Learning Bayesian Statistics

Proudly sponsored by PyMC Labs, the Bayesian Consultancy. Book a call, or get in touch!

In this episode, we dive deep into gravitational wave astronomy, with Christopher Berry and John Veitch, two senior lecturers at the University of Glasgow and experts from the LIGO-VIRGO collaboration. They explain the significance of detecting gravitational waves, which are essential for understanding black holes and neutron stars collisions. This research not only sheds light on these distant events but also helps us grasp the fundamental workings of the universe.

Our discussion focuses on the integral role of Bayesian statistics, detailing how they use nested sampling for extracting crucial information from the subtle signals of gravitational waves. This approach is vital for parameter estimation and understanding the distribution of cosmic sources through population inferences.

Concluding the episode, Christopher and John highlight the latest advancements in black hole astrophysics and tests of general relativity, and touch upon the exciting prospects and challenges of the upcoming space-based LISA mission.

Our theme music is « Good Bayesian », by Baba Brinkman (feat MC Lars and Mega Ran). Check out his awesome work at https://bababrinkman.com/ !

Thank you to my Patrons for making this episode possible!

Yusuke Saito, Avi Bryant, Ero Carrera, Giuliano Cruz, Tim Gasser, James Wade, Tradd Salvo, William Benton, James Ahloy, Robin Taylor,, Chad Scherrer, Zwelithini Tunyiswa, Bertrand Wilden, James Thompson, Stephen Oates, Gian Luca Di Tanna, Jack Wells, Matthew Maldonado, Ian Costley, Ally Salim, Larry Gill, Ian Moran, Paul Oreto, Colin Caprani, Colin Carroll, Nathaniel Burbank, Michael Osthege, Rémi Louf, Clive Edelsten, Henri Wallen, Hugo Botha, Vinh Nguyen, Marcin Elantkowski, Adam C. Smith, Will Kurt, Andrew Moskowitz, Hector Munoz, Marco Gorelli, Simon Kessell, Bradley Rode, Patrick Kelley, Rick Anderson, Casper de Bruin, Philippe Labonde, Michael Hankin, Cameron Smith, Tomáš Frýda, Ryan Wesslen, Andreas Netti, Riley King, Yoshiyuki Hamajima, Sven De Maeyer, Michael DeCrescenzo, Fergal M, Mason Yahr, Naoya Kanai, Steven Rowland, Aubrey Clayton, Jeannine Sue, Omri Har Shemesh, Scott Anthony Robson, Robert Yolken, Or Duek, Pavel Dusek, Paul Cox, Andreas Kröpelin, Raphaël R, Nicolas Rode, Gabriel Stechschulte, Arkady, Kurt TeKolste, Gergely Juhasz, Marcus Nölke, Maggi Mackintosh, Grant Pezzolesi, Avram Aelony, Joshua Meehl, Javier Sabio, Kristian Higgins, Alex Jones, Gregorio Aguilar, Matt Rosinski, Bart Trudeau, Luis Fonseca, Dante Gates, Matt Niccolls, Maksim Kuznecov, Michael Thomas, Luke Gorrie, Cory Kiser and Julio.

Visit https://www.patreon.com/learnbayesstats to unlock exclusive Bayesian swag 😉

Takeaways:

 ⁃    Gravitational wave analysis involves using Bayesian statistics for parameter estimation and population inference.

    ⁃    Nested sampling is a powerful algorithm used in gravitational wave analysis to explore parameter space and calculate the evidence for model selection.

    ⁃    Machine learning techniques, such as normalizing flows, can be integrated with nested sampling to improve efficiency and explore complex distributions.

    ⁃    The LIGO-VIRGO collaboration operates gravitational wave detectors that measure distortions in space and time caused by black hole and neutron star collisions.

    ⁃    Sources of noise in gravitational wave detection include laser noise, thermal noise, seismic motion, and gravitational coupling.

    ⁃    The LISA mission is a space-based gravitational wave detector that aims to observe lower frequency gravitational waves and unlock new astrophysical phenomena.

    ⁃    Space-based detectors like LISA can avoid the ground-based noise and observe a different part of the gravitational wave spectrum, providing new insights into the universe.

    ⁃    The data analysis challenges for space-based detectors are complex, as they require fitting multiple sources simultaneously and dealing with overlapping signals.

    ⁃    Gravitational wave observations have the potential to test general relativity, study the astrophysics of black holes and neutron stars, and provide insights into cosmology.

Links from the show:

Transcript

This is an automatic transcript and may therefore contain errors. Please get in touch if you’re willing to correct them.

Transcript
Speaker:

In this episode, we dive deep into

gravitational wave astronomy with

2

00:00:08,540 --> 00:00:12,840

Christopher Berry and John Vich, two

senior lecturers at the University of

3

00:00:12,840 --> 00:00:16,270

Glasgow and experts from the LIGO -VIRGO

collaboration.

4

00:00:16,270 --> 00:00:20,360

They explain the significance of detecting

gravitational waves, which are essential

5

00:00:20,360 --> 00:00:23,710

for understanding black holes and neutron

stars collisions.

6

00:00:23,710 --> 00:00:27,980

This research not only sheds light on

these distant events, but also helps us

7

00:00:27,980 --> 00:00:28,494

grasp

8

00:00:28,494 --> 00:00:31,154

fundamental workings of the universe.

9

00:00:31,154 --> 00:00:36,114

Our discussion focuses on the integral

role of Bayesian statistics, detailing how

10

00:00:36,114 --> 00:00:40,604

they use nested sampling for extracting

crucial information from the subtle

11

00:00:40,604 --> 00:00:42,814

signals of gravitational waves.

12

00:00:42,814 --> 00:00:47,524

This approach is vital for parameter

estimation and understanding the

13

00:00:47,524 --> 00:00:51,974

distribution of cosmic sources through

population inferences.

14

00:00:51,974 --> 00:00:56,780

Concluding the episode, Christopher and

John highlight the latest advancements,

15

00:00:56,780 --> 00:01:02,470

in black hole astrophysics and tests of

general relativity, and touch upon the

16

00:01:02,470 --> 00:01:09,210

exciting prospects and challenges of the

upcoming space -based LISA mission.

17

00:01:09,250 --> 00:01:16,170

So strap on for episode 101 of Learning

Bayesian Statistics, recorded February 14,

18

00:01:16,170 --> 00:01:17,450

2024.

19

00:01:20,990 --> 00:01:24,230

Hello my dear Bayesians!

20

00:01:24,230 --> 00:01:26,574

Today, I want to thank Julio

21

00:01:26,574 --> 00:01:29,914

joining the Good Basion tier of the show's

Patreon.

22

00:01:29,954 --> 00:01:34,374

Julio, your support is invaluable and

literally makes this show possible.

23

00:01:34,374 --> 00:01:39,354

I really hope that you will enjoy the

exclusive sticker coming your way very

24

00:01:39,354 --> 00:01:39,994

soon.

25

00:01:39,994 --> 00:01:43,194

Make sure to post a picture in the slide

channel.

26

00:01:43,214 --> 00:01:46,374

And now, on to the show.

27

00:01:49,974 --> 00:01:54,614

Christopher Barry, John Vich, welcome to

Learning Basion Statistics.

28

00:01:55,274 --> 00:01:56,852

Thank you very much for having us.

29

00:01:57,902 --> 00:02:07,732

Yes, thank you a lot for taking the time,

even more time than listeners suspect, but

30

00:02:07,732 --> 00:02:10,842

we're not gonna expand on that.

31

00:02:11,442 --> 00:02:18,642

But yeah, I'm super happy to have you on

the show and we're gonna talk about a lot

32

00:02:18,642 --> 00:02:23,842

of things, physics, of course

astrophysics, black holes and so on.

33

00:02:23,842 --> 00:02:25,062

But first,

34

00:02:25,966 --> 00:02:31,386

How would you both define the work you're

doing nowadays and how did you end up

35

00:02:31,386 --> 00:02:33,004

working on this?

36

00:02:35,534 --> 00:02:37,334

I can go first.

37

00:02:37,334 --> 00:02:39,654

I guess I'm slightly older than

Christopher.

38

00:02:39,734 --> 00:02:44,694

I started doing gravitational waves when I

was a physics student at Glasgow.

39

00:02:45,294 --> 00:02:50,474

I got involved with the LIGO, actually the

GEO experiment first of all, which is the

40

00:02:50,474 --> 00:02:52,774

Gravitational Weight Detector in Germany.

41

00:02:52,874 --> 00:02:57,104

Its bigger brother is the LIGO and the

LIGO detectors that we're going to talk

42

00:02:57,104 --> 00:02:59,154

about more today.

43

00:03:00,734 --> 00:03:03,754

And ever since then, I mean, thought the

project was fantastic.

44

00:03:03,754 --> 00:03:04,974

I'll you all about it.

45

00:03:04,974 --> 00:03:14,314

I just wanted to get involved in the

discoveries of gravitational waves and

46

00:03:14,314 --> 00:03:17,474

what they can tell us about black holes

and so on.

47

00:03:18,574 --> 00:03:22,174

I got involved back in my PhD.

48

00:03:22,194 --> 00:03:27,194

My PhD was largely about gravitational

waves we could maybe detect in the future

49

00:03:27,194 --> 00:03:33,514

with an upcoming space -based mission

called LISA, due for launch in the 2030s.

50

00:03:33,514 --> 00:03:34,350

I remember

51

00:03:34,350 --> 00:03:36,830

my advisor telling me, I hope you're OK.

52

00:03:36,830 --> 00:03:38,350

There's not going to be any real data.

53

00:03:38,350 --> 00:03:39,580

And I was like, yes, that's great.

54

00:03:39,580 --> 00:03:42,510

I just want to play around with the theory

stuff.

55

00:03:43,090 --> 00:03:45,290

And then I guess fate conspired against

me.

56

00:03:45,290 --> 00:03:48,030

After my PhD, I moved to the University of

Birmingham.

57

00:03:48,030 --> 00:03:50,120

That's where I first started working with

John.

58

00:03:50,120 --> 00:03:51,810

We were at University of Birmingham

together.

59

00:03:51,810 --> 00:03:55,830

And I got involved in LIGO, VEGO data

analysis.

60

00:03:55,830 --> 00:03:59,000

And we happened to make our first

detection just a couple of years after I

61

00:03:59,000 --> 00:04:01,710

joined in 2015.

62

00:04:01,710 --> 00:04:06,230

And we've been very busy since then

analyzing all the signals, figuring out

63

00:04:06,230 --> 00:04:08,330

the astrophysics of them.

64

00:04:08,330 --> 00:04:12,930

So each individual source and then putting

them together to understand the population

65

00:04:12,930 --> 00:04:14,570

underneath.

66

00:04:14,570 --> 00:04:19,610

So now we're both at the University of

Glasgow working on analyzing these

67

00:04:19,610 --> 00:04:24,440

gravitational wave signals and

understanding what they can teach us about

68

00:04:24,440 --> 00:04:25,530

the universe.

69

00:04:25,910 --> 00:04:26,810

Yeah.

70

00:04:27,230 --> 00:04:30,570

So as Liesner can already tell, I guess,

71

00:04:30,670 --> 00:04:36,590

Fascinating topics, lots of things to talk

about and dive into.

72

00:04:36,970 --> 00:04:44,970

But maybe to give us a preview of things

we're going to talk about a bit more.

73

00:04:45,970 --> 00:04:53,130

You guys are also using some patient stats

writing in these analysis, am I right?

74

00:04:54,957 --> 00:05:02,057

Yeah, so I think we look at, I guess, two

levels of Bayesian stats.

75

00:05:02,057 --> 00:05:05,517

So the first is what we refer to as

parameter estimation.

76

00:05:05,517 --> 00:05:11,477

So given a single signal trying to figure

out what are the properties of the source.

77

00:05:11,477 --> 00:05:16,817

So the signals we most often see are, say,

two black holes spiraling in together.

78

00:05:16,817 --> 00:05:20,897

So we look at the patterns of

gravitational waves that it emits.

79

00:05:20,897 --> 00:05:24,205

And from this, we can match templates and

then infer.

80

00:05:24,205 --> 00:05:26,045

These are the masses of the two black

holes.

81

00:05:26,045 --> 00:05:29,905

This is the orientation of binary, the

distance to the binary, and parameters

82

00:05:29,905 --> 00:05:30,265

like that.

83

00:05:30,265 --> 00:05:36,855

So we use Bayesian stats and the sampling

algorithms like nested sampling to mop out

84

00:05:36,855 --> 00:05:39,045

a posterior probability distribution.

85

00:05:39,245 --> 00:05:42,915

And then I guess the second level of this,

we do what we call a population inference,

86

00:05:42,915 --> 00:05:48,525

so a hierarchical inference of given an

ensemble of different detections,

87

00:05:48,525 --> 00:05:52,735

correcting for our selection effects that

we can detect some signals easier than

88

00:05:52,735 --> 00:05:53,585

others.

89

00:05:53,613 --> 00:05:55,973

What is the underlying astrophysical

distribution?

90

00:05:55,973 --> 00:06:01,353

So what is the distribution of masses of

black holes out there in the universe?

91

00:06:02,113 --> 00:06:02,973

Yeah.

92

00:06:03,013 --> 00:06:05,193

So fascinating things.

93

00:06:05,233 --> 00:06:09,033

And John, you want to maybe add something

to that?

94

00:06:09,033 --> 00:06:13,203

Just as a teaser, we're going to dive a

bit later in the episode into what you

95

00:06:13,203 --> 00:06:14,533

guys actually do.

96

00:06:14,533 --> 00:06:20,073

So as Christopher just said, nested

sampling, population inferences, but

97

00:06:20,073 --> 00:06:21,709

anything you want to add?

98

00:06:21,709 --> 00:06:22,539

teaser for Easter.

99

00:06:22,539 --> 00:06:27,949

I would add something about the background

of how it works within the LIGO scientific

100

00:06:27,949 --> 00:06:28,589

collaboration.

101

00:06:28,589 --> 00:06:36,639

So when I started doing my PhD, my

advisor, Graham Wohn, taught me about

102

00:06:36,639 --> 00:06:40,089

Bayesian statistics, Bayesian inference,

and I never learned it as an undergraduate

103

00:06:40,089 --> 00:06:41,069

at all.

104

00:06:41,069 --> 00:06:45,009

I just leave my mind, like, here we have

this mathematical theory of learning.

105

00:06:45,009 --> 00:06:46,829

Why are we using this everywhere?

106

00:06:46,889 --> 00:06:51,021

And in those days, it really wasn't being

used very much in LIGO.

107

00:06:51,021 --> 00:06:53,971

because a lot of the people that started

the collaboration were coming from a

108

00:06:53,971 --> 00:06:57,121

physical perspective and they were very

frequentists.

109

00:06:57,121 --> 00:07:02,441

They were counting and cutting all of

their events to try and measure the

110

00:07:02,441 --> 00:07:04,701

discovery that takes place on it or

whatever.

111

00:07:04,761 --> 00:07:08,041

So it was kind of novel in that patient's

way back then.

