Interview with Miguel Alcubierre / Alberto Vecchio

0:00 I have to be careful to make sure everything's going, and it is. Now that I'm used to it, it's actually quite easy to tell if it's working properly, but I was always worried that I wouldn't see the thing turning like a tape recorder. Anyway, today is the 11th of February, and it's 10.30 in the morning, and I'm talking with Miguel Alcubierre. Or Alcubierre. Alcubierre. Alcubierre. Alcubierre. So, well, let me think, I guess you were a graduate student at Cardiff? Yes, I was. And this is your first position since then and did you come directly? I had a postdoc at Cardiff also. You had a postdoc at Cardiff. Then I came here. So in Cardiff where you're working here, you're with Ed Seidel's group. And as we were talking the other day, you're working on numerical relativity, looking at railways and formation. Railway collapse, yeah. And is that similar to what you've been working on in Cardiff? It's related, but in Cardiff I was, during my PhD I concentrated mainly on numerical methods more than any physical application. so I didn't actually do any relativity as such it was just numerical methods for the wave equation and for hyperbolic equations implicit methods, stability analysis things like that and then also when I was a postdoc I was working in collaboration with Gabrielle Allen and we wrote actually a three-dimensional code a relativity three-dimensional code code, which we never actually used for anything, because it was unstable. Now, after working here for a while, I've discovered that maybe we should probably dig out that code and use it, because all the codes people use are unstable, so ours wasn't an exception, but we didn't know this before. People always have this instability, they just hope they can do some basics before the instability kicks in. So, maybe our code would work, but that's what I was doing kind of, I mean, it was, I mean, the code could run in parallel, so it was a pretty interesting code, but it was different from the main approach of everyone else, which is the one we're using now, because it was using the second-order system. It used the Einstein equations written in terms of the metric only, and not the metric and the instincts in curvature as it's done normally. So and the methods that you're using now then?

2:30 We're now using both the standard methods and the Bonner-Massell formalism, but both of them are first-order formalisms in which you have the metric and the extrinsic curvature as independent variables, and that gives you equations that are first-order in time. They can still be second-order in space in the standard formulation, but first-order in time. So, did the formulas of yours and previously have any disadvantages or advantages? Well, we thought that the approach had disadvantages because it wasn't stable. Now I'm not sure anymore. It might just be as good as any other thing. But we haven't tried it again, we haven't dug up the old code, because we came here and we started working with the codes they had here, specifically with the Cactus code. So you were saying that the work that you did while you were in Cardiff as a general rule was more just numerical methods and it wasn't on to the stage sort of when you were applying physics to it. We looked at a lot of numerical methods, specifically implicit methods that we're not using now, but we always have them in mind. Our code used them. The code that Gabriella and I wrote was based on implicit methods. Those are generally more stable than explicit methods, but they require a lot of work. implicit methods now meaning implicit numerical methods are basically divided into explicit and implicit an explicit method is the one in which you have you can solve explicitly for the value of a given function in the next unknown time level using just values in the old time level so that's explicit an implicit method you cannot do that all you have because of the way you approximated your differential equation you end up with a matrix equation on the new time level that you have to then you have to invert a matrix That's why you call them implicit. You only have an implicit expression for the new values in terms of the old, but you have still to invert a matrix. So it's generally more difficult to solve, but usually they're more stable. So as you evolve the system from one time splat to the next one? From time splat to the next one, yeah. Okay, I see. So one of the virtues of the implicit method would hopefully be stability, but... Yes, at the moment nobody's using them here, because the cactus has no provision. implicit methods. Implicit methods are difficult to apply even in the linear case. In the nonlinear case you can't even use them because you have to solve something that's not a matrix equation, it's a nonlinear equation in the next time level, so it's just terribly complicated. But you can get around this thing by just

5:00 using implicit methods in the... I'm sorry. Yeah, seems to be how it works, okay. So as I was saying, implicit methods for nonlinear equations, you can get away with them if you just use them in the principal part, which is quasi-linear, at least for Einstein's equations. So you can still use them, but they're not being used now. There's nothing I was going to say in response to that, but I think... I also did some work in Cardiff and gauge conditions. I was also interested in looking at that, and I discovered a phenomena that I called initially Coordinate Shocks, though I don't like the name anymore, but it seems to have stuck, and it's related with the gauge conditions used in the new hyperbolic formulations. The new hyperbolic formulations need a specific family of gauge conditions in order to make the whole system hyperbolic, and what I discovered is that that family of gauge conditions develops pathologists that look very much like shocks in the sense of the metric functions. You have a wavefront, something that looks like a wave moving and it develops as continuities, like the wave wants to break. And when that happens, then you're done for because your coordinate system doesn't work anymore, it's singular. That sounds interesting. But you say you don't like the name anymore? I don't like the name because the word shock in hydrodynamics means something that you can actually integrate past the shock infinity so there's many methods that allow you to integrate past this thing and you can even talk of shock waves as just as a shock moving. In this case you can't because as soon as this discontinuity appears then your coordinate system went wrong and there's nothing you can do anymore because your coordinates are just crazy. So you cannot integrate past this with this coordinate system. So in that sense it's a deceiving name. But it's nice in the sense that at least it looks very much like a shock is about to happen in the quantities that you're evolving. It looks very much like a shock until the discontinuity appears. So you think you actually have a shock wave, I mean, something that turns out it's only a coordinate effect. And this is a particular characteristic of these hyperbolic? It's a characteristic not of hyperbolic formulations, but of the gauge conditions they use. Okay. So even if you don't use a hyperbolic formulation, if you use that same gauge condition, you'll have the same problem.

7:30 So what type of gauge conditions are you used? Those gauge conditions are one, it's a whole family, but it's a family that generalizes the harmonic time slicing. So these are similar to harmonic gauges? Yes, basically just the harmonic time slicing. We don't worry normally about harmonic spatial conditions, just the time conditions. So it's that one plus a generalization of that one. Most hyperbolic formulations require either harmonic or something similar. to work. It's interesting because I know, first of all, I know nothing about, I've no experience myself with numerical methods in the real sense of the term, and the occasion I solve an equation on the computer and that's it. So it's interesting to find out, because this one particularly interests me because historically, when people first used at the very beginning of the study of gravitational waves people started using harmonic gauges right away they discovered that they would see these waves that it turns out were really just flat space but they thought they were waves because of the coordinate system so you have to be very careful with the gauge we have a lot of experience here so we still worry sometimes we have a good feeling for when things are just gauged so that would be one of the coordinate effects from... Yes, it's a very big problem. And the instabilities you spoke of, again, how... is there some particular type of problem in numerics that tends to make these arise, or do they come out in all situations? There's many different instabilities that we normally encounter. There are the typical numerical instabilities, which the most typical one is known as a violation of the Courant condition. The Courant condition I mean, the speed at which information propagates in your numerical grid should not be larger than the speed of the waves, in this case, the speed of light, the real physical speed. If it's larger, then things will go unstable numerically. And typically, you see that. But that one we know about and we're aware of it, so we take care not to violate that one. There's other numerical instability that we might not know the origin of that might pop out. And there's also instability that is more of a differential system of equations.

