Some observations/questions about the Delft Experiment

[minkwe]

I can also help getting hold of data from researchers who like only to give their data to persons whom they consider to be reliable researchers.

It will be great if you can help. Although having access to the full raw data is desired, initially all I need is just the 4xN spreadsheet of matched outcomes {a,b,x,y} used in demonstrating a violation. Since you have done analysis on those spreadsheets, it appears you do have the data. If you can share it, it will allow me to do the analysis while waiting for permission to include it in my paper, rather than waste time waiting for the data. In any case, if I'm not allowed to use the data in my paper, I will have to state in the paper that the researchers refused to share it, since any sensible reviewer would want to know why I did not include it.

In the field of crystallography, which pays my bills, we spend millions doing experiments and yet all the data is public. I just don't understand the type of science that involves researchers hogging data from published publicly funded experiments. I really applaud the NIST group for their transparency.

[Joy Christian]

... all I need is just the 4xN spreadsheet of matched outcomes {a,b,x,y} used in demonstrating a violation.

I can't wait to see that.

***

[minkwe]

... all I need is just the 4xN spreadsheet of matched outcomes {a,b,x,y} used in demonstrating a violation.

I can't wait to see that.

***

It is actually easy to produce, since a = alice setting, b = bob setting, x = alice outcome, y = bob outcome. This is the end result of every experiment irrespective of the variant (CHSH or CH or J). I have some questions about matching algorithms but for this test, I don't really care about the matching, I only need the matched sequences.

[local]

I just don't understand the type of science that involves researchers hogging data from published publicly funded experiments. I really applaud the NIST group for their transparency.

It's not science. And it's not about truth. It's about grant money, positions, fame, prizes, and reputations. Obviously, any experiment that does not disclose the full raw data should be immediately rejected and, if the forum allows me to say, ridiculed. That it was published at all is a disgrace.

How's it going Richard? Any progress in getting the Hensen et al full raw data? So kind of you to offer to get it for us.

[gill1109]

I just don't understand the type of science that involves researchers hogging data from published publicly funded experiments. I really applaud the NIST group for their transparency.

It's not science. And it's not about truth. It's about grant money, positions, fame, prizes, and reputations. Obviously, any experiment that does not disclose the full raw data should be immediately rejected and, if the forum allows me to say, ridiculed. That it was published at all is a disgrace.

How's it going Richard? Any progress in getting the Hensen et al full raw data? So kind of you to offer to get it for us.

Of course those who do science have to work to fund their science, and everyone likes a bit of fame, and everyone craves for a good reputation. Scientists are humans too, and they are embedded in a complex society. Do I detect a note of sarcasm, Mr Local? You call me "Richard" but to me you are a pseudonym. I prefer transparency and openness.

Banter aside, i.e., seriously, I will get onto this as soon as possible. I am pretty sure that Jan-Åke Larsson has the data we want; he also has some hefty criticism, not published, on the Hensen et al experiment. Though I believe that Hensen et al., who did know of that criticism from Jan-Åke, did do some further analyses and put out a further publication so as to keep one step ahead of the competition. Maybe they had independently thought of the same issues.

The "full raw data" is ginormous. It is really a three party experiment, the "source" is actually a destination as well as generating its own data. The experiment continues for a day or two and at three locations there are streams of input and output data (settings/inputs and measurement outcomes/output)

Michel is asking for a 245 x 4 spreadsheet which is extracted from the full data according to a published algorithm. I think it is not difficult to download a really big data set from Delft, and the Python programs from Delft, and then modify the code to get what Michel wants. Other people have already done this. Delft too is bound by the Netherlands Science Organisations' (government ministry rules; academy of science rules) on data-sharing.

[minkwe]

The "full raw data" is ginormous. It is really a three party experiment, the "source" is actually a destination as well as generating its own data. The experiment continues for a day or two and at three locations there are streams of input and output data (settings/inputs and measurement outcomes/output)

Richard I deal with large data everyday. I doubt that it will be worse. I generate and transfer terabytes of data each day.

Michel is asking for a 245 x 4 spreadsheet which is extracted from the full data according to a published algorithm. I think it is not difficult to download a really big data set from Delft, and the Python programs from Delft, and then modify the code to get what Michel wants. Other people have already done this. Delft too is bound by the Netherlands Science Organisations' (government ministry rules; academy of science rules) on data-sharing.

