(Note: This "email" is not intended for posting, as PRA has declared the matter closed and it says nothing I had not told them before.)
Date: Sun, 8 Aug 1999
From: c.h.thompson@newscientist.net
To: PR/A-email
Subject: ac7179 and av6320
Cc: pra@aps.org
AV6320
How Bell tests fail: The ``loopholes'' that favor local realism
Thompson,Caroline H./
AC7179 Subtraction of ``accidentals'' and the validity of Bell tests Thompson,
Caroline H./
Dear Dr. Crasemann:
I am sorry that you have again rejected my paper, and again, it seems, without giving me the opportunity to discuss points with the referees. Though I had tried to accomodate the points raised in relation to the previous submission, there are some matters that are best discussed between author and referee. These are the numerous matters that might be classed as "opinion". Whilst I should be happy to revise the manuscript to clarify if necessary, I cannot accept that the referees should have more rights in this respect than the authors.
There are clearly some points on which the referees have no first-hand information and have therefore resorted to quoting accepted beliefs, i.e. they appear to think that their opinions should carry more weight than mine.
I give below my detailed responses to their reports.
Yours sincerely
Caroline H Thompson
__________________________________________________________________
Response to Report of the First Referee:
I wonder how you arrive at your opinion that "AC7197 ... contains no new results, no practical suggestions, no proposals for new experiments"?
New results:
The main results that are effectively new are those on accidentals in Alain Aspect's experiments. These they have hitherto appeared only in unsorted and unsummarised form in his PhD thesis - a document that is still, I believe, available only in French. So far as the world at large is concerned, Aspect's experiments represented a critical point in the history of science - the point at which it was established experimentally that weird quantum effects actually happened. I maintain that the fact that, in two (the first and last) out of his three published experiments, the "violations of Bell tests" were achieved only as a result of adjusing data in a manner that has never been justified would, if it were widely known, cause many people to think again. As the third (central) experiment used a test that was invalid for other reasons, being susceptible to the "detection loophole", weird effects have not been demonstrated!
Practical Suggestions and New Experiments:
I may not have spelled these out as clearly as I intended, but I certainly mentioned the suggestion that Aspect's experiments should be repeated using reduced emission rates, so that accidentals became negligible.
More recent experiments such as Tittel's 1997 one should clearly be repeated in conditions in which there was less "noise", which could perhaps be achieved by omitting the long fibre cables. Though his 1998 experiment did not require the subtraction of accidentals, this is not because of any improvement in this respect but only because he used a different test. He used one that had the same critical weakness as that used in Aspect's central experiment.
It is not clear from the published papers which other experiments are biased by accidentals, as not all mention the point.
As mentioned in the discussion section of my paper, I think that all demonstrations of entanglement need to be reinvestigated. I may not have made it clear that what I mean is that comprehensive sets of experiments are needed, looking at all loopholes simultaneously. Recent experiments such as that of Weihs et al (quant-ph/9810080) cannot establish non-locality as they block only one loophole at a time.
Referee 1:It does however contain some irrelevant information (see for example the end of the Introduction, page 3; there are many more examples in the text and in the footnotes) ...
CHT: This is largely a matter of opinion.
Referee 1: ... as well as 3 appendices in which previous work (by other authors) is reported. The author claims that these derivations "are rarely reproduced." I wonder why they should be reproduced: they were published in PR and PRL, they are available on the web, and I have been able to find them even in universities of developing countries.
CHT: In a perfect world, agreed, there would be no call to reproduce them. My correspondence with various people has convinced me, though, that (a) most quantum theorists are happy with the Clauser/Horne/Shimony/Holt 1969 paper and prefer to ignore the Clauser/Horne 1974 one and (b) most people are too busy to look up references. My purpose in making the material so accessible is to illustrate a proof of Bell's theorem that is mathematiclly rigorous, based on clear assumptions, easy to follow, and applicable (given the absence of timing problems and "no enhancement") to real situations.
Referee 1: The model in Eqs.(3)-(5) has been discussed many times in the literature.
CHT: I reproduce the standard realist model as it is evident that it is not known to everyone concerned. If it were, it would be universally understood that the mere observation of periodic (approximately sinusoidal) variations in a coincidence curve was not evidence of quantum correlations. Indeed, it is clear, as I point out, that a slight relaxation of the assumptions (consistent, I might add, with the general philosophy of realism and with a pure wave theory of light) leads to coincidence curves of 100% visibility.
Referee 1: I would like to stress that one of the main criteria for a paper to be published in PRA is that it contains "new results in physics."
CHT: My aim in this paper is to quantify a fact that is, in my opinion, universally true: that all quantum entanglement experiments have achieved their "violations of Bell inequalities" purely as a result of the various loopholes. Thus my "new" result is that a set of results that has hitherto been accepted as supporting quantum theory does not in fact do so. Yes, others have many times tried to put this message across, but I believe that the data I present is new, stronger, evidence than has ever been published before. Others have presented realist theories, without explaining the numerical facts. I have shown that the subtraction of accidentals can mean truly enormous changes in Bell tests. This information is new.
__________________________________________________________________
Responce to Report of the Second Referee on MS AC7179 by C. H. Thomson
Referee 2: In this paper the author questions the validity of the procedure consisting in substracting accidental coincidences in experimental tests of Bell's inequalities performed by A. Aspect and collaborators.
As far as I know, this question was previously raised by T.W. Marshall and coll. (ref [3] of the present paper), and it was answered by Aspect and coll. in ref. [6], by doing an experiment where a clear violation of Bell's inequalities was obtained without any substraction of accidentals. This confirmed a previous experiment done in 1972 by Freedman, who used another source and different polarizers.
