2

I am trying to understand the following statement from Lee's book introduction so smooth manifolds:

enter image description here

Why does the matrix of $g:TM\to T^{*}M$ have entries $(g_{ij})$ and not $g_{ij}E_{k}^{i}$?

As far as I am concerned I have the following:

Let $E_{i}$ be some cooridnate frame. Then I would say $g(E_{k})=g_{ij}E_{k}^{i}dx^{j}$, $E_{k}^{i}$ coordinate functions. Therefore the matrix of $g:TM\to T^{*}M$ has entries $g_{ij}E_{k}^{i}$. Since there is are no restrictions on $E_{i}$ I dont' see how I can get rid of the coefficients $E^{i}$.

What I get is $(g_{ij}E_{k}^{i})_{ij}=(g_{ij})_{ij}^{T}(E_{k}^{i})_{ik}$, which would be $(g_{ij})^{T}=(g_{ij})$, if $(E_{k}^{i})_{ik}=Id$.

Many thanks in advacne!

Hans
  • 902
  • 1
    Isn't $E_k^i$ just the Kronecker delta? – Gibbs Nov 06 '23 at 16:10
  • @Gibbs I don't think so, since there is no mentioning of $E_{i}$ being something like (0,...,i,...0), as far as I know – Hans Nov 06 '23 at 16:42
  • But the component of $E_k$ with respect to $E_i$ should be $\delta_{ik}$, shouldn't it? For example, in dimension $2$ you would have the frame ${E_1,E_2}$, and $E_1 = 1\cdot E_1+0\cdot E_2$, so $E_1^1 = 1$ and $E_1^2=0$, etc. – Gibbs Nov 06 '23 at 16:46
  • I am not sure I follow. When Lee writes coordinate frame, he refers to the basis $\partial/\partial x^i$ on $TM$ and its dual basis $dx^i$ on $T^*M$.

    The matrix of a a linear map depends on the chosen basis. For a coordinate frame, we have $g_{ij}:=g(\partial/\partial x^i,\partial/\partial x^j)$. If you choose a different frame $E_i=E^j_i \partial/\partial x^j$, your matrix in this new basis will not be $g_{ij}$ but rather $\tilde g_{ij}=g(E_i,E_j)$.

    – WishYouTheBest Nov 06 '23 at 16:47
  • To talk about the matrix of a linear transformation $V\to W$, you first of all need to specify bases for $V$ and $W$. Perhaps Lee should have specified that, given a local frame (basis) for $TX$, he was using the dual frame (basis) for $T^*X$. – Ted Shifrin Nov 06 '23 at 18:04
  • @Gibbs yes you are right. I don't know what I was thinking. – Hans Nov 06 '23 at 22:59

3 Answers3

3

Alright let me see if I can clear things up a little bit. Let $(M,g)$ be a Riemannian manifold, and say we have a coordinate chart $(U,\phi)$ with coordinates $x^i$. Then locally, in this coordinate chart we can write vector fields as: $$X=X^i\frac{\partial}{\partial x^i}$$ where the $\partial/\partial x^i$ are the coordinate frame on $U$, and the $X^i$ are smooth functions on $U$.

We also have a coordinate coframe given by the differential of the coordinate functions, i.e. any one form can be locally written in coordinates as: $$\omega=\omega_idx^i$$ where again $\omega_i$ are smooth functions on $U$, and as mentioned the $dx^i$ form a coordinate coframe. These $dx^i$ are dual to the coordinate frame in the sense that: $$dx^i\left(\frac{\partial}{\partial x^j}\right)=\delta^i_j$$ where $\delta^i_j$ is the constant function on $U$.

