# Proofs where V is a finite-dimensional vector space

Printable View

• Sep 30th 2010, 10:09 AM
Runty
Proofs where V is a finite-dimensional vector space
These are probably easier than I think they are.

Suppose $V$ and $W$ are vector spaces over a field $F$.

Now suppose $W_1,...,W_k$ are subspaces of $V$ and that $V$ is finite-dimensional.

a) Prove that $\dim(W_1+...+W_k)\leq\dim(W_1)+...+\dim(W_k)$.

The question recommends proving this via induction (although proving via bases is possible, albeit more difficult). I think what I have below works, but if it needs correction, I'd like to know.

For $k=1, \dim(W_1)\leq\dim(W_1)$ is true. (base case)
Assume the result is true for $k=n$,
i.e. $\dim(W_1,...,W_n)\leq\dim(W_1)+...+\dim(W_n)$ (hypothesis)
For $k=n+1$, (inductive case)
$\dim(W_1,...,W_n,W_{n+1})\leq\dim(W_1)+...+\dim(W_ n)+\dim(W_{n+1})$
$=\dim(W_1,...,W_{k-1},W_k)\leq\dim(W_1)+...+\dim(W_{k-1})+\dim(W_k)$
Therefore, $\dim(W_1,...,W_k)\leq\dim(W_1)+...+\dim(W_k)$

b) Prove that $V=W_1\oplus...\oplus W_k$ if and only if every vector $v\in V$ is equal to $w_1+...+w_k$ for a unique choice of $w_j\in W_j, 1\leq j\leq k$.

I don't have an answer yet for this half, but I'm working on it.

Any help on these would be appreciated.
• Oct 2nd 2010, 01:43 PM
HappyJoe
Your inductive step looks weird.

You have typed commas instead of plusses between the W_i. Anyway, I suggest that you prove your base case _also_ for k=2.

Then the inductive case could go like: Suppose the result is true for k=n, and let k=n+1. Then

$dim(W_1+...+W_n+W_{n+1})$
$= dim((W_1+...+W_n)+W_{n+1})$
$\leq dim(W_1+\cdots+W_n)+dim(W_{n+1})$
$\leq dim(W_1)+\cdots+dim(W_n)+dim(W_{n+1}),$

where the first equality is just putting in parentheses, the first inequality is using the base case k=2 (with W_1+...+W_n and W_{n+1} being the two vector spaces), and the next inequality is using the induction hypothesis that the result is true for k=n.

As for b), what is your definition of direct sum?
• Oct 3rd 2010, 10:17 AM
Runty
Quote:

Originally Posted by HappyJoe
Your inductive step looks weird.

You have typed commas instead of plusses between the W_i. Anyway, I suggest that you prove your base case _also_ for k=2.

Then the inductive case could go like: Suppose the result is true for k=n, and let k=n+1. Then

$dim(W_1+...+W_n+W_{n+1})$
$= dim((W_1+...+W_n)+W_{n+1})$
$\leq dim(W_1+\cdots+W_n)+dim(W_{n+1})$
$\leq dim(W_1)+\cdots+dim(W_n)+dim(W_{n+1}),$

where the first equality is just putting in parentheses, the first inequality is using the base case k=2 (with W_1+...+W_n and W_{n+1} being the two vector spaces), and the next inequality is using the induction hypothesis that the result is true for k=n.

As for b), what is your definition of direct sum?

Thanks for clearing up my answer on the first half. I knew I'd done SOMETHING wrong.

As for the second half, I have quite a bit done, but again it could use some proof-reading. I probably have a few things out of proper order, which I feel would be the main issue.

We first have $v=w_1+...+w_k$, as provided by the question.
Then, we suppose that $v=w'_1+...+w'_k$.
This leads to the following: $w_1+...+w_k=w'_1+...+w'_k$
This implies that $0=(w_1-w'_1)+...+(w_k-w'_k)$,
$\Rightarrow w_1-w'_1=0,...,w_k-w'_k=0$
$\Rightarrow w_1=w'_1,...,w_k=w'_k$
Assuming the above is true, since each $w_j\in W_j$ is a unique choice, there could only be one solution to $w_1+...+w_k=0$, which is that $w_1=0,...,w_k=0$ for all $k>0$.

Therefore, $V=W_1\oplus...\oplus W_k$.

-----

Tell me if I've messed up anything. I'm suspecting right now that my proof just might be a bit out of order.
• Oct 3rd 2010, 12:47 PM
HappyJoe
Here's a quickly written answer, I hope you can read my non-LaTeX'ified scribbles.

What's your definition of direct sum? Is it that V is the direct sum of W_1, W_2, ..., W_k, iff W_1+...+W_k = V and the W_i have only the 0-vector in common?

For the first part of your proof, where you show that (V is a direct sum of the W_i) implies (each v is a unique sum of elements w_i), how do you argue from 0 = (w_1-w_1') + ... + (w_k - w_k') to w_1-w_1'=0, ..., w_k-w_k'=0?

I don't quite get the other part. It may be because I'm not sure which definition you are working from, or maybe because I'm busy.