Let g be a μ-measurable function on X. The adjoint of the multiplication operator mg is mg.
For every integer n⩾ 1, let φn be the characteristic function of the set of elements x∈ X such that ∣g(x)∣⩽n, and let φn be its class in L2(X,μ). Let f∈L2(X,μ) be such that its class f belongs to dom(mg∗), and let ψ be a function whose class is mg∗(f).
For every h∈L2(X,μ) whose class h belongs to dom(mg), we have also hφn∈ dom(mg), and therefore ⟨f∣mg(hφn)⟩=⟨mg∗(f)∣hφn⟩. This yields the equality
∫X(fg−ψ)φnhdμ=0
Since the domain of mg is dense in L2(X,μ), it follows that (fg−ψ)φn is zero μ-almost everywhere. Since n is arbitrary, this means that mg∗(f) is the class in L2(X,μ) of fg. In particular, since mg∗(f)∈L2(X,μ), one concludes that f belongs to the domain of mg and that mg∗(f)=mg(f).
The adjoint of mg is therefore an extension of mg. Moreover, one has
⟨f∣mg(h)⟩=∫Xf⋅(gh)dμ=∫Xfghdμ
for every f∈L2(X,μ) and h∈ dom(mg), which proves that the linear form h↦⟨f∣mg(h)⟩ is continuous whenever mg(f)g belongs to L2(X,μ). Consequently the domain of mg is contained in that of mg∗, which concludes the proof.
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.