Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 91
Session 2 Panel Discussion Robert Lichter, Camille & Henry Dreyfus Foundation: During this joint presentation, all of you in one way or another talked about the way in which people will be an integral part of the whole revolution by 2020. I would like to hear your comments on how we will get there. That is, what are the implications for the education of the people who would be part of that revolution that would presumably begin now? David Dixon: I will try and make a stab at it, although I am not in the formal education business these days. Much of what we are going to see will actually be on-the-job training. When I was at DuPont, for example, we did not see the broad training you need to actually solve some of these problems in new staff coming directly out of universities. In the future, we are going to be seeing teams of people working together. We are going to have to change fundamentally the concept of how individual research is done. If we want to solve complex software issues and things like that, we need to look at team approaches. We have been developing very complex parallel code at PNNL, and the only way we have been able to do it is by putting teams of applied mathematicians, computer scientists, and chemistry domain specialists and users together to solve the problem. Fundamentally, academics will have to change to allow teaming to happen and for that to be a profitable part of the university curriculum. David Smith, DuPont: Can I just try to add to that incrementally? The issue seems to me that the university must and will continue to rely on the individual contribution in the thesis area, and that is an essential part of getting your Ph.D. On the other hand, when you come to work at a company like DuPont, and I am sure, Dow, teamwork is important, and I think that most of the exciting work that is going on in our corporation today is not so much in any one particular field, but in the intersection between fields where interesting things are happening. So, the ability to have people from different disciplines work together and to be able to understand one another's language is really an important part of the process, and it takes time, and that time comes from DuPont's time. I do not think that there is any way the university can give that to us.
OCR for page 92
Gintaris Reklaitis: I would like to add to that point from the perspective of the software tools. The question that I have is, How do you teach people to use the computing tools that are getting increasingly more complex? I really think we need a lot of new ideas in this area. I am struck by the fact that we have tools like process simulators that are very time consuming to master. If you look at the user's manual for complex software tools, the first thing the user is asked to do is to spend 2 hours going through all the menus and clicking on all of the commands. Have you tried this with a group of undergraduates? It lasts about 5 minutes. This is absolutely not the way anyone wants to learn to use a software tool. We have to come up with intelligent ways around this training problem. Specifically, I think that there is a lot of creative work that we need to do with our education colleagues to devise new models for learning and intelligently using complex software tools that have many options and possibilities. This is particularly relevant for the casual user, which is what most industrial users are. Most engineers are focused on projects rather than tools and will only revisit the tools periodically as the need arises. Such a practitioner surely does not want and does not have the time to reinvestigate all of the menus every time to recall what is available. Evelyn Goldfield, Wayne State University: I want to follow up on the previous question because I think that a lot of universities, including my own, are rethinking some of the ways that we train graduate students and really do want to focus on some of these interdisciplinary team approaches, particularly in computational science. We found that there were a lot of different people at the university who should be and could be training together and working together but until recently were almost totally isolated. There tends to be duplication of effort, of reinventing the wheel. Also, too many students are not taking certain courses that would benefit them because these courses are offered in a different department or a different part of the university, whereas they are all using basically the same algorithms. And so, I wondered if you had any comments for how that these sorts of team and interdisciplinary approaches at the university would have any beneficial effects on your projects? Gregory McRae, Massachusetts Institute of Technology: I would like to address that question from an MIT perspective because, in fact, just last week a whole new division has been formed in the Engineering School called the Engineering Systems Division, which is specifically directed at dealing with the issues that you talked about. It is across disciplines. It involves not only interaction with the Engineering School but also with the social sciences as well as with the management sciences. It is by no stretch of the imagination an easy thing to do in a university, but there are people with vision to basically say this has to be done and they are putting the faculty slots on the table to actually make it happen. So, I think that there are some universities quite committed to doing that. David Smith: I would just like to comment that the Design Center at Carnegie Mellon University had exactly the same strategy and was very, very effective for those of us who participated in it. It was a very good experience. Judith Hempel, University of California, San Francisco: I was just going to ask you, the panel, what you see for the year 2020 in the sort of division that we currently see in the chemistry modeling area between what some people call materials research and on the other side, biological materials research. Many of the techniques are very similar when you go all the way from pharmaceutical modeling over to materials modeling. In 2020 will there be a division, do you think, of this kind? Or will it come together? David Dixon: In principle, one would hope that they would come together. I would not guarantee it at
OCR for page 93
