Interview with Hideyuki Tagoshi

0:00 So, what are you going to ask me? Well, it won't be anything too outrageous. Let me quickly say that it's a quarter to three in the afternoon of the 27th of October, 1999, and I'm talking with Hideyuki Togashi. So, well, I mentioned earlier today one I'm sort of interested in at the moment, which is contemporary theoretical work, which is concerned with gravitational-related detection. And so, obviously, one of the things that I'm most interested in is data analysis, because many theorists work in data analysis in this field. And so I guess I wanted to start actually by asking how you got involved in the data analysis for Tanama 300. Okay. Well, it was not very straightforward that I became involved in this project. So, basically, this particular work began, I think, last year, and so, well, last year, Tama plans to take data this year, but they know that there is no one who analyzes data in Tama projects. So there's one person, Kanda-san, I think he was the only person who was really seriously trying to do the data analysis. But apparently one person is not enough.

2:30 And so Seiji Kawamura, you know, he was in Caltech before, Seiji Kawamura, so they're trying to do, trying to find people, persons who work on data. At that time, I was in Tokyo National Astronomical Observatory as a postdoc, but I didn't work I work on the data analysis at all, only working on purely theoretical, alcohol perturbation or post-Nitanian thing. But my position was over at March last year. So they talked to me, but I... You were living? Yeah. So they suggested that I would become the postdoc for the TAMA project. But I didn't accept. I choose to go abroad. So they tried to find other people, and Seiji Kawamura came here, came to Osaka, and gave a seminar, and he recruited.

5:00 So there are three possibilities, Tanaka or Masaru Shibata or I, but Shibata had already plan to go to America, the United States, so Tanaka became real. I don't know how much it is official or not, but so he decided he became born in TAMA data analysis But around that time, I was chosen to become assistant professor here. And so apparently one person is not enough, a number of people. So you... I decided, after I came here, I decided to work on data analysis. Did Tanaka ask you to join in there? Were you, are you attracted by the idea of working on data analysis or would you rather be doing... Yeah, I have been interested in data analysis, but I didn't imagine that I really evolved to analyze real data. So what has it been like handling real data? What has it been like handling or analyzing real data? Of course it's very interesting to see the real data and how much it is non-Gaussian.

7:30 Probably I wasn't willing to do the real data analysis. So I wanted to do the other thing. To do other things? To do other things. You were saying that you hadn't been expecting to be working so much with the real data but maybe, do you mean that you'd expected more to work on the source modeling end? So how does it work? I'm curious. In what way, in what form do you get the data from the detector? The data was recorded on DLT tape. It was a frame format, the common data format of interferometer detector around the world.

10:00 And that tape, DLT tape was sent to our office by mail, ordinary mail. And does it just come off the detector and it's then stored in this particular way in frames of some length? And then it's just sent to you like that. So what do you have to do to the day to make it? Do you have to do some... What do you have to do before you can actually start analyzing it? Well, so there is a special C language library called FrameLib, and this is a set of sublutines to read the data from that particular data format. And so we read using that subroutine, we read a particular channel of data. And do you typically use just one channel, the main detector output, or do you use one channel? Yes, so TAMA has basically eight channels, but now I only looked at only one channel, which is the main output of the detector response. And when you have extracted the data from that channel, from the frames, from the tape, you then have a file somewhere.

12:30 You then have your file, and do you look at it on the screen, or listen to it, or do you never bother to look at it visually at all? I didn't look at visually. I didn't listen to it by sound. You did listen to it? No, I didn't. Probably people working on the detector. Sure. But you just take the file and then you analyze it? So you mentioned how non-gaussian it is. What do you do initially when analyzing it to prepare What's the first thing you do once you've got a file? So you get a file which contains the main output, and about how long a stretch would you do, would you analyze at one point, at one time? How long a period of time? At once? So, basically the frame format, one file of frame format is about 1 minute and it's divided into 20 pieces of data. So each small piece is about 3.2 seconds, but since we want to analyze very small mass binary, the length of the template is 30 minutes or 40 minutes. So the data must be longer than that. We picked up the data about 20 seconds.

