Short question about diffeomorphism invariance

  • Thread starter nrqed
  • Start date

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
There is a clear write up in Wuthrich's thesis, p49. http://philosophy.ucsd.edu/faculty/wuthrich/pub/WuthrichChristianPhD2006Final.pdf [Broken]
Thanks for finding that source Atyy. I've been busy with other stuff and haven't looked at it.
I did find confirmation in Sean Carroll's thing of what you see in Rovelli. Rovelli's certainly not presenting anything unusual.
In Carroll "Intro to GR" (which seems pitched to undergraduates, style-wise) on page 138 it says:
==quote==
Let’s put some of these ideas into the context of general relativity. You will often hear it proclaimed that GR is a “diffeomorphism invariant” theory. What this means is that, if the universe is represented by a manifold M with metric gµν and matter fields ψ, and φ : M → M is a diffeomorphism, then the sets (M, gµν , ψ) and
(M, φ* gµν , φ* ψ) represent the same physical situation.

==endquote==

That is, you can totally moosh and morph and it is still the same physical reality. You aren't limited to an "isometric" diffeo.

In other words physical realities correspond to equivalence classes under diffeomorphism of metrics. They don't correspond to single metrics but to diffeomorphism classes of metrics. All those that you can make by morphing the one you started with.

So there is no rubber sheet. The equiv class carries the more abstract idea of a geometry-without-an-underlying-manifold.

Note that Carroll gives the mathematical truth, but then reassures his undergrads with some condescending pablum which basically says "don't worry your little heads about this". He gives the impression that diffeo is "just a highbrow change of coordinates" the basic message is "I gave you the equation, we aren't going to use it, so no need to think much about it."

Carroll is carrying on the "half century of confusion" that MTW complained about---diffeomorphism invariance going incognito, anonymous (as MTW put it) under the mask of "just like a change of coordinates".

Carroll is a master of comfortable communication---gift of the gab. The important thing is he gives the equation and says the two represent the same physical situation. His spin after that can be ignored.

Above just my humble view of course :biggrin:

“Dadurch verlieren Zeit & Raum den letzter Rest von physikalischer Realität.”
 
Last edited by a moderator:

atyy

Science Advisor
13,501
1,610
Above just my humble view of course :biggrin:

“Dadurch verlieren Zeit & Raum den letzter Rest von physikalischer Realität.”
Yes, I believe Einstein was confused.
 

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
...What I dislike about Rovelli's work is that he makes simple things so obscure so that he can say LQG is really addressing deep conceptual issues that other quantum gravity people don't.
Atyy try thinking it from the other direction! According to what you say here, you read Hawking Ellis, and Wald, and Carroll and you didn't get the idea! You thought Rovelli was saying something different! Only Rovelli made the idea of diffeo invariance clear enough to get through to you. So you thought he was saying different:

But since GR models physical reality by manifold and metric, only manifolds that are isometric describe the same physical reality (eg. Hawking and Ellis, p56). In the most common sense of diffeomorphism (Hawking and Ellis, Wald, Carroll, but not Rovelli's), two manifolds related by a diffeomorphism do not necessarily represent the same physical spacetime - only manifolds related by isometric diffeomorphisms .
The objective evidence would indicate that you got a wrong understanding from reading H&E and Wald and Carroll. You thought only isometries gave the same physical situation. That means that H&E Wald Carroll expository writing was obscure. It left you with a radically erroneous conception.

Actually Rovelli is a pretty good writer. I think the main reason some people (not necessarily you) find him difficult to read might be that they start with an attitude of disbelief and resentment. If you obstinately doubt everything you read it will make it more difficult to "get it."

For example people coming from just being overwhelmed by string mathematics, much of which depends on postulating a prior geometry---depends on a set geometric background---will naturally be reluctant to accept the idea that nature is not that way.

The disapprobrium seems in part like a classic case of punishing the messenger. The bringer of cognitive dissonance.
 