112

00:07:08,041 --> 00:07:12,851

But since then, as Christopher said, it's

been applied all over the place to all

113

00:07:12,851 --> 00:07:14,161

kinds of different problems.

114

00:07:14,161 --> 00:07:17,401

So it's been quite exciting to watch that

back over the years.

115

00:07:18,301 --> 00:07:20,877

I remember we had a...

116

00:07:20,877 --> 00:07:25,027

We had our first detection and we were

lighting up our results.

117

00:07:25,027 --> 00:07:28,607

And I think at that time still a lot of

the collaboration was very frequent.

118

00:07:28,607 --> 00:07:33,207

So we were writing in our papers, we have

a posterior probability distribution for

119

00:07:33,207 --> 00:07:36,197

the masses and there are people going,

hey, what's that?

120

00:07:36,197 --> 00:07:37,037

What's a posterior?

121

00:07:37,037 --> 00:07:38,327

We've never come across this before.

122

00:07:38,327 --> 00:07:41,057

Can you explain it to us?

123

00:07:41,057 --> 00:07:43,977

And now it is very much accepted.

124

00:07:44,917 --> 00:07:47,537

And yeah, everyone has a new detector.

125

00:07:47,537 --> 00:07:48,537

Where are the masses?

126

00:07:48,537 --> 00:07:50,477

I want to see this probability

distribution.

127

00:07:50,477 --> 00:07:58,177

What do you think drove that evolution and

that change?

128

00:07:58,677 --> 00:08:02,497

I think Bayesian statistics is very

popular in other parts of astronomy.

129

00:08:02,497 --> 00:08:07,987

So in a sense, it was kind of inevitable

that it would make its way over to

130

00:08:07,987 --> 00:08:11,597

gravitational wave astronomy as it's only

a matter of time.

131

00:08:11,757 --> 00:08:18,797

But I think the problems that we were

trying to solve, particularly for the

132

00:08:18,797 --> 00:08:20,077

parameter estimation,

133

00:08:20,077 --> 00:08:25,507

type analysis did lend itself to a

Bayesian analysis because you have a

134

00:08:25,507 --> 00:08:26,217

unique event.

135

00:08:26,217 --> 00:08:30,717

You you're not, we only see a very small

number of gravitational waves.

136

00:08:30,957 --> 00:08:35,137

We see them all the time, but it's still

measurable.

137

00:08:36,457 --> 00:08:38,977

The dozens, not the millions.

138

00:08:39,237 --> 00:08:42,837

So we have to make the most of every

single one.

139

00:08:43,117 --> 00:08:45,897

The second one, the ratio is rather low.

140

00:08:46,657 --> 00:08:49,453

So graphs from the other side.

141

00:08:49,453 --> 00:08:53,793

is also very important if you want to do

science.

142

00:08:54,153 --> 00:08:57,613

Yeah, that makes sense.

143

00:08:57,613 --> 00:09:05,813

And so it was mainly driven by just

patient stats entering a need in what you

144

00:09:05,813 --> 00:09:15,073

wanted to do basically, which is something

I often see in fields where psychology,

145

00:09:15,821 --> 00:09:21,011

from what I've seen in the last few years,

for instance, psychometrics, things like

146

00:09:21,011 --> 00:09:25,011

that, have seen a big rise in patient

statistics because they have been able to

147

00:09:25,011 --> 00:09:28,981

answer the questions that researchers had

and that they could not answer with the

148

00:09:28,981 --> 00:09:30,501

tools they had before.

149

00:09:30,581 --> 00:09:36,361

So basically, a very practical, oriented

view of things.

150

00:09:36,361 --> 00:09:41,901

And then afterwards, let's say the more

epistemological philosophical side of

151

00:09:41,901 --> 00:09:44,861

things enters to also justify that.

152

00:09:44,861 --> 00:09:45,195

But...

153

00:09:45,195 --> 00:09:52,735

But most of the time it's a very practical

driven mindset, which is great, right?

154

00:09:52,735 --> 00:09:57,775

Because in the end, why you care about

that is just, is that the right tool to

155

00:09:57,775 --> 00:10:01,125

answer the questions I have right now?

156

00:10:01,645 --> 00:10:03,665

And yeah, for what it's worth.

157

00:10:04,085 --> 00:10:05,105

Yeah, go ahead, John.

158

00:10:05,105 --> 00:10:11,505

The pragmatism is what's put in the table

at the end of the day, but during my PhD,

159

00:10:11,505 --> 00:10:12,653

I was trying to look...

160

00:10:12,653 --> 00:10:16,823

or not the kind of black hole binaries

that we'll talk about later, but from

161

00:10:16,823 --> 00:10:18,613

monochromatic waves.

162

00:10:19,133 --> 00:10:22,993

So you imagine doing a Fourier transform

of some data and you have a single spike,

163

00:10:22,993 --> 00:10:25,053

and it has a bit of modulation on it.

164

00:10:25,313 --> 00:10:30,333

But really there's no information about

that spike in any area of the prime space

165

00:10:30,333 --> 00:10:31,973

outside of the spike.

166

00:10:32,093 --> 00:10:36,013

So I learned about Bayesian statistics and

tried to do MTMC on this problem, which

167

00:10:36,013 --> 00:10:40,163

was kind of like the most pathological

problem that you'd be trying to do MTMC

168

00:10:40,163 --> 00:10:40,941

on.

169

00:10:40,941 --> 00:10:44,921

that basically reduces itself to doing an

exhaustive search for the ground or space.

170

00:10:44,921 --> 00:10:50,441

So I was kind of convinced by the

epistemology originally rather than the

171

00:10:50,441 --> 00:10:56,881

thesis and the only nature that we used

for the sake.

172

00:10:57,621 --> 00:10:59,001

Yeah, yeah.

173

00:10:59,101 --> 00:10:59,821

Yeah, yeah.

174

00:10:59,821 --> 00:11:01,401

That makes sense.

175

00:11:01,401 --> 00:11:09,471

And as you were saying also that patient

studies is popular in other parts of

176

00:11:09,471 --> 00:11:10,125

physics.

177

00:11:10,125 --> 00:11:16,405

That's definitely true in a sense that,

for instance, in the core developers of

178

00:11:16,405 --> 00:11:23,725

MC, of Stan, you have a lot of physicists,

often coming from statistical physics and

179

00:11:23,725 --> 00:11:28,825

historically even the algorithms that we

even use, MCMC algorithms, have been

180

00:11:28,825 --> 00:11:32,825

developed mainly by physicists or for

physics purposes.

181

00:11:32,825 --> 00:11:37,019

So there is really this integration here

almost historically.

182

00:11:37,069 --> 00:11:43,689

And that made me think that if listeners

are interested, there is an interesting

183

00:11:43,689 --> 00:11:46,449

package that's called Exoplanet.

184

00:11:46,449 --> 00:11:52,209

And that's basically a toolkit for

probabilistic modeling of time series data

185

00:11:52,209 --> 00:11:56,889

in astronomy, but with a focus on

observations of exoplanets.

186

00:11:56,889 --> 00:12:02,539

So that's different from what you guys do,

but that's using PIMC as a backend.

187

00:12:02,539 --> 00:12:04,889

So that's why I know it.

188

00:12:05,049 --> 00:12:05,989

And that's...

189

00:12:06,157 --> 00:12:09,997

mainly developed by Dan, Firm and Macky,

if I remember correctly.

190

00:12:09,997 --> 00:12:16,017

I'll put that in the show notes for people

who are interested because that is

191

00:12:16,017 --> 00:12:21,537

definitely something to check out if you

are doing that kind of models.

192

00:12:21,897 --> 00:12:29,817

And that made me think that I didn't even

thank our matchmaker because today is

193

00:12:29,817 --> 00:12:36,017

February 14th, but actually this episode

was made possible thanks to a matchmaker,

194

00:12:36,245 --> 00:12:41,025

Cupid, if you want, of patient statistics,

Johnny Highland.

195

00:12:41,025 --> 00:12:45,925

Thanks a lot for putting me in contact

with Christopher and John.

196

00:12:45,985 --> 00:12:52,285

Johnny is a faithful listener and I am

very grateful for that and for putting me

197

00:12:52,285 --> 00:12:54,565

in contact with today's guests.

198

00:12:55,865 --> 00:13:02,415

And so you mentioned already, Christopher,

that you two work on the LIGO -VIRGO

199

00:13:02,415 --> 00:13:03,645

collaboration.

200

00:13:04,185 --> 00:13:05,069

Maybe...

201

00:13:05,069 --> 00:13:08,679

Yeah, tell us a bit more about that

collaboration, what that is about, and

202

00:13:08,679 --> 00:13:13,409

what the goal is, so that listeners have a

clear background.

203

00:13:13,409 --> 00:13:15,789

And then we'll dive into the details.

204

00:13:16,969 --> 00:13:20,989

So yes, LIGO -VIRGO -CAGRA is a

collaboration of collaborations.

205

00:13:20,989 --> 00:13:25,939

So each of LIGO -VIRGO and CAGRA operate

their own gravitational wave detectors.

206

00:13:25,939 --> 00:13:30,249

So these are remarkable experimental

achievements.

207

00:13:30,249 --> 00:13:33,749

We're talking devices that can measure

208

00:13:35,423 --> 00:13:38,193

Distortions in space and time is what

we're looking for.

209

00:13:38,193 --> 00:13:42,023

So in effect, what we do is we time how

long it takes a laser to bounce up and

210

00:13:42,023 --> 00:13:45,533

down between some mirrors in one direction

compared to another.

211

00:13:45,533 --> 00:13:49,973

We're looking for a part of less than one

part in 10 to the 21.

212

00:13:49,973 --> 00:13:54,233

So it's equivalent to measuring the

distance between the Earth and the sun to

213

00:13:54,233 --> 00:13:58,583

the diameter of a hydrogen atom, or the

distance from here to Alpha Centauri to

214

00:13:58,583 --> 00:14:00,313

the width of a human hair.

215

00:14:00,313 --> 00:14:02,473

So over many decades,

216

00:14:02,917 --> 00:14:06,257

experimentalists have developed the

techniques to build these detectors to

217

00:14:06,257 --> 00:14:07,377

design them.

218

00:14:07,377 --> 00:14:11,967

And we're now in a very fortunate

situation that we have multiple of these

219

00:14:11,967 --> 00:14:13,577

detectors operating across the world.

220

00:14:13,577 --> 00:14:18,677

So we have two LIGO detectors in the US,

one in Livingston, Louisiana, one in

221

00:14:18,677 --> 00:14:21,017

Washington, in Hanford.

222

00:14:21,017 --> 00:14:28,317

And we've got Virgo in Italy, just outside

Pisa, Kagra underground in Japan, and

223

00:14:28,317 --> 00:14:31,395

coming decade another LIGO to be built in

India.

224

00:14:31,885 --> 00:14:35,625

And each of these observatories is looking

for gravitational wave signals.

225

00:14:35,625 --> 00:14:40,525

The ideal source for gravitational waves

would be a binary of two black holes or

226

00:14:40,525 --> 00:14:45,225

two neutron stars, very dense objects

coming together, merging very quickly,

227

00:14:45,225 --> 00:14:48,105

very strong gravity, very dynamical

objects.

228

00:14:48,105 --> 00:14:52,305

And we can detect these gravitational

waves and with those do astronomy.

229

00:14:52,345 --> 00:14:57,285

So instead of using a telescope to make

observations with light, we're using these

230

00:14:57,285 --> 00:15:01,485

gravitational wave detectors to look for

gravitational waves in undercover.

231

00:15:01,485 --> 00:15:04,185

the astrophysics of these sources.

232

00:15:04,845 --> 00:15:05,165

Yeah.

233

00:15:05,165 --> 00:15:05,505

Yeah.

234

00:15:05,505 --> 00:15:05,795

Thanks.

235

00:15:05,795 --> 00:15:07,895

So that's a very clear explanation.

236

00:15:07,895 --> 00:15:13,995

It's a bit like being able to hear the

universe itself only looking at it, right?

237

00:15:13,995 --> 00:15:20,095

So that's another way of getting

information about the universe that maybe

238

00:15:20,095 --> 00:15:26,445

allows us to also answer questions that we

had, but we were not able to answer only

239

00:15:26,445 --> 00:15:28,065

with a telescope data.

240

00:15:28,065 --> 00:15:30,871

Is that the case or is that mainly

241

00:15:30,871 --> 00:15:34,661

information that's parallel and similar.

242

00:15:34,661 --> 00:15:35,041

Yes.

243

00:15:35,041 --> 00:15:36,881

Go ahead.

244

00:15:37,481 --> 00:15:40,581

Yeah, I think that's one of the most

exciting things about this is completely

245

00:15:40,581 --> 00:15:49,151

set in the electron spectrum using the

structural squeezing space itself by these

246

00:15:49,151 --> 00:15:50,621

buckles and neutrons.

247

00:15:52,021 --> 00:15:55,721

The waves that we've been offering are of

an oil from the sea.

248

00:15:55,721 --> 00:15:58,981

So, as you said, the detectors are picking

up

249

00:15:59,405 --> 00:16:04,505

essentially the equivalent of sound waves,

bulk motion of the material rather than

250

00:16:04,505 --> 00:16:06,105

the jiggling of atoms.

251

00:16:06,105 --> 00:16:10,985

We're talking about the jiggling of whole

stars, movement of them in their orbits.

252

00:16:11,585 --> 00:16:15,665

And because you're looking at the bulk

motion rather than the surface of the

253

00:16:15,665 --> 00:16:21,115

object, you can see right into the heart

of what's going on in some of these very

254

00:16:21,115 --> 00:16:22,505

violent events.

255

00:16:23,765 --> 00:16:27,145

In principle, we should be able to see

also inside supernovae if there are

256

00:16:27,145 --> 00:16:27,725

enough...

257

00:16:27,725 --> 00:16:30,545

motion of material during the core

collapse.

258

00:16:30,545 --> 00:16:34,485

That would also give off gravitational

waves that we could see, although their

259

00:16:34,485 --> 00:16:37,505

thoughts would be much weaker than those

that we're looking at in the moment.

260

00:16:40,909 --> 00:16:41,389

I see.

261

00:16:41,389 --> 00:16:46,339

One thing that's particularly nice we can

do as well is really test how gravity

262

00:16:46,339 --> 00:16:49,689

behaves in very extreme environments.

263

00:16:50,749 --> 00:16:53,449

John, I don't know if you want to mention

something about looking at the ring down

264

00:16:53,449 --> 00:16:55,149

of black holes.

265

00:16:55,869 --> 00:16:56,199

Sure.

266

00:16:56,199 --> 00:17:03,659

I mean, as Christopher says, there's a

very detailed prediction for how two stars

267

00:17:03,659 --> 00:17:06,709

should approach each other in their own

spiral over time.

268

00:17:07,809 --> 00:17:10,317

And the equations are horrendously

complicated.

269

00:17:10,317 --> 00:17:13,377

talking about the full view of general

relativity.

270

00:17:14,017 --> 00:17:18,677

But once they've collided and they form a

larger black hole, suddenly everything

271

00:17:18,677 --> 00:17:25,077

becomes rather simple and acts just like a

wine glass that's been excited with a fork

272

00:17:25,077 --> 00:17:30,917

and then it actually decays down and

settles into its final state.

