The (still) open one-sided bet.

[gill1109]

Sorry, but I've put Joy Christian on my list of "foes" so I don't see his postings here unless I take extra action. When he's posted a computer file of directions on internet, he can personally email me the URL. Or somebody else can let me know.

[gill1109]

Request to Fred Diether:

I should have named this new topic "The (still) open one-sided bet" a bit differently - namely as "The (still) open one-sided challenge". It's a subtle but perhaps important distinction. Is it possible for you, respected super-user and owner, to rename a topic without changing URL's etc?

It would be much appreciated.

But if it's too much bother, no problem. And if it's technically impossible, sorry that I bothered you with the request.

Richard

[minkwe]

Sorry, but I've put Joy Christian on my list of "foes" so I don't see his postings here unless I take extra action. When he's posted a computer file of directions on internet, he can personally email me the URL. Or somebody else can let me know.

We've seen statements like this in the past. We know how long the tantrum will last. Joy's R-code is very clear. You do not say what is wrong with it. First it was the suggestion that 'b' was fixed at 0, which was swiftly snuffed, then the suggestion that "good" was doing some magic. That too was swiftly snuffed. Then next was the suggestion that Joy had produced a curve rather than a surface. That line of argument was long abandoned. All of a sudden the master of R is no longer interested in R but in two separate files and a different method of calculation. As I suggested all along, the real issue is not whether Joy can reproduce QM correlations, Richard's real interest is in playing games with method of calculation, as we've already established elsewhere.

Some questions are begging to be asked:

1) Is Joy's simulation local-realistic or not? If not state precisely where in the code.
2) Does any of the correlations E(a,b) deviate from the QM prediction by more than 0.2 or not? (Simply look at the plot. or point out precisely why you believe the calculation of the plot is in error, the code is public).

I suspect you won't answer these questions. I suspect you would resort to a different method of calculation, and blame Joy for the failure of your method to reproduce the results/calculation he has posted in the open. Not only that, I suspect you will then claim that Joy has agreed to your method, and you will then try to deflect by accusing me of not having read Joy's paper. But the above questions will remain as a dark cloud over your head, and you will know it is there, and it will bother you until you address them.

[Joy Christian]

Sorry, but I've put Joy Christian on my list of "foes" so I don't see his postings here unless I take extra action. When he's posted a computer file of directions on internet, he can personally email me the URL. Or somebody else can let me know.

We've seen statements like this in the past. We know how long the tantrum will last. Joy's R-code is very clear. You do not say what is wrong with it. First it was the suggestion that 'b' was fixed at 0, which was swiftly snuffed, then the suggestion that "good" was doing some magic. That too was swiftly snuffed. Then next was the suggestion that Joy had produced a curve rather than a surface. That line of argument was long abandoned. All of a sudden the master of R is no longer interested in R but in two separate files and a different method of calculation. As I suggested all along, the real issue is not whether Joy can reproduce QM correlations, Richard's real interest is in playing games with method of calculation, as we've already established elsewhere.

Some questions are begging to be asked:

1) Is Joy's simulation local-realistic or not? If not state precisely where in the code.
2) Does any of the correlations E(a,b) deviate from the QM prediction by more than 0.2 or not? (Simply look at the plot. or point out precisely why you believe the calculation of the plot is in error, the code is public).

I suspect you won't answer these questions. I suspect you would resort to a different method of calculation, and blame Joy for the failure of your method to reproduce the results/calculation he has posted in the open. Not only that, I suspect you will then claim that Joy has agreed to your method, and you will then try to deflect by accusing me of not having read Joy's paper. But the above questions will remain as a dark cloud over your head, and you will know it is there, and it will bother you until you address them.

Thank you, Michel. I couldn't have put this better than you have.

[Heinera]

2) Does any of the correlations E(a,b) deviate from the QM prediction by more than 0.2 or not? (Simply look at the plot. or point out precisely why you believe the calculation of the plot is in error, the code is public).

