Definition

Let (A,H,D)(\mathcal A,H,D) be a , with DD densely defined and self-adjoint. Its bounded transform is the bounded self-adjoint contraction

FD=D(1+D2)1/2,F_D=D(1+D^2)^{-1/2},

defined by functional calculus. gives FD2I=(1+D2)1F_D^2-I=-(1+D^2)^{-1} compact, and the bounded-commutator axiom implies that [FD,a][F_D,a] is compact for every aAa\in\mathcal A. Thus (H,π,FD)(H,\pi,F_D) is a in the compact-defect convention. If the spectral triple is even, its grading anticommutes with FDF_D; otherwise the transform is odd.

Why the transform is bounded

The scalar function

f(x)=x1+x2f(x)=\frac{x}{\sqrt{1+x^2}}

is real-valued and bounded by 11. The self-adjoint functional calculus therefore turns the generally unbounded DD into a bounded self-adjoint operator. Its square tends to 11 at spectral infinity, while

IFD2=(1+D2)1I-F_D^2=(1+D^2)^{-1}

retains the compactness information. The transform preserves the sign of every nonzero spectral value but sends large magnitudes toward ±1\pm1.

The compactness of [FD,a][F_D,a] follows from the resolvent integral for (1+D2)1/2(1+D^2)^{-1/2} and the boundedness of [D,a][D,a]. This is the bounded-transform theorem of Baaj and Julg; Connes, Appendix A, Theorem 15 states the Hilbert-module version and the integral argument.

K-homological role

The bounded transform forgets metric scale but retains the stable Fredholm data needed for . Homotopies satisfying the appropriate uniform unbounded-cycle hypotheses descend to operator homotopies of bounded cycles. The resulting class can then be paired with KK-theory by the even or odd Fredholm index.

For a nonunital unbounded Kasparov cycle, compact resolvent is replaced by local compactness a(1+D2)1K(H)a(1+D^2)^{-1}\in K(H). The same formula then gives the local compact-defect conditions a(FD2I)K(H)a(F_D^2-I)\in K(H). Connes, Appendix A, Theorem 15.

Example and distinction from the phase

For the spin on a closed Riemannian spin manifold,

D(1+D2)1/2D(1+D^2)^{-1/2}

is an order-zero bounded operator and defines the manifold's analytic K-homology class. Its commutators with smooth functions are compact.

The bounded transform is not literally the phase sgn(D)\operatorname{sgn}(D). On kerD\ker D, both may be assigned 00, but on nonzero finite spectral values x/1+x2sgn(x)x/\sqrt{1+x^2}\neq\operatorname{sgn}(x). The two yield related normalized cycles only after an additional compact perturbation or kernel convention.

References