Starting with these definitions

  • A curve on a manifold $\mathcal M$ is a smooth (i.e. $C^{\infty}$) map $\sigma $ from some open interval $(-\epsilon,\epsilon)$ of the real line into $\mathcal M$

  • Two curves $\sigma_1$ and $\sigma_2$ are tangent at a point $p$ in $\mathcal M$ if (a) $\sigma_1(0) = \sigma_2(0) = p$ and (b) In some local coordinate system $(x^1,x^2,\ldots,x^m)$ around $p$, two curves are tangent in the usual sense as curves in $\mathbb R^m$, $$ (x^i \circ \sigma_1)'(0) = (x^i \circ \sigma_2)'(0) $$ here, $i=1,\ldots,m$

  • The tangent vector is defined as the equivalence class of curves in $\mathcal M$ where the equivalence relation between two curves is that they are tangent at point $p$.

  • The tangent space is $T_p\mathcal M$ to $\mathcal M$ at point $p$ is the set of all tangent vectors at point $p$

I am trying to prove the tangent space at the point $p$ in a manifold $\mathcal M$ is a vector space.

I'm starting with $v_1 \in T_p\mathcal M$, and $v_2 \in T_p\mathcal M$, and I have the following definitions $$ v_1 + v_2 := [\phi^{-1}\circ \ (\phi\ \circ \sigma_1 + \phi\ \circ \sigma_2 )] \\ r \ v_1 := [\phi^{-1}\circ \ (r \phi\ \circ \sigma_1)]\ \forall r \in \mathbb R $$

I want to show that $v_1 + v_2 \in T_p \mathcal M$ and $r \ v_1 \in T_p \mathcal M$

As $v_1 ,v_2 \in T_p\mathcal M$, then $$ \sigma_1(0) = \sigma_2(0) = p $$

Now, for $v_1 + v_2$ be a vector at $p$ , $\phi^{-1}\circ \ (\phi\ \circ \sigma_1 + \phi\ \circ \sigma_2 )(0) = p$ $$ \phi^{-1}\circ \ (\phi\ \circ \sigma_1 + \phi\ \circ \sigma_2 )(0) = \phi^{-1} \ (\phi\ ( \sigma_1(0)) + \phi\ (\sigma_2(0)) ) \\ = \phi^{-1}((\phi\ ( p) + \phi\ (p) )) \\ = \phi^{-1}( \ 2\phi\ ( p) ) \neq p $$

I can't prove the closure relations starting from the definitions, what I'm doing wrong?


The book I am following "Isham, Chris J. Modern differential geometry for physicists. Vol. 61. World Scientific, 1999.", takes a special chart $(U,\phi)$ such that $\phi(p) = \mathbf 0 \in \mathcal M$, using this choice

$$ \phi^{-1}\circ \ (\phi\ \circ \sigma_1 + \phi\ \circ \sigma_2 )(0) = \phi^{-1}( \ 2\phi\ ( p) ) = \phi^{-1}(0) = p $$ So, the closure is proven under addition. But this chart is a special choice. But the definitions hold for any charts around $p$, so another choice of charts should give the same result.

  • 1
    $\begingroup$ The definition of a manifold means it's always possible to find a chart that maps any point in the manifold to a defined point in the underlying vector space, so without loss of generality we can always choose a chart that maps $p$ to $0$. See Definition 2.1 in the book and figure 2.2 $\endgroup$
    – PeteBabe
    Aug 21, 2020 at 12:52
  • $\begingroup$ @PeteBabe Now, I get it, the functions $\phi : U \to \mathbb R^m$ is defined as per my need. So I can choose $\phi(p) = \mathbf 0$ $\endgroup$
    – Galilean
    Aug 21, 2020 at 14:56

1 Answer 1


Tangent vectors a $p \in M$ are equivalence classes of smooth curves $\sigma : (-\epsilon,\epsilon) \to M$ such that $\sigma(0) = p$ ("smooth curves in $M$ through $p$"). Here $\epsilon = \epsilon (\sigma)$ is a parameter which may vary from curve to curve. The equivalence relation is given by $\sigma_1 \sim \sigma_2$ if $(\phi \sigma_1)'(0) = (\phi \sigma_2)'(0)$ for some chart $\phi$ around $p$. It is easy to verify that $\sigma_1 \sim \sigma_2$ iff $(\phi \sigma_1)'(0) = (\phi \sigma_2)'(0)$ for all charts $\phi$ around $p$.