I already have data from Delft, just not the raw data. I have 421 kilobytes from Delft labelled "bell_open_data.txt". . This is not the full raw dataset by any chance but it is sufficient for my present analysis. I also have data of two experiments from NIST. I have nothing from the Vienna group who have also worked with your friend Jan despite, except for the very early Weihs experiment. And, nothing from the Rosenfeld experiment.

[gill1109]

I suggest you write directly to my friend Jan-Åke. He is a co-author of several Vienna papers so he must have some rights and authority regarding the data. He also is very strongly in favour of open access etc.

But I will also raise the question with him, myself, one of these days.

[local]

So your offer to assist was just bluster; we all know what "one of these days" means.

And my requests for the raw data have already been declined. That's why I was hoping the great Richard could succeed where we "unreliable researchers" could not.

Finally, your "the dataset is too large" excuse is pathetic. Shalm et al do not seem to be subject to this problem, as they have published multiple gigabytes of their raw data.

[minkwe]

Okay, here is what I have so far on the MI calculation from QM.






We know that the probabilities according to QM are the following.




It turns out that for a photonic Bell test experiment, the typical settings are
, and therefore , with corresponding probabilities . Note that for simplicity, since and are even functions, we can ignore the sign of .

Therefore, we end up with



A close inspection of equation (1) above shows substituting will give us according to QM.

Unless I've made a serious error here, this is very surprising. According to QM, Alice's outcomes should be independent of Bob's outcomes. Of course with the assumption that the settings are identically distributed. If this holds, the fact that the experiments are showing clear dependence between the outcomes is concerning. The good thing here is even small deviations in the settings distributions can be quantified and using the same methodology, we can derive exact QM I(xy) for the experiment to compare with the values observed.

[minkwe]

Now I realise this should not be surprising at all. It simply tells us that Bob cannot obtain any information about Alice's outcomes simply by looking at his own outcomes. Could anyone point me to a paper that suggests otherwise? I would like to investigate the origin of the idea that outcome dependence is allowed in QM.

[local]

The quantum solution for EPRB (the full cosine curve) requires both parameter and outcome dependence.

Pawlowski, M., Kofler, J., Paterek, T., Seevinck, M., and Brukner, C., “Non-local setting
and outcome information for violation of Bell’s inequality”, New Journal of Physics 12,
083051 (2010).

Also

https://arxiv.org/pdf/1607.01808.pdf

Studying your derivation...

[minkwe]

The quantum solution for EPRB (the full cosine curve) requires both parameter and outcome dependence.

If this is true, then it should be possible to send some information about Alice's measurement to Bob. In other words, by simply measuring his outcomes without any knowledge about Alice's settings or outcomes, Bob should be able to learn some information about Alice's outcomes. This information is exactly what the Mutual Information represents, in units of bits. If my calculation is correct, then Bob can't learn anything about Alice's measurements.

Pawlowski, M., Kofler, J., Paterek, T., Seevinck, M., and Brukner, C., “Non-local setting
and outcome information for violation of Bell’s inequality”, New Journal of Physics 12,
083051 (2010).

Also

https://arxiv.org/pdf/1607.01808.pdf

Studying your derivation...

Thanks, I'll check it out. Note that Pawlowski are assuming that Bell's inequality is violated by QM. But my analysis is pointing to the suggestion that in fact QM does not violate it. So in some way their paper re-enforces what I'm finding. Let me take a closer look at it.

[local]

Don't forget that an EPRB experiment involves two separated measurements, not a single joint measurement. The first measurement projects the state at the other side. This projection must include information about the angle setting and the outcome, if the QM correlations are to be obtained. This is clearly developed in the second citation I gave above.

[minkwe]

Don't forget that an EPRB experiment involves two separated measurements, not a single joint measurement. The first measurement projects the state at the other side. This projection must include information about the angle setting and the outcome, if the QM correlations are to be obtained. This is clearly developed in the second citation I gave above.

Yes. But that does not matter for calculating the MI. All we need are the individual and joined probabilities from QM and we should be able to calculate the MI. If the resulting MI is zero, then there is no dependence, irrespective of how you arrived at the probabilities.

Additionally, when you do QM calculations like that, we always have a specific pair of settings in mind. in an experiment in which the settings stay fixed, then definitely the MI of the outcomes will not be zero. But in that case, we already have very strong setting dependence. In Bell test experiments, settings are randomly selected on both sides and the MI calculation is for the whole experiment involving all the settings and all the outcomes. It is this scenario that appears to have zero MI for QM but significantly higher MI from the Bell-test experiments I've analysed so far.

