Yet more elementary questions about the tangent space

In summary, the author is trying to learn differential geometry concepts from old-style index games to update his knowledge from years ago and is having difficulty understanding the concepts. He is looking for conceptual help, not rigorous explanations.
  • #1
orion
93
2
I am trying to self-study some concepts in differential geometry to try to update my knowledge from the old-style index games to something more meaningful. I know that there are many threads that have in some way addressed this, but I am still not understanding it completely. I'm new to this, and I could be way off base so please bear with me as I am feeling my way in the dark. I'm looking for more conceptual help so rigor is not that important.

Consider a tangent space of a manifold M at point p. For simplicity, assume TpM is Euclidean and of dimension 2. I understand that {∂i} forms a basis of TpM for some coordinate chart around p.

Question 1: What does the dot product between two vectors in this space look like when the basis vectors are partial derivatives?

Question 2: How can I show that the ∂i form a basis?

Question 3: According to what I have read, the dxi form a basis of Tp*M, the dual of TpM. The Euclidean metric (for the two dimensional case) is expressed differentially as ds2=dx2+dy2.

This looks like length is formed from covectors in Tp*M and not vectors. Why is this?

Question 4: Consider the following definition:

dxi(∂j) = ∂jxi

I realize that this is a definition, but is there any background to this definition? Where did it come from? (I have a suspicion that the directional derivative has something to do with it, but I'm not there yet.)
 
Physics news on Phys.org
  • #2
orion said:
Question 1: What does the dot product between two vectors in this space look like when the basis vectors are partial derivatives?

There are no standard basis vector for tangent spaces in general so there usually is no natural choice of inner product. Once you have one fixed, however, you just need information about <∂i,∂j> to compute.

Question 2: How can I show that the ∂i form a basis?

This is a local question so it suffices to assume M = Rn. Now clearly the ∂i are linearly independent. So only the spanning condition remains. For this assume D is an arbitrary derivation at p and let vi = D(xi). The claim now is that D = ∑vii. To see this note that every function f:RnR has an expansion of the form:

f(x) = f(p) + ∑∂if(p)(xi-pi) + ∑gi(x)(xi-pi)

where gi(p) = 0. Now just apply D to this expansion and simplify. The result is Df = ∑viif(p).

Question 3: According to what I have read, the dxi form a basis of Tp*M, the dual of TpM. The Euclidean metric (for the two dimensional case) is expressed differentially as ds2=dx2+dy2.

This looks like length is formed from covectors in Tp*M and not vectors. Why is this?

This is essentially because we can integrate covectors. The same is true in ordinary calculus. In that context, anytime you wanted to integrate a vector field, you first used the natural Riemannian metric (a consistent choice of inner product between all the tangent spaces) to turn the vector field into a covector field and integrated from there.

Question 4: Consider the following definition:

dxi(∂j) = ∂jxi

I realize that this is a definition, but is there any background to this definition? Where did it come from? (I have a suspicion that the directional derivative has something to do with it, but I'm not there yet.)

I cannot speak authoritatively about the historical reasons for the definition. I can say, however, that we want the natural pairing to between vectors and covectors to be nice and that this definition accomplishes just that.
 
  • #3
jgens said:
There are no standard basis vector for tangent spaces in general so there usually is no natural choice of inner product. Once you have one fixed, however, you just need information about <∂i,∂j> to compute.

Thank you very much for your reply and I really do appreciate it though I must admit that a lot of what you wrote went over my head. What I'm trying to do is to relate these concepts from differential geometry to what I already know from elementary vector analysis. What usually aids in my understanding is to consider a particular, concrete case and then later generalize it to higher levels of abstraction.

I'd like to consider a sphere with a tangent plane at a particular point p. I assume the tangent plane is Euclidean with the standard Euclidean metric. (I am thinking of the Earth and a local street map lying in the tangent space.) Assume that I have placed some type of coordinates around p. Now, I want to take two vectors from that tangent space and form a dot product. I would like to understand how to relate this case to base vectors that are partial derivatives and what such a dot product would look like. It seems to me that the approach from differential geometry should reduce somehow to the familiar approach from vector analysis.

Thanks again for your reply.
 
  • #4
orion said:
I'd like to consider a sphere with a tangent plane at a particular point p. I assume the tangent plane is Euclidean with the standard Euclidean metric.

Here is the problem. The expression for the standard Euclidean metric depends heavily on our choice of basis. For Euclidean spaces we have a natural choice of basis so this poses no problem. For arbitrary tangent spaces we have no natural choices however. Each chart around p is going to give us a new set of basis vectors and there is no reason to prefer one of these over the other.

Now, I want to take two vectors from that tangent space and form a dot product.

The dot product refers to a very specific inner product on Euclidean space. So the relevant concept here is an inner product. You form these guys in exactly the same ways you formed them in vector analysis.
 
  • #5
I guess what I am driving at is how can it ever be possible that ∂ij = δij assuming a basis of ∂k and a tangent space that is Euclidean. In other words, I want to see how base vectors ∂k can be related to the basis B = {(1,0), (0,1)} of a Euclidean plane or how we can get the basis B from the partial derivatives.

I want to take the abstract formalism of differential geometry and apply it to a specific, well-known case.
 
  • #6
orion said:
I guess what I am driving at is how can it ever be possible that ∂ij = δij assuming a basis of ∂k and a tangent space that is Euclidean.

You can simply define your inner product that way. Just beware that with a difference choice of coordinates your inner product will no longer have this nice form.
 
  • #7
Even in Euclidean space, you don't need to use coordinate axes that are perpendicular to one another, and the variations of the coordinate values along the coordinate axes to not have to be equal to actual distances in space. Consider a set of affine coordinate axes that are straight lines and make a fixed angle with one another. Before you move on to arbitrary tangent spaces, first work everything out and understand it for Euclidean space. After you have finished with affine coordinate axes, consider orthogonal curvilinear coordinate axes. Then, finally consider arbitrary curvilinear coordinate systems. This should solidify your understanding.
 
  • #8
One of the issues here is that tangent spaces at different points of ## \mathbb R^n ## with basis vectors ## e_i ## and with the standard rectangular system, are naturally-isomorphic to each other; naturally isomorphic meaning that there is an iso. between them that does not depend on the choice of basis. Specifically, tangent spaces at any two points a,b are just translates of the tangent space at the origin, and this translation does not affect the essential vector space structure. This natural isomorphism makes the process of differentiating very simple; just translate the vector at (x+h) to that at x . This is not the case with most manifolds; while any two tangent spaces are isomorphic ( being of the same dimension) , they are not naturally-so. To make up for that, you make a choice of isomorphism between them by using a connection.

Re the issue of the basis, you may also want to consider the case where the manifold is embedded in some ambient ## \mathbb R^n ## . Then you can use the curve definition of tangent vectors; jgens covered the case when the manifold is stand-alone, where a tangent spaces makes no sense, since the tangent space at p is not part of the manifold ( outside of the point p, of course ). Then you can show that every derivation is a directional derivative, so that the basis you mentioned spans every derivation.

For question 3, notice that your standard Riemann integral is the integral of the n-form ## fdx_!dx_2..dx_n##. If you know some geometric algebra, i.e., the algebra/geometry associated to wedge products, I think will help make sense of the integration of forms.
 
  • #9
WWGD said:
Re the issue of the basis, you may also want to consider the case where the manifold is embedded in some ambient ## \mathbb R^n ## . Then you can use the curve definition of tangent vectors

Just a quick note: Defining tangent vectors via curves works even when the manifold is not embedded in some ambient Euclidean space. This is actually quite an important fact since often times we want to consider derivatives and vector fields associated to curves.
 
  • #10
jgens said:
Just a quick note: Defining tangent vectors via curves works even when the manifold is not embedded in some ambient Euclidean space. This is actually quite an important fact since often times we want to consider derivatives and vector fields associated to curves.