Now, a Riemannian metric $g$ is just a smoothly varying inner product on each tangent space, i.e. a global smooth section of $T^*M\otimes T^*M$ which is symmetric, non degenerate, and positive definite. We define the components of the metric tensor $g$ with respect to the coordinate frame on $U$ by: $$g\left(\frac{\partial}{\partial x^i},\frac{\partial}{\partial x^j}\right)=g_{ij}$$ These $g_{ij}$ are literally just smooth functions on $U$. Moreover, we have a coordinate frame for $T^*M\otimes T^*M$ is given by the set $\{dx^i\otimes dx^j\}$, so we can write $g$ locally as: $$g=g_{ij}dx^i\otimes dx^j$$ Indeed, if we plug in a pair of coordinate vectors $\partial/\partial x^k$, $\partial/\partial x^l$ we have: $$g\left(\frac{\partial}{\partial x^k},\frac{\partial}{\partial x^l}\right)=g_{ij}dx^i\left(\frac{\partial}{\partial x^k}\right)dx^j\left(\frac{\partial}{\partial x^l}\right)=g_{ij}\delta^i_k\delta^j_l=g_{kl}$$ In particular, if we vector fields $X=X^i\partial/\partial x^i$ and $Y=Y^j\partial/\partial x^j$, then: $$g(X,Y)=g_{ij}X^iY^j$$

We can now define a smooth vector bundle isomorphism $TM\rightarrow T^*M$ by taking an element $v_p\in T_pM$ and defining a covector in $T^*_pM$ by: $$g(v_p)(w_p)=g(v_p,w_p)$$ This is smooth because on local sections of $TM$ (i.e. local vector fields) we obtain that $g(X)$ is defined by: $$g(X)(Y)=g(X,Y)$$ which as we saw earlier was a sum of smooth functions. Now, in the coordinate frame, this must be given by: $$g(X)=g_{ij}X^idx^j$$ so the matrix in the coordinate frame is given by $g_{ij}$.

Now suppose we have a different frame $E_i$, then we can define new functions $\tilde{g}_{ij}$ by: $$g(E_i,E_j)=\tilde{g}_{ij}$$ With this new frame, we have an new coframe which we denote by $E^i$, so in this frame we have that: $$g=\tilde{g}_{ij}E^i\otimes E^j$$ If we want to write this in terms of our original $g_{ij}$, note that there is a matrix of smooth functions $B$ such that: $$E_i=B_i^j\frac{\partial}{\partial x^j}$$ then: $$ \begin{align} g(E_i,E_j)=&g\left(B_i^k\frac{\partial}{\partial x^k},B_j^l\frac{\partial}{\partial x^l}\right)\\ =&B_i^kB_j^lg\left(\frac{\partial}{\partial x^k},\frac{\partial}{\partial x^l}\right)\\ =&B_i^kB_j^lg_{kl} \end{align}$$ hence we have that: $$\tilde{g}_{ij}E^i\otimes E^j=B_i^kB_j^lg_{kl} dx^i\otimes dx^j$$ so: $$g(X^iE_i)=\tilde{g}_{ij}X^iE^j=B_i^kB_j^lg_{kl}X^idx^j$$ So to sum up, it doesn't matter what frame you use, you just have to be able to switch between them when the need arises. Oftentimes it is convenient to use coordinate frames, other times it easier to use orthonormal frames. I suggest you derive for yourself what the transformation law is locally between two coordinate frames. When working in coordinates, or locally in general things can get confusing sometimes I find, so it's best to work out a lot of this on your own to get a clear idea of what is going on.