10:00 The Einstein system of equations is pretty complicated. So you have six evolution equations, and then you have four constraint equations. Now, in the exact differential case, you can prove easily, using the Bianchi identities, that if the constraint equations are satisfied in the initial time slice, then they will be satisfied forever. But that's only in the differential case. As soon as you do a numerical approximation, that's no longer true. Even if your constraints are exactly satisfied initially, they will be typically violated soon afterwards during the numerical evolution. And that violation can actually be unstable. It can easily turn out that you diverge exponentially from the surface of the constraints. And that kills you eventually numerically. And we don't know what to do about it. We don't even know how to describe this phenomenon very well. Nobody studied this very much in detail. but I think that's something that's happening now typically if that is what's happening then if you refine your grid if you have more resolution then the effect will be smaller and you will be able to run for longer which is what we do most of the time so that's we have this very brute force approach at the moment if it doesn't work and your code crashes double the resolution and then it will go on for longer that typically works but in three dimensional computations because that memory requirements is just tremendous. So a great deal of the instabilities that arise are directly linked to the resolution that you have? Improving the resolution always makes things better. In the case of the current condition that you mentioned, do you, looking for something that's moving faster than light, Do you simply look for anything that's moving faster than light at the moment in the coordinate conditions that you're using? No, it's simpler than that. I mean, you can have... You just look at the grid spacing, the delta X and the time step, the delta T. And the grid's velocity is defined as just the delta X over delta T. It's the grid spacing over your time step. And that gives you the speed at which things propagate numerically just because of the way you build your computational molecules. see that a point that's very far away in the numerical grid cannot be influenced by something close by just because of the way you build your molecules. So that's basic quantity,

12:30 and that quantity has to be small, smaller than the speed of propagation of the waves. So typically you want delta T over the, sorry, delta X over delta T to be smaller than the speed of light. And there's a factor that comes in depending on the number of dimensions. It's always the speed of light times one over the square root of the number of dimensions. So in three dimensions, it's 1 over the square root of 3, which is 0.57 or something. So you want delta t over delta x to be 0.5 or less. And that's typically what we do. Of course, things are complicated because the speed of light depends on the core of the system. So you have the lapse function and the metric, and even if things are working nice initially, then it might turn out that all the dynamics of the geometry will make this condition violated, to be violated later, right? So you have to look for that and be careful. So basically you're looking for anything that's propagating across the grid too quickly, any of the numerical tests. Yeah, but you don't look at it like that. You run it knowing that you satisfy the condition initially, and then look for... These sort of instabilities have a very typical signature. Suddenly something starts, you have a very short wavelength noise that just grows exponentially without moving. You suddenly see a very short wavelength wave that's blowing up exponentially, but it doesn't move, just stays there. That's a typical current instability file. So when you see something like that, then you know that you violated the condition somewhere. And you better take a smaller time step. So for the most part, then, the kind of instabilities that you're dealing with have certain signatures that tend to show up in the output that you recognize from experience. And then you have certain steps that, as a first approximation, you immediately take to try and do it. So this one, the current violation instability, is one we know how to solve. like that is happening, we take a smaller time step and typically cure, so... The other ones, the ones related with constraint violation, we don't know what to do about them except just throwing in more resolution. And eventually we could just push them down, but in 3D it's turning out to be very difficult. And the solution for the current violation is in some sense a kind of a resolution thing as well in the sense of decreasing the time step. Yeah, but in that case it's resolution in time only. And if you decrease the time step with the same grid spacing, then it will be cured. is more like decreasing everything, the time step and the grid spacing, and keeping the ratio constant. So you restrict everything, and in the current case,

15:00 you just restrict the distance between your slides. So those differences. So is it fair to say then that, by and large, the instabilities, the problematic instabilities that you encounter tend to be connected with the resolution invariably that solving the equations as such is not so much of a problem our resolutions are too bad in one dimension probably we can use thousands even millions of points in one dimension and that's it, you just kill them no problem even in two dimensions we can do fairly well but in three dimensional cases we really can't because we can Even with the largest supercomputer available to us now, and with this new code, the Cactus Code, which is very memory efficient, we can only run about 500 points cubed. And that's not a lot for resolution. And that's the most we can do. So generally it doesn't take long enough for your instability to start killing. Yeah, in a typical black hole simulation, it's very difficult to go beyond about 30 or 40 m where I miss the mass of a black hole. before the code crashes. Unless you do very clever things. The code essentially crashes because you develop very large gradients in the metric functions just inside the horizon because you use a slicing condition. I mean, you don't want to hit the singularity because then the code will crash anyway. So what we do is we slow down the evolution inside the black hole and let it go outside. That makes the metric functions develop large gradients. and those gradients eventually will kill you because you won't be able to resolve them and the code will crash again. But these gradients always seem to happen inside the horizon, the apparent horizon. So there's an idea that's called apparent horizon boundary conditions in which you actually excise the interior of the black hole. Everything inside the apparent horizon you throw away and the rationale behind that is it cannot affect the outside anyway because it's already inside the horizon. So it's causally disconnected. it's not easy to implement in practice but there's people here working on that Paul Walker is writing his thesis about that and the hope is that that will help a lot because all these large gradients will be hidden inside and then you'll just throw them away and does that

17:30 so does the kind of do the boundary conditions themselves then present a problem? The boundary conditions inside do present a problem. It's a problem of implementing very complicated, in 3D there's very complicated different boundaries because you have the points that are not in the sphere and this is essentially a spherical or roughly a spherical horizon. So actually getting all the points right is difficult. But not as much as which boundary condition to apply. That's relatively simple because, again, inside the horizon you know that information is only flowing in and not out. so you can always solve for the points inside the horizon using just information from outside the horizon so you don't need any artificial boundary condition, the boundary condition in that sense is natural, there's nothing artificial, so physically we know exactly what boundary condition to apply, it's just as much of doing it in America that's a bit complicated now that's in contrast with what we do outside the grid outside the grid we don't even know what boundary condition to apply physically because there's no consistent boundary condition that tells you only how it's going on waves or anything that doesn't exist in relativity so we just try to do things that we know are wrong that will at least be stable and that's our approach at the moment at that time so did I understand you correctly that part of the problem with the inner boundary condition is that you have a sort of spherical geometry for your boundary but then your grid is kind of a you have all these funny corners with all different you have to implement that very very carefully I suppose then there's some reason why you can't for instance try and do your grid with a different geometry that would presumably make it much more complicated everywhere else after many years of experience in axisymmetric spacetimes it turns out that in an axisymmetric spacetime your coordinate system is singular of course along the axis and that produces lots nobody's been able to make any code stable for a long time because of that. The axis always introduces instabilities. So that experience tells us that using spherical coordinate systems in 3D is probably going to be much worse. So the whole rationale behind going to 3D is to use partition coordinate systems. Coordinate systems are not singular. Because any coordinate singularity will just cause an instability and then you can't cure. That's interesting.