CHT: This is all undisputed and clearly stated in my paper. It is my opinion that T.W.Marshall et al accepted Aspect and Grangier's "new data" too easily. The data I present in my paper has not, to my knowledge, previously been analysed. This is a grave failure.
Referee 2: However, the author simply discards the experiment of ref. [6], just because it is done with two-channel rather than one-channel polarizers. This argument appears to me to be very unphysical.
CHT: These statements show misunderstanding Clauser and Horne's 1974 paper. It might help if you were to read my Chaotic Ball paper (quant-ph/9611037). It is not so much a matter of the difference between one and two-channel experiments as simply that the standard test used in two-channel experiments has no theoretical justification. Its origins are now lost is the mists of time, but it was never "derived" by anyone, let alone John Bell.
Aspect quoted a paper by Garuccio and Rapisada (Garuccio, A and Rapisada, V, Nuovo Cimento A 65, 269 (1981)) in justification of the test, but had he read it carefully? It in fact merely shows that realist models can be found that satisfy the "fair sampling" assumption. Most realist models - ones that reflect actual experimental conditions - do not.
Aspect also (as is perhaps scarcely known to anyone) thought he could show that any bias introduced as a result of failure of fair sampling could not be large. In his thesis he derives a modified Bell test, and presumably he is referring to this in a footnote to the relevant paper (PRL 49, pp 91-94, 1982, footnote 9, concerning generalised Bell inequalities "to be published elsewhere"). The derivation ought to be checked, as I am reasonably sure that it is flawed.
Referee 2: It seems that when getting the answer to one question, the author refuses to hear it by shifting to another question.
CHT: I hope I have now clarified this.
Referee 2: It is not so clear why Freedman's experiment is also discarded, except again if the author does not want to hear the answer.
CHT: I do mention that there is the possibility of bias due to timing problems, with his use of a very short coincidence window (6 ns, in circumstance where at least some of the signals he was detecting were likely to have been longer than this). In ealier editions of my paper (including AV6320) I make it clear that I think these are the most likely explanation for Freedman's results. The matter could be investigated.
Referee 2: I have a few more technical remarks : * the tests of IB done with one-channel polarizers must use the so-called "no enhancement" hypothesis, while the tests done with two-channel polarizers may use either the "no enhancement" hypothesis (if the normalization is done by withdrawing the polarizer), or the "fair sampling" hypothesis (if the normalization is done by the sum of the + and - channel). For standard two-channel multidielectric polarizers, it is quite obvious that the two normalization will give very similar results.
CHT: Yes, the normalization may be "very similar", but the way the terms are used in the tests is logically different, and far too little investigation has been done. It is easy to see from natural generalisations of the standard realist formula (my expression (5)) that the sum of the + and - coincidences is not a suitable denominator for estimates of probabilities. Due to low detector efficiencies it is always too small, and in many cases, depending on detector characteristics, will vary with the angle between detector settings. Gregor Weihs (private correspondence) has experiments currently in progress to check the characteristics of the detectors.
For the improved, C/H 1974 test, it is not important to be able to interpret terms as probability estimates at all, as it is only ratios of coincidence counts that are used in the actual test.
Referee 2: Actually, the total number of pairs entering the polarizer (which is used for normalization purpose) ...
CHT: No, this is wrong. Nobody knows the number of pairs entering the polarizer, only the number detected ...
Referee 2: ...does not involve the polarization properties of the pairs, but only the properties of the source and the overall collection efficiency : unless many things are wrong with physics, the different possible ways to measure this number should give the same result.
CHT: I believe that "many things are wrong with physics". In particular, it is not possible to characterise a detector sufficiently by just one figure, its "efficiency". But, as explained above, it is not only in the method of normalization that the various tests differ.
Referee 2:
* from what is written in the text it appears that the author does not have a very clear idea about what is "correlated" or "uncorrelated" emission by the source. In the present context "uncorrelated" means that the photon scattering events of the laser light by the different atoms are independant. The fact that photons are scattered independantly, despite the fact that all atoms are irradiated by the same lasers, is a basic property of spontaneous emission. But maybe the author does not admit that either.
CHT: I think I have a right to doubt the independence of the scattering events. It is an unnecessary assumption. How can it be proved? Even if true for the idealised case, how are we to know that Aspect achieved this? It all depends on quantum-theoretical assumptions, and these would not be needed if the experiment were performed with lower emission rates so that accidentals were negligible.
Referee 2:
* the main goal of the author seems to advocate that more tests, and better ones, are carried out. Since 1981 this occurred anyway ...
CHT: Each such test has concentrated on blocking just one loophole. It is necessary to block all at once to get a valid Bell test.
Referee 2: ... and all tests, including the recent "time varying" ones, agreed with QM.
CHT: See above: as is well known, all have had loopholes, so numerical agreement with QM is meaningless.
Referee 2:
This is maybe an indication that the world IS quantum mechanical.
CHT:
You are entitled to your opinion!
Referee 2:
In case the author still disagrees, I would recommend to concentrate criticisms on the most recent experiments, which are usually also the most advanced technically.
CHT: I shall bear this in mind, and am currently working on realist explanations of 1998 experiments by Gregor Weihs and Paul Kwiat. However, I see it as of the greatest importance to present these hitherto unpublicised facts about early experiments. Possibly they were not previously known, even to Aspect, being buried in unanalysed tables. Had the importance of accidentals been fully realised, the world might have decided to chose realism back in 1981. Surely revelation of more complete information is better late than never?
Yours sincerely
Caroline H Thompson