TITLE: Dimension of infinite product of vector spaces
QUESTION [31 upvotes]: This question is motivated by the question link text, which compares the infinite direct sum and the infinite direct product of a ring.
It is well-known that an infinite dimensional vector space is never isomorphic to its dual. More precisely, let $k$ be a field and $I$ be an infinite set. Let $E=k^{(I)}=\oplus_{i \in I} k$ be the $k$-vector space with basis $I$, so that $E^{*}$ can be identified with $k^I = \prod_{i \in I} k$. Then a stronger result asserts that the dimension of $E^{*}$ over $k$ is equal to the cardinality of $k^I$. This is proved in Jacobson, Lectures in Abstract Algebra, Vol. 2, Chap. 9, $\S$ 5 (Jacobson deduces it from a lemma which he attributes to Erdös and Kaplansky). Summarizing, we have
\begin{equation}
\operatorname{dim}_k (k^I) = \operatorname{card} k^I.
\end{equation}
Now, if $V$ is any $k$-vector space, we can ask for the dimension of $V^I$. Does the Erdös-Kaplansky theorem extend to this setting ?

Is it true that for any vector space $V$ and any infinite set $I$, we have $\operatorname{dim} V^I = \operatorname{card} V^I$ ? More generally, given a family of nonzero vector spaces $(V_i)$ indexed by $I$, is it true that $\operatorname{dim} \prod_{i \in I} V_i = \prod_{i \in I} \operatorname{card} V_i$ ?

If $V$ is isomorphic to $k^J$ for some set $J$, then the result holds as a consequence of Erdös-Kaplansky. In the general case, we have $V \cong k^{(J)}$, and we can assume that $J$ is infinite. In this case I run into difficulties in computing the dimension of $V^I$. I can only prove that $\operatorname{dim} V^I \geq \operatorname{card} k^I \cdot \operatorname{card} J$.

REPLY [18 votes]: Here is a self-contained version of Todd Trimble's wonderful answer.
Let $K$ be a field. "Vector space" shall mean "$K$-vector space", "linear" shall mean "$K$-linear", $\dim$ shall mean $\dim_K$, $\operatorname{Hom}$ shall mean $\operatorname{Hom}_K$, and $|X|$ shall denote the cardinal of $X$ for any set $X$.
Let $V$ be the product of a family of nonzero vector spaces $(V_i)_{i\in I}$:
$$
V=\prod_{i\in I}V_i.
$$
As we have
$$
\dim V=\sum_{i\in I}\dim V_i
$$
if $I$ is finite, we can (and will) assume from now on that $I$ is infinite.

Main Theorem. We have, in the above notation, $\dim V=|V|$. In words: the dimension of the product of an infinite family of nonzero vector spaces is equal to its cardinal.

As a corollary, let us express explicitly the dimension of the product $V$ of the $V_i$ in terms of the $d_i:=\dim V_i$. Setting
$$
\mu:=\max(\aleph_0,|K|),\quad\alpha:=|\{i\in I\ |\ d_i < \mu\}|,
$$
we get
$$
\dim\prod_{i\in I}V_i=|K|^\alpha\prod_{d_i\ge\mu}d_i.
$$
Let us prove the Main Theorem.

Lemma. If $V$ is a vector space which is infinite as a set, then we have
  $$
|V|=|K|\cdot\dim V.
$$

Proof. It is easy to see that, if $S$ is an infinite generating subset of a group $G$, then $S$ and $G$ are equipotent. Putting
$$
G:=V,\qquad S:=\{\lambda b\ |\ (\lambda,b)\in K\times B\},
$$
where $B$ is a basis of $V$, we get the conclusion. QED
Let $V$ be an infinite dimensional vector space.
Say that $V$ is large if $\dim V\ge \max (|K|, \aleph_0)$.
By the lemma, $V$ is large if and only if $\dim V=|V|$.

Erdős-Kaplansky Theorem. The vector space $K^{\mathbb N}$ is large.

It is clear that the Erdős-Kaplansky Theorem implies the Main Theorem. So we are left with proving the Erdős-Kaplansky Theorem.
Proof of the Erdős-Kaplansky Theorem. Let $B$ be a $K$-basis of $K^{\mathbb N}$, and suppose by contradiction $|B|<|K|$. Let $K_0$ be the prime subfield of $K$, and put
$$
K_1:=K_0(\{b_j\ |\ b\in B,\ j\in\mathbb N\}).
$$
As $|K_1|<|K|$ and $K$ is infinite, there is an $x$ in $K^{\mathbb N}$ whose coordinates are $K_1$-linearly independent. There are $c_1,\dots,c_n$ in $B$ such that $x$ is a $K$-linear combination of the $c_j$. Since $c_{ij}$ is in $K_1$ for all $i,j$, there is a nonzero $\lambda$ in $K_1^{n+1}$ such that
$$
\sum_{j=0}^n\lambda_j\,c_{ij}=0
$$
for $1\le i\le n$, and we have
$$
\sum_{j=0}^n\lambda_j\,x_j=0,
$$
in contradiction with the choice of $x$. QED