20:00 and so at the at the inner boundary conditions you have this kind of geometry issue then further out with the boundary conditions well say your outside boundary conditions you were saying that the problem is yeah we don't know what boundary condition to apply physically because it's a completely artificial boundary of course base goes up all the way to infinity so what condition do you put there nobody knows any anything you can put there that is it's a very complicated problem what is done normally is just to apply something trivially simple that will make things stable and not very bad the cactus code for example you have this cube and it chooses the normal derivative of all the geometric quantities along the cube is 0 which is of course ridiculous but it works and it's stable it is consistent in the sense that your grid, you're only actually copying the last, the value at the boundary you're copying it from one value in and as you refine your grid in the limit those two would be equal anyway so you're actually not doing anything that will be inconsistent in the long run if you really have an infinite resolution, but still it introduces errors not errors, but it's stable so that's one of the things we're interested now, because if you try anything a bit more fancy, it typically goes non-stable now let me see if I've got a good picture in some kind of abstract analytical sense you might say that the boundary conditions you want to impose are sort of at scry no incoming or outgoing ways but presumably the practical problem here is that you're not dealing with any infinities you're starting at a certain point we don't have scry in the grid we are at a finite distance you had to scry then the boundary conditions would be very natural. So there's a few approaches in that direction. There's one approach using the conformal field equations of Friedrich that goes out all the way to scry because you use a compactification of the grid. And this one person here is doing exactly that. Who is that? It's Peter Hibner. But he's the only one here. There's another approach that we're not doing here at all. that's called the Cochi characteristic matching in which people use a Cochi surface all the way to a finite radius and then they match with a characteristic cone outside

22:30 and that cone goes all the way to Scry so again in that case your boundary condition is very natural but then you have this problem of the matching of the two grids in the middle which is specifically very complicated so there's a lot of people working on that but not here that would be for instance the groups of Pittsburgh and Southampton right Right, so it seems to me that that sort of approach using this closely characteristic matching is a little bit analogous, or maybe a lot analogous, to analytic methods where you have kind of your near zone and then your far zone techniques with some sort of matching between the two. So there's those two approaches, but the standard approach that most people use and the one we use here in the Cactus Code doesn't do that, it just goes out to a finite distance in a space like Slice to apply some artificial boundary conditions there. So there's really, with the approach using the cactus code, at the moment there's no matching, you just apply your boundary conditions. You hope that if these boundaries are far enough away, they won't ruin your evolution. Now, about this question of resolution, being one of the big problems that you face, you mentioned that in the 1D case, where you obviously have a much easier time, fine resolution that you don't have to worry so much about these instability problems. How do you see things progressing? I mean, is it likely that, at all likely, that computers will advance enough in memory and power that you'll arrive at a similar stage in a 3D case and you won't have to worry about it, or is it more than you? I think we'll never arrive to the same stage, but everybody hopes that we'll do much better soon. I mean, that's the hope. A few years ago, we could only do 200 points cube. five years time we'll be able to do a thousand points cubed, and things will certainly improve. So the hope is that they'll improve fast enough so we won't have to worry too much. But we're not sure, that's just hope. I mean, unless somebody comes along and has a very clever idea to do these evolutions with no instabilities, that's the only hope we have. Right. So, yeah, obviously there's the idea that something new might come up. So, exactly, but presumably in any case, you'll need to develop your techniques more and more. I mean, how, sort of at your current level of knowledge, can you give me a rough idea of how big a grid, how, you know, fine a grid you'd need to be able to have to suddenly have all of these problems disappearing?

25:00 I don't really know, because the experience hasn't been very good in that direction. I mean, you can, if you increase the resolution, things look nicer, but you typically don't have to run for much longer. I mean, you can run for 30M with 100 points cube and maybe 35M with 200 points cube. And maybe 40M with 400 points cube. So the time of run is increasing, but very slowly. And that's really disappointing. I mean, that's been a surprise. I think people expected that if you doubled the resolution, you could integrate twice as long in time, and you can't. So I don't really know how much resolution we'll need. So the technological fix, in any case, is not likely to be the whole answer? Yeah, I think we need something new. I have a meeting. Yeah, I'll let you go. Thanks very much. That was very... ...because the microphone is on too many gadgets in it. So now we're actually going, and it's the 12th of February at 9.50 a.m. And I'm talking with Alberto Vecchia. I'm also just interested in the work that you're doing at the moment I believe you're working with Kurt on, for instance, some data analysis techniques for geodes and other things. Yeah, well, yeah, I've started basically because I come from a gravitational wave detection experiment in space, basically. So I started doing my work in Italy with people working on double tracking of interplanetary spacecraft like, I don't know, probably you heard about Ulysses, Cassini, Mars Observer and stuff like that. So I did my PhD basically working on low-frequency experiments for the detection of gravitational waves. And then, well, during the PhD, I moved to Carbic to spend there some time. So I started working with people involving interferometers and instruments that likely will detect gravitational waves, not like Doppler tracking of interplanetary spacecraft, which are not dedicated instruments.

27:30 so I decided to basically shift towards high frequency experiments and so I started doing some sort of doing some work on basically techniques for not simply detection of gravitational waves but for parameter estimation so more More for the future than not for the immediate need of the detectors where you want, of course, to first pick up the signal, but so my work was trying to understand exactly how well you could estimate the parameters of, for example, coalesce and binaries and what are the best techniques that you can use for doing this job, basically. So people were always working on maximum likelihood estimation of parameters, and then they found out that probably that's not the optimal way of doing this. So with David Nicholson there, I started to look at different techniques. really it was really a very first approach to the problem just to have an idea because people find out that the estimation for the accuracy in parameter extraction was probably much better what was probably much worse than you expect so I would say that gravitational wave astronomy probably is less powerful than we thought two or three years ago and the problem was to having realized that probably the expectation were optimistic about parameter extraction algorithm, we wanted to try to see whether we could find something in the literature, some more efficient techniques, basically, that we could implement in the data and in the data analysis in order to improve this, the performances, basically, of the data analysis for this problem. So that's, I I would say more my personal history, how I got involved into this business, shifting from, as I said, space project to, well, earth-based project.