15:00 So we read each small piece of frame and just combine them to the 20 seconds, 200 seconds. What are the seconds in there? And... And we... So... We convert... It's a... It's an original data output. Data is just an integer. So we first convert it into a voltage. and using the transfer function, so we take the FFT of that data and we put it into the And we multiply them by the transfer function to convert them into the strain equivalent amplitude. So you take about 200 seconds at a time, and before you transform that, you go within frequency space. Convert them into the strain amplitude. And that's all done just on the raw data that you had in your first place. And you said that the templates that you would use would actually be as long as 30 minutes. Well, of course it depends on the mass, the frequency band, but if we take the lowest frequency cutoff, such as 100Hz,

17:30 and the minimum mass in the binary star to 0.3 mass. The length of the template is about 30 seconds. So 30 seconds would be about the longest template, is it? So about 30 seconds template for... So that's for 2.3 solar mass macho black ones. So anyway, now you have your 200 seconds of Fourier transform data, and what do you do then? Do you simply start comparing it with the templates at that point, or what's the So, you know, one problem is that in matched filtering, the filter has to be weighted by the inverse of the power spectrum density. So we have to estimate the power spectrum noise. And that's the thing. Maybe nobody knows the best way to estimate the power spectrum. because our spectrum changing with time right so so what do what what do you currently use country we we just take the average of our spectrum before the data

20:00 we want to analyze. So this is the data we want to analyze and we take the, we estimate the average of our spectrum before this data, around 15 minutes. So you have Fourier transform for 15 minutes of data before the bit that you're actually going to search through, and you just take the average spectral noise from those 15 minutes and take that to be your noise spectrum. So then you invert that and do you simply at that point plug in the stretch of data you want to analyze and your template and see what number you get? Okay. And presumably you actually use a bank of templates? So at the moment, how many templates are you using? Uh, so if, so if, since we are working, we are trying to do the matchdo filtering with two steps, so one step, first step, first step, we use only a small number of templates with the low threshold. That's the first step. We currently working using about 100, nearly 100. About 100. And the number of all templates which we need to analyze on one-step search

22:30 So there's an order of magnitude difference between the numbers. But what you actually do with this hierarchical method is that if you see a point that might be a possible detection at the first search, then you go down to a second grid of a hundred or so will the templates more finely spaced. And how many parameters do you have for your templates? How many? Just two parameters. The mass. The mass is of two bodies. And how much variation is there in those parameters? What mass range do you have? So it doesn't Minimum mass is about 0.3, 0.4, maximum is about 3, but the maximum mass is not very important. And the number of template doesn't depend on the maximum mass very much. Do you mean that you cover the parameter space for the smaller masses more finely? So your grid is smaller? For smaller masses. So the templates that you have, how do you derive them? So you presumably, including there, so you have equal mass binaries and also unequal mass binaries in that. And what waveforms do you use to generate the templates? You mean the post-international order ones? Yeah, basically. We use 2.5 post-Newtonian waveform, but basically with restricted post-Newtonian approximation, which means that amplitude contains only Newtonian approximation part,

25:00 But the phase evolution contains up to the 2.5 post-Nitonian order. And do you use the form that the form was, say, from Balanchet-Dumor and Ayer group? Balanchet-Dumor, Will-Weisman. Yeah, the whole five. That's right. Okay. So you just use the 2.5 phase evolution? And so using that and your Newtonian order amplitude, you just generate a series of waveforms? Yes, but we directly generate waveforms in frequency domain using stationary phase. So you actually just get the, oh, you use a stationary phase approximation, and then you directly derive a spectrum of each, of each possible wave. Okay. The, and how, so you mentioned that the longest waveforms that you have would be about 30 seconds or so. And what, roughly speaking, what period does that cover in the in-spiral of the sources? I mean by that, how close did they get? Is it still just the in-spiral phase? For smaller mass, yes. It's before the merging phase, just in spiral phase. So you're not, are you in the length of the way of the templates that you're using right now, I mean, are you at all close to running into a point at which, for instance, the 2.5 post-natonian approximation might be somewhat inaccurate or is it kind of comfortably within it you mean that good post-natonian approximation is good it's good for the whole length of the template or is it getting towards a a dangerous region at the end. Yeah. So, for, but probably for three solar mass binary, so maximum frequency is already in the,

27:30 already less than one kilohertz so so maybe post-intentional approximation is not good enough for such binary but we cut for such binary we cut off the high frequency part. At some frequency, it corresponds roughly a bit larger than the last circular cerebral orbit. So in the case of the more extreme binaries, you actually determined the cut-off point of the inspire, by the effectiveness of the theoretical. It is just a priori guess about A to N. So it's one of the most difficult It's a great color. Yeah. Number. Right. It's in judgements. Yeah. Sure. But in the case of the other templates, you actually just go to the frequency that you think is the highest that the detector can see? Well, so highest frequency, I mean, maximum frequency is determined. So highest frequency for that detector is about 1 kHz. 10 kHz. Sorry, the highest frequency for a small... Because the sampling rate of the detector is

30:00 20 kHz. So the highest frequency for that is 10 kHz. But the applicability of the transfer function It doesn't cover such high frequency, so we can only use maybe less than 5 kHz, but there's another discussion about this is that if we lower the maximum frequency too much, we will lose signal to noise. the output of the match filter decreases because of the lower in the maximum frequency. So we choose 2500 Hz as a maximum frequency. And this number is determined so that we don't lose very much signal to noise, of the signal. And they are still covered by the transfer function. You mentioned that you calculate the spectrum rather than the way for itself. Of the temperate. Of the temperate, yeah, sorry. I was so maybe this I was curious if when calculating the waveform and you have some kind of a cut off if you just made the cut off very sharp whether you attempted to smooth the end point because I guess unless I presume you don't try to put in any model of the merger So you just let the waveform go to a certain point and then just cut it off sharply there in the spectrum?

32:30 You don't try to smooth the end? I don't use any smoothing at the end point. It's pretty safe. You don't get any, I'm not here, I'm just not too sure, when you calculate the spectrum, you don't get any extra high frequency stuff because effectively there's a sharp end to the, there's probably some extra high frequencies. I was just wondering if when you have your waveform, coming in like that and then you just end it kind of suddenly here if that means that when you calculate the spectrum as you do directly there will be some extra especially high frequency noise but some extra signal from this sort of high frequency jump there, but that's not, there's nothing very important that... So what we did is, we cut the high frequency in frequency. So you just cut everything off sharp in the spectrum itself? In the spectrum, yes. Since you don't have to do that. So the spectrum is just cut off at a certain high frequency, and is there a low frequency as well? So about what is the frequency range? So from 100 to 2500 Hz. So, why did you pick on 2.5 post-Newtonium as that was basically the best available since the 3 plan hasn't been really published yet?

35:00 Ultimately, do you think that it would be necessary to have more precise templates than...? So, for...? I mean, in principle, do you think you have a chance of seeing any binaries at the current level of precision of the templates? Well, so when the detector developed, I think we need 3pn template, but for now we don't need it. Yeah, sure. You're not expecting to see anything yet. but you think you probably will need 3pm and for the type of sources that you're looking for these macho binaries with the relatively small mass you don't think you'll need numerical modeling of the merger phase or anything like that we don't have just the inspiral of all of you So you have your bank of templates and you have them for your transform noise and so on, and you basically then have a large, you have a code that, I was wondering how much you automate the process. Do you take a given stretch of data and run your entire bank of templates on it and you have a code written up to do that? Yes, basically yes, but for the two-step search, the decision of the stretch of data the first step is important. And that is not determined automatically at the moment.

37:30 So at the moment... At the moment, so... Well, at the moment, uh... Well... Uh... So, basically, it's a process to optimize the computation time. So, threshold and the spacing at the first step is determined simultaneously. So we put some signals into the data, fake signal, and try to do some simulation to detect to determine the first step threshold. The first step threshold is determined so that we don't lose most of the signal, which can be detected at the second step. So you actually work out your threshold for the first step, especially by putting in some fake signals and seeing if you can find them. And by changing the spacing, we look for the first step threshold, which by which we don't lose any signal, which can be detected at the second step.

40:00 So once you've determined the threshold using that system, you then just run the data without anything put into it and try and see what you can see. Yes, that's right. So how many hits would you get? Of course it depends on how much data you want to keep, you want to record. Basically we can record all the output, but it's rather meaningless. So we decide some number of signal-to-noise above which we record. We write the event in the file. So what signal-to-noise do you normally use as your threshold? We try to record all the events more than signal-to-noise is equal to about 5 or 4 or 5. So in, say, a 200-second stretch of data, how many events would you find greater than that? It very much depends on the data. What would be the most that you would see?

42:30 I don't know. Well, I mean, I'm just sort of wondering, is it possible that you'd see, like, a dozen hits or a dozen events in a given stretch of data, or...? Yes, yes, but the one point is that once at some mass and at some time, we... event exceed the threshold, and so as a parameter, mass parameter, at that point, so they also exceed the threshold. Right. So if one, with a lot of these events which are caused by some big peak in the noise, you'll set up a whole load of templates all at once so I mean counting all that as one event how many but basically anyway there are quite a lot of very noisy of sort of burst of noise that you were just discussing at the workshop there are quite a lot of these bursts of noise that actually will trigger a whole bank of And, um, I was just curious if, um, well, first of all, at this point, at this point, how do you just sort of discriminate those bursts? I guess right now you presume that they're not, uh, that they're not gravitational waves anyway. So we use the chi-square test as well as simple match filter. So chi-square test is designed so that it can discriminate the time frequency profile So by using the Chi-Square test, we can reject more than 99% of the event.

45:00 So the test, you just decide if it looks anything like a chirp? So, by just sort of running the chi-square test, you don't again even have to look at the data any further than that. That'll just get rid of most of the false. That'll get rid of, like, 99% of the false positives. So what about the remaining ones? Well, I don't know. At the moment you just leave them as they are. Yeah. But such event can be caused by the purely Gaussian noise. So I think they are due to noise. So there's some, every so often there's just going to be a little Gaussian clip or, you know, a Gaussian thing. Yeah. So what happens with the data that then, after you've done all that, is that with, say, this data from this eight-hour run that you had, do you then, do you at some point after you've done all that, send it back to the experimenters? I was wondering who do you show the results to once you've recorded a bunch of events and gotten rid of 99% of them using a chi-square test. If I understand it correctly, at that point you have a list of events. which very likely are just Gaussian noise. And is that in some sense the final output of the signal analysis?

47:30 Is that sort of the end of what you do with the data? Well, I don't get the question. Well, I just mean... Well, okay, so let me try and add two points. So when you've done the analysis, or when you've finished with a given stretch of data, analyzing a given stretch of data, what do you have left? What's the output, as it were, of your analysis? Is it a record of some events? Well, it's a list of events. A list of events. A list of events. Right, yeah. And who do you then show your list of events to? I don't know. You don't send it off anywhere? It just ends there and nobody's particularly interested in seeing it at this point? Of course. I showed them to the other people in the TAMA project. Right. So, meeting Professor Kanda, for instance. Yes, of course. And the experimenters in general. And I was sort of curious, because I tried talking to Professor Kandon, but I didn't talk to most of the experimenters. when they look at the results of the data analysis do they react as if it tells them something about the detector and the way it's working or are they mostly just interested in seeing how the data analysis is going at this point Well, I think that the people in the detector is interested in the sources of the Nonggao