Last edited:

MTd2

Gold Member
2,028
25
Let me see from a spin foam point of view. Minimal size means the term locality is restricted the link be nodes, so a ruler can measure a minimum size between nodes. But this size can be arbitrarily small if inferred in other reference frame, so the set node + link + node is just another example of extended object. Right?
I would really want an answer about this :eek:
 

atyy

Science Advisor
13,501
1,610
In other words physical realities correspond to equivalence classes under diffeomorphism of metrics. They don't correspond to single metrics but to diffeomorphism classes of metrics. All those that you can make by morphing the one you started with.
This is also true in special relativity.
 

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
But this size can be arbitrarily small if inferred in other reference frame,
I would really want an answer about this :eek:
What does it mean for a length (or an area) to be arbitrarily small if you cannot measure that small?

I must have cited the same 2003 Rovelli paper dozens and dozens of times here at this forum.

The title is something like "Reconciling discrete area spectrum with Lorentz invariance."

You have some physical object that defines an area and you observe the area from a stationary and a moving frame.
The expectation value of the area operator can be made arbitrarily small even though the spectrum (the possible results of any particular measurement) is discrete and has a smallest possible value.

so a ruler can measure a minimum size between nodes. But this size can be arbitrarily small if inferred in other reference frame
No this is not true.
Take the case of area. No matter how fast the other reference frame is going it cannot measure and get a positive area for an answer that is smaller than the minimum eigenvalue of the operator. Sometimes it will get the answer ZERO (so the average or expectation value can be quite small) but it will never measure a positive area smaller than the minimum.

This is the kind of thing you encounter in a quantum theory of geometry. A quantum theory is about measurement and observation. It is not about "what is there" at micro scale. It is about what we can measure and what we can SAY about the micro world. It is about the limitations on the information which we can get.

So I would say simply that you are trying to reason about the length of a link, the separation between two nodes. We don't do that. The question is not well posed.
 
Last edited:

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
This is also true in special relativity.
? I don't understand. Could you be joking? SR works in Minkowski space which is itself not invariant under diffeomorphism. Diffeos don't preserve flatness. You must have some obscure idea in mind. Please explain.
 

atyy

Science Advisor
13,501
1,610
? I don't understand. Could you be joking? SR works in Minkowski space which is itself not invariant under diffeomorphism. Diffeos don't preserve flatness. You must have some obscure idea in mind. Please explain.
I am not joking. If you move the metric with a diffeomorphism, you will preserve flatness.

So SR is invariant under diffeomorphisms that move manifold, metric and matter. Diffeomorphisms that move the metric are isometries.

SR is not invariant under diffeomorphisms that move manifold and matter without moving the metric. The point is that GR is invariant under such diffeomorphisms, because the metric has become matter, and you move it automatically once you move matter ("dynamical fields").
 

MTd2

Gold Member
2,028
25
Sometimes it will get the answer ZERO (so the average or expectation value can be quite small) but it will never measure a positive area smaller than the minimum.
I agree with you. The minimum size can be made as small as possible due Lorentz contraction. I don't understand what you understood from me.

BTW, what is a well posed question?
 

MTd2

Gold Member
2,028
25
BTW, That thing about E8XE8 inside SO(3,1) doesn`t make sense because that E8 is the dynkin diagram of the group of symmetries of the equation of the ALE space whose the hypersurface correspond to the group of rotations of the icosahedron, which is a finite subgroup of SU(2). So, we would be talking about an internal space of a node, at best. So, forget about this idea.
 

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
The minimum size can be made as small as possible due Lorentz contraction.
No, that is precisely the point. The spectrum of the area observable cannot be changed by Lorentz contraction. "Minimum size" can only mean the smallest positive eigenvalue. This does not change.

It does not make any sense to refer to an expectation value as a minimum size. There is no minimum positive expectation value.

This is explained simply and clearly in the 2003 Rovelli paper. You would save yourself some time if you looked at it. It is short. I gave a paraphrase of the title earlier, but maybe you were unable to find it. Here is the link and abstract. (It was published 2003 but the arxiv abstract is from the previous year),

http://arxiv.org/abs/gr-qc/0205108
Reconcile Planck-scale discreteness and the Lorentz-Fitzgerald contraction
Carlo Rovelli, Simone Speziale
12 pages, 3 figures Physical Review D67 (2003) 064019
(Submitted on 25 May 2002)
"A Planck-scale minimal observable length appears in many approaches to quantum gravity. It is sometimes argued that this minimal length might conflict with Lorentz invariance, because a boosted observer could see the minimal length further Lorentz contracted. We show that this is not the case within loop quantum gravity. In loop quantum gravity the minimal length (more precisely, minimal area) does not appear as a fixed property of geometry, but rather as the minimal (nonzero) eigenvalue of a quantum observable. The boosted observer can see the same observable spectrum, with the same minimal area. What changes continuously in the boost transformation is not the value of the minimal length: it is the probability distribution of seeing one or the other of the discrete eigenvalues of the area ..."

In LQG context, the minimal area is the minimal positive eigenvalue.
That does not change for the boosted observer.
The boosted observer sees the same minimal length (or minimal area).
What changes is the probability distribution (i.e. if you repeat the experiment many times, the average or expectation value.)
 
Last edited:

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
I agree with you. The minimum size can be made as small as possible due Lorentz contraction.
I think you did not understand and you did not agree with me when you wrote that.

It is false that (in Lqq context) the minimum size can be made as small as possible due to Lorentz contraction.

I don't understand what you understood from me.
That is correct. I believe I understand what you tried to say. It is what someone unfamiliar with LQG would expect---that boosting would cause the min length to contract. But actually it does not cause the min length to contract.

BTW, what is a well posed question?
I will try to think of one. :biggrin:
 

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
...
To really understand the invariance under diffeomorphisms (as you, and Rovelli, define it) I would need to see the proof. It is probably very short. Do you know a good reference where it is done clearly?
...
I'll try to sketch a proof. Recall that a diffeo φ: M --> M moves points in the manifold around. A coord change leaves the manifold unaffected and just moves points in Rd..

We already know that the Einstein equation is invariant under coord change. That can just be cranked out. Solutions remain solutions if you just remap the coordinates with a function k: Rd --> Rd.

OK now let's pick a point in the manifold and take a very modest diffeomorphism that slightly moves that point and its immediate neighbors, but doesn't take them out of a particular coordinate chart f: U --> Rd

(See wiki, definition of a manifold, atlas of coordinate charts, as a convenience we stay in one region U so we only need one coordinate function mapping U, into Rd)

Now we can define a "fake coordinate change" function k : Rd --> Rd

k = f(φ(f-1(x)))

Start with x in Rd
Go with f-1 up into the open set U in the manifold.
Then move stuff around with phi
Then come back down with f, and you are back in Rd.

Since this is a map from Rd to Rd, it can be treated as a coordinate change. And as usual it preserves solutions to Einstein equation. All coordinate changes do.
But see what the coordinate change does! If you look at how m gets mapped in the new coordinates k(f(m)) it give the same answer as f(φ(m))

kf = fφ

Keeping the points in the manifold the same and using the new coords kf gives the same result as doing the diffeomorphism φ (moving points in the manifold) and using the old coordinate function f.
Now the first of these (changing to new coords kf) preserves solutions, so therefore the diffeomorphism must also preserve solutions.

This is just the sketch of a proof. I think it is how a proof should go, if one were to write out all the greek letters and the arrows.

I think Atyy already found an online source where the proof is presumably written out, not just a sketch.
αβγδεζηθικλμνξοπρσςτυφχψωΓΔΘΛΞΠΣΦΨΩ∏∑
 
Last edited:

MTd2

Gold Member
2,028
25
Marcus, I wanted an argument using spin foams. How about this:
matter fields is something that lives on a spin foam like electical fields on a electrical grid structure. Any possible experiment done by whoever lives on that grid will be like obtaing an eigenvalue of a surface operator. But what moves is the matter field, so any measurement of an area will be irrespective of its energy-momentum.

Right?
 

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
If you look back you see Nrqed started this thread not about QG but about GR. It has always been about GR.

The difference between (1) diffeomorphism and (2)coordinate change leaving the manifold unaffected.
The difference between (1) diffeo invariance and (2) invariance under change of coords.
The fact that in GR a geometry is an equivalence class under diffeomorphism. Not bound up with any particular manifold or any particular metric on that manifold.
The ontological consequence of that:
In GR spacetime does not exist, it has no objective or physical existence. What exists in GR are the relationships among events. The geometry itself--like the smile on the face of the cat after the cat is gone. A web of geometric relationships is information but it is not a thing. No rubber sheet, in other words.

Nrqed was also asking how one can use (2) to prove (1). How coord change invariance can be used to finesse diffeo invariance. I sketched a proof and Atyy may have found an online source. I think it's a fairly trivial thing to show.

Out of respect for the topic, I believe we should not start chatting about QG stuff like spin foams in this thread. If you have an idea about spinfoam models, why not start a separate thread about it?
 
Last edited:

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
...I wanted an argument using spin foams. How about this:
matter fields is something that lives on a spin foam like electical fields on a electrical grid structure. Any possible experiment done by whoever lives on that grid will be like obtaining an eigenvalue of a surface operator. But what moves is the matter field, so any measurement of an area will be irrespective of its energy-momentum...
Nothing can move on a spin foam. A spin foam depicts a possible course of evolution.
In that very limited sense, is like a trajectory, or a world-line. It IS the motion. So nothing moves on it.

I can see you are pursuing some analogy. But the analogy is not clear yet. It might work better if you were talking about spin networks rather than spinfoam.
Spin networks describe geometry.
Spin foams are somewhat like Feynman diagrams, or the hypothetical paths in a path integral. They are alternative possible histories of geometry, so to speak, not geometry themselves.
They depict various ways that some change in a spin network might have happened.

But I still think that if you want to discuss your idea you should start a separate thread, since it doesn't fit in here (as far as I can tell.)
 
Last edited:
346
7
I was interested in the ongoing debate between you and Atyy - did it get resolved?

In Carroll "Intro to GR" (which seems pitched to undergraduates, style-wise) on page 138 it says:
==quote==
Let’s put some of these ideas into the context of general relativity. You will often hear it proclaimed that GR is a “diffeomorphism invariant” theory. What this means is that, if the universe is represented by a manifold M with metric gµν and matter fields ψ, and φ : M → M is a diffeomorphism, then the sets (M, gµν , ψ) and
(M, φ* gµν , φ* ψ) represent the same physical situation.

==endquote==

That is, you can totally moosh and morph and it is still the same physical reality. You aren't limited to an "isometric" diffeo.

In other words physical realities correspond to equivalence classes under diffeomorphism of metrics. They don't correspond to single metrics but to diffeomorphism classes of metrics. All those that you can make by morphing the one you started with.
My worry is that t these constructions really do little more than change the identity of the points of the manifold, and that the identity of points in M NEVER mattered, in GR, SR or Newtonian theory. On the qualitative questions that physically matter, whether a manifold is flat, curved, whether the path an object travels is inertial, the constructions make no difference. Although φ is not an isometry, it doesn't change the fact that the two models constructed above are isometric - that there is some function φ' that is an isometry.
 

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
“Dadurch verlieren Zeit & Raum den letzter Rest von physikalischer Realität.”
Yes, I believe Einstein was confused.
I was interested in the ongoing debate between you and Atyy - did it get resolved?...
It is kind of you to ask. But for me the wonderful thing about talking with Atyy is the stimulating unresolution. The brilliant chimaera. The changeling aspect. We never quite agree but he forces me to think.