You clearly have some programming experience. Look at this piece of code from Joy's public program:

Code: Select all

for (i in 1:(K - 1)) {
    alpha <- angles[i]
    a <- c(cos(alpha), sin(alpha))  ## Measurement vector 'a'
    for (j in 1:(K - 1)) {
        beta <- angles[j]
        b <- c(cos(beta), sin(beta))  ## Measurement vector 'b'
        ca <- colSums(e * a)  ## Inner products of cols of 'e' with 'a'
        cb <- colSums(e * b)  ## Inner products of cols of 'e' with 'b'
        N <- length(ca)
        corrs[i] <- sum(sign(-ca) * sign(cb))/N
        Ns[i] <- N
    }
}

First the code loops on i (the alpha angles), and then on j (the beta angles).

But the assignment to the correlations corrs[i] within the j-loop only has one index i. So for each new value of j, the previous assignment to corrs[i] is simply forgotten and overwritten. So at the end of the loop, we end up with corrs[i] dependent only on angles[K-1] (the last assignment to beta). All other values of beta are irrelevant. Agree?

[gill1109]

OK. Joy and I are talking again (by email) and I also removed his name from my "foe list" here.

The challenge as formulated at the beginning of this topic is taken up by Joy. He's even submitted his file of directions but I have not analysed them yet. First I will draft a letter to the persons we have in mind for the adjudicating committee, and put it to Joy for his approval.

If he approves, I'll send it to them.

Then I will analyse the data which Joy has submitted, post it on internet, and also post my initial conclusion.

If it's negative then Joy will ask the adjudicators for a "second opinion".

If it's still negative we are done. If however the adjudicators judge in Joy's favour I owe him 10 000 Euro.

Is that fair or is that fair?

Probably we are finished in a few days ... perhaps a couple of weeks. At the most we will be finished by June 11, I'm sure. It will be mainly up to the adjudicators.

[minkwe]

First the code loops on i (the alpha angles), and then on j (the beta angles).

But the assignment to the correlations corrs[i] within the j-loop only has one index i. So for each new value of j, the previous assignment to corrs[i] is simply forgotten and overwritten. So at the end of the loop, we end up with corrs[i] dependent only on angles[K-1] (the last assignment to beta). All other values of beta are irrelevant. Agree?

That is a fair criticism.

[Joy Christian]

First the code loops on i (the alpha angles), and then on j (the beta angles).

But the assignment to the correlations corrs[i] within the j-loop only has one index i. So for each new value of j, the previous assignment to corrs[i] is simply forgotten and overwritten. So at the end of the loop, we end up with corrs[i] dependent only on angles[K-1] (the last assignment to beta). All other values of beta are irrelevant. Agree?

That is a fair criticism.

How is this equivalent to "b" being fixed? It seems to me that the last assignment of beta can take any value.

[Heinera]

First the code loops on i (the alpha angles), and then on j (the beta angles).

But the assignment to the correlations corrs[i] within the j-loop only has one index i. So for each new value of j, the previous assignment to corrs[i] is simply forgotten and overwritten. So at the end of the loop, we end up with corrs[i] dependent only on angles[K-1] (the last assignment to beta). All other values of beta are irrelevant. Agree?

That is a fair criticism.

How is this equivalent to "b" being fixed? It seems to me that the last assignment of beta can take any value.

No, the last assignment of beta always takes the value angles[K-1], which is constant (K is a constant).

[gill1109]

This is the letter I am thinking of sending to the adjudicators:

Dear ...

Recently I emailed you telling you that a certain bet I had made with Joy Christian was called off. At the time I was rather angry and pretty determined not to have anything to do with him again, but in the meantime my anger abaited somewhat and after some further contacts we came to a new agreement concerning the following challenge: ...

We would like to ask you if you would be prepared to adjudicate a submission to this challenge, just once. We moreover believe that very little work is required of you, and all that you will be required to produce is a yes/no answer to a simple question.

The procedure is: Christian submits a computer file to Gill who processes it to the best of his ability according to his interpretation of the rules set out in the challenge. (Gill and Christian have formulated these rules together, and both believe they are unambiguous).