50:00 Yeah, they want to look for the source, the fake false events that are caused. Especially these non-Gaussian births. Non-Gaussian births. so in some sense instead of throwing them out those are the interesting events from their point of view but are they interested in seeing those events from your data then or I mean when you you know you mentioned that you used the cot square test to eliminate those events are the experimental people interested in seeing those events or looking at the largest non-Gaussian bursts or do they just, you know, they're sort of working on the machine and the instrument and they don't need to see? Well, so probably people in the detector and data analysts should get together and and talk to each other and discuss how we are working on looking for. So it's possible that they might be interested in having you, as it were, find the fake burst, the non-Gaussian burst as signals but at the moment you haven't been doing that at the moment you haven't really been looking for the non-Gaussian bursts or trying to classify them I mean you've mostly been throwing them and so I was curious in fact do the do the other people well Tanaka at the moment is in Spain The people in Tokyo, like Professor Kanda, who are working also on data analysis and data acquisition, do you meet them regularly?

52:30 Do you meet the other data analysis people regularly? Well, I think we meet. We have a meeting once in two months in Tokyo. But we don't meet very often. Sure. And do you meet the other experimenters then, or is it just the data analysis? Just the data analysis. So I was curious, you mentioned that you have a number of events left over that may be the results of non-Gaussian noise. is would there be any way I mean supposing there was a signal a real signal in this data would there be any way to tell it from the remaining events that you have if one of the remaining events that you have when you've finished after the chi-square you have some remaining events and if one of those was really a gravitational wave Would it be possible to tell at this point? Would there be any way that one could... Well, so... So if we brought the event... If we brought the event on the histogram, So we, in any case, we probably have such smooth shape of events, number of events, histogram. So if a real event is within that smooth distribution, we don't have any way to tell this is a real event. But if there is an event which is very far away from that distribution, very large signal to noise, and still they satisfy the chi-square test.

55:00 That would be a candidate. Yeah, that's a very good candidate. So do you already do that when you get your final list of events, you already plot your histogram? Yes, we plot the histogram, but so far, nothing. There's nothing. Well, that's good, I guess. But from that platform, I think that if there is an event about signal to noise rate is 10, then I think we can say something. So at the moment you are only searching for macho binaries? Macho or any compact binary. Any compact binary, but, right, sure. Within that... Macho. Macho, yeah. And the particular reason for thinking it might be Macho is that It's possible that they're closer than, yeah, that there are some close by. Are you planning to do any other types of searches? Uh, yeah, probably. Like for pulsars, for instance, or what? I don't think I will do the pulsar search, because I believe that someone will not detect any pulsar. I want to look for some burst source by correlating other information, like the Gamma-re burst.

57:30 Okay, so you might take records of gamma ray burst events or something similar and look for coincidences. Are there any other possible types of data or detector that you might use in coincidence searches besides gamma-rebirths, like, for instance, neutrino detectors or gravitational wave bar detectors? We don't have any particular problem yet. That's a possibility. Yeah, the coincidence between the neutrino detector will also be very important. Right, sure, looking for supernovas. Supernova. So at the moment you've already started on match-filtered searches for black hole biaries, and you're thinking in terms of maybe doing coincident searches for perverse sources. Are there any other types of search that you think are worthwhile doing with TAMA? Or would that probably be it, just those? Pardon? Are they the only types of searches you think would be useful with TAMA? I don't know, but... My understanding from the workshop is that after next year, that TAMA is applying for money to run the experiment for two years starting next year. Yeah. If it does get the money to run for that time, do you think you'll still remain doing data analysis? from the time it runs or would you rather do something else well I don't know but so for the next one year I can do the data analysis right but after the

