Perturbation of Identity is isomorphism

676 Views Asked by At

Let $X$ be a Banach space and $K:X\to X$ be a linear, bounded operator with $\Vert K\Vert<1$. Show that $(I-K)^{-1}$ exists and is bounded.

Let $\Vert K\Vert = 1-\varepsilon$. Then we get $$\varepsilon\Vert x\Vert \leq \Vert (I-K)x\Vert \leq (2-\varepsilon)\Vert x\Vert,$$ so $I-K$ is bounded and injective. If we have surjectivity, it follows that $I-K$ is an isomorphism since $X$ is Banach. While it seems clear that it should be surjective, I have so far failed to prove it.

1

There are 1 best solutions below

0
On

Consider the sequence

$S_n = \displaystyle \sum_0^n K^i = I + K + K^2 + \ldots + K^n \in \mathcal B(X), \tag 1$

where $\mathcal B(X)$ is the Banach space of bounded linear operators $X \to X$, with the operator norm. I claim $S_n$ is Cauchy in $\mathcal B(X)$; for with $m > n$ we have

$S_m - S_n = \displaystyle \sum_0^m K^i - \sum_0^n K^i = \sum_{n + 1}^m K^i = K^{n + 1} \sum_0^{m - n - 1} K^i, \tag 2$

whence

$\Vert S_m - S_n \Vert = \left \Vert K^{n + 1} \displaystyle \sum_0^{m - n - 1} K^i \right \Vert \le \Vert K \Vert^{n + 1} \left \Vert \displaystyle \sum_0^{m - n - 1} K^i \right \Vert \le \Vert K \Vert^{n + 1} \displaystyle \sum_0^{m - n - 1} \Vert K \Vert^i; \tag 3$

now since

$\Vert K \Vert < 1, \tag 4$

it follows that the strictly increasing geometric series

$L = \displaystyle \sum_0^\infty \Vert K \Vert ^i \tag 5$

converges; furthermore, we have

$L - 1 = \displaystyle \sum_1^\infty \Vert K \Vert^i = \Vert K \Vert \sum_0^\infty \Vert K \Vert^i = \Vert K \Vert L, \tag 6$

whence

$L(1 - \Vert K \Vert) = 1, \tag 7$

or

$L = \dfrac{1}{1 - \Vert K \Vert}; \tag 8$

incorporating formulas (5)-(8) into (3) yields

$\Vert S_m - S_n \Vert \le \Vert K \Vert^{n + 1} \displaystyle \sum_0^{m - n - 1} \Vert K \Vert^i \le \Vert K \Vert^{n + 1} \sum_0^\infty \Vert K \Vert^i = \Vert K \Vert^{n + 1} L = \dfrac{\Vert K \Vert^{n + 1}}{1 - \Vert K \Vert}; \tag 9$

since $\Vert K \Vert < 1$, the right-hand side of this equation may be made arbitrarily small by taking $n$ sufficiently large; which shows that $S_n$ is indeed Cauchy in $\mathcal B(X)$; since $X$ is Banach, so is $\mathcal B(X)$ in the operator norm, hence $S_n$ is possessed of a limit $S \in \mathcal B(X)$; and indeed, said limit is in fact

$S = \displaystyle \sum_0^\infty K^i, \tag{10}$

since, in similarity to (9),

$\Vert S - S_n \Vert = \left \Vert \displaystyle \sum_{n + 1}^\infty K^i \right \Vert \le \displaystyle \sum_{n + 1}^\infty \Vert K \Vert^i = \Vert K \Vert^{n + 1} \sum_0^\infty \Vert K \Vert^i = \dfrac{\Vert K \Vert^{n + 1}}{1 - \Vert K \Vert} \to 0 \; \text{as} \; n \to \infty. \tag{11}$

Finally, we have

$(I - K)S_n = (I - K) \displaystyle \sum_0^n K^i = \sum_0^n K^i - \sum_1^{n + 1}K^i = I - K^{n + 1}; \tag{12}$

thus,

$\Vert (I - K)S - I \Vert = \displaystyle \lim_{n \to \infty} \Vert (I - K)S_n - I \Vert = \lim_{n \to \infty} \Vert K^{n + 1} \Vert \le \lim_{n \to \infty} \Vert K \Vert^{n + 1} = 0, \tag{13}$

and we see that, by virtue of the fact that $KS = SK$ (cf. (10)),

$S(I - K) = (I - K)S = I, \tag{14}$

or

$(I - K)^{-1} = S = \displaystyle \sum_0^\infty K^i. \tag{15}$

It is easy to see that $S$ is bounded, since

$\Vert S \Vert = \left \Vert \displaystyle \sum_0^\infty K^i \right \Vert \le \displaystyle \sum_0^\infty \Vert K \Vert^i = L. \tag{16}$

This whole spiel is of course a re-iteration of the standard theory.

In closing, I have a brief query of my own about our OP Analysis801's question; I would like to know how the left-most inequality in

$\epsilon \Vert x \Vert \le \Vert (I - K)x \Vert \le (2 - \epsilon) \Vert x \Vert \tag{17}$

comes about. I can derive the right-hand inequality myself; from

$\Vert K \Vert = 1 - \epsilon, \tag{18}$

we have

$\Vert I - K \Vert \le \Vert I \Vert + \Vert K \Vert = 1 + 1 - \epsilon = 2 - \epsilon; \tag{19}$

thus,

$\Vert (I - K) x \Vert \le \Vert I - K \Vert \Vert x \Vert \le (2 - \epsilon) \Vert x \Vert; \tag{21}$

what I haven't been able to figure out is why

$\epsilon \Vert x \Vert \le \Vert (I - K)x \Vert; \tag{22}$

the best I have been able to manage is,

$\epsilon \Vert x \Vert = (1 - (1 - \epsilon)) \Vert x \Vert = \vert \Vert I \Vert - \Vert K \Vert \vert \Vert x \Vert \le \Vert I - K \Vert \Vert x \Vert, \tag{23}$

but I haven't been able to show that

$\epsilon \Vert x \Vert \le \Vert (I - K)x \Vert; \tag{24}$

so if anyone wants to leave remarks, proofs or counterexamples in the comments, I would be most appreciative.