30:00 and so at the moment I'm still working on well I'm still working on on that problem but considering that is first of all people want to to extract the to to find gravitational waves and let's do a gravitation wave astronomy I got involved also another problem of that analysis just very recently and in particular in problem of searches for pulsar and for let's say unknown signal signal that you for what Kurt called UCO unidentified chirping object so I decided to use this fancy word yes apparently people are starting to use it And so for the Pulsar search, there are already many people here involved in the problem. I'm sure you know that there is this huge computational, that basically the search of Pulsar is limited by computer speed, by the amount of templates that you need to cross-correlate with the data. So there are several different approaches that have been proposed really recently. And first of all, to do a hierarchical search of signals, then to try different techniques in order to compress in a suitable way the data and then to be able to use this different way of basically writing down the data with parallel computers. so the idea is not to use a whole set of I don't know one month of observation but to divide this observation in different chunks then analyze each chunk individually and then produce some time frequency table and then look for

32:30 pattern in this in this table so you this is you save a lot of computer time because in this way because you can parallelize this problem because you you are have also basically dots in a map and not you all have t and so on so so these are techniques that people are looking at and I'm well I'm I'm getting basically involving this problem and at the same time I'm trying to to find out in different fields basically of information theory of data analysis what kind of technique people are using to extract signals of completely unknown form basically where your prior information on the signal or on the waveform is really very low yes or probably you haven't got a clue on what you're looking for and you just see whether there is some characteristic pattern in the data so that's now it's everybody's have been talking about the that the searches people are implementing at the moment are really focused on some object that you know basically in advance like coalescing binaries or pulsar or stuff like that and most of the time in astronomy you detect something that you were not expecting at all and so if we are really limited by templates then this this is a big problem so people would like to know for searches of completely unknown stuff if there is some strange event that you cannot predict because your fantasy is limited basically. So I think many groups around the world are trying to get into this business and so that's what I'm doing. And also with Kurt I'm working on laser interferometry in space. So we started to consider that there are projects like Omega and Lisa that at present, at the moment, it's really difficult to know if and when this project will fly.

35:00 But I think there is, well, there is an effort from both USA and Europe to try to see if it is possible to get money to fly some sort of space version of the Interferother. I know that Ron Hellings and JPL is trying really hard to fly Omega, maybe a very, some sort of very limited version, but just to see a few signals, considering that you know that there are sources there. And then to, well, if you have detected gravitational waves, then you can propose a much bigger project. And so with Kurt, basically we wanted to check how accurately you can extract information about the sources, considering that you could have very complex signals from collecting binaries of massive black holes, even more from a small object orbiting around the black hole, for example, in galactic casps and so on. So you could have eccentricity, very high spin, very large spin, and so on. So much more complex waveforms. And also because you want to do cosmology probably with LISA, so you want to estimate the parameters, the cosmological parameters using gravitational wave observation, it would be important to know whether you can more or less locate the source in the sky, identify the source in the sky. So you want to know the angular resolution of these instruments as well. And so we are trying to figure out these numbers. Also because everybody was saying that these instruments are absolutely great, you can do whatever you want, but of course it's not so. So we found out that, for example, the angular resolution is pretty much degraded because you have to, from only one instrument, extract all the parameters of the signal.

37:30 So basically these are the main projects I'm working on. And, well, I'm still collaborating with the people involved in space, in an experiment with spacecraft, Doppler tracking of spacecraft, because now there is probably the final giant mission from NASA, which is Cassini. And, well, the main goal of this mission is to study Saturn and the planetary system of Saturn. But there will be the longest test acquisition run for searches of gravitational waves in the low band frequency and also with what is expected to be a much better sensitivity. So you have this instrument and you would like to see what you can do basically in the low frequency band. But the sensitivity is really limited because these are not dedicated instruments. And so there would be three experiments in 2002, 2003, and 2004, where for 40 days there would be a data acquisition, basically. And then you offline analyze the data. The main problem is that the search is not, well, there is the problem of the even rate. So you need to have a very strong signal. So, for example, collecting of massive or supermassive black holes of mass around 10, well, between 10 to the 6 and 10 to the 9. And the maximum distance is a few hundred megaparsec. So you don't expect to have, you can reach, for example, the Virgo cluster, you can go probably at, I don't know, 100 or 200 megaparsecs with a signal-to-noise ratio of 5 or 10, but the event rate is probably extremely low. So, well, you do it, but you don't expect anything in particular.

40:00 And the other thing that people want to target is the galactic center, because it seems that there is a massive black hole in the galactic center, and there could be a low mass compact object falling into this black hole, and here the sensitivity is much quieter, so you could detect masses down to a few solar masses. So this is astrophysically interesting, even though the event rate is extremely low, because I don't know, it's 10 to the minus 6 per year per galaxy. So you have three times 40 days, so you have a third of a year basically of data. And so you can access basically only the galactic center, maybe some of the galaxies in the local group for this kind of system. But so the sensitivity is really limited. What instrument, the ships, is it the spacecraft's normal communication? Yeah, basically they have the communication, they have a special radio link on the spacecraft and they have the usual telemetry but this is not useful for the experiment because they need the real problem is that you need to track the spacecraft continuously so you need to have all the station of the all the antenna of the the space network basically working, covering the spacecraft 24 hour days for let's say a month. And so for this spacecraft they have a new carrier at about 32 gigahertz, what they call Ka band frequency and so for for these 40 days basically they they transmit back and forth this signal and they measure the Doppler shift between

42:30 the transmitted signal at the earth and the received signal yes and of course the spacecraft and the earth are moving with respect to each other so the is due to the motion. And then there is a Doppler, well, an effect of shift of the frequency because the frequency, the light is not traveling in the vacuum, but there is interplanetary plasma, there is the atmosphere and so on. But you can also show that if there is a gravitational wave passing through the detector, then you move these two masses of the Earth and the spacecraft, so there is a Doppler shift, with a characteristic shape, which is a three-pulse shape. So... What causes the three pulses? Because basically, because you have, basically, the effect of the wave, you have one photon that travels from the Earth to the spacecraft and then coherently back to the Earth. If you have a gravitational wave passing through this detector, basically this photon is shifted three times when it's emitted at the Earth and is received there and then comes back. So the signal at the detector output basically is a convolution of the gravitational waveform and this three-pulse detector response function. So there is a clear signature, basically, of the signal, which depends on the distance of the spacecraft and on the position of the source in the sky. and so there could be some hope if you have a signal a subcandidate to try to understand whether it is well ask a cosmic signal or just an artifact of it all the instrumentation and And that's interesting. Omega, is that just an acronym now for the American version of Liga, Lisa? Yeah, Omega is orbiting, I can tell you exactly what it means, medium explorer Omega.

