Statement

Let EE be a right over a CC^*-algebra AA, with inner product linear in the second variable. For all x,yEx,y\in E,

x,yAx,yAx,xAy,yA\langle x,y\rangle_A^*\langle x,y\rangle_A \leq \lVert\langle x,x\rangle_A\rVert\,\langle y,y\rangle_A

in the order on positive elements of AA. Taking norms gives

x,yAxy,x=x,xA1/2.\lVert\langle x,y\rangle_A\rVert \leq \lVert x\rVert\,\lVert y\rVert, \qquad \lVert x\rVert=\lVert\langle x,x\rangle_A\rVert^{1/2}.

This is the Hilbert-module Cauchy--Schwarz inequality. It makes the algebra-valued inner product jointly continuous and justifies the triangle inequality for the induced norm.

Proof idea

After adjoining a unit if necessary, positivity of yxa,yxaA\langle y-xa,y-xa\rangle_A is applied with

a=(x,xA+ε1)1x,yA.a=(\langle x,x\rangle_A+\varepsilon1)^{-1}\langle x,y\rangle_A.

Expanding the positive element and letting ε0\varepsilon\downarrow0 yields the displayed order inequality. The scalar proof by direct division cannot simply be copied, because y,yA\langle y,y\rangle_A need not be invertible or commute with x,yA\langle x,y\rangle_A Lance, Proposition 1.1.

Consequences

For fixed xx, the map yx,yAy\mapsto\langle x,y\rangle_A is bounded with norm at most x\lVert x\rVert. Hence the inner product extends uniquely across completion, and

x,yAx,yAxxy+xyy.\lVert\langle x,y\rangle_A-\langle x',y'\rangle_A\rVert \leq \lVert x-x'\rVert\lVert y\rVert +\lVert x'\rVert\lVert y-y'\rVert.

The theorem also proves the for the module norm by expanding x+y,x+yA\langle x+y,x+y\rangle_A.

Conventions and scope

If the inner product is linear in the first variable, the order-valued formula is written with the variables and factors reversed. Unlike the scalar , equality does not in general admit a simple characterization by linear dependence, because coefficients lie in a possibly noncommutative algebra.

References
  1. E. C. Lance, Hilbert CC^*-Modules: A Toolkit for Operator Algebraists, Cambridge University Press, 1995. DOI record. Relevant: Proposition 1.1 on the module Cauchy--Schwarz inequality and induced norm.