1:00:00 next September. I don't know. But would you mind, has it been fun doing the death analysis? Have you enjoyed it? Yes, yes. But you know that Japanese people so express in different way. what they experienced. It's very contrary to the European or American people. So Bruce Allen said that for the last one month after TAMA data came, I had very terrible days working on a ferry for a long time for each day. Yeah, working hard, yes. Yeah. And it was very terrible days. But Bruce Allen said that I enjoyed analyzing real data very much, but I didn't mean to say that I didn't enjoy the look at the real data. you did enjoy that in some sense in some sense we don't say that we enjoy analyzing data so Bruce Allen was saying that he enjoyed analyzing real data but that's the wrong word to use for you but I suppose a more practical question is would you do it again

1:02:30 would you would you repeat the experience and would you continue working on real data I think so for a while I don't know how long it will continue. I guess I was going to ask the last question on data analysis is what part of it involves the most work? Is it writing code beforehand or is it getting the data and sort of preparing it and getting it into the right form? I think. Or is it actually running and working on them? I think the most difficult is coding and debugging. As usual. Right. Great. Well, most of your work previous to doing this data analysis has been on the radiation reaction problem, I think. In some way. Well, emission of gravitation, generation of gravitation, radiation. Yeah, yeah. That's right, yeah. And a lot of your work has been with perturbation analysis, which is relevant to space-based detectors especially. Do you see yourself continuing with that line of work? Do you think that there's a good chance that there will eventually be gravitational wave detectors in space, for instance, that could see signals from subsistence like stars falling into supermassive black holes? Do you think that's likely to be an important... In the future? In the future, yeah. I think so.

1:05:00 Yeah. The... So... So do you plan to work more on analysis of systems like that? Do you think you'll get back more into that? Well, yeah, someday. Yeah, but the problem is that the problem becomes difficult more and more, and it takes more time than before. Sure, it's difficult. I was going to say you've already done all the easy stuff, but in fact you've done the That's something in the future, I'd like to come back to this field, especially the a program of radiation reaction in a car space time. Do you think probably using a radiation reaction force type of approach? I don't know. Whatever works. I was curious, since you've done calculations to very high post-Newtonian order in some of this work, I was curious about how you worked doing this very complicated analytic work. Do you use Mathematica or one of those other algebraic software systems? Mathematica and Maple. And how do you use them? Do you do some calculations by hand and then check them with Mathematica or do you do them first with Mathematica or what way do you?

1:07:30 Well, basically, except the very basic equation, we worked on mathematics from the beginning. So, basically, except deriving the basic formula of the Tchukorsky equation for basic formula It's impossible to work on them by mathematical, and so basic equation of the solution of Tchaikovsky The homogeneous solution is derived mostly by hand, but it turns out that most of the The derivation of homogeneous equations can be done by some mathematical programming. Most of them are automatically automatically. So you write, you create a program to tell it how to do a given stretch of calculation, and then you let it run and do that. And about checking the formula, checking the result, so what we have done is to calculate

1:10:00 the same thing by more than two people. Right. So that's it. So they, and those two people have their different codes that they've written, their different programs, and do they use the same analysis software, or did they... Usually, yeah. Somebody was mentioning to me that they had one person use Mathematica and the other person used Maple as a check. Well, I think in most cases, to use Mathematica, only Mathematica or Maple is safe. But sometimes there is a very trivial, not the bug, exactly the bug, but a very trivial character of a particular software. So we must be careful of such a special character of each software. Right. So you have to be familiar enough with the way that the given software package behaves to know what the pitfalls might be, what little quirks that it has. Yes. And those type of behaviors, those quirks that the software has, is that usually something that you just learn about from using it? Yes, basically yes. Right, so people in the group will have sort of a number of these that they've noticed but it's not like you've heard about them from outside probably. It's just what you've noticed in your own experience. So, in fact, when you're working on one of these big papers where you do calculations out to, I guess, 8th order post-netonium in some cases, or 8th order in V over C. Would