Chris
  • 2,921
  • 1
    Thank you very much for your long answer! I tried to find an answer, which I posted below. How do you get from $E^{i}\otimes E^{j}$ to $dx^{i}\otimes dx^{j}$. I get additional terms from the transformation of the $E^{i},E^{j}$. I am using https://math.stackexchange.com/questions/3947762/transformation-law-for-metric-tensor – Hans Nov 07 '23 at 13:39
  • @Hans you get the transformation as you do for all covariant objects. If $A$ is an automorphism on a vector space $V\rightarrow V$, then $A$ acts on covariant objects via the pullback. I.e. in the case of a metric tensor you have $A^g$ is defined implicitly by $(A^g)(v,w)=g(Av,Aw)$. – Chris Nov 07 '23 at 16:55
  • I am afraid I don't see how the pull-back is related to changeing coordinates and I can not find any book that adresses this. Would using the formula in https://math.stackexchange.com/questions/3796079/pull-back-of-a-smooth-covariant-tensor-field-is-smooth/3796109#3796109 with F=Id give the same answer? – Hans Nov 08 '23 at 15:34
  • Comparing your result with $\enspace g_{ij}(y) , \frac{\partial y^i}{\partial x^k} , \frac{\partial y^j}{\partial x^{\ell}}$, I still wonder what happens to the terms $\frac{\partial y^i}{\partial x^k} , \frac{\partial y^j}{\partial x^{\ell}}$. https://math.stackexchange.com/questions/3947762/transformation-law-for-metric-tensor – Hans Nov 08 '23 at 17:00
  • @Hans This is exactly what I meant when I said you should check this result on coordinate transformations. In this case the matrix of functions is the jacobian matrix of the coordinate transformation, i.e. $B^i_k=\frac{\partial y^i}{\partial x_k}$. – Chris Nov 08 '23 at 17:32
  • Oh okay! So changing coordinates as it is done in section 7,p.50, of https://kconrad.math.uconn.edu/blurbs/linmultialg/tensorprod.pdf can be shown to be done using the pull-back, i.e. changing coordinates is a special case of the pull-back? What would the map F be? – Hans Nov 09 '23 at 11:15
  • @Hans I suggest you go to the simple case. Take a vector $v=v^i\frac{\partial }{\partial x_i}$ in $T_pM$. Use a different coordinate system $\partial/\partial y^i$, what is the transformation? Well it is the Jacobian, so $v=v^i\frac{\partial y^j}{\partial x^i}\frac{\partial}{\partial y^j}$. Now take a one form, $\omega=\omega_idy^i$, if we want to rewrite this in the $x$ coordinate frame, then we do $\omega(\partial/\partial x^i)=\omega(\frac{\partial y^j}{\partial x^i}\frac{\partial}{\partial y^j})=\frac{\partial y^j}{\partial x^i}$, so $\omega=\omega_i\frac{\partial y^i}{\partial x^j}dx^j$. – Chris Nov 09 '23 at 15:05
  • I already did that. I plugged everything in and the used multilinearity. I still am not sure how the pull-back formula ties in. If I look at $F^A = F^(A_{j_1 \dots j_k}) \cdot F^(dy^{j_1}) \otimes \cdots \otimes F^(dy^{j_k}) = \left[(A_{j_1\dots j_k} \circ F) \cdot \dfrac{\partial (y^{j_1}\circ F)}{\partial x^{i_1}} \cdots \dfrac{\partial (y^{j_k}\circ F)}{\partial x^{i_k}} \right]dx^{i_1}\otimes \cdot\otimes dx^{i_k}$. I guess F=Id? – Hans Nov 10 '23 at 11:27
  • @Hans when looking at the change of coordinates you don’t need to use that formula for the pull back. Just rewrite your vectors in a different frame using the jacobian and you’ll get what the new coefficients of the covariant tensor field are in the corresponding dual frame. – Chris Nov 10 '23 at 18:11
0

Hope this works:

Let $E_{i}$ be a frame and $\epsilon^{i}$ the dual frame. $\tilde{g}[X]=g_{ij}X^{i}\epsilon^{i}$.

Therefore $\tilde{g}[E_{k}]=g_{ij}E_{k}^{i}\epsilon^{j}=g_{ij}\delta_{ik}\epsilon^{j}=g_{kj}\epsilon^{j}$, hence the k-th row of $(a_{ij})$ consists of $g_{kj}$, which is the k-th column of $(g_{ij})$, because it is symmetric, i.e. $(a_{ij})=(g_{ij})^{T}$.

we could also say that $g=g_{ij}\epsilon^{i}\otimes \epsilon^{j}\cong g_{ij}\epsilon^{i}(X)*\epsilon^{j}(Y)=:\tilde{g}[X](Y)$. Since $g_{ij}\epsilon^{i}\otimes \epsilon^{j}$ is an abstract object that a priori doesn't act on anything.