[gill1109]

So your offer to assist was just bluster; we all know what "one of these days" means.

And my requests for the raw data have already been declined. That's why I was hoping the great Richard could succeed where we "unreliable researchers" could not.

Finally, your "the dataset is too large" excuse is pathetic. Shalm et al do not seem to be subject to this problem, as they have published multiple gigabytes of their raw data.

I object to your language, local. “One of these days”, written by me a few days ago on my return from a long and exhausting lecture tour in China, means “today”.

[local]

“One of these days”, written by me a few days ago on my return from a long and exhausting lecture tour in China, means “today”.

Sweet! So what was the outcome of your enquiry? Your post was 3 days ago; surely your friend Jan-Ake would have responded by now.

[gill1109]

“One of these days”, written by me a few days ago on my return from a long and exhausting lecture tour in China, means “today”.

Sweet! So what was the outcome of your enquiry? Your post was 3 days ago; surely your friend Jan-Ake would have responded by now.

Well, it was weekend. Some of us spend our weekends with our families or have lots of household chores to take care of. I'll let you know when I get a reply. I'm confident that Michel can get access to that data, one way or another.

By the way, on another thread Michel has said:

I think this thread has run its course and can be closed. The two undisputed conclusions are the following:

1. As long as the symbols mean the same thing as defined by Bell, the number represented by ... can not possibly exist because it involves mutually contradictory premises.

2. Richard Gill claims there will never be a local-realistic simulation of the Delft, Munich, Vienna, or NIST experiments.

Fred followed up Michel's request and closed the thread. But this present thread is about the Delft experiment, and I still need to say the following about it. Firstly, Michel's conclusion “1.” was disputed, by me, and I still do dispute it. No problem. That discussion is closed as far as I am concerned, too. Regarding Michel's conclusion 2, it is easy to make a local realistic simulation of those four particular experiments when the streams of settings are taken to be the same as those actually used. It is also possible to reproduce them exactly, by pure chance, when the settings are re-randomised. So I think he should agree that my claim does need further specification. I am happy to discuss details of such a specification with anyone who is interested.

The Delft p-value of say 3% means that the experimenters themselves admit that 3% of local realistic simulations (with new, random, settings), would reproduce their finding. What I am talking about is a local realistic simulation which scores an equally large violation of Bell inequalities a good deal more than 3% of the time. Yes, I do indeed believe there will never be a local-realistic simulation of the Delft, Munich, Vienna, or NIST experiments. I've told you why. Bell's theorem, seen as a theorem in computer science (in which field it is indisputably a true theorem, as far as any mathematical theorem is "indisputible". Maybe someone, sometime, will come up with a new interpretation or new definition which makes us re-think the situation. But I am not holding my breath in anticipation).

I already issued a 64 thousand dollar challenge. Michel, Joy and Fred have all declined to take it up. Any other takers?

[local]

Nobody cares about your silly challenge that you now bring up here to obscure your failure to do any better than we have in getting the raw data.

Just for the record, it's not just Michel, I am requesting the full raw data from Hensen et al. Why are the authors hiding it?

[gill1109]

Nobody cares about your silly challenge that you now bring up here to obscure your failure to do any better than we have in getting the raw data.

Just for the record, it's not just Michel, I am requesting the full raw data from Hensen et al. Why are the authors hiding it?

Some people do care about that challenge. It is not thought by everyone to be silly. I don't like insinuations as to my motives.

I hope to gain access to the data soon. I will be delighted to share it with Michel. I will keep everyone here informed.

[gill1109]

Vienna data: I understand that the Vienna people do provide the data, but from one distribution point, which used to be the first author of the experimental paper. That would be Marissa Giustina, but she is now at Google's Quantum A.I. lab. So maybe someone else is now responsible for the data in Vienna. I will enquire further. Marissa's email address in Vienna was marissa.giustina@univie.ac.at , I don't know what it is with Google. She's on LinkedIn.

Delft data: Michel said that he already had what he needed.

Here are links to what is directly available:

https://data.4tu.nl/repository/uuid:53644d31-d862-4f9f-9ad2-0b571874b829
Loophole-free Bell test using electron spins in diamond: second experiment and additional analysis

https://data.4tu.nl/repository/uuid:6e19e9b2-4a2d-40b5-8dd3-a660bf3c0a31
Loophole-free Bell-inequality violation using electron spins separated by 1.3 kilometres