He either accepts that Christian has succeeded, or he claims that Christian has failed.

In the latter case, and in the latter case only, and once only, you are asked to adjudicate.

Christian might submit some supporting material arguing why he disagrees with my conclusion. Up to you what you make of it.

Your decision is final and binding on all parties and no appeal is possible, no correspondence with the jury is allowed (on the subject of the challenge) either during their deliberations or after their conclusion is announced.

Christian and Gill expect to submit materials in a few days and would be delighted if the jury could come to a conclusion before the coming Vaxjo conference.

On Gill's side the only material expected would be the data file (which is not going to be very large at all) and an R script generating the numbers on the basis of which he comes to his initial decision. He has Python, Perl, and Excel versions available too (which have been tested on similar data sets to all give the same answer). These scripts are short and transparent. (It is a question of calculating four correlations). However, it must be emphasized that the jury is free to use whatever computer tools they like to get the answers. Their essential task is to independently calculate the four numbers described in the challenge.

[Joy Christian]

This is the letter I am thinking of sending to the adjudicators:

Nice letter, but in the mean time a new argument has emerged that "b" may be fixed in my simulation. I would like to investigate that further before we proceed.

[minkwe]

First the code loops on i (the alpha angles), and then on j (the beta angles).

But the assignment to the correlations corrs[i] within the j-loop only has one index i. So for each new value of j, the previous assignment to corrs[i] is simply forgotten and overwritten. So at the end of the loop, we end up with corrs[i] dependent only on angles[K-1] (the last assignment to beta). All other values of beta are irrelevant. Agree?

That is a fair criticism.

How is this equivalent to "b" being fixed? It seems to me that the last assignment of beta can take any value.

The criticism is that the value is overwritten by the next assignment. The outer loop runs K times, the inner loop runs K*K times, there are only K values of corrs. The only effective corrs value is then only the last one calculated.

[gill1109]

This is the letter I am thinking of sending to the adjudicators:

Nice letter, but in the mean time a new argument has emerged that "b" may be fixed in my simulation. I would like to investigate that further before we proceed.

Sure. No hurry. I don't want to send them the letter if we are not pretty sure we are going to be going ahead, pretty soon.

The challenge is open. It stands. It's a Randi-type challenge. Everything is on record. I respect your wish to temporarily freeze and possibly withdraw your present submission. I did not do anything with it yet ... You may withdraw it or give it the go ahead, as you please, when you are ready. If after *withdrawal* of this submission you wish to make a new submission, you are always welcome (well - unless someone else in the meantime has been successful). But once you proceed you only have one chance to win the hearts and/or minds of the adjudicators.

The draft letter to the adjudicators is on record. We know who we have in mind to ask for this job. I'm rather pleased with the letter, myself! We made a very great deal of progress today.

[gill1109]

1) Is Joy's simulation local-realistic or not? If not state precisely where in the code.
2) Does any of the correlations E(a,b) deviate from the QM prediction by more than 0.2 or not? (Simply look at the plot. or point out precisely why you believe the calculation of the plot is in error, the code is public).

Regarding the second question: A plot of one curve is not enough if we want to inspect four points on a surface. Four points which do not lie above a straight line in the x-y plane. The four points lie on the corners of a square in the x-y plane (a square aligned with the x and y axes).

It is not difficult to plot the correlation surface and then you immediately see massive failure of many of the correlations.

Regarding the first question: The present simulation is local-realistic, that is easy to see.

I suspect that that is why it fails to deliver! (At least, as far as I can see).

Before rushing in to tell me how mean and stupid I am, you ought at least to read the posting early on in this topic where the challenge (not a bet, really) is spelt out. The wording of the challenge was agreed, for 100%, by Christian and me. We drew it up together.

You can advise Christian to withdraw, if you think he was unwise to accept that wording.

The challenge makes perfectly clear what each of us has to do, in turn. So you don't have to tell me what I have to do. I receive a file from Christian, process it according to the challenge's instructions, deliver my verdict. If Joy disagrees with my verdict we refer to our adjudicators. They decide and either I pay or I don't. Pretty clear, don't you think?

