I A new realistic stochastic interpretation of Quantum Mechanics

[Fra]

I'll think and see if I can explain more clearly in some other way and still keep it general.
...
But let me think if I can find some other "simple" made up example to explain the concepts of the framework without making specuations of physical interactions.

I have been thinking howto to explain the "mechanism" where the memory effect can explain correlation via a kind of "hidden variables", and yet escape beeing subject to bells inequality, but without speculation or detail as that blurs the conceptual overview.

I think to make up an explicit toy example, would involve lots of assumptions about how information are encoded and stored, and how actions are randomly chosen by guided by probabilities, and making the examples realistic without engaging in specuialtion I have hard to see doable in a simple way.

So I tried to explain again conceptually, but found out that I still need to first list a number of say axiomatic assumptions, not about details, but about constrainin principles on causality and interactions in an agent based model view, that in themselves can be interpreted as "speculations", so I deleted it

Therefore I will pass try to explain the function of memory effect and the proble of dividisiblity. If I find paper that does something similar I can get back to it, but all the papers I have find are all tangenting this thing from different angles, many are related to AI research, and many are relating to agent based modelling, but none put the right things together yet.

Happy new year!

/Fredrik

 

[GentDave]

The notion of wether "information exists" becomes more complicated if you add learning and emergence/organisation to parts of the system.

If we assume that "information exists" means that, there is at least one observer (at some point), or someone that "knows", but it's just not me?

Then this becomes subject to evolution, as initially perhaps noone knows, but with time someone can learn or find out; it may emerge as a result of some learning/inference process. Also the only way for someone to find out if someone else has the information, is by interacting with others; which again changes things. I think at some point interacting observers can evolve an emergent effective agreement of consenscus, this is then "objectivity", but that is only effective locally, it has no global rooting.

Except of course in the space of mathematics of our models. But that is I think redundant embedding lacking physical basis, but it is what keeps seducing theorists.

This mechanism, is critical at least for my conceptual understanding, and it is what I see behind the memory effect.

/Fredrik

Hi Thank you for your reply I mean “information exists” in a different sense. When I’m asking about "objective probability" I'm asking "does the universe 'know' the result?” That is - even if you had all information in existence and infinite computing time can ,in this case, the next exact position of the particle be known ? If it can't then its next position is objectively undetermined. - As is assumed in a mathematical stochastic process. If it could be known, but we just lack information then that is a subjective uncertainty. Every probability we know of outside of quantum mechanics is a subjective uncertainty, although the abstract random generator in math is always assumed to be an objective uncertainty. What DOES complicate this a bit in my view is when you start thinking about nonlocal relations and the role "future knowledge" plays.

I’m also sensing you are coming from a global entanglement point of view – like "many worlds" perhaps. I don’t share that idea though. In those models the universe is deterministic and there is never any objective uncertainty. So it is a concept that separates various interpretations.

If I understand the author he is saying the next position of the photon is objectively undetermined but its current position is always only a subjective uncertainty.

 

[Fra]

Hi Thank you for your reply I mean “information exists” in a different sense. When I’m asking about "objective probability" I'm asking "does the universe 'know' the result?” That is - even if you had all information in existence and infinite computing time can ,in this case, the next exact position of the particle be known ?

For me this is not a question I ask, it is not physical for me. Your hypothetical question implies infinite computational capacity. For me "physical questions" must be constructible from what the questioner (ie the observer or agent) has to work with; and subject to any actual constraints in terms of memory of information capacity with reduces the "set of possible questions" that are physical.

If it can't then its next position is objectively undetermined. - As is assumed in a mathematical stochastic process. If it could be known, but we just lack information then that is a subjective uncertainty. Every probability we know of outside of quantum mechanics is a subjective uncertainty, although the abstract random generator in math is always assumed to be an objective uncertainty. What DOES complicate this a bit in my view is when you start thinking about nonlocal relations and the role "future knowledge" plays.

I’m also sensing you are coming from a global entanglement point of view – like "many worlds" perhaps. I don’t share that idea though. In those models the universe is deterministic and there is never any objective uncertainty. So it is a concept that separates various interpretations.

Ouch no I am not into mwi

If we are talking about understanding "normal descriptive QM", I an more into statistical or copenhagen interpretation. But in this context, of trying to understand the QM foundations deeper, my stance is a kind of qbist inspired interpretation where there action under irreducible uncertainty is a key.

/Fredrik

 

[GentDave]

For me this is not a question I ask, it is not physical for me. Your hypothetical question implies infinite computational capacity. For me "physical questions" must be constructible from what the questioner (ie the observer or agent) has to work with; and subject to any actual constraints in terms of memory of information capacity with reduces the "set of possible questions" that are physical.

Ouch no I am not into mwi

If we are talking about understanding "normal descriptive QM", I an more into statistical or copenhagen interpretation. But in this context, of trying to understand the QM foundations deeper, my stance is a kind of qbist inspired interpretation where there action under irreducible uncertainty is a key.

/Fredrik

It make sense that your comments come from a Copenhagen interpretation. Initially it was sort of founded on the idea that anything we can't observe does not count. and there is only subjective uncertainty. I think that was an early wrong turn. An example I'd use is a black box that we can not open. We can test its function with inputs and observing outputs. But our best models may postulate moving bits inside we can't see. What we can't measure may still be important for forming our best theories.

 

[Fra]

An example I'd use is a black box that we can not open. We can test its function with inputs and observing outputs. But our best models may postulate moving bits inside we can't see. What we can't measure may still be important for forming our best theories.

This is fully in line with how I prefer to think as well. In fact i take it probably eve more serious, in that from the perspective of an agent/observer the future is always a "black box". And the learning, such as inference to best explanation, of what is not "directly visible", is a physical process.

In the simplifications of quantum mechanics, where we study small subsystems, which can be preparted many times and whose dynamics are shortlive, relative to the time of the "inference" (including tomoghraphic processes and acquiring statistics), it is indeed possible to infer an internal structure and dynamical law that does constitute the effective inference to best explanation.

In QM as we know it, this is presumed in the starting point. The "process" of inferring the state space and dynamical laws (states preps and hamiltoninans) are not considerd subject to "dynamics". It is treated outside. And this is precisely the unacceptable idealisation that prevents us from getting deeper.

If I understand the author he is saying the next position of the photon is objectively undetermined but its current position is always only a subjective uncertainty.

This makes sense yes.

I would put it so that the FUTURE is objectively undertermined because the abduction or inference to best explanation requires TIME so the future is fundamentally undetermined; all we have is our "best expectations of the future".

Put perhaps that PAST can be said to be the "subjectively undetermined" (beacuse even one might say that history is a "fact" no single subsystem/observer could have inferred and encode in a non-lossy way all this). So the "best description of the past" we have is still subject to uncertainty as any recording is necessarily incomplete and lossy.

This I agree resonates well with the memory effect, divisibility and non-markov features as well.

Copenhagen interpretation or statistical interpreation is what you get as as "limiting case"; for small short-lived "quantum-systems"; as described from a dominany macroworld, where information processing capacity is practically unlimited. And as this, and nothing else philosohical, was what the founders of QM worked on, I still think the old Copenagen or statistical interpretation is the best and most honest one, at least in the historical perspective.

/Fredrik

 

[GentDave]

This is fully in line with how I prefer to think as well. In fact i take it probably eve more serious, in that from the perspective of an agent/observer the future is always a "black box". And the learning, such as inference to best explanation, of what is not "directly visible", is a physical process.

In the simplifications of quantum mechanics, where we study small subsystems, which can be preparted many times and whose dynamics are shortlive, relative to the time of the "inference" (including tomoghraphic processes and acquiring statistics), it is indeed possible to infer an internal structure and dynamical law that does constitute the effective inference to best explanation.

In QM as we know it, this is presumed in the starting point. The "process" of inferring the state space and dynamical laws (states preps and hamiltoninans) are not considerd subject to "dynamics". It is treated outside. And this is precisely the unacceptable idealisation that prevents us from getting deeper.


This makes sense yes.

I would put it so that the FUTURE is objectively undertermined because the abduction or inference to best explanation requires TIME so the future is fundamentally undetermined; all we have is our "best expectations of the future".

Put perhaps that PAST can be said to be the "subjectively undetermined" (beacuse even one might say that history is a "fact" no single subsystem/observer could have inferred and encode in a non-lossy way all this). So the "best description of the past" we have is still subject to uncertainty as any recording is necessarily incomplete and lossy.

This I agree resonates well with the memory effect, divisibility and non-markov features as well.

Copenhagen interpretation or statistical interpreation is what you get as as "limiting case"; for small short-lived "quantum-systems"; as described from a dominany macroworld, where information processing capacity is practically unlimited. And as this, and nothing else philosohical, was what the founders of QM worked on, I still think the old Copenagen or statistical interpretation is the best and most honest one, at least in the historical perspective.

/Fredrik

I would agree the future is objectively undetermined, but many interpretations do not agree. The wavefunction delelops deterministicly and the future is fully determined by the present state of things in many interpretations. The one statement of yours I disagreed with was when you described one person and then another and another learning about a quantum result and that that changes the result (if I understood correctly). That implies a global entanglement and I don't thing that is the case.

 

[iste]

I would put it so that the FUTURE is objectively undertermined because the abduction or inference to best explanation requires TIME so the future is fundamentally undetermined; all we have is our "best expectations of the future".

Put perhaps that PAST can be said to be the "subjectively undetermined" (beacuse even one might say that history is a "fact" no single subsystem/observer could have inferred and encode in a non-lossy way all this). So the "best description of the past" we have is still subject to uncertainty as any recording is necessarily incomplete and lossy.

This I agree resonates well with the memory effect, divisibility and non-markov features as well.

Been reading, thinking more deeply Caticha's papers recently has some interesting stuff to say: (top two papers below)

https://scholar.google.co.uk/schola..._vis=1&q=caticha+entropic+dynamics&oq=caticha

If time is constructed purely based on informational entropic assumptions it must be naturally irreversible and is described in a Markovian sense, independent of the past. He asks then how can some processes be time-reversible and dependent on the past - like quantum mechanics - if time is irreversible. Interestingly, the conditions for / that prevent time-reversibility actually seem to follow from Bayes theorem, and can be cashed out in the following way: the Markovianity conditions only work in one direction, which contradicts time-reversibility and so the system will look different forward in time vs backward - the difference between backward and forward is actually in Bayes' theorem. Seems things shouldn't work in a non-Markovian way under informational-entropic time; as he puts it - there is no symmetry between the inferential past and inferential future. And this statement is inherently connected to Markovianity.

So it seems from Caticha's papers that you cannot get time-reversibility and non-Markovianity without something extra. He requires extra-variables in addition to particles in his quantum mechanics. Dynamics of extra variables influence dynamics of particles and dynamics of particles influence dynamics of extra-variables so its reciprocal, and this stops the irreversibility of time. But the way he does it seems kind of tautological in the sense that he just asserts a condition where something called "energy" is conserved; but, "energy" here is literally defined as just whatever is conserved in the dynamics due to invariance in time-translation - which seems to be another way of saying... uhh, it just is.

But I think it can still be made more tangible in the following sense - if you assume that the Markovian time is occuring under an open system which interacts with its environment through force and energy. If we don't assume anything about what it gets back from the environment we have Markovianity as the random changes to the system described by entropy maximization are not counterbalanced - we also don't know where the force goes because we have decided to arbitrarily focus on this small compartment of a larger whole isolated system . Adding the extra variables may just be a way of modeling (even if just an idealization) isolatedness - in the sense that if the system is isolated, then the force being exerted elsewhere hasn't left the system - but it needs to go somewhere, so we can talk about these extra variables as where this force goes. We can then effectively assume that everything lost by the particles, and exerted on the extra-variables, is returned - nothing is leaked out the system and so it is no longer Markovian. Because the informational-entropic time is based around the idea that the transition probability distributions maximize entropy, but only in one direction, I guess one could see that as dynamics over time are inherently characterized by the loss of information which maybe is returned by the extra-variables.

The coupling of the particles to the extra variables to my eyes then looks, at least superficially, kind of like a recurrent neural network, especially an Elman network.

https://en.wikipedia.org/wiki/Recurrent_neural_network#Elman_network

The y units in the Elman depiction receive inputs that change their state at one time-step, then outputs from the y unit go into the  u unit which recycle y's outputs back to it at the next time-step - this means y not only depends on its immediate inputs but its history - what happened in the past. This recurrence then is the very basic principle of how recurrent neural networks (and indeed brains where recurrence is fundamental) can learn sequential, history-dependent information (e.g. what comes next in the sentence based on what came before, but modern large language models are based on a more efficient way of doing this than recurrent neural networks though). Maybe this is a way of an intuition for the non-Markovianity - the particle outputs are fed back onto them so that nothing leaks out (In terms of physical energy, neural activity), but simultaneously rendering the system history-dependent and time-reversible.

The extra-variables take the role of the phase in quantum mechanics. Caticha later plays down the extra-variables because he doesn't have an interpretation at hand; but there is a trade-off in the sense that by playing down the extra-variables, you just have to make certain assumptions about the existence of phase and related things - but the extra-variables give you them all for free.

What perhaps is most interesting is that quantum-type non-locality appears to occur in Caticha's Newtonian theory (first two papers below but specifically second paper, section 5):

https://scholar.google.co.uk/schola...is=1&q=caticha+newtonian+inference+to+physics

You see in his Newtonian system for multiple particles in an isolated system, their Newtonian behaviors depend on each other in a similar way to Bohmian mechanics (but note equation (39) and the following note: particles still are statistically conditionally independent). Seems to me this is from the fact that the Newtonian behavior conserves energy. And interestingly, in later papers you see that both the quantum potential in Bohmian mechanics, responsible for quantum non-locality; and the quantum configuration space both are consequences of the information metric which characterizes his entropic dynamics even in the Newtonian case (and curvedness is related to information): i.e.

https://scholar.google.co.uk/scholar?cluster=2247357911529834348&hl=en&as_sdt=0,5&as_vis=1

Caticha's Newtonian dynamics are obviously time-reversible and conserve energy but it still emerges from the lossy entropic dynamics, evident in the fact that the Newtonian dynamics is only along the most probably path: i.e. analogous to path of least action.

 

[Fra]

Been reading, thinking more deeply Caticha's papers recently has some interesting stuff to say: (top two papers below)

https://scholar.google.co.uk/scholar?hl=en&as_sdt=0,5&as_vis=1&q=caticha+entropic+dynamics&oq=caticha

If time is constructed purely based on informational entropic assumptions it must be naturally irreversible and is described in a Markovian sense, independent of the past. He asks then how can some processes be time-reversible and dependent on the past - like quantum mechanics - if time is irreversible. Interestingly, the conditions for / that prevent time-reversibility actually seem to follow from Bayes theorem, and can be cashed out in the following way: the Markovianity conditions only work in one direction, which contradicts time-reversibility and so the system will look different forward in time vs backward - the difference between backward and forward is actually in Bayes' theorem. Seems things shouldn't work in a non-Markovian way under informational-entropic time; as he puts it - there is no symmetry between the inferential past and inferential future. And this statement is inherently connected to Markovianity.

So it seems from Caticha's papers that you cannot get time-reversibility and non-Markovianity without something extra. He requires extra-variables in addition to particles in his quantum mechanics. Dynamics of extra variables influence dynamics of particles and dynamics of particles influence dynamics of extra-variables so its reciprocal, and this stops the irreversibility of time. But the way he does it seems kind of tautological in the sense that he just asserts a condition where something called "energy" is conserved; but, "energy" here is literally defined as just whatever is conserved in the dynamics due to invariance in time-translation - which seems to be another way of saying... uhh, it just is.

But I think it can still be made more tangible in the following sense - if you assume that the Markovian time is occuring under an open system which interacts with its environment through force and energy. If we don't assume anything about what it gets back from the environment we have Markovianity as the random changes to the system described by entropy maximization are not counterbalanced - we also don't know where the force goes because we have decided to arbitrarily focus on this small compartment of a larger whole isolated system . Adding the extra variables may just be a way of modeling (even if just an idealization) isolatedness - in the sense that if the system is isolated, then the force being exerted elsewhere hasn't left the system - but it needs to go somewhere, so we can talk about these extra variables as where this force goes. We can then effectively assume that everything lost by the particles, and exerted on the extra-variables, is returned - nothing is leaked out the system and so it is no longer Markovian. Because the informational-entropic time is based around the idea that the transition probability distributions maximize entropy, but only in one direction, I guess one could see that as dynamics over time are inherently characterized by the loss of information which maybe is returned by the extra-variables.

The coupling of the particles to the extra variables to my eyes then looks, at least superficially, kind of like a recurrent neural network, especially an Elman network.

https://en.wikipedia.org/wiki/Recurrent_neural_network#Elman_network

The y units in the Elman depiction receive inputs that change their state at one time-step, then outputs from the y unit go into the  u unit which recycle y's outputs back to it at the next time-step - this means y not only depends on its immediate inputs but its history - what happened in the past. This recurrence then is the very basic principle of how recurrent neural networks (and indeed brains where recurrence is fundamental) can learn sequential, history-dependent information (e.g. what comes next in the sentence based on what came before, but modern large language models are based on a more efficient way of doing this than recurrent neural networks though). Maybe this is a way of an intuition for the non-Markovianity - the particle outputs are fed back onto them so that nothing leaks out (In terms of physical energy, neural activity), but simultaneously rendering the system history-dependent and time-reversible.

The extra-variables take the role of the phase in quantum mechanics. Caticha later plays down the extra-variables because he doesn't have an interpretation at hand; but there is a trade-off in the sense that by playing down the extra-variables, you just have to make certain assumptions about the existence of phase and related things - but the extra-variables give you them all for free.

What perhaps is most interesting is that quantum-type non-locality appears to occur in Caticha's Newtonian theory (first two papers below but specifically second paper, section 5):

https://scholar.google.co.uk/schola...is=1&q=caticha+newtonian+inference+to+physics

You see in his Newtonian system for multiple particles in an isolated system, their Newtonian behaviors depend on each other in a similar way to Bohmian mechanics (but note equation (39) and the following note: particles still are statistically conditionally independent). Seems to me this is from the fact that the Newtonian behavior conserves energy. And interestingly, in later papers you see that both the quantum potential in Bohmian mechanics, responsible for quantum non-locality; and the quantum configuration space both are consequences of the information metric which characterizes his entropic dynamics even in the Newtonian case (and curvedness is related to information): i.e.

https://scholar.google.co.uk/scholar?cluster=2247357911529834348&hl=en&as_sdt=0,5&as_vis=1

Caticha's Newtonian dynamics are obviously time-reversible and conserve energy but it still emerges from the lossy entropic dynamics, evident in the fact that the Newtonian dynamics is only along the most probably path: i.e. analogous to path of least action.

Lots of things to comment here... not sure where to begin.

I like Caticha, I have read most of his papers, and ET Jaynes half-ready book but this was many years time ago and he was an early inspiration, and while I share alot of the idea of Ariel when it comes to inference and physics, I different in some critical parts in his thikning. He largely thinks of inference as as human probabilistic description of nature, and that this in itself has implicates on the laws. While this is true, I unlike Caticha, have a problem to start with the real numbers as an intrinsic measure of degree of belief because it requires the "memory constraint" to be patch onto things, in a way that I think gets complicated and risks leading to pathologies of divergences later in development.

I share parts of the entropic dynamics thinking, but I don't share the way he does it completely. I think the way that each agent/observer see and "entropic flow"; but as this is conditional to the agent, different observers have different parameterisation of time. Ie this time is relative to the obsever, and only valid differentially (before new information update) like in relativity. So I think the parameterisation of the "expected change" is entropic, the global change is not; it is evolutionary and the difference is that the "microstructure" relative to which entropic flow can be defined, is in principled evolving, this is why they entropic time makes sense for me only for short time.

But as I think different in many basic things, while I love Catichas papers, I stopped comparing his details to how I think years ago. I traced things back to this introduction of the degree of belief, and he basically goes with continuum probabiliti and real numbers, I find this probolematic and wants to "reconstruct" measure itself from ang agent with memory constraints to get an instrinsic measure. Here an uncountable number system is a nono for me. it is a valid embedding, but as the embedding is obvioulsy way too large, it creates the problem of tracking limts, and the tracking is unfortunately lost fast, which requires that you keep patching things with normalizations and other constraints all the time.

/Fredrik

 

[Fra]

The coupling of the particles to the extra variables to my eyes then looks, at least superficially, kind of like a recurrent neural network, especially an Elman network.

https://en.wikipedia.org/wiki/Recurrent_neural_network#Elman_network

The y units in the Elman depiction receive inputs that change their state at one time-step, then outputs from the y unit go into the  u unit which recycle y's outputs back to it at the next time-step - this means y not only depends on its immediate inputs but its history - what happened in the past. This recurrence then is the very basic principle of how recurrent neural networks (and indeed brains where recurrence is fundamental) can learn sequential, history-dependent information (e.g. what comes next in the sentence based on what came before, but modern large language models are based on a more efficient way of doing this than recurrent neural networks though). Maybe this is a way of an intuition for the non-Markovianity - the particle outputs are fed back onto them so that nothing leaks out (In terms of physical energy, neural activity), but simultaneously rendering the system history-dependent and time-reversible.

Yes, although there are many options for how to encode the past dependence, a neural network if this type, where the hidden layer is a memory, is indeed the kind of intuition I tried to describe as well in previous posts. There is more one can discuss here, but as far as I recall Caticha is not taking this very far.

But in these cases, the most obvious "natural" intuition to a conserved quantity is memory. Ie. an agent (or neural network) can reorganize, say add additional layer, but doing so requires sharing finite memory, and it requires additional computations, so at some pint it seems reasonble to think that - there is an optimal internal microstructure of hte agent/network or what we calle it, where the agent has maximal fitness in a given environment (which defines the inputs and feedback to outputs). This may even allow for future associations of concepts such as mass~inertia and energy, in terms of such things. The question of mass generation in phusics, might thus relate to how can a neural agent increase it's "effective memory"? Maybe by cooperating with the environment, exploit patterns and use instead "implicit memory". when you start in this direction, we can release ourselves from Ariels view of physics are inference, in the sense of human tools. Perhaps we can undo some of the ad hoc constrains or assumttions we initially add. Perhaps physics can even be inference in the sense of inter-material tools, to handle relations? But solving everything at once is hard, so a first step is perhaps to see, is what Ariel is doing. Then the next step would be to relase ourselves from the broad human context or mathematics, and try to see what natures instrinsic "mathematics" is like? Perhaps then we need less ad hoc constraints?

Here there is plenty of "intuition" for non-markovianty and also emergence.

/Fredrik

 

[pines-demon]

J. Barandes did another podcast with Curt Jaimungal:

He seems to discuss INMS. I have not listened to it yet, I will report back when I can and only if it brings anything new to the discussion.

