Talk:Tensor density

• Talk:Tensor density/Archive 1

Definition from elsewhere than physics?

Is there a definition of the thingies in the article coming from a pure math source, like a definition that does not relate to transformation properties? Physicists tend to dislike to formally define the concepts they are using, and when they (most reluctantly) do define them, they don't always agree among themselves. YohanN7 (talk) 14:06, 29 November 2013 (UTC)

For example, a tensor (w/o all pseudo, etc qualifiers) may be defined without reference to transformation properties. YohanN7 (talk) 19:23, 28 November 2013 (UTC)

I have a possibly different, but essentially related issue – the chosen wording. The concept of a tensor used in this article is the matrix of coefficients with respect to a chosen basis, and it excludes the coordinate-free concept of a tensor. I've seen this concept mathematically formalized somewhere in WP as a matrix function of the space of basis vectors, thus formally capturing the transformation requirement. In this context (whether formally or in the less formal physicists' approach), tensor densities are easily defined. In a coordinate-free approach to tensors, or even with wording that is agnostic to which approach is to be used, tensor densities don't work. For example, the concept of a determinant of a tensor in the coordinate-free formulation only makes sense in the case of a linear transformation (necessarily a type-(1,1) tensor; see Determinant#Abstract_algebraic_aspects). This article (to some extent unlike the article Tensor), in its wording implicitly denies the alternative abstract definition of tensors. For example, the phrase "the determinant of the metric tensor" does not make sense using the abstract geometric concept of tensor, where what is meant is "the determinant of the metric tensor coefficient matrix [with respect to a chosen basis]". By way of establishing the context in which a reader is to interpret this article, it should at least point out at the start that it is using the formalism in which the matrix of coefficients representing a tensor with respect a basis, and that here this matrix is referred to as the tensor.

A second point, related in that it suffers from the same encyclopaedic style problem, namely that it makes the implicit assumptions of a certain approach without stating the assumptions, is that the definitions assume a holonomic basis. A coordinate-based approach still makes sense with a nonholonomic basis, and thus tensor densities can defined in the same way in terms of the determinant of the transformation matrix, but the transformation matrix cannot be expressed as the matrix of partial derivatives as assumed in this article.

Am I being reasonable in saying that this article simply does not adequately outline the context and assumptions sufficiently before launching into the detail? —Quondum 14:28, 14 June 2014 (UTC)

See Spivak's book listed in the references. (It should be easy to find a digital copy using google)TR 09:00, 3 July 2014 (UTC)

Sign of the weight of the determinant of the metric tensor

I have been in an edit-war with an unregistered user with a constantly changing IP address. This makes for a very difficult situation as it is almost impossible to communicate with such people.

Although there are several things I dislike about his edit, the main problem is that he is getting the sign of the weight of the determinant of the metric tensor wrong. Essentially he has reversed the definition of weight given in the first section of the article (without changing it there). I think that changing that definition would be wrong as it would not be the usual definition and would conflict with the usage of the word "density" in ordinary speech. A weight of +1 should be used for "density" in the ordinary sense.

The integral of a scalar field over a hyper-volume of space-time is not an invariant. To make an invariant integral, one has to integrate a scalar density instead. How does this scalar density transform? For simplicity, let us consider a change of coordinate system in which each of the coordinates is doubled. Then the hyper-volume of the region over which one is integrating would seem to increase by a factor of 16 (if one ignores the metric tensor). So to compensate for that and make the integral invariant, we must multiply the scalar density by 1/16. Under this doubling of coordinates, an arrow (contravariant vector) would have its components double. So the metric tensor (a covariant matrix) would have its components multiplied by 1/4. The determinant of the metric would be multiplied by 1/256. So the square-root of its absolute value would be multiplied by 1/16. This is just the factor needed to change an ordinary scalar into a scalar density. So that square-root should have weight +1. JRSpriggs (talk) 05:54, 1 July 2014 (UTC)

Page protection?

It would force him/her to either quit or get an account and discuss. YohanN7 (talk) 13:35, 1 July 2014 (UTC)

It does not seem necessary or useful to apply page protection. I support JRSpriggs's perspective, and hope that the IP will choose to discuss here. Using informative edit summaries and the opening of this thread may be helpful in this direction. —Quondum 14:32, 1 July 2014 (UTC)

The IP(s) doesn't seem to be interested in discussion. This has been going on for some time. Therefore page protection is both necessary and useful (useful in that he/she(/they) is forced to discuss). YohanN7 (talk) 15:52, 1 July 2014 (UTC)

He reverted. This guy is not here to discuss. Semi-protect! YohanN7 (talk) 20:41, 1 July 2014 (UTC)

Hi, This is the unregistered user with the changing IP address. I only found the 'editing talk' section now and I do apologize for not writing here sooner. There are two main resons why (as you said) I 'reverse the sign of the weight'.

1) When I took a General Relativity (GR) course we used the GR textbook by Sean Carroll. (http://arxiv.org/pdf/gr-qc/9712019v1.pdf) In that book (and course) we went over tensor densities. One thing we showed/learned was that the square-root of the determinant of the metric is a scalar density of weight -1. (Sean Carroll shows it on p.52).

2) The volume element is a (0,n) tensor density of weight +1. (http://ned.ipac.caltech.edu/level5/March01/Carroll3/Carroll2.html, equation 2.44). Now, in curved space-time we want the expression

$\int \phi (x^{\mu })Ad^{n}x$

to be a intergral of a scalar over a proper n-form. Here $A$ is some constant to be determined. The scalar is a type (0,0) tensor density of weight 0 (i.e. a scalar), so that is good. Since the volume element is a tensor density of weight +1 we require $A$ to be a object of weight -1 so that the combination is a (0,0) tensor of weight 0. It so happens that ${\sqrt {\vert g\vert }}$ fulfills this requirement as seen by (http://ned.ipac.caltech.edu/level5/March01/Carroll3/Carroll2.html, equation 2.41).

As further proof that this is correct, consider a famous action, say the Einstein-Hilbert (EH) action. For simplicity lets look at the EH-action with zero cosmological constant and no matter. The EH action is $S\propto \int R{\sqrt {-g}}d^{4}x$. Do you deny this is correct? According to you ${\sqrt {-g}}$ has weight +1 which (since the volume element has weight +1) means that the intergral has overall weight $+2\not =0$. But it is a known fact that the EH-action is a true scalar. Or if you insist that it has weight 0 then you are saying that the EH-action should be $S\propto \int {\frac {R}{\sqrt {-g}}}d^{4}x$ which, obviously is wrong. QED

P.s. I use absolute value on g to ensure positivity even in a Riemannian manifold. — Preceding unsigned comment added by 206.45.187.125 (talk) 21:58, 1 July 2014 (UTC)

Hi, and welcome to the talk page. YohanN7 (talk) 22:42, 1 July 2014 (UTC)

Three things you need to know (for a start):

• You need to update the article with references supporting your claims (or you are likely to be reverted).
• Lecture notes do generally not constitute acceptable references. Nor does your own arguments - even if they are correct.
• It is customary here at Wikipedia that when an edit war occurs, then the matter should be settled on the talk page before any changes are made.

YohanN7 (talk) 00:09, 2 July 2014 (UTC)

Hi Yohan, thank you for your reply.