1:12:30 that mean that actually most of the time or the majority of the time you'd actually be working on the computer with Mathematica rather than with pencil and paper or pen and paper? I'm not sure. So initially, when I began to work on this, I was not familiar with Mathematica or software. So I had to take a lot of time and just purely programming. Just learning, just getting the skills. But once we learned how to use it, such time it becomes very... It's actually quite quick. Yeah, quite quick. programming so so apart from this learning the skills using Mathematica there's still actually in a given calculation there's still much more time spent learning how to do the calculation as it were there's still more time spent on a given calculation working with pen and paper to figure out to do the calculation. So it's a matter of still, you know, figuring out how to go through the calculation and then you then just do certain parts of it. I'm not explaining myself So I'm wondering if, considering that Mathematica is used for some parts of the calculation, if you end up spending a lot of time sitting in front of the computer, or if, leaving aside as you say, getting used to using Mathematica,

1:15:00 or if in fact you still mostly spend most of your time with pen and paper but you actually work out the whole calculation on pen and paper and then you leave certain details blank because you haven't worked through a certain tedious stretch of calculation and then you just go and do that bit on the computer but you've actually worked out most of the calculation before you even get to using Mathematica. Is that sort of the way it works, or, for you? Well, in my case, I think, in that case, I try to do most of it on Mathematica. Okay. So you might actually sit down at the computer and use Mathematica to sort of explore the problem and see how you can go working your way through the problem. So you mentioned that at a certain point you had to get familiar with using Mathematica, so when you started off, did you mostly, you were usually just using pen and paper, just writing everything out? I think so, yeah. As a graduate student, for instance? So you did all the calculations by hand, except for... For some long series expansion. Right, yeah. So you did it all in my hand. But then later on, you got used to Mathematica, and so you found that you used that more and more. That definitely makes things a lot easier. It's nice not to have to do those really tedious trenches.

1:17:30 So, when you were an undergraduate at Kyoto University, and then you stayed on there as a graduate student, and what attracted you to working on gravitational waves? Well, so why did I start to work on the gravitation wave? Well, I think it sounds like very interesting. I think that's the only reason. Well actually, my first paper is about nothing to do with gravitational waves. Was it an astrophysics paper? on gravitational arranging of KUESA. But... I think... And I wrote the master thesis on gravitational arranging. So after that I can choose, I could choose my topic, or some other, something else. So at that time there are some number of people in Kyoto who began to work on a gravitational wave.

1:20:00 Working with Professor Nakamura? Nakamura and Misawa Sasaki was also there. And Shibata Tanaka. Yeah, so there was that big group. So, and I think the influence of Nakamura was very important. Because he was interested in gravitational waves. Yes. So I began, he has a seminar every week on Friday afternoon. So I began to attend that meeting. And a lot of these seminars were on gravitational waves or that was a topic that was discussed. And was part of the attraction or do you think part of the reason for the interest in gravitational waves with this big group was part of the reason, the fact that gravitational wave detectors were just beginning, like LIGO, were just beginning to be built or do you think that the detectors played any role in that or was it really just for theoretical reasons that it was interesting? Well, I think that, you mean that, why did they begin to work on gravitational wave? Yes, I mean, well, either was it something that you were aware of, well, let's say, Were you aware of LIGO and these other detectors as being something important? I mean at the seminars that might take place, if gravitational waves were mentioned, would LIGO or some similar projects be mentioned as well? Yeah, so at that time, LIGO and the Japanese small project already began. So already the Japanese project was there?

1:22:30 Yeah. And so that was probably a choice? That was something you were aware of anyway? Yeah. And Nakamura was already then maybe the PI? Nakamura was the PI of the initial small project. So the group was somehow closely connected to the idea of the detector project. and because I asked because so I started with Kip's group at Caltech about the same time and when we I'm one of the very first meeting group meetings so he would have a group meeting similar to that which everyone would be there and one of the very first ones that I was at And Kip came in and he had typed out a number of sheets of paper, and quite a few, and he had worked out a whole load of problems that he thought needed to be worked on from the point of view of making LIGO and solar detectors work. and he wanted the whole group to work on this if they wanted to some people were finishing and had done other things and so right from that moment on the group was really working on the problem of detecting gravitation so this paper the last three minutes was written as a result of that So I was curious to know if your experience starting at the moment... Well, I was just explaining that in the case of being in KIPPS group that we were... I was really made aware right at the start. Oh, you know, we really ought to work on this stuff because it's going to be very important for LIGO and that kind of thing. So I was wondering if you had a similar impression when you were starting off as a graduate student. Yes. But actually, our work was very much motivated by the last three minutes paper.