this point, and the reason is that it is not clear to me yet that the potential function needs for both are going to be the same. I think you will have similarities, but right now if you look at industry, for example, I would say that in the use of computational chemistry throughout industry, including chemistry and pharmaceuticals, you probably have 80 percent of the people in the pharmaceutical/biological and about 20 percent in the materials/chemical side. And I think it is going to depend on a lot of features. One would hope that they would come together, but one does not know that yet. Stanley Sandler, University of Delaware: The one thing I was surprised to see on your list was the comment for the year 2020 of the need for efficient methods for generating accurate potentials for molecular simulation. Intermolecular potentials generally imply two-body effects, not considering pairwise nonadditivity. Wouldn't you think by 2020 we would have done away with that completely, and be using quantum mechanics and simulation together to calculate the total energy functions, and not calculating individual two-body potentials? David Dixon: One would hope that we would be able to do the quantum mechanics then, but, as Peter Taylor mentioned this morning, it is extremely expensive to get those very small energies correctly. It is not clear when we will be able to carry out the kind of simulations that Peter Cummings is doing in order to obtain thermophysical properties that will not require potential functions. The potential function issue has come up at about three or four workshops we have been at over the last year as being one of the key foci for the next 10 years, if we are going to make headway in predicting thermophysical properties. Stanley Sandler: How do you take into account nonpairwise additivity or non two-body effects? David Dixon: That is part of generating the potential functions. Do we put in polarization potentials, and how do we put in three-body terms? There is a whole branch of science in how to do this correctly. People are now putting in polarization potentials so you can actually treat the electrostatics of a molecule as it interacts with other molecules and solve for the electric field changing over time with the dynamics. We need to worry about three-body effects, four-body, and up to n-body effects. These are areas that have to be studied in order for the field to progress. Christos Georgakis, Lehigh University: Besides the industrial need for people who have been trained in several disciplines, maybe one other issue that we need to discuss is the type of academic degree(s) these people should have: B.Sc., M.S., or Ph.D.? More specifically, do we need more M.S. graduates than Ph.D.s? David Smith: I will make an attempt at answering that. The first thing I would like to comment on is that chemical plants have a very long lifetime. They are on the ground for anywhere from 25 to 50 years. No matter what we do today in research, those plants are still going to be making adipic acid well into the next century, so we still need students trained in traditional chemical engineering who are going to go out and run those plants and manage our businesses in the traditional way. We just cannot walk away from that responsibility. There are too many billions of dollars of equipment on the ground to think that we should change the way we have been doing chemical engineering because in the next 10 years we are going to be building biological processes to replace them. This is just unrealistic. Now, because of where we believe the growth to be, there is a trend in DuPont toward hiring more in the biological sciences, but that does not mean that we are going to stop hiring Ph.D.s, well trained in
OCR for page 94
chemical engineering and other disciplines, especially chemistry, to do our work. So I do not see DuPont R&D shifting away from hiring Ph.D.s. I see us perhaps not hiring quite as many as we did 10 years ago, but there is another complicating factor here, and that is the demographics. If you walk into the DuPont cafeteria at lunchtime and take a look at the people there, there are a lot of chronologically gifted people present. We face a problem in terms of maintaining our technical capability in the next decade as these people retire. We have had a hard time getting through to our HR people that they cannot solve that problem in one year. There are not enough high-quality Ph.D.s in chemical engineering graduating in any one year for us to replace the people we are going to lose at that particular time. There is a fundamental problem but I am not sure that it is being addressed in the best possible way. Tom Edgar: We talked a little bit about what the factory of the future is going to look like and the sort of technical demands that are going to be placed on people who are there. There is also this ongoing pressure to reduce the number of personnel in chemical plants as they become more automated. What is the operator of the future going to look like? Is that person going to need a B.S. in chemical engineering rather than a community college degree? Someone in an editorial recently likened that person to essentially having the same responsibility as an airplane pilot in terms of the financial implications as well as possible safety implications. So, it is, again, something to think about. The differential costs between an operator and a B.S. chemical engineer are not all that great. David Smith: I would like to say that the salary for a senior operator it is not that much different from the starting salary for a Ph.D. I had a young Ph.D. go to the plant and discover what the lead operator was making, and he came back absolutely incensed and wanted to know why I wasn't paying him more. I simply asked him, “Can you run the plant as well as he can?" The answer, of course, was no. Gregory McRae, Massachusetts Institute of Technology: There is also another dimension of your question that I do not think we should ignore, especially if we are looking at 2020. The training that students have in chemical engineering makes them very attractive to a lot of people other than those who make chemicals. In a typical graduating class at MIT, not many of those kids finish up as chemical engineers because of the tremendous salary offers they are getting from many other disciplines. So, it is not only the question of how you can feed people into the existing industry to deal with the problem that Dave is describing, it is also how to make the industry itself attractive to these people, because they are able to go to many other different places. For example, the last three of my Ph.D. students have gone to Wall Street, and for one of them, his first-year bonus is more than I will make in my lifetime. Christos Georgakis: Let me retry a more focused question. Fathers and mothers see the incentive to pay for a B.S. education. University professors who have to do research see the incentive to educate Ph.D. students. Where is the financial incentive to educate M.S. students, which, I personally believe, will be in much larger demand? These M.S. graduates are needed to operate model-directed plants, or to utilize and apply the sophisticated computer technologies that the 2020 framework envisions. David Smith: Christos, I am not sure whether it is a chicken and egg problem. Today if I wanted to, I could not go out and hire master's degree candidates in the numbers that we might be interested in hiring. So if they were available and we had experience dealing with that situation I might be able to give you a meaningful answer. In point of fact, I think that we are going to hire more B.S. engineers to