 

[pines-demon]

In the new video he gives two or three thoughts that could be important here:

1. simple non-chaotic Newtonian mechanics is non-Markovian. You need the position $\mathbf r(t)$ at some time $t$ but also the position an infinitesimal time before ($\dot{\mathbf r}(t)\mathrm d t$), the present state does not suffice. The way that we make the dynamics Markovian is to expand into the configuration space $(\mathbf r,\dot{\mathbf r})$ (from 3D to 6D). Quantum mechanics according to him is the same idea, you have some extremely non-Markovian system, and forcing it to be Markovian (Schrödinger equation) introduces phases and interference terms.
2. He seems to have dropped the idea of memory. He says
   

   These are the prices you pay [interference terms to make it Markovian]. We talked before about, like, are these memory effects? Memory is not quite the right metaphor for indivisible stochastic processes. Traditional non-Markovian processes require that you specify sort of conditional probabilities conditioned in arbitrarily many previous times. That's very complicated and has a lot of structure and contains a lot of information. And you can legitimately ask, where is that information being stored? An indivisible process is actually much simpler. It doesn't entail the specification of all those higher order non-Markovian probabilities. It's much sparser.

   He does not say much more unfortunately. Barandes compares his discovery with path integral formulation which is equivalent to other formulations.
3. He is working on a draft on the "Hello-world example", a qubit example. Is he reading PF? (he actually hand-wave explains it, but does not provide the interpretation part).

That's what I got so far, he explains where linearity and unitarity comes from (the latter comes from a relation with unistochastic processes). But I am still trying to dissect if there is an interpretation here. He takes more time to discuss the importance of analytic philosophy in science. I will add more if I get to watch it completely.

He left Bell's inequalities to a future video...

 

[iste]

simple non-chaotic Newtonian mechanics is non-Markovian. You need the position r(t) at some time t but also the position an infinitesimal time before (r˙(t)dt), the present state does not suffice. The way that we make the dynamics Markovian is to expand into the configuration space (r,r˙) (from 3D to 6D). Quantum mechanics according to him is the same idea, you have some extremely non-Markovian system, and forcing it to be Markovian (Schrödinger equation) introduces phases and interference terms.

I do also believe he says it isn't non-Markovian in the same sense though, if iirc - the quantum case is not the same as that procedure which I believe is closely related to what is said in your next quote about conditioning on arbitrarily many times.

He seems to have dropped the idea of memory.

It really depends what you actually mean by memory. I was under no illusions that he meant the kind of memory he describes in that quote.

But I am still trying to dissect if there is an interpretation here.

I think it just has to be accepted that Barandes has no interpretation of how it happens even though he clearly states in the video that at all times the system has a definite configuration, even during periods ordinarily described by superposition. He can point to indivisible stochastic processes as the explanation of quantum mechanics but the correspondence theorem is so general that it can never explain exactly why something is indivisible because it applies to literally any indivisible system. If physical systems are indivisible they can be turned into a quantum system, if cognition is indivisible it can be turned into a quantum system, if the dynamics of ecological systems are indivisible they can be turned into quantum systems. Barandes seems to just go down the route that he believes this may just be a fundamental physical law which doesn't need explanation because he thinks that indivisible stochastic systems are plausibly the most general kind of dynamical system.

If you want a more palatable interpretation (imo) of why an indivisible stochastic process might behave the way it does, I suggest looking at stochastic mechanics which will be equivalent to Barandes' formulation when looking at regular physical quantum systems and is extremely well-established as a formulation and research programme involving many physicists and mathematicians, stretching back well into the mid-twentieth century. Recent review:

https://www.mdpi.com/2218-1997/7/6/166

Beyer gives three assumptions used to construct quantum mechanics from a stochastic process / diffusion:

1. The form of the diffusion coefficient, featuring Plank's constant and inversely proportional to particle mass.

2. The diffusion is non-dissipative - i.e. it conserves energy on average.

3. The system behaves according to Newton's second law, F= ma, with regard to its average motion.

From a physical system constructed this way, all quantum behavior can be reproduced.

Assumption 2 is enforced by something called the osmotic velocity, which was invoked by Einstein in his studies of Brownian motion and is defined as "the velocity acquired by a Brownian particle, in equilibrium with respect to an external force, to balance the osmotic force". If you want to take the meaning of the osmotic velocity literally, as originally defined in the Einsteinian context of stochastic systems, then the natural interpretation implied is that the randomness in the stochastic system is being caused by its interactions with another background system. This is conceptually no different to the archtypical Brownian motion of a dust pollen in a glass of water being pushed about by the background water; but in the quantum case, the particle's exchanges with the environment conserve energy on average. This is the broad consensus in stochastic mechanics on how to make quantum mechanics intelligible as a stochastic system, the background system mentioned a couple times passingly in the Beyer paper.

Hydrodynamic pilot-wave experiments provide an interesting toy model for how the stochastic mechanical particle-background system could produce quantum behavior in an intelligible way. These toy models produce quantum-like, nonlocal-like, non-markovian behaviors in oil droplets which are being pushed around by their interactions with the fluid bath. The behaviors only manifests when viscous dissipation in the fluid bath is counteracted by vibrating it, injecting energy to balance the losses from dissipation. This mechanism therefore looks like it is artificially mimicking assumption 2 from earlier, and enables information about the global configuration of the system to be stored in the fluid bath and affect droplet behavior. Some reviews of the behaviors in these toy models: Review1, Review2, Review3.

I think this is the most intelligible way of interpreting a stochastic formulation of quantum mechanics, like Barandes'. The strangeness of the double slit experiment would then be mediated by the influence of the background system on the particle. The slit configuration would indirectly influences the particles via the background once the whole system has relaxes into the stochastic mechanical quantum equilibrium (which takes a very short but finite amount of time, and is seen in stochastic mechanics simulations).

The background system seems like quite a huge assumption to make in order to explain the indivisible stochastic process; but there is already vacuum energy and fluctuations in quantum field theory. A resemblance to the kind of pervasive background ontology that stochastic mechanics postulates is then already alluded to in orthodox quantum theory. At worst, the stochastic mechanical background hypothesis is no more radical than what you would otherwise have to still entertain, even if only in an instrumentalist capacity, in quantum theory. I also prefer the intelligibility of such a hypothesis to Barandes' alternative that potentially this indivisibility just represents some brute fundamental law. Ofcourse, the non-dissipative, conservative nature of the background may also warrant explanation; but imo it doesn't immediately beg for an explanation in the same way that the bizarre indivisibility does. Nor does a possible explanation in the future seem implausible imo.

