Penrose Gurzadyan fire back: new paper today

[marcus]

Penrose and Gurzadyan are not giving up without a fight. They posted a second paper today, answering their critics:

http://arxiv.org/abs/1012.1486
More on the low variance circles in CMB sky
V.G.Gurzadyan, R.Penrose
2 pages
(Submitted on 7 Dec 2010)
"Two groups [3,4] have confirmed the results of our paper concerning the actual existence of low variance circles in the cosmic microwave background (CMB) sky. They also point out that the effect does not contradict the LCDM model - a matter which is not in dispute. We point out two discrepancies between their treatment and ours, however, one technical, the other having to do with the very understanding of what constitutes a Gaussian random signal. Both groups simulate maps using the CMB power spectrum for LCDM, while we simulate a pure Gaussian sky plus the WMAP's noise, which points out the contradiction with a common statement [3] that 'CMB signal is random noise of Gaussian nature'. For as it was shown in [5], the random component is a minor one in the CMB signal, namely, about 0.2. Accordingly, the circles we saw are a real structure of the CMB sky and they are not of a random Gaussian nature. Although the structures studied certainly cannot contradict the power spectrum, which is well fitted by LCDM model, we particularly emphasize that the low variance circles occur in concentric families, and this key fact cannot be explained as a purely random effect. It is, however a clear prediction of conformal cyclic cosmology."

 

[Chronos]

I would love to hear from a hard core statistician on this one. I do not believe you can prove a non-random distribution within the error bars Penrose is claiming. Apparently he is asserting you can not prove a purely random distribution in a finite data set - which is always true in statistics.

 

[Chalnoth]

Ugh, what they responded with was truly, truly sad. I'm trying to be nice here, but there's not much of anything good to say about a couple of sentences like this:

Both groups simulate maps using the CMB power spectrum for LCDM, while we simulate a pure Gaussian sky plus the WMAP's noise, which points out the contradiction with a common statement [3] that 'CMB signal is random noise of Gaussian nature'. For as it was shown in [5], the random component is a minor one in the CMB signal, namely, about 0.2. Accordingly, the circles we saw are a real structure of the CMB sky and they are not of a random Gaussian nature.

Apparently they are conflating "Gaussian" with "uncorrelated". The two are not the same thing. As I explained in the other thread, $\Lambda$CDM predicts a very specific correlation pattern on the CMB. This pattern of correlations is basically an interference pattern of the pressure waves in the early universe, and depends solely upon the angular separation.

Without modeling these correlations, you get the wrong answer. Period.

 

[George Jones]

Ugh, what they responded with was truly, truly sad. I'm trying to be nice here, but there's not much of anything good to say about a couple of sentences like this:
Apparently they are conflating "Gaussian" with "uncorrelated". The two are not the same thing. As I explained in the other thread, $\Lambda$CDM predicts a very specific correlation pattern on the CMB. This pattern of correlations is basically an interference pattern of the pressure waves in the early universe, and depends solely upon the angular separation.

Without modeling these correlations, you get the wrong answer. Period.

Okay, I'm going in way over my head here; I'm going to talk about stuff about which I know nothing.

I think that Gurzadyan and Penrose are well aware of the "specific correlation pattern on the CMB."

From the second paper by Gurzadyan and Penrose:

First, there is difference between our treatments of the Gaussian significance of the low variance circles. In [3,4] the authors simulate the CMB maps, as is commonly done, using the power spectrum parameters for LCDM plus the WMAP noise. As a result, they find standard deviation around 5μK, and hence conclude that there is a lower significance to the circles, e.g. around 3σ at 15μK depth, than that which we have found, where we adopted a more appropriate procedure, namely to simulate an isotropic Gaussian signal, to see whether such circles are result of random fluctuations. Below, in addition what is given in [1], we represent the standard deviations for 3 types of simulated Gaussian maps: Gaussian map simulated for parameters of (1) W-band (of all horns); (2) corresponding foreground reduced; (3) foreground reduced plus the WMAP’s noise:

          Source                                 Sigma (μK)

Simulated Gauss W                                  2.5
Simulated Gauss W FR                               2.2
Simulated Gauss W FR + WMAP’s noise                2.4

From this, we see, as is also correctly mentioned in [3,4], that the WMAP signal is indeed dominated by a cosmological signal rather than by an instrumental noise. However, we see that the sigma in the simulated data is only half the one claimed in [3] for LCDM, hence the corresponding increase in the Gaussian significance of the signal that we find in the actual CMB data.

