MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $A$ be an $n\times n$ matrix which depends smoothly on a variable $x\in \mathbb{R}^n$ and such that there are constants $C_\alpha > 0 $ so that $\| \partial ^\alpha A \| \le C_\alpha $ for all multi-indices $\alpha \in \mathbb{N}_0^n$ (i.e. $A$ together with all its derivatives are bounded in matrix norm). Call the set of such matrices $S$. Assume moreover that at some point $x_0 \in \mathbb{R}^n$ we have $A(x_0) = I$ (identity matrix).

I now want to prove that we can find a matrix $B\in S$ that is supported away from $x_0$ and such that $A + B$ is invertible for all $x\in \mathbb{R} ^n$.

My approach was that, since $A$ is smooth, we can find neighborhoods $V_0$ and $V_1$ of $x_0$ with $\overline{V_0}\subset V_1$ such that $$ \|A - I\| \le 1/3 \text{ on } V_0 \quad \text{ and } \quad \|A - I\| \le 2/3 \text{ on } V_1 $$ and then take $B=t\chi I$ for some suitable $t\in \mathbb{R}$ where $\chi $ is smooth and $\chi = 0$ on $V_0$ and $\chi = 1$ outside $V_1$. I tried making $A+B$ within distance 1 to identity but did not succeed.

Any ideas?

share|cite|improve this question
What does it mean for $B$ to be supported away from $x_0?$ – Igor Rivin Apr 20 '12 at 14:12
All I want is that $x_0$ does not belong to the support of $B$ where the support of $B$ is the closure of $\{x: B(x)\neq 0 \}$. – flavio Apr 20 '12 at 14:33
up vote 2 down vote accepted

Define $f(x)$ to be the least real eigenvalues of $A(x)$, then $f$ is defined in a neighborhood $B(x_0,\epsilon)$ of $x_0$ and $f(x)>1/2$ for any $x\in B(x_0,\epsilon)$. Suppose $M>\sup_x||A(x)||$, then choose $\chi(x)$ to be a nonnegative function such that $\chi$ is smooth, $\chi(x)=0$ for any $x\in B(x_0,\epsilon/2)$, and $\chi(x)=M$ for any $x\notin B(x_0,\epsilon)$. Put $B(x)=\chi(x)I$, then $B$ satisfies your requirement, since all real eigenvalues of $A+B$ is always positive, hence $A+B$ is always invertible.

share|cite|improve this answer
Thanks a lot! Your solution appears to work very well. – flavio Apr 21 '12 at 9:57
Is there a similar argument if A is merely assumed to be invertible at the point $x_0$, so there is also the possibility of one eigenvalue being negative at $x_0$? – flavio Apr 28 '12 at 20:13
Apply the proof to $\overline{A}(x)=A(x)(A(x_0))^{-1}$, since we have $\overline{A}(x_0)=I$, we can always find a $\overline{B}(x)$ such that $\overline{A}(x)+\overline{B}(x)$ is invertible, then put $B(x)=\overline{B}(x)A(x_0)$. – Yuchen Liu Apr 29 '12 at 9:16
Of course, so obvious once you see it! Thanks! – flavio Apr 30 '12 at 13:57

Your idea works. Instead of trying to put A+B within distance 1 of the identity, just make t large and positive.

share|cite|improve this answer
Yes, I realized this. Thanks! – flavio Apr 21 '12 at 9:58

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.