 

[JC_Silver]

1. He is working on a draft on the "Hello-world example", a qubit example. Is he reading PF? (he actually hand-wave explains it, but does not provide the interpretation part).

I wasn't aware PF existed until I Googled "Jacob Barandes Stochastic Interpretation" and this forum was on the first page, so I wouldn't doubt he did read some posts here.

I'm excited about the paper he says he's writing, I want to try doing some calculations using his method, for fun, but I need to see an example done first, as I haven't finished reading his sugested paper on the indivisible process. (this one: https://journals.aps.org/prxquantum/abstract/10.1103/PRXQuantum.2.030201)

 

[physika]

Realism these days seems to be something related to pre-xistence of values before

Not necessarily.
It refers to entities that may or may not have properties. But such entity exists, whether it has properties or not, and therefore it is REAL.

......

 

[Fra]

i havent' had time to wach the long clip yet, but it's on my todolist

1. He seems to have dropped the idea of memory. He says
   He does not say much more unfortunately. Barandes compares his discovery with path integral formulation which is equivalent to other formulations

These are the prices you pay [interference terms to make it Markovian]. We talked before about, like, are these memory effects? Memory is not quite the right metaphor for indivisible stochastic processes. Traditional non-Markovian processes require that you specify sort of conditional probabilities conditioned in arbitrarily many previous times. That's very complicated and has a lot of structure and contains a lot of information. And you can legitimately ask, where is that information being stored? An indivisible process is actually much simpler. It doesn't entail the specification of all those higher order non-Markovian probabilities. It's much sparser.

What is written in bold here, makes sense to me. Clearly it makes no sense to consider ALL history retained, I think that was never what is meant by memory effect either. So his questioning of where this information is stored is the right quesion. And in my view, this is stored in part in the agents microstructure and implicitly in the environment; there is necessarily a lossy retention at play here, so that we can answer the questions "where is it stored". This in a way is "much sparser"; buy they exact way it is sparser is likely evolutionary emergent and remains to be understood.

This is similar to how the result of billions of years worth of evolutionary training, is "retained" in simple organisms. Clearly the history of all the universse in detail can not possible be encoded in these parts. But they retained the critical parts (sparse is an understatement!) relative to the similary evolving environment that supports the "simple encoding"; where the higher orders are truncated, beacuse of limited memory. So I think the fact that there is a "memory effect" and that the memory is limited, and explicitly constrains the effect are both equally important.

I think the idea of "memory" is similar in foundations of QM, or it is at least how I see the logic beging behindes formulation.

I will try to listend to the talk when i get time and get back with impressions of Barandes talk.

/Fredrik

 

[Fra]

J. Barandes did another podcast with Curt Jaimungal:

I now went through this talk and my impression is that I really have no objections to anything Barandes says, I like his angles and emphasis on philosophical perspective, and his explanations of the correspondence between hilbert space, unitary evolution, and the non-divisible markovian picture which has a natural correspondence to a basis choice. Ie. the hilber space and "wave interference" is and apparent artifact and price you pay from reformulating the probabilistic non-divisible non-markovian original picture.

But he does not address the question of the origin and physical nature of the "probability spaces" and the transitiion amplitudes. He however notes that if we assume this; the hilber formalism simplifies that math; and the side effect is some of the "interfereance weirdness" and complex numbers. I fully agreee on that.

But this only transforms the questions into; what is the physical basis and ontology and preference for the particular probability spaces and their transition amplitudes? And how is this "selected" or "tuned" in the perspective of unification? My original impression on Baranders still is the same, I think it is good work, but many deep issues remains, and from the youtube clip these seems to be uncommented.

/Fredrik

 

[JC_Silver]

Seems like the second half of the interview is out, but for members only.
It's a bummer since paying for anything in dollars is beyond what I can afford.

 

[Fra]

I think it just has to be accepted that Barandes has no interpretation of how it happens even though he clearly states in the video that at all times the system has a definite configuration, even during periods ordinarily described by superposition.

I agree he seems to have no clear interpretation; this is why it's hard to follow him. All I can do is "fill in" with my own interpretation, and then it makes sense and harmonized, but what Barande himlself actually things is not clear.

But I think that there is a key here, that is often confused: Ontology and the nature of causality is I think one of the things that are most often confused in the normal paradigm. It tends to be understood "mechanistically" rather than via "information". This is the case even in quantum mechanics, as long as we still don't have a better foundation.

This also makes sense, as Jacob Baranders says himself he is also interested in "metaphysics of causation", and this is exactly what is deeply insatisfactory in the classical paradigm, and IMO at least, this is also exactly where the indivisibility enters.

immediately beg for an explanation in the same way that the bizarre indivisibility does

Indivisivilibly is bizarre only if you have a "simplistic view" on the nature causation.

I would say that if you instead see the nature of causation in the context of interacting and information processing agents(observers) then indivisiblity of the "expected evolution" - which is exactly what the schrödiner equation is, nonthing else - it becomes even almost obvious, that a division, BREAKS the expectation. There is nothing bizarre about this. They insight is rather than expectations is always contextual, or observer dependent.

So IMHO, the only way to see clearly why this is natural, is to complement his ideas with an interpreation, and I think acqknowledging the importance of the observers the "stochastic process" is the most obvious solution. But none of this is in his talks. But as someone that has been thinking alot about this, I know there is a big can of worms also, coming with new problems if you go this route, so perhaps he wants to refrain from exposing this can of worms in public. I do not blame him, as it might be held against.

So keeping the exposition as a more mathematical equivalence and refrain from "speculating" too much how to interpret it, is perhaps a clever professional strategy.

