SciPhysicsForums.com

Simulation programs for EPR-type experiments have the problem that the simulated model is not obvious. In particular, it is not obvious whether locality is violated. Also the separation of laws of "nature" (i.e., of model), experimenter's choices and analysis of results. The latter two should be independently modifiable in order to allow comprehensive analysis of the simulation program.

In order to separate analysis of the simulation model from the model itself, the simulation program should be separated from the analysis program. In addition, the settings of the detectors should be read from input files. The simulation should produce output files that can be read to an analysis program.

The following recommendations enforce factual definiteness and locality. In addition, they separate laws of "nature" from analysis. In addition, the recommended file formats permit integrity checks.

The simulation model shall consist of two or three distinct programs: one that simulates the source that produces pairs of particles; and one or two that simulate detection of particles - the two detectors shall be simulated separately but the same program may be used.

The emission simulation program shall input the number of particle pairs to be emitted (e.g., from command line) and produce two files, below called particle files.

The mesurement simulation program shall input one of the particle files produced by the emission programs and a settings file, and output a result file.

In addition an analysis program is needed. It is not part of the simulation model and need not be created by the author of the simulation program. Any number of analysis programs may be used. An analysis program shall read two settings files and two result files and output the results of the analysis (to a file or stdout or display or whatever).

All files mentioned above shall be ASCII files that contain header lines, data lines, and a footer line in this order. The header and footer lines begin with the '#' charcter followed by one or more spaces. The data lines begin with one or more spaces followed by the number of particle pair (starting from 1) followed by one or more spaces. The rest of the line depends on the file type. Files may also contain blank lines but they mean nothing and shall be ignored. Any lines after the footer shall also be ignored.

The first header line shall contain the name of the file. The footer line shall contain the same name.

A particle file shall have three other header lines. The second one identifies the simulation model (e.g., the name of author, the name of model). The third lines contain date and time for indentification of the run. These two lines must be identical in the two particle files produced at the same time. The fourth header line be different in the two files and contain the letter 'A' or 'B'.
Each data line shall end with "hidden" variables of the particle, which shall be encoded with visible ASCII characters and spaces.

A settings file shall have one additional header line that contain the letter 'A' or 'B'. Each data line shall end with a setting value in degrees (as an integer or real number).

The resul file shall have second, third, and fourth header lines that are exact copies of the second, third, and fourth header lines of the particle file. The fifth header line shall be an exact copy of the second header line of the settings file. Each data line shall end with result of the detection as a '+' or '-' character except if the particle is not detected at all, which shall be encoded with a '?' character.

These recommendations can be followed if the simulation model satisfies locality, factual definiteness and counterfactual definiteness. Other simulations should follow as many of these recommendations as the model allows.

Mikko wrote:Simulation programs for EPR-type experiments have the problem that the simulated model is not obvious. In particular, it is not obvious whether locality is violated. Also the separation of laws of "nature" (i.e., of model), experimenter's choices and analysis of results. The latter two should be independently modifiable in order to allow comprehensive analysis of the simulation program.

I agree.

The following recommendations enforce factual definiteness and locality. In addition, they separate laws of "nature" from analysis. In addition, the recommended file formats permit integrity checks.

The simulation model shall consist of two or three distinct programs: one that simulates the source that produces pairs of particles; and one or two that simulate detection of particles - the two detectors shall be simulated separately but the same program may be used.

Very reasonable recommendations.

The emission simulation program shall input the number of particle pairs to be emitted (e.g., from command line) and produce two files, below called particle files.

Given that in real experiments we can not tell a source how many particles to emit, I believe it is more reasonable to provide as input the duration of time to run the source. Of course the number of particles emitted can be determined from the output files such that requiring as input the source duration will still fulfill the spirit of your recommendation while adhering better to realizable experiments. Agreed?

The mesurement simulation program shall input one of the particle files produced by the emission programs and a settings file, and output a result file.

Very reasonable.

In addition an analysis program is needed. It is not part of the simulation model and need not be created by the author of the simulation program. Any number of analysis programs may be used. An analysis program shall read two settings files and two result files and output the results of the analysis (to a file or stdout or display or whatever).

Very reasonable.