The sentence after the table indicates that they are using isotropic Gaussian sources to model instrument noise.

What does the last quoted sentence mean? Does it mean that they infer that their concentric circles are not the result of isotropic Gaussian instrument noise? If it does, is this the wrong way to look at things?

Any answers should assume I know little about statistics in general (Many, many years ago, I took an introductory statistics course, and I have not used statistics in any (statistically) significant way since.), and almost nothing about statistics applied to the CMB.

 

[Chalnoth]

Okay, I'm going in way over my head here; I'm going to talk about stuff about which I know nothing.

I think that Gurzadyan and Penrose are well aware of the "specific correlation pattern on the CMB."

Sadly, they don't write as if they are.

The problem is that they pooh-pooh the analysis by Wehus, Eriksen, Moss, Scott, and Zibin for doing precisely this. The analysis done by these two other groups did the same thing my mini-analysis did: simulated a CMB sky in harmonic space (the usual thing), then converted it to real space.

The reason this is done is that in harmonic space, because the CMB is statistically isotropic, the components are statistically independent. So you don't need any sort of covariance matrix or anything fancy like that: you just need the variance of the fluctuations for each scale (the power spectrum).

Now, when you convert this back to real space, the fact that different scales have different-amplitude fluctuations leads to correlations on the sky. Gurzadyan and Penrose clearly failed to recognize this crucial point. In the body text they say they use an "isotropic distribution," as a contrast to what the others did (as you quoted). The thing is, the analysis the others did was an isotropic distribution already. So the only conclusion I can make is that they think isotropic means uncorrelated. Which it doesn't.

The sentence after the table indicates that they are using isotropic Gaussian sources to model instrument noise.

What does the last quoted sentence mean? Does it mean that they infer that their concentric circles are not the result of isotropic Gaussian instrument noise? If it does, is this the wrong way to look at things?

I'm not entirely sure. Sounds like they did something wrong. It could be down to this exact same point of not taking into account the correlations.

 

[Lievo]

Let's assume that Moss et al. and Wehus et al. correctly demonstrated that single concentric low-variance features are as often in randomly reconstructed CMB and in the true one. This is not a strong assumption given that they are described as very smart for this kind of stuff.

However, the fact is that this is not Gurzadyan and Penrose's main claim.

Our final point is a key issue that appears not to be taken seriously into account by [3,4]. What we find are families of concentric low-variance circles (as predicted by the cosmological scheme of [2]), not just individual low-variance circles. The probability of finding such families is clearly immensely smaller, for Gaussian random fluctuations, than finding the same number of unrelated individual low-variance circles.
The simulations obtained by [3], including those using equilateral triangles rather than circles, show no indication of the low-variance instances occurring in concentric families, which is what we indeed find with the circles.

Too bad Moss et al. and Wehus et al. apparently both forgot to test the main claim. Is there any reason they miss this point?

PS: fun to see that basically Chalnoth did the same job days before (with https://www.physicsforums.com/showpost.php?p=3015529&postcount=30" but that's because it was just a quick look with no intent to publish -I have to say I'd have expected better from published refutation)

 

[Chalnoth]

Except for the fact that Gurzadyan and Penrose don't actually show evidence that there are such concentric circles, let alone compare the same sort of search with a Gaussian realization of the CMB.

 

[Lievo]

Except for the fact that Gurzadyan and Penrose don't actually show evidence that there are such concentric circles, let alone compare the same sort of search with a Gaussian realization of the CMB.

I do agree this a bad.

 

[twofish-quant]

Also I have the problem with the statement that the probability of finding concentric circles is "clearly immensely lower" since it's not clear to me that this is the case. Once you have correlations, then the probability of having these sorts of connections is higher than you would naively expect (which is in fact the definition of correlation).

 

[Lievo]

Once you have correlations, then the probability of having these sorts of connections is higher than you would naively expect (which is in fact the definition of correlation).

I couldn't say better. G&P claimed that if x is the probability to find one circle, then the probability to find n circles is xⁿ. This is clearly naive. In fact, Chalnoth found at leasthttps://www.physicsforums.com/showpost.php?p=3015028&postcount=29" in a random reconstruction of the CMB, and maybe 7 depending on how you count it. If one believes G&P, the probability that this happen is 10^-21 for n=3 and 10^-49 for n=7. Does Chalnoth use to be hitten by a meteorit every morning or every microsecond?

