Interview with Jeffrey Winicour, Richard Isaacson & Nigel Bishop (contd.)

0:00 You know, code development and debugging, you know, wanting to involve in this production environment. Right. But we can't even debug the problems sufficiently well because computers aren't big enough to make access to them. Right. They're out there, but they're not yet widely available to people in the design community. In some sense, we're back where we were in, I don't know, 10 or 15 years ago when Larry Samar was stirring the country, crying out for the supercomputer family. We have big computers available, but if you define a supercomputer as the fastest, largest computer at any given time, those are not widely available to people in the academic community. They are by people who have contacts and get approval to run codes on things that are on the national basis. Wait, wait, it wasn't that time. Yes, I was just in Penn State and had a little bit of attention. So even though, for instance, Larry Smart is the director of the NCSA, it doesn't necessarily mean that... He doesn't have a computer to give anybody on this project that's equal to some of the things that people who do bond-testing. Salomos probably has his computers, but yeah. So I think Pablo and Richard Manson were well plugged in for some of us. But his code is not running for those computers. I would say the whole environment right now is pretty fragile, you know, to this field because you don't have too many people working, one thing is a personnel project. What's been able to be accomplished over the last four or five years is due to the fact that a lot of new personnel have been brought into the project due to the grand geology. No one really knows how that's going to evolve over the, you know, now the grand challenge is going to be a lot. It's not in a national sense, but in terms of the number of people who've been doing the work well before, you know, it's sort of like doubling the number of people in the field by giving the senior people and the students to postdocs. Sure.

2:30 And so they've had a tremendous impact. And the students and postdocs are the ones who do a little hard work. So you've got, you see, so personnel-wise, you've got all this, you know, details of what's been accomplished in the minds of a lot of students and postdocs around the country. And it's not clear exactly, you know, how that, you know, it's like throwing a few seeds of grass down on territory that you never grew grass on before. Right. You know, and the wind's blowing, and, you know, times are changing, and you don't know what's going to happen. It's been coming on for a long time. We're doing this in winter, really. The Grand Challenge came along at a time when everyone else has been pretty much cut, I think. And so, it created a temporary summer for us. You know, it's coming to an end now, and it's not clear at all how long-term, you know, that have been accomplished yet, have a long-term impact. What are the, the dramatic challenge, what's going to end next year? So what, what are the planning for continuing on with the planaries that go? You know, the proposals are trying to keep some of the manpower, the young people, you know, still active. But no one knows how, you know, what's going to happen to those young. Actually, I'm going to mention that it might happen now that there would be two sort of groups working on different approaches, which yourselves and Penn State and Texas would be continuing the sort of ADM, general ADM approach, and then that Cornell, Illinois, North Carolina would be pursuing hyperpolic. Those are the two main proposals to continue with that. One thing that he mentioned also, which is just what you were saying, and maybe you guys have some comments to make on this, is that he was worried that one problem for people doing a lot of American work is that given the amount of time that had to be invested in working in the code and developing and stabilizing the code, that it was difficult for them to produce as many papers as other theorists.

5:00 I think our group has been, it's done well in our respect. We've generated a lot of papers. We'll give you a supply of papers, if you care to talk with them, before you go. It's been a lot of physics. But I think Ken Wilson went around the country giving lots of talks pointing out that computation has a lot in common with experimental physics and some time scales and difficulties. the need for more massive groups to get some results and actually, those are all characters, those are experimental methods. Here we have theorists who begin to big science. It's like facilities, big budgets, lots of people, you know, multi- multi-university, what the ship, it looks like high energy and computer physics now, at least this is people who are hearing. So there's a lot of social engineering and I guess changes in sociological patterns that go along with working in this remote. It truly takes a while. It's taken every community a while. They all go from things they have a picture all over to go kicking and screaming, that's the way it comes in the future. I think that transition to general activity has been sharper probably than in other communities. General activity was better known for people being individual researchers, long rangers, than any other field you can imagine practically. I think what you're saying really has been exaggerated and all those problems have been exaggerated in the case of the Grand Challenge. You always have two people together in groups with different anxieties. you create a lot of anxieties brings it out that you never normally worry about you're not you can't see them in the other people maybe they're there but you don't see them and they don't come all into conflict

