Find precisely the coefficients of product matrix AB as from B.

518 Views Asked by At

In the book by Serge Lang, titled 'Introduction to Linear Algebra', there is given on pg. #88, as below:

Let $A = (a_{ij}), B = (b_{jk})$ & let $AB =C$ with $C =(c_{ik})$. Let $C^k$ be the $k$-th column of $C$. Express $C^k$ as a linear combination of the columns of $A$. Describe precisely which are the coefficients, coming from the matrix $B$.

Let us take an example with given size of matrices $A,B$, such that the sizes differ for all $3$ matrices; e.g. $A$ is a $2*3$ matrix, $B$ is a $3*2$ matrix; while their product is a $2*2$ one; while the terms are general ones.
I hope this will help understand the general case of matrices: $A$ of size $m*n$, $B$ of size $n*k$, $AB$ of size $m*k$.

$A = \begin{bmatrix} a_{11} & a_{12} & a_{13}\\ a_{21} & a_{22} & a_{23} \\ \end{bmatrix} $, $B = \begin{bmatrix} b_{11} & b_{12}\\ b_{21} & b_{22}\\ b_{31} & b_{32} \end{bmatrix} $, $C = AB = \begin{bmatrix} a_{11}b_{11} + a_{12}b_{21} + a_{13}b_{31} & a_{11}b_{12} + a_{12}b_{22} + a_{13}b_{32}\\ a_{21}b_{11} + a_{22}b_{21} + a_{23}b_{31} & a_{21}b_{12} + a_{22}b_{22} + a_{23}b_{32}\\ \end{bmatrix}$

Let $k=2$ (out of the two possible values for $k=1,2$), for $C^2$ or the $2$-nd column of $C$, and follow the approach given here to express as a linear combination of $A$.
Let us express the matrix $A$ in terms of columns only as : $A = \begin{bmatrix} A^1 & A^2 & A^3\\ \end{bmatrix} $. So, for $C$ have $\begin{bmatrix} A^1 & A^2 & A^3\\ \end{bmatrix} \begin{bmatrix} b_{11} & b_{12}\\ b_{21} & b_{22}\\ b_{31} & b_{32} \end{bmatrix} = \begin{bmatrix} C^1 & C^2 \end{bmatrix} = \begin{bmatrix} A^1 b_{11}+ A^2 b_{21}+ A^3 b_{31} & A^1 b_{12}+A^2 b_{22}+A^3 b_{32} \end{bmatrix},$
with $C^2$ being the second term .

I feel tempted here (due to lack of clarity due to generalized terms) to consider a practical example of two linearly independent vectors $A,B$, with given dimensions. Taking actual value based example for $A,B$, get:

$A = \begin{bmatrix} 2 & 3 & 4\\ 5 & 6 & 7 \\ \end{bmatrix} $, $B = \begin{bmatrix} 1 & 2\\ 3 & 4\\ 5 & 6 \end{bmatrix} $, $AB = \begin{bmatrix} 2\times 1 + 3\times 3 + 4\times 5 & 2\times 2 + 3\times 4 + 4\times 6\\ 5\times 1 + 6\times 3 + 7\times 5 & 5\times 2 + 6\times 4 + 7\times 6\\ \end{bmatrix}$ $AB = \begin{bmatrix} 2 + 9 + 20=31 & 4 + 12 + 24=40\\ 5 + 18 + 35=58 & 10 + 24 + 42=76\\ \end{bmatrix}$

Want verify the above expression $C^2$, for the given example with $A^1 = \begin{bmatrix} 2 \\ 5 \\ \end{bmatrix}, A^2 = \begin{bmatrix} 3 \\ 6 \\ \end{bmatrix}, A^3 = \begin{bmatrix} 4 \\ 7 \\ \end{bmatrix}$.

So, $C^2 = \begin{bmatrix} A^1b_{12}+A^2b_{22}+A^3b_{32} \end{bmatrix} = \begin{bmatrix} 2 \\ 5 \\ \end{bmatrix}\cdot 2 + \begin{bmatrix} 3 \\ 6 \\ \end{bmatrix}\cdot 4 + \begin{bmatrix} 4 \\ 7 \\ \end{bmatrix}\cdot 6 $.

