Bourbaki marks this one as harder than the rest.
(a) Let (λ,x)↦λ.x be a law of continuous left operation of R on T. Then one of the two following holds: (1) either there exists a fixed point and the stabilizer of every point is either R or {0}; (2) or there exists a point whose stabilizer is an infinite discrete subgroup of R and R operates transitively on T. (If {λ∈R∣λ.x=x}={0}, the orbit of x is homeomorphic to R and has frontier points limλ→+∞λ.x which are fixed. If {λ∈R∣λ.x=x}=Za with a=0, the orbit of x is homeomorphic to T and R is transitive.)
(b) If f is a homeomorphism of T onto itself and k an integer ⩾1, a point x∈T is called periodic of period k for f if fk(x)=x and fh(x)=x for 1⩽h<k. In the notation of (a), let fλ be the homeomorphism x↦λ.x. Then the set of periodic points of period k>1 for fλ is empty or equal to T. (If there exists a periodic point of period k>1 for fλ, its stabilizer is distinct from {0} and R, hence R operates transitively on T and all the points of T are periodic of period k for fλ.)
(c) Let Ω be a neighbourhood of e in Diff∞(T) (Differentiable and Analytic Manifolds, R, 15.3.8, Note (1)). There exist an integer k>1 and f∈Ω without fixed point such that the set of periodic points of period k for f is neither empty nor equal to T. (Let p:R→T be the canonical mapping. Let ρ be the mapping p(x)↦p(x+k2π) of T into T. Let Ω′ be a neighbourhood of e in Diff∞(T) such that Ω′2⊂Ω. For sufficiently large k, ρ∈Ω′. On the other hand, let σ be a mapping of T into T such that σ(y)=y for
y∈/p(]0,k2π[) and
σ(p(x))=p(x+λ(x)) for
x∈]0,k2π[, where
0<λ(x)<k2π−x; can be chosen such that
σ∈Ω′. Then
f=ρ∘σ has the required properties.)
Such an f cannot, by (b), be in the image of a continuous morphism of R into Diff∞(T).
(d) Let Y be a compact differentiable manifold of dimension ⩾1. Every neighbourhood of e in Diff∞(Y) contains an element h with the following property: h belongs to the image of no continuous morphism of R into Diff0(Y). (The manifold Y contains an open submanifold of the form T×D, where D is the open Euclidean ball of radius 1 and centre 0 in Rn (n=dimY−1). Let g∈Diff∞(D) be such that g(0)=0, ∥g(x)∥>∥x∥ for 0<∥x∥<21 and g(x)=x for ∥x∥⩾21 and g very close to e in Diff∞(D). Let f be as in (c), and, for 0<t<1, let ft:T→T be defined as follows: for x∈T, y∈p−1(x), z∈p−1(f(x)) and ∣z−y∣<π, we write ft(x)=p(ty+(1−t)z). Then there exists h∈Diff∞(Y) such that h(x,y)=(f2∥y∥(x),g(y)) for x∈T, y∈D and h(u)=u for u∈Y−(T×D); moreover, we can choose f and g such that h is arbitrarily close to e in Diff∞(Y). The points of T×{0} are the y∈Y such that when n tends to +∞, hkn(x) tends to a point not fixed under h (in fact it is periodic of period k). Therefore, every homeomorphism of Y onto Y which commutes with h leaves T×{0} invariant. Then apply (c).)
(e) Let Y be a compact differentiable manifold of dimension ⩾1. There do not exist a Lie group G and continuous morphisms Diff∞(Y)→G, G→Diff0(Y) whose composition is the canonical injection of Diff∞(Y) into Diff0(Y). (Use (d).)