Throughout, the functions 0, f+g, min(g,f), f−g, cf, f+ and f− formed below from elements of Cc(X) lie in Cc(X) by Continuous Functions with Compact Support: the Zero Function, Closure under Operations, Vanishing off the Support, Existence of the Maximum Modulus, and Uniform Continuity (i) and Continuous Functions with Compact Support: the Zero Function, Closure under Operations, Vanishing off the Support, Existence of the Maximum Modulus, and Uniform Continuity (ii). Write M=∥Λ∥∗; the dual norm is a bound for Λ,
∣Λg∣≤M∥g∥u(g∈Cc(X)).
Let Cc+ be the set of f∈Cc(X) with f≥0.
The supremum. Let f∈Cc+ and let S(f)={Λg:g∈Cc(X), 0≤g≤f}. It contains Λ0=0. If 0≤g≤f, then ∣g(x)∣≤f(x)≤∥f∥u for every x∈X, so ∥g∥u≤∥f∥u, and Λg≤M∥g∥u≤M∥f∥u. Thus S(f) is nonempty and bounded above, and has a supremum p(f)=supS(f). Since 0∈S(f) and Λf∈S(f),
0≤p(f)≤M∥f∥u,Λf≤p(f)(f∈Cc+).(1)
Moreover S(0)={Λ0}, so p(0)=0.
Additivity on nonnegative functions. Let f1,f2∈Cc+. If 0≤g1≤f1 and 0≤g2≤f2, then 0≤g1+g2≤f1+f2, so Λg1+Λg2=Λ(g1+g2)≤p(f1+f2); taking suprema over g1 and then over g2 gives p(f1)+p(f2)≤p(f1+f2). Conversely, let 0≤g≤f1+f2, and put
g1=min(g,f1),g2=g−g1.
Then 0≤g1≤f1 and g2≥0. At a point x with g(x)≤f1(x) we have g2(x)=0≤f2(x), and at any other point g2(x)=g(x)−f1(x)≤f2(x); so 0≤g2≤f2. Hence Λg=Λg1+Λg2≤p(f1)+p(f2), and taking the supremum over g,
p(f1+f2)=p(f1)+p(f2)(f1,f2∈Cc+).(2)
Homogeneity on nonnegative functions. Let f∈Cc+ and c≥0. For c=0, p(0f)=p(0)=0. For c>0, g↦cg is a bijection from {g∈Cc(X):0≤g≤f} onto {g∈Cc(X):0≤g≤cf}, and Λ(cg)=cΛg; hence S(cf)={cs:s∈S(f)} and
p(cf)=cp(f)(f∈Cc+, c≥0).(3)
Extension to Cc(X). Every f∈Cc(X) is f=f+−f− with f+,f−∈Cc+. If also f=u−v with u,v∈Cc+, then f++v=u+f−, and (2) gives p(f+)+p(v)=p(u)+p(f−), that is,
p(f+)−p(f−)=p(u)−p(v).(4)
Define Λ+f=p(f+)−p(f−) for f∈Cc(X). For f∈Cc+ we have f+=f and f−=0, so Λ+f=p(f), which is the displayed supremum formula.
Linearity and positivity of Λ+. Let f,g∈Cc(X). Since f+g=(f++g+)−(f−+g−) with both terms in Cc+, (4) and (2) give
Λ+(f+g)=p(f++g+)−p(f−+g−)=Λ+f+Λ+g.
Let c∈R. If c≥0, then cf=cf+−cf−, and (4) and (3) give Λ+(cf)=cp(f+)−cp(f−)=cΛ+f. If c<0, then cf=(−c)f−−(−c)f+, and likewise Λ+(cf)=(−c)p(f−)−(−c)p(f+)=cΛ+f. So Λ+ is a linear functional on Cc(X), and it is positive by (1).
The negative part. Define Λ−f=Λ+f−Λf for f∈Cc(X). As a difference of linear functionals, Λ− is a linear functional, and Λf=Λ+f−Λ−f for every f∈Cc(X). For f∈Cc+, (1) gives Λ−f=p(f)−Λf≥0, so Λ− is positive. For f∈Cc+ and 0≤g≤f we have 0≤f−g≤f, hence ∥f−g∥u≤∥f∥u as in the first step, and
Λg−Λf=−Λ(f−g)≤M∥f∥u;
taking the supremum over g gives Λ−f=p(f)−Λf≤M∥f∥u.
Bounds. Let Q be Λ+ or Λ−. By (1) and the previous step, 0≤Qf≤M∥f∥u for every f∈Cc+. Let f∈Cc(X). For every x∈X, 0≤f±(x)≤∣f(x)∣≤∥f∥u, so ∥f±∥u≤∥f∥u. By linearity Qf=Qf+−Qf−, a difference of two numbers in [0,M∥f∥u], so
∣Qf∣≤M∥f∥u(f∈Cc(X)).
Thus M≥0 is a bound for Q: Q is bounded, so Q∈Cc(X)∗, and M lies in the set of nonnegative bounds of Q, whose infimum is ∥Q∥∗ by The Dual Norm (i). Hence ∥Λ±∥∗≤M=∥Λ∥∗, which completes the proof of (i).