In reference to your three items...

i) I did support my claims both on the talk and the article.

ii) The reference I used were from a textbook called 'Spacetime and Geometry: An Introducton to General Relativity' by Sean Carroll which is a graduate level textbook and widely accepted in physics today. It so happens that Carroll made lecture notes as well (which I referenced). However his 'lecture notes' are identical to his textbook. Here is his blog: http://preposterousuniverse(dot)com/spacetimeandgeometry/

iii) That I did not know and I apologize! — Preceding unsigned comment added by 205.200.233.195 (talk) 01:00, 2 July 2014 (UTC)

It looks like things are going to work out just fine. I'm not an expert on the subject, so I'll leave it to you guys who are to sort it out. But I have a hunch that it is just a matter of definition or convention. If this is the case, then both conventions can be mentioned in the article, with the most common one being the "main" definition to be used throughout. YohanN7 (talk) 01:54, 2 July 2014 (UTC)

To anonymous:

Your teacher was mistaken, if he equated what Carroll called the "naive volume element" with a density of weight +1. In ordinary language density is characterized by

$Density={\frac {d\,Mass}{d\,Volume}}$

So density is inversely related to volume, not proportional to volume. That is, your basic assumption is wrong.

Also, density is used in lieu of four anti-symmetric covariant indices. So you should not combine both a notion of weighted density and tensor indices in your volume element. JRSpriggs (talk) 06:22, 2 July 2014 (UTC)

Saying that people adhering to a different convention are wrong is silly.TR 10:49, 2 July 2014 (UTC)

To JRSpriggs

Carroll SHOWS that ${\sqrt {\vert g\vert }}$ is a scalar density of -1!!! Furthermore (Carroll also shows this) the volume element transforms under a co-ordinate transformation with a determinant as such: $d^{n}x'=\left\vert {\frac {\partial x'}{\partial x}}\right\vert d^{n}x$ Thus it is a density of +1.

So the combination ${\sqrt {\vert g\vert }}d^{n}x$ has total weight $-1+1=0$. So it is a ordinary $(0,0)$ tensor.

Also I highly doubt that both Carroll and my professors are wrong...espcially that they have a PhD in physics, specilizing in GR. — Preceding unsigned comment added by 205.200.233.195 (talk) 08:06, 2 July 2014 (UTC)

It is useful to realize that there are different conventions for the definition of weight of a tensor density that differ by a sign. Carroll (like for example Weinberg) use a convention in which the Levi-Civita symbol with lower indices has weight +1. Consequently, the square root of g has weight -1 in that convention. The definition currently appearing in the article uses the opposite convention, where the square root of g has weight +1. This is not a matter of right or wrong, but a matter of convention, and whatever convention is used in the article should be used consistently.TR 10:49, 2 July 2014 (UTC)

To JPSpriggs

Ok, this lasted too long and I am sick of it. You know what? I give up. I raise the white flag. YOUR RIGHT AND I (with my tiny miniscule insignificant pee brian) AM WRONG!!!! HAPPY?

FYI, Personally I will keep believing what I believe but you probably don't care about that. I will continue physics and continue with my convention and if I fail university and life and become a hobo on the streets because of it, then SO BE IT, (You can come by and laugh at my stuipidity then). By the looks of it, my miniscule insignificant pee brain and I deserve it. (and according to you it will happen)

At the beginning you said there were other things you didn't like about my post. I was going to ask you what but I don't need to because I can guess...it is EVERYTHING.

Have a good life, this future stupid IDIOTIC loser (aka me) won't bother you again!!!! — Preceding unsigned comment added by 206.45.29.3 (talk) 19:08, 2 July 2014 (UTC)

Ok I just realized that my argument was valid if you defined the tensor transformation as ${\mathfrak {T}}_{\nu '_{1}\cdots \nu '_{m}}^{\mu '_{1}\cdots \mu '_{n}}=\left|{\frac {\partial x'}{\partial x}}\right|^{w}{\frac {\partial x^{\mu '_{1}}}{\partial x^{\mu _{1}}}}\cdots {\frac {\partial x^{\mu '_{n}}}{\partial x^{\mu _{n}}}}{\frac {\partial x^{\nu _{1}}}{\partial x^{\nu '_{1}}}}\cdots {\frac {\partial x^{\nu _{m}}}{\partial x^{\nu '_{m}}}}{\mathfrak {T}}_{\nu _{1}\cdots \nu _{m}}^{\mu _{1}\cdots \mu _{n}}$

(then g has weight -2)

BUT you defined it as

${\mathfrak {T}}_{\nu '_{1}\cdots \nu '_{m}}^{\mu '_{1}\cdots \mu '_{n}}=\left|{\frac {\partial x}{\partial x'}}\right|^{w}{\frac {\partial x^{\mu '_{1}}}{\partial x^{\mu _{1}}}}\cdots {\frac {\partial x^{\mu '_{n}}}{\partial x^{\mu _{n}}}}{\frac {\partial x^{\nu _{1}}}{\partial x^{\nu '_{1}}}}\cdots {\frac {\partial x^{\nu _{m}}}{\partial x^{\nu '_{m}}}}{\mathfrak {T}}_{\nu _{1}\cdots \nu _{m}}^{\mu _{1}\cdots \mu _{n}}$

(then g has weight +2)

So I again apoligize. Truthfully, looking back, I am appalled at how stupid and mentally challenged I am. I really should be in a mental institution come to think of it. I know I graduated with honors but, honestly, I am SHOCKED that I passed elementry school with my mentally challanged brain.

Have a good life! — Preceding unsigned comment added by 206.45.181.160 (talk) 02:30, 3 July 2014 (UTC)

Hi, we all make mistakes. An important thing on Wikipedia is to follow etiquette so that you do not make life difficult for other editors. I know this because I found out the hard way! — Cheers, Steelpillow (Talk) 08:29, 3 July 2014 (UTC)

I'm not sure what the current state of this discussion is. I only just became aware of it after editing Tensor#Tensor densities because the convention used there seemed to me to be wrong (or at least inconsistent with most conventions). Let me explain why JRSpriggs is "right" and other conventions are "wrong". I will just discuss scalar densities. (Here by "density" I just mean "1-density"). A density is a quantity on a manifold that is integrable. Examples are charge densities, mass densities, population densities, (even "volume densities", although if we think of this, we are likely to confuse ourselves). The main point is that a density is something that can be integrated over small pieces of the manifold (in a coordinate-independent manner), producing a single number (scalar). The Jacobian change of variables formula dictates what the transformation law for a density must be under a change in coordinates. It must vary by the inverse of the Jacobian of the coordinate change. That is,

$f(y)=f(x)|\det(\partial y/\partial x)|^{-1}.$

This is the convention adopted in the present article, and in JRSpriggs' comment above.

Put another way, a density is something that defines a scalar field when it is multiplied by the wedge product of the coordinate functions. Thus if the wedge product of the coordinate functions has units of volume, densities have units of 1/volume (or (whatever unit)/volume). Sławomir Biały (talk) 13:32, 3 July 2014 (UTC)

@Sławomir Biały: I'm slightly confused. Both the Tensor article and the present article follow the mnemonic "new variables downstairs in weighting factor" and they are presumably using the "correct" convention. Your change of variables formula above have "new variables upstairs". YohanN7 (talk) 15:50, 3 July 2014 (UTC)

yes, you're right. It should be the inverse of the jacobian. Sławomir Biały (talk) 16:28, 3 July 2014 (UTC)

The long-term edit war (which I think justifies semiprotection), has spilled over to the Tensor article as well, fwiw. Sławomir Biały (talk) 19:38, 3 July 2014 (UTC)

The convention where the weight of g is -2 is NOT I repeat NOT wrong...it is just a convention (Which I prefer) — Preceding unsigned comment added by 142.132.71.246 (talk) 21:29, 3 July 2014 (UTC)

Fine. Except in that convention, the word "density" is unrelated to its usual mathematical meaning. Sławomir Biały (talk) 22:35, 3 July 2014 (UTC)

To JRSpriggs

For your information I spoke to THREE different university professors who all have a PhD IN THEORETICAL PHYSICS specializing in GRAVITY. They ALL said that the transformation of tensor densities defined as:

${\mathfrak {T}}_{\nu '_{1}\cdots \nu '_{m}}^{\mu '_{1}\cdots \mu '_{n}}=\left|{\frac {\partial x'}{\partial x}}\right|^{w}{\frac {\partial x^{\mu '_{1}}}{\partial x^{\mu _{1}}}}\cdots {\frac {\partial x^{\mu '_{n}}}{\partial x^{\mu _{n}}}}{\frac {\partial x^{\nu _{1}}}{\partial x^{\nu '_{1}}}}\cdots {\frac {\partial x^{\nu _{m}}}{\partial x^{\nu '_{m}}}}{\mathfrak {T}}_{\nu _{1}\cdots \nu _{m}}^{\mu _{1}\cdots \mu _{n}}$

is a convention that is RIGHT. By THIS convention $g$ is a scalar density of $-2$. This is NOT WRONG. It is just a convention. So by saying (and I quote) "Your teacher was mistaken, if he equated what Carroll called the "naive volume element" with a density of weight +1. ...That is, your basic assumption is wrong." your simply telling everyone that you are arrogant!

Good day! — Preceding unsigned comment added by 207.161.69.254 (talk) 23:11, 3 July 2014 (UTC)

Fine. If three PhDs agree that a "mass density" or "charge density" is not a density at all, clearly they must be right. Sławomir Biały (talk) 23:22, 3 July 2014 (UTC)

To be honest, I'm reacting a bit against the right-or-wrong formulation. Chose convention as suits the occasion. It is little more (actually no more) than choosing a basis in Rⁿ. Sometimes you don't even want it to be orthogonal because it would complicate things. YohanN7 (talk) 00:41, 4 July 2014 (UTC)

In the end, any vocabulary is a matter of convention. But not all vocabularies are equally useful. In this case, calling something a "density" should agree with the premathematical use of the term, in my opinion. The convention should be chosen so that mass, charge, population, etc, are densities. Wouldn't you agree? Choosing the opposite convention therefore seems arbitrary and unmotivated. Also it is not a common one in my experience. Sławomir Biały (talk) 01:41, 4 July 2014 (UTC)

The premathematical (in the sense of me not properly understanding the detail) use would lead me to expect that a scalar density (which I'd assume would be said to have weight +1) would be one that when integrated over a region (using the volume element), yields a true scalar. Charge and mass densities do not qualify, since they are components of tensors, not scalars, unlike the trace of the stress–energy tensor or the Ricci scalar. Here there is room to define what "of weight w" means, but I guess this'd be the most natural. Is this the sense in which it is currently used? —Quondum 02:14, 4 July 2014 (UTC)

@Sławomir, I agree of course. I was speaking in general terms. I don't have a problem at all with the Wikipedia articles making internal choices that could be said are natural. But other choices in common existence could (and should imo) be mentioned, and not be called wrong, and, likewise, the Wiki-choice shouldn't be called right, just natural or appropriate or whatever, preferably with an explanation. It is a potential blunder to call, say, Weinberg, wrong. It is hard to find even the slightest typo is his books, errors in equations are almost unheard of, and would he make gross conceptual errors? (Having said that, his mathematical taste is horrendous )

So, I suggest, in the present case, that we motivate briefly our choice of definition (so that it goes into the top part of the article), and there also mention the alternative, with pop-up citations. This would increase the quality of the article, and keep the peace longer. YohanN7 (talk) 03:26, 4 July 2014 (UTC)

I think it is misleading to think of the volume form as a density in this context, since the "volume form" might refer to a scalar or a density, depending on what "volume form" means. If "volume form" means the wedge product of the differentials of the coordinate functions, then it is not coordinate invariant (it picks up a factor of the Jacobian). But of course, it does not do so in a way that results in an invariant integral. For that, one needs an actual scalar. This scalar is obtained by taking the product of this "volume form" with a density! This is what the slogan "densities have a well-defined integral" means.

I don't mean to say that Weinberg is mathematically wrong in his choice. I have been using "wrong" in quotes, because it is after all only my opinion. But no one so far seems to have produced a clear motivation for the opposite convention, aside from just appeal to authority (an authority which I think is fairly limited as far as it goes). Maybe Weinberg did have good reasons, but Weinberg is somewhat unconventional in his entire approach to GR, and he often uses the opposite conventions of others working in that area (cf. Penrose, Hawking, for example). I have no doubt that he had his reasons for choosing these conventions, but so far no one has been able to give a compelling account of those reasons. Sławomir Biały (talk) 12:45, 4 July 2014 (UTC)

I'm not getting my point through. I am not defending unconventional choices, nor do I say every choice is as good as any other (though I wrote something that could be misinterpreted to that effect above).

The point is this: The articles are supposed to reflect what is in the literature, not what we think ought to be in the literature. When there are multiple conventions around, the article should make its choice. At that point a motivation could well be given if there (like in our case) is a compelling reason for making the choice. Then, in the interest of completeness, list some other common choices (naturally without motivation unless there is one), and then get on with the article. This pattern reflects what it looks like in (good) textbooks as well. Alternatives are mentioned. But don't misunderstand me as proposing that every fringe convention should be mentioned. Is Weinberg's authority enough to make his choice of convention get mention in this case? Don't know. I haven't read his work on GR, just his QFT vol 1,2,3. (Yes, these were totally unconventional - and great.) Just trying to reason in general terms here. YohanN7 (talk) 15:31, 4 July 2014 (UTC)

I agree with Yohan completely. As an aside: I have been thinking and (I will probably be criticised severely for this), I think (personally) the name 'Tensor density' is a misnomer. The reason being is, when one hears the word 'density' or the words '____ density', our minds automatically think that this is something that (once integrated) is a physical thing, e.g. can be visualized or picked up...in other words one can interact with it, be it in implicit way.

For example mass density can be integrated to give a mass of a object that we can pick up...charge density can be integrated to give charge of an object which we can see (some of it) for example electric charge and electricity, etc.

But Tensors are purely abstract and mathematical concepts which can be written on paper and were INVENTED to understand GR yes, but one can't see tensors, pick them up, or interact with them at all in any way. (Like numbers for example...you cannot pick up 2)...[FYI I don't consider working with them on paper as interacting lol]

So by calling something a 'tensor density' one assumes that one can get a tensor which is a physical thing. But it's NOT. It makes more sense for it to be called just a 'weighted tensor'

What do you guys think? Or am I off my rockers? — Preceding unsigned comment added by 206.45.181.185 (talk) 04:13, 4 July 2014 (UTC)

Often it is said that definitions and conventions are arbitrary and that it does not matter which version one uses. This is wrong. It is important to make them as intuitive as possible, especially for people just learning the subject. Just imagine if when you swept your finger to the right over a cell phone, the image moved upward or spun around or shrank, rather than moving to the right.

Tensor densities (of weight +1) are called "densities" precisely because they are used as the mathematical representation of things which are called densities in ordinary life (classical physics). Sławomir Biały is right about that. Action (actually the Lagrangian density, as anonymous pointed out below) is a scalar density. The electric charge density and current density form a vector density. It could be represented either as a contravariant vector density with 4 components or equivalently as an anti-symmetric three-times-covariant ordinary tensor with 64 (mostly redundant) components. JRSpriggs (talk) 04:36, 4 July 2014 (UTC)

Sorry but I disagree with you JRSpriggs: 1) Conventions ARE arbitrary and just because person A uses a different convention from person B doesn't make it wrong. As long as they are consistent with their conventions the physics that the two are describing are COMPLETELY EQUIVALENT. The math will be different but the answer/interpretation will be the same.

Consider the sign convention for the metric (-,+,+,+) vs (+,-,-,-). I personaly use the mostly plus convention, but as long as people are consistant they are both right. By saying that "Often it is said that definitions and conventions are arbitrary and that it does not matter which version one uses. This is wrong." you are saying that either ALL particle physicists are wrong or ALL relativity physicists are wrong, which (if you ask me) is arrogant.

Math IS arbitrary, numbers are arbitrary...Just look at complex numbers for example. You won't see a '2+i' walking down the street and smiling at you now would you? But we use numbers to describe the world, so how can it be wrong?

2) Action is not a density but rather a functional. You probably meant to say Lagrangian density. — Preceding unsigned comment added by 206.45.181.185 (talk) 05:07, 4 July 2014 (UTC)

Perhaps we should avoid the use of the terms "right" and "wrong" in this discussion. And I interpret JRSpriggs differently – he seems to be saying that we do ourselves a disservice if we choose to completely sweep the benefits of selecting certain conventions under the rug, with which I agree. Essentially, since the word "density" is used, it will trigger certain natural associations as its selection was presumably intended to do. Thus, I would expect the components a tensor density of weight +1 to scale exactly as the components of something that I would call a density, and hopefully this would suggest a convention.

This is where I fall off the wagon, though, since it seems to me that scalar densities (in the everyday sense) are invariant, and all tensors transform with weight 0. Thus, the concept of a "tensor density" as described in this article is an artefact of retaining a form of integration that is coordinate-dependent, and thus (quite unnecessarily) undermines the fundamental motivation of tensors: coordinate-independence. It leads to defining densities in the sense of this article (of weight +1?) as entities that have been scaled to allow integration (using what I would call a flawed form of integration, or more specifically, volume element) over a hypervolume region to yield an invariant quantity. This article describes tensor densities, but fails to explain why they have any value, or what motivates them. As far as I can tell, the motivation will relate to a specific form of integration, where it was presumably decided to retain a form that rescaled the units with a change of coordinates, and dumped the responsibility for compensating this on the quantity being integrated. I would encourage a knowledgeable editor to insert such a motivation into the article; surely it must be sourceable. —Quondum 18:48, 4 July 2014 (UTC)

Maybe I am interpreting JRSpriggs incorrectly, but it sounds like he is trying to state that there is the one and only convention. (Whatever that convention is) and all other conventions and by extention people who use these "wrong" conventions are stupid/wrong... (In analogy it sounds like religion - there is only one God and all other Gods are false Gods and if you belive in them then you go to hell). There are MANY physicists who work in the same feild (eg Gravity) who use different conventions.

• Steven Weinburg in his book Gravitation and Cosmology: Principles and Applications of the General Theory of Relativity uses the $(-,+,+,+)$ signature and says that g has weight $-2$
• Misner, Thorne, Wheeler in their book Gravitation uses the $(-,+,+,+)$ signature and says that g has weight $+2$
• Sean Carroll in his book Introduction to General Relativity Spacetime and Geometry uses the $(-,+,+,+)$ signature and says that g has weight $-2$
• Ray D'Inverno in his book Introducing Einsteins's Relativity uses the $(+,-,-,-)$ signature and says that g has weight $+2$

Now all these physicists understand and know GR and (more to the point) come to the same conclusions. I highly doubt anyone would say that any of these people are wrong simply because they use a convention that someone dislikes. Yet by JRSpriggs argument he is basically saying that people like me, Weinburg and Carroll who use the $(-,+,+,+)$ signature and uses the definition that g has weight $-2$ are wrong/stupid/etc... — Preceding unsigned comment added by 207.161.172.197 (talk) 02:32, 5 July 2014 (UTC)

I suggest you mellow your interpretation; it is a principle in Wikipedia to assume good faith on the part of other editors. I've given an interpretation that differs from yours and is compatible with none of these authors, conventions or you being slighted. In any event, Wikipedia does not go by what is "right" or "true", but rather by what is published by notable sources. Given that both conventions are clearly used by notable sources, you can proceed from the assumption that they should both be mentioned in the article. For internal consistency, however, the article does need to select one only of the prevailing conventions for its presentation of the details. A further principle is not to change the convention in use in an article without consensus. This essentially means that arguments about the best convention have little point other than to sway the consensus prior to any change. JRSpriggs and Sławomir have both expressed their preference, YohanN7 is essentially saying the same as I've said now, and as you will gather from what I've said above, I fail to grasp the utility of the concept of tensor densities, nor do I see a natural preference for either convention. Simply put, if you wish to change the article, you will need to persuade the other involved editors to agree to the changes that they are presently opposed to, and taking a combative stance is not going to help with that. Editing works best through mutual understanding and cooperation. As you settle in as a Wikipedia editor, I expect you'll learn to enjoy being a part of a mutual, cooperative endeavour that we all find rewarding. Your attention to detail and energy for researching specifics are a plus here. You may find Wikipedia:Five pillars and the numerous articles and essays that it links to illuminating. —Quondum 04:17, 5 July 2014 (UTC)

On the utility of tensor densities. The usefulness of tensor densities becomes apparent when you apply the variational principle to a gravitational action. The invariant (weight 0) volume form is not invariant under variations of the metric, the coordinate 4-form is. Consequently, it is the Lagrangian density (weight +/- 1) that naturally appears in your Euler-Lagrange equations. TR 08:34, 7 July 2014 (UTC)

@Anonymous: The choice of signature is AFAIK arbitrary, while the choice of convention for tensor density isn't completely arbitrary. In the latter case there is a natural choice. The other choice isn't wrong because it will produce identical results (apart from a sign switch for weights). But it goes counter the intuitive when it comes to the word "density". The convention doesn't match the terminology. That is all there is to it. The reason this is so is made clear in this thread, especially in JRSprigg's posts.

There's nothing stopping you from editing the article and mentioning the other convention. I can't promise it wont be reverted, but you might as well try. But don't get the idea that there should be a parallel development of the two conventions (I'd be quick to revert) or anything like it. I'd try to describe it in words in one sentence with one reference (hint: Weinberg has higher weight than Carroll), the more economical, the bigger chance it lasts, but that's really up to you. YohanN7 (talk) 09:56, 5 July 2014 (UTC)

Actually, upon closer inspection, the article was already mentioning the other convention before the definitions. All that was needed was a reference. I added Weinberg (and got rid of a reference needed tag in the process). YohanN7 (talk) 11:45, 5 July 2014 (UTC)

I also added a brief motivation for the choice of convention in the article along with the difference in the other convention. The latter is contained in the popup reference.

Correct me if I'm wrong, but there is little conceptual difference and no substantial difference at all between the two conventions. The conceptual difference is that "ordinary" densities get weight +1 with our convention, and -1 with the other. Both are equally invariantly integrable since while our convention assigns the weight -1 to the "coordinate volume element" (wedge of coordinate differentials), the other assigns +1; hence the net result is a true scalar (vector, tensor) when multiplied together with the density.

Ironically, there is one possible convention that seems to me to be better than the present choice of ours. That would be to use the other convention, but with an additional flip in the sign of the weight. That way Jacobian determinants of the inverse transformation are kept out of sight, and ordinary densities are assigned weight +1. (Admittedly, this is our convention, but with the defining formula reformulated in a more natural way (Jacobian determinant of forward transformation.))

At the risk of making myself more unpopular than I already am, I'll (hesitantly) say that there seems to be very little real substance in what has been said (with close to religious conviction) in this thread. What it seemingly boils down to is where to put minus-signs and inverses. I feel little religiosity about having +1 for weight for an ordinary density though I prefer it before -1. A "weight" is anyhow something that is over, above and beyond the concept of a classical density.