But, how do you define the vector fields? Their range is not necessarily in the manifold itself.
 
  • #11
A vector field is just a smooth section M→TM. Sure the range is not the manifold itself, but nothing about the definition requires that M be embedded in some Euclidean space either.
 
  • #12
Ah, I see, I misunderstood ( misunderestimated?) your post.
 
  • #13
Thanks everyone for the replies. I appreciate the discussion.

I'm not sure, but I think that I have found the answer in Introduction to Smooth Manifolds by John M. Lee, page 53 (as part of a proof). If we assume a rectangular coordinate system and allow the ∂i vectors to operate on the coordinates we can obtain the three standard basis vectors, ek because ∂ixj = 0 except when i=j when it is equal to 1.
 
  • #14
orion said:
If we assume a rectangular coordinate system and allow the ∂i vectors to operate on the coordinates we can obtain the three standard basis vectors, ek because ∂ixj = 0 except when i=j when it is equal to 1.

As has been mentioned several times already. Yes there is an isomorphism of vector spaces taking ∂k to ek. From there one can define <∂i,∂j> = δij. Just beware: in another choice of local coordinates this inner product will look very different.
 
  • #15
jgens said:
As has been mentioned several times already. Yes there is an isomorphism of vector spaces taking ∂k to ek. From there one can define <∂i,∂j> = δij. Just beware: in another choice of local coordinates this inner product will look very different.

Ah, yes, I'm sorry. My problem was actually seeing the isomorphism.

I still have a question involving inner products, but I need to think on it a bit.
 
Last edited:
  • #16
jgens said:
As has been mentioned several times already. Yes there is an isomorphism of vector spaces taking ∂k to ek. From there one can define <∂i,∂j> = δij. Just beware: in another choice of local coordinates this inner product will look very different.

I've given it more thought, and my question is that I don't want to just define <∂i,∂j> = δij and leave it at that, but I want to show that it is true in the same way that one can demonstrate that <ej,ek>= δjk for an suitable local coordinate system.

In other words, I want to show that it is true by the action of the differential operators.
 
  • #17
The problem is without a Riemannian metric on your manifold there is no natural choice of inner product on the tangent spaces. So you have to define an inner product first, before you can show anything! If simply setting <∂i,∂j> = δij is unsatisfactory, then how do you propose defining it?
 
  • #18
I am assuming a Riemannian metric. I think that the problem was that I was mistaking ∂ij to be equal to ∂2/∂xij when it is ∂if∂jf. If we let f be the standard coordinate functions, we have ∂ixkjxk = δikδjk = δij.
 
  • #19
orion said:
I am assuming a Riemannian metric.

In that case to each coordinate chart corresponds a symmetric matrix (gij) where gij = <∂i,∂j> and one need only apply elementary linear algebra to diagonalize this matrix.

An important note: The resulting local vector fields are not guaranteed to be the "partial derivatives" corresponding to any coordinate system. In most applications this introduces no real problems, since local frames with the desired properties are sufficient. But if you wanted these frames to be induced from a choice of coordinates, then you are out of luck.
 
  • #20
jgens said:
In that case to each coordinate chart corresponds a symmetric matrix (gij) where gij = <∂i,∂j> and one need only apply elementary linear algebra to diagonalize this matrix.

This is the point of confusion for me. I don't understand how one can use the metric to "define" <∂i,∂j> when the ∂k are operators. It seems like it must be possible to obtain the metric elements from letting the operators operate on something. That was the point of my last post.
 
  • #21
orion said:
I don't understand how one can use the metric to "define" <∂i,∂j> when the ∂k are operators.

The metric is a map <-,->p:TMp x TMpR. Since ∂i and ∂j are elements of TMp it makes complete sense to consider <∂i,∂j>.

It seems like it must be possible to obtain the metric elements from letting the operators operate on something.

This is probably not feasible. You can endow your manifold with two very different Riemannian metrics and these operators will remain unchanged. So it is probably too much to hope for this approach to work out.
 
  • #22
I need to backtrack a bit because I am still trying to come to grips with operators being a basis and having their inner product defined by the metric. Well, no, I'm ok with having the inner product defined by the metric. But what really is troubling me is whether I should I consider the ∂k as operating on something?

I will tell you how my mind wants to think of this, and I hope you will point out the error in my thinking. My mind wants to say that the ∂k operates on coordinate functions and out pops a basis like we are familiar with from elementary linear algebra. The confusion right now for me is whether I should consider the ∂k as operating on something (what?) or whether I should think of it mostly as a symbol like ej. The problem is that I can express the ej in a particular representation (eg. a column vector) if I must.
 
  • #23
orion said:
But what really is troubling me is whether I should I consider the ∂k as operating on something?

Yes. They operate on maps f:M→R. The Riemannian metric is an additional structure, on top of this operator property, so it would be folly to assume it is somehow determined by the way these ∂k act on functions.

My mind wants to say that the ∂k operates on coordinate functions and out pops a basis like we are familiar with from elementary linear algebra.

The coordinate functions induce the basis. Not the action of ∂k on the coordinate functions.

The confusion right now for me is whether I should consider the ∂k as operating on something (what?) or whether I should think of it mostly as a symbol like ej.

Both. As mentioned before the ∂k act on functions and at times understanding this action is crucial. Other times, like the case at hand, the operator property is unimportant. The important fact is that these guys form a basis for the vector space TMp.

The problem is that I can express the ej in a particular representation (eg. a column vector) if I must.

You can do the same with the ∂k. In general this will not help you much, but it can be done.
 
Last edited:
  • #24
Thanks. I'm getting there, but I'm not quite there yet. Can you recommend any good books where this is covered at an elementary (well, as elementary as possible) level?
 
  • #25
The book Introduction to Smooth Manifolds by Lee is probably the most easy-going exposition about this material I know. His Riemannian Manifolds book is also good (and considerably more on-point for this topic) but it assumes quite a bit more.
 
  • #26
Thanks, I do appreciate it.
 

1. What is the tangent space?

The tangent space is a mathematical concept that describes the local behavior of a surface or curve at a specific point. It is a vector space that contains all possible tangent vectors to the surface or curve at that point.

2. How is the tangent space related to the concept of derivative?

The derivative of a function at a point can be thought of as the slope of the tangent line to the function at that point. In a similar way, the tangent space can be thought of as containing all possible tangent lines to a surface or curve at a specific point, making it closely related to the concept of derivative.

3. Can you explain the difference between the tangent space and the normal space?

The tangent space is a vector space that contains all possible tangent vectors to a surface or curve at a specific point. The normal space, on the other hand, is a vector space that contains all possible normal vectors to the surface or curve at a specific point. While tangent vectors are perpendicular to the surface or curve at a point, normal vectors are perpendicular to the tangent space at that point.

4. How is the tangent space useful in understanding the local behavior of a surface or curve?

The tangent space allows us to approximate the behavior of a surface or curve at a specific point using tangent vectors. This can help us understand the direction and rate of change of the surface or curve at that point, as well as its curvature and other important properties.

5. Can the concept of tangent space be extended to higher dimensions?

Yes, the concept of tangent space can be extended to higher dimensions. In fact, it is a fundamental concept in differential geometry, which deals with the study of curves and surfaces in higher dimensions. In higher dimensions, the tangent space is a vector space that contains all possible tangent vectors to the surface or curve at a specific point in that higher-dimensional space.

Similar threads

  • Differential Geometry
Replies
21
Views
552
Replies
4
Views
1K
  • Differential Geometry
2
Replies
55
Views
7K
  • Differential Geometry
Replies
4
Views
2K
  • Differential Geometry
Replies
12
Views
3K
  • Differential Geometry
Replies
10
Views
639
  • Differential Geometry
Replies
6
Views
3K
  • Differential Geometry
Replies
6
Views
6K
Replies
9
Views
3K
  • Differential Geometry
Replies
6
Views
1K
Back
Top