45:00 so Lisa Lisa basically was proposed by Europe, by ESA and then there was and so it's basically a European project but there are also people from USA involved in it for example there is Peter Bender who probably had for the very first time the idea flying an interferometer in space, and there are people like Faulkner at, I think I can find it somewhere in a minute. But then Ron Hellings at the JPL decided to go for a basically smaller mission less expensive and so with also less powerful but the same time he thought it could be able to fly this instruments naturally so that's so Omega is the smaller yeah it's just basically a smaller ratio so instead of the arms instead of being five million kilometers are only one million And also the orbit is less complex because Liza is, I don't know what you know about it, but it's orbiting around the sun, and then it's tilted by 60 degrees and counter-rotates. So basically, you sweep the sky during one year of rotation because the angle of the orbital plane of the instrument is moving around the sun, so basically there is a precession in the sky. So even though you have only one instrument, you can have, for example, information about the position of the source because you change the the orient you have the Doppler effect of the antenna moving around the sensor with respect to the source but also the antenna pattern changes because the orientation of the antenna changes. Omega is not an heliocentric instrument in an heliocentric orbit

47:30 so it's in a geocentric orbit it is around the earth and is not tilted by with respect to the plane of the ecliptic so it keeps all it's basically on the same plane of the ecliptic and it stays on that plane for all time so you lose so it goes around the Sun with one year period it rotates around itself with more about I think the period is 50 56 days so it's rotating faster but at the same time it keeps only the same orientation with respect of a source in the sky yeah so you lose a lot of angular resolution basically OK, so omega means orbiting medium explorer for gravitational astrophysics. So it was proposed for, maybe you want to have a look at this. I don't know. I think you can find it on the web or. I think last year, two years ago, it was selected at that time, but I know that Helix is going to resubmit the project this year and also is thinking about, for example, tilting the orbit because he realized that, for example, you lose all having this, let's say, simple orbit. So probably with no extra cost you could get a better instrument just while choosing a different orbit. But nobody knows whether this instrument will fly or not in some decent time scale. Because now, at the moment, well, the time scale for LISA is 2017 or 2025, something like that.

50:00 And all the projects with LISA were slowed down by the crash of clusters. Because there was the mission of the five spacecraft that crashed. Well, during the launch basically the rocket exploded and so they decided to rebuild all this mission and therefore all the, with of course extra cost, time delay for all the other projects and so on. And Lisa apparently hasn't got a lot of planning also for so large projects like this. But I'm not aware of all the political problems with Lisa, so it's very nice to be the person to talk to her. You mentioned earlier that one of the conclusions you had, you reached out while at Cardiff working with David Nicholson, I guess, was the gravitational astronomy, I guess, parameter extraction and so on from binary systems might be less powerful than the original. What were some of the factors for that? So, all the studies that have been carried out so far, basically, on parameters extraction are based on the computation of the Fisher matrix. So, you have the variance covariance matrix and you invert it and the diagonal elements of this matrix basically tells you the error associated with the measurement. it but so what what's what is called the Kramer rail Kramer rail bound on parameter estimation errors but it is known and everybody knows it that this this method works in in the in the limit of very high signal to noise ratio now

52:30 Now, very high signal-to-noise ratio doesn't mean very much, it depends with respect to what. So, the problem was to check whether a signal-to-noise ratio of about 10 was for that problem of coalescing binary a high or low signal-to-noise ratio. From the point of view of extracting? Yeah, from the point of view of extracting a parameter. You wouldn't say it's very large because you need something around 7 just for detection. So probably 10 is not a large signal-to-noise ratio. So the problem was that we tried to understand was instead of using an approach that was valid only this asymptotic limit, to use some more powerful approach that was more complex, but that could give results that are valid in any regime of signal-to-noise ratio. And also that is a prediction of the error on the parameter extraction, which is independent of the algorithm. So you basically, you theoretically work out how good can be an estimator, basically, and then you can figure out many different estimators and try to see which one performs, as you would expect, computing this theoretical limit. So, basically, we used a different technique to compute the theoretical best performance of an estimator without computing the estimator. And what we found out was that the Cramer-Rey bound was underestimating, basically, the error, and the error should be actually larger. So at the same time, actually six months or a year before, the people in India, in Ayuka, just made a simple Monte Carlo simulation where they estimated the parameters of qualescent binaries using maximum likelihood techniques. And they found out that the error that they got were a factor of two or three larger than the errors that are predicted using the Cramer-Rail, using the Cramer-Rail-Bound.

55:00 So basically what people started to realize that the Cramer-Row bound was underestimating the accuracy in parameter estimation and then that the maximum likelihood techniques were producing error much larger than this theoretical, let's say, unbeatable limit, which was not really the unbeatable limit because it was underestimating the error in a typical signal-to-noise ratio regime. And we tried to understand basically where is this unbeatable limit. And we found something that is more or less in between the two. So still there is a discrepancy between what we think could be the best optimal accuracy parameter estimation and what you can get from real estimation techniques. So the problem is to figure out some estimator which is more powerful than maximum likelihood estimator and try to reduce these errors. And so, this is something that I'm working on at the moment, so to use different estimation approaches, basically. When working with these calculations, are you assuming perfect knowledge of the signal beforehand? Yeah, I'm working only on the side of the data analysis technique, not on the side of the possible bias introduced by the lack of knowledge of the signal. So I'm assuming that, I don't know, 2 p.m., 3 p.m., 5 p.m., whatever people can compute is the actual waveform that is produced. I'm not I'm actually using waveforms that are only Newtonian or 1.5 post-Newtonian approximation because I'm not concerned at the moment with the accuracy of the of the template that people give to me but on the on the

57:30 power of the efficiency of the algorithm so the algorithm could introduce much larger error than the actual lack of knowledge of the signal so we thought that's that was part of the problem that needed to to be understand better understood better because all the work was on computation of waveforms and nothing on basically on the algorithm and especially since as you say that you may actually reduce some further accuracy with the algorithm because you probably won't have probably had perfect of course and also you want to know the bias basically of this of these estimators you want maybe maybe you can have a very small error but actually you are far away the error the statistical error is very small but you have a large bias so you are well probably more in trouble in this situation so at least we want to start having some some idea if of control on this, on these techniques basically. Now, I think you mentioned also, so when you were, well this is on a slightly different subject then, connected with these UCOs, you mentioned looking for existing techniques that may be used in other fields, Is there any evidence so far that there are other fields that have techniques that might be useful? I'm trying to have a fair answer to this question. Well, probably I should say I don't know. In the sense that I just started very recently to look at these techniques. So there are a lot of techniques that have been developed, for example, in geophysics, for example, for modeling of earthquakes, where it's really difficult to have a prediction, a very good model of the waveform.