All files mentioned above shall be ASCII files that contain header lines, data lines, and a footer line in this order. The header and footer lines begin with the '#' charcter followed by one or more spaces. The data lines begin with one or more spaces followed by the number of particle pair (starting from 1) followed by one or more spaces. The rest of the line depends on the file type. Files may also contain blank lines but they mean nothing and shall be ignored. Any lines after the footer shall also be ignored.

This is a well meaning requirement, except that when dealing with millions of lines of data, ASCII is not the proper format as far as efficiency is concerned. There are lots of standard formats which can easily be converted to text for verification. I've used the python ".npy" binary array format for this purpose.

The resul file shall have second, third, and fourth header lines that are exact copies of the second, third, and fourth header lines of the particle file. The fifth header line shall be an exact copy of the second header line of the settings file. Each data line shall end with result of the detection as a '+' or '-' character except if the particle is not detected at all, which shall be encoded with a '?' character.

Given that in real experiments you can not know when a particle is detected or not detected, it is not reasonable to expect to be told "?" for undetected particles. Although a simulation can tell you this, it introduces a lot of potential for confusion and misunderstanding, because it obscures assumptions that will always have to be made in real experiments. Therefore, I would suggest that you ditch this requirement and leave it up to the data analysis program to make all assumptions it needs to make when analyzing the data. The same applies to the source. You can not assume that all particles will be produced in pairs, so you will have to leave it up to the data analysis to make the necessary assumptions.

These recommendations can be followed if the simulation model satisfies locality, factual definiteness and counterfactual definiteness. Other simulations should follow as many of these recommendations as the model allows.

Overall, a pretty decent set of requirements, other than the one or two issues I raised. I noticed also that you said nothing about the data analysis programs. Do you have a series of recommendations for them too. I assume the recommendations will be applicable also to data from actual experiments. In other words, we should have on one hand results from actual experiments and on the other hand results from a simulation and be able to apply the same analysis guidelines to both.

If you would agree to issues I raised above, I have a simulation almost ready.

Excellent discussion.

I think we need to distinguish two kinds of experiments. Ideal clocked or pulsed experiments, or experiments with event ready detectors, can be thought of as running in discrete time. At each time step, two particles are generated at the source and go to the detectors. Two settings are supplied "from outside". Two outcomes are generated at the two detectors.

So each setting and each outcome has an externally determined identifier, which is just the run number: run 1, run 2, run 3, ...

Such experiments have probably never been done though some come close and I think that everyone agrees that this is a kind of ideal model.

In real world experiments with photons everything is happening (more or less) in continuous time. Particles leave the source at random times. They are detected by the detectors, sometimes, at times which cannot be predicted in advance. Settings are being switched rapidly according to some external scheme and each particular setting value is then in force for a whole time interval (till the next switch).

So the experiment generates four streams of random times with labels identifiying the kind of event.

My point here is that one can have different aims with a simulation model; it seems to me that both kinds of experiment are legitiate and interesting. There is not "one size fits all". But there are two main types. For the two different main types one can draw up a separate list requirements.

Further recommendations: programs should be written in programming languages which are widely used and freely available and preferably open source and endorsed by international scientific bodies. They should run on different common computing platforms with only minor modifications.

Programs should be written using the principles of "literate programming" and "reproducible research" ... the basic principle being that a human being can read and understand what is supposed to be going on, as well as a computer. Code should be made publicly available.

So: Java, C, C++, Python, R (open source implementation of S), Octave (open source implementation of Matlab) ... are eligible.

Mathematica is not eligible but probably alternatives can be found in Sage for those who wish to use symbolic computation.

Random number generation should be done using reliable, well-documented, pseudo random generators and the implementation should include set / save / restore random seed facilities, so that a program can be run twice with identical inputs and generate identical output (even on different computers, even on different platforms).

Finally some comments on some of the issues Minkwe raised.

Clocked experiments (ie those with a definite "run number" 1, 2, ... identifiying subsequent, linked, particles settings and outcomes, can and do, both in the real world and in simulation models, generate outcomes "no detection" as well as +1, -1. I would suggest simply to code these with a 0. After all, it's up to the analyst how the data is going to be processed. So a "?" and a "0" are just as good as one another.

Continuous time experiments generally do not tell us when a particle was not emitted. The measurement outcomes on Alice's side is a list of times of detections which all by definition are +1 or -1 detections. The settings file is a list of times at which the setting was switched and a list of what the new setting is for the next time interval.