273

00:17:30,917 --> 00:17:34,547

Therefore a black hole that happens

extremely fast because they want to

274

00:17:34,547 --> 00:17:37,997

settle down as quickly as they possibly

can, if you like.

275

00:17:38,257 --> 00:17:42,447

So the notes that we give off are

milliseconds long rather than seconds

276

00:17:42,447 --> 00:17:43,437

long.

277

00:17:43,497 --> 00:17:48,317

But the frequencies in the damping times

of those notes are measurable with their

278

00:17:48,317 --> 00:17:49,337

picture waves.

279

00:17:49,577 --> 00:17:53,757

And by looking at them and comparing them

to each other, we can check to see that

280

00:17:53,757 --> 00:17:59,897

the predictions of the theory are indeed

what we would see in the world.

281

00:18:00,317 --> 00:18:03,689

So far, they seem to be the case.

282

00:18:03,853 --> 00:18:05,213

Yeah.

283

00:18:08,373 --> 00:18:13,613

Something I'm wondering is that these

collisions that you're talking about, they

284

00:18:13,613 --> 00:18:17,593

are happening millions of light -wears

away.

285

00:18:18,153 --> 00:18:24,413

How are we even able to study them and

also maybe tell us what we already have

286

00:18:24,413 --> 00:18:25,955

learned from them?

287

00:18:27,821 --> 00:18:32,621

They're really quite rare events, the

types of collisions that we're seeing.

288

00:18:32,621 --> 00:18:37,001

I mean, this is why there are millions,

hundreds of millions or even billions of

289

00:18:37,001 --> 00:18:43,201

light years away is because they're so

rare in the universe that we need to look

290

00:18:43,201 --> 00:18:50,061

out a very long way before we see one

often enough to make the detections often.

291

00:18:50,061 --> 00:18:54,981

So they do happen in local galaxies as

well as the reason to think it wouldn't,

292

00:18:54,981 --> 00:18:57,067

but it's just they're so rare.

293

00:18:58,763 --> 00:19:00,953

I've seen one near black source.

294

00:19:00,973 --> 00:19:01,833

Yeah.

295

00:19:02,273 --> 00:19:06,093

Yes, they are remarkably energetic.

296

00:19:06,453 --> 00:19:11,253

The amount of energy that is output as

gravitational waves when you've got, say,

297

00:19:11,253 --> 00:19:14,233

two black holes coming together is

phenomenal.

298

00:19:14,233 --> 00:19:21,103

For just that moment as they smash in

together, more energy, so the luminosity,

299

00:19:21,103 --> 00:19:26,113

the amount of energy per unit time emitted

right at that peak is higher in

300

00:19:26,113 --> 00:19:28,237

gravitational waves than if you were to

add up.

301

00:19:28,237 --> 00:19:32,197

or the visible light from all the stars

that you could see in the universe.

302

00:19:32,237 --> 00:19:36,177

So it's a phenomenal amount of energy just

over a very short way.

303

00:19:36,177 --> 00:19:41,507

So yeah, we just need to be listening to

the universe to see these, to discover

304

00:19:41,507 --> 00:19:45,037

these sources and find out what they're

trying to tell us.

305

00:19:45,217 --> 00:19:49,697

The energy flux from these black hole

collisions, despite the fact that they're

306

00:19:49,697 --> 00:19:53,577

hundreds of millions of light years away,

is actually comparable to the flux from

307

00:19:53,577 --> 00:19:54,857

the full moon.

308

00:19:54,977 --> 00:19:58,001

So the brightest object in the night sky,

309

00:19:58,001 --> 00:20:02,681

is surpassed by gravitational wave

signals, except we can't see the

310

00:20:02,681 --> 00:20:06,901

gravitational waves because they don't

interact very strongly with matter.

311

00:20:07,601 --> 00:20:12,281

And it's only by building these incredibly

sensitive detectors to pick up their

312

00:20:12,281 --> 00:20:16,181

effect on distances that we can still look

at.

313

00:20:17,221 --> 00:20:25,421

Yeah, that's just fascinating to me that

we're even able to see...

314

00:20:25,421 --> 00:20:29,141

like hear these waves in a way.

315

00:20:29,421 --> 00:20:34,141

So, yeah, just to finally point home,

there's so much energy that you're

316

00:20:34,141 --> 00:20:39,701

carrying away, but the effect is so tiny,

as Chris said, 10 to the minus 21, no

317

00:20:39,701 --> 00:20:40,461

less.

318

00:20:40,461 --> 00:20:41,001

Yeah.

319

00:20:41,001 --> 00:20:44,921

If you think about how those two things

could be true at once, it's telling you

320

00:20:44,921 --> 00:20:49,801

that it takes an enormous amount of energy

to produce a tiny distortion in space.

321

00:20:49,801 --> 00:20:53,061

So it's very, very difficult to walk

space.

322

00:20:53,061 --> 00:20:54,245

And that's...

323

00:20:54,273 --> 00:20:56,371

the consequences of general malpractice.

324

00:20:57,901 --> 00:20:58,781

Yeah.

325

00:20:59,521 --> 00:21:05,381

And then, so I think now it's a good time

for you to tell us.

326

00:21:05,781 --> 00:21:09,301

So maybe Christopher, you can tell us

that.

327

00:21:09,401 --> 00:21:15,541

How do you use patient stats to extract as

much information as possible from these

328

00:21:15,541 --> 00:21:17,481

tiny wave signals?

329

00:21:17,521 --> 00:21:20,341

How is base useful in this field?

330

00:21:20,341 --> 00:21:23,581

And how do you also actually do it?

331

00:21:24,441 --> 00:21:27,581

Are you able to use any...

332

00:21:27,927 --> 00:21:31,857

widespread open source packages or do you

have to write everything yourself?

333

00:21:31,857 --> 00:21:33,777

How does that work concretely?

334

00:21:34,577 --> 00:21:40,547

Yes, so for the type of sources we've been

seeing these binaries, we have predictions

335

00:21:40,547 --> 00:21:42,097

for what the signal should look like.

336

00:21:42,097 --> 00:21:48,117

So we have a template that is a function

of the parameters and we have a decent

337

00:21:48,117 --> 00:21:51,397

understanding of the properties of our

noise.

338

00:21:51,397 --> 00:21:56,689

So the data is a combination of the signal

plus some noise which you assume to be

339

00:21:56,689 --> 00:22:03,949

stationary over the short time scales that

we're analyzing and characterized such

340

00:22:03,949 --> 00:22:06,909

that the noise at individual frequencies

is uncorrelated.

341

00:22:06,909 --> 00:22:10,669

So if you like, you get your data,

transform it to the frequency domain,

342

00:22:10,769 --> 00:22:14,479

subtract out your template, you should be

just left with noise, which is Gaussian at

343

00:22:14,479 --> 00:22:15,929

each frequency bin.

344

00:22:16,169 --> 00:22:20,169

And so you have a lot of Gaussian

probabilities that you combine to get.

345

00:22:20,249 --> 00:22:22,789

So that gives us our likelihood.

346

00:22:22,789 --> 00:22:25,707

You map that out, you change your

parameters for your template.

347

00:22:25,707 --> 00:22:29,617

evaluate that at another point in

parameter space, map that out with your

348

00:22:29,617 --> 00:22:33,857

suitable prior, and you end up with your

posterior probability for a single event.

349

00:22:34,257 --> 00:22:37,667

The number of parameters that we're

typically dealing with is something like

350

00:22:37,667 --> 00:22:39,947

15 for typical binary.

351

00:22:39,947 --> 00:22:43,937

Maybe that goes up to 17 when we add in a

couple of extra ones, a few more if we're

352

00:22:43,937 --> 00:22:46,257

maybe looking at tests of general

relativity.

353

00:22:46,477 --> 00:22:52,327

So it's enough that exploring the

parameter space can't be just done by

354

00:22:52,327 --> 00:22:54,381

gridding it up and exploring it.

355

00:22:54,381 --> 00:22:57,461

We generally use some kind of stochastic

sampling algorithm.

356

00:22:57,461 --> 00:23:02,471

But it's not one of these problems, at

least yet, where we've got millions of

357

00:23:02,471 --> 00:23:04,981

parameters and it's a really high

parameter space.

358

00:23:05,601 --> 00:23:10,191

In terms of the algorithms that we use to

explore parameter space, we've got a long

359

00:23:10,191 --> 00:23:16,061

history of using MCMC and nested sampling

for these.

360

00:23:16,061 --> 00:23:18,701

And John's really the expert on this.

361

00:23:18,701 --> 00:23:20,421

So, John, do want to say some more about?

362

00:23:20,421 --> 00:23:22,027

We'll get to that, yeah.

363

00:23:24,941 --> 00:23:26,381

Oh, you are done?

364

00:23:26,381 --> 00:23:27,241

OK, perfect.

365

00:23:27,241 --> 00:23:30,281

So yeah, John, maybe if you can tell us.

366

00:23:30,941 --> 00:23:35,331

Yeah, maybe let's start with nested

sampling that you use a lot for your

367

00:23:35,331 --> 00:23:36,121

inferences.

368

00:23:36,121 --> 00:23:42,421

So can you talk about that, why that's

useful, and also why you end up using that

369

00:23:42,421 --> 00:23:44,661

a lot in your work?

370

00:23:44,781 --> 00:23:47,101

Which problem does that solve?

371

00:23:48,001 --> 00:23:53,261

So nested sampling is an alternative to

MCMC.

372

00:23:53,261 --> 00:23:54,221

I don't know if you're...

373

00:23:54,221 --> 00:23:58,001

listeners will all have encountered it

before.

374

00:23:58,201 --> 00:24:01,701

If you're a regular user of MCMC though,

it's definitely worth a look.

375

00:24:01,701 --> 00:24:08,281

It was invented in 2006 -2007 by John

Scaling.

376

00:24:08,281 --> 00:24:10,121

He was a physicist.

377

00:24:10,701 --> 00:24:16,351

The idea is that you're actually trying to

evaluate the evidence, the normalization

378

00:24:16,351 --> 00:24:21,921

constant of the posterior to allow you to

do model selection in a basic way.

379

00:24:22,541 --> 00:24:29,761

But as a by -product, it can generate

samples from the Bistidia as well.

380

00:24:30,501 --> 00:24:36,681

So this popped up around about the time

that I started a full stock position in

381

00:24:36,681 --> 00:24:41,621

Birmingham and thought, well, why don't we

give it a go and apply it to the problem

382

00:24:41,621 --> 00:24:43,181

of compact binaries.

383

00:24:43,361 --> 00:24:48,281

So at that point, there was no off -the

-shelf package available to do this.

384

00:24:48,281 --> 00:24:50,295

And so we had to create our own.

385

00:24:50,317 --> 00:24:53,397

That was all coded up in C for so time.

386

00:24:53,397 --> 00:24:55,077

It wasn't such a big thing.

387

00:24:55,337 --> 00:24:58,497

There was thousands of lines of code and

all that.

388

00:24:58,657 --> 00:25:04,117

But yeah, so the reason is that you might

prepare it for MCMC.

389

00:25:04,357 --> 00:25:08,817

People were trying to solve the same

problem with parallel tempered MCMC.

390

00:25:08,937 --> 00:25:13,687

The compact binary parameter space has a

fair amount of degeneracies, multiple

391

00:25:13,687 --> 00:25:17,357

data, and in amongst those modes.

392

00:25:17,357 --> 00:25:22,547

They make it difficult to sample the

waveforms are facilitated in a nonlinear

393

00:25:22,547 --> 00:25:23,657

problem.

394

00:25:23,937 --> 00:25:25,957

It can be quite complicated.

395

00:25:26,517 --> 00:25:32,377

Getting a decent exploration of the prior

was proving to be difficult for the MCMC.

396

00:25:32,797 --> 00:25:35,877

Hence the need for parallel tempering.

397

00:25:36,237 --> 00:25:39,187

And this is something that works a little

bit differently because it starts off by

398

00:25:39,187 --> 00:25:41,977

sampling the whole prior in the first

place.

399

00:25:42,017 --> 00:25:46,757

So you know, say thousands of points,

they're called live points.

400

00:25:47,033 --> 00:25:51,903

scatter them across the entire prior and

then compute the likelihood for every one

401

00:25:51,903 --> 00:25:52,933

of those.

402

00:25:53,093 --> 00:25:57,863

If you then eliminate the point that has

the lowest likelihood and replace it with

403

00:25:57,863 --> 00:26:02,683

one that has the higher likelihood of the

lowest one, then people still have a

404

00:26:02,683 --> 00:26:07,133

thousand points, so they will all have a

likelihood higher than the worst one.

405

00:26:07,533 --> 00:26:14,123

And you can see that, roughly speaking,

the volume of that remaining set of points

406

00:26:14,123 --> 00:26:15,533

will be about

407

00:26:15,533 --> 00:26:22,733

999 thousandths of the original one just

by random large numbers.

408

00:26:23,593 --> 00:26:28,453

And so if you repeat that process, always

replacing the point of the next iteration,

409

00:26:28,453 --> 00:26:34,733

you'll have 999 thousandths of 999

thousandths of the original.

410

00:26:34,733 --> 00:26:41,143

And so eventually you'll shrink in a

geometric fashion the volume that your

411

00:26:41,143 --> 00:26:43,813

points are contained within.

412

00:26:43,813 --> 00:26:44,929

And...

413

00:26:44,929 --> 00:26:49,629

In doing so, you're walking uphill, you're

moving towards the peak of the posterior.

414

00:26:49,989 --> 00:26:56,639

So, what I have seen to see it is

guaranteed to terminate once you have held

415

00:26:56,639 --> 00:26:59,749

the climb up, which was a nice feature.

416

00:27:00,029 --> 00:27:02,909

And it gives you the evidence for doing

multi -selection.

417

00:27:03,689 --> 00:27:08,329

Once you've done the entire chain, you can

resample those points from the chain and

418

00:27:08,329 --> 00:27:13,809

weight them according to the posterior to

produce either independent samples or

419

00:27:13,809 --> 00:27:14,285

weighted.

420

00:27:14,285 --> 00:27:17,585

posterior samples are to meet.

421

00:27:18,645 --> 00:27:21,185

Yeah, so it's a really effective

algorithm.

422

00:27:21,185 --> 00:27:23,325

I like it because it's reliable.

423

00:27:24,465 --> 00:27:26,735

And as I say, your run is guaranteed to

finish.

424

00:27:26,735 --> 00:27:30,225

It might take a long time, but it will get

there.

425

00:27:30,465 --> 00:27:33,885

There are of course, places where it falls

down.

426

00:27:33,885 --> 00:27:37,225

If you don't have an upline, you can end

up missing a mode.

427

00:27:37,965 --> 00:27:43,805

The challenge is really how do you explore

that constrained prior distribution.

428

00:27:44,301 --> 00:27:47,641

And so over the years, there have been

different approaches to doing that.

