a) Let g′ be a μ-measurable function on X such that ∣g∣⩽∣g′∣. One has dom(mg′)⊂ dom(mg) and dom(mg′) is a core of the partial operator mg;
b) Let F be a subspace of LK2(X,μ) whose intersection with K(X;K) is dense in K(X;K) and whose image G in LK2(X,μ) is contained in dom(mg). If ∣g∣2 is locally μ-integrable, then K(X;K) is contained in the domain of mg and G is a core of mg.
Let us prove a). If f∈LK2(X,μ) belong to the domain of mg′, so that fg′∈LK2(X,μ), the assumption implies that fg∈LK2(X,μ), whence the result.
Let us prove that dom(mg′) is a core of mg, that is to say that the domain of mg′ is dense in the Hilbert space Emg. Let h∈LK2(X,μ)
whose class h belong to Emg and is orthogonal to dom(mg′). This means that
(h∣h′)mg=∫Xhh′(1+∣g∣2)dμ=0
for every function h′∈LK2(X,μ) whose class h′ belongs to dom(mg′).
Let C be a compact subset of X and let φ be its characteristic function. Let n∈N. Denote by φn the characteristic function of the μ-integrable set Cn of the x∈C such that ∣h(x)∣⩽n and put hn′=φnh. The class of hn′ belongs to the domain of mg′ since ∣g′hn′∣⩽nφ; hence
0=∫Xhhn′(1+∣g∣2)dμ=∫X∣h∣2φn(1+∣g∣2)dμ
This implies that h is zero for μ-almost every x∈Cn and therefore, since n is arbitrary, that h is zero for μ-almost every x∈C. It follows finally that h is zero μ-almost everywhere, since C is arbitrary and h is moderate (INT, V, p. 9, § 1, no3, cor.).
Let us now consider assertion b). Since ∣g∣2 is locally μ-integrable, the function fg belongs to LK2(X,μ) if f∈K(X;K), hence K(X;K) is contained in the domain of mg.
Let h∈LK2(X,μ) whose class h belongs to Emg and is orthogonal to G. One then has
0=(h∣h′)mg=∫Xhh′(1+∣g∣2)dμ
for every h′∈F of class h′. In view of the assumption on F, this means that the measure h(1+∣g∣2)⋅μ is zero, hence that h is zero μ-almost everywhere since h is moderate.
Let p be a real number ⩾1. Let h be an element of LK∞(X,μ). The multiplication operator mh by h is an endomorphism of LKp(X,μ)(IV, p. 186, No.o5). Suppose that the set Y of the x∈X such that h(x)=0 is locally μ-negligible. Then the endomorphism mh is injective (Lemma 7 of IV, p. 186). Let us denote by h−1 the function on X equal to 0 on Y and to x↦1/h(x) on X- Y. The reciprocal partial operator mh−1 is the multiplication operator by h−1 from LKp(X,μ) into LKp(X,μ), that is to say, mh−1=mh−1. In fact, the image of mh is the space of classes of functions g∈LKp(X,μ) of the form g=hf for f∈LKp(X,μ). This condition is equivalent to g(x)/h(x)=f(x) for every x∈X- Y and g(x)=0 if x∈Y. This implies that the domain of mh−1 in LKp(X,μ) is the domain of mh−1, and that the equality mh−1=mh−1 is valid.
In what follows, we shall sometimes denote simply by mh the partial multiplication operator by h from LKp1(X,μ) into LKp2(X,μ).
Bourbaki sets the proof directly after the statement, and the corpus does not separate them, so what is above is the statement and its proof together.