7:30 so the I couldn't possibly begin to give you any feel for the sociology of the grand challenge but you're In a sense, you're missing one of the more interesting areas by not going to the Grand Challenge meetings. If you had come there and just sat as a passive bystander through these meetings, you could have caught objectively what was going on would have made an interesting tale to see what was happening. Someone should have thought of that at the beginning, you don't know how it thinks about it. Let's sort of illustrate it. With our matching problem, a large part of it is outside our control because it depends upon, if you're talking about the stability of a boundary condition, It also depends very much on how the interior part of that code is written. So we have to interact very much with the people working on ID and cut. That brings in some interesting sociology. So it has to be worked on so that the codes are ultimately going to be compatible as you go along? Yeah, they have to be compatible, but it's at a more technical level than that as well. There are various different ways that the Koshi people could write things, and that very much affects certain aspects of that very much affect boundaries to it. If they're going to provide a boundary condition that will be useful, they have to know something about the details of how the code can take the data and what points on the grid it needs it on. I don't know how long these codes are and how many thousand lines. Okay. So it's a complicated code, you just can't poke in and understand it all. them to cooperate, at least to tell you what their needs are, but then if you say, well that's hard for us, can we give you this instead, you need somebody who understands it that doesn't

10:00 have a discussion. And it was all that structured in this at the beginning, because no one really knew how to do the out-of-boundary, or the problem, you know, we thought we knew, but other people thought they knew, and their ideas weren't the same as ours, so they said, okay, we're to do the out of boundary turbulent you know that's the nation you know there was a group centered about ncsa that we wanted to do it we wanted to do it but it's not you know but it's not a linear characteristic evolution so that so that there are always these you know these two approaches say how to do deal with that boundary bond at the same time there were there was the the same kind of contract going on with how you do the ADM evolution, the interior evolution. There was the group at North Carolina that came up with the idea that we'll solve all problems with instabilities if we do a hyperbolic formulation of equations. And then there's the traditionalist that says, what do you mean? We have to sit down and go through all the detail of doing all that for the next five years? we have a code array that we can you know that we can do it you know run adm code that's a there's a lot of confusing views what was the best way to see you know so on top on top of all these scientific issues i had didn't any one particular approach there's all these uncertainties about what approach is going to be taken, and the same with the software, people, they, it wasn't clear exactly what direction that was going to take at the beginning. So, you can see why I said that it would have been interesting to come to those meetings and see how that worked out. So, was it largely a faith meeting that issues of... Face-to-face? Yeah, I mean, given the communication between, say, groups working on it, and the inner region work versus characteristic work in the other region, given that obviously you had to communicate back and forth what you needed for each from the other, was that largely done during face-to-face meetings or...? No, there's face-to-face, there's email, telephone conferences.

12:30 So they all work briefly? Yeah. It's not only the problems with the hard thing about it, no matter how you did it, you know, You know, I try to communicate with her at meetings, you know, cut it out and say, on the phone, email. I think email, in a way, was the less fractious means of communication. Because people have a chance to think about what they're saying. You know, I'm going to think again about what the impact on the other person will be. You know, if you don't have that luxury when you talk face-to-face, you know. I think right now the network is self-supporting, the university is kicking money at him. But who's in charge of it? Who's in charge of it? Who's the budget president? There's no focus. Who's the I? That's a strange organization, isn't it? Well, at least you know who's the blame. It could be Bryce because of AOL. Microsoft. Well, I suppose I'll usually ask what kind of a time frame you see for solving the full It's hard to say, I mean, it could be without six months, you know, I'd say six months minimum. But after that, it's hard to say, you couldn't actually get, it's conceivable that we have results, in something like a time period of six months of indefinite matching one.

