In Linear Algebra Done Right, the book proves that every subspace of $V$ is part of a direct sum equal to $V$. I generally follow the proof, but do not understand some points.

Suppose $V$ is finite-dimensional and $U$ is a subspace of $V$. Then there is a subspace $W$ of $V$ such that $V = U\bigoplus W$.

Proof Because $V$ is finite-dimensional, so is $U$ (proved in 2.26. I am OK). Thus there is a basis $u_1,...,u_m$ of $U$. Of course $u_1,...u_m$ is a linearly independent list of vectors in $V$ (I am OK). Hence this list can be extended to a basis $u_1,...,u_m, w_1,...,w_n$ of $V$ (I am not OK. Does it mean $w_1,...w_n$ can only be vectors that extend to the basis of $V$? If so, why the title of the proof said EVERY subspace of $V$ is part of a direct sum equal to $V$).

To prove $V = U\bigoplus W$, we need only show that $$V = U+W \text{ and } U \cap W = \{0\}$$ (I am OK).

To prove the first equation above, suppose $v \in V$. Then, because the list $u_1,...,u_m, w_1,...w_n$ spans $V$, there exist $a_1,...a_m, b_1,...b_m \in \mathbb{F}$ such that $$v=a_1u_1+...+a_mu_m+b_1w_1+...+b_nw_n$$.

In other words, we have $v=u+w$, wheere $u \in U$ and $w \in W$ are defined as above. Thus $v \in U + W$, completing the proof that $V = U + W$. (I am also OK.)

To show that $U \cap W = \{0\}$, suppose $v \in U \cap W$. Then there exist scalars $a_1,...,a_m,b_1,...b_n \in \mathbb{F}$ such that

$$v=a_1v_1+...+a_mv_m = b_1w_1+...+b_nw_n$$. Thus $$a_1u_ + ...+a_mu_m - b_1w_1-...-b_nw_n = 0$$

Because $u_1,...u_m,w_1,...,w_n$ is linearly independent, this implies that $a_1=...=a_m=b_1=...=b_n = 0$. Thus $v=0$, completing the proof that $U \cap W = \{0\}$.

Is the proof means for every subspace $U$ of finite-dimensional $V$, we can find a $W$ that is the direct sum of $V$?

  • $\begingroup$ It means that every subspace has a complementary subspace. $\endgroup$
    – Bernard
    Jan 1 '19 at 23:09
  • $\begingroup$ I'm not entirely sure what you are confused about but it seems like you are under the impression that $V$ only has one basis. Consider for example $\mathbb{R} ^2$. We have the standard basis $\{(1,0),(0,1)\}$ but we can replace $(1,0)$ with $(1,1)$ and it is still a basis. In fact replacing $(1,0)$ with anything in the form $(a,b)$ where $a\neq 0$ will result in a basis. $\{(1,2),(1,3)\}$ is another example of a basis. $\endgroup$
    – Fortox
    Jan 1 '19 at 23:13

You have a subspace $U$ with basis $u_1,...,u_m$.

Extend the basis to a basis of $V$ by adding vectors $w_1,...,w_n$. There is some freedom to choose the $w_k$ but they must be linearly indepdendent and the collection must span $V$.

Let $W=\operatorname{sp} \{w_k\}$.

Since the whole collection spans $V$, we must have $V = U +W$. If $u\in U, w\in W$ and $u+w = 0$, we must have $u=w=0$ since the whole collection is linearly independent.

Note that $W$ is not unique.


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.