429

00:27:47,641 --> 00:27:53,301

The one that I started, I was coding in

Oracle Struct, was using MCMC inside the

430

00:27:53,301 --> 00:27:54,181

nested sampling.

431

00:27:54,181 --> 00:28:00,431

So just do a little MCMC chain to draw the

next sample, which works fine, especially

432

00:28:00,431 --> 00:28:04,361

because we already knew how to do MCMC's

for this problem quite well.

433

00:28:05,301 --> 00:28:12,261

But other people have invented the

ellipsoidal multiness algorithm, was one

434

00:28:12,261 --> 00:28:14,317

of the first very popular.

435

00:28:14,317 --> 00:28:18,277

off -the -shelf solutions and that was

used also for gravitational waves.

436

00:28:19,217 --> 00:28:26,377

These days there are more modern, I

packages that do everything you need,

437

00:28:26,477 --> 00:28:31,577

either with MCMC or with side sampling or

more complicated things like normalizing

438

00:28:31,577 --> 00:28:32,537

flows.

439

00:28:33,117 --> 00:28:39,447

I should mention the Bowman or most of the

gravitational wave using dynasty, which is

440

00:28:39,447 --> 00:28:42,277

next to sampling.

441

00:28:43,901 --> 00:28:47,481

myself and the students, there's no force

with it.

442

00:28:47,501 --> 00:28:51,661

That's the image that connects it

something with artificial intelligence

443

00:28:51,661 --> 00:28:57,101

that attempts to use some machine learning

to accelerate this whole process.

444

00:28:58,161 --> 00:28:59,741

Well, that sounds like fun.

445

00:28:59,741 --> 00:29:05,701

Yeah, I'm definitely going to link to

Genesty.

446

00:29:05,941 --> 00:29:11,085

So the package you're using right now to

do the NST sampling in the show notes and

447

00:29:11,085 --> 00:29:16,245

If you have anything you can share on this

new package you're working on, for sure,

448

00:29:16,245 --> 00:29:18,125

please add that to the channel.

449

00:29:18,245 --> 00:29:21,265

These listeners will be very interested.

450

00:29:21,625 --> 00:29:25,345

And maybe you want to add a bit more about

this project.

451

00:29:25,345 --> 00:29:33,115

So how would you use machine learning in

this way to help you do the nested

452

00:29:33,115 --> 00:29:33,985

sampling?

453

00:29:34,565 --> 00:29:36,305

Yeah, I can say something about that.

454

00:29:36,305 --> 00:29:38,545

It's a cool idea.

455

00:29:38,545 --> 00:29:40,683

I mean, the...

456

00:29:40,683 --> 00:29:45,213

Enabling technology for this is a tool

called the normalizing flow.

457

00:29:45,453 --> 00:29:49,433

And I don't know if you've talked about it

in podcast before, but they have a way of

458

00:29:49,433 --> 00:29:54,843

approximating complicated distributions

using single ones with a remapping of the

459

00:29:54,843 --> 00:29:56,113

coordinate system.

460

00:29:56,433 --> 00:30:05,023

So in that context, we were trying to make

a good fit to the jump proposal for the

461

00:30:05,023 --> 00:30:08,493

sample, if you like, because that has to

evolve.

462

00:30:08,493 --> 00:30:11,913

with the scale of the problem as the

nested sample proceeds.

463

00:30:11,953 --> 00:30:17,483

The mode shrinks and it can shrink by a

factor of 10 to the 20 over the course of

464

00:30:17,483 --> 00:30:18,473

the run.

465

00:30:18,473 --> 00:30:24,893

So you're going to need something adaptive

to continue to have good efficiency.

466

00:30:25,453 --> 00:30:31,033

So we took this normalizing flow technique

and applied it to this problem of fitting

467

00:30:31,033 --> 00:30:32,733

the existing samples.

468

00:30:32,733 --> 00:30:36,853

And then the advantage being that it

allows you to draw independent samples, a

469

00:30:36,853 --> 00:30:38,381

bit like the ellipsoidal.

470

00:30:38,381 --> 00:30:43,721

technique, but it doesn't require a fixed

shape.

471

00:30:43,721 --> 00:30:48,021

So it's able to make more complicated

shapes for distribution.

472

00:30:48,261 --> 00:30:54,121

Yeah, I'll pop the link in and people are

very welcome to give a go.

473

00:30:54,501 --> 00:30:55,741

Yeah, for sure.

474

00:30:56,661 --> 00:30:57,381

Yeah.

475

00:30:57,581 --> 00:31:00,321

So folks give it a go, try it.

476

00:31:00,321 --> 00:31:03,161

If you see issues, report them on the

GitHub, even better.

477

00:31:03,161 --> 00:31:07,461

If you can do a PR, I'm sure John will

appreciate it.

478

00:31:07,693 --> 00:31:16,293

And actually, so that could not be better

because I will refer people to episode 98

479

00:31:16,293 --> 00:31:24,923

of the podcast where I talked with Maridu

Gabriel, who is one of the persons

480

00:31:24,923 --> 00:31:27,693

developing these kinds of methods.

481

00:31:27,693 --> 00:31:29,503

And we talked exactly about that.

482

00:31:29,503 --> 00:31:34,933

Adaptive MCMC augmented with normalizing

flows.

483

00:31:35,013 --> 00:31:35,533

And we...

484

00:31:35,533 --> 00:31:38,983

talked in the episode about how it offers

a powerful approach, especially for

485

00:31:38,983 --> 00:31:45,003

sampling multimodal distributions, how it

also scales the algorithm to higher

486

00:31:45,003 --> 00:31:52,063

dimensions, how you can handle discrete

parameters, and how all these ongoing

487

00:31:52,063 --> 00:31:53,453

challenges in the field.

488

00:31:53,453 --> 00:31:58,773

So if you're interested in the nitty

gritty details of what John just

489

00:31:58,773 --> 00:32:03,253

mentioned, I recommend listening to

episode 98 because, well, Marilou is

490

00:32:03,253 --> 00:32:04,013

really a f***.

491

00:32:04,013 --> 00:32:06,633

One of the persons developing all that

stuff.

492

00:32:07,133 --> 00:32:08,973

Sounds super interesting Alex.

493

00:32:08,973 --> 00:32:12,893

I'm amazed at the power of some of these

new techniques.

494

00:32:12,893 --> 00:32:16,433

There's a revolution going on at the

moment in this area.

495

00:32:16,733 --> 00:32:18,913

It's a good time to be involved.

496

00:32:18,973 --> 00:32:20,753

Yeah, I know for sure.

497

00:32:20,833 --> 00:32:22,253

I will link to that.

498

00:32:22,253 --> 00:32:30,243

Also, Colin Carroll, who is one of the

PIMC developers, he also has a new

499

00:32:30,243 --> 00:32:34,125

package, well, working on a new package

called Biox.

500

00:32:34,125 --> 00:32:40,305

And I know that they implemented these

normalizing flow algorithm.

501

00:32:40,305 --> 00:32:47,685

And so now you can use that in PyMC

directly through BIOX and to use that kind

502

00:32:47,685 --> 00:32:54,145

of algorithm and handle your multi

-dimensional, multi -model distributions

503

00:32:54,145 --> 00:32:55,645

more easily.

504

00:32:55,645 --> 00:33:00,685

So I also link to that because it's

definitely super interesting if you have

505

00:33:00,685 --> 00:33:04,125

lots of weird distributions.

506

00:33:04,429 --> 00:33:06,029

Like that.

507

00:33:07,829 --> 00:33:14,539

And Christopher, to come back to you, you

also mentioned that you guys do population

508

00:33:14,539 --> 00:33:15,909

inferences.

509

00:33:16,329 --> 00:33:22,559

And that's hierarchical models where you

use a bunch of observations to infer the

510

00:33:22,559 --> 00:33:28,389

underlying distribution of the sources of

the signal, if I understood correctly.

511

00:33:28,389 --> 00:33:30,959

So what does that look like?

512

00:33:30,959 --> 00:33:33,315

What do you guys do here?

513

00:33:33,869 --> 00:33:40,289

Yeah, so we do the calculation in a couple

of stages that we always run the parameter

514

00:33:40,289 --> 00:33:44,969

estimation to get the events parameters

for just one signal of time first.

515

00:33:44,969 --> 00:33:50,519

And so the result of that is a set of

posterior samples calculated with a

516

00:33:50,519 --> 00:33:51,789

fiducial prior.

517

00:33:52,469 --> 00:33:58,369

And what we want to do is then divide out

that prior, put in a population model, see

518

00:33:58,369 --> 00:33:59,749

how well that fits.

519

00:33:59,749 --> 00:34:02,349

So calculate the, I guess, the evidence.

520

00:34:03,181 --> 00:34:07,361

under the assumption of a particular set

of hyperparameters.

521

00:34:07,361 --> 00:34:13,661

And then we have an inference one level up

where we vary the population parameters,

522

00:34:13,661 --> 00:34:20,751

the hyperparameters for the population

model, explore that to see what fits work.

523

00:34:20,751 --> 00:34:23,741

So that really is starting to get the

astrophysics.

524

00:34:23,741 --> 00:34:27,861

So looking at the distribution of masses,

are there more low mass black holes and

525

00:34:27,861 --> 00:34:28,771

high mass black holes?

526

00:34:28,771 --> 00:34:30,981

How does that scale?

527

00:34:30,981 --> 00:34:31,565

Is there a little?

528

00:34:31,565 --> 00:34:34,905

bumps in the distribution and things like

that.

529

00:34:35,405 --> 00:34:37,845

So, yeah, it's next level up.

530

00:34:37,845 --> 00:34:42,805

The likelihood isn't quite as expensive as

evaluating the waveforms, but we have some

531

00:34:42,805 --> 00:34:46,785

data handling issues of reading in order

of the posterior samples.

532

00:34:46,785 --> 00:34:51,205

And key to this is, as I alluded to, is

correcting for the selection effects so

533

00:34:51,205 --> 00:34:54,905

that we need to account for the fact that

with our gravitational wave detectors, we

534

00:34:54,905 --> 00:34:58,865

can preferentially see some sources over

other sources.

535

00:34:58,865 --> 00:35:00,941

So if you were just to look at our

536

00:35:00,941 --> 00:35:04,781

distribution of sources that we detect,

you'll see, hey, there are lots of 30

537

00:35:04,781 --> 00:35:09,671

solar mass black holes, there aren't too

many 10 solar mass black holes, and if you

538

00:35:09,671 --> 00:35:12,351

didn't know about our selection effects,

you can actually assume, okay, the

539

00:35:12,351 --> 00:35:16,601

universe is full of 30 solar mass black

holes, and 10 solar mass ones are much

540

00:35:16,601 --> 00:35:17,801

rarer.

541

00:35:18,081 --> 00:35:21,821

Whereas because our detectors are more

sensitive to the high mass signals, those

542

00:35:21,821 --> 00:35:25,331

are intrinsically louder, so we can see

them further away, we can see more of

543

00:35:25,331 --> 00:35:26,141

them.

544

00:35:26,301 --> 00:35:28,971

Once you correct for the selection

effects, you actually see it's the other

545

00:35:28,971 --> 00:35:30,509

way around, there are many more

546

00:35:30,509 --> 00:35:33,969

At least there should be many more 10

solar mass black holes than 30 solar mass

547

00:35:33,969 --> 00:35:34,869

black holes.

548

00:35:34,869 --> 00:35:39,379

And the fact we don't see so many 50 solar

mass black holes, 90 solar mass black

549

00:35:39,379 --> 00:35:43,289

holes, tells you that the distribution

does drop off quite rapidly.

550

00:35:44,049 --> 00:35:49,209

So this is a field that's growing quite

nicely as we get more and more detections.

551

00:35:49,209 --> 00:35:54,369

Your uncertainties on the population

basically go as the square root of the

552

00:35:54,369 --> 00:35:55,885

number of detections.

553

00:35:56,141 --> 00:36:02,881

what we're seeing a lot of work on is what

does one assume for the population model.

554

00:36:02,901 --> 00:36:08,201

So when we started off with, I guess,

following what is common in astronomy, we

555

00:36:08,201 --> 00:36:13,671

put a power law through for the masses,

just infer the power law index basically

556

00:36:13,671 --> 00:36:18,101

in the normalization for the overall rate

and see how that worked.

557

00:36:18,101 --> 00:36:19,881

Then we like that's a bit simplistic.

558

00:36:19,881 --> 00:36:21,961

Let's add in a couple more parameters.

559

00:36:21,961 --> 00:36:22,765

Let's have

560

00:36:22,765 --> 00:36:26,615

say a little peak, a Gaussian add on top

of that to get peak.

561

00:36:26,615 --> 00:36:29,765

Let's say have two parallels with the

break, see how those fit.

562

00:36:29,765 --> 00:36:31,885

Let's put in another peak.

563

00:36:31,885 --> 00:36:35,885

And now people are looking at semi

-parametric models.

564

00:36:35,885 --> 00:36:37,975

So OK, what if we add a spline on top of

that?

565

00:36:37,975 --> 00:36:39,225

See how we can vary that.

566

00:36:39,225 --> 00:36:47,075

Or what if we do something really

flexible, so allow a bunch of kernels to

567

00:36:47,075 --> 00:36:51,265

come together and further the population

to get out of there?

568

00:36:51,265 --> 00:36:52,429

So a lot of.

569

00:36:52,429 --> 00:36:58,969

A lot of the work at the moment is trying

to see what is a good fit for the data and

570

00:36:58,969 --> 00:37:00,619

then checking is this overly complex?

571

00:37:00,619 --> 00:37:01,569

Are we overfitting?

572

00:37:01,569 --> 00:37:03,069

Is there a little bump there?

573

00:37:03,069 --> 00:37:06,899

Is that just because of a pass on

fluctuation that we've only seen so many

574

00:37:06,899 --> 00:37:07,789

events?

575

00:37:07,789 --> 00:37:11,769

So a small number of statistics means

there's a few more here and a few fewer

576

00:37:11,769 --> 00:37:12,529

there.

577

00:37:12,529 --> 00:37:17,849

Or is there actually some feature of the

underlying population, which may be a hint

578

00:37:17,849 --> 00:37:19,935

to how stars are formed?

579

00:37:20,205 --> 00:37:24,125

I think it's quite an interesting time at

the moment from this testing out models,

580

00:37:24,125 --> 00:37:28,145

trying to determine do they fit the

observations quite well.

581

00:37:28,185 --> 00:37:32,045

And I'm very excited for getting the

results of our upcoming observing runs

582

00:37:32,045 --> 00:37:36,865

when we're having a much larger number of

detections and we'll really be able to

583

00:37:36,865 --> 00:37:40,365

constrain the models to higher accuracy

and precision.

584

00:37:41,605 --> 00:37:43,885

Yeah, so that's super interesting.

585

00:37:43,885 --> 00:37:50,085

And so here to understand what you're

doing, it's like your...

586

00:37:50,189 --> 00:37:57,469

hearing different sounds and you're trying

to infer not really what the sound is

587

00:37:57,469 --> 00:38:01,489

about, but what is emitting that sound?

588

00:38:01,489 --> 00:38:04,249

What is the source of that sound?

