In Euclidean space, we usually don't distinguish between vectors and covectors (or dual vectors or 1-forms or whatever you want to call them) -- because the spaces overlap. However, a physicist friend of mine (I'm a physics major, too, BTW) was trying to convince me that not only are Euclidean vectors and covectors the same thing, but ANY vector is equivalent to a covector. His reasoning is that because there is an isomorphism between them and we can convert one to the other with the metric tensor, that there's no real difference between them. My argument is that an isomorphism between mathematical objects is not enough to call two things the same. For example, there is an isomorphism between the additive groups of $\Bbb R$ and $\Bbb R^2$, but surely no one would say that $\Bbb R$ is the SAME THING as $\Bbb R^2$.

So which of us is right? Are vectors the same thing as covectors?

  • 4
    $\begingroup$ An isomorphism is an equivalence relation, so it's technically correct to say that a vector is equivalent to its covector, though it's not equal to its covector. I really hope this is what your friend meant, else he's a really bad physicist. Physicists should be more acutely aware than anyone else of the vital distinction between the two. $\endgroup$
    – David H
    Commented Jul 12, 2014 at 13:23
  • $\begingroup$ @DavidH I'm not quite sure if he was making that distinction or not. Basically I read a short paper on tensors and told him that the only thing I didn't really like about it was the it didn't distinguish between vectors in a vector space vs its dual space (though it was only discussing Euclidean vectors for the most part). Then he said that there's "only one type of vector, both vectors and covectors are really the same thing." $\endgroup$
    – phizzy
    Commented Jul 12, 2014 at 13:28
  • 5
    $\begingroup$ Your friend's remark suggests that he doesn't understand tensor calculus. $\endgroup$
    – TonyK
    Commented Jul 12, 2014 at 13:37
  • 3
    $\begingroup$ No one has pointed out that the only reason you get the isomorphism is that Euclidean space comes equipped withh a natural Riemannian metric. In an abstract manifold, without a metric or a nondegenerate two-form, you don't get such an isomorphism, which for Riemannian manifolds is called the "musical isomorphism." This is usually first learned when you're taught that you're really integrating forms instead of vector fields.... en.m.wikipedia.org/wiki/Musical_isomorphism $\endgroup$ Commented Jul 12, 2014 at 14:14
  • 2
    $\begingroup$ Perhaps in physics the difference between a vector and a covector are best emphasized by the terms contravariance and covariance. For example, a generalized coordinate transforms contravariantly (ok, essentially by defininition) whereas a generalized momentum transforms covariantly. These transformation laws are not the same. $\endgroup$ Commented Jul 12, 2014 at 19:16

3 Answers 3


To add to all the answers above, there is a delightful example in the text "Mathematics for Physics" by Stone and Goldbart, (Appendix A.3), to clarify the difference between vectors and co-vectors, which I can't resist quoting here.

One way of driving home the distinction between $V$ and $V^*$ is to consider the space $V$ of fruit orders at a grocers. Assume that the grocer stocks only apples, oranges and pears. The elements of $V$ are then vectors such as

$x = 3 \text{ kg apples }+ 4.5 \text{ kg oranges } + 2 \text{ kg pears.} $

Take $V^*$ to be the space of possible price lists, an example element being

$f = (\$3.00/\text{kg}) \text{ apples}^* + (\$2.00/\text{kg}) \text{ oranges}^* + (\$1.50/\text{kg}) \text{ pears}^*$

The evaluation of $f$ on $x$

$f(x) = 3 \times \$3.00 + 4.5 \times \$2.00 + 2 \times \$1.50 = 21.0$

then returns the total cost of the order. You should have no difficulty in distinguishing between a price list and box of fruit!


If $V$ is a finite-dimensional vector space then there exists an isomorphism $V\cong V^*$, under which every vector in $V$ can be interpreted as a covector (linear map $V\to $ the scalar field). However, there are going to be many such maps, and there is in general no canonical such isomorphism, so given a vector $v\in V$, which covector in $V^*$ it corresponds to depends on an ultimately arbitrary choice of isomorphism $V\cong V^*$.

Your friend does observe a valid mental tool that is useful in mathematics: identifying things when they are "essentially the same." If ever there is a canonical isomorphism $A\cong B$ between two objects satisfying some set of desirable properties (what's desirable depends on context), then we can think of any $a\in A$ as a corresponding $b\in B$. (Often we're really speaking about a collection of objects $A$ that are built or result from some kind of process or characterization, and we may use the language of category theory (universal properties, natural isomorphisms). In particular, there does not exist a natural isomorphism between the identity endofunctor and "dual space" endofunctor on the category of finite-dimensional vector spaces over a given field.)

But this identification idea doesn't work when there is no unique or canonical isomorphism between two given objects or structures. Indeed, $\{1,2,3\}\cong\{\rm\color{Red}{red},\color{Green}{green},\color{Blue}{blue}\}$. But if someone were to claim that "the numbers $1,2,3$ are basically colors," you could simply ask "if $1$ is a color, which color is it specifically?" and there would be no correct answer. Whenever there is a unique or canonical isomorphism $A\cong B$, given a $a\in A$, the question "which $b\in B$ is $a$ basically?" does have a unique correct answer; it's $\phi(a)$, where $\phi:A\to B$ is the aforementioned isomorphism.


Expanding just a touch on my comment:

One simple way to see that vectors and covectors are not as naively identifiable as your friend says is to realize that there are settings where the spaces they form aren't naturally isomorphic. They are developed independently on abstract manifolds, via the tangent and cotangent structures, respectively.

In particular, the only reason they are naturally isomorphic in Euclidean space is the presence of a natural Riemannian metric. It's this metric that gives you the sort structure you need to convert covectors (actually covector fields) into vectors (actually vector fields). This mapping is called the "musical isomorphism." You can get a similar mapping any time you have a non-degenerate two-form, e.g. a symplectic form would also work. But in the absence of such structure -- on an abstract finite-dimensional smooth manifold -- there is no such natural isomorphism, and you can't naturally identify the tangent and the cotangent structures (although any given tangent and cotangent space are isomorphic, by a dimensional argument).

This is a really important distinction, even in physics, as many have pointed out. It will show up when you learn how to integrate differential forms; you can't talk about "gradients" unless you have a Riemannian metric. It will also show up when you learn the symplectic structure of Hamiltonian mechanics: the natural setting for that is the cotangent bundle, but you can use the Legendre transform, sort of like the musical isomorphism, to push that physics into the tangent bundle, which gives you Lagrangian mechanics.


You might also be interested in the following discussion of how to "visualize" a covector, which can help to give an intuitive sense of why, though isomorphic, covectors and vectors are different things:



You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .