enter image description here

Question 1

In calculus, we visualize the tangent space $T_p(\mathbb R^n)$ at $p$ in $\mathbb R^n$ as the vector space of all arrows emanating from $p$. What is the motivation of creating of $T^*_p(\mathbb R^n)?$ How can we visualize covectors?

Question 2 How do I prove the sets $T^*_p(\mathbb R^n)$ are all disjoint? where $p\in U$

Considering $p,q\in U$ and $p\neq q$, suppose f is a non zero linear functional and $f \in T^*_p(\mathbb R^n)\cap T^*_q(\mathbb R^n)\implies f\in T^*_p(\mathbb R^n)$ and $f\in T^*_p(\mathbb R^n)$.How do I proceed further?. Can you please help me?

  • $\begingroup$ they are just linear functionals $T_p{\Bbb R^n}\to{\Bbb R}$ $\endgroup$ – janmarqz Jul 7 '18 at 15:30
  • $\begingroup$ What about the geometrical meaning? $\endgroup$ – Unknown x Jul 7 '18 at 15:32
  • $\begingroup$ They are employed to decompose the domain into level sets, which are lines $\endgroup$ – janmarqz Jul 7 '18 at 15:33
  • $\begingroup$ Can you explain? How is it coming? I am not able to visualize :( $\endgroup$ – Unknown x Jul 7 '18 at 15:39
  • $\begingroup$ my previous comment is a little wrong in the sense that I was thinking level set for $n=2$. For dimension $3$ the level sets are plane, for dimension $4$ hyperplanes an so on... In a while I will widen my comment into an answer. $\endgroup$ – janmarqz Jul 7 '18 at 15:55

I only given an answer to the first question. My answer is an addition to the answer given by janmarqz. Basically I just provide some plots for his second point.


I like the visualization given in V. Arnolds, Mathematical Methods of Classical Mechanics.

Suppose we have a covector $\omega \in T^*_0\mathbb{R}^2$, we then draw the hyperplanes $\dots, \omega^{-1}(\{-1\}),\omega^{-1}(\{0\}),\omega^{-1}(\{1\}), \omega^{-1}(\{2\}), \dots$ (Let's call them the contour lines of $\omega$.)

If we now want to see how the covector acts on a vector $\mathbf{v} \in T_0\mathbb{R}^2$, we just draw the arrow representing $\mathbf v$ and look how many contour lines we hit.

Below I have drawn this method for a basis $\{\omega, \eta\}$ of $T^*_0\mathbb{R}^2$.

Covectors in two dimensions

Why don't we draw arrows for covectors?

A typical difference between tangent vectors and cotangent vectors is their transformation behavior!

Let's use the linear transformation $\Phi:\mathbb R^2 \to \mathbb R^2: (x,y)^T \mapsto (x,2 y)^T$.

Vectors transform via the differential $T\Phi: T \mathbb R^2 \to T \mathbb R^2$ (sometimes also denoted by $D\Phi$). In our case this is just the Jacobian of $\Phi$ i.e. $\mathbf v \mapsto \begin{pmatrix}1&0\\0 & 2\end{pmatrix} \mathbf v$.

Covectors transform differently! An example is the gradient of a smooth function $f:\mathbb R^2 \to\mathbb R$. In classical calculus, we have see, that $\nabla f$ is always orthogonal to the contour-lines of $f$! But if you apply the transformation $\Phi$ and transform $\nabla f$ like a vector, it will not be orthogonal anymore!

enter image description here

But, if we visualise covectors by their contour-lines, they behave correctly! For example, if we draw the contour-lines of $df$, they will stay tangent to the contour-lines of $f$ after applying a transformation to the visualization, which is the correct behavior!

This is one of the reasons, why we treat vectors and covectors differently. And also the reason why we draw contour-lines of covectors instead of arrows.

Higher Dimensions

The reasoning above is still valid in higher dimensions, here the contour-lines become hyperplanes instead. (The example with the gradient is only plausible for Riemannian manifolds, but the example should just give a feeling how covectors are different than vectors. In general covectors transform according to the inverse of the adjoint of $T\Phi$, i.e. $\omega \mapsto ((T\Phi)^*)^{-1}$. )

  • $\begingroup$ What is $M$ here? $\endgroup$ – Unknown x Jul 8 '18 at 17:39
  • $\begingroup$ +1 excellent blowup! $\endgroup$ – janmarqz Jul 8 '18 at 18:56
  • 1
    $\begingroup$ $M = \mathbb R^2$, which plays the role of the manifold in this example. Note, that in this special case the manifold $\mathbb R^2$ and the 'attached' vector space $T_0 \mathbb R^2$ can be identified (also in a exact way), which allows use to draw contour lines of $f$ and contour lines of $df$ on top of each other. $\endgroup$ – Steffen Plunder Jul 8 '18 at 20:00
  • $\begingroup$ @janmarqz I'm glad that you like it. Thanks! $\endgroup$ – Steffen Plunder Jul 8 '18 at 20:02