589

00:38:04,249 --> 00:38:10,809

And the issue is that these sounds can be

emitted by a lot of entities and a lot of

590

00:38:10,809 --> 00:38:15,849

these sources you don't really care about

because I know they are on earth, they are

591

00:38:15,849 --> 00:38:19,821

like, but what you're interested in are

the sources.

592

00:38:19,821 --> 00:38:27,821

outside, which are in space and which tell

you something about the universe, which

593

00:38:27,821 --> 00:38:32,181

here would be mainly neutron stars and

black holes colliding.

594

00:38:32,761 --> 00:38:36,341

How weird was that characterization?

595

00:38:37,421 --> 00:38:42,901

Yes, I guess maybe a nice analogy might

be, imagine you have a room full of people

596

00:38:42,901 --> 00:38:46,255

and you're trying to judge the composition

of the room.

597

00:38:46,349 --> 00:38:50,849

And some of the people there, you have a

bunch of librarians who are very quiet.

598

00:38:50,929 --> 00:38:56,629

And you have some heavy metal stars who

are very, very loud.

599

00:38:56,629 --> 00:39:01,159

And so you've made your recording of the

audio in the room, and then you need to

600

00:39:01,159 --> 00:39:02,509

try and reconstruct that.

601

00:39:02,509 --> 00:39:05,589

OK, I can only hear one librarian.

602

00:39:05,809 --> 00:39:09,099

But given that the librarians are very

quiet, there's probably a whole host of

603

00:39:09,099 --> 00:39:12,829

other librarians who I just missed because

they're being too quiet.

604

00:39:13,677 --> 00:39:18,497

and I can hear lots of electric guitars

going on, so I know there's some rock

605

00:39:18,497 --> 00:39:22,577

stars here, but I know they're very loud

and easy.

606

00:39:22,637 --> 00:39:30,097

I probably will have detected 100 % of

those, so correct for those bias from the

607

00:39:30,097 --> 00:39:30,837

detection.

608

00:39:30,837 --> 00:39:34,007

We're very fortunate actually in

gravitational wave detection that we can

609

00:39:34,007 --> 00:39:35,517

calculate our selection effects.

610

00:39:35,517 --> 00:39:40,537

It's quite easy for us to determine what

sources we can detect and what we can't.

611

00:39:40,537 --> 00:39:43,533

This is a standing problem in astronomy

that you're

612

00:39:43,533 --> 00:39:49,433

We only have one universe, so we need to

make sure we understand what we're seeing.

613

00:39:50,253 --> 00:39:53,213

And you can know what you detect, but it's

very hard to know what you're not

614

00:39:53,213 --> 00:39:54,173

detecting.

615

00:39:54,333 --> 00:39:57,593

So a lot of astronomy is trying to correct

for these.

616

00:39:58,033 --> 00:40:00,903

And if you have a telescope, that can be

very difficult because you've got to

617

00:40:00,903 --> 00:40:04,723

calculate, OK, not just what did I see,

but what could I have seen?

618

00:40:04,723 --> 00:40:06,803

So that would depend on where I was

pointing the telescope.

619

00:40:06,803 --> 00:40:10,813

It would depend on the weather on a

particular day and how cloudy it was.

620

00:40:10,829 --> 00:40:13,969

Whereas with our gravitational wave, it's

much simpler.

621

00:40:13,969 --> 00:40:18,489

What we do is we can inject the

terminology we use.

622

00:40:18,489 --> 00:40:22,879

We simulate signals, put those into our

data, run our detection pipelines on that,

623

00:40:22,879 --> 00:40:26,879

and see what fraction of the signals that

we injected would we recovered and from

624

00:40:26,879 --> 00:40:28,249

that work out.

625

00:40:28,249 --> 00:40:32,629

As a function of source parameters, what

was the probability that something was

626

00:40:32,629 --> 00:40:33,229

detected?

627

00:40:33,229 --> 00:40:39,341

And then use that in renormalizing our

likelihood to establish.

628

00:40:39,341 --> 00:40:43,571

Okay, how many of these sources should

have there have been given that we saw

629

00:40:43,571 --> 00:40:44,821

this money?

630

00:40:45,661 --> 00:40:51,521

Okay, it helps a lot that gravitational

waves are not blocked by anything in the

631

00:40:51,521 --> 00:40:56,621

universe that we know about except for

other black holes But even then other

632

00:40:56,621 --> 00:41:03,421

black holes tend to be very small So when

we are able to calculate exactly what the

633

00:41:03,421 --> 00:41:08,841

source is doing it means that we've got a

very good idea of what we will see.

634

00:41:09,101 --> 00:41:12,461

It doesn't really matter what's in the

entropy space.

635

00:41:12,461 --> 00:41:17,491

The two veins of astronomy are dust and

magnetic fields, and gravitational waves

636

00:41:17,491 --> 00:41:20,181

are just don't really care about any of

those two things.

637

00:41:22,361 --> 00:41:23,161

Yeah, okay.

638

00:41:23,161 --> 00:41:23,591

I see.

639

00:41:23,591 --> 00:41:25,661

And that's actually a good thing.

640

00:41:25,661 --> 00:41:30,841

Indeed, that's quite a luxury to be able

to compute your own selection bias.

641

00:41:30,841 --> 00:41:32,341

That's pretty amazing.

642

00:41:32,581 --> 00:41:37,561

Me, who've done a lot of political

science, you usually cannot do that, so

643

00:41:37,561 --> 00:41:38,899

I'm very jealous.

644

00:41:38,899 --> 00:41:44,109

And can you tell us actually where does

that noise come from?

645

00:41:44,109 --> 00:41:47,849

Because it seems like you're saying there

is a lot of noise in your observations.

646

00:41:47,849 --> 00:41:52,389

Thankfully, you are able to tame that

somewhat easily.

647

00:41:52,389 --> 00:41:55,009

Can you tell us a bit more about that?

648

00:41:55,149 --> 00:41:58,969

And John, it seems like you want to add

something about that.

649

00:41:59,029 --> 00:42:05,569

Most of the noise, all the noise is not of

extraterrestrial origin.

650

00:42:06,669 --> 00:42:10,869

It's coming from the detectors and coming

from the environment around the detectors.

651

00:42:10,969 --> 00:42:15,149

So in order to understand that you have to

know a little bit about how to light over

652

00:42:15,149 --> 00:42:16,669

a porp.

653

00:42:17,069 --> 00:42:23,299

So imagine a giant in all shape, four

kilometres long, in bits of light, with

654

00:42:23,299 --> 00:42:29,989

the letters at the ends of the arms

shining a laser into the coin, if like.

655

00:42:29,989 --> 00:42:34,569

It gets split into two and sent down both

arms, bounces off them into the end and

656

00:42:34,569 --> 00:42:35,745

then comes down.

657

00:42:36,045 --> 00:42:45,555

and if they aren't the same length then

the light will constructively interfere or

658

00:42:45,555 --> 00:42:47,525

destructively, I may have that wrongly

written.

659

00:42:47,525 --> 00:42:50,685

The point is if they aren't at different

lengths or if they're changing lengths

660

00:42:50,685 --> 00:42:55,585

then the pattern of the light that comes

out will change over time.

661

00:42:55,765 --> 00:43:02,525

So we are really worried about anything

that can change that output of the laser

662

00:43:02,525 --> 00:43:04,125

in the detector.

663

00:43:04,173 --> 00:43:07,133

And so that could be due to the laser

itself.

664

00:43:07,313 --> 00:43:10,393

All lasers have some noise in them.

665

00:43:10,913 --> 00:43:14,213

So the lasers that they use in these

detectors are some of the most stable

666

00:43:14,213 --> 00:43:15,653

lasers that you can use.

667

00:43:17,837 --> 00:43:21,157

have been invented from scratch basically

for this one.

668

00:43:21,577 --> 00:43:27,677

It could be the thermal motion of the

atoms in the matrix of the complex.

669

00:43:27,677 --> 00:43:33,197

It would be better in that, simply having

a wide enough laser beam approaching the

670

00:43:33,197 --> 00:43:41,917

whole surface of the metal, cancelling out

the mean motion to the low enough level to

671

00:43:41,917 --> 00:43:43,657

get it ready.

672

00:43:44,777 --> 00:43:46,769

But the laser also

673

00:43:47,277 --> 00:43:52,467

You know, there's energy and that energy

fishes on the mirrors of radiation, which

674

00:43:52,467 --> 00:43:55,337

causes the mirrors to move a little bit.

675

00:43:55,537 --> 00:44:01,777

And now, think about the algorithms, the

laser energy is carried by photons, which

676

00:44:01,777 --> 00:44:08,057

are ultimately quantum objects, so they

get off the radar distinctly.

677

00:44:08,217 --> 00:44:13,017

Kind of raindrops on the roof, if you

imagine, or if you're in a tent, you get

678

00:44:13,017 --> 00:44:14,257

raindrops of rain.

679

00:44:14,257 --> 00:44:15,297

That's kind of what it's like.

680

00:44:15,297 --> 00:44:16,845

The lasers are enormously hard.

681

00:44:16,845 --> 00:44:20,665

still they are made of individual photons.

682

00:44:20,825 --> 00:44:26,505

And so there's a shot noise associated

with them, just due to the statistical

683

00:44:26,505 --> 00:44:30,365

fluctuation in the number of photons that

are writing per second.

684

00:44:32,405 --> 00:44:38,595

Then we've got the environment as well,

which is especially dominant at low

685

00:44:38,595 --> 00:44:39,385

frequencies.

686

00:44:39,385 --> 00:44:45,695

So we can't sense anything below about 10

Hertz with these detectors that are above

687

00:44:45,695 --> 00:44:46,763

the ground.

688

00:44:46,965 --> 00:44:49,685

because of seismic motion.

689

00:44:49,925 --> 00:44:54,425

Now we do have a lot of techniques to try

and screen the mirrors out in the motion

690

00:44:54,425 --> 00:44:55,665

of the Earth.

691

00:44:55,665 --> 00:45:02,325

They're hung on suspended optics, which

act as a natural filter to prevent ground

692

00:45:02,325 --> 00:45:04,965

motion from propagating through to the

mirror.

693

00:45:05,025 --> 00:45:11,405

But even so, we need to have active

oscillation systems as well.

694

00:45:11,405 --> 00:45:15,485

And on top of all of that, even if you

manage to screen out all the mechanical

695

00:45:15,485 --> 00:45:16,621

coupling,

696

00:45:16,685 --> 00:45:21,135

There's unfortunately the gravitational

coupling that we can't spin out because we

697

00:45:21,135 --> 00:45:23,725

actually want to measure gravity in the

first place.

698

00:45:23,725 --> 00:45:29,205

So if you imagine a seismic wave as a

pressure wave in the rock, I mean, when

699

00:45:29,205 --> 00:45:32,645

pressure is high, the rock is actually

compressed slightly.

700

00:45:32,705 --> 00:45:36,385

And because it's compressed, it's denser

than average.

701

00:45:36,385 --> 00:45:42,105

And because it's denser than average, it

exerts a gravitational pull on the mirrors

702

00:45:42,105 --> 00:45:44,641

that tends to pull them along.

703

00:45:44,877 --> 00:45:46,577

with the seismic waves.

704

00:45:46,577 --> 00:45:50,697

So this tiny effect, I mean, you've

probably never even thought about it, but

705

00:45:50,697 --> 00:45:57,957

it's there as a small gravitational

coupling of seismic waves to the detector.

706

00:45:58,277 --> 00:46:01,317

And you can't really get around these

things tall on the earth.

707

00:46:01,317 --> 00:46:06,217

And so that's why one of the challenges

that we're working on at the moment is

708

00:46:06,217 --> 00:46:11,817

looking at sending a detector into space,

which is hopefully going to open up a

709

00:46:11,817 --> 00:46:13,477

whole new range of...

710

00:46:13,605 --> 00:46:15,587

objects for us to look at.

711

00:46:18,477 --> 00:46:25,057

Yeah, thanks a lot, That's definitely

clear, and I didn't have, indeed, any idea

712

00:46:25,057 --> 00:46:32,117

of all these sources of noise, which is

pretty incredible that we're able to

713

00:46:32,117 --> 00:46:40,177

filter that out, knowing that already the

signals you're looking at are already so

714

00:46:40,177 --> 00:46:41,077

weak.

715

00:46:41,077 --> 00:46:46,727

So it feels pretty incredible to still be

able to do it, even though the signals are

716

00:46:46,727 --> 00:46:47,425

weak.

717

00:46:47,425 --> 00:46:49,245

and the result of noise.

718

00:46:49,325 --> 00:46:55,405

It's really amazing the technology that is

required to do these experiments has been

719

00:46:55,405 --> 00:47:01,035

developed decades and decades for people

to develop it and almost all aspects of

720

00:47:01,035 --> 00:47:04,465

the detectors have to be invented for that

purpose.

721

00:47:04,465 --> 00:47:09,355

There's very little off -the -shelf

technology and of course the spinoffs from

722

00:47:09,355 --> 00:47:14,745

that then taken up in other areas but it's

the pure science that was driving the

723

00:47:14,745 --> 00:47:15,909

development of the law.

724

00:47:18,701 --> 00:47:19,501

Yeah, exactly.

725

00:47:19,501 --> 00:47:24,251

It's like, it's not even as if the all the

engineering of these was already available

726

00:47:24,251 --> 00:47:27,451

and you could just go on Amazon and buy

it, right?

727

00:47:27,451 --> 00:47:33,101

You have like everything has to be

developed custom for these and you don't

728

00:47:33,101 --> 00:47:37,061

even know if that's going to work before

you actually try it out.

729

00:47:37,061 --> 00:47:40,961

So that's like all these endeavors are

absolutely incredible.

730

00:47:41,381 --> 00:47:46,951

And so that makes me think and I think on

these Christopher, you will have stuff to

731

00:47:46,951 --> 00:47:47,871

add.

732

00:47:48,121 --> 00:47:55,651

Because, so if I understood correctly, all

these detectors that we have right now are

733

00:47:55,651 --> 00:47:56,841

on Earth.

734

00:47:56,901 --> 00:47:59,501

These gravitational waves detectors.

735

00:48:00,161 --> 00:48:07,121

Hopefully, we'll be able to do a video

documentary on Learned Bay stats in one of

736

00:48:07,121 --> 00:48:07,861

these detectors.

737

00:48:07,861 --> 00:48:12,051

It's just some of the backstage I'm

telling to the listeners.

738

00:48:12,051 --> 00:48:13,921

We'll see if that's possible.

739

00:48:14,161 --> 00:48:17,655

But, so these detectors are on Earth.

740

00:48:17,933 --> 00:48:24,513

If you go to space and were able to put

one of these detectors around the earth or

741

00:48:24,513 --> 00:48:29,693

I don't know, in space floating somewhere,

I'm guessing that solves these problems,

742

00:48:29,693 --> 00:48:34,573

even though there are other sources of

issues if you do that in space.

743

00:48:34,973 --> 00:48:40,513

But if I understood correctly, the LISA

mission is space -based.

744

00:48:41,573 --> 00:48:44,363

And so is that a way of doing that?

