Definition

Let XX and YY be over the same scalar field R\mathbb R or C\mathbb C. A T:XYT:X\to Y is a finite-rank operator if its range T(X)T(X) is finite-dimensional. Its rank is

rankT=dimT(X).\operatorname{rank}T=\dim T(X).

Thus the zero operator has rank 00, and TT has rank nn precisely when its range has a of nn vectors. Boundedness is part of the operator-theoretic convention used here; an algebraic can have finite-dimensional range without being continuous.

Rank-one decompositions

An operator has finite rank exactly when it can be written

Tx=j=1nφj(x)yjTx=\sum_{j=1}^{n}\varphi_j(x)y_j

for finitely many continuous linear functionals φjX\varphi_j\in X' and vectors yjYy_j\in Y. Choosing the yjy_j as a basis for T(X)T(X) gives such a representation, while every displayed sum has range contained in their span. Rank-one operators xφ(x)yx\mapsto\varphi(x)y are the elementary building blocks Conway, Chapter II.

Approximation and examples

Finite-rank operators form a of B(X,Y)B(X,Y), are compact, and are stable under composition with bounded operators. Their operator-norm closure is the class of approximable operators; it need not contain every for arbitrary . On , compact operators are operator-norm limits of finite-rank operators, and finite-rank operators lie in every .

The identity on XX is finite-rank exactly when XX is finite-dimensional. A nonzero functional φX\varphi\in X' and vector yYy\in Y give a rank-one operator xφ(x)yx\mapsto\varphi(x)y.

References
  1. John B. Conway, A Course in Functional Analysis, 2nd ed., Graduate Texts in Mathematics 96, Springer, 1990. Springer DOI record. Relevant: Chapter II, “Operators on Hilbert Space,” and the discussion of finite-rank and compact operators.
  2. Michael Reed and Barry Simon, Methods of Modern Mathematical Physics I: Functional Analysis, revised and enlarged ed., Academic Press, 1980. Elsevier book record. Relevant: Chapter VI, “Bounded Operators.”