1:00:00 So you have the time series and you want to figure out the waveform. There are techniques that have been used, for example, in economy, trying to understand trends of some indicators. That can be, well, the Dow Jones Index or some bond, but some technique, for example, has been already used in astronomy, where you have, for example, variability of AGN or QSO that have some strange time behavior and you cannot model, you don't have a very good physical model of the system, so you want to try to have a look at the waveform basically that you detect but my problem at the moment so there are several techniques that have been suggested my problem at the moment is is our different or I think I'm facing a problem at this time because you know these cases basically they have already a signal for example for the case of the earthquake they have there the time the time series so they just looking by yeah you can you can see that there is there is not not just noise but also a signal or for if you look at data from AGNs you can see a light curve or something like that so these techniques it seems are techniques that are used to obtain a model of the waveform using a different a different approach the problem here is we have let's say one year of data from Geo, LIGO, Virgo, forever, and you want to see whether there is a signal or not coming from some unidentified object or unknown object. And so it's not clear to me at the moment, for example,

1:02:30 whether you can use these techniques and how the statistics work, basically, for this technique. So the literature is really huge and And I'm, well, I'm trying to understand basically whether there is some direction, some, yeah, where to go basically, if we can pursue some technique, if there is something that we can use. So the likely problem in this case that they wouldn't have any others would be that when you look at your years' worth of data from GEO, it's not going to be obvious that there's a signal. Exactly. So the first stage is not, is, you say, for example, you search for everything you know, Maybe you find something, maybe you don't. Then you want to search for something that you don't know. Now, the problem is basically to decide whether you have seen something or not, whether you have detected something or not, and not just looking at the... I'm sure you can find signals in the data, but probably these signals are coming from wires or oscillators of the mirrors So, signals that are instrumental and not astrophysical signals. So, the problem is to get an algorithm for detection and not, let's say, a parameter estimation or a model of something that has already been detected. As it seems to me is the case for most of these techniques that I've seen around. But this is a really premature statement because really I don't know anything, so I'm trying to constrain the number of papers or of books that I have to read to try to understand something. I know that people, well, I had an exchange of emails with Kip Thorin, I think who's, it's a long time he's trying to figure out whether there are out there techniques that can be used for this purpose and he says that well he he doesn't know basically what one could do at the moment people suggest for example time frequency techniques

1:05:00 it's really nobody has really done some quantitative work, maybe with some very simple signal, but just to have an idea how different technique can work or how confident you can be applying the two legs to this data stretch or the two legs to the same data stretch. So I think in this field we are really at the very beginning because all the attention has been focused basically on coalescing binaries. Now all the attention is moving toward pulsar, also for projects like geo that are, they have only one detector and so at the beginning they can search probably on for continuous signals. Because of the problem of finding coincidence. Yeah, exactly. I mean, you can find a coalescing binary, but nobody probably will believe you, yeah. And also because, for example, GEO is very limited for searches of coalescing binaries because we probably can get at the virulent cluster but not beyond it. I don't know. I don't remember because nobody... because basically the low frequency cutoff is at, well, rather large frequency, something around 50 Hz. So, well, the noise sensitivity curve moves quite a bit, and you don't know how is the real one. But anyway, GEO is not an instrument for coalescing bandwidth. It might detect coalescing bandwidth. But at the same time, you have one event. And if it is not an extremely high signal-atomized radiation event, it might be you have also different way of doing coincidences with other, well, electromagnetic instruments or other instruments, it's going to be difficult that people will basically believe you. And also you believe you are yourself.

1:07:30 Unless you can get a coincidence with something like a gamma-ray burst. Exactly. But, for example, for gamma ray bursts, if they are cosmological, there is no way that you can detect this stuff. I was going to ask you about your work with the doctor tracking device. the people you're working with there, are they largely the people performing the experimenter or are they other theorists who are helping with the data now? So, basically, the number of people working on Doppler tracking is really strict. I think there are 10 people, no more than 5 maybe in the world. In the sense that people started—well, this technique was proposed, I think, in 1975 by Easterbrook and Warquist. And there was a paper by, I think, Thorne and Braginsky or Braginsky—I don't know how to pronounce it in 1976 on where they said ok the doctor tracking is the only basically the only technique we have at the moment for gravitational experiment of the low frequency regime and if you add merger and collapses of large masses at large distances you could have burst from basically massive blackball so that's an important um direction of research basically yes and so also there were this uh spacecraft already flying so voyager pioneer from nasa and it was re it was reasonably inexpensive to do basically, and to do these experiments. So they started, I think the first paper was published in 1980, 1980-1981, I think by Hemmings, and they did an experiment, or one day, I

1:10:00 think experiment, with I think Pioneer, either Pioneer or Bajer, I don't remember exactly. So these were non-dedicated experiments. And so very low, basically you had everything already there. You just needed to ask for time from the deep space network to track the spacecraft for some time. And you needed a very stable measure as standard frequency in your transmitting and receiving stations. So I think things started in the late 70s, early 80s. And then, mainly, yeah, were people from JPL, so, well, used to work on Walpist, but I don't think they never get involved very much with the experiments. They proposed the method, but I don't think they did any experimental work. Then there was, well, Ron Hellings and John Armstrong, basically at JPL, that started working on this project. And then Bruno Bertotti from Italy was involved, got involved basically in this kind of project when Europe I think started to collaborate with NASA. and to have instrumentation on the spacecraft and also contribute with money to the missions. So I think early in the 80s it was proposed to fly Ulysses, the Ulysses mission, to study the pole of the sun and so on. And so with the Ulysses mission, also people from Italy got involved in the Doppler experiment. And I think Bruno Bertotti is the people from Europe. I think he was actually the principal investigator for the Ulysses mission.

1:12:30 But I'm not, that was scheduled I think sometime in mid-80s, but then there were several delays, also because the space shuttle, well there was the accident with the space shuttle in 87, I don't know, except that would be the date. So this mission that was a step forward with respect to previous missions, because there was some instrumentation devoted to these experiments, that is basically carrier frequency at higher frequency, basically. So the transmission was not in what is called S-band, so at about 2 gigahertz, but was in X-band, that is at about 8 gigahertz. And consider that one of the main disturbances is the plasma noise, and the plasma noise is proportional to 1 over the square of the frequency, so the higher the frequency, the lower the noise. So the problem was to get higher frequency carriers. And so actually the experiment was done with Ulysses, was done in 19... So there was one Tata acquisition in 1991, then one in 1992. And then there were the two spacecraft from NASA, the Mars Observer and Galileo, that were launched early 90s. And so they had a, both spacecraft had, I think, a large program of data acquisition for searches for gravitational waving of frequency. But there was the problem with the Galileo where the main antenna didn't open at all. So the high gain antenna basically didn't work for the old mission. they had to use only the low-gain antenna and saw that the sensitivity of the spacecraft was reduced drastically.

1:15:00 And there was a coincidence experiment between the three spacecrafts, so Galileo, Ulysses and Mars Observer, between February and March 1993. and so I think so John Armstrong I think is in charge was in charge of the gravitational wave experiment for Mars Observer Frank Easterbrook from for Galileo and Bruno Bertotto for Ulysses I think it's right So that was considered a major step for these non-dedicated experiments because for 20 days three spacecraft in three different positions in the sky were continuously tracked by the space network. And there were two additional stations, one in Italy and one in Japan, in order to have the best possible tracking, basically, of this spacecraft. people are so so that's some sort of historical background but the if you first of all still the people are looking at the data of the 93 coincidence experiment because there is John Armstrong at JPL working I think more or less full time on this problem and there is Luciano Yes in Italy who is working on this problem. And I think they are the two people that are really working on this step. And then the step forward for this mission was the Cassini mission. Originally, I think Cassini was proposed to have not only one spacecraft, but two spacecraft. I think it was called Cassini craft mission. So two spacecraft in two different positions. But then there was the problem of the budget and NASA.