OCR for page 95
run our plants. I do not see us requiring more master's degrees. I think a lot of the mining that we want is going to be on-the-job training because of the increasing interdisciplinary nature of the work we will be doing in the future. It is too difficult to target that. I just think we are going to rely on more on-the-job training. Gintaris Reklaitis: One might ask whether industry will continue to be wanting to invest in on-the-job training because at universities, we hear from our industrial colleagues that they want graduates ready to hit the ground running. They do not want to invest in training or in career development. David Smith: Oh wow. I do not work at one of those places, I am happy to say. Sam Kounaves, Tufts University: Parents pay for B.S. degrees and faculty pay for the Ph.D. degrees. At our university nobody wants to pay for M.S. degrees because no master's student is willing to encumber another $20,000 tuition bill on top of the bill for the B.S. degree, and we cannot support our master's students because they are not there long enough to do any research for us. At least that is my perspective on the problem. The other question I have is more general. Greg talked about a balanced approach to doing modeling, both in terms of software and hardware, and I was just curious about your opinion on the allocation of resources for supporting computational chemistry. At first it sounded like we need all the high-powered hardware to do advanced modeling in all branches of chemistry and then later it sounded like the problem is actually in the software. I am curious about what your opinions are in terms of the resources. Are we at the point now where the hardware is really all right except for specific cases and that a larger effort needs to be put into funding programming and algorithms? I think people would rather put funding into hardware in the computer sciences, but software has been relegated down to the bottom. But that is exactly where the resources need to be put in order to address some of the problems, like interfacing and operating systems and interoperability. David Dixon: Actually, I do not think that there has been a decision actually to put all the money only into hardware. I think one needs to continue to have hardware that is going to give us new ways of solving larger and larger problems, but you have to have a balance between the software and the hardware. I think ASCI and the Strategic Simulation Plan are trying to be balanced on both. I think the point that I was trying to make earlier is that the operating system part of the hardware has been the weakest part for us in terms of making it available to a broad range of users. We are significantly investing in all of the software pieces, the algorithm development and the theory, and we are trying to have a balanced approach. I think everybody has been trying to do that. Paul Messina, California Institute of Technology and Department of Energy: But, yes, in terms of the investments, even the ASCI program, which is known for buying big machines, is spending less than half of its money on hardware. The rest of it is on developing the software and the applications. And your comment about computer scientists wanting the hardware, I am afraid, is wrong, because computer scientists typically do not want any hardware. Tom Edgar: There is one other aspect of computational speeds that I would like to clarify. In real-time process control, there is an imperative to try to generate an answer within a sampling time or within a time constant of a process. That is a little bit different than for a simulation environment that is off-line. So there actually is a fair amount of pressure to have hardware that is really fast. But, of course, having said that, with the doubling of processor speeds every 18 months according to Moore's law, you can
OCR for page 96
always figure that with a good algorithm you can just wait 18 or 36 months and the computers will be fast enough so you can use your algorithm then. Gintaris Reklaitis: From the perspective of combinatorial problems arising in supply-chain management and scheduling, doubling of computer time every 18 months will not suffice. These are problems for which computational effort grows exponentially with problem size, at least in the worst case. For the kinds of increasingly larger applications that people want to solve in these domains, waiting for the hardware to become faster is not the solution. The improvements must be found through algorithm research. David Smith: I do not know. I used to have a group that did supply chain optimization, and in the reorganization that group went elsewhere. I would say that for a lot of very good sized, really significant supply chain problems that we were tackling, we were getting solutions in a couple of hours. The problem that we ran into was that, for reasons I could not understand, the people who wanted the answers were upset because they had to wait 2 hours for them; they did not realize that the business time scale that were dealing with probably involved days or weeks. These guys are used to running Excel spreadsheets in minutes, and the fact that they had posed a problem to the computer and had to wait for 2 hours for an answer was a very difficult cultural thing for them to deal with. In general, business people are not trained in optimization, and they are usually very defensive when we bring those kinds of solutions to them. So it is really an educational problem that we have internally. Judith Hempel: Are they senior staff of long standing? David Smith: No. Neither were the people who were solving those problems for me. The young people in support positions in DuPont supplying that kind of information to business managers do not have the right kind of training and background to solve those problems. Solving the problem for them turns out to be an iterative process because they do not really understand how much information they have to give us so we can give them a good solution to their business problem.
Representative terms from entire chapter: