Linear Algebra I MT22, Change of basis theorem


Composition of linear maps as matrix multiplication

If

\[T : V \to W\]

What does the notation

\[A = {} _ \mathcal{U}T _ \mathcal{V}\]

mean?

The transformation matrix of $T$ with respect to basis $\mathcal{V}$ for $V$ and $\mathcal{U}$ for $W$.

There is a theorem that states

Let $U, V, W$ be finite-dimensional vector spaces with dimensions $m, n, p$ and ordered bases $\mathcal{U}, \mathcal{V}, \mathcal{W}$. Let $S : U \to V$ and $T : V \to W$ be linear. Let $A = {} _ \mathcal{V}S _ \mathcal{U}$ and $B = {} _ \mathcal{W}T _ \mathcal{V}$. Then $BA = {} _ \mathcal{W}TS _ \mathcal{U}$.

What does this actually mean?

Matrix multiplication represents the composition of linear maps.

Let $V$ be an $n$-dimensional vector space with ordered basis $\mathcal{V} = [v _ 1, \ldots, v _ n]$. Let $W$ be an $m$-dimensional vector space with ordered basis $\mathcal{W} = [w _ 1, \ldots, w _ m]$. Let $T : V\to W$ be a linear transformation. Let ${} _ \mathcal{W}T _ \mathcal{V}$ be the matrix representing this linear transformation.

Say $v \in V$ has coordinate representation $\underline{x}$.

What does ${} _ \mathcal{W}T _ \mathcal{V} \underline{x}$ equal?

$\underline{y}$ where $\underline{y}$ is the coordinate representation of $T(v)$.

Let $V$ be an $n$-dimensional vector space with ordered basis $\mathcal{V} = [v _ 1, \ldots, v _ n]$. Let $W$ be an $m$-dimensional vector space with ordered basis $\mathcal{W} = [w _ 1, \ldots, w _ m]$. Let $T : V\to W$ be a linear transformation. Let ${} _ \mathcal{W}T _ \mathcal{V}$ be the matrix representing this linear transformation. Denote the entries of ${} _ \mathcal{W}T _ \mathcal{V}$ with $(a _ {ij})$.

What’s the maximum value $i$ could be?

\[m\]

Let $V$ be an $n$-dimensional vector space with ordered basis $\mathcal{V} = [v _ 1, \ldots, v _ n]$. Let $W$ be an $m$-dimensional vector space with ordered basis $\mathcal{W} = [w _ 1, \ldots, w _ m]$. Let $T : V\to W$ be a linear transformation. Let ${} _ \mathcal{W}T _ \mathcal{V}$ be the matrix representing this linear transformation. Denote the entries of ${} _ \mathcal{W}T _ \mathcal{V}$ with $(a _ {ij})$.

What’s the maximum value $j$ could be?

\[n\]

Let $V$ be an $n$-dimensional vector space with ordered basis $\mathcal{V} = [v _ 1, \ldots, v _ n]$. Let $W$ be an $m$-dimensional vector space with ordered basis $\mathcal{W} = [w _ 1, \ldots, w _ m]$. Let $T : V\to W$ be a linear transformation. Let ${} _ \mathcal{W}T _ \mathcal{V}$ be the matrix representing this linear transformation. Denote the entries of ${} _ \mathcal{W}T _ \mathcal{V}$ with $(a _ {ij})$.

What is in the $i$-th column of the matrix?

$\underline{y _ i}$ where $\underline{y _ i}$ is the coordinate representation of $T(v _ i)$.

Let $V$ be an $n$-dimensional vector space with ordered basis $\mathcal{V} = [v _ 1, \ldots, v _ n]$. Let $W$ be an $m$-dimensional vector space with ordered basis $\mathcal{W} = [w _ 1, \ldots, w _ m]$. Let $T : V\to W$ be a linear transformation. Let ${} _ \mathcal{W}T _ \mathcal{V}$ be the matrix representing this linear transformation. Denote the entries of ${} _ \mathcal{W}T _ \mathcal{V}$ with $(a _ {ij})$.

What property characterises $T(v _ i)$?

\[T(v _ i) = \sum^m _ {k=1} a _ {ki} w _ k\]

Let $V$ be an $n$-dimensional vector space with ordered basis $\mathcal{V} = [v _ 1, \ldots, v _ n]$. Let $W$ be an $m$-dimensional vector space with ordered basis $\mathcal{W} = [w _ 1, \ldots, w _ m]$. Let $T : V\to W$ be a linear transformation. Let ${} _ \mathcal{W}T _ \mathcal{V}$ be the matrix representing this linear transformation. Denote the entries of ${} _ \mathcal{W}T _ \mathcal{V}$ with $(a _ {ij})$.

How can you remember the property characterises $T(v _ i)$, namely

\[T(v _ i) = \sum^m _ {k=1} a _ {ki} w _ k\]

?

This is just a mathematical way of stating the fact that the $i$-th column is the coordinate representation of $T(v _ i)$.

Let $U, V, W$ be finite-dimensional vector spaces with dimensions $m, n, p$ and ordered bases $\mathcal{U}, \mathcal{V}, \mathcal{W}$. Let $S : U \to V$ and $T : V \to W$ be linear. Let $A = {} _ \mathcal{V}S _ \mathcal{U}$ and $B = {} _ \mathcal{W}T _ \mathcal{V}$. Then what is true?

Then $BA = {} _ \mathcal{W}TS _ \mathcal{U}$.

Let $U, V, W$ be finite-dimensional vector spaces with dimensions $m, n, p$ and ordered bases $\mathcal{U}, \mathcal{V}, \mathcal{W}$. Let $S : U \to V$ and $T : V \to W$ be linear. What transformation would the matrix ${} _ \mathcal{W}TS _ \mathcal{U}$ represent?

\[(T \circ S)\]

When proving the following

Let $U, V, W$ be finite-dimensional vector spaces with dimensions $m, n, p$ and ordered bases $\mathcal{U}, \mathcal{V}, \mathcal{W}$. Let $S : U \to V$ and $T : V \to W$ be linear. Let $A = {} _ \mathcal{V}S _ \mathcal{U}$ and $B = {} _ \mathcal{W}T _ \mathcal{V}$. Then $BA = {} _ \mathcal{W}TS _ \mathcal{U}$.

You start by writing down the property that characterises the matrices $A$ and $B$, namely

\[S(u _ i) = \sum^n _ {k=1} a _ {ki} v _ k\]

and

\[T(v _ i) = \sum _ {k=1}^p b _ {ki} w _ k\]

What do you need to end up showing?

\[\begin{aligned} (T \circ S)(u _ i) &= T(S(u _ i)) \\ &= \ldots \\ &= \sum _ {k=1}^p (BA) _ {ki}w _ k \end{aligned}\]

There’s a very important result about matrices and the composition of linear maps that’s useful for proving the change of basis theorem. @State it in full.

Let $U, V, W$ be finite-dimensional vector spaces with dimensions $m, n, p$ and ordered bases $\mathcal{U}, \mathcal{V}, \mathcal{W}$. Let $S : U \to V$ and $T : V \to W$ be linear.

Let $A = {} _ \mathcal{V}S _ \mathcal{U}$ and $B = {} _ \mathcal{W}T _ \mathcal{V}$. Then $BA = {} _ \mathcal{W}TS _ \mathcal{U}$.

Change of basis theorem

@State the change of basis theorem in terms of $V, \mathcal{V}, \mathcal{V}', W, \mathcal{W}, \mathcal{W}'$ and $T$.

Let $V$ be a finite-dimensional vector space with ordered bases $\mathcal{V}, \mathcal{V}'$. Let $W$ be a finite-dimensional vector space with ordered bases $\mathcal{W}, \mathcal{W}'$. Let $T : V \to W$ be a linear map. Then

\[{} _ \mathcal{W'}T _ \mathcal{V'} = ({} _ \mathcal{W'}I _ \mathcal{W})({} _ \mathcal{W}T _ \mathcal{V})({} _ \mathcal{V}I _ \mathcal{V'})\]

For a matrix like ${} _ \mathcal{V}T _ \mathcal{V'}$, what’s the basis of the input?

\[\mathcal{V}'\]

Bite-sized

For an endomorphism $T : V \to V$ and two bases $\mathcal V, \mathcal V'$, the change-of-basis formula gives a similarity transformation: $[T] _ {\mathcal V'} = P^{-1} [T] _ {\mathcal V} P$, where $P$ is the change-of-basis matrix from $\mathcal V'$ to $\mathcal V$.

Source: Basicblock cards above in this entry; foundational for diagonalisation and the QR algorithm.

@bite~

Why is similarity the right notion of equivalence between matrices in NLA?

Similar matrices represent the same linear transformation in different bases. So invariants of the linear map — eigenvalues, determinant, trace, characteristic polynomial, minimal polynomial — are preserved. Algorithms like the QR algorithm work by applying a sequence of similarity transformations (specifically orthogonal ones, for stability) to reveal the eigenvalues on the diagonal of the limit. The whole structure-revealing program of NLA is “find an orthogonal similarity that diagonalises (or triangularises) $A$”.

Source NLA MT25 §8.1 Schur decomposition, §9 QR algorithm.

@bite~