A Uncovering the Truth Behind LIGO's Gravity Wave Detection: A Critical Analysis

[auou]

they don't have anything else as strong as the signal in the individual detectors

That's not clear to me, they have published the results of events they have filtered out as 'PeterDonis' said:

Signals which look like GW signals but only appear in one detector, which is how an "unknown origin" signal would be expected to behave, don't pass this filter.

There is no talk of possible other strong signals, that doesn't mean that they have none, and that was exactly my question.

Yes, we know for those events that the signal was more than clear. But what outside of these filtered time periods?

 

[mfb]

That filter is used to search for signal events, not for the background estimate.

There is no talk of possible other strong signals, that doesn't mean that they have none, and that was exactly my question.

They have a histogram of the significance distribution.

 

[auou]

They have a histogram of the significance distribution.

What is this, do you have a link?

 

[mfb]

In the original publication, figure 4.

Note how the background estimate at intermediate detection statistics in 4b is completely dominated by combining one of the signal spectra with random noise from the other detector.

 

[auou]

The graph gives the filtered results that 'PeterDonis' mentioned and not the individual signals that I asked about:

The search reconstructs signal waveforms consistent with a common gravitational-wave signal in both detectors using a multidetector maximum likelihood method.

Do you know of a reference of other non-filtered events?

 

[mfb]

There is nothing filtered - apart from data not used in either the search or the background estimate. The whole dataset used to search for events is used for the background estimate as well.

 

[auou]

There is nothing filtered - apart from data not used in either the search or the background estimate. The whole dataset used to search for events is used for the background estimate as well.

Sorry, but I am not seeing it in Fig. 4 of the paper. The first graph has value ηc and the second one ρˆc and both formulas are based on combined data of the two detectors:

• "The statistic ηc thus quantifies the SNR of the event and the consistency of the data between the two detectors."

• "The final step enforces coincidence between detectors by selecting event pairs that occur within a 15-ms window and come from the same template. The 15-ms window is determined by the 10-ms intersite propagation time plus 5 ms for uncertainty in arrival time of weak signals. We rank coincident events based on the quadrature sum ρˆc of the ρˆ from both detectors."

 

[mfb]

If you combine a strong signal-like pattern in one detector with random noise in the other, you get intermediate $\hat \rho_c$ values sometimes. That's what you see in 4b between 13 and 21 for this value: Combinations of the GW signal from one detector with random data from the other (from a different point in time). After they remove the two signal-like patterns from GW150914 from the analysis, nothing of this remains. If there would be another similar signal-like pattern (from a single detector) somewhere in the dataset you would still get some entries in this intermediate range. But there is absolutely nothing.

 

[auou]

Yes, but this doesn't answer my original question:

Would there be a list of 'single' strong detections …

Fig. 4 doesn't tell us anything about any other signals that appeared in only one of the detectors. The graphs are both about combinations as you point out.

I am not sure if you understood my question.

 

[mfb]

Fig. 4 doesn't tell us anything about any other signals that appeared in only one of the detectors.

It does. The graph would look completely different if there would be other signal-like features (of comparable strength) in only one of the detectors.

 

[auou]

It does. The graph would look completely different if there would be other signal-like features (of comparable strength) in only one of the detectors.

Let's say there was such a signal in only one of the detectors where would it be positioned, because as mentioned in my previous post the formulae used are based on data of the two detectors. So if it doesn't show up in the other the 'statistic' goes all the way down.

This intrigues me as to were an exactly similar signal detected at only one detector would show up in the graph?

 

[mfb]

If there would have been such a signal, figure 4b "without GW150914" would have entries in the intermediate test statistics range. It does not have them.

So if it doesn't show up in the other the 'statistic' goes all the way down.

It is lower than 32, sure. But still higher than all the background they actually see.

This intrigues me as to were an exactly similar signal detected at only one detector would show up in the graph?

In the intermediate test statistics range. The entries "with GW150914" are exactly the result of such a signal from one detector correlated with unrelated data from the other one.