Question 2 is easy: We already consider the tangent spaces $T_p{\mathbb R}^n$ as disjoint: Each point $p$ has its own tangent space. Of course all these tangent spaces are carbon copies of one and the same "model space" ${\mathbb R}^n$. (If we would not make the spaces $T_p{\mathbb R}^n$ disjoint a vector field would become a hedgehog!) Now each $T_p{\mathbb R}^n$ has its own dual space $\bigl(T_p{\mathbb R}^n\bigr)^*$, and these dual spaces are considered to be disjoint as well. This is matter of definition; no proof is needed.

Question 1 is more difficult. Number 1 there is no universally agreed way to "draw" a cotangent vector in a picture showing the vector space $V$, e.g., $T_p{\mathbb R}^n$. Sometimes you have a scalar product in $V$. It is then possible to identify each $\phi\in V^*$ with a vector $a_\phi\in V$, such that the identity $$\phi(x)=a_\phi\cdot x\qquad\forall\>x\in V$$ holds. E.g., if $\phi:=df(p)$ is the differential of a scalar function $f$ at $p$ then $a_\phi=\nabla f(p)$.

If you want an "intuitive" description of covectors $F\in V^*$ think of physical forces. Physicists cannot explain what a force is, for the same reasons that you have no intuitive feeling for covectors. They then draw the vector $a_F\in V$ and call it the force vector. In reality an (e.g., gravitational) force acting at $0\in V$ is a covector $F\in V^*$. If you want to move along the segment $\sigma$ connecting $0\in V$ with the point $x\in V$ then you have to perform the work $W_\sigma=F.x$ (or get back $W_\sigma$, depending on the sign conventions), whereby the $.$ denotes evaluation.

  • $\begingroup$ could you give your opinion about using level sets to attach some geometry to a 1-form, a covector at each position? $\endgroup$ – janmarqz Jul 7 '18 at 20:43
  • $\begingroup$ Are we considering $(T_p\mathbb R^n)^*$ disjoint or Can we able to rigorously prove it? If we consider functions from $(T_p\mathbb R^n)^*$ and $(T_q\mathbb R^n)^*$. the domain of consideration is different. hence, No two functions are equal. Am I correct? $\endgroup$ – Unknown x Jul 8 '18 at 2:32
  • $\begingroup$ We can rigorously prove it, your proof is fine; and it only depends on the $T_p(\mathbb{R}^n)$ being pairwise disjoint; and for this the proof depends on your initial definition of the tangent space $\endgroup$ – Max Jul 8 '18 at 9:31
  • $\begingroup$ Once we've decided that the tangent spaces are disjoint, we can prove that the cotangent spaces are disjoint. More generally, if two vector spaces $V$ and $W$ are disjoint (or merely distinct), then their duals $V^*$ and $W^*$ are disjoint also. $V^*$ consists of certain functions with domain $V$, and similarly for $W^*$. No function can have two different domains. $\endgroup$ – Andreas Blass Jul 8 '18 at 13:17

There are few steps to consider:

1) How are the level sets of a linear maps $\Bbb R^2\to\Bbb R$?

2) How are the level sets of a linear maps $\Bbb R^3\to\Bbb R$?

3) How are the level sets of a linear maps $\Bbb R^n\to\Bbb R$?

4) The tangent spaces $T_p\Bbb R^n$ are vector spaces associated to a position $p$.

5) Different points in $\Bbb R^n$ receive different tangent spaces so they are disjoint.

6) The dual of a tangent space $T^*_p{\Bbb R^n}$ are linear maps $T_p{\Bbb R^n}\to\Bbb R$, so a 1-form is an association of a linear map to a copy of $\Bbb R^n$ attached to $p$ in which one gets the picture explained in points 1), 2) or 3) above.

What about an example?

Let's illustrate with a simple one-form $\omega=(5+x^2-y)dx+(9+x+y^2)dy$. With it you can prescribe to a points in $\Bbb R^2$ a linear functional, let's pick at the origin $\omega|_{0,0}=5dx+9dy$ which is the same as the linear map in $\Bbb R^2\to\Bbb R$ given by $$(x,y)\mapsto\omega|_{0,0}(x,y)=5x+9y$$ Here we can experiment a see how a little horizontal variation, say at $(\epsilon,0)$ what covector gonna be? This will be $$\omega|_{\varepsilon,0}(x,y)=(5+\varepsilon^2)x+(9+\varepsilon)y$$ which is $\approx5x+9y$, for $0<\varepsilon<\!\!<1$.


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.