1:17:30 And I think they tried to kill more than once the mission, but the mission survived. And so now there is only one spacecraft anyway that the mission is going. And so there is another upgrade in the transmission frequency. So you expect to improve by more than an order of magnitude because you are going from 8 to 32 gigahertz basically. You have to square this quantity. So you expect more than an order of magnitude of improvement in sensitivity. You have much longer than acquisition time. You have three different experiments. so it's people are well that's the back basically the best chance for a low frequency experiment even though problem is the last for Doppler tracking because at least at the moment there is no no mission of schedule after after Cassini and I think also the policy of NASA is to move from giant mission to smaller missions so so maybe this is the final final one and but then and you don't know whether it will get there safe and we'll be able to do all this all the experiments anyway And for this, I think John Armstrong, Bruno Bertotti, Frank Kistelbruck, and Buchanan-Eyes are the people in charge for the Gertrisch-Floer experiment. But I'm not sure exactly how formally the agreement is, so this is something you should check. When you were working in Italy on the dumper tracking originally, were you involved on the data acquisition side as well or just...? Well, I did basically every possible thing, in the sense that there is really the problem of manpower. So there are no, for example, postdocs hired to work on this experiment. So I, well, I spent a lot of nights basically acquiring the data from the spacecraft in Italy.

1:20:00 Near Bologna there is a radio telescope that was used for it. So I stayed there basically trying to see that everything was going right. There wasn't very much to do, just to be sure that the antenna was tracking basically the spacecraft, just to start the operation and to finish the operation. So I did this and then I also worked on the data analysis. So basically the first problem is of course to try to remove the Doppler effect induced by the relative motion between the Earth and the spacecraft, because it's really huge, of course. And so I work on the preprocessing of the data before you can actually search for stuff. And then we searched for nearly sinusoidal signals, basically, and also we use this data for estimate of the gravitational wave background in the frequency band between 10 to the minus 5 and 10 to the minus 2 and of course you will get only go for their walk between 1 and 10 to the 4 so that's it's not It's not significant at all, of course, as an astrophysical limit. And, yeah, so... With the Doppler Tracking Experiments, pretty much everybody did... Yeah, everybody did something, you know. Also, permanent professors were basically forced to stay awake at least for one night and try to to to get some data from the spacecraft so at that time I was a PhD student so I spent there I think 20 nights for both the experiments and so also also everything is I mean the software is designed it well is written

1:22:30 by people that work maybe for a year or two during their thesis for the PhD or for the undergraduate and then you use it. So it's not like for example in LIGO where there is a team working on the software then someone who validates the software and then you check it and it's much more, it's really a very small group of people. So with all the problems self-dedicated project, basically. Then, in contrast with the ground-based experiment, there seems to be, well, obviously, there's a definite distinction between experimentless and theorists. And I get the impression that it's primarily the theorists at the moment who are looking at data analysis. Yeah. Is there a reason for that, that the experimentalists are not interested at this point, since they don't You mean for ground-based experiments? I mean, are there any of the experimentalists that you work closely with in any way? It seems that there is... I think there is some sort of historical reason that probably Kip Thorne and Bernard Schutz The people who started looking at data analysis issues, they came from theory, basically, from really theoretical work, and most of their students also came from theoretical work, and they had this dream also towards data analysis problems. I think that for a gravitational wave experiment, I think there is a big difference with respect to other experiments, for example in astronomy, me because here you you are fighting a lot against the the signal to noise right against the noise basically you're going to search for the lowest SNR signals and so all the effort of the experiment it seems to me is on the

1:25:00 instruments, trying to get the best performance they can out of the instruments. And the data analysis, and then there is no community, a data analysis community for gravitational experiments probably because there are not real data at the moment and and the data analysis was triggered basically by the fact that people realized that to to explain So the people working on the instruments basically take the point of view that I don't care about what's happening in the universe, I'm just trying to get the best instrument I can. If I get the perfect one, I'm sure I can attack something. People working on theory, I think, started to realize that even though you can cover the entire universe, It might be a year of observation is not enough for observing a signal at the level of sensitivity that you have now. And I think that the data analysis problem was very linked to the astrophysical one and to the theoretical one. Because you can do that analysis if you know in advance the waveform, at least for the sorts that people were only looking at until maybe five years ago. I don't know, like all those in binaries. So people thought if we know exactly the waveform, we can do that analysis. Probably also theorists got really excited because it was a sort of renaissance of general relativity. And so data analysis was also a good excuse for doing really GR instead. And it seems to me that now there is this completely lack of communication between a huge community of data analysis in completely different fields

1:27:30 and the gravitational work experiment, because I'm sure that for radar problem, sonar problem, in engineering, they had to face problems really difficult for signal processing. And we should try to get these people involved, I think, in these searches for gravitational waves. But for some reason the community is still very, very restricted. And until one year ago, a conference for gravitational wave data analysis said, I don't know, it went to 50 participants. Now, well, the last meeting on gravitational wave at Mars, we were about 150. So I'm sure... Yeah, in Paris. So I'm sure that in one or two years there will be many more people. But the problem is also the background of these people. I think that we could have a lot of inputs from really experts in signal processing. I think if you explain them just a few things, they could give you very good advices. And I'm sure that most of the work people are doing now in the analysis for gravitational waves has already been done in some form somewhere else. So are almost all, let's say the 150, are there almost still people from GR? no now is starting to let's say there are new faces even though it's difficult for me to understand I mean I think people are involved in all the country involved in a gravitational wave experiments well they see this experiment that are funded there is a lot of money in these experiments and so it's I think also for for for for

1:30:00 their career could be a good thing to get involved in these problems even though they they're not they they come from from some completely different field of research i i must say that well you you i'm looking especially at the italian situation for example what's happening with virgo yes where there is a lot so let's say my my My judgment is, my comment about the situation is, I think is mainly focused on the Italian side of the situation, where I think there are a lot of really bugs in the project, in the sense that there is an awful lot of money involved in it. And so people basically are rushing into this project, trying to get important positions there, because they know that they can stay there for 10, 20, or 30 years. And so, well, with all the problem of recruiting the people, so I think that there is no really a scientific reason why so many people are getting involved in this project. But it's just that the competition is growing so much in science these days. It's so difficult to find a job. That is a great opportunity, is an open field, basically. Nobody is an expert in gravitational wave astronomy. So everybody could become a great expert. So for example, just because if you look at the participant in the meeting in Paris that was organized by Virgo, probably two-thirds of the people were from the Virgo project, or less related to the Bureau project. And I think that there is, I think this is an important point to understand how people get involved, at least from my point of view,

1:32:30 in these days in this project. Because it seems, for example, that in LIGO there are A lot of people coming from particle physics, well, experiment, large experiment, were basically killed. And so there was this shift of people toward gravitational wave experiment. In that case, I'm sure it's very good. I'm much more skeptical about the Italian situation. So the projects have reached the stage at this point where it's becoming unattractive to you? I think, yeah. I think, well, I remember that when I was doing my PhD, so two years ago, people basically were laughing at me because I was working on gravitational waves, also because I did the PhD in astronomy, so that was really something strange. Now, I'm not saying that people take you seriously, but it's getting really an attractive field. Also because, I mean, a lot of manpower is required, and nobody basically knows what to do, in the sense that there there is a there are so many open questions and everything basically is still to start so so there are a lot of people moving from different fields also a lot of people working for example in engineering uh that are becoming involved in this in this project In the U.S., as you say, many of the people moved from particle physics to... Yeah, but I don't know exactly the situation there. I know just by suddenly... Well, LIGO started to hire people, and a lot of people... I think, I hear it comes from particle physics, I think that the super collider was the big project basically to get the money. And so they saw in LIGO basically a wonderful opportunity to work in a project.