 

[auou]

If there would have been such a signal, figure 4b "without GW150914" would have entries in the intermediate test statistics range. It does not have them.

Mh, this seems to conflicts with what the Danish group wrote in their paper:

"There are, however, a few events that are morphologically similar to the GW150914 event for the 0.1 s case with cross correlators at the level of 0.3–0.6. In order to illustrate these features of the H4096 and L4096 records, we have considered the following four events from Fig. 7 with the approximate coordinates A = (0.6,0.4), B = (0.5,−0.5), C = (0, −0.7) and D = (−0.65, −0.3). All of these events are characterized by a relatively large high level of positive or negative cross-correlations with GW150914. Event A is pointed in the direction of GW150914 and seems to be morphologically close to it." - https://arxiv.org/abs/1609.08346

 

[mfb]

What the Danish group did is not what LIGO did. The Danish group didn't even consider how LIGO estimated the background.
You can get large correlation coefficients with noise alone. That is not the right way to search for signals. You have to consider the amplitude. LIGO did that, the Danish group did not.

Where exactly is the text you quoted?

 

[auou]

Where exactly is the text you quoted?

Sorry, I had pasted the link of their other paper it should have been this one: https://arxiv.org/pdf/1604.06211.pdf?

at topic 5 'The search for similar morphology in the 4096 second records' on page 9

 

[mfb]

Thanks. See above: That is not the right way to estimate a background.

 

[auou]

Thanks. See above: That is not the right way to estimate a background.

Mh, have you read the response from the Danish group to Harry that I posted earlier:
http://www.nbi.ku.dk/gravitational-waves/gravitational-waves.html

In short they say that there's an issue with the cleaning of the noise. The correlation of the noise is in the same time window as the signal, while:

The purpose in having two independent detectors is precisely to ensure that, after sufficient cleaning, the only genuine correlations between them will be due to gravitational wave effects.

Anyway to come back to my question, there is also a new article by 'Quantamagazine' where they write:

Lots of little bumps and vibrations can mimic a gravitational-wave signal … https://www.quantamagazine.org/strange-noise-in-gravitational-wave-data-sparks-debate-20170630/

 

[mfb]

They make a lot of claims.
Their claim that you should see exactly (?) zero correlation in the residuals is one of the worst ones. You only expect this if the template matches exactly. No one expects that, and LIGO explored possible deviations. Again something the Danish authors seem to ignore.

the only genuine correlations between them will be due to gravitational wave effects.

And random fluctuations. Not every non-zero correlation is directly something interesting. You expect a lot of non-zero correlation events in noise just by random chance. They will all have small amplitudes. Did I mention that you have to look at the amplitude?

Lots of little bumps and vibrations can mimic a gravitational-wave signal

But not with the signal strength LIGO saw.

 

[auou]

They make a lot of claims.
Their claim that you should see exactly (?) zero correlation in the residuals is one of the worst ones.

…

But not with the signal strength LIGO saw.

The article claimed the same thing:

The noise at each detector should be completely uncorrelated — a jackhammer going off in the town near one detector won’t show up as noise in the other.

BTW the signal of 'a jackhammer going off' should at least have the same 'signal strength' if not stronger. I speak of personal experience they have been doing construction works in my street the past few weeks and at times I could feel my apartment shake, in contrary I have never felt a GW.

 

[mfb]

The article claimed the same thing:

It is not the same thing, and the difference is what the Danish authors are missing.

"has no expected correlation" and "will be measured with a correlation coefficient of exactly zero" are different things, and "noise" and "residuals after template subtraction" are different things as well.

BTW the signal of 'a jackhammer going off' should at least have the same 'signal strength' if not stronger.

And the seismic sensors detect it and it doesn't enter data analysis.

they have been doing construction works in my street the past few weeks and at times I could feel my apartment shake, in contrary I have never felt a GW.

Can we keep the discussion scientific please?

 

[auou]

the seismic sensors detect it and it doesn't enter data analysis.

Question here is were/how to draw the line?

First you mention that the other signals are not strong enough, but it turns out other (strong) signals are not accepted.

 

[mfb]

Question here is were/how to draw the line?

The sensitivity depends on that line, but in the worst case you miss a signal - you don't get a significant signal where there is no signal.

First you mention that the other signals are not strong enough, but it turns out other (strong) signals are not accepted.

You are mixing two completely different analysis steps.

 

[auou]

You are mixing two completely different analysis steps.

OK. I guess the article is a bit misleading when they write: 'a jackhammer going off in the town near one detector won’t show up as noise in the other'. It gave me the impression that the drill showed up at least in the noise of that one detector.

So there are 2 (or more) separate levels of noise, how many filters are there?

BTW I took an other look at this remark:

Their claim that you should see exactly (?) zero correlation in the residuals is one of the worst ones.

Their claim seems to be rather that the GW signal is small compared to the overall noise, and that it's difficult to get a signal so different from a GW waveform noise correlation. They even allowed 10% leeway for the templates and it change nothing for the correlation.

 

[mfb]

So there are 2 (or more) separate levels of noise, how many filters are there?

LIGO published a lot how exactly they analyze their data.

The GW signal is huge compared to the residuals where the Danish group is looking for correlations. Even tiny differences between fitted template and the gravitational wave will lead to large correlations in the residuals.

They even allowed 10% leeway for the templates and it change nothing for the correlation.

10% in what? The template is not a one-dimensional number.
LIGO checked all this, as every good scientific analysis does.
The Danish group did not. Or if they did, they didn't discuss the results, which would be even worse.

 

[auou]

10% in what?

'10% scaling of the amplitude of the templates'

 

[mfb]

That's my point. The amplitude is not everything.

 

[auou]

Ok, one last thing that kept hanging in my mind was this comment in the Quantamagazine article about the raw data:

“The only persons qualified to analyze this paper are in the LIGO Scientific Collaboration,” said https://physics.stanford.edu/people/faculty/robert-wagoner, a theoretical physicist at Stanford University who is not affiliated with LIGO. “They are the only ones who have had access to the raw data.”

And this related to your comment on the seismic sensors that filter out the vibrations of a drill:

the seismic sensors detect it and it doesn't enter data analysis.

Does this mean that some signals like that of the drill nearby aren't in the provided raw data on the LIGO website? Or are those sensors automatically filtering out seismic vibrations with a kind of mechanical damping detector?

 

[mfb]

If the seismic sensors detect some activity, the measurements in this time period are not used to search for gravitational waves. I guess the raw data provided by LIGO is only from times where no seismic activity was detected.

 

[Auto-Didact]

They make a lot of claims.
Their claim that you should see exactly (?) zero correlation in the residuals is one of the worst ones. You only expect this if the template matches exactly. No one expects that, and LIGO explored possible deviations. Again something the Danish authors seem to ignore.
And random fluctuations. Not every non-zero correlation is directly something interesting. You expect a lot of non-zero correlation events in noise just by random chance. They will all have small amplitudes. Did I mention that you have to look at the amplitude?

The problem is not that there are (or seem to be) non-zero correlations, the problem is that the non-zero cross-correlations which are seen are extremised and close to |1| nearly exactly for the delay times between detectors for GW signals, i.e. for 7 ms, 1 ms and -3 ms for respectively the first, second and third GW signals.

These exact same correlations are also found using null output sets, which are null input sets made by subtracting the unfiltered theoretical GW templates from the raw data sets and subsequently cleaning these input sets per LIGO's cleaning prescriptions.

Their claim seems to be rather that the GW signal is small compared to the overall noise, and that it's difficult to get a signal so different from a GW waveform noise correlation. They even allowed 10% leeway for the templates and it change nothing for the correlation.

Their claim is neither that you should see zero correlation nor that the GW signal is small compared to the actual noise. Their claim is instead:

While our findings do not contradict the previous statement about near Gaussianity during the time of the events, this is to be contrasted with the present demonstration that the residuals show apparent correlations between the detectors. It is striking that these correlations are maximized by applying nearly the same time shifts as found for the GW events themselves — for all three GW events reported to date. The purpose in having two independent detectors is precisely to ensure that, after sufficient cleaning, the only genuine correlations between them will be due to gravitational wave effects. The results presented here suggest this level of cleaning has not yet been obtained and that the detection of the GW events needs to be re-evaluated with more careful consideration of noise properties.

