Find a rank 1 matrix that better approximates a given matrix

121 Views Asked by At

I have been working on a problem that asks the following: given a matrix

$$\begin{pmatrix} 0 & 1 & 3\\ 3 & 1 & 0 \end{pmatrix},$$ I need to find a matrix $M$ such that $rank(M) = 1$ that better approximates A. That is, I have to find a matrix of type $$M = \begin{pmatrix} a & b & c\\ ra & rb & rc \end{pmatrix},$$ such that the (real) values $a, b, c, r$ are such that it minimizes the following expression:

$(a^2) + (1-b)^2 + (3-c)^2 + (3-ra)^2 + (1-rb)^2 + (rc)^2$ (that is, the Frobenius norm of the difference $A-M$).

I know that I can do it by finding the minimmum of a function on 4 variables. But I was trying to find a better and clever way to do this, but I wasn't able to do so.

I would appreciate if you could come up with any ideas that could make this problem more easy to solve.

(Edit 1) Maybe an idea could arise out of a decomposition of this matrix into a product (for instance, LU decomposition or something like this).

1

There are 1 best solutions below

0
On BEST ANSWER

The EYM theorem states that this approximation may be obtained using the singular value decomposition. In particular, we proceed as follows.

Begin by computing the singular value decomposition $A = U \Sigma V^T$, where $$ A = \pmatrix{0 & 1 & 3\\ 3 & 1 & 0}, \quad U = \frac 1{\sqrt{2}}\pmatrix{1&-1\\1&1}, \quad \Sigma = \pmatrix{\sqrt{11}\\ & 3}, \quad \\ V = \pmatrix{ 3/\sqrt{22} & 1/\sqrt{2} & 1/\sqrt{11}\\ \sqrt{2/11} & 0 & -3/\sqrt{11}\\ 3/\sqrt{22} & -1/\sqrt{2} & 1/\sqrt{11}}. $$ Then, compute the rank-1 "truncated SVD" $M = U \hat \Sigma V$, where $$ \hat \Sigma = \pmatrix{\sqrt{11}&0\\0&0}. $$ Equivalently, we find $$ M = \sqrt{11} \cdot \frac 1{\sqrt{2}}\pmatrix{1\\1}\pmatrix{3/\sqrt{22} & \sqrt{2/11} & 3/\sqrt{22}} \\ = \left(\begin{matrix}1.5 & 1.0 & 1.5\\1.5 & 1.0 & 1.5\end{matrix}\right). $$