1:25:00 So KIPP had a big influence there too. So when the first three minutes paper came out, that had quite a big impact. Yeah. I mean, last three minutes paper. That's right. First three minutes, something completely different. So, for instance, did the last few minutes encourage you in that type of work? Yes. So, in that paper and in the next paper, maybe you involved one? I don't know. They show some. And in America, the analytic work of the post-Newtonian expansion was done by a person, and the medical work was done by about four people. That's right. Yeah, I came in a little later on that particular work. So, the results were written on the last three minutes paper. So, because Nakamura has worked on Tchaikovsky or Lechivira homorism, They were very familiar with me. And he suggested me to work on that particular problem in a different way. So, although you were influenced by the Last Three Minutes paper and by Eric's paper, and the Ducalski form was in the Schwarzschild's circle orbit, did you just use those papers as a guide? And you didn't, and you had the experience of Sasaki and Nakamura to go on, you didn't contact any of the people outside.

1:27:30 Because I seem to remember when I, so after Eric did that paper, and there was that second paper with Sussman, Cutler, and Plussman, and the other four persons, I started working with Eric and others for the work. And I remember when your group at Kyoto came out on eccentric orbits in Georgia. I seem to remember that, well I certainly didn't, I don't think Eric either knew that you guys were working on that. so you just sort of took it took those papers and and you already have the experience of Sasaki and Epimora and just sort of and worked from those when you were when you got when you started working on these poems you didn't I mean to say that it was just enough, you know, to read the paper and then decide, well, let's work on that. Without any... Well, I... I mean, I know that Eric, for instance, visited, but that wasn't until much later. Yes. So, what question is... Well, I guess the question is that... Well, about the eccentric COVID paper, I think, as far as I remember that Takashi Nakamura visited a conference somewhere, maybe in the United States, And he met Kurt Cutler and talked about that problem.

1:30:00 And after he came back to Japan, he suggested to work on the eccentric orbit. I was interested if there was any And also Misao Sasaki visited one of Israel's 60-year-old memorial conference in Canada. And he listened to Kip's talk and at that conference Kip said that Boston Junior expansion doesn't converge at all. And after that, that's the last three minutes paper game. Well, probably, eccentric qualities between this and last three minutes paper. Did you work mostly from when you began your work on particles in Schwarzschild, did you work mostly from Eric's paper or did you go back to Sasaki Nakamura or to earlier papers when you were sort of beginning?

1:32:30 Well, I think I both saw a person's paper for a couple of person. You, you, you? Do you have a paper on a numerical list square fitting method to determine the question of the post-junior expansion? Oh, I know. I know I wasn't on that. And also on the old paper by Nakamura Sasaki. Yeah, and my first paper on the Boston Union expansion of the core perturbation theory is mainly because of the That's four people's, four or five people's paper. The eccentric? No, not the eccentric orbit. Ah, yes. This paper.

1:35:00 Okay, so you actually mostly used that one as a guide, but... I took the paper. I used that too. The, well, you guys, I must say you guys worked really fast, because you waited until after the last three minutes, the paper was published, I guess, and it took quite a while for that paper to get into print, because, really, and, but you really, your papers were not that long, well, not much longer later. You work very quickly. We were impressed. Yes, you did very good work. Yes, it was very impressive. Well, thank you very much. I guess we'll leave it there and we can turn the machine off. You're welcome.