 

[mfb]

The problem is not that there are (or seem to be) non-zero correlations, the problem is that the non-zero cross-correlations which are seen are extremised and close to |1| nearly exactly for the delay times between detectors for GW signals, i.e. for 7 ms, 1 ms and -3 ms for respectively the first, second and third GW signals.

That is exactly what you expect if the templates are not perfect.

 

[Auto-Didact]

Why exactly?

 

[mfb]

Every deviation between GW and template will lead to correlated residuals, with the same time delay as the GW and the template.
No one expects that the template fits perfectly, so yes, what the authors observe here is expected to some extent. I don't understand what all the noise is about (scnr).

 

[Auto-Didact]

Would such correlations be expected as well in raw or cleaned data sets of both detectors which do not contain any GW signals?

And can you elaborate on the phase correlations mentioned here:

The cleaned data should contain signal and residual noise. Although the strength of the signal and noise are comparable during the 0.2 s of the GW150914 event, noise is completely dominant for the remainder of this 32 s record. In other words, the Fourier amplitudes shown here are noise dominated. Since random noise would have a uniform distribution of phases, it is evident that the noise is neither stochastic nor even roughly stochastic. We also note that plots of the phases of the 4096 s data show similar correlations. Such correlations can either be numerical artifacts or a genuine consequence of the physically meaningful coupling of nearby frequencies. We point out the rather surprising fact that the phase correlations in the Livingston detector (middle row in Fig. 3), already present in the raw data, are significantly increased by the cleaning procedure. This suggests that their origin may be physical. Whatever its origin, the non-stochastic behavior indicated by phase correlations in the supposedly clean data immediately presents an a priori challenge to the reliability of any significance estimates of possible GW events.

 

[mfb]

Would such correlations be expected as well in raw or cleaned data sets of both detectors which do not contain any GW signals?

As random fluctuations, sure, but not beyond that. And that's what the Danish authors find. A strong correlation within the signal, a very weak correlation outside.

And can you elaborate on the phase correlations mentioned here:

I think the blog article linked earlier covers that.

 

[timmdeeg]

During the visit at LIGO Lousiana quite recently (organized by the german Journal "Bild der Wissenschaft" for our group) there was the opportunity to ask the scientist who spoke to us how LIGO would respond to the Danish group (primary source mentioned in #4). He said that there isn't any correlation, if one looks at the whole noise signal, which wasn't shown in the LIGO publication and that this was stated already in response to the paper of the Danish group. I'm not sure however if he meant the article of one of the LIGO postdocs mentioned in #21 or an official answer of LIGO which I'm not aware of.

 

[auou]

There's a new signal detected now along with VIRGO in Italy:

https://dcc.ligo.org/LIGO-P170814/public/main

It keeps on being a bit weird, in the SRN graphs they look very different and in the Whitened Strain* they suddenly look the same.

*The whitening emphasizes different frequency bands for each detector, which is why the reconstructed waveform amplitude evolution looks different in each column.

 

[mfb]

The SNR plots have a single peak for all three detectors, as you would expect for a single event. It is an integrated SNR.

I wonder what the Danish group will do now. Admitting that the LIGO analysis method is sound? I doubt it.

 

[auou]

The SNR plots have a single peak for all three detectors, as you would expect for a single event. It is an integrated SNR.

I wonder what the Danish group will do now. Admitting that the LIGO analysis method is sound? I doubt it.

Again they show just a very small sample, and have probably used templates to get the match for the withened strain, the peaks in the SRN are still very different.

I'd like to see what a 30 sec. stretch looks like.

Also how fast is the Earth moving (turning) in relation to the signal and how much does the angle changes of the location in Italy vs. those in the US over that period?

 

[Ibix]

Also how fast is the Earth moving (turning) in relation to the signal and how much does the angle changes of the location in Italy vs. those in the US over that period?

The Earth is 12,800km in diameter. Maximum possible delay between reception is therefore a little over 0.04s (for a lightspeed signal). That works out to about 0.64 seconds of arc as an upper bound on angle change.

 

[auou]

The Earth is 12,800km in diameter. Maximum possible delay between reception is therefore a little over 0.04s (for a lightspeed signal). That works out to about 0.64 seconds of arc as an upper bound on angle change.

Yes, but we are also moving at 627 km/s relative to the CMB. So how much would the signal be skewed?

 

[Ibix]

Yes, but we are also moving at 627 km/s relative to the CMB. So how much would the signal be skewed?

Skewed in what sense? I think that the difference between Earth-centred and FLRW co-moving coordinates on the scales LIGO cares about is just a Lorentz transform, although I could be being overconfident there.

 

[auou]

Skewed in what sense?

That one detector runs into the wave head on and the other sideways.

 

[mfb]

That is just a relative motion of the source and the detectors. It has nothing to do with the CMB. And it is negligible at the current level of sensitivity. Redshift is relevant (~10%).

Again they show just a very small sample, and have probably used templates to get the match for the withened strain, the peaks in the SRN are still very different.

Of course they used templates. That's the analysis method.
The peaks in the SNR are not expected to be the same. The different detectors have different sensitivities and they have different orientations.

You are trying to make up issues that don't exist.

 

[Ibix]

That one detector runs into the wave head on and the other sideways.

Unless there's some reason you can't use standard aberration formulae, failing to distinguish between the Earth-centred inertial frame and the local FLRW co-moving frame will knock your angle estimates off by about 1/500 of a radian, or around a tenth of a degree (so says the back of an envelope).

I find it difficult to imagine people capable of planning on generating gravitational wave solutions for different source types forgetting about the possibility that the sources are in motion.

 

[auou]

It has nothing to do with the CMB.

Of course it has got nothing to do with the CMB itself, but the CMB acts like a reference frame.

The detectors could be moving away or running into the GW, or be hit from the side, skewing the signal. So it all depends on running into the wave at a particular time of day (orientation), a few hours later and we wouldn't notice a thing if the signal is perpendicular.

 

[auou]

I find it difficult to imagine people capable of planning on generating gravitational wave solutions for different source types forgetting about the possibility that the sources are in motion.

Imagine having a flag to define where the wind comes from, now start waving that flag around, and things become a lot more complicated.

 

[Nugatory]

Imagine having a flag to define where the wind comes from, now start waving that flag around, and things become a lot more complicated.

Yes, but is that a good analogy for this observational setup? If the flag were planted on a glacier, you wouldn't worry so much that the glacier's movement might mess up your measurements of wind speed and direction.
So do the calculation. What is an upper bound on the error introduced from the relative motion between different points on the surface of the rotating and orbiting earth? Can this effect be large enough toffee to the validity of the reported results?

 

[auou]

Yes, but is that a good analogy for this observational setup? If the flag were planted on a glacier …

No, it's not perfect. A flag has many 'points' that serve as a reference to define direction etc. LIGO and Virgo combined only 3.

 

[Nugatory]

No, it's not perfect. A flag has many 'points' that serve as a reference to define direction etc. LIGO and Virgo combined only 3.

Yes, so we have an example of why we do science with math instead of analogies.

Until you've quantitatively demonstrated an upper bound on the possible error produced because we only have three points and they are in relative motion you're just speculating idly.

 

[Dr. Courtney]

Yes, so we have an example of why we do science with math instead of analogies.

Until you've quantitatively demonstrated an upper bound on the possible error produced because we only have three points and they are in relative motion you're just speculating idly.

I don't consider that to be how confidence works in science if we apply the maxim that extraordinary claims require extraordinary evidence.

I tend to put the burden of proof that potential sources of error are smaller than their error bars on those publishing or defending extraordinary new experimental claims rather than on skeptics suggesting potential sources of error.