Do you think the challenge is stupid because no one can win it? Well if so please try and explain that to our friend Christian. I tried, but did not succeed. This shows that it is not a stupid challenge. It has a pedagogical function. Nobody loses (except I might lose a lot of money), everyone can learn something. The challenge remains open till someone wins it, or I die or am legally declared insolvent or insane or equivalent.

[gill1109]

Zen asked me to run and post an R script which draws Christian's correlation surface

http://rpubs.com/gill1109/ZenSurface

Red is JC, blue is QM

[Joy Christian]

[quote="gill1109"]Zen asked me to run and post an R script which draws Christian's correlation surface

http://rpubs.com/gill1109/ZenSurface

Nice!

Now we are making progress. I see these images as both good news and bad news. Note that my LHV surface is a massive improvement over the traditional LHV surface (or Bell-CHSH surface). But the images also reveal that the surfaces do not match perfectly.

Good. So now I go back and work on the current model a little more. Note that, despite the confusion by Satori (who is a reincarnation of Zen) on the FQXi blog, the current model is for the very specific purpose of generating the 2 x N vectors involved in my experiment. The most accurate simulation of my analytical model is still this one: http://rpubs.com/jjc/13965.

[gill1109]

Chán is the Chinese form of the Japanese Zen, Sankskit Dhyāna, English closest equivalent: "enlightenment"
Satori is Japanese for Sanskrit Prajñā, English closest equivalent: "meditation"

Mr (or as far as I know, Ms) Chán Satori is an excellent programmer and a very sharp analyst.

I hope that soon that he'll have his own Rpubs site too ...

It would be fun to add the four crucial points to these two surfaces, so that we can better see where the maximal pain is being experienced. I love playing with R scripts myself but I really must break my addiction for a couple of days and do some other pressing work ... and after that go on vacation without my computer ...

Alongside of R, I highly recommend Buddhism and in particular Vipassana meditation. Though I am only a pre-novice myself. Programming is a form of meditation. It teaches one how to de-program, and to be aware of hidden programs. Some time I might write a book "Zen and the art of R programming" (if it hasn't already been done).

The Zen of R has been done: http://rpubs.com/gill1109/ZenOfR

[Heinera]

Nice!

Now we are making progress. I see these images as both good news and bad news. Note that my LHV surface is a massive improvement over the traditional LHV surface (or Bell-CHSH surface). But the images also reveal that the surfaces do not match perfectly.

Actually not. For some pints it is an improvement, for orher points it performs worse. If you take the average absolute difference between your surface and the QM surface, it persforms as bad as the picewice linear Bell surface.

[Joy Christian]

Nice!

Now we are making progress. I see these images as both good news and bad news. Note that my LHV surface is a massive improvement over the traditional LHV surface (or Bell-CHSH surface). But the images also reveal that the surfaces do not match perfectly.

Actually not. For some pints it is an improvement, for orher points it performs worse. If you take the average absolute difference between your surface and the QM surface, it persforms as bad as the picewice linear Bell surface.

I disagree. I have run the R script for 10^6 and 10^7 trials, and the wrinkles in the LHV surface smooth out considerably for these larger number of trials. They do not go away, however, and so there are indeed specific points where my (current) LHV model does worse than the Bell-CHSH model.

[Heinera]

Nice!

Now we are making progress. I see these images as both good news and bad news. Note that my LHV surface is a massive improvement over the traditional LHV surface (or Bell-CHSH surface). But the images also reveal that the surfaces do not match perfectly.

Actually not. For some pints it is an improvement, for orher points it performs worse. If you take the average absolute difference between your surface and the QM surface, it persforms as bad as the picewice linear Bell surface.

I disagree. I have run the R script for 10^6 and 10^7 trials, and the wrinkles in the LHV surface smooth out considerably for these larger number of trials. They do not go away, however, and so there are indeed specific points where my (current) LHV model does worse than the Bell-CHSH model.

I have published an R script where average performance is computed:

http://rpubs.com/heinera/16559