I agree that for really big experiments, an ascii file of output can be inconvenient. Fine that it is saved in a convenient alternative form, as long as the programmer also supplies a utility for conversion to ascii ... which at the very least, is convenient for someone else testing and playing with the program.

gill1109 wrote:Mathematica is not eligible but probably alternatives can be found in Sage for those who wish to use symbolic computation.

Ah... so the QRC is not valid then since it uses Mathematica? Get real! Mathematica is just a fancy calculator. There is no problem with it.

gill1109 wrote: At each time step, two particles are generated at the source and go to the detectors. Two settings are supplied "from outside". Two outcomes are generated at the two detectors.

This will never be accomplished in a real experiment so it is pointless to introduce it here, other than to muddy the waters. I've asked you questions about this in the past which you promised to answer but never did. There is no way in any experiment to tell if in fact two particles were emitted, and at what time, or if the emissions were simultaneous or not.

So each setting and each outcome has an externally determined identifier, which is just the run number: run 1, run 2, run 3, ...

This is fantasy. Each station can have it's own identifier based on what it detects, but there is no way in any experiment for any station to know that a particle was emitted but did not reach it, or a particle was not emitted, or a particle was emitted but not detected. Identifying which particle at Alice corresponds to which particle at Bob is the domain of data analysis and should stay that way, unless you can describe a feasible experiment in which Alice knows for sure that a particle is on its way to her and Bob knows for sure that a particle is on it's way, and they know at what time to expect their respective particle. If you can not think of an experiment like this, you will have to give up this unrealistic requirement. It will not advance the discussion. On the other hand, you may be able to frame a specific question you are trying to answer, understanding that such a question will not be verifiable experimentally. In other words, you can not use any arguments based on that kind of simulation as reasons why the simulation is wrong.

Such experiments have probably never been done though some come close and I think that everyone agrees that this is a kind of ideal model.

We are discussing about the possibility of hidden variables, which means, if they exist, nobody as yet knows the exact nature for sure. You can not purport to know what an ideal model means in this case without introducing more hidden assumptions based on your own understanding/misunderstanding of what might be actually going on.

In real world experiments with photons everything is happening (more or less) in continuous time. Particles leave the source at random times. They are detected by the detectors, sometimes, at times which cannot be predicted in advance. Settings are being switched rapidly according to some external scheme and each particular setting value is then in force for a whole time interval (till the next switch).

So the experiment generates four streams of random times with labels identifiying the kind of event.

My point here is that one can have different aims with a simulation model; it seems to me that both kinds of experiment are legitiate and interesting. There is not "one size fits all". But there are two main types. For the two different main types one can draw up a separate list requirements.

I agree, the requirements must reflect the question we are trying to answer. However, requirements that will never be realized experimentally are not useful for distinguishing between a simulation of a model and QM. Ultimately, experiments will be the judge, and if the prescribed requirements are impossible to measure, the exercise is fruitless.

gill1109 wrote:Further recommendations: programs should be written in programming languages which are widely used and freely available and preferably open source and endorsed by international scientific bodies. They should run on different common computing platforms with only minor modifications. Programs should be written using the principles of "literate programming" and "reproducible research" ... the basic principle being that a human being can read and understand what is supposed to be going on, as well as a computer. Code should be made publicly available.

This is useful but unnecessary. If you are trying to answer the question whether it is possible to simulate QM-like correlations in a local realistic manner, it suffices to have black-boxes for the source, particles and the stations with only the input parameters at controllable (cf "hidden" variables). That the source code is not available, should not be a reason to disbelief a simulator, if it reproduces the correlations within reasonable requirements. In fact, it could sometimes be better to only reveal the source code, after everyone agrees that the challenge has beet met, so as to avoid irrelevant nitpicking peripheral to the main issue at hand. For this, the minimum requirements are:

1. No communication from Station to Source
2. No communication from Station A to Station B
3. No recording of information at any Station that is independently unmeasurable. For example, no station is allowed to record particle indices or unmeasured particles (my epr-simple simulation fails in this respect, but it was written for a different purpose).

This to me are the necessary requirements. Plus one more for data analysis.

4. Data analysis should not treat the data in a manner that is significantly different from the procedures employed in analysis of real experimental results. In other words, if statistical mistakes are made for real experiments, those same mistakes should also be made while analysing the simulation results.