15:00 What we've been able to show is that if you can do the auto-boundary by matching, then you can also do the individual black hole inter-boundary by matching, too. We've got a characteristic O now that will run on an incoming when we do the auto-boundary But we could also match to a kosher evolution of the inner boundary with incoming inner bonds that will start outside the horizon where you do your matching and then come right inside the horizon so that you can do the inner boundary where you find a modular trapped surface. This is a standard technique for avoiding the singularity, the stop when you find a modular trapped surface. on the characteristic surface as well as you can on a Cauchy surface and the Cauchy evolution has been having a problem one of the problems is that the Cauchy evolution is going to go unstable when you get down near these apparent horizons whereas we we've shown now that we can run a characteristic code forever with an apparent horizon boundary condition I wish Louise my other brother's is here this is really sure the animations it is but they're on the web let's see them so we've been able to actually uh do uh you know run run things like black holes that are wobbling around and run them forever and these cones run as long as you want You know, we've been able to do black holes that have outgoing, you know, radiation on them. So the black hole is very, things that are very, very non-linear. And we can run these things safely for as long as we like, you know, while we're tracking and finding this apparent horizon. you know so all this all this is you know is a done deal now we can do this inside where the black holes are moving around completely stable we can do the outside completely stately the adm holds are i think are proven to be very fairly stable now so if we can do the matching then

17:30 we've got all the ingredients that we needed i think that this you know getting the matching accomplished is the missing leaf right now, and we'll be able to do a two black hole problem in a short amount of time if we can handle it naturally. There's also an issue, isn't it the DAN code, that it's difficult to run over a long period of time to evolve the system? that's that's an issue so we but a lot of it is a lot of that is in the context of strong fields we're trying to run with you know in the neighborhood of black holes near black holes or near horizons but see if we could if we could do that in a region forum and i think they could run long enough maybe we'd have to find out exactly how long but maybe at least do a few of a binary problem, which would be, you know, once you've done that, then you're on your way. You know, doing one or two orbits, it won't be long, you know, at that stage, you know, everyone had confidence that they could do 10 orbits within the next year, right? You know, you'd find ways to improve it. So once you have the pieces together, I think that progress would go fast. So the pieces are there. The latching is probably a huge. So you need the pieces and you need the hardware. Right. But that will count. I don't know how much I do, but we run these characteristic codes on what are basically workstations now, this RG2000 machine at NCSA, is something that Roberto's wife, who's a new faculty at Duquesne, just got as part of her, you know, what do you call it? Startup. Startup funds, yeah, she got it as part of her startup package, you know. as a you know basically as a workstation for office and i said we're running this thing basically on what nowadays is a workstation mobile machine and we're running these things for significant times you know and i'm much you know enough time to do a few wallets of a black hole or binary system so you know so i don't i don't see where the computing capability is that that far out of reach if you just want to do a rough problem like a show that you can do it

20:00 and I think once you show you can do it then the computers will fall into place if you could really show you how to code that can run two black hole problems and get away from the LIGO need it, you probably can convince someone in Los Alamos with this bomb computer you get your time We were just talking to Kit yesterday, and he really thinks one of the possible early sources for LIGO might be the black hole could have been 10 to 25 solar masses, and that will be right in the middle of the most sensitive part of the LIGO bandwidth, and if we had a few numerical simulations just to get order of magnitude estimates from the bandwidth and some rough spectral shape. That would help a whole lot in writing software to get, you know, another factor, 30% better resolution somehow. And, you know, that might double the event right at the store. And I think they provide more detail and then they can go back for five or ten better segments at once. Sure. So anything that this group can do between now and 2002 is going to feed right into the experiments. So it's a real challenge. It's a challenge for the experimentalists. It's a challenge in the theorists as well. About the matching problem, so I can kind of get a feel of it, is it fairly analogous to the type of matching techniques used in analytic radiation natural problems such as matching between the near field and the near zone and the far movement? It's that idea, the American version of that idea. I think it's more pedestrian. I think it's the textbook description of, you have a manifold, a coordinate that catches, it's something that can overlap, and you can make transformations from one to the other, and that's what's involved. Basically, in one approach, usually it's the perturbation of the calculations that are the complicated part of it, right?