By the above $\tilde{g}[E_{i}](Y)=g(E_{i},Y) = g_{ij}\epsilon^{i}(E_{i})*\epsilon^{j}(Y)=g_{ij}\epsilon^{j}(Y)\cong(g_{ij}\epsilon^{j})(Y)$ ( I guess it is overly pedantic)

In other words the i-th basis vector gets mapped to $g_{ij}\epsilon^{j}$. Thus, the ith-row of $(a_{ij})$ consists of $g_{i1},...,g_{in}$, which is the ith column of $(g_{ij})$,i.e. $(a_{ij})=(g_{ij})^{T}$

Hans
  • 902
  • What is the object $E^i_k$? You shouldn't use $dx^i$ as a dual frame unless the frame is the coordinate frame, as otherwise they are not dual to one another. – Chris Nov 06 '23 at 23:11
  • It is the coordinate function. I am not sure why I shoudn't use $dx^{i}$. I statet that it is corresponding dual frame, but okay, could have named it $\epsilon^{i}$ as well. – Hans Nov 07 '23 at 09:02
  • if it is the coordinate function it's probably best to write the frame as $\partial/\partial x^i$. – Chris Nov 07 '23 at 16:52
0

Let ${M}$ be a smooth metric manifold. Then, the covariant metric tensor field ${g}\in\Gamma^{\infty}({M},\mathrm{Sym}^{2}({T}^{\star}{M}))$ induces an isomorphism \begin{align*} \mathrm{flat}:\Gamma^{\infty}({M},{T}{M})\rightarrow\Gamma^{\infty}({M},{T}^{\star}{M}){\,}{,} \end{align*} that is defined within in a coordinate chart through \begin{align*} \mathrm{flat}({A})_{\mu}={g}_{\mu\nu}{A}^{\nu}{\,}{,}{\,\,}\mathrm{where}{\,\,}{A}\in\Gamma^{\infty}({M},{T}{M}){\,}{.} \end{align*} Similarly, the contravariant metric tensor field ${G}\in\Gamma^{\infty}({M},\mathrm{Sym}^{2}({T}{M}))$ induces an isomorphism \begin{align*} \mathrm{sharp}:\Gamma^{\infty}({M},{T}^{\star}{M})\rightarrow\Gamma^{\infty}({M},{T}{M}){\,}{,} \end{align*} that is defined within in a coordinate chart through \begin{align*} \mathrm{sharp}(\omega)^{\mu}={G}^{\mu\nu}\omega_{\nu}{\,}{,}{\,\,}\mathrm{where}{\,\,}\omega\in\Gamma^{\infty}({M},{T}^{\star}{M}){\,}{.} \end{align*} Now, the contravariant metric tensor is often referred to as the "inverse" of the covariant metric tensor in the literature, since \begin{align*} &{A}^{\mu}=\mathrm{sharp}(\mathrm{flat}({A}))^{\mu}={G}^{\mu\nu}{g}_{\nu\varrho}{A}^{\varrho}{\quad}\forall{A}\in\Gamma^{\infty}({M},{T}{M}){\quad}\mathrm{and}\\[0.2em] &\omega_{\mu}=\mathrm{flat}(\mathrm{sharp}(\omega))_{\mu}={g}_{\mu\nu}{G}^{\nu\varrho}\omega_{\varrho}{\quad}\forall\omega\in\Gamma^{\infty}({M},{T}^{\star}{M}){\,}{,} \end{align*} from which, one can follow that ${g}_{\mu\varrho}{G}^{\varrho\nu}=\delta^{\nu}_{\mu}$. It also follows that \begin{align*} {g}\langle\mathrm{sharp}(\omega),\mathrm{sharp}(\eta)\rangle&={G}\langle\omega,\eta\rangle{\quad}\forall\omega,\eta\in\Gamma^{\infty}({M},{T}^{\star}{M}){\,}{,}\\[0.5em] {G}\langle\mathrm{flat}({X}),\mathrm{flat}({Y})\rangle&={g}\langle{X},{Y}\rangle{\quad}\forall{X},{Y}\in\Gamma^{\infty}({M},{T}{M}){\,}{.} \end{align*}

DeVoyd
  • 96