So: Java, C, C++, Python, R (open source implementation of S), Octave (open source implementation of Matlab) ... are eligible.

The way I see it, anything is eligible. I wouldn't recommend it but you could even use humans to do the simulation of you like, so long as the above 3 requirements are kept and the data files are available to be analysed by anyone who chooses without onerous requirements, and several people can verify by rerunning the software and obtaining similar results.

Random number generation should be done using reliable, well-documented, pseudo random generators and the implementation should include set / save / restore random seed facilities, so that a program can be run twice with identical inputs and generate identical output (even on different computers, even on different platforms).

This requirement makes no sense. So long as there is no communication between Station A and Station B and no communication from the stations to the source, it doesn't matter. Secondly, Richard keeps talking about saving and restoring random number seeds, which is completely pointless for this exercise. We are interested in almost identical correlations, EVEN WHEN, individual results results appear random, just like is the case in the real world. For this purpose random number seeds are just red-herrings. Making sure that you get identical outputs for individual events is just calling for unnecessary confusion. Unless there is an unstated agenda in this requirement, I fail to see why it keeps popping up.

Finally some comments on some of the issues Minkwe raised.
Clocked experiments (ie those with a definite "run number" 1, 2, ... identifiying subsequent, linked, particles settings and outcomes, can and do, both in the real world and in simulation models, generate outcomes "no detection" as well as +1, -1. I would suggest simply to code these with a 0. After all, it's up to the analyst how the data is going to be processed. So a "?" and a "0" are just as good as one another.

I have answered this issue several times already but I will have another try in case it was not clear. In order to identify "?" or "0" event the Station making that recording must be able to know:
i) That a particle was emitted at the source
ii) That that particle was supposed to arrive at the given time
iii) That no other stray particle found it's way into the station at the given time
iv) That that particle was not detected.

Of those 4 requirements, none can be accomplished in a feasible experiment, and you need all to know all 4 in order to record "?" or "0" at a station. Now it has been suggested that the experiment can be clocked with time slots, implying that the experimenter can control exactly when the particles are emitted and how many are emitted. Just because the experimenter can select time slots for pulsing does not guarantee that the particles are produced in pairs and only in pairs, without any time delays between them.

Therefore the reasonable thing will be to simply record each detection (+1 or -1) with a time stamp and the active setting at the time of detection. If in the future it becomes possible to know exactly when particles are emitted with time slots and you know for a fact that no other stray particle should be present, then during data analysis, you will know that a particle was missing by not having anything in the corresponding time slot. It makes no sense to expect the simulation to tell you this information.

Minkwe: you are thinking only on experiments with photons. What about the experiments on trapped ions, atoms, Josephson junctions and other qubits; what about the experiments with pulsed lasers, what about the experiments with event-ready detectors?

There are definitely different kinds of experiments and my broad classification into two main kinds seems to me convenient.

You seem *only* interested in computer simulations which mimic as accurately as possible one particular kind of actually performed experiment.

I am also interested in thought experiments.

I am interested in testing computer simulation models in all kinds of ways, including ways which do not have a laboratory parallel.

I am interested in testing claims of the writer of the computer program which can be performed without in depth study of the code, without specialist knowledge of the computer language it is written in. The set/save/restore random seed facility is a basic requirement of a thorough testing programme for simulation software. A requirement from the independent agency which performs "stress tests".

Frankly speaking, your objection to this requirement is very suspicious.

Fred: any computer language is just a pocket calculator. The point is that everyone should be able to test and run scientific software to its limits. Requiring them to pay so many thousand dollar for commercial software for which the underlying code (even the algorithms) is secret is not scientific.

gill1109 wrote:Fred: any computer language is just a pocket calculator. The point is that everyone should be able to test and run scientific software to its limits. Requiring them to pay so many thousand dollar for commercial software for which the underlying code (even the algorithms) is secret is not scientific.

LOL! Thousands? I paid US $295 for Mathematica 9 Home Edition. The built in documentation is well worth it. And... there is a tremondous amount of information about how it works.

http://reference.wolfram.com/mathematic ... rview.html

Scientists at CERN use Mathematica to Create "Quark-Gluon Plasma" State of Matter

If it is good enough for CERN, it is good enough for me. I doubt that any open source stuff can match it. Ya get what you pay for!