The direct answer to your question is "no". I'm happy with that and hope to hear more. :biggrin:
 

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
...
I did find confirmation in Sean Carroll's thing of what you see in Rovelli. Rovelli's certainly not presenting anything unusual.
In Carroll "Intro to GR" (which seems pitched to undergraduates, style-wise) on page 138 it says:
==quote==
Let’s put some of these ideas into the context of general relativity. You will often hear it proclaimed that GR is a “diffeomorphism invariant” theory. What this means is that, if the universe is represented by a manifold M with metric gµν and matter fields ψ, and φ : M → M is a diffeomorphism, then the sets (M, gµν , ψ) and
(M, φ* gµν , φ* ψ) represent the same physical situation.

==endquote==

That is, you can totally moosh and morph and it is still the same physical reality. You aren't limited to an "isometric" diffeo.

In other words physical realities correspond to equivalence classes under diffeomorphism of metrics. They don't correspond to single metrics but to diffeomorphism classes of metrics. All those that you can make by morphing the one you started with....
My worry is that t these constructions really do little more than change the identity of the points of the manifold, and that the identity of points in M NEVER mattered, in GR, SR or Newtonian theory. On the qualitative questions that physically matter, whether a manifold is flat, curved, whether the path an object travels is inertial, the constructions make no difference. Although φ is not an isometry, it doesn't change the fact that the two models constructed above are isometric - that there is some function φ' that is an isometry.
I think φ* has a substantial effect on the metric tensor. It does not simply pick gµν up and copy it to another point of the manifold. I think we probably agree on that, so let me think about your function φ'.

I don't see how I would construct φ', Yossell.
 
346
7
I think φ* has a substantial effect on the metric tensor. It does not simply pick gµν up and copy it to another point of the manifold.
Oh - then I must have misunderstood the quote from Carroll. When we start with φ, a mapping from M to M, and then we compare (M, gµν , ψ) and (M, φ*gµν , φ*ψ), what is the new metric φ*gµν of this second model but a copying of g to other points of the manifold?
 

atyy

Science Advisor
13,501
1,610
My worry is that t these constructions really do little more than change the identity of the points of the manifold, and that the identity of points in M NEVER mattered, in GR, SR or Newtonian theory.
Yes, I understand similarly. It's not that the manifold doesn't exist, it's that without putting non-dynamical or dynamical fields on it, the points are experimentally identical. Just like electrons - different electrons are identical, but electron number is still something that we can measure experimentally.

And yes, the manifold is needed to define Newtonian physics, SR and GR.

Newtonian physics and SR both have a non-dynamical metric field - this corresponds to matter which does not interact with the dynamical fields of the theory. In Newtonian gravity, the non-dynamical metric could correspond to light rays. In Maxwell's equations on flat spacetime, the dynamical fields would be electromagnetic, while the non-dynamical metric is represented by measuring rods (although we know these ultimately interact with electromagnetic fields, at the everyday level, these are inert, since the charges have all clumped together and neutralized each other at large distance scales). The distinction of GR is that we deal with a field whose coupling is universal, and so it must be dynamical.
 

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
Oh - then I must have misunderstood the quote from Carroll. When we start with φ, a mapping from M to M, and then we compare (M, gµν , ψ) and (M, φ*gµν , φ*ψ), what is the new metric φ*gµν of this second model but a copying of g to other points of the manifold?
I think Carroll would explain. Don't take my word for it.

I'll tell you how I think of it, which isn't necessarily the way Carroll does, or the way that would be right for you.

I was taught that the differential structure gives you smooth real valued functions f(m) defined on the manifold. And the tangent vectors X at point m are "derivations" defined on the functions, that satisfy a few simple conditions (linearity....). This makes the tangent vector essentially be the operation of taking directional derivative in some direction.

That makes the operation of "pushing forward" very easy to define.
If φ(m) = n and X is a tangent vector at m, then one gets a new tangent vector at n by taking a function f defined around n, and pulling it back by φ and operating by X on it.