Although am clueless, yet follow the hint given here at mse, even though my chosen matrix $A$ would not lead to an identity ($I$) matrix (as, the $rref$ form works for rectangular matrices. But, it will 'only' produce leading row entry (left to right, across rows) as $1$).
Still would follow it to find $rref$ form of the augmented matrix $[A|AB]$, given as : $[A|AB] = \begin{array}{ccc|cc} a_{11} & a_{12} & a_{13}&a_{11}b_{11} + a_{12}b_{21} + a_{13}b_{31} & a_{11}b_{12} + a_{12}b_{22} + a_{13}b_{32}\\ a_{21} & a_{22} & a_{23} & a_{11}b_{11} + a_{12}b_{21} + a_{13}b_{31} & a_{11}b_{12} + a_{12}b_{22} + a_{13}b_{32}\\ \end{array} $
which for the example at hand is:
$[A|AB] = \begin{array}{ccc|cc} 2 & 3 & 4 & 31 & 40\\ 5 & 6 & 7 & 58 & 76\\ \end{array}\implies \begin{array}{ccc|cc} 2 & 3 & 4 & 31 & 40\\ 0 & -1.5 & -3 & -19.5 & -24\\ \end{array}\implies \begin{array}{ccc|cc} 2 & 0 & -2 & -7 & -16\\ 0 & -1.5 & -3 & -19.5 & -24\\ \end{array}\implies \begin{array}{ccc|cc} 1 & 0 & -1 & -7/2 & -8\\ 0 & 1 & 2 & 13 & 36\\ \end{array}$

Please help me with : (i) vetting for any error above,
(ii) please tell if have expressed above $C^2$ correctly as a linear combination of $A$,
(iii) how to find out the coefficients coming from matrix $B$.

2

There are 2 best solutions below

4
On BEST ANSWER

Say $A$ has $n$ columns $A^i$, $1\le i\le n$. We can then describe $A$ using the columns of $A$ thus: $A = \begin{bmatrix}A^{1} & A^{2} & \dots & A^{j} \dots & A^{n} \end{bmatrix}$. To multiply $A$ and $B$, $B$ necessarily has $n$ rows, and so the $k$th column of $C=AB$, or $C^k$, is formed by multiplying $A$ with the $k$th column of $B$, or $B^k$, thus $$ \begin{bmatrix} A^{1} & A^{2} & \dots & A^{j} \dots & A^{n} \end{bmatrix} \begin{bmatrix} b_{1k}\\ b_{2k}\\ \vdots\\ b_{jk}\\ \vdots\\ b_{nk}\\ \end{bmatrix}=b_{1k}A^1+b_{2k}A^2+\dots+b_{jk}A^j+\dots+b_{nk}A^n\tag{1}$$

Only square matrices have the possibility of being invertible, so you can't row reduce to isolate $B$ in the non-square case.

Once you have a linear combination as in the left hand side of $(1)$ you can find the $b_{jk}$ entries of the $k$th column of $B$ by comparing with the original columns of $A$, if you know the $A^i$ first, i.e., if you already know $A$. For this consider your example: $A^1b_{12} = \begin{bmatrix}4 \\ 10 \\ \end{bmatrix}$, so $A^1=\frac{1}{b_{12}}\begin{bmatrix}4 \\ 10 \\ \end{bmatrix}$. But you know $A^1 = \begin{bmatrix}2 \\ 5 \\ \end{bmatrix}=\frac{1}{2}\begin{bmatrix}4 \\ 10 \\ \end{bmatrix}$, so $b_{12}=2$ follows.

As to why $b_{ij}A^i=A^ib_{ij}$, $b_{ij}$ is a scalar here, and $A^i$ is a column vector so this is merely commutativity of multiplication of scalars thus: $$A^1b_{12} = \begin{bmatrix}2 \\ 5 \\ \end{bmatrix}\cdot2=\begin{bmatrix}2\cdot2 \\ 5\cdot2 \\ \end{bmatrix}=2\cdot \begin{bmatrix}2 \\ 5 \\ \end{bmatrix}=\begin{bmatrix}2\cdot2 \\ 2\cdot5 \\ \end{bmatrix}=b_{12}A^1$$

0
On

Hint: Show that $C^k = AB^k$ using the definition of matrix multiplication. (Here $C^k$ and $B^k$ denote the $k$-th column of $C$ and $B$, respectively.)