1:35:00 And that also had a lot of money, at least at that time. I don't know now the funding situation. But in the Virgo case, lots of the new people from one particular field at heart physics are just different. Well, I don't know the French side of Virgo. I know a little bit better the Italian side. people come from well many different areas of of physics uh well there is the the hardcore of virgo which is the uh the people working on the on the instruments and everything started just i think by looking at the some new technique of isolation of the of suspension so i think that's how people how Virgo started but nowadays I think there are that people come from really many different fields in physics they for some reason they they they get involved in in the project or they they get interested in the project and it's You have to consider how is the Italian situation where basically you have only permanent jobs, so you get your position and then even though you are working on some project that eventually finishes, then you still have your position. So, there are some projects that go on for 30 years, but not all the projects. So I think that people that already have a job, they can also find some good opportunity to recycle themselves. But I don't want to use this word in a better way. So, I think this could be one of the reasons, even though I'm not so tightly linked at the moment with Italy, so I don't have a really good idea of what's going on there.

1:37:30 Do you, in the work that you're doing at the moment in data analysis, do you make use of data from any of the experiments like Geo? No, actually I'm working only on simulated data. So I'm not, I'm planning to work, well, pretty soon on data that are available from prototypes. or for example here we have data from the bars also detectors and but at the moment I'm so in trouble with the with the design of the algorithms that well very nice Gaussian stationary simulated noises is much better than real data But, yeah, but let's say I really hope that in six months time I will be working also, applying these techniques also on real data. So, also, yeah, also because there are trouble with the true noise, so there's also this, this other side of the problem that one has to tackle, not only the signals, but also the noise because everybody is basically assuming Gaussian and stationary noise and we know that he's not Gaussian and he's not stationary. So that's the next step. Do you find that the experimentalists or do you have enough communication with the experimentalists to know their feelings on what are the important noise? as I know when I heard them talk about it they were always emphasizing the non-Gasene noise so it was clear as I was talking about Gasene you mean whether they have an idea about what are the reasons of non-stationary or non-Gasene noise I'm not sure I've understood I guess I mean I guess I mean whether they have I guess I mean

1:40:00 whether they have types of noise that they see or feel they're going to be there in the detectors that might be particularly problematic from the point of view of... Kurt, for instance, the other day mentioned that somebody had pointed out that one possible use of looking for unexpected signals was that you would also be picking out non-Gaussian noise potentially you know, sort of, which might, which might, you know, be seen as a sort of signal in the attack or something. Yeah, exactly. All this, in fact, I think that most of the technique we are looking at at the moment will be very useful for, let's say, diagnostic or for cleaning, basically, the data stream. because, of course, there are signals that are not from a cosmic source but present with a very high signal-automized ratio there. So if you have some algorithm which is able to pick up some signal, you can use it for cleaning the data. One of the main problems with the instruments at the moment is that maybe you have one hour or one day of data from one instrument and then you start analyzing it. Then you find out some particular feature that you don't understand. So you go back to the people at the instrument and you say, look, I have these problems here. Do you have any idea? And unfortunately, you cannot rerun the instrument, basically, Because these are all prototypes, so basically people are working continuously on the instrument. They are putting new stuff, changing the configuration, add parts, remove parts. So there is never the same instrument there. And I think it's really difficult to, I'm really scared about the problem of moving from a 1 meter or 10 meter instrument up by 2 or 3 yards of magnitude in size.

1:42:30 people are saying well more or less the same stuff you see on the prototypes you will see on the real instrument even though everything will be much better because we will have all the perfect well the best possible system of isolation of control there I'm not so I'm not so sure maybe I'm not trusting too much the people working on the instruments I hope I'm wrong but at the same time it seems to me that there are a lot whenever I've looked at the data for example my experiences on the well for the Doppler data there are thousand really thousands of things that you really don't understand and you you cannot find out where they came from yes and there are extremely strange stuff that can cause these problems for example different length of cables I remember that this was one problem that we had for for the for the for the We were using cable, for example, in some setup of the instrument of different length, and this causes problems in the data that we were acquiring. Just because there was a tiny time delay in two different channels. And if you have an instrument of, I don't know, a couple of kilometers, three kilometers, four kilometers, and all that instrumentation, I can think about billions of stuff that can go wrong. Hopefully everything goes right. So I think that it's really premature to say something about the predictability of the noise. I think that the phase of test of this instrument, we need to be long and careful. So I think the best thing we can do is to try to get experience on powerful technique basically. This is at least my feeling. I'm sure that the people working on the instrument will try to get,

1:45:00 they will have a beautiful feeling about the instrument and all the mechanical or instrumental stuff. But we are... I think we need to work much harder on the data analysis side because I think we are in some sense late because in one year or two year at least GEO for example will be somehow working so and you want to be ready to get the best data out of the instruments. So I think there is really an awful lot of work to do in this field. So for instance, your experience might even tell you that in the early stages of the experiment getting up and running, assuming the theorists have some access to the data, they could even be of some use from the diagnostic point of view. Yeah. I think, at least for GEO, I know that there will be, I think, several months of, for example, or even a year of testing of only one arm of the instrument, just to understand. Because, well, they were building up the second arm, but in the meantime, you can start looking only at one arm. And you are not yet doing interferometry but at least you can look at what happened in one side of the instrument and try to get information from that side. I think it's also a matter of experience because probably you cannot trust only statistics and signal processing but also really a feeling. sure the people actually well just looking at well friends of mine working in astronomy looking at the data they could understand exactly the mathematics was saying something but they could actually understand something different just by experience by looking at the data after thousands and thousands of stars observed and so on so I think that that's really yeah I appreciate that but my life is an astronomer After a while, yeah, they have to look for... Yeah, I guess.

1:47:30 They are great. They look at the... Just at the raw data, they can understand that there is a periodicity of 1.32 seconds or something like that, without any algorithm. So, yeah, that's the... I think there is a completely lack of experience. Sure. So... And we need it, I guess. Thank you very much, it was very interesting.