So now I have screwed up badly YohanN7 (talk) 14:41, 5 July 2014 (UTC)

Not religious conviction, but usually on the literature the bare word "density" means "1-density". From this basic premise, the rest follows. Sławomir Biały (talk) 16:21, 5 July 2014 (UTC)

Ah, but that is far away from intuitive notions about classical densities and, furthermore, already has a choice of convention (ours) built in. The motivation would then be commonality, not naturality. But let us not argue . Article ok (last sentence in lead + popup)? YohanN7 (talk) 16:41, 5 July 2014 (UTC)

Sławomir, I can't shake the feeling that different concepts are being conflated under a shared name, causing a confusion. Looking at Density on a manifold, it seems to me that the concept is best understood in the absence of a metric tensor, and the crucial feature appears to be that a density is a quantity that can be integrated (intrinsically) over a region of the manifold. I deduce that this is a coordinate-independent concept, which puts it at odds with this article. If one defines a volume form in terms of the coordinates as say dx ∧ dy ∧ dz, one has an inherently coordinate-dependent volume form, but if volume is to be a coordinate-independent concept, one is forced to multiply this expression by the Jacobian determinant J, thus arriving at an expression v = J dx ∧ dy ∧ dz that is a coordinate-independent "density of volume" (though one needs an initial frame for a density reference). Now I am guessing here, but this invariant "density of volume" is a 3-form, and thus an ordinary tensor; this is deserving of the name "density". This is, however, not the concept represented by a "tensor density" of this article; this article considers the quantity dx ∧ dy ∧ dz as a density of weight ±1, and v as a density of weight 0. Ergo: the word "density" is being used with two distinct meanings. In light of this, I do not follow you when you say "the rest follows". —Quondum 18:54, 5 July 2014 (UTC)

In your example, it is J that we would call the density (a "volume density"). The three-form itself is a scalar (it has weight zero and no indices). Now there is also a canonical identification of 3-forms on the manifold with the bundle of skew-symmetric tensors of type (0,3). Skew forms of the top degree do have a transformation law under change of coordinates, one that naturally identifies them as densities (at least on an oriented manifold). Sławomir Biały (talk) 14:02, 6 July 2014 (UTC)

The transformation law is $d{\bar {x}}^{1}\land \ldots \land d{\bar {x}}^{n}=\operatorname {det} \left({\frac {\partial {\bar {x}}^{j}}{\partial x^{i}}}\right)dx^{1}\land \ldots \land dx^{n}\qquad {\text{Lee (12.13)}},$ so it is a density of weight −1.

You misunderstand me. A skew symmetric tensor of type (0,n) has n downstairs indices. It transforms as a 1-density. Sławomir Biały (talk) 15:39, 6 July 2014 (UTC)

I think I understand what you mean. This here is surely related to the topic #Invariantly integrable quantities below. What do you and what do you not include when you "measure" the weight. If we have a tensor, then the total weight of the thing must be 0 as far as I can see. YohanN7 (talk) 17:36, 6 July 2014 (UTC)

Right. There are three ways of thinking of a skew (0,n) form. One is as a tensor (weight zero), one is as a scalar (a differential n-form of weight zero), and one is as a density (a scalar with weight 1, in our convention). All of these perspectives are connected by a "natural isomorphism" (that is, an isomorphism that does not require any choices to be made or additional structure). Sławomir Biały (talk) 18:02, 6 July 2014 (UTC)

---

Do you still have access to Lee's book? You can find that kind of density in chapter 14 (old ed). It allows you to do invariant integration on non-orientable manifolds. Sorry for interrupting. YohanN7 (talk) 19:06, 5 July 2014 (UTC)

Hi Yohan

Ok I will put at the bottom of the article a side note about the other convention and *hope* that it doesn't get deleted. (Though I suspect it will) Does that sound ok?

The article is updated already. YohanN7 (talk) 19:20, 5 July 2014 (UTC)

Yohan

I added a secton titled "Note about conventions" Is this fine? I hope it won't get deleted...though I bet JRSpriggs will delete it...lol — Preceding unsigned comment added by 207.161.101.195 (talk) 19:47, 5 July 2014 (UTC)

No, JRSpriggs won't delete it. I will. First off, I told you the article is updated already with enough on other conventions already by me. You can find it as the last paragraph of section Tensor density#Definition. That is perfectly adequate for a convention that the article doesn't make use of. Secondly, and more importantly, your attempt is a one page long continuation of your angered argumentation on this talk page. Did you honestly expect it to stay? YohanN7 (talk) 20:32, 5 July 2014 (UTC)

To Anon

Do you remember my item 3 above? You wrote "That I did not know and I apologize!" in response too it. You seem to understand the principle, but just don't care about following it.

Please don't do any more edits to the article before consensus is reached. YohanN7 (talk) 13:53, 2 July 2014 (UTC)

I'm not saying you are wrong (or right). As I suspected, this is much a matter of definition and convention. Even so, since you have been edit warring, the original version should stand as long as no consensus is reached. YohanN7 (talk) 14:43, 2 July 2014 (UTC)

To Yohan

Yes and I am sorry about that. This did last too long and should have stopped....I am sorry and I'll stop contributing to this page.

Now you screw up pretty badly. I had written "I'm not saying you are wrong (or right)." You changed that to "I'm not saying you are right." Here's diff: [1]

Do not ever edit what other people have written on a talk page! YohanN7 (talk) 19:18, 2 July 2014 (UTC)

Fine, I am sorry I'll stop contributing. — Preceding unsigned comment added by 206.45.181.160 (talk) 21:15, 2 July 2014 (UTC)

Comparing to Lee's definition

(one post copied from above thread, separating as above thread is getting unwieldy)

Do you still have access to Lee's book? You can find that kind of density in chapter 14 (old ed). It allows you to do invariant integration on non-orientable manifolds. Sorry for interrupting. YohanN7 (talk) 19:06, 5 July 2014 (UTC)

I presume you're referring to John M Lee's Introduction to Smooth Manifolds. I don't have complete access. He claims "a density is not a tensor, but I assume that this is because of the nonlinearity of the absolute value. There are (IMO) cleaner ways to deal with non-orientability than via the absolute value. The sign of the Jacobian is not of concern to the weight mentioned above, only its scaling. If one uses the Jacobian J instead of |J| (as one can do on an orientable manifold, or locally on any manifold), then Lee's density is R-linear, and hence a tensor. Aside from the issue with the sign, Lee's definition is compatible that of an n-form. So yes, I'm saying that "that kind of density" (Lee's) is entirely different from "this kind of density" (this article's), but that only "that kind" can naturally be called a "density", that "this kind" should have a different name, and that there is thus no natural selection between the two weight conventions, and indeed no need for "this kind" at all (a statement for which I'll probably be taken to task). —Quondum 21:14, 5 July 2014 (UTC)

I think Lee's density is a density in the sense of the present article with weight -1 (our convention). Haven't checked thoroughly though. YohanN7 (talk) 21:57, 5 July 2014 (UTC)

If a density in Lee's sense is something that can be integrated, it has weight +1 in our convention (reams of text explain this above). I only have the old electronic version of his book on me. I'm away from my dead tree copies at the moment. He doesn't discuss densities in that version unfortunately. On oriented manifolds, the bundle of densities in naturally isomorphic to the bundle of n-forms, where n is the dimension of the base manifold. In general, a non-oriented manifold M has an oriented double cover, N→M, together with an action of the group with two elements Z_2. The bundle of densities is naturally isomorphic the bundle of Z_2 equivariant maps from N to the bundle of n-forms on M. Such a thing has an invariant integral because it can be pulled back to an n-form on the oriented manifold N, integrated there, and then divided by two. Notice that everything in this construction is tensorial: there are no absolute values appearing. But the densities are tensors on the bundle N. Sławomir Biały (talk) 23:09, 5 July 2014 (UTC)

I find the double cover of a manifold idea a rather beautiful way of dealing with the integrability orientation issue – I was toying with it but was unsure; I'm glad you mentioned it. In any event, Lee's use of densities appears to be a way of dealing with the orientability problem. Here is a link (or this newer edition) to a publicly viewable page in his book. It should be clear that he is defining it in a coordinate-independent fashion. In a nutshell, Lee's density is defined to scale its result according to the determinant of a transformation applied to its input vectors, but this article's density is defined to scale its result according to the Jacobian determinant of the transformation of the basis (potentially in addition to the scaling as per Lee's density, if it is an n-form density), and cannot be made tensorial via your double cover construction. These are two very different scaling dependencies. Thus, the two meanings of "density" are entirely distinct. —Quondum 00:55, 6 July 2014 (UTC)

Thanks for the link. The usual bare word "density" refers to an even 1-density (in the terminology of this article). That is, one for which the Jacobian appears in the absolute value. This is probably worth mentioning. Also, I wonder whether the distinction between even and odd densities is a useful one at all. Sławomir Biały (talk) 02:41, 6 July 2014 (UTC)

@Q Lee's density measures volumes spanned by n vectors in the tangent space at each point of the manifold. That's, for instance, what every n-form does, some of them with a sign, this one is unsigned. These things can be integrated and that's the point. (The same is true for volume forms on orientable manifolds.) This article addresses what happens when the coordinates are changed. Edit: What I'm trying to say is that there aren't two meanings of density in Lee's density. YohanN7 (talk) 03:35, 6 July 2014 (UTC)

The article ought to address why 1-densities are invariantly integrable. As far as I can see it is because the "usual" volume element dx⁴ has weight -1. When multiplied with something with weight +1, these weights add and the net result is a true scalar with weight 0 that is immune to coordinate changes. Also, Lee's density appears to be just an even 1-density. YohanN7 (talk) 03:09, 6 July 2014 (UTC)

Invariantly integrable quantities

I don't think it is clear cut in our discussion what is required for an invariant integral to exists. It is hinted that objects with weight +1 are invariantly integrable, no other. I'd say that it depends on what pre-exists in the integrand. If there is an invariant volume element in place, then objects of weight 0 will be invariantly integrable with respect to it. If, on the other hand, only dx⁴ is in place, then an object of weight +1 will be invariantly integrable.

The default seems to be that dx⁴ is in place in the integrand, no invariant element, i.e. not √(−g)dx⁴.

By extension if the integrand has weight N (including everything!), then it can be multiplied by an object of weight −N and then invariantly integrated.

Are these observations correct? YohanN7 (talk) 04:31, 6 July 2014 (UTC)

Yes, and you're hopefully starting to get my point. If we use the (invariant) weight-0 √−gdx⁴ as our volume element, the whole point of multiplying by the Jacobian determinant elsewhere disappears. The volume element becomes tensorial, and we need only tensors, never "weighted densities" in the sense of this article. In short, we never need to work with anything not of weight 0. The whole, complex treatment of weights appears to be an artefact of failing to identify an invariant volume element at the start. (I've said this before above.) Furthermore, the nontensorial even/odd/pseudo sign shebang can be abolished by using a double cover of the manifold, as per Sławomir. —Quondum 05:34, 6 July 2014 (UTC)

But identifying a volume form at the start is not a canonical choice. The bundle of densities is trivial, but not canonically trivial. Sławomir Biały (talk) 13:30, 6 July 2014 (UTC)

On the utility of tensor densities

(again splitting out what seems like a separate topic, by copying a post from a thread above)

On the utility of tensor densities. The usefulness of tensor densities becomes apparent when you apply the variational principle to a gravitational action. The invariant (weight 0) volume form is not invariant under variations of the metric, the coordinate 4-form is. Consequently, it is the Lagrangian density (weight +/- 1) that naturally appears in your Euler-Lagrange equations. TR 08:34, 7 July 2014 (UTC)

JRSpriggs did mention this example before when I tried to claim that tensor densities have no intrinsic mathematical value in any context other than as a calculation tool, some time ago in another thread somewhere. At first blush your (and JSpriggs's) claim seems to be a claim of a substantive (i.e. with mathematically significant) meaning to tensor densities. As I commented elsewhere, this formulation appears to be a peculiarity of the formulation, and cannot have mathematical significance. In particular, if one looks at the action

$S={1 \over 2\kappa }\int R{\sqrt {-g}}\,\mathrm {d} ^{4}x\;,$

one sees that the the volume form ${\sqrt {-g}}\,\mathrm {d} ^{4}x$ (with a fillip to deal with sign on non-orientable manifolds) is the invariant quantity, and that as long as one does not try to work separately with the inherently basis-dependent quantity $\mathrm {d} ^{4}x$, even here the use of tensor densities vanishes. I'm going to (not for the first time) claim that in a properly formulated basis-independent treatment where this simple point is kept in mind, the entire tensor density treatment is superfluous. Okay, lotsa people are going to turn in their graves and it will offend many others, but please prove me wrong with a counterexample. —Quondum 13:39, 7 July 2014 (UTC)

Well, actually, I haven't addressed your comment and appear to be stating simply the contrary (I actually do not know how to deal with the variational principle), so I'll try to rephrase what I suspect underlies your comment. Presumably, for your statement to be true, one must have made the (arbitrary) choice to coordinatize the manifold, and then vary the metric on the manifold without "changing the shape" of the manifold under the chosen coordinatization. This allows one to vary and determine the components of the metric in terms of the chosen "fixed" coordinatization at each point. Other (non-"fixed") coordinatizations would be possible. So in this case, it has a particular calculational convenience based on a simplifying assumption on the choice of coordinates. I this picture it helps to group the "variable" expressions together. I still don't see that this motivates the concept of a standalone definition of a tensor density, though. —Quondum 14:30, 7 July 2014 (UTC)

Your presumptions are somewhat naive. The crux here is that in GR the metric is a dynamical quantity (i.e. you are dealing with a family of mathematically distinct (pseudo)-Riemannian. In order to make sense of the variational principle it is necessary to introduce an external frame of reference that you keep fixed while varying the metric. Of course, a coordinate frame is not the only way to do this (you could try it using non-coordinate frames). It is however it is by far the most intuitive because it allows you to use the ordinary Euler-Lagrange equations for fields.

I also do not buy your "it is only a computational tool" argument. This is true for many mathematical concepts. You could argue that derivatives are only computational tools.TR 19:50, 7 July 2014 (UTC)

In my naïvity, I cannot distinguish what you describe from what I was picturing. I can see that before a metric is determined, a coordinate-determined volume element has utility. But the utility of the whole theory of tensor densities is clearly eluding me. —Quondum 22:30, 7 July 2014 (UTC)

If there is exactly one (nonsingular) metric tensor field, then one can avoid using tensor densities (with some inconvenience). However, in alternatives to general relativity which have zero or two (or more) metric tensors, this is not possible. In those cases, tensor densities are essential because Lagrangian densities must be scalar densities of weight +1 and there is no unique metric which one can use to convert them to the "equivalent" ordinary scalar. JRSpriggs (talk) 03:40, 8 July 2014 (UTC)

That is an interesting complication. However, in such theories one probably cannot transfer the theory of tensor densities directly without some complications. In nonmetric and other cases, would n-forms not be the equivalent of choice, thus not using scalar densities at all? I've got a feeling (quite possibly wrong) that without at least this, integration becomes meaningless. —Quondum 05:29, 8 July 2014 (UTC)

A 4-form IS a scalar density. It is just a different way of talking about the same thing. Personally, I prefer to talk about tensor densities than about differential forms. It makes the transformation properties clearer. And while you think that my method is unnecessarily coordinate-dependent, I think that your method is unnecessarily dependent on abstract nonsense (category theory) and does not allow you the flexibility to easily mix tensor indices with densities (i.e. differential forms). Also, Maxwell's equations are more naturally represented using densities for the ρ, J, D, H, P and M variables. JRSpriggs (talk) 10:30, 8 July 2014 (UTC)

Ah, well, this is moderately a subtle point, and a case where normal tensors do exactly the right thing in terms of transformations, without the need to introduce additional mathematical machinery. It is called the Hodge dual. If one wants to represent an n-form (which is not a 1-density as defined in this article) as its Hodge dual (also a true tensor) so that one is working with a vector for convenience, then only the desire to use the same component values (rather than differing by the determinant of the basis transformation (or some mix), not the Jacobian of the coordinate transformation – the distinction is significant for non-holomorphic bases) pushes one into the concept of densities as defined in this article. I would personally work with the true Hodge dual, not the scaled version (being a purist), but from a practical perspective, simply rearranging the coordinates into a vector (or scalar etc.) form with a mental tag that these are from the n-form is far simpler theoretically, and does not need new rules to deal with the construction of densities. Use of the Hodge dual in the Maxwell's equations is natural; one needs both the quantity and its Hodge dual at various places. Any "density" other than the Hodge dual equivalent I'm guessing does not have a natural use anywhere. The article also makes no reference to the Hodge dual, but should since it is the natural interpretation for your example. —Quondum 13:28, 8 July 2014 (UTC)

Why would there be any particular problems with tensor densities with metric gone? Their definition doesn't make use of the metric. YohanN7 (talk) 11:34, 8 July 2014 (UTC)

If a tensor density is used in the place of the Hodge dual, what is really happening is that the dual was found using the Levi-Civita symbol instead of the volume form of the manifold. The volume form depends upon the metric, but the Levi-Civita symbol depends upon a choice of basis. The Levi-Civita symbol is non-tensorial but can be defined without a volume form or metric, whereas the volume form is tensorial but is best defined in terms of the metric. The definition of a density weighting is the function giving the proportionality between them, which is consequently only defined when both a basis is specified and there is a volume form. This severely limits the definition of densities in metric-independent contexts. —Quondum 15:27, 8 July 2014 (UTC)

Hm. I think I understand parts of what you mean, but not all. I don't see what the Hodge dual has to do with this, I'll take it as homework. I though the point with tensor densities is that we know what they are already and that those with density 1 can be used with the "usual" volume element (not form) right away. Needless to say, there seems to be little point in manufacturing a tensor density if a true tensor is available, except under unusual circumstances (e.g. variation of action). (The Levi-Civita symbol, b t w, is a tensor density, that much I know.)

A bit on terminology. Consider a n-form Ω, a true tensor. Then, in coordinates

${\begin{aligned}&\Omega =fdx^{1}\land dx^{2}\land \cdots \land dx^{n}&{\text{is a tensor (density of weight 0)}}\\&f&{\text{is a scalar density with weight 1}}\\&dx^{1}\land dx^{2}\land \cdots \land dx^{n}&{\text{is a tensor density of weight -1}}\end{aligned}}$

Yay or nay? YohanN7 (talk) 16:19, 8 July 2014 (UTC)

More hm. How do you compute the Hodge dual properly without a metric? An inner product will do, but then you have a metric, right? The Hodge dual uses the Levi-Civita tensor, (or the metric volume form for fields) and without inner product there is no Hodge dual, at least not from what I can glean from the Hodge dual article. I still don't see what Hodge duals have to do with this. YohanN7 (talk) 18:12, 8 July 2014 (UTC)

Looks good. In this example, as you say, if one has the 4-form Ω, why would one want to consider the other quantities? This perspective generalizes to pretty much every case. Also on terminology, let's distinguish between natural tensorial densities and the densities as defined in this article. I propose calling them "n-forms" and "w-densities" respectively.

The Levi-Civita symbol is, as you say, a ±1-density, the sign depending on whether the indices are raised or lowered. To facilitate clarity, I'd suggest moving away from a holonomic basis. The Jacobian of the change of coordinates and the tensor transformation matrix then diverge.

The Hodge dual is not defined on a manifold without a volume form, which one can derive from the metric tensor and an orientation. You don't need the Hodge dual in circumstances in which it is not defined ;-) It only arose due to the mention of Maxwell's equations, where the Levi-Civita symbol is sometimes used as a sloppy Hodge dual (4-current density is a 3-form, but it is often expressed as its Hodge dual for simplicity). Ignore it. —Quondum 18:44, 8 July 2014 (UTC)

Okay, but the question remains. Why would there be any particular problems with tensor densities with metric gone? These are given to us, e.g. like in the case of EM if I understand what JRSpriggs means correctly.

@JRSpriggs, I'd like to know too what you mean by a 4-form being a scalar density. Do you identify it with the f of above? YohanN7 (talk) 19:06, 8 July 2014 (UTC)

@Quondum, may I ask what you mean by "natural tensorial densities"? You exemplify by n-forms. And by w-densities you mean tensor densities? why not simply tensors and tensor densities (occasionally scalar densities)? But then we really need to know what JRSpriggs means by 4-form = scalar density or misunderstandings will result. YohanN7 (talk) 19:31, 8 July 2014 (UTC)

Above, JRSpriggs said "A 4-form IS a scalar density", and Sławomir has made a similar-sounding assertion. In a context such as this, such statements become very confusing. With suitable nomenclature, in my mind such a statement should be impossible. A 4-form is a 0-density and a scalar density is a 1-density. They are not the same mathematical object. By "natural tensorial densities" I mean tensors that may be intrinsically integrated over a region. Sorry, scratch the term n-form; I'm too unsure of what objects qualify, though I suspect that they are all k-forms, 0 ≤ k ≤ n, and I'm guessing that the natural dimensionality of the region of integration is k. —Quondum 22:44, 8 July 2014 (UTC)

Best is to avoid inventing terminology, "may intrinsically be integrated" only adds to the confusion. What exists suffices, and not only tensors can be integrated. B t w, a 1-density is not the same as a scalar density if I get it right. The "1" refers to the weight. Such tensor densities can be integrated invariantly using the "usual" volume element because it has weight -1, making the whole package invariant. I can only assume that the "n-forms are 1-densities" is standard jargon/notational abuse.YohanN7 (talk) 05:11, 9 July 2014 (UTC)

I am not inventing that particular terminology, I am taking it from Density on a manifold, which says: "... a density is a spatially varying quantity on a differentiable manifold which can be integrated in an intrinsic manner". Let's also be clear on something: in the sense of this definition, multiplying by anything before integrating does not fit the definition. You can integrate a volume form over a region of a manifold. You don't first multiply it with a volume element. No w-density with w ≠ 0 fits this definition.

Confusion is being caused by allowing sloppy abuse of terminology in this discussion. When I choose to distinguish between a density as defined in Density on a manifold (an article which unfortunately then proceeds to confuse the meanings), and a density as defined in Tensor density, I need language to draw attention to the distinction. Without that, this discussion is not worth pursuing.

At this juncture I'll note that Penrose, in The Road to Reality §12.5 and §12.7 gives an encapsulation of integration, p-forms and their integrability over p-surfaces, as well as defining a volume element. It may be noted that his "volume element" is an n-form that is part of the structure of the manifold, and his p-density is not what is defined in this article, it is something that can (on its own) be integrated. He says that p-forms are p-densities. He does not once refer to a metric, a weight, coordinates, or transformation properties. WP and Penrose could just as well be from different planets. —Quondum 06:13, 9 July 2014 (UTC)

Yes, you are right. There is confusion. Here is what Sławomir wrote earlier:

There are three ways of thinking of a skew (0,n) form. One is as a tensor (weight zero), one is as a scalar (a differential n-form of weight zero), and one is as a density (a scalar with weight 1, in our convention). All of these perspectives are connected by a "natural isomorphism" (that is, an isomorphism that does not require any choices to be made or additional structure).