/Fredrik

 

[pines-demon]

Seems like the second half of the interview is out, but for members only.
It's a bummer since paying for anything in dollars is beyond what I can afford.

It will release in a couple of days, and then we have to watch about 1-2 hours of it, it can wait.

 

[JC_Silver]

It will release in a couple of days, and then we have to watch about 1-2 hours of it, it can wait.

Oh, I wasn't aware that's how that channel worked.
Sorry for sharing anticipated worry

 

[iste]

They insight is rather than expectations is always contextual, or observer dependent.

Sure, but the question is why should there be observer-dependence? Why do interference patterns occur due to adding slits? Neither you or Barandes have offered a deeper explanation of why these things should be considered the norm or non-bizarre in contrast to everyday experience.

and I think acqknowledging the importance of the observers the "stochastic process" is the most obvious solution. But none of this is in his talks.

I think that because the observers are all explicitly included in the model as just sub-systems of the overarching stochastic process there is not really anything else to say about them. It isal laid bare in the formulation and the distinction between observer and non-observer is not special - they are just a kind of physical system which is capable of a precise coupling to whatever states of another system that we find interesting.

 

[Fra]

Sure, but the question is why should there be observer-dependence? Why do interference patterns occur due to adding slits? Neither you or Barandes have offered a deeper explanation of why these things should be considered the norm or non-bizarre in contrast to everyday experience.

You are right of coruse, I see two ways to "answer" this in a big perspective...

1) Via an interpretation where this is preferred

2) To reformulate or reconstruct current theory in a way where the observers have a different role, and show that this reconstructed theory solves many problems; thus this will "prove" the point, just like other reformulations in history proved very viable, such as geometric methods etc. one might as well, why on earth would "geometrization" be better? The answer is I think that someone tried it, and it proved to be very powerful method. I think Baranders point is that again, there may be new formulations that with time will probe powerful and solve old problems

(2) is an open question, and NOone has the answer and published it, and speculations aren't allowed here, so we are left with sniffing around (1), but these arguments does not convinced anyone who is already comitted deeply to anoher interpretation, but it can provide an explanation to those that are not.

Not sure at else I can say that isn't already said in the "mode" of (1). But I had some things i could add, but be back later or tomorrow when i get more time

/Fredrik

 

[gentzen]

You are right of coruse, I see two ways to "answer" this in a big perspective...

1) Via an interpretation where this is preferred

2) To reformulate or reconstruct current theory in a way where the observers have a different role, and show that this reconstructed theory solves many problems; thus this will "prove" the point, just like other reformulations in history proved very viable, such as geometric methods etc. one might as well, why on earth would "geometrization" be better? The answer is I think that someone tried it, and it proved to be very powerful method. I think Baranders point is that again, there may be new formulations that with time will probe powerful and solve old problems

(2) is an open question, and NOone has the answer and published it, and speculations aren't allowed here, so we are left with sniffing around (1), but these arguments does not convinced anyone who is already comitted deeply to anoher interpretation, but it can provide an explanation to those that are not.

Not sure at else I can say that isn't already said in the "mode" of (1). But I had some things i could add, but be back later or tomorrow when i get more time

I am so upset that even you talk of „observers“ now that I have to comment: One unintuitive feature of QM is that you have to interact with a system to observe it. And this interaction alters the system. Classically, I can also alter a system without learning anything about its state prior to the interaction. How does QM change this part? Why can‘t I just forget what I learned during the interaction, if QM should force me to always learn something? Perhaps because my own state was altered nevertheless? (Now you will edit to replace observer by agent, or maybe you already edited. But you and the QBists need to do more to liberate us from this unhealthy focus on observation.)

 

[gentzen]

I would say that if you instead see the nature of causation in the context of interacting and information processing agents(observers) then indivisiblity of the "expected evolution" - which is exactly what the schrödiner equation is, nonthing else - it becomes even almost obvious, that a division, BREAKS the expectation. There is nothing bizarre about this. They insight is rather than expectations is always contextual, or observer dependent.

OK, this was a different post which you edited. But still, one of the "possible actions" I thought about was to write to Chris Fuchs and beg him to exand on the non-observer part of agents, after realizing that Barandes had done good work on QM foundations before he fouled himself into believing that he has a new formulation of QM (similar to Feynman's path integral formulation) or solved the interpretation of the measurement problem.

A much later thought (after watching the recent ToE video with Jacob Barandes and Manolis Kellis) was that Jacob's point that Hilbert spaces give a misleading picture of QM is similar to saying that binary digits give a misleading picture of computers. You might object that our computers today really are based on binary states, and that all attempts to use tenary states instead proved less reliable or less efficient. But the DNA uses quaternary base units, and tenary alternatives can be more efficient or unbiased for survey questions. Tenary also allows less symmetry breaking answers, which is one reason to rather avoid it in surveys. (Faced with the answer options "yes", "no", and "undecided", some people would go for "undecided" as their default answer for most questions. Which could be a point about the randomness in QM, it also feels like symmetry breaking.)

 

[iste]

Barandes had done good work on QM foundations before he fouled himself into believing that he has a new formulation of QM (similar to Feynman's path integral formulation) or solved the interpretation of the measurement problem.

Why isn't it a new formulation? I would say it solves the measurement problem in its own context as a full formulation, Bohm does similar, and Many Worlds if I am not mistaken. Maybe an emergence of classical world is needed for a full explanation, which I think for Barandes involves decoherence.

 

[gentzen]

Why isn't it a new formulation?

It is not a new formulation, because it is not comparable to Feynman's path integrals, to Schrödinger's wave mechanics, to the matrix mechanics of Born and Heisenberg, or to Dirac's transformation theory. And it is simply missing the resources to go from physical situation to mathematical description.

I would say it solves the measurement problem in its own context as a full formulation, Bohm does similar, and Many Worlds if I am not mistaken. Maybe an emergence of classical world is needed for a full explanation, which I think for Barandes involves decoherence.