745

00:48:44,363 --> 00:48:46,033

Can you tell us a bit more about that?

746

00:48:46,033 --> 00:48:47,589

Christopher and...

747

00:48:47,773 --> 00:48:53,213

Yeah, mainly tell us what the discoveries

will be with that.

748

00:48:53,253 --> 00:49:03,143

Also the data analysis problems that will

engender, especially when it comes to the

749

00:49:03,143 --> 00:49:05,873

size of the data, I'm guessing.

750

00:49:06,793 --> 00:49:07,373

Yeah.

751

00:49:07,373 --> 00:49:11,683

So Lisa's Space Space Gravitational Wave

mission, it's led by the European Space

752

00:49:11,683 --> 00:49:15,233

Agency with NASA as a junior partner

there.

753

00:49:15,233 --> 00:49:17,037

And the idea is we...

754

00:49:17,037 --> 00:49:21,367

launch a constellation of satellites, so

three satellites that will orbit around

755

00:49:21,367 --> 00:49:26,267

the Sun lagging behind the Earth in a

triangular formation and we bounce the

756

00:49:26,267 --> 00:49:30,867

lasers between them to make the same sort

of measurements that we do for

757

00:49:30,867 --> 00:49:37,617

gravitational waves but over a much larger

scale, so really massive arms.

758

00:49:38,437 --> 00:49:44,837

So this is great because we can avoid the

ground -based noise that John mentioned

759

00:49:44,837 --> 00:49:46,157

and this

760

00:49:46,157 --> 00:49:47,277

is really good.

761

00:49:47,277 --> 00:49:51,367

So for Lisa, we're not trying to see

exactly the same sources as with our

762

00:49:51,367 --> 00:49:54,977

ground -based detectors, but we're trying

to look for lower frequencies.

763

00:49:55,137 --> 00:49:58,787

So one of the things we've learned in

astronomy over the last century or so is

764

00:49:58,787 --> 00:50:02,167

that each time you're observing the

universe in a new way, you discover new

765

00:50:02,167 --> 00:50:02,517

things.

766

00:50:02,517 --> 00:50:07,597

So we want to look at a different part of

the spectrum of gravitational waves.

767

00:50:07,597 --> 00:50:12,217

So Lisa's most sensitive is the millihertz

range, so much lower frequencies.

768

00:50:12,557 --> 00:50:15,309

And a much lower frequency gravitational

wave,

769

00:50:15,309 --> 00:50:17,969

corresponds to a bigger source.

770

00:50:17,969 --> 00:50:23,429

So these could be the same type of binary,

but just much further apart in that orbit,

771

00:50:23,429 --> 00:50:27,689

so much earlier before they come in and

merge much further apart.

772

00:50:27,689 --> 00:50:33,009

Or we could be looking at much more

massive objects, so massive black holes.

773

00:50:33,009 --> 00:50:36,609

We believe at the center of every galaxy

is a massive black hole.

774

00:50:36,609 --> 00:50:40,469

Our own galaxy has one about four million

solar masses, four million times the mass

775

00:50:40,469 --> 00:50:41,469

of our sun.

776

00:50:41,589 --> 00:50:44,813

And we think galaxies merge, and so the

massive black hole should merge.

777

00:50:44,813 --> 00:50:48,213

And so we'd be able to see these out to a

much greater distance.

778

00:50:49,013 --> 00:50:56,073

So Lisa's objective is to see what we can

observe in the gravitational wave sky at

779

00:50:56,073 --> 00:50:58,413

these much lower frequencies.

780

00:50:58,613 --> 00:51:00,563

And there's a whole host of different

sources.

781

00:51:00,563 --> 00:51:04,543

So these massive black hole mergers we

should be able to see out across the

782

00:51:04,543 --> 00:51:06,213

entire history of the universe.

783

00:51:06,273 --> 00:51:09,573

We should be able to see regular stellar

mass black holes.

784

00:51:09,573 --> 00:51:11,179

So black holes formed from.

785

00:51:11,349 --> 00:51:15,309

stars at the end of their lives spiraling

into these supermassive black holes.

786

00:51:15,309 --> 00:51:16,949

It's a topic I've studied quite a lot.

787

00:51:16,949 --> 00:51:19,439

Those signals are extremely complicated.

788

00:51:19,439 --> 00:51:25,199

The orbits they undergo are very

intricate, which is great if we observe

789

00:51:25,199 --> 00:51:30,229

one because we can measure the parameters

to tiny, tiny precision, to one part in a

790

00:51:30,229 --> 00:51:31,529

million, something like that.

791

00:51:31,529 --> 00:51:35,669

But it's a huge pain from a data analysis

point of view because you've got to find

792

00:51:35,669 --> 00:51:38,409

the part of parameter space where this is.

793

00:51:38,489 --> 00:51:40,529

And we're also going to see

794

00:51:40,589 --> 00:51:47,179

huge numbers of binaries in our own galaxy

of white dwarfs, maybe neutron -style

795

00:51:47,179 --> 00:51:50,389

white dwarfs, so the wide binaries here.

796

00:51:50,669 --> 00:51:57,759

And so the real data analysis problem for

LISA will be how to fit all of this

797

00:51:57,759 --> 00:52:01,849

information all at once, because with our

ground -based detectors, at least at the

798

00:52:01,849 --> 00:52:06,149

moment, we basically just see here's a

signal and then here's another signal.

799

00:52:06,149 --> 00:52:08,767

So you can analyze each signal in

isolation.

800

00:52:09,005 --> 00:52:14,105

With Lisa, you cannot you see everything

all at time.

801

00:52:14,105 --> 00:52:16,905

Some of these lights, they don't

supermassive black hole mergers might be

802

00:52:16,905 --> 00:52:20,505

quite short to compare to place a

localized in time, but they will still be

803

00:52:20,505 --> 00:52:21,825

overlapping these long lives.

804

00:52:21,825 --> 00:52:26,865

So the the in spiraling objects or the

very wide bindings will basically be there

805

00:52:26,865 --> 00:52:30,345

for the entire mission or a large fraction

of the mission.

806

00:52:30,585 --> 00:52:35,255

So to analyze the data, you need to fit

everything or this is what we call a

807

00:52:35,255 --> 00:52:36,845

global fit problem.

808

00:52:36,845 --> 00:52:37,773

And you

809

00:52:37,773 --> 00:52:43,073

So you potentially have hundreds of

thousands of sources, each with a dozen

810

00:52:43,073 --> 00:52:49,533

parameters or so, maybe less than simpler

sources.

811

00:52:50,033 --> 00:52:53,533

But you've got to do all of these all at

the same time.

812

00:52:53,733 --> 00:52:58,043

And it potentially does matter how you do

this, because things like the massive

813

00:52:58,043 --> 00:53:03,573

black hole binaries are extremely loud, so

signal -to -noise ratios of thousands.

814

00:53:03,573 --> 00:53:06,727

So if you get that wrong by just a little

percentage,

815

00:53:07,053 --> 00:53:11,363

residual power in your data stream would

be enough to bias your measurements of the

816

00:53:11,363 --> 00:53:12,933

quieter signals underneath.

817

00:53:13,593 --> 00:53:18,193

So this is a huge, I think possibly the

most complicated data analysis problem in

818

00:53:18,193 --> 00:53:23,793

astronomy and we're just starting to

figure out how we're going to tackle this.

819

00:53:24,073 --> 00:53:28,113

So yeah, space -based detectors I think

extremely exciting, a whole host of new

820

00:53:28,113 --> 00:53:32,883

sources that we can see, a new host of

astrophysics that we can unlock through

821

00:53:32,883 --> 00:53:35,725

these observations, but also

822

00:53:35,725 --> 00:53:41,025

some extremely complicated data analysis

challenges that need to be tackled and

823

00:53:41,025 --> 00:53:44,293

solved before the mission launches in the

2030s.

824

00:53:45,741 --> 00:53:50,829

And what's the timeline on this mission?

825

00:53:52,513 --> 00:53:55,713

Are we close to launch?

826

00:53:55,813 --> 00:53:58,513

Where are things right now?

827

00:53:58,513 --> 00:54:04,953

So just in the last couple of months, the

mission was approved by ESA.

828

00:54:04,953 --> 00:54:09,493

So that's them looking at the designs and

going, OK, we think we can build this.

829

00:54:09,533 --> 00:54:13,593

And now the serious work on putting it

together comes.

830

00:54:13,593 --> 00:54:20,433

So it's due to launch in the 2030s,

exactly when that be, I'm sure.

831

00:54:21,805 --> 00:54:24,725

People are very confident on when it will

be, but we know space -based missions are

832

00:54:24,725 --> 00:54:25,645

hard.

833

00:54:25,845 --> 00:54:31,165

So it might, maybe, maybe it's a little

early to say exactly what date it will

834

00:54:31,165 --> 00:54:32,385

launch.

835

00:54:33,165 --> 00:54:37,835

But it will go up and then there'll be a

little period of commissioning and then it

836

00:54:37,835 --> 00:54:38,675

will start observing.

837

00:54:38,675 --> 00:54:44,505

So in the late 2030s, we should hopefully

get the observations from that.

838

00:54:44,505 --> 00:54:49,613

So the current timeline, 2035 for launch,

which I guess is...

839

00:54:49,613 --> 00:54:54,063

Good news to any of your listeners who are

inspired by the problems that we're

840

00:54:54,063 --> 00:54:57,553

talking about and think this is really

cool and think that maybe they'd like to

841

00:54:57,553 --> 00:54:58,413

tackle these problems.

842

00:54:58,413 --> 00:55:05,753

There's certainly enough time to go out,

get a degree, start a PhD in the field

843

00:55:05,753 --> 00:55:08,813

before we get the real data.

844

00:55:08,973 --> 00:55:10,413

Yeah, for sure.

845

00:55:10,413 --> 00:55:11,373

Exactly.

846

00:55:12,133 --> 00:55:17,833

And also, historically, these kind of huge

missions tend to take a bit of delay.

847

00:55:17,833 --> 00:55:19,501

So, you know, like...

848

00:55:19,501 --> 00:55:23,681

You can start your PhD on this.

849

00:55:24,441 --> 00:55:31,311

I mean, that's better to launch later than

to launch on time, but have a mission that

850

00:55:31,311 --> 00:55:32,441

fails, right?

851

00:55:32,581 --> 00:55:33,581

Yes.

852

00:55:33,581 --> 00:55:37,411

We're talking a billion euro cost of these

things.

853

00:55:37,411 --> 00:55:40,321

So you definitely don't want to explode on

the launch pad.

854

00:55:40,441 --> 00:55:41,161

Exactly.

855

00:55:41,521 --> 00:55:46,371

Way better to take a few more months and

do some double checks than just launch

856

00:55:46,371 --> 00:55:49,453

because we said we would launch on that

arbitrary date.

857

00:55:49,453 --> 00:55:53,063

Yeah, the space agencies do take these

things.

858

00:55:53,063 --> 00:55:57,543

It's been fascinating seeing the order,

the things that needed to be rubber

859

00:55:57,543 --> 00:56:00,253

stamped to get the approval for the

mission.

860

00:56:00,253 --> 00:56:03,493

So very good work people getting that

done.

861

00:56:04,113 --> 00:56:08,353

So there are also other proposed space

-based missions, some potential ones in

862

00:56:08,353 --> 00:56:08,623

China.

863

00:56:08,623 --> 00:56:14,533

There's a potential follow -up mission, I

guess, slightly in the future, maybe in

864

00:56:14,533 --> 00:56:16,645

Japan that's been proposed for a few

years.

865

00:56:16,645 --> 00:56:21,445

status of these, I guess, it's difficult

getting the funding for these things.

866

00:56:21,905 --> 00:56:26,065

So I think it's an exciting time in the

field.

867

00:56:26,065 --> 00:56:31,335

Hopefully we'll expand the range of

gravitational waves we can detect and

868

00:56:31,335 --> 00:56:32,785

that'll be great.

869

00:56:33,605 --> 00:56:35,325

Yeah, yeah, for sure.

870

00:56:35,565 --> 00:56:37,085

And I mean, that must be...

871

00:56:37,085 --> 00:56:42,295

So I don't know how directly involved you

are on these, Lounch, but I'm guessing

872

00:56:42,295 --> 00:56:45,389

that if you're still working on these

when...

873

00:56:45,389 --> 00:56:53,189

the mission launches, I'm pretty sure the

day of the launch, you will be pretty

874

00:56:53,189 --> 00:56:54,789

nervous and excited.

875

00:56:54,789 --> 00:56:59,569

Have you already lived that actually, or

would that be new to you?

876

00:57:00,369 --> 00:57:07,689

So I mean, the closest analogy would have

been there was a technology mission to

877

00:57:07,689 --> 00:57:12,799

test some of the key components of Lisa

called Lisa Pathfinder that went up a few

878

00:57:12,799 --> 00:57:14,957

years ago, an extremely successful

mission.

879

00:57:14,957 --> 00:57:18,637

And so watching that from the sidelines,

my PhD was on LISA.

880

00:57:18,637 --> 00:57:21,197

If this mission didn't work, then there'd

be no LISA mission.

881

00:57:21,197 --> 00:57:23,697

So all my PhD work would be in vain.

882

00:57:23,757 --> 00:57:29,157

But thankfully, it worked very well and

worked better than what was hoped for, in

883

00:57:29,157 --> 00:57:29,287

fact.

884

00:57:29,287 --> 00:57:30,877

So that was great.

885

00:57:31,197 --> 00:57:35,717

And I guess that's a real testament to the

experiment, as saying I was feeling

886

00:57:35,717 --> 00:57:37,107

worried because it was my PhD work.

887

00:57:37,107 --> 00:57:41,007

But there really people in the field who

have spent their entire careers working on

888

00:57:41,007 --> 00:57:43,017

this technology, you know, multiple

decades.

889

00:57:43,017 --> 00:57:44,391

So it's all.

890

00:57:44,391 --> 00:57:48,651

Yeah, real testament to their

determination, I guess, their vision going

891

00:57:48,651 --> 00:57:53,981

into a field right at the beginning before

anything worked to look at these things.

892

00:57:54,261 --> 00:57:57,761

It's also honestly quite remarkable that

we somehow managed to convince the funding

893

00:57:57,761 --> 00:58:03,481

agencies to fund these things for so long

before there would be scientific returns.

894

00:58:03,581 --> 00:58:09,291

So, yeah, we're extremely grateful that

they had the forethought and the patience

895

00:58:09,291 --> 00:58:13,101

to invest in something so long before it

would give returns.

896

00:58:15,253 --> 00:58:17,333

Yeah, definitely.

897

00:58:17,813 --> 00:58:21,413

Yeah, that must be absolutely fascinating.

898

00:58:21,873 --> 00:58:25,013

John, anything you want to add on that?

899

00:58:25,953 --> 00:58:31,263

I think Christopher is doing a great

overview of WISA, which indeed will be an

900

00:58:31,263 --> 00:58:33,433

enormous challenge on the ground.

901

00:58:33,433 --> 00:58:40,673

There are also plans to take things

forward into the 2030s and beyond.

