Let the space of all matrices over $\mathbb R$ of size $ n^2 $ , with the natural metric of $\mathbb R^{n^2 }$.

Prove that there exist a neighborhood of the identity matrix, such that all the matrices in that ball have a square root, i.e $$ \exists \varepsilon > 0\,:\forall X \in B\left( {I,\varepsilon } \right)\,\,\exists \,Y:\,Y^2 = X . $$

Hint: Consider the function $ F(X) = X^2 $ and use the Inverse Function Theorem

Help with this problem !! )=

  • 2
    $\begingroup$ Let $n=1$. Then not all matrices (= numbers) in the neighborhood of $0$ have a square root. $\endgroup$ – Christian Blatter Nov 18 '11 at 15:40
  • $\begingroup$ Wouldn't the following be a counterexample? Take any matrix with no square roots, and scale it (by a sufficiently small $\delta > 0$) so that the resulting matrix falls inside the ball. $\endgroup$ – Srivatsan Nov 18 '11 at 15:42
  • $\begingroup$ Sorry!! The center of the ball is not the zero matrix, it´s the identity I $\endgroup$ – August Nov 18 '11 at 19:41
  • $\begingroup$ Compute the derivative of $F$ at the identity and show that it is invertible... Since $F(I) = I$ the inverse function theorem then does the rest. $\endgroup$ – t.b. Nov 18 '11 at 19:44
  • $\begingroup$ How can I don´t know how to calculate the derivate of a matrix D: $\endgroup$ – August Nov 18 '11 at 19:47

You want to solve the equation $(I+X)^2=I+Y$ for the $(n\times n)$-matrix $X$ when $Y$ is a given matrix near $0$. This amounts to

$$Z(X):=2X+X^2\ =\ Y\ ,\qquad (*)$$

an equation in the $n^2$ unknowns $X_{ik}$. When $Y=0$ then $X=0$ is a solution of this equation. According to the assumptions of the implicit function theorem we have to check whether the $(n^2\times n^2)$-matrix

$$J:=\Bigl({\partial(Z_{11},Z_{12},\ldots,Z_{nn})\over \partial(X_{11},X_{12},\ldots,X_{nn})}\Bigl)_{X=0}$$

is regular. As $Z_{ik}=2 X_{ik} +$ terms of degree $2$ in the $X_{ik}$, the matrix $J$ is $2$ times the $n^2\times n^2$ identity matrix; so it is certainly regular. It follows that the equation $(*)$ has a solution $X=F(Y)$ such that $F(0)=0$ and $F(\cdot)$ is a differentiable matrix valued function of (the matrix elements of) $Y$.

By the way, the square root $I+X$ of $I+Y$ for $\|Y\|<1$ is given by the binomial series

$$\sum_{k=0}^\infty\ {1/2 \choose k}\ Y^k\ .$$


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.