are included in the written paper, and sometimes they're the essence that's needed to view the phenomenon. However, information like such-and-such a computer has a certain peculiarity is not made public. On the contrary, it's kept hidden.
Suwa : It would be great if there were a forum where everyone could freely upload that kind of information and people could consult it if they encountered a problem. I think having a Q & A section on the CMSI website would be helpful.
Ohno : It would be good if the circle of young researchers could be expanded as well.
S u w a : I ' d a l s o l i k e t o s e e m e t h o d development get more recognition. The development of methods doesn't get much acknowledgement as compared to overseas.
Ohno : I get the impression that it's difficult to get funding for method development. But without methods, there would be no progress.
Lastly, I'd like you to tell us your expectations for CMSI and any candid advice you may have.
Misawa : It's not clear to me what shape the organization is taking.
Shiba : I don't have a clear overall picture of what gets decided, and in what manner.
Ohno : Originally, the most important mission of CMSI was to discuss and decide who should use the next-generation supercomputer and for what purpose. Subsequently, a variety of objectives were added, but I'd like the focus to be on how to select embryonic research in a way that will make everybody happy.
Suwa : In that case, by all means let me use the next-generation supercomputer.
Ohno : In April 2011, use of the K Computer will begin at last in Kobe. In the near future, we'll have to nail down what will be needed after that. Each generation of a supercomputer lasts five years, so I'd like you to not be shy about speaking up as soon as possible regarding the next generation.
All : Understood. We'll do that.
Ohno : I have high hopes for your work at CMSI.
(Recorded on February 24, 2011 at the Department of Science, The University of Tokyo)
Text : Etsuko Furukori
Photos : Shuichi Yuri
