6
$\begingroup$

Let matrix $C_{n\times‎n}$ and equation $X^{2}=C$ be given, i want to find matrix $X$.

For $n=2$, $X$ is obtained by solving a system of equations;

$$\left\{ \begin{array}{l} x_{11}^2 + {x_{12}}{x_{21}} = {c_{11}}\\ {x_{11}}{x_{12}} + {x_{12}}{x_{22}} = {c_{12}}\\ {x_{21}}{x_{11}} + {x_{22}}{x_{21}} = {c_{21}}\\ {x_{21}}{x_{12}} + x_{22}^2 = {c_{22}} \end{array} \right. $$

but does analytical method (without solve a system of equations) for solving such equations exist?

Thanks.

$\endgroup$
  • 1
    $\begingroup$ Diagonalization is a possible method. $\endgroup$ – Seirios Jan 13 '13 at 17:59
  • $\begingroup$ @M.Sina firstly you should ask yourself is such a matrix $X$ always exists. I doubt it. Diagonalization won't help if $C$ isn't diagonalizable. $\endgroup$ – Git Gud Jan 13 '13 at 18:01
  • 3
    $\begingroup$ I second @GitGud's comment. What structure do you have on $C$? For example, $C=\begin{bmatrix} 0 & 1 \\ 0 & 0 \end{bmatrix}$ does not have a square root. $\endgroup$ – copper.hat Jan 13 '13 at 18:01
  • 3
    $\begingroup$ @M.Sina In this case, note that $\det(X^2)=(\det X)^2\geq 0$. So there is no solution for every $C$ such that $\det C<0$. $\endgroup$ – Julien Jan 13 '13 at 18:31
  • 1
    $\begingroup$ For $n=2$ , someone has done a complete analysis in tw.knowledge.yahoo.com/question/question?qid=1010020803771. $\endgroup$ – doraemonpaul May 17 '13 at 0:46
4
$\begingroup$

No, an "analytical method (without solve a system of equations)" for solving such equations does not exist. This is because any matrix equation is by definition a system of equations.

However, considering when a matrix is already diagonalized, finding the solution is very simple. Read ahead for some considerations of the non-diagonalizable case.

Every matrix may be orthogonaly reduced to triangular. Take then the equation $Y^2 = R$ where $R$ is triangular and $C=QRQ^*$ where $QQ^*=I$. Then the solution for $X$ may be found

$$X^2=C=QRQ^* = QY^2Q^*$$ So that $$ X = QYQ^*$$

So it comes down to finding for a triangular matrix. Lets look at the $3 \times 3$ case. $$\pmatrix{d_0 & 0 & 0 \\ a_1 & d_1 & 0 \\ a_2 & b_2 & d_2}\pmatrix{d_0 & 0 & 0 \\ a_1 & d_1 & 0 \\ a_2 & b_2 & d_2} = \pmatrix{d_0^2 & 0 & 0 \\ a_1(d_0 + d_1) & d_1^2 & 0 \\ a_2(d_0 + d_2) + a_1b_2 & b_2(d_1 + d_2) & d_2^2}$$

This would not necessarily give a result " analytical method (without solve a system of equations)" but it does show that the square root does exist, and there are $2^n$ different solutions, where $n$ is the number of non-zero eigenvalues.

EDIT: I will rollback if this is wrong as it is ripped from a text, but...

Functions of Jordan Blocks

For a $k \times k$ Jordan block $J_*$ with eigenvalue $\lambda$, and for a function $f(z)$ such that $f(\lambda), f'(\lambda), \dots, f^{(k-1)}(\lambda)$ exist, $f(J_*)$ is defined to be $$f(J_*) = f\pmatrix{\lambda & 1 \\ & \ddots & \ddots \\ & & \ddots & 1 \\ &&& \lambda}=\pmatrix{f(\lambda) & f'(\lambda) & \frac{f''(\lambda)}{2!} & \cdots & \frac{f^{(k-1)}(\lambda)}{(k-1)!}\\ & f(\lambda)& f'(\lambda)& \ddots & \vdots\\ & & \ddots & \ddots & \frac{f''(\lambda)}{2!}\\ & & & f(\lambda) & f'(\lambda)\\ & & & & f(\lambda)\\}$$

$\endgroup$
  • $\begingroup$ @copper though I suppose your example shows the problem with this when $d_i=0$ for one or more $d_i$. $2^n=0$ in that case though, doesn't it, so no roots exist... $\endgroup$ – adam W Jan 13 '13 at 18:18
  • $\begingroup$ Why did you assume the square root of a triangular matrix is also triangular? $\endgroup$ – Arin Chaudhuri Jan 13 '13 at 18:21
  • $\begingroup$ I assumed that the square of a triangular is also triangular. $\endgroup$ – adam W Jan 13 '13 at 18:23
  • $\begingroup$ @ACARCHAU you are correct, the square root need not be triangular. See jspecter's nice answer to my question. $\endgroup$ – adam W Jan 14 '13 at 2:42

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.