Given a smooth curve $\sigma : (-\epsilon,\epsilon) \to M$ through $p$, you can of course define $r \cdot \sigma : (-\epsilon/\lvert r \rvert,\epsilon/\lvert r \rvert) \to M, (r \cdot \sigma)(t) = \sigma (rt)$. Unfortunately there is no similar definition of $\sigma_1 + \sigma_2$ for curves $\sigma_i$ in $M$ trough $p$. You try to add them via the definition $$\sigma_1 + \sigma_2 = \phi^{-1}(\phi\sigma_1 + \phi \sigma_2).$$ This exploits the fact that the chart $\phi : U \to V \subset \mathbb R^n$ take values in $\mathbb R^n$, but in general it does not work because you cannot be sure that $\phi\sigma_1(t) + \phi \sigma_2(t) \in V$ for $\lvert t \rvert$ sufficiently small. Not even $\phi\sigma_1(0) + \phi \sigma_2(0) = \phi(p) + \phi(p) = 2\phi(p)$ is in general contained in $V$.

The solution is to consider only charts such that $\phi(p) = 0$. This can always be achieved if we replace an arbitrary chart $\phi$ by $T\phi$ where $T$ is the translation by $-\phi(p)$. The same holds for your definition of $r \cdot \sigma$.

Doing so, you will see that you get in fact the structure of a vector space on $T_p M$. Formally I suggest to proceed as follows:

  1. Show that $\phi_* : T_pM \to T_0V, \phi_*([\sigma]) = [\phi\sigma]$, is a bijection.

  2. Show that $T_0V$ becomes a vector space via $[\tau_1] + [\tau_2] = [\tau_1 + \tau_2]$ and $[r \cdot \tau] = [r \cdot \tau]$, where $(\tau_1 + \tau_2(t) = \tau_1(t)+ \tau_2(t)$ and $(r \cdot \tau)(t) = r \cdot \tau(t)$. Note that there always exist a maximal interval on which $\tau_1(t)+ \tau_2(t) \in V$ and $r \cdot \tau(t) \in V$; we take these intervals as the domains of $\tau_1 + \tau_2$ and $r \cdot \tau$. It is then easy to see that the map $\mathbb R^n \to T_0V, v \mapsto \tau_v$ with $\tau_v(t) = tv$, gives an isomorphism of vector spaces whichs shows that $\dim T_0V = n$.

  3. Observe that $\phi_*$ induces a unique structure of a vector space on $T_pM$ such that $\phi_*$ becomes an isomorphism of vector spaces.

  4. At first glance it seems that the vector space structure on $T_pM$ depends on the choice of $\phi$. The final step will therefore be to prove that any two charts $\phi_1, \phi_2$ around $p$ with $\phi_i(p) = 0$ produce the same vector space structure on $T_pM$.

  • $\begingroup$ wouldn't it be possible to, instead, define the sum of equivalence classes by sending $[\tau_1]+[\tau_2]=[t\mapsto \phi^{-1}(\phi(\tau_1(t))+\phi(\tau_2(t))-\phi(p))]$? In other words, using an affine transformation to sum the $\mathbb R^n$ vectors rather than just their sum. This would at least give a path mapping $0\mapsto p\in M$, as required $\endgroup$
    – glS
    Dec 23, 2021 at 17:16
  • $\begingroup$ @gls Yes, you can do that. It results in the same what I have done: You replace the chart $\phi$ by the translated chart $\psi = \phi - \phi(p)$ which satisfies $\psi(p) = 0$. Then you take the sum $\psi(\tau_1(t)) + \psi(\tau_2(t))$ and compute $\psi^{-1}(\psi(\tau_1(t)) + \psi(\tau_2(t)))$. $\endgroup$
    – Paul Frost
    Dec 24, 2021 at 8:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.