If $A$ and $H$ is a two $n\times n$ matrixs, such that $\det(A)\neq 0$ and $ \det(H)\neq 0$, what happen for $H A H^{-1}$, if $h(i,j)\longrightarrow 0 $? Is $H A H^{-1}\cong A$?$ \ \ \ (1\leq i,j\leq n).$ Any idea can be helpful.thanks
-
What is the meaning of $\simeq$? – user251257 Aug 25 '15 at 21:44
-
Does $\cong$ mean that they are congruent or asymptotically equivalent? What do you mean by $h(i,j) \rightarrow 0$ - that there is a sequence of $H$'s whose components tend to zero? – Chester Aug 26 '15 at 00:04
-
What is $h(i, j)$? – anomaly Aug 26 '15 at 06:49
2 Answers
This is a wrong question because the limit of $HAH^{-1}$ in general may not exist. The right question should be: provided that $\lim_{H\to 0}HAH^{-1}$ exists, is it necessarily similar to $A$?
The answer, unfortunately, is no. Here is a counterexample: $$ HAH^{-1} = \pmatrix{x^2&0\\ 0&x} \pmatrix{1&1\\ 0&1} \pmatrix{\frac1{x^2}&0\\ 0&\frac1x} = \pmatrix{1&x\\ 0&1}. $$
If you mean that a certain entry of $H$ approaches to zero while the others are kept constant, the answer is still no. We only need to modify the above counterexample by a little bit: $$ HAH^{-1} = \pmatrix{x&0\\ 0&1} \pmatrix{1&1\\ 0&1} \pmatrix{\frac1x&0\\ 0&1} = \pmatrix{1&x\\ 0&1}. $$
Edit. Note that the $A$ above is not diagonalisable. In general, the similarity orbit of a real or complex matrix $A$ is closed if and only if $A$ is diagonalisable over $\mathbb C$.
I assume that $h(i,j)$ denotes the coefficients of $H$.
In the case $n=1$ the product $HAH^{-1}$ is a product of numbers, in which $H$ converges to $0$. Then obviously the product converges to $A$.
In the case $n>1$ one first observes, that matrix multiplication is a continous operation considering for example the euclidean metric on $\mathbb{R}^{n\times n}$ (or any $p$-metric, $p\in\mathbb{N}\cup\infty$). Convergence with respect to that metric is the same as coefficient-wise concergence.
By the theorem of Cayley-Hamilton $H^n+a_{n-1}H^{n-1}+\ldots +a_0=0$ for some real numbers $a_k$, where $a_0\neq 0$ by assumption (up to sign $a_0$ is the determinant of $H$). Hence
$H^{-1}=-\frac{1}{a_0}(H^{n-1}+a_{n-1}H^{n-2}+\ldots +a_1)$.
Consequently
$\begin{array}{rcl}\lim_{H\rightarrow 0}(HAH^{-1})&=&(\lim_{H\rightarrow 0}H)A(\lim_{H\rightarrow 0}-\frac{1}{a_0}(H^{n-1}+a_{n-1}H^{n-2}+\ldots +a_1))\\ &=&0\;A\;(-\frac{a_1}{a_0}E)=0, \end{array}$
where $E$ is the unit matrix.

- 8,932
-
1that can't be right. Take $A = I$, the identity, $H=hI$ for some scalar $h\ne 0$. Then, $HAH^{-1} = I$. – user251257 Aug 25 '15 at 23:52
-
1This isn't right. One can construct a counterexample where some of the elements of $HAH^{-1}$ diverge. – Chester Aug 26 '15 at 00:06
-
@ Hagen Knaf . What a beautiful mistake ! Your limit $-a_1/a_0 I_n$ varies because it depends on $a_0,a_1$ that both tend to $0$. I think it's better that you remove your answer. – Mar 31 '16 at 15:07