On every locally compact group, there exists a left (resp. right) Haar measure, and, up to a constant factor, there exists only one.
A) Existence. — Set K(G)=K, K+(G)=K+, K+∗=K+−{0}. If C is a compact subset of G, we denote by K+∗(C) the set of f∈K+∗ with support in C. For f∈K and g∈K+∗, there exist numbers c1,…,cn≥0 and elements s1,…,sn of G such that f≤∑i=1nciγ(si)g: for, there exists a nonempty open set U in G such that infs∈Ug(s)>0, and the support of f can be covered by a finite number of left-translates of U. Let (f:g) be the infimum of the numbers ∑i=1nci for all systems (c1,…,cn,s1,…,sn) of numbers ≥0 and elements of G such that f≤∑i=1nciγ(si)g. Then:
(i) (γ(s)f:g)=(f:g) for f∈K, g∈K+∗, s∈G; (ii) (λf:g)=λ(f:g) for f∈K, g∈K+∗, λ≥0; (iii) ((f+f′):g)≤(f:g)+(f′:g) for f∈K, f′∈K, g∈K+∗; (iv) (f:g)≥(supf)/(supg) for f∈K, g∈K+∗; (v) (f:h)≤(f:g)(g:h) for f∈K, g∈K+∗, h∈K+∗; (vi) 0<(f0:f)1≤(f0:g)(f:g)≤(f:f0) for f,f0,g in K+∗; (vii) let f,f′,h be in K+ with h(s)≥1 on the support of f+f′, and let ε>0; there exists a compact neighborhood V of e such that, for every g∈K+∗(V),
(f:g)+(f′:g)≤((f+f′):g)+ε(h:g).
The properties (i), (ii), (iii) are obvious. Let f∈K, g∈K+∗; if f≤∑i=1nciγ(si)g with the ci≥0, then supf≤∑i=1ncig(si−1s) for some s \in G, therefore supf≤(∑i=1nci)supg, whence (iv). Let us now prove (v); let f∈K, g,h in K+∗; if f≤∑i=1nciγ(si)g and g≤∑j=1pdjγ(tj)h (ci≥0,dj≥0,si,tj in G), then f≤∑i,jcidjγ(sitj)h, therefore (f:h)≤∑i,jcidj=(∑ici)(∑jdj); thus (f:h)≤(f:g)(g:h). Applying (v) to f0,f,g on the one hand and to f,f0,g on the other, one obtains (vi). Finally, let f,f′,h be in K+ with h(s)≥1 on the support of f+f′, and let ε>0. Set F=f+f′+21εh; the functions φ,φ′, that coincide respectively with f/F and f′/F on the support of f+f′ and are zero outside it, belong to K+; for every η>0, there exists a compact neighborhood V of e such that ∣φ(s)−φ(t)∣≤η and ∣φ′(s)−φ′(t)∣≤η for s−1t∈V. Then let g∈K+∗(V); for every s∈G,
φ⋅γ(s)g≤(φ(s)+η)⋅γ(s)g;
for, this is obvious at the points where γ(s)g is zero, therefore outside of sV; and in sV, φ≤φ(s)+η; similarly, φ′⋅γ(s)g≤(φ′(s)+η)⋅γ(s)g. This stated, let c1,…,cn be numbers ≥0 and s1,…,sn elements of G such that F≤∑i=1nciγ(si)g; then
f=φF≤i=1∑nciφ⋅γ(si)g≤i=1∑nci(φ(si)+η)⋅γ(si)g
and similarly for f′; consequently
(f:g)+(f′:g)≤i=1∑nci(φ(si)+φ′(si)+2η)≤(1+2η)i=1∑nci
since φ+φ′≤1. Applying the definition of F, then (ii), (iii) and (v), one infers that
(f:g)+(f′:g)≤(1+2η)(F:g)≤
(1+2η)[((f+f′):g)+21ε(h:g)]≤
(((f+f′):g)+21ε(h:g)+2η((f+f′):h)(h:g)+εη(h:g)
and, if η has been chosen so that η[2((f+f′):h)+ε]≤21ε, one obtains (vii).
As V runs over the set of compact neighborhoods of e, the K+∗(V) form a base of a filter B on K+∗. Let F be an ultrafilter on K+∗ finer than B. On the other hand, let us fix f0∈K+∗ and let us set, for f∈K+∗ and g∈K+∗,
Ig(f)=(f0:g)(f:g).
By (vi), limg,FIg(f)=I(f) exists in the compact space [1/(f0:f),(f:f0)].
By (iii), I(f+f′)≤I(f)+I(f′). By (vii), I(f)+I(f′)≤I(f+f′)+εI(h) for all ε>0 provided h is ≥1 on the support of f+f′; it follows that I(f+f′)=I(f)+I(f′). By Ch. II, §2, No. 1, Prop. 2, I is extendible to a linear form on K; this linear form is a nonzero positive measure on G, left-invariant by (i); this is the sought-for left Haar measure. Passing to the opposite group, one deduces from this the existence of a right Haar measure.
B) Uniqueness. — Let μ be a left Haar measure, ν a right Haar measure. Then νˇ is a left Haar measure. We are going to show that μ and νˇ are proportional. This will prove that any two left Haar measures are indeed proportional.
Let f∈K be such that μ(f)=0. By Lemma 1, the function Df defined on G by the formula
Df(s)=μ(f)−1∫f(t−1s)dν(t)
is continuous on G. Let g∈K. The function (s,t)↦f(s)g(ts) is continuous with compact support in G×G. By Ch. III, §4, No. 1, Th. 2,
μ(f)ν(g)=(∫f(s)dμ(s))(∫g(t)dν(t))=∫dμ(s)∫f(s)g(ts)dν(t)=∫dν(t)∫f(s)g(ts)dμ(s)=∫dν(t)∫f(t−1s)g(s)dμ(s)=∫g(s)[∫f(t−1s)dν(t)]dμ(s)=μ(g⋅μ(f)Df),
whence
ν(g)=μ(Df⋅g).
This proves, first, that Df does not depend on f. For, if f′∈K is such that μ(f′)=0, then Df⋅μ=Df′⋅μ, therefore Df=Df′ locally almost everywhere for μ, hence everywhere, since Df and Df′ are continuous and the support of μ is G. We may therefore write Df=D. The formula (16) gives
μ(f)D(e)=νˇ(f).
The formula (19) may be extended by linearity to the functions f∈K such that μ(f)=0. We have D(e)=0 since νˇ=0. This indeed establishes the proportionality of μ and νˇ.
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.