4
$\begingroup$

Comments below What does "GPU-accelerated butterfly matched filtering over dense bank of time-symmetric chirp-like templates" mean? (GW170817) suggest that for this technique a library of strain "chirps" is built, then in real time a fast computer (in this case GPU-assisted) constantly compares the the most recent signals from a gravitational wave detecting interferometer to all of them, looking for matches above a chosen matching threshold.

As long as the matching rate is not too high, these can be passed on to more sophisticated analyses to better decide if a chirped gravitational wave event might have been detected, and if so, to start figuring out where it was so that alerts can be generated and other instruments (e.g. optical and X-ray telescopes) can start looking for a signature for a source.

Question: Are chirped gravitational wave events generally first identified by searching through libraries of chirps?

Is this how they are first identified in general, or does the paper cited in the linked question describe just one possible method, and library searches are not always a part of the process?

Improve this question
$\endgroup$

2 Answers 2

Reset to default
3
$\begingroup$

LIGO/Virgo has multiple detection pipelines. Several of them (GstLAL, MBTA, PyCBC Live and SPIIR) are "modelled" searches which use large grids of pre-computed models to compare the signal against. There are also "unmodelled" searches (cWB) which simply look for something "bright" that occurs in multiple detectors at the same time.

See https://emfollow.docs.ligo.org/userguide/analysis/searches.html for further details

Improve this answer
$\endgroup$
3
  • $\begingroup$ Is it possible to say if it's the library/template search that identifies most of them, or is it mostly a red herring? "Are chirped gravitational wave events generally first identified by searching through libraries of chirps?" $\endgroup$
    – uhoh
    Commented Mar 25, 2022 at 15:52
  • $\begingroup$ @uhoh I think it's generally not true. Detection depends on many things, e.g., false alarm rate, signal to noise of the detectors, the properties of the source itself, etc... For example, see Table IV of the third catalog which shows that many sources are detected by many methods, but some only by others. arxiv.org/abs/2111.03606 $\endgroup$
    – Daddy Kropotkin
    Commented Mar 25, 2022 at 16:20
  • $\begingroup$ @DaddyKropotkin then perhaps your link provides enough information to complete an answer to my question since I'm not asking for the whys or hows. Thanks! $\endgroup$
    – uhoh
    Commented Mar 25, 2022 at 16:23
2
$\begingroup$

If I’m understanding everything about what you’re asking and how it relates to this other thing, I believe a kind of library you’re thinking of refers to a surrogate model.

The idea behind a surrogate model is that while post-Newtonian approximations can capture gravitational wave signals up to the ring down, that often time that isn’t enough, and so surrogate models are used as the ‘library’ to compare gravitational wave signals to for identifying the possible source.

The idea behind a surrogate model is that you can solve Einstein’s equations to get a pretty good looking waveform that has features like ringdown, but that just one run of the numerical model can take up to weeks, and so it’s not practical to run at every possible set of parameter points. Instead, people will run a sample of (not evenly spaced) points in the parameter space, and used sophisticated interpolation schemes and sometimes machine learning to try to fill in the gaps, creating something of an effectual library of gravitational waveforms in parameter space that can then be referenced against incoming signals.

As far as which surrogate model is used or even how popular they are amongst those who actually do the signal processing, I’m not sure, but I do know it’s a very active area of research that has had many papers about improving them come out in recent years.

Furthermore, as to your main question, a lot of times (using whatever model it might be, surrogate, PN, etc) these libraries serve as a primary method for finding these waveforms in the noise, because my (limited) understanding is that it is much easier to check if a given signal is present in data than to find a previously ungenerated waveform, and that once these models are construed, it's pretty quick to compare a signal against one.

I say all of this with some uncertainty, as working with these was only a brief side project a year ago, so if anyone sees I've misrepresented anything, please let me know.

Improve this answer
$\endgroup$
5
  • $\begingroup$ Thank you for your answer! My term "library" refers to their "dense bank of templates" so I'm really asking if the early identification of potential GW events from mergers is done by searching template banks or not; i.e. is this now how it's done. I didn't realize until now that the ringdown was such an important factor for near-real time flagging potential events, but including it certainly could reduce the rate of false positives. $\endgroup$
    – uhoh
    Commented Mar 25, 2022 at 3:14
  • 1
    $\begingroup$ btw What does "GPU-accelerated butterfly matched filtering over dense bank of time-symmetric chirp-like templates" mean? (GW170817) currently has a bounty, and you may find What can be learned from, or noted in this LIGO Orrery video? interesting as well. $\endgroup$
    – uhoh
    Commented Mar 25, 2022 at 3:17
  • 1
    $\begingroup$ @uhoh Rather conveniently, I was able to attend a guest lecture series from Dr. Nicolas Yunes on the matter today and it does seem seem that looking for a specific template is the principal method for determining a match, and he referenced a theorem for stating that this can always be done given a signal, although he did not reference the name of the theorem $\endgroup$
    – Justin T
    Commented Mar 26, 2022 at 0:51
  • 1
    $\begingroup$ He also mentioned that while surrogate models work great where they exist, they actually don’t really exist in high mass ratios currently, and so perturbative theory based templates like those of post-Newtonian seem to be the principal method for determining those methods, and then using a Markov Chain Monte Carlo algorithm, they find the parameters that best match the signal from the templates $\endgroup$
    – Justin T
    Commented Mar 26, 2022 at 0:54
  • $\begingroup$ Excellent! That's great to hear, thanks for update and insight and for sharing the anecdote. I don't know what the theorem is, but it might be simply something about "spanning the space" of possible functions, or related to proofs for (some) wavelet transforms being (over-)complete sets. I'm just using my "math words" without any specific expertise though. :-) $\endgroup$
    – uhoh
    Commented Mar 26, 2022 at 1:03

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .