logo

Extensions of Bounded Linear Operators 📂Banach Space

Extensions of Bounded Linear Operators

Theorem1 2

Let V1,V2V_{1}, V_{2} be a Banach space. Let WV1W \subset V_{1} be a dense subspace. And let T:WV2T : W \to V_{2} be a bounded linear operator. Then for all vW\mathbf{v} \in W, there exists a unique bounded linear operator that satisfies

T~:V1V2 \widetilde{T} : V_{1} \to V_{2} .

Moreover, the following holds:

T~=T \| \widetilde{T} \| = \left\| T \right\| .

Explanation

T~\widetilde{T} is called the extension of TT.

1.gif Although it is not specified what T~\widetilde{T} looks like in the theorem, it is basically defined through the convergence of sequences in dense subspaces during the proof process. By extending this way, the operator T~\widetilde{T} can be naturally denoted as TT.

When there are bounded linear operators T:WVT : W \to V and wWW\mathbf{w} \in \overline{W}\setminus W, we wish to define TwT \mathbf{w}, but since the domain of TT is limited to WW, it’s not possible. However, the sequence {wk}\left\{ \mathbf{w}_{k} \right\} converging to w\mathbf{w} exists within WW, so TwT \mathbf{w} can be ‘indirectly’ defined in the following plausible manner:

T(w):=limkT(wk) T (\mathbf{w}) := \lim \limits_{k \to \infty} T(\mathbf{w}_{k})

To paraphrase,

Since w\mathbf{w} is originally not part of the domain of TT, TwT \mathbf{w} cannot be defined. However, there exists a sequence {wk}\left\{ \mathbf{w}_{k} \right\} that converges to w\mathbf{w} within the domain of TT. Therefore, when wkw\mathbf{w}_{k} \to \mathbf{w}, it will be TwkTwT \mathbf{w}_{k} \to T \mathbf{w}, so defining it like Tw:=limkTwkT \mathbf{w} := \lim\limits_{k\to \infty}T \mathbf{w}_{k} is natural and plausible. And this is indeed possible.

Thus, it is guaranteed from the above theorem that a bounded linear operator T:WVT : W \to V defined on a Banach space WW can expand its domain in the same way up to the closure W\overline{W} of WW while maintaining the mapping to VV.

Proof

Let’s denote by vV1\mathbf{v} \in V_{1}. Assuming that WW is dense in V1V_{1}, there exists a sequence {vk}\left\{ \mathbf{v}_{k} \right\} that converges to v\mathbf{v}.

limkvk=v \lim \limits_{k \to \infty} \mathbf{v}_{k} = \mathbf{v}

Since TT is bounded and linear, the following holds for ,kN\ell, k \in \N:

TvkTv=T(vkv)Tvkv \left\| T \mathbf{v}_{k} - T \mathbf{v}_{\ell} \right\| = \left\| T (\mathbf{v}_{k} - \mathbf{v}_{\ell}) \right\| \le \left\| T \right\| \left\| \mathbf{v}_{k} - \mathbf{v}_{\ell} \right\|

Now, since {vk}\left\{ \mathbf{v}_{k} \right\} is a Cauchy sequence, by the above equation, {Tvk}\left\{ T \mathbf{v}_{k} \right\} also becomes a Cauchy sequence. Therefore, since V2V_{2} is a Banach space, {Tvk}\left\{ T \mathbf{v}_{k} \right\} will converge to some element within V2V_{2}. From this, define T~\widetilde{T} as follows:

T~v:=limkTvk \widetilde{T} \mathbf{v} := \lim \limits_{k \to \infty} T\mathbf{v}_{k}

Then, due to the properties of the bounded linear operator for all vW\mathbf{v} \in W, TvkTvT \mathbf{v}_{k} \to T \mathbf{v} holds, thus satisfying T~v=Tv\widetilde{T} \mathbf{v} = T \mathbf{v}. Now, consider the case of vV1W\mathbf{v} \in V_{1}\setminus W.

  • Part 1. The irrelevance of sequence choice

    Consider two sequences {vk},{uk}\left\{ \mathbf{v}_{k} \right\}, \left\{ \mathbf{u}_{k} \right\} of WW that converge to v\mathbf{v}. If v\mathbf{v} were an element within WW, since TT is a bounded linear operator, when vk,ukvW\mathbf{v}_{k}, \mathbf{u}_{k} \to \mathbf{v} \in W, TvkT \mathbf{v}_{k} and TukT \mathbf{u}_{k} would converge to the same value TvT \mathbf{v}. However, in the case of vk,ukvV1W\mathbf{v}_{k}, \mathbf{u}_{k} \to \mathbf{v} \in V_{1}\setminus W, this property cannot be used, so it must be directly confirmed whether they converge to the same value.

    Now consider the following sequence {wk}\left\{ \mathbf{w}_{k} \right\}:

    {wk}={v1,u1,v2,u2,} \left\{ \mathbf{w}_{k} \right\} = \left\{ \mathbf{v}_{1}, \mathbf{u}_{1}, \mathbf{v}_{2}, \mathbf{u}_{2}, \dots \right\}

    Then limkwk=v\lim \limits_{k \to \infty} \mathbf{w}_{k} = \mathbf{v} holds, and TwkT \mathbf{w}_{k} also converges to some value T~v=limkTwk\widetilde{T} \mathbf{v} = \lim \limits_{k \to \infty} T\mathbf{w}_{k}. Since TvkT \mathbf{v}_{k} and TukT \mathbf{u}_{k} are subsequences of TwkT\mathbf{w}_{k}, they both must have the same limit value. Therefore, regardless of which sequence converging to v\mathbf{v} is taken, T~v\widetilde{T} \mathbf{v} is uniquely determined.

  • Part 2. Linearity

    Let v,wV1\mathbf{v}, \mathbf{w} \in V_{1} and suppose vkv,wkw\mathbf{v}_{k} \to \mathbf{v}, \mathbf{w}_{k} \to \mathbf{w}.

    T~(αv+βw)= limkT(αvk+βwk)= αlimkT(vk)+βlimkT(wk)= αT~(v)+βT~(w) \begin{align*} \widetilde{T} \left( \alpha \mathbf{v} + \beta \mathbf{w} \right) =&\ \lim \limits_{k \to \infty} T \left( \alpha \mathbf{v}_{k} + \beta \mathbf{w}_{k} \right) \\ =&\ \alpha \lim \limits_{k \to \infty} T ( \mathbf{v}_{k} ) + \beta \lim \limits_{k \to \infty} T ( \mathbf{w}_{k} ) \\ =&\ \alpha \widetilde{T} \left(\mathbf{v}\right) + \beta \widetilde{T} \left(\mathbf{w} \right) \end{align*}

  • Part 3. Bound, T~=T\| \widetilde{T} \| = \left\| T \right\|

    Let’s denote by vV1\mathbf{v} \in V_{1} and suppose vkv\mathbf{v}_{k} \to \mathbf{v}. Since the norm is continuous, limits can be interchanged. Using this fact and the boundedness of TT,

    T~v= limkTvk= limkTvklimkTvk= Tlimkvk= Tv \begin{align*} \left\| \widetilde{T} \mathbf{v} \right\| =&\ \left\| \lim \limits_{k \to \infty} T \mathbf{v}_{k} \right\| \\ =&\ \lim \limits_{k \to \infty} \left\| T \mathbf{v}_{k} \right\| \\ \le& \lim \limits_{k \to \infty} \left\| T \right\| \left\| \mathbf{v}_{k} \right\| \\ =&\ \left\| T \right\| \left\| \lim \limits_{k \to \infty} \mathbf{v}_{k} \right\| \\ =&\ \left\| T \right\| \left\| \mathbf{v} \right\| \end{align*}

    Therefore, T~\widetilde{T} is bounded, and T~T\| \widetilde{T} \| \le \left\| T \right\| is valid. Now let’s show that the opposite inequality holds. For vectors that are vW\mathbf{v} \in W, the following holds:

    T~v=TvTv \| \widetilde{T} \mathbf{v} \| = \| T \mathbf{v} \| \le \| T \| \| \mathbf{v} \|

    However, for vectors that are vV1W\mathbf{v} \in V_{1} \setminus W, there can be vectors that are not bound as above by T\| T \|. Hence, the validity of T~T\| \widetilde{T} \| \ge \left\| T \right\| can be known. Therefore, since both sides of the inequality hold, we obtain:

    T~=T\| \widetilde{T} \| = \left\| T \right\|


  1. Ole Christensen, Functions, Spaces, and Expansions: Mathematical Tools in Physics and Engineering (2010), p55 ↩︎

  2. Erwin Kreyszig, Introductory Functional Analysis with Applications (1989), p99-101 ↩︎