I recommend you try Sage. Mathematica isn't bad. But it's closed and it's commercial. Scientists all over the world are working on rewriting the thing so that *we* (the scientists) have control over our own tools. It has happened before so many times: S became R, Matlab is becoming Octave, SPSS became PSPP. We use Python, remember? Java and Javascript? We don't pay for them. *We* wrote Python, *we* wrote R, *we* wrote Octave...

Sage will replace Mathematica in a few years. In the mean time feel free to pay your 295 dollar to Wolfram.

By the way, a clocked experiment is a special case of unclocked: the settings are changed by Alice and Bob regularly at 1 time unit intervals, so can be described by two lists of settings. The measurement stations generate outcomes at some of the measurement times: so their data can be represented by two lists of the same length, each containing only blanks, +1 or -1.

Thus to some extent, clocked and unclocked experiments can be dealt with in a common format. Actually the important thing is how the data is analysed. An unclocked experiment can be analysed like a clocked one by breaking continuous time in a deterministic way into short intervals of fixed length, and then fixing some algorithm to reduce the input data (settings) form Alice for each interval to a single representative setting; the output data (times and types of events, if any) to a single representative outcome +1, -1, or blank.

I still have no idea why Minkwe is not prepared to divulge how Python sets, saves and restores random seeds, when you use its numerical library to simulate random variables. But no problem if he doesn't know how to do it, one can always read the manual and find out. This requirement of mine is simply a minimal requirement for doing *reproducible research* when the research involves Monte Carlo computations.

I have in mind that we have some kind of agency like Moody's, which rates simulation programs instead of national economies. You know, anything between A+++ to E--- (or whatever). The rating agency's work is made a whole lot more simple when (a) the software is multi-platform and more or less free, and (b) the coding is decent, which in particular, means that the coding can be put through routine tests for which one needs to be able to give the code the same random seed in combination with other inputs being the same. One can then objectively and statistically test that the proper separation has been enforced between source, measurement stations, and setting generation. One does not have to rely on specialist knowledge of particular software protocols or, for that matter hardware protocols.

The programs which are written with this kind of facility built in, simply get higher ratings, and for that matter, cost less time and money to get rated. A win-win situation.

About the price of Mathematica: one thing you are paying for is eye-candy. I'm afraid that you are paying a lot for the eye-candy (which indeed is quite impressive). Indeed, CERN scientistis also need eye-candy from time to time, to promote their projects. Good PR is vital to science nowadays.

gill1109 wrote:I still have no idea why Minkwe is not prepared to divulge how Python sets, saves and restores random seeds, when you use its numerical library to simulate random variables.

What are you talking about? Python is open-source and the documentation is available. Anyone can find out for themselves how Python saves and restores random seeds, whether or not I "divulge how" or not. The issue which you still fail to understand is that allowing a simulation to reproduce identical results is equivalent to remeasuring already measured particles which is practically impossible in real life. And it is precisely for this reason that Bell's inequalities are violated (the inability to measure counterfactual results) so unless you are trying to rig the game, you will cease insisting on such an irrelevant requirement.

Why don't you explain exactly why you want this requirement, and how it helps answer the question you are trying to answer.

minkwe wrote:

gill1109 wrote:I still have no idea why Minkwe is not prepared to divulge how Python sets, saves and restores random seeds, when you use its numerical library to simulate random variables.

What are you talking about? Python is open-source and the documentation is available. Anyone can find out for themselves how Python saves and restores random seeds, whether or not I "divulge how" or not. The issue which you still fail to understand is that allowing a simulation to reproduce identical results is equivalent to remeasuring already measured particles which is practically impossible in real life. And it is precisely for this reason that Bell's inequalities are violated (the inability to measure counterfactual results) so unless you are trying to rig the game, you will cease insisting on such an irrelevant requirement.

Why don't you explain exactly why you want this requirement, and how it helps answer the question you are trying to answer.

I have explained this many times but the answers do not seem to get across.

Here's one answer:

(a) so that software can be rigorously tested.

Here's another:

(b) in order to mathematically prove that a loophole free local realistic simulation programme cannot ever be written. That is sections 2 and 9 of my paper which we were discussing earlier ... I put it to you as a *thought experiment*. A thought experiment about an imaginary computer simulation program; a thought experiment which we can perform in our heads, and which we can use to logically deduce something about them.

It is totally irrelevant for either of these purposes that "remeasuring already measured particles is practically impossible in real life".

Incidentally, that might be true for photons, but it is false for the colourful exploding balls in Christian's proposed experiment. In fact he asks the experimenters to measure the spins of both hemispheres in all directions all at once!

Are you saying that the EPR paper should never have been published because no-one ever had done, nor ever will do, the EPR experiment, exactly as described there?

Are you saying that the mathematical theorem that S^0, S^1, S^3 and S^7 are the only parallelizable spheres is not true, because 7 dimensional spheres do not exist in the three dimensional space of a physicist's laboratory?

Please try to use your imagination, please try to use your capacity for abstract and analytic thought.

And tell me, what do you think: do you think it's possible to write a local realist computer simulation of a *clocked* experiment with no "non-detections", and which reliably reproduces the singlet correlations? (By reliably, I mean in the situation that the settings are not in your control but are delivered to you from outside; the number of runs is large; and that this computer program does this not just once in a blue moon, by luck, but most times it is run on different people's computers.)

gill1109 wrote:(a) so that software can be rigorously tested.

Rigorous testing of what? What exactly do you want to test, that the software can reproduce the correlations or that the software can reproduce individual results reproduceably. This is the part you do not get. It is possible to rigorously test the correlations and yet not have individual results. This is evident in the fact that everyone who has reimplemented my simulation is confirming the results without any saved random number seeds, and I'm sure their results are not identical but the correlations are confirmed. So this reason is not an excuse to insisist on being able to save random number seeds.

(b) in order to mathematically prove that a loophole free local realistic simulation programme cannot ever be written.

This by itself is a nonsensical endeavor unless you can also at the same time prove that a so-called "loophole-free" experiment can ever be performed. Think for a second how naive the "loophole" business is:
- Detection "loophole": By insisting that this is a loophole in an experiment or simulation, you are effectively assuming without any experimental justification that nature is not allowed to arrange for certain hidden parameters to be unmeasurable at certain detector settings. On what basis do you make such an assumption?
- Coincidence time "loophole": By insisting that this is a loophole in an experiment or simulation, you are effectively assuming contrary to experimental evidence that particles leave the source at exactly the same time, and arrive the two stations at exactly the same time. On what basis do you make such an assumption?

I've already explained to you multiple times why no experiment can ever measure the terms in the CHSH, and why the CHSH can never be violated by anything if the terms are calculated as they should be. So if you want to talk about loopholes, why don't you talk about that, let us call it the "substitution" loophole. The loophole of substituting actual results from a different set of particles for counterfactual outcomes from a single set of particles. Isn't it remarkable, that well meaning physicists, by substituting actual results from a different set of particles for counterfactual outcomes from a single set of particles proclaim the demise of counterfactual definiteness? Don't you realize how misguided this exercise is?

Now you want to use random number seeds to prove mathematically what everyone already agrees, that the CHSH cannot be violated by anything if the terms are calculated as they should be calculated. You want to force simulations to measure counterfactual terms by saving and restoring random number seeds so that you can proclaim that the CHSH was not violated. But as I've already told you umpteen times, this is cheating. (1) it is cheating because no experiment will ever measure counterfactual terms on the same set of particles. (2) it is cheating because QM does not predict counterfactual outcomes for the same set of particles (3) It is cheating because nobody has claimed that counterfactual terms violate the CHSH, and numerous mathematical proofs exist demonstrating this fact. So I ask you again, what are you trying to prove, and what question will that answer?

The only important question is whether the QM correlations can be reproduced by a local realistic model, and whether experimental reports which were published in high-impact journals claiming to have proved "non-locality" or "non-reality" were valid, and whether it is possible contrary, to claims by famous physicists to simulate quantum phenomena in a local realistic manner. I'm sure you can find the quotes from Gull, Feynmann, etc, and I'm sure you have made such claims yourself that it is impossible to simulate quantum phenomena in a local-realistic manner.

minkwe wrote:You want to force simulations to measure counterfactual terms by saving and restoring random number seeds so that you can proclaim that the CHSH was not violated.

No. You are *completely* missing the point.

minkwe wrote:The only important question is whether the QM correlations can be reproduced by a local realistic model, and whether experimental reports which were published in high-impact journals claiming to have proved "non-locality" or "non-reality" were valid, and whether it is possible contrary, to claims by famous physicists to simulate quantum phenomena in a local realistic manner. I'm sure you can find the quotes from Gull, Feynmann, etc, and I'm sure you have made such claims yourself that it is impossible to simulate quantum phenomena in a local-realistic manner.