902

00:58:40,833 --> 00:58:44,851

Currently, there are two major...

903

00:58:45,101 --> 00:58:49,141

detectors in the kind of scoping design

stage.

904

00:58:49,141 --> 00:58:55,821

One is led by the Europeans called the

Einstein Telescope and the other one is

905

00:58:55,821 --> 00:58:59,081

led by the US called Cosmic Explorer.

906

00:59:00,001 --> 00:59:02,041

They're taking different approaches.

907

00:59:02,181 --> 00:59:04,041

They're both going by detectors.

908

00:59:04,061 --> 00:59:07,301

The challenge there is to lower the noise

floor.

909

00:59:07,301 --> 00:59:11,051

So giving them a sort of order of

magnitude improvement in the range that

910

00:59:11,051 --> 00:59:14,285

you can see things to, which translates to

911

00:59:14,285 --> 00:59:20,765

thousand -fold increase in the volume that

you can see things to, more or less.

912

00:59:20,765 --> 00:59:24,105

At these kinds of distances, you do

actually have to worry about the size of

913

00:59:24,105 --> 00:59:27,085

the universe, getting in the way of these

calculations.

914

00:59:27,945 --> 00:59:34,465

But yeah, these new experiments will

require a new infrastructure.

915

00:59:34,765 --> 00:59:40,525

So they're also going to require a new

batch of experiments from national,

916

00:59:40,525 --> 00:59:43,257

indeed, European land.

917

00:59:43,949 --> 00:59:45,029

best friend.

918

00:59:47,437 --> 00:59:51,967

A lot of the data analysis challenges for

those are kind of similar to the ones that

919

00:59:51,967 --> 00:59:56,337

we're tackling with the current generation

of ground -based detectors.

920

00:59:56,497 --> 01:00:01,367

But the major difference is that the

signals would be much longer because the

921

01:00:01,367 --> 01:00:05,097

low frequency end is really the target for

improvement.

922

01:00:05,157 --> 01:00:08,997

I think that's the way that the binaries

chop.

923

01:00:08,997 --> 01:00:14,687

I mean, okay, I told you that they sort of

make this characteristic, whoop, type

924

01:00:14,687 --> 01:00:15,697

noise.

925

01:00:15,697 --> 01:00:17,325

Maybe you can find a sample.

926

01:00:17,325 --> 01:00:21,025

and pluck out my pale imitation.

927

01:00:21,945 --> 01:00:26,825

The lower in frequency you start, the

longer the signal will be.

928

01:00:26,825 --> 01:00:31,205

That multiplies the amount of data that

you have to analyze, which with a Bayesian

929

01:00:31,205 --> 01:00:32,785

problem can be a bit challenging.

930

01:00:32,785 --> 01:00:37,595

If you're doing many millions of light

-weighting evaluations, you don't want

931

01:00:37,595 --> 01:00:40,425

each light -weighting evaluation to be

expensive.

932

01:00:42,245 --> 01:00:45,365

And also the signal -to -noise ratio will

be huge.

933

01:00:45,365 --> 01:00:47,269

Least effects are 10 higher.

934

01:00:47,545 --> 01:00:53,765

So you will run into problems with our

uncertainties on the nature of the

935

01:00:53,765 --> 01:00:54,585

sources.

936

01:00:54,585 --> 01:00:58,265

So the models that we have are very good

theoretical models at the moment and

937

01:00:58,265 --> 01:01:03,555

they're good enough for the current

generation of detectors, but they will

938

01:01:03,555 --> 01:01:08,045

break down once observations become good

enough.

939

01:01:08,045 --> 01:01:14,735

They will probably show the crops in

theories, which I should say is probably

940

01:01:14,735 --> 01:01:17,101

not a fundamental part in the theory.

941

01:01:17,101 --> 01:01:20,101

I think most people probably would put

their money on general relativity being

942

01:01:20,101 --> 01:01:20,881

correct.

943

01:01:20,881 --> 01:01:25,631

The problem is that there is a translation

layer between general relativity and the

944

01:01:25,631 --> 01:01:32,041

types of temperament we can use it that

requires approximations and shortcuts and

945

01:01:32,041 --> 01:01:34,541

models to be created.

946

01:01:35,021 --> 01:01:41,261

So there's challenges with modeling and

balance that are quite difficult to

947

01:01:41,261 --> 01:01:45,671

overcome and people are searching that as

well at the moment.

948

01:01:48,685 --> 01:01:50,125

Yeah, fantastic.

949

01:01:50,265 --> 01:01:51,365

Thanks a lot, guys.

950

01:01:51,365 --> 01:01:58,325

That's really fantastic to have all these

overviews of the missions.

951

01:01:59,285 --> 01:02:07,445

And actually, I'm wondering, so with all

that work that you've been doing, all

952

01:02:07,445 --> 01:02:13,825

these studies that you've been talking

about since we started recording, we've

953

01:02:13,825 --> 01:02:16,301

been able to study actually what

954

01:02:16,301 --> 01:02:17,421

we want to do, right?

955

01:02:17,421 --> 01:02:22,321

So study the astrophysics of black holes

and also some tests of general relativity,

956

01:02:22,321 --> 01:02:23,861

as you were saying, Christopher.

957

01:02:24,221 --> 01:02:30,931

Can you tell us about that and mainly what

are the current frontiers on those fronts?

958

01:02:30,931 --> 01:02:34,669

What are we trying to learn with the

current missions?

959

01:02:37,291 --> 01:02:38,881

That's a big question.

960

01:02:39,041 --> 01:02:45,091

So general relativity, I guess, we really

want to find somewhere where it doesn't

961

01:02:45,091 --> 01:02:46,001

work.

962

01:02:46,001 --> 01:02:54,031

So for the point of view of understanding

gravity, there's this tension within

963

01:02:54,031 --> 01:02:59,141

physics that how do you reconcile general

relativity with quantum theory?

964

01:02:59,141 --> 01:03:03,921

And that is rather tricky and the whole

host of different theoretical frameworks

965

01:03:03,921 --> 01:03:05,421

to try and reconcile this.

966

01:03:05,421 --> 01:03:07,641

But we don't know for certain what the

answer is.

967

01:03:07,641 --> 01:03:12,591

And finding some hint where general

relativity breaks down would give a

968

01:03:12,591 --> 01:03:14,281

pointer in the right direction.

969

01:03:14,441 --> 01:03:18,561

Of course, finding a place where general

relativity breaks down is very difficult.

970

01:03:18,561 --> 01:03:24,161

The place where I think it makes sense to

look most is the most extreme environment.

971

01:03:24,161 --> 01:03:25,961

So where is gravity strongest?

972

01:03:26,041 --> 01:03:27,681

Where is the spacetime most dynamical?

973

01:03:27,681 --> 01:03:29,441

Where do things change the quickest?

974

01:03:29,581 --> 01:03:33,521

So black hole mergers, I think, are

really, and the gravitational wave

975

01:03:33,521 --> 01:03:35,117

signals, they admit, are the

976

01:03:35,117 --> 01:03:36,937

best place to look for that.

977

01:03:36,937 --> 01:03:39,047

So that's why we're looking there.

978

01:03:39,047 --> 01:03:43,447

And what we'd really love to find is some

deviation from general relativity that we

979

01:03:43,447 --> 01:03:47,937

could actually be certain is a deviation

from general relativity and not just a

980

01:03:47,937 --> 01:03:49,257

noise artifact.

981

01:03:49,357 --> 01:03:53,507

So I think we're pursuing a whole host of

different things to look for deviations

982

01:03:53,507 --> 01:03:54,537

there.

983

01:03:54,897 --> 01:04:01,207

On the astrophysics point of view, there's

just so much we don't know about the

984

01:04:01,207 --> 01:04:03,617

progenitors of these sources.

985

01:04:03,617 --> 01:04:04,781

So how do

986

01:04:04,781 --> 01:04:08,081

we end up with black holes and neutron

stars.

987

01:04:08,941 --> 01:04:12,141

So stars are pretty important in

astronomy.

988

01:04:12,141 --> 01:04:14,981

Exactly how they work is kind of

complicated.

989

01:04:14,981 --> 01:04:17,371

So there's a lot of uncertainties in that.

990

01:04:17,371 --> 01:04:21,421

And I think it's really quite remarkable

how rapidly the field has progressed.

991

01:04:21,421 --> 01:04:27,621

So back in 2015, before we made our first

detection, it wasn't at all certain that

992

01:04:27,621 --> 01:04:31,661

we would find pairs of black holes

orbiting each other and merging.

993

01:04:32,061 --> 01:04:34,189

We knew there would be neutron stars.

994

01:04:34,189 --> 01:04:37,049

But we didn't know they're black holes

because we'd never seen them.

995

01:04:37,049 --> 01:04:39,899

They're really hard to see other than

gravitational waves.

996

01:04:39,899 --> 01:04:43,009

That's kind of why we built the

gravitational wave detectors.

997

01:04:43,449 --> 01:04:45,049

But we hadn't seen any of them.

998

01:04:45,049 --> 01:04:47,389

So our first detection confirmed, yes,

they exist.

999

01:04:47,389 --> 01:04:51,069

And they exist in sufficient numbers that

we can actually detect them.

Speaker:

01:04:51,069 --> 01:04:53,909

And then the follow up was when we

measured the masses, they were about 30

Speaker:

01:04:53,909 --> 01:04:55,409

times the mass of our sun.

Speaker:

01:04:55,409 --> 01:04:58,329

We'd never seen black holes in that mass

range before.

Speaker:

01:04:59,249 --> 01:05:01,529

We now know, yep, there's quite a few of

them.

Speaker:

01:05:01,529 --> 01:05:04,045

But whether you can form black holes that

big,

Speaker:

01:05:04,045 --> 01:05:07,685

tells you something about the way that

stars live, how much mass they lose

Speaker:

01:05:07,685 --> 01:05:09,365

through their lifetime.

Speaker:

01:05:09,425 --> 01:05:13,905

So that's a key uncertainty that we don't

really understand about how stars evolve.

Speaker:

01:05:14,005 --> 01:05:18,805

So now, as we're building up statistics,

really teasing out the details of the mass

Speaker:

01:05:18,805 --> 01:05:21,665

distribution, what is the biggest black

hole that you can build?

Speaker:

01:05:21,805 --> 01:05:25,925

Currently, we know there are these black

holes that form from stars collapsing.

Speaker:

01:05:26,065 --> 01:05:32,065

And we know there are these massive stars,

massive black holes, millions of solar

Speaker:

01:05:32,065 --> 01:05:32,945

masses.

Speaker:

01:05:33,215 --> 01:05:36,185

lightest ones, hundreds of thousands, tens

of thousands.

Speaker:

01:05:36,185 --> 01:05:39,965

But we don't know, is there a continuous

distribution of black holes in between?

Speaker:

01:05:39,965 --> 01:05:42,725

So are there hundreds of thousands of mass

black holes?

Speaker:

01:05:42,725 --> 01:05:44,485

So that's one of the key things to figure

out.

Speaker:

01:05:44,485 --> 01:05:45,585

Is there a key thing?

Speaker:

01:05:45,585 --> 01:05:49,025

Where do these big, really big, massive

black holes come from?

Speaker:

01:05:49,725 --> 01:05:51,825

And how do stars evolve?

Speaker:

01:05:51,825 --> 01:05:55,815

The details of all the different ways that

you could end up with massive black holes

Speaker:

01:05:55,815 --> 01:05:57,085

that people theorized?

Speaker:

01:05:57,085 --> 01:05:58,305

Which ones are correct?

Speaker:

01:05:58,305 --> 01:06:00,905

In what ratio out there?

Speaker:

01:06:01,165 --> 01:06:02,733

And then I guess one...

Speaker:

01:06:02,733 --> 01:06:07,023

One additional key thing, we talked about

black holes in nature gravity.

Speaker:

01:06:07,023 --> 01:06:09,773

We've talked about how you form black

holes in neutron stars.

Speaker:

01:06:10,253 --> 01:06:13,533

But there's also what neutron stars are

really made of.

Speaker:

01:06:13,953 --> 01:06:18,613

So neutron stars, from the name you might

suggest, OK, they're made of very neutron

Speaker:

01:06:18,613 --> 01:06:19,773

-rich matter.

Speaker:

01:06:19,833 --> 01:06:23,933

But actually, what happens inside the core

of a neutron star, we get a whole host of

Speaker:

01:06:23,933 --> 01:06:28,653

different phase changes, really quite

exotic matter going on that we can't hope

Speaker:

01:06:28,653 --> 01:06:30,363

to replicate in the lab here on Earth.

Speaker:

01:06:30,363 --> 01:06:31,897

So we really don't know.

Speaker:

01:06:32,203 --> 01:06:33,083

this behaves.

Speaker:

01:06:33,083 --> 01:06:37,033

If we did, that would be really

informative for understanding the dynamics

Speaker:

01:06:37,033 --> 01:06:39,053

of the particles that make those.

Speaker:

01:06:39,053 --> 01:06:43,633

So by making measurements of the neutron

stars we observe, how much they stretch

Speaker:

01:06:43,633 --> 01:06:48,123

and squeeze, we can hopefully get some

constraints on what neutron stars are made

Speaker:

01:06:48,123 --> 01:06:52,073

of, which would be an exciting frontier

there.

Speaker:

01:06:53,533 --> 01:06:54,393

John?

Speaker:

01:06:55,353 --> 01:07:00,603

One thing that I think we can zoom out

from looking at the individual black holes

Speaker:

01:07:00,603 --> 01:07:02,275

and neutron stars and

Speaker:

01:07:03,213 --> 01:07:08,313

Still with the theme of trying to

understand gravity is on the other scale

Speaker:

01:07:08,313 --> 01:07:15,443

is cosmology, the very, very largest

scales, how is the universe evolving over

Speaker:

01:07:15,443 --> 01:07:16,453

time?

Speaker:

01:07:17,833 --> 01:07:23,393

Hopefully with the current generation and

the next generation, we'll be able to do

Speaker:

01:07:23,393 --> 01:07:28,333

cosmology in a completely different way

than what we have done up until now.

Speaker:

01:07:28,333 --> 01:07:32,013

By looking at the gravitational wave

signal, so those...

Speaker:

01:07:32,013 --> 01:07:37,193

properties of those signals, the fact that

we know exactly what they look like, their

Speaker:

01:07:37,193 --> 01:07:41,873

amplitude and how it would case with

distance means that they can be used as an

Speaker:

01:07:41,873 --> 01:07:43,613

independent co -coxmology.

Speaker:

01:07:43,613 --> 01:07:48,923

Now we've already done this with the

prying intrastar signal and with black

Speaker:

01:07:48,923 --> 01:07:53,843

holes that we've seen up to now,

relatively low numbers of sources such

Speaker:

01:07:53,843 --> 01:07:58,863

that the constraints that we're able to

cook are not yet competitive with the best

Speaker:

01:07:58,863 --> 01:08:01,753

constraints that we can get from other

techniques.

Speaker:

01:08:01,933 --> 01:08:06,313

But going forward, as the numbers improve,

as the SNRs and the applies ratio

Speaker:

01:08:06,313 --> 01:08:09,593

improves, this is going to get better and

better over time.