22:30 Trying to get the solutions by some approximation. If you do it analytically, in this case, that part is just replaced by a finite difference. and so matching so the nice thing about a matching approach it's no harder from if you've done it for the linear system it's no how to do it with the nominee system because you just add a bunch of terms and and the terms you're adding on it's not the principal part of the equation about the second derivatives you know how it is in the nominee in terms of first products of first derivatives and they don't determine the stability properties of the system it's the second derivative terms that that it determines the stability properties of the system when you try to plan items so so at some level it's the same idea two coordinate patches for different domains that are you know devoted to different domains but it's uh it's a different problems right in one case you're worrying about convergence of the perturbation series and in the other case you're worrying about not so much the accuracy of the scheme because you can always make a numerical scheme accurate you know by finite differencing your property but you're worrying about the stability it doesn't blow up on you one of the year when I get out of here. That's the rule of thumb. given, as you say, matching is the main problem at the moment, or the key to the problem at the moment. How does that affect how you work? Does that mean you're sort of, at the moment, heavily dependent on the interactions with people working on both of the clothes? In a way yes, in a way no, because everyone still has their independent individual problems. I may probably not as a matching, but the ADM people have as their problems first time to get their codes stable.

25:00 Increase the stability properties of the code. The second thing to implement is SAG-H, which has not been fully implemented in the code yet. and so we're really you we're really trying to match to an older version of the ADM code and we'd like to there's the better versions of it now but it's hard to keep on changing your target you know you know they've updated and they have better versions maybe we do better with some of the versions now but on the other hand we can't ask those guys to stop trying to put it into DAG H and stabilize it so that we have a version that you know that we're sure we should be dealing with so it's there's all these timing questions the last grand challenge meeting they promised us they'd have bad dates ready in two weeks that age version of the code and that so i said okay we won't test you for two weeks you know that for you know for the latest version in order to try mashing with it but you know those time tables never never very predictable and they've been having a headache trying to finish our book get all the bugs out of the bag they have a new boundary have various amounts of successful But the code has problems with the inner and the outer, so maybe it runs well, but even that doesn't need to be tested for sure. That is the main purpose of integrating the back page code into the ADN code to parallelize I don't know how detailed you want to get into things, but the original ADM code was written as a leapfrog card this is the you know the first thing you would try is a standard numerical technique for evolving from one level to another so you just keep you have two levels and then you jump up to this level up to this level and you keep on going like that so that's

27:30 standard technique is it's a technique that has that's known to have trouble with boundaries the boundary conditions you've got you've better have dissipated boundary conditions have the boundaries act stable so the interior code might be unstable in the boundary the interior code might be stable actually for the boundary leads towards instability any of my I put on a boundary condition this station see that's the stage we're gonna latch one of these lead from codes which is which might be trying to do it with no likely the hand tag behind the back but in the meantime they've gone on and they've got what are called Frank Nicholson codes. I don't know how much you put in the area, but these are more dissipated codes, intrinsically through these product codes, not just at the boundary, but everywhere. They tend to be more stable. And this is what we're really waiting for right now to try matching to it but the current nicholson codes are are much more complicated than than the reply codes and we want to wait until it stabilizes before we go through the pain of writing you know writing all the doing all the programming so so in a way you know in a way we're we're stalled by the by the coaching people but on the other hand we have lots of things we can do You know, so that's, you know, you do the things that are productive to do while you're waiting. Sure. Do you want a table? Sure. Want to get one with me? These are all things you can catch on our left hand.

30:00 What this is, is this is a boosted microphone. So you just started moving across the grid. Let me, let me make a picture of what that actually did. This one is actually, yeah, yeah, it's quite true. So here's the first part right here. Okay? And it's not off of the initial slice. It's sliced like this, okay? The data, obviously, coming from it. And it's bigger. And it's quite true here. And it's quite true here. Okay? So in terms of this, usually the way of dealing with the Priscope picture, you can't deal with that because she's all disturbed with symmetry by these little Priscope spheres. So you're really coming out of the plane of this picture in some way, and you can see the whole book. So what we're doing is robbing this in some three dimensional sense.