I have never ever made any such claims. I have always said that an experimental proof still has not been provided. All the experts agree on this, by the way. Some leading experts have said that they believe that a definitive experiment will be succesfully concluded in about five years. I think that would be exciting.

I think that the important questions are quite a different ones!

I think it is important to understand what the experiment-of-the-century will look like and why, and in what way (if it succeeds), it will be different from all predecessors. For this purpose it is important to understand the logic of the Bell argument, and it is clear to me that you don't: in fact, from some strange matters of principle you refuse even to entertain thought experiments about computer simulation experiments. If we can't discuss the logic of experiments on computers, we are not going to get far discussing the logic of experiments on photons.

gill1109 wrote:Are you saying that the EPR paper should never have been published because no-one ever had done, nor ever will do, the EPR experiment, exactly as described there?

Are you saying that the mathematical theorem that S^0, S^1, S^3 and S^7 are the only parallelizable spheres is not true, because 7 dimensional spheres do not exist in the three dimensional space of a physicist's laboratory?

Please try to use your imagination, please try to use your capacity for abstract and analytic thought.

If you are interested in abstract thought experiments, I have one for you which you can easily verify in your R version of my simulation. I hope you will do it and post the results because it will prove exactly what I just described to you, and it may be an eye-opener for you finally. I encourage anyone reading this who can do the simulation in R, Python, Mathematica or any other language to do this and post their results. You only need a small modification of the various versions of my simulation.

The question we are trying to answer is: "What happens if instead of substituting actual results from a different set of particles, we use counterfactual results from the same set of particles as was intended in the original CHSH?"

We will proceed as follows:
* Generate pairs of particles as done previously.
* Instead of measuring at just "alice" and "bob", we will add two more "ghost" station called "cindy" and "dave". We will send a ghost of Alice's particle to Cindy and a ghost of Bob's particle to Dave. This way we will have counterfactual results for Alice's particle at Cindy, and the same for Bob at Dave.
* We will do the data analysis in two steps. In the first step, we will ignore Cindy and Dave and simply use Alice and Bob as we have been doing until now. This scenario is equivalent to substituting actual results on different sets of particles for counterfactual results on a single set.
* The next step of data analysis will involve using all 4 outcomes for calculating the correlations. So that we use Alice and Bob to calculate C(a,b), Cindy and Bob to calculate C(a',b), Alice and Dave to calculate C(a,b') and Cindy and Dave to calculate C(a',b'). This step is equivalent to using counter-factual correlations just as is intended in the CHSH.
* We will then compare the results between the two scenarios and be able to answer our main question.

Now this is an abstract simulation worth doing. I await your results.

And tell me, what do you think: do you think it's possible to write a local realist computer simulation of a *clocked* experiment with no "non-detections", and which reliably reproduces the singlet correlations? (By reliably, I mean in the situation that the settings are not in your control but are delivered to you from outside; the number of runs is large; and that this computer program does this not just once in a blue moon, by luck, but most times it is run on different people's computers.)

Are you willing to claim that it is impossible?

gill1109 wrote:

minkwe wrote:You want to force simulations to measure counterfactual terms by saving and restoring random number seeds so that you can proclaim that the CHSH was not violated.

No. You are *completely* missing the point.

What point exactly, I've explained the issues very clearly.

gill1109 wrote:Incidentally, that might be true for photons, but it is false for the colourful exploding balls in Christian's proposed experiment. In fact he asks the experimenters to measure the spins of both hemispheres in all directions all at once!

Not true! Where did you get such an absurd idea from? Please read my papers carefully: http://libertesphilosophica.info/blog/e ... taphysics/.

Joy Christian wrote:

gill1109 wrote:Incidentally, that might be true for photons, but it is false for the colourful exploding balls in Christian's proposed experiment. In fact he asks the experimenters to measure the spins of both hemispheres in all directions all at once!

Not true! Where did you get such an absurd idea from? Please read my papers carefully: http://libertesphilosophica.info/blog/e ... taphysics/.

I got this idea by reading your paper carefully. But it doesn't matter, we are planning the definitive experiment, right?