Let $\Lambda^k(V)$ be the space of alternating $k$-linear tensors on $V$. Consider the map $f: \left(\mathbb{R}^n\right)^{n-k} \to \Lambda^k(\mathbb{R}^n)$ given by $\left(v_1,v_2, ..., v_{n-k}\right) \mapsto \text{Det}(v_1,v_2,...,v_{n-k},\cdot)$, i.e. the $k$-form you get is a partially applied determinant.

Is there a name for the image of $f$? I am having trouble computing even basic things like the dimension of this space. I would imagine this space must have been studied classically.

I am thinking about this because the surjectivity of the map for $n=3$ seems to underlie a lot of multivariable calculus in $3$ dimensions (it lets you translate 2 forms into vector fields for instance, which is what the textbooks all seem to do).


Using the determinant is something of a distraction. In completely coordinate-free terms, and after applying the observation in abx's answer, the map you're trying to understand is the wedge product map $$V^k \ni (v_1, ..., v_k) \mapsto v_1 \wedge ... \wedge v_k \in \Lambda^k(V)$$

where $V$ is a finite-dimensional vector space (of dimension, say, $n$). So the variety you're trying to understand is the variety of pure tensors sitting inside $\Lambda^k(V)$.

This variety has definitely been studied classically. The pure tensors are characterized by a set of quadratic relations called the Plücker relations. They cut out a subvariety of the projective space $\mathbb{P}(\Lambda^k(V))$ into which the Grassmannian $\text{Gr}_k(V)$ embeds via the Plücker embedding

$$\text{Gr}_k(V) \ni \text{span}(v_1, ..., v_k) \mapsto v_1 \wedge ... \wedge v_k \in \mathbb{P}(\Lambda^k(V)).$$

It follows that the pure tensors are the affine cone of $\text{Gr}_k(V)$ in this embedding, and so the space of pure tensors has dimension one greater than $\text{Gr}_k(V)$, hence $k(n-k) + 1$.

Example. The first nontrivial Plücker relation occurs when $\dim V = 4$ and $k = 2$. We can describe a pure tensor $v \wedge w$ using a $2 \times 4$ matrix, say

$$\left[ \begin{array}{ccc} c_1 & c_2 & c_3 & c_4 \\ d_1 & d_2 & d_3 & d_4 \end{array} \right],$$

where the first row gives the components of $v$ in some basis and the second row gives the components of $w$. The components of $v \wedge w$ are then given by the $2 \times 2$ minors of this matrix, and the only nontrivial Plücker relation, given by expanding $(v \wedge w) \wedge (v \wedge w) = 0$, is

$$p_{12} p_{34} - p_{13} p_{24} + p_{14} p_{23} = 0$$

where $p_{ij} = \det \left[ \begin{array}{cc} c_i & c_j \\ d_i & d_j \end{array} \right]$. Hence the pure tensors have dimension $6 - 1 = 2(4 - 2) + 1 = 5$ in this case.

Identifying forms and vector fields is a somewhat different matter. The canonical identifications are as follows. As abx says, the wedge product gives a nondegenerate pairing $\Lambda^k(V) \otimes \Lambda^{n-k}(V) \to \Lambda^n(V)$. The partial determinant gives a nondegenerate pairing $\Lambda^k(V) \otimes \Lambda^k(V^{\ast}) \to 1$ ($1$ the ground field) which allows us to commute taking duals with taking exterior powers. This gives identifications

$$\Lambda^k(V) \cong \Lambda^n(V) \otimes \Lambda^{n-k}(V)^{\ast} \cong \Lambda^n(V) \otimes \Lambda^{n-k}(V^{\ast}).$$

Hence to identify $\Lambda^k(V)$ with $\Lambda^{n-k}(V^{\ast})$ requires precisely the data of a nonzero element of $\Lambda^n(V)$. Globally this says that to identify vector fields with $n-1$-forms on an $n$-dimensional manifold requires precisely the data of a volume form. In particular you do not need a metric.

  • $\begingroup$ Awesome, thanks a lot. $\endgroup$ – Steven Gubkin Nov 15 '13 at 12:05

I think everything you ask boils down to the following: if $V$ is a vector space over a field $K$, say of dimension $n$, the wedge product $\ \wedge^pV\times \wedge^{n-p}V\rightarrow \wedge^nV\cong K\ $ is a perfect pairing, i.e. it identifies one space to the dual of the other.

  • $\begingroup$ How does it do this exactly? $Im(f)$ is not all of $\Lambda^k(\mathbb{R}^n)$, since the domain of $f$ is $n(n-k)$ dimensional, and $\Lambda^k(\mathbb{R}^n)$ is $\binom{n}{k}$ dimensional... $\endgroup$ – Steven Gubkin Nov 14 '13 at 18:58
  • $\begingroup$ @abx: it does not quite identify one space with the dual of the other, e.g. as $\text{GL}(V)$-representations we do not get duality because the action of $\text{GL}(V)$ on $\Lambda^n(V)$ is nontrivial. $\endgroup$ – Qiaochu Yuan Nov 14 '13 at 19:20
  • $\begingroup$ @Qiaochu Yuan: Right. But the OP is working with $\Bbb{R}^n$, so there is a canonical identification $\wedge^n\Bbb{R}^n\cong \Bbb{R}$. $\endgroup$ – abx Nov 14 '13 at 19:34
  • $\begingroup$ @QiaochuYuan I understand that $f$ is a multilinear map, and not a linear one. It is still a function from a $n(n-k)$ dimensional vector space to a $binom{n}{k}$ dimensional vector space, so I do not see how it is going to be a surjective map. $\endgroup$ – Steven Gubkin Nov 14 '13 at 20:27
  • $\begingroup$ @Steven: oh, I see, you're not looking at the induced linear map but at the image of the pure tensors. My apologies. I think the image is described by Plucker relations of some kind. $\endgroup$ – Qiaochu Yuan Nov 14 '13 at 22:02

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.