The thermal interpretation of quantum physics

  • #1
A. Neumaier
Science Advisor
Insights Author
2019 Award
7,262
3,144

Main Question or Discussion Point

I think Dr Neumaier has a good point - QFT may indeed be a better place for interpretations. I do not know enough of his thermal interpretation to comment on its specifics.
A complete description of the thermal interpretation of quantum physics can be found in my just finished papers (for the bare bones, see Section 2.5 of Part II)

Foundations of quantum physics I. A critique of the tradition,
Foundations of quantum physics II. The thermal interpretation,
Foundations of quantum physics III. Measurement.

They are also accessible through the arXiv at arXiv:1902.10778 (Part I), arXiv:1902.10779 (Part II), arXiv:1902.10782 (Part III). The simplest quantum system, a qubit, was already described by Stokes 1852, in terms essentially equivalent to the thermal interpretation.

This is a very long thread. DarMM gave in post #268 a nice summary of the thermal interpretation. Post #484 is my very short summary and post #479 contains links to my explanations of the connections between theory and experiments. Open problems related to the thermal interpretation are discussed in post #293.

An additional fourth part (from April 27, 2019) is announced here, and a fifth part (from May 2, 2019) completing the series is here. Reviews (Part I, Part II, Part III, Part IV, Part V) are on PhysicsOverflow, together with some comments by me. The book
gives a polished exposition based on these papers.

Related threads are about the thermal interpretation of the Stern-Gerlach experiment and the double-slit experiment, about the derivation of atomic and molecular spectra in the thermal interpretation, about the relation to decoherence (which gives an averaged description of the description of measurements given by the thermal interpretation), and about an example illustrating the differences in the interpretation of measurement results in the thermal interpretation and in Born's statistical interpretation. Another thread is about an attempt to disprove a misunderstood version of the thermal interpretation.
 
Last edited:
  • Like
Likes Lynch101, mattt, Auto-Didact and 10 others

Answers and Replies

  • #2
  • #3
A. Neumaier
Science Advisor
Insights Author
2019 Award
7,262
3,144
Can you share link to your early published work
Click on my picture and you'll find on the profile page (another click) an information button (another click) where you can find my web site with links to my publications.
 
Last edited:
  • #4
Are you a qualified endorser in arXive Quantum physics or general Physics. Can you share any link to your arXive publication.
I couldn't find anything to arXive from your web
 
  • #5
Peter Morgan
Gold Member
244
42
Copy editing comment: On pages 17-18 of the first paper, there are a couple of *** notes to self ***, which perhaps you intend to be there or perhaps not.

Your discussion of Born's rule is very interesting. I like it a lot. I have been working for some time, however, from a different version, which I suggest avoids many of the problems you list, albeit at the cost of adopting a philosophical position you may think too instrumental:
  1. An experiment or a sequence of experiments generate a (typically large or very large) list of raw experimental data that is stored in some kind of computer, which is strongly intersubjective if not objective in that if I display that experimental raw data on my computer screen I will very strongly expect to see exactly the same numbers, schematic diagrams, photographs of the apparatus, et cetera, as anyone else. A journal editor may well insist that they or their referee can read the data and confirm the claims made in a paper that describes the experiment(s).
  2. From this raw experimental data, I can construct summary statistics of any kind at all, using any mathematical operation.
  3. Given a very large list of summary statistics ##S_{ij}##, indexed by the preparation apparatus ##i## and by the measurement apparatus ##j##, I look for a set of density operators ##\hat\rho_i## and measurement operators ##\hat A_j## for which ##S_{ij}=\mathsf{Tr}[\hat\rho_i\hat A_j]##. If we choose the dimensionality of the Hilbert space large enough, we can always solve this set of linear differential equations. There may be constraints, which may be nonlinear, if, for example, a given summary statistic ##S_{ij}## is a mathematical function of other summary statistics or is a higher moment of the same raw experimental data.
  4. At any point we may introduce an idealization that converts the finite amount of data we have into effectively an infinite dataset, as would be the case if we were, say, to extrapolate a cubic spline approximation to a given data set. Some such idealizations will work better than others.
  5. There is also a need for engineering rules. That is, before we build an experiment, certainly after we have characterized its parts and built the whole apparatus, we want to predict what a given summary statistic of the raw experimental data will be. There's the problem and there's the inverse problem. This process, however, works as well as it does for pragmatic reasons and to a pragmatically decided number of standard deviations, not because QM/QFT is true, well-founded, or whatever.
The point here is to start with the raw experimental data and work towards a model, whereas your account largely follows the common convention that the theory comes first (indeed, I discovered in Section 4.2 explicitly follows that convention). I only come to the theory at point 5, which can be whatever works, but finds something of a pragmatic justification in the previous points. If an 8-dimensional Hilbert space always works well enough for a given type of well-controlled quantum optics experiment, then we'll use it, even though we know that in the wild we might have to consider line widths, perhaps very complicated deviations of the laser beams we use away from a coherent state, et cetera.
Perhaps I should note that this only implicitly suggests the idea that there are systems and subsystems. The engineering rules of QM may well find it worthwhile to think in such terms, but (anticipating your second paper) QFT, which I take to be much more a signal analysis formalism than is QM, is grounded in measurements associated with regions of space-time, not in there being objects occupying those regions, so it arguably has no such concept in principle.
With a construction from the raw data of a set of operators ##\hat A_j##, we can use all the usual rules of linear algebra to derive the sample space and probability density associated with it in any given state ##\hat\rho_j##.

Your discussion in Section 3.6 gets three cheers from me. I look forward to your second paper.

Your Section 4.2 rather denies that I can construct QM/QFT in anything like the way I have, when it states "Thus measurement must be grounded in theory". I think I'm much more comfortable to suggest, as experimenters usually do, that there is a dialog between experiment and theory. Another difference arises in that there is no concept of nondestructive or destructive measurements: there is just experimental raw data, and, indeed, there are no systems to be destroyed.
Section 4.3's discussion of "beables" also has a rather simple resolution: if we increase the dimensionality of the Hilbert space enough, any experimental raw data can be presented as a system of commuting measurement operators and diagonal density matrices. That's effectively to introduce ancillas and contextuality, which is not very helpful at all for engineering, but it can be done if we are determined to have beables.

I'm looking forward to discovering what your "thermal interpretation" is, in your second paper. I decided I would comment on the first paper, then perhaps comment separately on the second and third, before I discovered I would have to wait until the second paper for that.
I hope you won't mind me approaching your paper by introducing this contrasting discussion; if you do, say so and I won't comment further. Feel free, as well, not to respond at all to this comment: writing it out is its own reward. I'm very grateful for the impetus you gave me to write it out in response to your writing!
 
  • Like
Likes AlexCaledin and Lynch101
  • #6
Peter Morgan
Gold Member
244
42
I have to react immediately to this, at the head of the second paper: "Quantum physics is used to determine the behavior of materials made of specific molecules under changes of pressure or temperature, their response to external electromagnetic fields (e.g., their color), the production of energy from nuclear reactions, the behavior of transistors in the microchips on which modern computers run, and a lot more." That "determine" is too loud for me! Certainly QM is used to model all those cases, but it seems too much to prejudge whether the world is just as it is or is really determined by a few equations. This barely makes any difference to the instrumental level of physics, however, so let's suppose we don't differ over this.

You may have noticed, but here I'll make explicit, that my discussion of experimental raw data allows summary statistics to be computed using any subset of the data. With experience we will come to know that a particular summary statistic of a particular subset of the data will be more useful or more interesting than some other statistic, or have some other merit. I see this as a plausible response, "it's whatever a physicist says it is, if they can convince other physicists to listen", to the problem you struggle with in Section 2.4 of the second paper, "What is an ensemble?", I think inconclusively.

I find the lack of mention of the experimental raw data in your Section 3 of your second paper quite striking, which perhaps reaches its peak when you say "Statistics is based on the idea of obtaining information about noisy quantities of a system by repeated sampling from a population of independent systems with identical preparation". I would prefer to say something much more in the style of signal analysis: "Summary statistics of experimental raw data are used whenever the dynamics of a representation of the summary statistics is more tractable than the dynamics of the noisy experimental raw data". "Information about noisy quantities of a system" is already too far from the experimental raw data we really have, which is as much or more associated with a given measurement apparatus than with a measured system.

Section 4.1:
"2-point correlations in quantum field theory are effectively classical observables" is only true for 2-point VEVs at space-like separation. At time-like separation, there is an imaginary component.
"it is impossible to repeat measurements" is not true in general. For a massive free quantum field, ##\hat\phi(f_1)## where the test function ##f_1(x)## has support near time ##t_1##, can be equivalent to the operator ##\hat\phi(f_2)## where the test function ##f_2(x)## has support near time ##t_2##, if the fourier transforms ##\widetilde{f_1}(k)## and ##\widetilde{f_2}(k)## have the same projection to the mass-shell, so the later measurement is formally identical to the earlier measurement. This may not be possible for interacting quantum fields, however the time-slice axiom (G in Haag's "Local Quantum Physics") is as much as to insist that it should be possible.

Hurrah for Section 4.2!!! Oh yes, the near-field can and should be discussed!!! Furthermore, quantum non-demolition measurements can allow some aspects of the dynamics to be discussed almost as if the quantum field is classical, as I lay out in my "Classical states, quantum field measurement", arXiv:1709.06711, which was this week recommended for publication by a referee, but with small changes requested that I've now resubmitted.
 
  • #7
Demystifier
Science Advisor
Insights Author
10,815
3,517
@A. Neumaier I have a question on paper II, page 34. What is ##p_{\nu}## in the covariant Schrodinger equation? (I didn't read the whole paper, so you can just pinpoint to the right part of the paper where it is explained.) I mean, if ##p_{0}## is the Hamiltonian, then what is the ##p_{i}## for ##i=1,2,3##?
 
  • #8
A. Neumaier
Science Advisor
Insights Author
2019 Award
7,262
3,144
How does QT in the minimal interpretation describe the state of the solar system?

We only have a single realization of the solar system, which has been prepared once in ancient times.
Hence we cannot apply rules that require a large ensemble of similarly prepared systems.
You should refer to your own interpretation, which you call the "thermal interpretation"! It perfectly describes how quantum theory is applied by usual physicists of all kind.
I know that my interpretation has no problem with the solar system (Subsection 3.4 of Part II). But my thermal interpretation is not the minimal statistical interpretation but one could call it the maximal nonstatistical interpretation since it is a completely deterministic (Subsection 2.1 of Part II) interpretation of everything (Section 5 of Part II).

All probabilistic aspects of quantum mechanics are derived in the same way (Sections 4 and 5 of Part III) as classical probabilities are derived from classical mechanics.

Your own interpretation (I'd somehow rename, because it goes far beyond thermal, i.e., equilibrium systems)
Equilibrium systems are not the only thermal systems - for example, all of fluid mechanics is thermal.
Part II (Introduction) said:
Essential use is made of the fact that everything physicists measure is measured in a thermal environment for which statistical thermodynamics is relevant. This is reflected in the characterizing adjective 'thermal' for the interpretation.
The "solar system", described by macroscopic physics [...] is an emergent phenomenon, and what we observed are grossly coarse-grained macroscopic observables which average at any macroscopic moment of time over zillions of quantum- and thermally fluctuating microscopic degrees of freedom.
But according to the minimal statistical interpretation,
a state is represented by the statistical operator and operationally as an equivalence class of preparation procedures. That's an objective notion of state since a preparation procedure is clearly defined
Thus if one takes your ''clear'' definition of the state at face value, the solar system has in the minimal interpretation no state (unless you are able to come up with an equivalence class of preparation procedures), and is thus outside the scope of minimally interpreted quantum mechanics. Only the microscopic degrees of freedom (which form a huge ensemble) are described by your minimal interpretation. Or not even these - since it would be difficult to come up with a single equivalence class of preparation procedures for these zillions of systems.

Thus I should probably take your definition liberally and should not insist on preparation. Let me assume instead that pp.21-23 of your Lecture Notes on Statistical Physics
Hendrik van Hees said:
The state of a quantum system is described completely by a ray in a Hilbert space [...] The observables of the system are represented by self-adjoint operators. [...] A possible result of a precise measurement of the observable O is necessarily an eigenvalue of the corresponding operator O. [...] the probability to find the value o when measuring the observable O is given by ##P_\psi(o)##. [...] This preparation of a system is possible by performing a precise simultaneous measurement of a complete complete set of observables.
are a faithful reflection of your views of the minimal interpretation. Thus the state of the solar system is supposed to be a ray in some Hilbert space, and the observables of the solar system are supposed to be certain self-adjoint operators on this Hilbert space. I have no idea how to prepare the solar system in the way you require, but let us perhaps assume that God did it. Then the sun qualifies as a quantum system according to your version of the minimal interpretation.

Now let us consider some observable consequences. The effective temperature of the photosphere of the sun is surely an observable since Wikipedia gives for it the value of 5772 K and the value of 5777 K. They don't agree, so there seems to be a probability distribution of possible measurement results. But where is the associated self-adjoint operator that would allow me to apply your minimal interpretation? I cannot see how this squares with your version of the minimal interpretation? One needs to stretch your words quite a lot by adding much nonminimal stuff....
 
Last edited:
  • Like
Likes dextercioby
  • #9
A. Neumaier
Science Advisor
Insights Author
2019 Award
7,262
3,144
@A. Neumaier I have a question on paper II, page 34. What is ##p_{\nu}## in the covariant Schrodinger equation? (I didn't read the whole paper, so you can just pinpoint to the right part of the paper where it is explained.) I mean, if ##p_{0}## is the Hamiltonian, then what is the ##p_{i}## for ##i=1,2,3##?
They are defined on the page before as the generators of space translations. (I missed a factor of ##c## in the second displayed formula of p.34.)
 
  • #10
Demystifier
Science Advisor
Insights Author
10,815
3,517
They are defined on the page before as the generators of space translations. (I missed a factor of ##c## in the second displayed formula of p.34.)
Can you write down the explicit expression for ##p_{\mu}##, e.g. for the free particle without spin?
 
  • #11
vanhees71
Science Advisor
Insights Author
Gold Member
2019 Award
14,715
6,242
I know that my interpretation has no problem with the solar system (Subsection 3.4 of Part II). But my thermal interpretation is not the minimal statistical interpretation but one could call it the maximal nonstatistical interpretation since it is a completely deterministic (Subsection 2.1 of Part II) interpretation of everything (Section 5 of Part II).
Obviously I haven't understood your interpretation then as you seem to mean it, as you start with standard notions of QT in the Hilbert-space formulation and then everything macroscopic is defined also in the usual way of quantum many-body physics. That's at least how I understand your bullet list at the beginning of this very Sect.

For me this IS the minimal interpretation, and the more I think about the foundations, I come to the conclusion that this is all there is to QT. As long as there's no other more comprehensible theory than QT we have to live with this irreducible statistical aspects of nature as we preceive and comprehend it.

All probabilistic aspects of quantum mechanics are derived in the same way (Sections 4 and 5 of Part III) as classical probabilities are derived from classical mechanics.
Which just underlines, how I understood it. So I don't see where my mistake should be.

Equilibrium systems are not the only thermal systems - for example, all of fluid mechanics is thermal.
Ok, maybe that's due to my embedding in the heavy-ion community, where "thermal" means "in or at least close to (local) thermal equibrium". Fluid mechanics in this sense is indeed thermal since it's precisely about systems that can be described as close to local thermal equilibrium.
But according to the minimal statistical interpretation,

Thus if one takes your ''clear'' definition of the state at face value, the solar system has in the minimal interpretation no state (unless you are able to come up with an equivalence class of preparation procedures), and is thus outside the scope of minimally interpreted quantum mechanics. Only the microscopic degrees of freedom (which form a huge ensemble) are described by your minimal interpretation. Or not even these - since it would be difficult to come up with a single equivalence class of preparation procedures for these zillions of systems.

Thus I should probably take your definition liberally and should not insist on preparation. Let me assume instead that pp.21-23 of your Lecture Notes on Statistical Physics

are a faithful reflection of your views of the minimal interpretation. Thus the state of the solar system is supposed to be a ray in some Hilbert space, and the observables of the solar system are supposed to be certain self-adjoint operators on this Hilbert space. I have no idea how to prepare the solar system in the way you require, but let us perhaps assume that God did it. Then the sun qualifies as a quantum system according to your version of the minimal interpretation.
The complete state of the solar system is undescribable. That's why we use the corresponding statistical operator appropriate for the relevant macroscopic observables as, e.g., the classical positions and momenta of the planets and moons etc. as described by (post-)Newtonian celestial mechanics.
Now let us consider some observable consequences. The effective temperature of the photosphere of the sun is surely an observable since Wikipedia gives for it the value of 5772 K and the value of 5777 K. Thus there seems to be a probability distribution of possible measurement results. But where is the associated self-adjoint operator that would allow me to apply your minimal interpretation? I cannot see how this squares with your version of the minimal interpretation? One needs to stretch your words quite a lot by adding much nonminimal stuff....
Well, this is a very nice example. Of course, it's impossible to describe the Sun in all microscopic detail. Rather it's a good assumption to use a quantum-statistical description of the Sun in thermal equilibrium with the radiation pressure counterabalancing the gravitational force.
 
  • #12
A. Neumaier
Science Advisor
Insights Author
2019 Award
7,262
3,144
at the cost of adopting a philosophical position you may think too instrumental:
I view any interpretation as inadequate that cannot account for the meaning of quantum physics at a time before any life existed.
The universe exists for billions of years - in a sense to be explained by any interpretation that allows for quantum cosmology.
But experimental data exist for a few thousands of years only.

"Thus measurement must be grounded in theory". I think I'm much more comfortable to suggest, as experimenters usually do, that there is a dialog between experiment and theory.
Well, how can you sensibly assert that a silver speck at a screen in a Stern-Gerlach experiment is the measurement of a particle with spin up, without having first a theory of how such a particle behaves when passing through a magnetic field?

if we increase the dimensionality of the Hilbert space enough, any experimental raw data can be presented as a system of commuting measurement operators and diagonal density matrices.
Well, already the Hilbert space of a harmonic oscillator has countably infinite dimension, Fock space too. Thus unless you want to work with nonseparable Hilbert spaces, you cannot increase the dimensiononality....
"Quantum physics is used to determine the behavior of materials made of specific molecules under changes of pressure or temperature, their response to external electromagnetic fields (e.g., their color), the production of energy from nuclear reactions, the behavior of transistors in the microchips on which modern computers run, and a lot more." That "determine" is too loud for me!
Determine - this is the goal of ab initio quantum chemistry, realized to a large extent. One can determine ab initio the color of gold, the melting point of mercury, the equation of state of small molecules, etc., and only computer power seems to limit the extent and accuracy with which this can be done.
"Information about noisy quantities of a system" is already too far from the experimental raw data we really have
The data produced in scattering experiments or experimental checks of Bell inequalites are heaps of noisy statistical raw data, from which a small number of reproducible (and hence scientifically relevant) data (cross sections, coincidence probabilities) are
produced.
so the later measurement is formally identical to the earlier measurement.
But these measurement measure the field at a different time, hence measure different field operators.
 
  • #13
A. Neumaier
Science Advisor
Insights Author
2019 Award
7,262
3,144
Can you write down the explicit expression for ##p_{\mu}##, e.g. for the free particle without spin?
In the momentum representation of the Fock space, it is just multiplication of ##\psi(p_1,\ldots,\p_N)## by ##p_1+\ldots+p_N##.
 
  • #14
Demystifier
Science Advisor
Insights Author
10,815
3,517
In the momentum representation of the Fock space, it is just multiplication of ##\psi(p_1,\ldots,p_N)## by ##p_1+\ldots+p_N##.
Isn't this a bit circular? To define the momentum operator ##p_{\mu}##, you must first know the momentum representation, as you defined above. But to know the momentum representation, you must first know what is the momentum operator.
 
  • #15
A. Neumaier
Science Advisor
Insights Author
2019 Award
7,262
3,144
Obviously I haven't understood your interpretation then as you seem to mean it, as you start with standard notions of QT in the Hilbert-space formulation and then everything macroscopic is defined also in the usual way of quantum many-body physics. That's at least how I understand your bullet list at the beginning of this very Sect.
Well, Section 5 of Part III is the summary of what emerges from the whole paper. Of course, my interpretation is consistent with the shut-up-and-calculate part (the formal core given in Part I) and with everything done in quantum many-body physics!

But it is derived from a completely deterministic dynamics without assuming Born's rule (which has only a restricted domain of validity).

For me this IS the minimal interpretation, and the more I think about the foundations, I come to the conclusion that this is all there is to QT. As long as there's no other more comprehensible theory than QT we have to live with this irreducible statistical aspects of nature as we preceive and comprehend it.
But what you wrote as foundation on p.21-22 of your statistical physics lecture note is far from what you now say the minimal interpretation is. It seems to me that in fact you really adhere to my thermal interpretation while only paying lipservice to your own formulation of the minimal interpretation.

All probabilistic aspects of quantum mechanics are derived in the same way (Sections 4 and 5 of Part III) as classical probabilities are derived from classical mechanics.
Which just underlines, how I understood it. So I don't see where my mistake should be.
If you really understand it in this way, why then do you postulate Born's rule and probabilities in the very foundations?

Ok, maybe that's due to my embedding in the heavy-ion community, where "thermal" means "in or at least close to (local) thermal equilibrium". Fluid mechanics in this sense is indeed thermal since it's precisely about systems that can be described as close to local thermal equilibrium.
And all measurements are done by instruments in local thermal equilibrium....

The complete state of the solar system is undescribable. That's why we use the corresponding statistical operator appropriate for the relevant macroscopic observables as, e.g., the classical positions and momenta of the planets and moons etc. as described by (post-)Newtonian celestial mechanics.
I have never seen a statistical operator for (post-)Newtonian celestial mechanics, which to me is pure classical mechanics. Could you please point me to a source?

Well, this is a very nice example. Of course, it's impossible to describe the Sun in all microscopic detail. Rather it's a good assumption to use a quantum-statistical description of the Sun in thermal equilibrium with the radiation pressure counterabalancing the gravitational force.
Even there, where is the self-adjoint operator associated to temperature that would allow me to apply the minimal interpretation and get a probability distribution for the measured temperatures?
 
Last edited:
  • #16
A. Neumaier
Science Advisor
Insights Author
2019 Award
7,262
3,144
Isn't this a bit circular? To define the momentum operator ##p_{\mu}##, you must first know the momentum representation, as you defined above. But to know the momentum representation, you must first know what is the momentum operator.
No. One starts with a definition of the Fock space in terms of single-particle momenta, and then defines the generators of translations as operators on Fock space. This is the usual building-up procedure.

If you prefer to work with the position representation in Fock space, you can define the generators of spatial translations instead as the sum of the 1-particle momentum operators. This is the way it is usually done in statistical mechanics, e.g., in Linda Reichl's book.
 
Last edited:
  • #17
DarMM
Science Advisor
Gold Member
2,370
1,397
@A. Neumaier , I have finished the papers and I am now beginning my closer second read through.
 
  • Like
Likes bhobba and Demystifier
  • #18
vanhees71
Science Advisor
Insights Author
Gold Member
2019 Award
14,715
6,242
Well, Section 5 of Part III is the summary of what emerges from the whole paper. Of course, my interpretation is consistent with the shut-up-and-calculate part (the formal core given in Part I) and with everything done in quantum many-body physics!

But it is derived from a completely deterministic dynamics without assuming Born's rule (which has only a restricted domain of validity).


But what you wrote as foundation on p.21-22 of your statistical physics lecture note is far from what you now say the minimal interpretation is. It seems to me that in fact you really adhere to my thermal interpretation while only paying lipservice to your own formulation of the minimal interpretation.


If you really understand it in this way, why then do you postulate Born's rule and probabilities in the very foundations?


And all measurements are done by instruments in local thermal equilibrium....


I have never seen a statistical operator for (post-)Newtonian celestial mechanics, which to me is pure classical mechanics. Could you please point me to a source?


Even there, where is the self-adjoint operator associated to temperature that would allow me to apply the minimal interpretation and get a probability distribution for the measured temperatures?
Well, perhaps I'm too biased with the traditional interpretation, but precisely from the quoted Sec. 5 I came to the conclusion that your thermal interpretation is nothing else than what every physicist using QT (however you call his/her interpretation) understands under the formalism: The formalism predicts probabilistic properties of measurement outcomes when measuring a real object with a real measurement device. Since the papers consist of a lot of text with sparse use of formulae maybe it's not precisely clear to me what you really mean, because for instance I didn't get that you have a deterministic view point as the foundation.

So let me summarize, how I understand your concept. The only difference between standard QM1 textbook treatments and your starting point is that instead of using the special case of pure states you start right away with the general case of states ("mixed states"). This is what I always understood as the state, because also for pure states it's way more convenient to use the projection operator to represent the state in terms of a stat. op. than as a ray in Hilbert space. So after the introductory heuristics has settled, for me (and as far as I see also for you according to your 3 papers) the undisputable mathematical formalism is as follows (everywhere, where you write "hermitian", I think one should read "self-adjoint" to be on the safe side, but that's a formality):

Kinematical part:

(1) There's a (separable) Hilbert space associated with the system to be described within QT. Time is a real parameter.
(2) The state of the is represented by a self-adjoint positive semidefinite operator with trace 1, ##\hat{\rho}(t)##.
(3) Any observable is represented by a self-adjoint operator.
(4) Possible outcomes of precise measurements of an observable are the spectral values of the corresponding self-adjoint operator.
(5) The expectation values of any observable ##A##, represented by the self-adjoint operator ##\hat{A}(t)## is given by ##\langle A(t) \rangle=\mathrm{Tr}[\hat{\rho}(t) \hat{A}(t)]##.

BTW. I'd call (5) Born's rule, while you seem to restrict the notion to apply only to the special case and the proabilities (or probability distributions) for pure states, but that's semantics.

Dynamical part (for not explicitly time-dependent observables as in your papers):

(6) For each system the dynamics is governed by an observable ##H##, the Hamiltonian of the system.
(7) For any observable operator ##\hat{A}(t)## the operator describe the time derivative of this observable is given by
$$\mathring{\hat{A}}(t)=\frac{1}{\mathrm{i} \hbar} [\hat{A}(t),\hat{H}(t)].$$

Note that in (7) ##\mathring{\hat{A}}## is not the time derivative of ##\hat{A}## (except in the Heisenberg picture).

The rest should follow from these axioms, among other things also Ehrenfest's theorem, which seems to be the key to your interpretation, and maybe that's the point, I don't understand correctly. Obviously you define a Lie algebra implying Lie derivatives on an abstract algebra of observables and then reconstruct the above postulates from them. On the other hand at least your notation suggests that what's meant as observables (or in Bell's language "beables") are the expectation values in the above QT sense.

Do you have a paper with more math and less text that shows the derivation from the Lie algebra to the Hilbert space formulation, so that I can follow the logic better?

Another, maybe much more difficult, question is, whether one can use these concepts to teach QM 1 from scratch, i.e., can you start by some heueristic intuitive physical arguments to generalize the Lie-algebra approach of classical mechanics in terms of the usual Poisson brackets of classical mechanics? Maybe that would be an alternative approach to QM which avoids all the quibbles with starting with pure states and then only finally arrive at the general case of statistical operators as description of quantum states?
 
  • Like
Likes Spinnor and PeroK
  • #19
A. Neumaier
Science Advisor
Insights Author
2019 Award
7,262
3,144
Well, perhaps I'm too biased with the traditional interpretation, but precisely from the quoted Sec. 5 I came to the conclusion that your thermal interpretation is nothing else than what every physicist using QT (however you call his/her interpretation) understands under the formalism:
Only in the sense stated in Part III on p.51, in the paragraph directly after the list of bullets:
Part III said:
The thermal interpretation is inspired by what physicists actually do rather than what they say. It is therefore the interpretation that people work with in the applications [...], rather than only paying lipservice to it.
The formalism predicts probabilistic properties of measurement outcomes when measuring a real object with a real measurement device.
But the formalism of the thermal interpretation is completely deterministic, with a conservative dynamics for the collection of all q-expectations. Itproduces statistical results only in its coarse-grained approximations where the dynamics is (as in all practical applications) reduced to a collection of relevant q-expectations.
I didn't get that you have a deterministic view point as the foundation.
This is stated explicitly at the top of Subsection 3.3 of Part II, where the discussion of the statistical aspects begins.
(everywhere, where you write "hermitian", I think one should read "self-adjoint" to be on the safe side, but that's a formality):
n
No. q-expectations are defined and real for any Hermitian linear operator. Self-adjointness is needed only for the spectral theorem, i.e., when referring to the spectrum or spectral projections.
(1) There's a (separable) Hilbert space associated with the system to be described within QT. Time is a real parameter.
Separability is nowhere needed. In fact, interacting quantum field theories typically need nonseparable Hilbert spaces. This can be most easily seen for a simple model, the relativistic massless scalar field in 1+1 dimensions.
(3) Any observable is represented by a self-adjoint operator.
In the thermal interpretation, any observable (though I avoid this word) is represented by a function of q-expectations.
(4) Possible outcomes of precise measurements of an observable are the spectral values of the corresponding self-adjoint operator.
In the thermal interpretation, this holds only for exact measurements of observables defined by self-adjoint operators, i.e., those where the theoretical uncertainty vanishes.
(5) The expectation values of any observable ##A##, represented by the self-adjoint operator ##\hat{A}(t)## is given by ##\langle A(t) \rangle=\mathrm{Tr}[\hat{\rho}(t) \hat{A}(t)]##.

BTW. I'd call (5) Born's rule, while you seem to restrict the notion to apply only to the special case and the probabilities (or probability distributions) for pure states, but that's semantics.
In Part I, five different forms of Born's rule are distinguished. The universal from looks like (5), but is explicitly related to a mean of measurement results over a large sample. For q-expectations without this measurement interpretation, (5) implies no connection to reality, hence is not an interpretation statement but a formal definition of what to call a q-expectation. The thermal interpretation of these is as beables,
that can be approximated by measurement results within the limits given by the uncertainty, as defined in eq. (15) of Section 2.4 of Part II. On the other hand, Born's rule in the minimal statistical interpretation requires (by its derivation from your postulates in your statistical mechanics lecture notes) that the q-expectation is a mean of a large number of actual measurement results.
Ehrenfest's theorem, which seems to be the key to your interpretation, and maybe that's the point, I don't understand correctly. Obviously you define a Lie algebra implying Lie derivatives on an abstract algebra of observables and then reconstruct the above postulates from them.
Section 2.1 has a large ratio of formulas to text, and explains the Ehrenfest picture in full detail. The Ehrenfest dynamics for expectations is clearly deterministic.
On the other hand at least your notation suggests that what's meant as observables (or in Bell's language "beables") are the expectation values in the above QT sense.
The q-expectations in the formal sense, yes, but without the interpretation as sample means.
Do you have a paper with more math and less text that shows the derivation from the Lie algebra to the Hilbert space formulation, so that I can follow the logic better? [...]
can you start by some heuristic intuitive physical arguments to generalize the Lie-algebra approach of classical mechanics in terms of the usual Poisson brackets of classical mechanics?
These are questions quite different from the ones an interpretation of quantum mechanics has to address. The answer to both questions is yes. This (among other things) will be discussed in Part IV, which answers the critique from Section 5.2 of Part I and gives a coherent synthesis. It exists in draft form but is not yet ready for making it public. Please wait a few more weeks....
 
Last edited:
  • #20
vanhees71
Science Advisor
Insights Author
Gold Member
2019 Award
14,715
6,242
No. q-expectations are defined and real for any Hermitian linear operator. Self-adjointness is needed only for the spectral theorem, i.e., when referring to the spectrum or spectral projections.
Well, that's the important point, why I think it should be "self-adjoint". If the spectral theorem is not valid, there's no sensible probability interpretation, at least not in the usual sense, and not considering the stricter condition of self-adjointness and sloppily dealing with Hermitean operators as if they were self-adjoint leads to misconceptions and misunderstandings (e.g., in the apparently simple infinite-potential-well model when momentum instead of energy eigenvectors are discussed although there's no self-adjoint momentum operator defined).

Separability is nowhere needed. In fact, interacting quantum field theories typically need nonseparable Hilbert spaces. This can be most easily seen for a simple model, the relativistic massless scalar field in 1+1 dimesnions.
Intersting, but doesn't one then run into the trouble with Haag's theorem, which however is of little practical relevance since it only occurs if not regularizing the model by introducing a finite quantization volume. I'm well aware of the fact that from a strictly mathematical point of view there's no proof for the existence of realistic QFTs. For (resummed) perturbative physicists' prescriptions, it's however enough to deal with the models in this non-strict way.

In the thermal interpretation, any observable (thoug I avoid this word) is represented by a function of q-expectations.
But then the question is, how this representation of expectation values is defined, and this has to be also given operationally. At least this point has been clarified a lot in recent years concerning the standard interpretation of QT with generalizing the idealized von Neumann measurements to the description of real-world experiments in terms of the POVM formalism.

For the same reason, I've still no clue what's behind the q-bism interpretation. They claim that the probabilities of QT have meaning for single realizations of an experiment but never give an operational definition of what's measured if nothing is averaged (neither in Gibbs's sense nor in the coarse-graining sense discussed above).

In the thermal interpretation, this holds only for exact measurements of observables defined by self-adjoint operators, i.e., those where the theoretical uncertainty vanishes.
Yes, indeed. That's also the case for traditionally minimally interpreted QT, and it's the starting point for understanding the theory as a physical theory to begin with. First one has to understand the most simple cases to understand the meaning of an interpretation.

The Stern-Gerlach experiment is a very good example for that. It can be treated analytically and exactly for Gaussian wave packets with using the approximate Hamiltonian
$$\hat{H}=\frac{\hat{\vec{p}}^2}{2m} + g_s \mu_B (\vec{B}_0 + \beta z )\hat{s}_z$$
which leads via the dynamics of the Ag atom to an strictly space-##s_z##-entangled state which lets you filter out the definite ##s_z=\pm \hbar/2##-states.

This is however approximate since the magnetic field close to the beam in fact is approximated by $$\vec{B}=\vec{B}_0 + \beta (z \vec{e}_z-y vec{e}_y)$$, and so far I could treat the "perturbation"
$$\hat{V}=-g_s \mu_B \beta y \hat{s}_y$$
only perturbatively, which leads of course to (small) mixing of the "wrong" ##s_z##-states into the regions which are however still approximately pure ##s_z## states.

I don't see, what's lacking with the standard minimal interpretation in this case since it predicts the outcome of measurements, and to do the experiment properly you need some amount of Ag atoms in the beams to accumulate "enough statistics" to be able to see the splitting at all.

So, how would the analogous calculation work with the thermal representation. Since the Hamiltonian is only maximally quadratic in the observables, this must be a pretty simple thing for the thermal interpretation since the Ehrenfest equations of motion for the expectation values are of course just the classical equations of motion for the classical Hamiltonian motion of an uncharged particle with magnetic moment in an as simple as possible approximate magnetic field applicable to the fine beams prepared in the typical textbook experiment.

In Part I, five different forms of Born's rule are distinguished. The universal from looks like (5), but is explicitly related to a mean of measurement results over a large sample. For q-expectations wihout this measurement interpretation, (5) implies no connection to reality,
hence is not an interpretation statement but a definition of what to call a q-expectation. The thermal interpretation of these is as beables,
that can be approximated by measurement results within the limits given by the uncertainty, as defined in eq. (15) of Section 2.4 of Part II.
Yes, 2.4 is precisely why I was misunderstanding your interpretation as being in fact the usual (minimal) interpretation, because you argue with classical phase-space distributions. For me that's already a coarse-grained description, approximating the one-body Wigner functions of many-body systems via the gradient expansion of the corresponding Kadanoff-Baym equations. This is the formal description of an "ensemble average" in the sense that one averages over the mircocopic fluctuations by just "blurring" the observation to the accuracy/resolution of typical macroscopic time and space scales, and thus "averaging" over all fluctuations at the microscopic space-time scales. Of course you don't need to take "ensemble average" in Gibbs's sense literally here. Otherwise we'd never ever have observed classical behavior of single macroscopic (many-body) systems to begin with.
Section 2.1 has a large ratio of formulas to text, and explains the Ehrenfest picture in full detail. The Ehrenfest dynamics for expectations is clearly deerministic.
Yes, and obviously I misinterpreted this section in thinking that, despite the somewhat unusual notation, you just describe usual quantum-theoretical averages. There are however no details given, how one deals with the fact that of course for functions of averages in genral you have ##f(\langle A \rangle) \neq \langle f(A) \rangle##. Maybe that's the reason, why I didn't understand the fact that you consider this Lie-algebra formalism for expectation values as the fundamental set postulates, because I always thought you'd need the quantum formalism to define expectation values to begin with. For me expectation values are given by the above quoted trace formula, and as you say, that's not different in your paper I.
The q-expectations in the formal sense, yes, but wihout the interpretation as sample means.

Yes, yes. This will be in Part IV, which answers the remaining critique from Part I and gives a coherent synthesis. It exists in draft form but is not yet ready for making it public. Please wait a few more weeks....
Great! I'm looking forward to it.
 
  • #21
A. Neumaier
Science Advisor
Insights Author
2019 Award
7,262
3,144
Well, that's the important point, why I think it should be "self-adjoint". If the spectral theorem is not valid, there's no sensible probability interpretation
Yes, but the expectation can still be defined via the trace. That's why the probability interpretation is secondary. In the thermal interpretation it is not assumed but (under additional assumptions) derived.
Interesting, but doesn't one then run into the trouble with Haag's theorem
It is needed in rigorous interactive relativistic QFT precisely because of Haag's theorem. One cannot use a Fock space (which is separable) but needs a nonseparable Hilbert space with superselection rules for unitarily inequivalent representations of the field algebra. I recommend studying the massless free scalar field in 1+1 dimensions, which is an exactly solvable toy example where the problem of superselection rules appears already in the free case.
But then the question is, how this representation of expectation values is defined, and this has to be also given operationally.
The definition of q-expectation is given by eq. (14) of Part II, and involves neither probabilities nor eigenvalues nor the spectral theorem. It works even for nonhermitian operators such as creation and annihilation operators, and is meaningful there. An operational meaning has to be given only for those q-expectations which actually correspond in a simple way to measurements (which is a small number of all possible ones). The operational meaning is therefore dependent on what you measure and how you do it. It is based on the general uncertainty principle (GUP) from page 15 of Part II, and made more concrete by the measurement principle (MP) on p.6 of Part III. For equilibrium thermodynamics, we may use the principle introduced by H.B. Callen in his famous textbook ''Thermodynamics and an introduction to thermostatistics'',
H.B. Callen said:
Operationally, a system is in an equilibrium state if its properties are consistently described by thermodynamic theory.
This implies that q-expectations of 1-particle operators and energy in QFT are measured by the established techniques of equilibrium thermodynamics. This is fully operational, using single measurements only, without any need to check whether the operator is self-adjoint (which is difficult to verify), and without ever having mentioned probabilities. Callen formulated the principle for thermodynamics, but its analogue is valid everywhere in scientific modeling:
  • Callen's principle: Operationally, a system is in a given state if its properties are consistently described by the theory for this state.
This together with (GUP) and (MP) is enough to find out in each single case how to measure a q-expectation.

q-probabilities are special cases of q-expectations, namely those of self-adjoint Hermitian operators with spectrum in ##[0.1]##. The weak law of large numbers, discussed in Subsection 3.3 of Part II applies to probabilities and gives operational recipes for measuring them in certain cases, see Subsection 3.5 of Part II; in particular, Born's probabilistic rule follows for ideal binary measurements; see Subsection 3.4 of Part II.
Thus, in the thermal interpretation, Born's rule is not assumed but derived (where appropriate)!
At least this point has been clarified a lot in recent years concerning the standard interpretation of QT with generalizing the idealized von Neumann measurements to the description of real-world experiments in terms of the POVM formalism.
See Subsection 2.5 of Part III for the ease with which POVMs arise in the thermal interpretation. Instead, in an approach based on statistical mechanics lecture notes (where POVM do not even appear, though they account for many more measurements than the ideal ones you formalize with your version of Born's rule!) one would need for the justification of POVMs as measurements a very unnatural exptension of the physical Hilbert space by adding a fictitious ancilla degrees of freedom.
The Stern-Gerlach experiment is a very good example for that. [...] So, how would the analogous calculation work with the thermal representation.
Maybe I'll discuss this in Part IV.
that's already a coarse-grained description, approximating the one-body Wigner functions of many-body systems via the gradient expansion of the corresponding Kadanoff-Baym equations.
For how the thermal interpretation views coarse-grained descriptions see Section 4.2 of Part III. One takes a subspace of q-expectations - in case of the Kadanoff-Baym equations the subspace of field expectations and pair correlation functions - and derives an approximate closed dynamics for these.
The approximation is in the equations of motion, not in the meaning of the expectations! Nowhere in the derivation of the Kadanoff-Baym equations is a need to interpret the q-expectations as ensemble averages. [Over which ensemble?? People successfully apply the equations to the early universe, of which there is only one observable case, not an ensemble; the ensembles are pure imaginations without any operational content, as discussed in Subsection 2.4 of Part II!]
Neither do the derivations (see, e.g., the slides of your 2002 lecture) invoke any observations or any average over fluctuations. Instead just a gradient expansion! Everything you do in your slides is shut-up-and-calculate - no probabilities, no observations, no measurements, no operational recipes for observing the bilocal field correlations. You just worked out approximate formulas for the dynamics of some q-expectations, without ever entering the interpretation of these q-expectations! And you end up with something where the q-expectations have a semiclassical, nonstatistical interpretation - as the thermal interpretation requires, without invoking any of the ghosts of the past that you invoked!
Yes, and obviously I misinterpreted this section in thinking that, despite the somewhat unusual notation, you just describe usual quantum-theoretical averages.
I describe indeed usual q-expectations, except that I don't interpret them probabilistically as averages, since this is unnecessary baggage that caused nearly a century of unsettled dispute. This makes all the difference:

Probability is not an irreducible input to quantum mechanics but a consequence of not being able to make observations at arbitrary scales: one needs to introduce approximations, and in many (but not all) cases, these approximation introduce stochastic aspects. Just as the classical dynamics of a bistable system coupled to a bath becomes in some approximation a binary stochastic hopping process. That this analogy is not far-fetched is discussed in Section 5.1 of Part III.
 
Last edited:
  • Like
Likes dextercioby
  • #22
736
34
@A. Neumaier: Just a dumb question: in your references, what does the number in the square brackets at the end of a citation mean (for example:
[1] A.E. Allahverdyan, R. Balian and T.M. Nieuwenhuizen, Understanding quantum measurement from the solution of dynamical models, Physics Reports 525 (2013), 1–166. https://arxiv.org/abs/1107.2138 [16])? So what does this [16] mean? Maybe you explain it somewhere, but I have not found such an explanation so far.
 
  • #23
A. Neumaier
Science Advisor
Insights Author
2019 Award
7,262
3,144
in your references, what does the number in the square brackets at the end of a citation mean.
It refers to the page(s) where the reference is cited.
 
  • #24
736
34
It refers to the page(s) where the reference is cited.
Thank you very much!
 
  • #25
736
34
@A. Neumaier: A quote from your work: " When a particle has been prepared in an ion trap (and hence is there with certainty), Born’s rule implies a tiny but positive probability that at an arbitrarily short time afterwards it is detected a light year away"

I guess this is only true if one assumes nonrelativistic equation of motion? (The argument about Newton-Wigner does not seem clear).
 

Related Threads on The thermal interpretation of quantum physics

Replies
16
Views
4K
Replies
21
Views
9K
Replies
1
Views
3K
  • Last Post
7
Replies
154
Views
20K
  • Last Post
Replies
15
Views
3K
Replies
5
Views
1K
  • Last Post
Replies
0
Views
1K
  • Poll
  • Last Post
6
Replies
141
Views
20K
Replies
3
Views
826
Replies
22
Views
2K
Top