Bohm and Many Worlds are interpretations. But Barandes himself already noticed that he doesn't have an interpretation in the same sense. So he called it a "formulation" instead. Maybe this was even fine for his first two papers, while he still tried to flesh it out. But in his third paper, he simply declared victory, without actually succeeding to flesh it out any further.

My guess is that what he actually could get if he successfully fleshed out what he has is a "quantum reconstruction", not an "interpretation" or a "formulation":

The pure formalism presented in the two older papers suffered from an unclear status of causal locality. I have not studied the newest paper in any detail yet, but if it manages to overcome this problem, then it constitutes nice incremental progress for this new formulation.

I read some relevant parts of the "new" paper now. I don't think that "it constitutes nice incremental progress for this new formulation". He does his calculation regarding locality in the normal Hilbert space formulation. I admit that he did his definitions regarding locality both in his formulation and in Hilbert space.

Barandes would do better to put his "new formulation" in the context of "quantum reconstructions" rather than "quantum interpretations". (Then he could check whether his math contains new insights, and also whether he was careful enough to show all "expected" continuity properties of his construction with respect to the time parameter t.) And his "Causally Local Formulation" turns out to be just the well known "no signaling" property of QM, verified with calculations done using normal QM (not his new formulation).

 

[Fra]

I am so upset that even you talk of „observers“ now that I have to comment:

I am a not sure I get what is making you upset? Is it about the word "observer"? By observer i mean a physical observer, an agent. If I ever mean a human, I will write physicists, not observer. But I don't think that's what upsetting you?

Maybe I am starting to defend what I wrote now based on misundersstanding your issue with this, but anyways...

One unintuitive feature of QM is that you have to interact with a system to observe it.

How can you possibly or reasonably observe something without interacting with it? For me this is as intuitive as it gets, the opposite is rather unintutive for me.

Classically, I can also alter a system without learning anything about its state prior to the interaction. How does QM change this part? Why can‘t I just forget what I learned during the interaction, if QM should force me to always learn something? Perhaps because my own state was altered nevertheless?

As I see QM, it is not about random (or specifically suboptimal) inference, as there is no lower bound on that. It is about optimal inference, it what is the maximal amount of information we can have about a another system; given some contraints.

In normal QM or Copenhagen interpretation, I think the "observer" is the whole macroscopic environment, so information capacity and processing capacity is not a constraint, the only constraints are that certain information simply does not commute, because they are not independent. Obviously the conjugate variables are not independent - by definition.

Also in normal QM, the backreaction of the whole macroscopic environemnt is ignored.

But when you consider a more realistic observer, ie agents with finite mass, then information capacity and backreaction can no longer be ignored. And this is when it becomes interesting, but this part is also which is not included in current theory and is an open problem. Taking this seriously in a future model, must associate with each "observer" a natural regulator, which implies lossy retention, and the "backreaction" is simply the update of what is normally considered the "background".

(Now you will edit to replace observer by agent, or maybe you already edited. But you and the QBists need to do more to liberate us from this unhealthy focus on observation.)

For me "observation" and "interaction" is just two perpectives of the same thing; there is no contradiction nor equivalence. There is a "tension", which translates into interaction terms, analogous to gauge interactions. So the interaction between observers, is loosely analgous to "gauge interactions". But the symmetries are never perfect, as they might evolve and be emergent. This is why - assuming them to be "complete" leads typically to fine tuning of too big spaces. This is the heart of the problem for me.

The intrinsic "inside view" realtive to one of the participating sytems is "observation", the "extrinsic view" is interaction. But the latter containts more information; and this perspective, when trying to unify forces leads to fine tuning.

The inside view avoids fine tuning as there is a natural normalization, the flip side side is that you get "subjectivity" or "obserber dependence", but there is supposed to be a solution to this - but which is not solved yet. This the "can of worms" I mentiond.

OK, this was a different post which you edited. But still, one of the "possible actions" I thought about was to write to Chris Fuchs and beg him to exand on the non-observer part of agents

I still see a possibilit that I misunderstand you, or that we misunderstand each other on the word observer.

What do you mean by "non-observer part of agents"? If I understand what you mean, I can at least add how I see it.

/Fredrik

 

[iste]

But Barandes himself already noticed that he doesn't have an interpretation in the same sense

Disagree. I think its clear he is espousing a similar interpretation to stochastic mechanics, he just doesn't have a strong explanation of why indivisibility occurs. But then again, neither Bohmians nor Evereytians know why quantum systems behave the way they do in any deeper fashion.

 

[gentzen]

I am a not sure I get what is making you upset? Is it about the word "observer"? By observer i mean a physical observer, an agent. If I ever mean a human, I will write physicists, not observer.

I see. It is not the "human" part which upsets me about the word "observer". It is the suggested absence of "interaction" and "altering"/"influencing" capabilities. The word "observer" suggest a passive viewer of some "TV show" or "cinematic movie", not an active participant in some "massively multiplayer online game".

Or maybe more aptly, it suggests an old time astronomer like in Newton's times watching the planets, moons, and stars without any possibility to alter their course. It does not suggests a modern NASA scientist designing swing-by (gravity assist) maneuvers to steal a tiny amount of the energy of some planet or moon to let his spacecraft save fuel.

What do you mean by "non-observer part of agents"? If I understand what you mean, I can at least add how I see it.

I mean the interaction and altering the system part, including but not limited to the preparation of some system. In QM one sometimes talks about preparing some system in its thermal state by waiting long enough, but other than that, the active part of macroscopic agents is rarely ever discussed.

 

[Fra]

I see. It is not the "human" part which upsets me about the word "observer". It is the suggested absence of "interaction" and "altering"/"influencing" capabilities. The word "observer" suggest a passive viewer

Ah now I see your objection! I do not mean to imply lack of "interaction" - on the contrary do i want to "involve" the observer more than the conventional picture, MUCH more. Which i think should be seem from my other comments.

So I dont think we have a disagreement here. It was misunderstandings.

More later about the actual issue.

/Fredrik