Speaker:

01:08:09,593 --> 01:08:14,833

And so even if we don't see anything on

the scale of the individual black holes,

Speaker:

01:08:14,833 --> 01:08:20,563

if this agrees with general relativity, it

could still help us en masse to pin down

Speaker:

01:08:20,563 --> 01:08:24,563

what's going on with cosmology, where

there are many things that we don't

Speaker:

01:08:24,563 --> 01:08:29,253

understand, including discrepancies in the

existing constraints we have.

Speaker:

01:08:31,757 --> 01:08:32,897

No one.

Speaker:

01:08:36,333 --> 01:08:44,203

And I'm curious, among all of these

burning issues, burning questions, if you

Speaker:

01:08:44,203 --> 01:08:49,493

could choose one that you're sure you're

going to get the answer to before you die,

Speaker:

01:08:49,493 --> 01:08:50,693

what would it be?

Speaker:

01:08:55,821 --> 01:09:00,421

I don't know how long I'm going to live,

but the thing that really motivates me is

Speaker:

01:09:00,421 --> 01:09:05,201

trying to understand whether the black

holes that we're seeing really are the

Speaker:

01:09:05,201 --> 01:09:08,661

things that you can write down with pencil

and paper when you're teaching people

Speaker:

01:09:08,661 --> 01:09:13,101

general relativity, or are they more

complicated than that in reality?

Speaker:

01:09:13,261 --> 01:09:17,781

I think if there was one problem I have to

choose in this field, that would be the

Speaker:

01:09:17,781 --> 01:09:19,981

one that I found the most interesting.

Speaker:

01:09:20,761 --> 01:09:24,907

I think I'd really like to know the answer

to that one as well.

Speaker:

01:09:24,941 --> 01:09:28,951

I think that might be one of the most

challenging to actually get the solution

Speaker:

01:09:28,951 --> 01:09:30,001

to.

Speaker:

01:09:30,441 --> 01:09:33,881

The best way to answer it might be to

travel into a black hole.

Speaker:

01:09:33,901 --> 01:09:38,741

But then the question of whether you

observe anything before you die becomes

Speaker:

01:09:38,741 --> 01:09:40,221

rather technical.

Speaker:

01:09:40,441 --> 01:09:41,361

Yeah.

Speaker:

01:09:41,821 --> 01:09:46,901

Certainly something not advised for your

listeners to give that a go.

Speaker:

01:09:46,901 --> 01:09:53,357

Yeah, I am not sure it would end up like

Matthew McConaughey in The

Speaker:

01:09:53,357 --> 01:09:54,237

What's the movie?

Speaker:

01:09:54,237 --> 01:09:55,437

You know that?

Speaker:

01:09:55,437 --> 01:09:56,657

Interstellar.

Speaker:

01:09:56,657 --> 01:10:01,097

So Kip Thorne is one of the founders of

LIGO.

Speaker:

01:10:01,097 --> 01:10:06,397

One of the recipients of the Nobel Prize

for Gravitational Analytics.

Speaker:

01:10:06,397 --> 01:10:08,717

He's behind Interstellar.

Speaker:

01:10:08,897 --> 01:10:10,857

So he advised on a lot of it.

Speaker:

01:10:10,857 --> 01:10:16,737

Yeah, the bit at the end is not backed up

by science.

Speaker:

01:10:18,297 --> 01:10:18,957

For sure.

Speaker:

01:10:18,957 --> 01:10:20,237

At least for now.

Speaker:

01:10:21,005 --> 01:10:26,385

They originally were going to have the

wormhole thing that opens up in

Speaker:

01:10:26,385 --> 01:10:26,605

Interstellar.

Speaker:

01:10:26,605 --> 01:10:29,505

They're going to have that detected with

gravitational waves at LIGO.

Speaker:

01:10:29,505 --> 01:10:32,165

Unfortunately, Christopher Nolan cut that

bit.

Speaker:

01:10:32,345 --> 01:10:33,075

Oh, that's a shame.

Speaker:

01:10:33,075 --> 01:10:34,625

It wasn't in the film.

Speaker:

01:10:35,185 --> 01:10:40,625

Maybe that would be for Interstellar 2.

Speaker:

01:10:40,625 --> 01:10:41,805

We don't know.

Speaker:

01:10:43,005 --> 01:10:44,415

So guys, thanks a lot.

Speaker:

01:10:44,415 --> 01:10:46,765

I've already taken a lot of your time.

Speaker:

01:10:47,085 --> 01:10:50,207

And I still have a good talk for you.

Speaker:

01:10:50,207 --> 01:10:54,817

hours because this is really really

fascinating but it's time to call it a

Speaker:

01:10:54,817 --> 01:11:00,147

show before that though as usual i'm gonna

ask you the the two questions i ask every

Speaker:

01:11:00,147 --> 01:11:05,517

guest at the end of the show first one if

you had unlimited time and resources which

Speaker:

01:11:05,517 --> 01:11:09,963

problem would you try to solve um who

wants to start

Speaker:

01:11:13,261 --> 01:11:20,981

I think if you're really serious about the

unlimited time resources, then the most

Speaker:

01:11:20,981 --> 01:11:25,791

pressing problem I think would be nothing

to do with adaptation waves, but it's more

Speaker:

01:11:25,791 --> 01:11:28,321

to do with the climate breakdown.

Speaker:

01:11:28,341 --> 01:11:34,477

So if you want an honest answer, that's my

answer, is solve climate change.

Speaker:

01:11:35,981 --> 01:11:38,401

That's a very popular answer.

Speaker:

01:11:38,941 --> 01:11:41,441

Get nuclear fusion working.

Speaker:

01:11:41,601 --> 01:11:44,381

That would be very nice.

Speaker:

01:11:45,001 --> 01:11:51,061

In our field with infinite resources, I

tackle the quantum theory of gravity and

Speaker:

01:11:51,061 --> 01:11:53,061

get the evidence for that.

Speaker:

01:11:53,061 --> 01:11:55,141

Would be nice.

Speaker:

01:11:56,461 --> 01:11:58,561

Yeah, definitely.

Speaker:

01:11:58,601 --> 01:12:02,341

That is a great answer.

Speaker:

01:12:02,341 --> 01:12:05,293

And I think also some people answered

that.

Speaker:

01:12:05,293 --> 01:12:08,433

So you're in good company, Christophe.

Speaker:

01:12:08,513 --> 01:12:13,833

And second question, if you could have

dinner with any great scientific mind

Speaker:

01:12:13,833 --> 01:12:17,509

dead, alive or fictional, who would it be?

Speaker:

01:12:20,887 --> 01:12:28,717

Maybe Chris, the only answer for, uh,

dead, I think for this podcast, and James

Speaker:

01:12:28,717 --> 01:12:29,897

would be my choice.

Speaker:

01:12:29,897 --> 01:12:32,257

You may know him if you're a Bayesian.

Speaker:

01:12:32,257 --> 01:12:32,877

Yeah.

Speaker:

01:12:32,877 --> 01:12:35,737

Um, I think he would be very good dinner

company.

Speaker:

01:12:35,957 --> 01:12:41,057

Um, his textbook was one of the formative

influences on me as a young Bayesian.

Speaker:

01:12:41,057 --> 01:12:41,977

Yeah.

Speaker:

01:12:41,997 --> 01:12:42,477

Yeah.

Speaker:

01:12:42,477 --> 01:12:43,417

Yeah, for sure.

Speaker:

01:12:43,417 --> 01:12:49,117

And, uh, there is a, there is a really

great, uh, YouTube.

Speaker:

01:12:49,117 --> 01:12:55,017

series playlist by Aubrey Clayton, who was

here on episode 51.

Speaker:

01:12:55,017 --> 01:13:00,827

So Aubrey Clayton wrote a book called

Bernoulli's Fallacy, The Crisis of Modern

Speaker:

01:13:00,827 --> 01:13:01,837

Science.

Speaker:

01:13:01,957 --> 01:13:02,957

Really interesting book.

Speaker:

01:13:02,957 --> 01:13:07,657

I'll link to the episode and also to his

YouTube series where he goes through E .T.

Speaker:

01:13:07,657 --> 01:13:11,417

Jane's book, Probability Theory, I think

it's called.

Speaker:

01:13:12,397 --> 01:13:17,807

which is a really great book, also really

well written and already goes through its

Speaker:

01:13:17,807 --> 01:13:22,987

chapters and explain the different ideas

and so on.

Speaker:

01:13:22,987 --> 01:13:27,947

So that's also a very fun YouTube playlist

if you want I'm definitely going to go and

Speaker:

01:13:27,947 --> 01:13:29,177

look that up.

Speaker:

01:13:29,177 --> 01:13:30,097

Awesome, yeah.

Speaker:

01:13:30,097 --> 01:13:31,877

I'll send that your way.

Speaker:

01:13:32,417 --> 01:13:32,957

And Christopher?

Speaker:

01:13:32,957 --> 01:13:35,217

One of my favorite books, yeah.

Speaker:

01:13:36,717 --> 01:13:42,317

I don't know, I think I might be somewhat

boring and just go for Einstein for the...

Speaker:

01:13:42,317 --> 01:13:46,037

of both gravity, I think he'd like to know

what we're up to.

Speaker:

01:13:46,037 --> 01:13:51,357

And also just to see what, you know, his

thoughts were being about being such a

Speaker:

01:13:51,357 --> 01:13:57,157

public intellectual and what it was like

being that would be being cool.

Speaker:

01:13:57,157 --> 01:14:00,557

I could invite a guest might be

interesting to get Newton along as well,

Speaker:

01:14:00,557 --> 01:14:02,097

and see what they think about gravity.

Speaker:

01:14:02,097 --> 01:14:07,837

But I think that would be quite awkward in

a conversation, I get the feeling, not the

Speaker:

01:14:07,837 --> 01:14:11,769

socially the most interactive.

Speaker:

01:14:12,525 --> 01:14:13,925

Yeah, yeah.

Speaker:

01:14:14,345 --> 01:14:20,435

Do you think Einstein would accept at that

point the, like all the advances in, like

Speaker:

01:14:20,435 --> 01:14:25,895

all the ramifications of actually general

relativity and so on and the crazy

Speaker:

01:14:25,895 --> 01:14:30,805

predictions that that was making and in

the end, most of them, like for now, at

Speaker:

01:14:30,805 --> 01:14:36,105

least were true, but at the end of his

career, he was not really accepting that.

Speaker:

01:14:36,105 --> 01:14:38,313

Do you think he would accept that now?

Speaker:

01:14:39,629 --> 01:14:43,989

I think he would accept the general

relativity and he would be delighted to

Speaker:

01:14:43,989 --> 01:14:49,049

find that we've seen some of the effects

that he never thought he observed.

Speaker:

01:14:50,669 --> 01:14:55,889

And again, he himself knew the general

relativity couldn't be the final answer to

Speaker:

01:14:55,889 --> 01:14:57,749

the correction of gravity.

Speaker:

01:14:57,929 --> 01:15:02,629

So he'd probably also be interested to

know how we've seen any signs of it

Speaker:

01:15:02,629 --> 01:15:03,319

breaking down.

Speaker:

01:15:03,319 --> 01:15:06,949

And I think the stuff that motivated him

towards the end of his career is

Speaker:

01:15:06,949 --> 01:15:07,725

probably...

Speaker:

01:15:07,725 --> 01:15:10,185

still what's motivating a lot of people.

Speaker:

01:15:15,085 --> 01:15:23,025

Well, if you are invited to such a dinner,

please let me know and I will gladly come.

Speaker:

01:15:24,585 --> 01:15:25,585

Awesome guys.

Speaker:

01:15:25,585 --> 01:15:28,135

I think it's time to call it a show.

Speaker:

01:15:28,135 --> 01:15:29,675

You've been wonderful.

Speaker:

01:15:29,675 --> 01:15:32,005

Thanks a lot for taking so much time.

Speaker:

01:15:32,005 --> 01:15:37,405

As usual, I will put resources and a link

to your websites in the show notes for

Speaker:

01:15:37,405 --> 01:15:38,861

those who want to dig deeper.

Speaker:

01:15:38,861 --> 01:15:42,801

The show notes are huge for this episode,

I can already warn listeners.

Speaker:

01:15:43,101 --> 01:15:45,781

So lots of things to look at.

Speaker:

01:15:45,781 --> 01:15:50,411

And well, thank you again, Chris and John

for taking the time and being on this

Speaker:

01:15:50,411 --> 01:15:51,059

show.

Speaker:

01:15:52,685 --> 01:15:54,085

Thank you very much.

Speaker:

01:15:54,745 --> 01:15:58,865

I may put in one thing that your listeners

might like.

Speaker:

01:15:58,865 --> 01:16:02,025

They're interested in trying gravitational

wave data analysis.

Speaker:

01:16:02,105 --> 01:16:03,225

Data are public.

Speaker:

01:16:03,225 --> 01:16:07,825

They can look up the Gravitational Wave

Open Science Center, download the data

Speaker:

01:16:07,825 --> 01:16:08,605

there.

Speaker:

01:16:08,605 --> 01:16:12,165

Also, they'll find links to tutorials.

Speaker:

01:16:12,405 --> 01:16:18,815

There are workshops held fairly regularly

that they can maybe sign up to to get some

Speaker:

01:16:18,815 --> 01:16:21,197

data analysis experience.

Speaker:

01:16:21,197 --> 01:16:24,577

And there's a whole list of open source

packages for gravitational wave data

Speaker:

01:16:24,577 --> 01:16:28,337

analysis linked from those so they can go

and have a look at themselves.

Speaker:

01:16:29,377 --> 01:16:31,477

Yeah, this is indeed a very good ad.

Speaker:

01:16:31,477 --> 01:16:32,697

Thank you very much, Christopher.

Speaker:

01:16:32,697 --> 01:16:37,717

I actually already put these links in the

show notes and forgot to mention them.

Speaker:

01:16:37,717 --> 01:16:39,637

So thank you very much.

Speaker:

01:16:39,637 --> 01:16:44,377

Because we're all very dedicated to open

source and open source here.

Speaker:

01:16:44,517 --> 01:16:49,297

So if any of the listeners are interested

in that, like how these things are done,

Speaker:

01:16:50,157 --> 01:16:57,247

You have all the packages we've mentioned

in the show notes, but also the open

Speaker:

01:16:57,247 --> 01:17:03,457

source and open science efforts from your

collaborations, Christopher and John.

Speaker:

01:17:03,457 --> 01:17:06,037

So definitely take a look at the show

notes.

Speaker:

01:17:06,037 --> 01:17:07,817

Everything is in there.

Speaker:

01:17:08,457 --> 01:17:09,657

Thank you guys.

Speaker:

01:17:09,717 --> 01:17:15,583

And well, you can come back on the podcast

any...

Speaker:

01:17:15,583 --> 01:17:25,197

Any time, hopefully around 2034 to talk

about Lysa and the space -based mission.

Speaker:

01:17:26,765 --> 01:17:28,625

I'll put it in my calendar.

Previous post
Next post