(φ*X)f = X(f.φ)

That's a quick way to see how φ maps tangent vectors.

Now a particular tensor might be a bilinear function of two tangent vectors, or it might (to take an even simpler example) be just a linear functional defined on tangent vectors.
You "pull back" an object like that by the diffeomorphism, by pushing forward some tangent vectors for it to eat (in its old location).

It's really not as complicated as it sounds and in some sense all the words I'm saying just obscure the central message, which is that tensors transform as they are pushed around by diffeomorphisms. It is not a simple copying operation.

Someone else may wish to correct me on this or describe this in some other way which they find preferable. I'm an old guy, my math courses were several decades ago. Happy to be exposed to anyone's alternative account.
 

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
Yossell, if the diffeo is going in the wrong direction for what you want to do, then of course use φ-1.
And people have different notational conventions. I would be happy to write out, if you would like to see it, how I think you ship a package from one point to another, where the package is a linear functional defined on the tangent space. For example.
But you may have already figured that out, or you may like some different approach to defining the tangent space/bundle. So I'll just wait and see if there are questions.
 

atyy

Science Advisor
13,501
1,610
Marcus, btw, I didn't say I disagreed with Rovelli, I just said his explanation is obscure, and I don't agree that he casts any light on the conceptual foundations of GR, and that the passge is poor motivation for LQG (ie. I do find LQG interesting, but not for Rovelli's reasons in fact Rovelli does give good reasons - but they are motivations for Aysmptotic Safety, and maybe string theory - not for LQG - so maybe I like Rovelli's argument after all, since I do like Asymptotic Safety and string theory - the former for its clarity of motivation, the latter for its visionary extension of GR!)

Also, the equation you quoted from Carroll is indeed an Rovelli agrees with - but it is Rovelli's definition of a passive diffeomorphism - which is Carroll's definition of an active coordinate transformation - so I agree with both Rovelli and Carroll that that is not what distinguishes GR from SR. ("Every field theory, suitably formulated, is trivially invariant under a diffeomorphism acting on everything." http://relativity.livingreviews.org/Articles/lrr-2008-5/ [Broken])
 
Last edited by a moderator:

marcus

Science Advisor
Gold Member
Dearly Missed
24,713
783
My worry is that these constructions really do little more than change the identity of the points of the manifold,..
Yossell, did you get over that worry yet? Diffeomorphisms do radically more than just change the identity of points, since by mooshing the manifold around they also change the metric and transform all the stuff based on the tangent spaces to the manifold at the various points. At least that is how I see it. Do you still see things differently?

BTW I might mention that at least in the Riemannian case you can realize the manifold in a "coordinate free" way as a set M with a distance function dg(m,n) defined for any two points in M. In pointset topology that would be called a "metric space". You take the Riemannian metric g and use it to find the shortest path distance between any two points, and you record all that d(m,n) information and then throw away the Riemannian metric g.
It is an intuitive way to think about a metric on a manifold. No need to imagine a bundle of tangent spaces---just picture the bare manifold and imagine that you know the distance d(m,n) between any two points.

The essential thing that a diffeomorphism does, in that picture, is that it maps any, say, triple of points into some other triple separated by completely different distances. It completely changes the distances amongst any bunch of points.
 

Related Threads for: Short question about diffeomorphism invariance

Replies
8
Views
17K
Replies
4
Views
1K
Replies
15
Views
744
  • Posted
2 3
Replies
68
Views
8K
Replies
1
Views
1K
Replies
8
Views
1K
Replies
4
Views
4K

Physics Forums Values

We Value Quality
• Topics based on mainstream science
• Proper English grammar and spelling
We Value Civility
• Positive and compassionate attitudes
• Patience while debating
We Value Productivity
• Disciplined to remain on-topic
• Recognition of own weaknesses
• Solo and co-op problem solving

Hot Threads

Top