It would be nice to see this detailed, both in terms of those natural isomorphisms and in terms of a breakdown of a 1-form into factors like I did above. YohanN7 (talk) 10:46, 9 July 2014 (UTC)

I have little issue with what Sławomir says: it makes sense, in the third instance provided that one accepts the definition of this article for "density", and that one must interpret a "represented as" as having been elided. My problem is that Penrose uses the terms "density", "p-density" and "volume element" to mean something incompatible with this usage. His is the abstract picture: that of differential geometry. I contend that Sławomir has used the term "density" in both (incompatible) senses without highlighting this, though I'd have to hunt out where. —Quondum 13:22, 9 July 2014 (UTC)

If you understand, then you can perhaps explain to me what he means? First problem: A n-form is a tensor (weight zero according to this article). This makes it impossible for it to have weight 1 as a density according to this article. YohanN7 (talk) 14:12, 9 July 2014 (UTC)

Penrose's meaning for "p-density" is a field that can be integrated over a region of a p-dimensional submanifold. It is therefore nothing to do with a weight, in the sense used here. You would want to be able to integrate the density of mass (a 3-density) in 3-space, for example, or a flux density (a 2-density) over a 2-surface in 3-space (for example, electric displacement field). Something that can be integrated along a curve (e.g. along a world line to yield proper time) is a 1-density. A scalar, by analogy, is a 0-density, and it can only be integrated on a 0-d manifold (i.e. summed over discrete points of the manifold). These p-densities are all p-forms, and hence all tensors. Weight is not a concept that enters into the picture in Penrose's version. —Quondum 15:18, 9 July 2014 (UTC)

I actually had Sławomir's quote in mind, not Penrose. YohanN7 (talk) 15:34, 9 July 2014 (UTC)

By implication (There are three ways of thinking of a skew (0,n) form), it is the same object that is being represented in each case. Sławomir, please feel free to shred my interpretation.

• One is as a tensor (weight zero)

  I interpret this to mean that one can represent it as a fully antisymmetric covariant tensor of order n, with nⁿ components in all but only one real degree of freedom, and satisfying the tensor transformation rules of such a tensor. This perspective will make sense those who like to think of tensors as a bunch of components.

• one is as a scalar (a differential n-form of weight zero)

  This perspective assumes that there is a natural duality, which is the case if there is a volume form as part of the defined structure of the manifold (what Penrose calls the volume element). I think of the implied equivalence between a scalar and a differential n-form of weight zero as meaning a scalar field that must be multiplied by the structural n-form. There is a natural isomorphism between this scalar field and the n-form; indeed, there is a direct mapping between the two: the Hodge dual.

• one is as a density (a scalar with weight 1, in our convention)

  This article chooses to define a volume element that depends on the choice of coordinatization with a holonomic basis, and is much like the Hodge dual. It non-tensorial, and unless an invariant structural volume element (what we've been calling the volume form) is also defined, other weightings are not defined.

—Quondum 17:10, 9 July 2014 (UTC)

This is what I think too, but the third item is shaky. It relies on the split I made (the only LaTex ITT) above of a 4-form into factors. It is rather inelegant and I doubt that a mathematician would call it "natural". They have bundles and smooth maps in mind, not a chop-up in coordinates. I have Penroses book b t w, so it's ok with me to refer to it. (I started working again on that thing for classical groups. Feel free to drop by, link is on my user page.) YohanN7 (talk) 17:43, 9 July 2014 (UTC)

To YohanN7, in response to your question above about Ω: Yes, you are essentially correct. If f is a scalar density (of weight +1), then

$\Omega =f\,dx^{1}\land dx^{2}\land \cdots \land dx^{n}$

is an invariant n-form. Otherwise, it is coordinate-dependent garbage.

If ω_αβγδ is an anti-symmetric covariant tensor (weight 0), then

$f={\frac {1}{24}}\epsilon ^{\alpha \beta \gamma \delta }\omega _{\alpha \beta \gamma \delta }=\omega _{1234}$

is a scalar density (weight +1).

In the notation system I prefer, only the coefficients are used, the basis vectors are discarded as redundant (and coordinate dependent) trash. JRSpriggs (talk) 19:27, 9 July 2014 (UTC)

Thank you for the concise answer. Now it all fits together as it should without any more guesswork on my part. YohanN7 (talk) 19:50, 9 July 2014 (UTC)

I just want to add to what JRSpriggs said. It is in this sense that I mean that a differential n-form (of weight zero) is a "scalar". A differential n-form has all of the dx's in it, so it's actually coordinate-invariant. That is what (in physics) distinguishes a scalar from a non-scalar. It's not a question of whether it is a "number" or "non-number" so much, which I think is how most mathematicians read the word "scalar". In particular, contrary to Quondum's interpretation, this does not require any kind of duality to be on hand. Sławomir Biały (talk) 20:56, 14 August 2014 (UTC)

So – do you agree that we have two colliding sets of terminology? And that we need to distinguish? —Quondum 23:41, 9 July 2014 (UTC)

Actually no. I have changed my mind for the tenth time in a few days now. Tensors, tensor densities, and densities are three distinct things with distinct names and distinct articles even if many conflate them more or less on purpose because the conflation has become second nature with time, like has been done here in this thread. Just as long as the articles are kept straight, there shouldn't be a problem. But I'm tired and can't think straight a t m. YohanN7 (talk) 00:45, 10 July 2014 (UTC)

The trickiest point may by be the density of Lee. These are ready to be invariantly integrated and are as such not tensor densities which need a volume element d⁴x attached before integration. But they are related. Densities contain a "factor" that is an even tensor density of weight +1. Conversely, if you have an even tensor density of weight +1, then attaching d⁴x to it yields a density. At least the formulas in Lee's book all indicate this, without exception, though I'm too lazy to bring out pencil and paper to compute myself so I may be wrong (and might go wrong even if I compute).

Oh yes, here's a real problem. "1-density" links to the density article. The tensor densities in the present article have (I think) also been referred to as 1-densities ITT. YohanN7 (talk) 01:21, 10 July 2014 (UTC)

Ah – a form of concession at last, at least from one person: density ≠ tensor density. Read the lede of Density on a manifold and see which of these each of the two paragraphs appears to be describing. —Quondum 04:02, 10 July 2014 (UTC)

On an oriented manifold a density is identical to a volume form. The same description as in the second paragraph in the lead (my browser tells me you spell lead wrong ) (I know, British English...) applies to true volume forms as well, apart from the absolute value. But I agree, the lead can be misinterpreted as saying that the integration isn't invariant. It is. The point is that, just like volume forms, there are two parts, one of which behaves like a tensor density of weight +1 and one of which behaves as a tensor density of weight -1. Compare the LaTeX ITT. There the f is a +1 tensor density and the wedge products of coordinate differentials is a -1 tensor density. The difference is that the density is endowed with an absolute value (making part of it an even tensor density, not an authentic tensor density) and the volume form not. YohanN7 (talk) 06:32, 10 July 2014 (UTC)

I give up. This is taking too many words to get anywhere. Already you are heading of in a direction on which there is no contention, suggesting that you've failed to realize where the problem is. —Quondum 13:13, 10 July 2014 (UTC)

Can someone who understands the subject please define it in the article?

The introductory section and the Definition section profoundly fail to explain what a tensor density is to the reader.

Can someone who actually understands this concept please rewrite these sections?

Note: Merely using the concepts is not the same as understanding them. To understand them it is necessary to grasp their mathematical underpinnings. To define a tensor by referring only to "how it transforms" and without ever saying what it is that does the transforming shows a complete lack of understanding of what tensors are.2600:1700:E1C0:F340:21E2:140D:C505:D61A (talk) 00:45, 12 November 2017 (UTC)