I am a first year student, studying linear algebra. In the lecture we briefly discussed double dual spaces and I am not sure if I understood it correctly: we take a function f that is an element of the dual space and we evaluate the function on a vector from V and its value is an element of the double dual space?

Thank you very much in advance.

  • $\begingroup$ That's right. The bidual space $V^{\star\star}$ is the the space of linear functions that assign a scalar value to linear functionals on $V$. Fixing a vector $v$ the assignment $V^\star\ni\lambda\mapsto\lambda(v)$ defines an element in $V^{\star\star}$. $\endgroup$ – AdLibitum Jan 14 '17 at 21:02

Nope, you're missing a bit here. Long story short: the elements of the double dual space are functions that take a function $f$ from the dual space, and evaluate the function on a vector from $V$. That is, an element of the double dual space is a function of the form $f \mapsto f(v)$. It's nice to think of this weird function as just being that vector $v$ from $V$.

Long story long: for simplicity, I'll talk about vector spaces over $\Bbb R$, but the same applies over arbitrary fields.

The first thing to understand is that the set $\mathcal L(U,V)$ of linear transformations between two vector spaces $U$ and $V$ forms a vector space. For example, if $U = \Bbb R^n$ and $V = \Bbb R^m$, then $\mathcal L(U,V)$ is canonically identified with the space $\Bbb R^{m \times n}$ of $m \times n$ matrices. In general, $\dim (\mathcal L(U,V)) = \dim(U) \cdot \dim(V)$. Dimension is important because any vector spaces of the same (finite) dimension are isomorphic.

Now, for any vector space $V$, $V^* = \mathcal L(V,\Bbb R)$ is the dual space of $V$. The elements of $V^*$ are called linear functionals; they are linear transformations that take vectors and produce numbers. Notably, $\dim(V^*) = \dim(V) \cdot \dim(\Bbb R) = \dim(V) \cdot 1 = \dim (V)$. So, any (finite dimensional) space is isomorphic to its dual space.

The double dual space is the dual of the dual. That is, $V^{**} = \mathcal L(V^*, \Bbb R) = \mathcal L(\mathcal L(V,\Bbb R),\Bbb R)$. The elements of this space are linear transformations that take linear functionals and produce numbers. If that seems weird and unintuitive, that's fine: it should.

Just like $V^*$, $V^{**}$ is isomorphic to $V$, since $\dim(V^{**}) = \dim(V^*)\cdot 1 = \dim(V)$. However, it turns out that $V^{**}$ is canonically isomorphic to $V$. That is (for our purposes), it is isomorphic in a "really nice way". In particular, there is a really nice invertible linear map that takes us from $V$ to $V^{**}$, and it's so slick that we can think of $V$ and $V^{**}$ as being "essentially the same space".

Let's describe that map $\alpha:V \to V^{**}$. For any vector $v \in V$, we want an element $\alpha(v) = \alpha_v \in V^{**}$, which is to say that $\alpha_v$ takes in functionals $f \in V^*$, and produces a number. So, we define $$ \alpha_v(f) = f(v) $$ In other words, the question of "is $V$ canonically isomorphic to $V^{**}$?" can be roughly translated as "is there a natural way to use $v$ to make an element $f \in V^*$ into a number?" Our answer is, "yes: plug $v$ into $f$". For any vector $v \in V$, $\alpha_v$ is the element of $V^{**}$ that tells you to plug in $v$.

  • $\begingroup$ I understood the words "weird" and took comfort in "that's fine". +1 $\endgroup$ – Antoni Parellada Jan 14 '17 at 21:36
  • 1
    $\begingroup$ Best answer I’ve seen on this site in a long time. $\endgroup$ – PossumP Apr 4 '19 at 22:34
  • $\begingroup$ @BenGrossmann: Although it is $V^{**}$ canonically isomorphic to $V$ but I don't see it naturally, i.e. $V\subset V^{**}$ or $V^{**}\subset V$. (the embedding of a vector space in its double dual) $\endgroup$ – C.F.G Nov 23 '20 at 18:46
  • $\begingroup$ @C.F.G Pinning down the definition of "natural" in this context is a bit tricky. Of course, it is not technically correct to say that $V \subset V^{**}$. However, it suffices for an introductory context to note that the injective map $v \mapsto \alpha_v$ can be defined using only the fact that $V$ is a vector space, and no "choices" (e.g. a selection of basis) need to be made $\endgroup$ – Ben Grossmann Nov 23 '20 at 18:52

Yes, $\psi_x : E' \to \Bbb{R}$ defined by $\psi_x(f) = f(x)$ is an element of $E''$.

That even give us a canonical injection from $E$ to $E''$ ( $x\mapsto \psi_x$)

But note that often not all elements of $E''$ can be written as such (there's not always a bijection between $E$ and $E''$)

  • $\begingroup$ This doesn't address the fundamental confusion of what a double dual space is $\endgroup$ – Ben Grossmann Jan 14 '17 at 21:04

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.