5 ms·

Yes, we carefully explain the statistical confidence of our detections in each discovery paper, and there are dozens of methods papers that go into excruciating

by pinko 5y ago

Yes, we carefully explain the statistical confidence of our detections in each discovery paper, and there are dozens of methods papers that go into excruciating detail on those techniques. Prior to our first detection, the overwhelming prime directive of our collaboration was _not to make a false detection_ and we went to insane lengths to avoid one; e.g., we had a small team of people secretly injecting false signals -- "blind injections" -- into our data, so that we all expected to be regularly seeing them and wouldn't be tempted to gin the analysis to find a detection where there wasn't one. (Amusingly, because of this, it took weeks for many of us to believe the first detection was real, even after the blind injection team swore it wasn't one of theirs. In the end, we charged an independent team to do a forensic analysis of everything from the security cameras and seismometers in the observatories, to every computer and disk the data passed through, to convince ourselves this couldn't possibly have been maliciously injected by hand by a rogue scientist. It was a wild few weeks!) Now that it's clear gravitational waves exist, however, we focus on optimizing the false alarm rate for astronomers (who want it well above zero so they don't miss anything) rather than optimizing for ~zero false detections.

ars 5y ago
When I see LIGO graphs they always appear linear - the change in frequency is linear with time. But with time dilation as black holes near each other, shouldn't the frequency change be exponential? Or does it cancel out - the frequency goes up as they get closer, and then time dilation lowers it back as it dilates into infinity?
im3w1l 5y ago
That sounds hilarious, but you must be quite unique in having this approach right? Or are the particule physics people playing similar games?
- pinko 5y ago
  Unique as far as I know, at least in a scientific collaboration like ours. But we were unusually sensitive to the risk of false detections given (a) skepticism of LIGO's prospects from other physicists, and (b) the lingering stain of a false detection claim, from a bar detector, a few decades earlier. We believed that if we had announced a discovery that turned out to be wrong, it would probably have meant the end of our experiment, and in practice the end of the field, at least for a long time. It was fortuitous that the first detection turned out to be gold plated and unambiguous, otherwise we would have probably published a bunch of "we saw something interesting but can't claim it as a GW discovery" papers on the next few weaker events before we would have felt comfortable making a confident claim.
  - airstrike 5y ago
    Thank you for all these replies and for the fantastic work you and the team do. This is now my favorite AMA
- jasonwatkinspdx 5y ago
  Not quite the same lengths, but AIRC LHC uses two detectors with substantial design differences, and partitions the teams from sharing information for some of what they target, as a way to reduce systemic errors.
matt123456789 5y ago
Are your security cameras sensitive enough to pick up rogue sophons as well? (Three body reference, sorry.)
dekhn 5y ago
Has LIGO ever responded to Sabine Hossenfelder's claims? http://backreaction.blogspot.com/2019/09/whats-up-with-ligo.html http://backreaction.blogspot.com/2019/09/whats-up-with-ligo....
- DrBazza 5y ago
  I thought that 2019 article is now "answered" by the fact that we have multi-messenger observations of gravitational wave events that confirm fading light curves.
  - dnautics 5y ago
    I believe we still only have one multi-messenger observation. One observation has infinite variance =P. They're ALSO not independent observations, though. The LIGO observation was made, and then the directive was "go look for correlating events". That's a dependent observation. To be truly independent, they would have to inject synthetic data (say 3x) into the observation reports and make sure that the multi-messenger results don't correlate with the synthetic data, much like GP says they did "at the beginning of LIGO" with internal (non-multimodal) signals. This would have the side benefit of making the analysis pretty simple to do (just a chi-squared analysis) and you don't have to have a PhD in signals analysis with a specialization in the filters used by LIGO to believe with a high confidence that 1) GWs exist and 2) LIGO is actually measuring GWs.
- PhantomGremlin 5y ago
  I have a basic problem with Sabine's claims. In a response comment she says: "No signal analysis can confirm that the signal was of astrophysical origin." I know nothing about the actual hardware. But I can reasonably speculate that LIGO can afford to have local atomic clocks. So they can timestamp their observations to nanosecond precision. Given the finite speed of light, the candidate astrophysical events are necessarily detected many milliseconds apart. There can only be two ways these potential events would correlate: 1) there are so many many possible false positives constantly happening that there is some reasonable probability of this correlation occurring by chance. or 2) a highly sophisticated "goof" or fraud. Someone could presumably set up some local source near each detector and spoof a signal. E.g. carefully "wiggle" large masses, each only a few km away from each detector. So, if it's not astrophysical, can Sabine tell us which it is? False positive? Goof?

by pinko 5y ago

ars 5y ago

When I see LIGO graphs they always appear linear - the change in frequency is linear with time. But with time dilation as black holes near each other, shouldn't the frequency change be exponential? Or does it cancel out - the frequency goes up as they get closer, and then time dilation lowers it back as it dilates into infinity?

im3w1l 5y ago

That sounds hilarious, but you must be quite unique in having this approach right? Or are the particule physics people playing similar games?

pinko 5y ago
Unique as far as I know, at least in a scientific collaboration like ours. But we were unusually sensitive to the risk of false detections given (a) skepticism of LIGO's prospects from other physicists, and (b) the lingering stain of a false detection claim, from a bar detector, a few decades earlier. We believed that if we had announced a discovery that turned out to be wrong, it would probably have meant the end of our experiment, and in practice the end of the field, at least for a long time. It was fortuitous that the first detection turned out to be gold plated and unambiguous, otherwise we would have probably published a bunch of "we saw something interesting but can't claim it as a GW discovery" papers on the next few weaker events before we would have felt comfortable making a confident claim.
- airstrike 5y ago
  Thank you for all these replies and for the fantastic work you and the team do. This is now my favorite AMA
jasonwatkinspdx 5y ago
Not quite the same lengths, but AIRC LHC uses two detectors with substantial design differences, and partitions the teams from sharing information for some of what they target, as a way to reduce systemic errors.

matt123456789 5y ago

Are your security cameras sensitive enough to pick up rogue sophons as well? (Three body reference, sorry.)

dekhn 5y ago

Has LIGO ever responded to Sabine Hossenfelder's claims? http://backreaction.blogspot.com/2019/09/whats-up-with-ligo.html http://backreaction.blogspot.com/2019/09/whats-up-with-ligo....

DrBazza 5y ago
I thought that 2019 article is now "answered" by the fact that we have multi-messenger observations of gravitational wave events that confirm fading light curves.
- dnautics 5y ago
  I believe we still only have one multi-messenger observation. One observation has infinite variance =P. They're ALSO not independent observations, though. The LIGO observation was made, and then the directive was "go look for correlating events". That's a dependent observation. To be truly independent, they would have to inject synthetic data (say 3x) into the observation reports and make sure that the multi-messenger results don't correlate with the synthetic data, much like GP says they did "at the beginning of LIGO" with internal (non-multimodal) signals. This would have the side benefit of making the analysis pretty simple to do (just a chi-squared analysis) and you don't have to have a PhD in signals analysis with a specialization in the filters used by LIGO to believe with a high confidence that 1) GWs exist and 2) LIGO is actually measuring GWs.
PhantomGremlin 5y ago
I have a basic problem with Sabine's claims. In a response comment she says: "No signal analysis can confirm that the signal was of astrophysical origin." I know nothing about the actual hardware. But I can reasonably speculate that LIGO can afford to have local atomic clocks. So they can timestamp their observations to nanosecond precision. Given the finite speed of light, the candidate astrophysical events are necessarily detected many milliseconds apart. There can only be two ways these potential events would correlate: 1) there are so many many possible false positives constantly happening that there is some reasonable probability of this correlation occurring by chance. or 2) a highly sophisticated "goof" or fraud. Someone could presumably set up some local source near each detector and spoof a signal. E.g. carefully "wiggle" large masses, each only a few km away from each detector. So, if it's not astrophysical, can Sabine tell us which it is? False positive? Goof?