The transfer operator PTP_T of a probability-preserving transformation is the PT=UT∗P_T=U_T^* on L2L^2. With inner products linear in the first argument, it is characterized by

∫(PTf)h‾ dμ=∫f h∘T‾ dμ.\int (P_T f)\overline h\,d\mu =\int f\,\overline{h\circ T}\,d\mu.

It is a positive contraction and PT1=1P_T1=1. If TT is invertible, PTf=f∘T−1P_Tf=f\circ T^{-1}.

Doubling-map example

For T(x)=2x mod 1T(x)=2x\bmod1,

PTf(x)=12(f(x/2)+f((x+1)/2)).P_T f(x)=\tfrac12\bigl(f(x/2)+f((x+1)/2)\bigr).

The operator averages over the two preimages. It is not composition with a nonexistent inverse map. The same duality extends to a positive L1L^1 operator transporting densities relative to the invariant probability measure.