So yes, G&P need to document their claim better. However, the fact remains that G&P's main claim has not been tested properly yet, neither by themselves nor their contradictor. Maybe next week?

 

[Lievo]

Some echo and another refutation:

http://www.nature.com/news/2010/101210/full/news.2010.665.html
http://arxiv.org/PS_cache/arxiv/pdf/1012/1012.1656v1.pdf

...still no test of the main claim yet.

 

[Chalnoth]

Some echo and another refutation:

http://www.nature.com/news/2010/101210/full/news.2010.665.html
http://arxiv.org/PS_cache/arxiv/pdf/1012/1012.1656v1.pdf

...still no test of the main claim yet.

Just showing that the specific circles that Gurzadyan and Penrose pointed out are not statistically significant deviations from the norm is a pretty succinct refutation of the claims in their paper, as far as I'm concerned.

 

[Lievo]

Sure it's pretty succinct. And maybe succint and wrong.

 

[Chalnoth]

Sure it's pretty succinct. And maybe succint and wrong.

Pretty sure it's succinct and correct :)

 

[Lievo]

Pretty sure it's succinct and correct :)

Well then there is something you should explain. Where can I find, in any of the published refutations, a statistical test excluding G&P main claim, i.e. that there exists anomalies in the CMB regarding multiple rings specifically?

 

[Chalnoth]

The primary point here is that these other people all replicated Gurzadyan and Penrose's supposed examples of concentric circles, and proceeded to show that they were entirely expected within $\Lambda$CDM. There's really no point in going further, as any claims made in the text are supposedly a consequence of the data. If they aren't a consequence of the data, then they aren't worth paying any attention to.

 

[Lievo]

The primary point here is that these other people all replicated Gurzadyan and Penrose's supposed examples of concentric circles, and proceeded to show that they were entirely expected within $\Lambda$CDM. There's really no point in going further, as any claims made in the text are supposedly a consequence of the data. If they aren't a consequence of the data, then they aren't worth paying any attention to.

Chalnoth, suppose I pretend having found 1000 circles while expecting 200, and that the display of the circle make the following sentence 'HEY IS THERE ANY BODY OUT THERE?WE ARE IN YOUR PRECEDING EON. PLEASE ANSWER BY SAME CHANNEL.' Would you really pretend that my main claim would be about having more circles than expected?

This is exactly the situation. G&P claim to have more circles than expected and that the circles form some pattern. They are likely wrong about the number of expected circles, but to say that it is enough to rebut their claim... this is just overstated.

 

[Chalnoth]

Chalnoth, suppose I pretend having found 1000 circles while expecting 200, and that the display of the circle make the following sentence 'HEY IS THERE ANY BODY OUT THERE?WE ARE IN YOUR PRECEDING EON. PLEASE ANSWER BY SAME CHANNEL.' Would you really pretend that my main claim would be about having more circles than expected?

Simply claiming something is completely insufficient. They have to actually show it. And they have completely and utterly failed to do that, as the response papers have demonstrated. Another way to say it is that the burden of proof is on the one making the positive claim. It's their claim, let them show it. It would require some work on their part. But they should be willing to put up with a little bit of work if they want to, you know, do science.

 

[marcus]

... display of the circle make the following sentence 'HEY IS THERE ANY BODY OUT THERE?WE ARE IN YOUR PRECEDING EON...

:rofl:

I like this way of talking about the curious discovery that the circles come in concentric families.

If born out, the concentricity is a striking observation.

It also would fit well with the idea of supermassive BH mergers sending out gravity ripples---since mergers would tend to occur repeatedly in large clusters of galaxies: hence repeatedly in roughly the same location.

Lievo, I gather this is the "sentence" you are talking about? Please correct me if I am mistaken.

I would imagine that eventually somebody will work out quantitatively the probability/improbability of circles coming in neat concentric families instead of just randomly scattered. If that gap in the picture eventually gets filled in by someone's paper (Gurzadyan or other) then maybe we don't have to make a big deal about the schedule on which the papers arrive?

 

[Lievo]

Lievo, I gather this is the "sentence" you are talking about? Please correct me if I am mistaken.

Yes. I'm happy you got it despite my poor english.

I would imagine that eventually somebody will work out quantitatively the probability/improbability of circles coming in neat concentric families instead of just randomly scattered. If that gap in the picture eventually gets filled in by someone's paper (Gurzadyan or other) then maybe we don't have to make a big deal about the schedule on which the papers arrive?

Sure.