Definition

Let XX and YY be , and let T:D(T)XYT:D(T)\subseteq X\to Y be linear on the D(T)D(T). The operator TT is closed if its

Γ(T)={(x,Tx):xD(T)}\Gamma(T)=\{(x,Tx):x\in D(T)\}

is a in X×YX\times Y. Equivalently, whenever xnD(T)x_n\in D(T), xnxx_n\to x in XX, and TxnyTx_n\to y in YY, one has xD(T)x\in D(T) and Tx=yTx=y. Closedness constrains simultaneous convergence of inputs and outputs; it neither says that D(T)D(T) is closed in XX nor implies that TT is bounded on D(T)D(T) with the norm inherited from XX.

Closed versus bounded

The says that a closed linear operator T:XYT:X\to Y defined on all of a XX, with YY Banach, is bounded. Properly defined closed operators can be unbounded. Differential operators on LpL^p- or are principal examples: their domains encode the regularity and boundary conditions needed for the graph to be closed.

Closure and closability

An operator is closable if the closure of Γ(T)\Gamma(T) is itself the graph of an operator. This happens exactly when

xn0,Txnyy=0.x_n\to0,\quad Tx_n\to y \quad\Longrightarrow\quad y=0.

The operator whose graph is Γ(T)\overline{\Gamma(T)} is the closure T\overline T, the smallest closed extension of TT. Hence “closed” and “closable” are not synonyms.

Domain-sensitive operations

For unbounded operators, algebraic expressions have domain conditions. The product STST is defined only on those xD(T)x\in D(T) for which TxD(S)Tx\in D(S), and a sum requires a common domain. Closed operators are therefore not automatically preserved by sums or products. Statements about adjoints, resolvents, or self-adjointness must likewise include density and domain hypotheses.

References
  1. Tosio Kato, Perturbation Theory for Linear Operators, Springer, 1995 reprint of the 2nd ed. Springer DOI record. Relevant: Chapter III.
  2. Michael Reed and Barry Simon, Methods of Modern Mathematical Physics, Volume I: Functional Analysis, Academic Press, 1980. Elsevier book record. Relevant: Chapter VIII.