Bourbaki

2. Continuous homomorphisms of Rn\mathbf{R}^n and its quotient groups

1. Continuous homomorphisms of the group Rm\mathbf{R}^m into the group Rn\mathbf{R}^n

Every linear mapping (cf. Chapter VI, § i, no. 3) of Rm\mathbf{R}^m into Rn\mathbf{R}^n is evidently a continuous homomorphism of the additive group Rm\mathbf{R}^m into the additive group Rn\mathbf{R}^n. Conversely:

Proposition 1

Every continuous homomorphism* ff of the additive group Rm\mathbf{R}^m into the additive group Rn\mathbf{R}^n is a linear mapping of Rm\mathbf{R}^m into Rn\mathbf{R}^n.

It is enough to show that f(tx)=tf(x)f(tx) = tf(x) for all x∈Rnx \in \mathbf{R}^n and all t∈Rt \in \mathbf{R}. The argument is the same as that of Proposition 5 of Chapter V, § i, no. 3 if we replace xx by xx and R\mathbf{R} by Rm\mathbf{R}^m.

2. Local definition of a continuous homomorphism of Rn\mathbf{R}^n into a topological group

Proposition 6 of Chapter V, § i, no. 4 may be generalized to all the groups Rn\mathbf{R}^n.

Proposition 2

Let* AA be a parallelopte in Rn\mathbf{R}^n which contains oo; and let ff be a continuous mapping of AA into a topological group GG (written multiplicatively) such that f(x+y)=f(x)f(y)f(x + y) = f(x)f(y) for each pair of points x,yx, y such that x∈A, y∈Ax \in A,\ y \in A and x+y∈Ax + y \in A. Then there exists a unique continuous homomorphism of Rn\mathbf{R}^n into GG which extends ff.

By the same reasoning as in Proposition 6 of Chapter V, § i, no. 4 we show that the homomorphism extending ff, if it exists, is unique. Next, the subgroup G1G_1 of GG, generated by f(A)f(A), is commutative; for if xx and yy are any two points of AA, then 12x,12y\frac{1}{2} x, \frac{1}{2} y and 12(x+y)\frac{1}{2} (x + y) belong to AA, therefore

f(12(x+y))=f(12x)(f12y)=f(12y)f(12x), f \left( \frac{1}{2} (x + y) \right) = f \left( \frac{1}{2} x \right) \quad (f \frac{1}{2} y) = f \left( \frac{1}{2} y \right) f \left( \frac{1}{2} x \right),

which shows that f(12x)f \left( \frac{1}{2} x \right) and f(12y)f \left( \frac{1}{2} y \right) commute; therefore so do f(x)=(f(12x))2f(x) = (f \left( \frac{1}{2} x \right))^2 and f(y)=(f(12y))2f(y) = (f \left( \frac{1}{2} y \right))^2, which proves the assertion.

Let a1,a2,…,ana_1, a_2, \ldots, a_n be nn non-zero vectors contained in AA and proportional to the basis vectors of the parallelootope, and for each index ii let DiD_i be the line through oo and aia_i, that is, the set of points taita_i as tt runs through R\mathbf{R}. Let AiA_i be the set of all t∈Rt \in \mathbf{R} such that tai∈Ata_i \in A; then AiA_i is an interval of R\mathbf{R} containing [0,1][0, 1], and the function

fi(t)=f(tai) f_i(t) = f(ta_i)

is defined and continuous on AiA_i and satisfies the relation

fi(t+t′)=fi(t)fi(t′) f_i(t + t') = f_i(t) f_i(t')

whenever t,t′t, t' and t+t′t + t' all belong to AiA_i. By Proposition 6 of Chapter V, § 1, no. 4, there exists a continuous homomorphism fˉi\bar{f}_i of R\mathbf{R} into GG which extends fif_i. Since Rn\mathbf{R}^n is the direct sum of the subgroups DiD_i, we can define a homomorphism fˉ\bar{f} of Rn\mathbf{R}^n into the abelian group G1G_1 by the rule fˉ(x)=∏i=1nfˉi(ti)\bar{f}(x) = \prod_{i=1}^n \bar{f}_i(t_i) where x=∑i=1ntiaix = \sum_{i=1}^n t_i a_i; fˉ\bar{f} is an extension of ff since, if x∈Ax \in A, all the components xix_i of xx on the lines DiD_i also belong to AA, by reason of the choice of the aia_i; moreover, fˉ\bar{f} is continuous on Rn\mathbf{R}^n, because it is continuous on each of the lines DiD_i, and xix_i is a linear function of xx (and therefore continuous).

Corollary 1

Let VV be a neighbourhood of oo in Rn\mathbf{R}^n and let ff be a continuous mapping of VV into a topological group GG such that

f(x+y)=f(x)f(y) f(x + y) = f(x) f(y)

for each pair of points x,yx, y such that x∈V,y∈Vx \in V, y \in V and x+y∈Vx + y \in V. Then there exists a unique continuous homomorphism of Rn\mathbf{R}^n into GG which agrees with ff at all points of a neighbourhood WW of oo.

Take WW to be an open box with centre oo, contained in VV, and apply Proposition 2 to WW.

We shall see later that this property of Rn\mathbf{R}^n extends to a larger class of topological groups, the "simply connected" groups.

Corollary 2

Let ff be a local isomorphism of Rn\mathbf{R}^n with a topological group GG. Then there exists a unique strict morphism of Rn\mathbf{R}^n onto an open subgroup of GG which agrees with ff at all the points of some neighbourhood of oo.

Let f‾\overline{f} be the continuous homomorphism of Rn\mathbf{R}^n into GG which agrees with ff at all points of a neighbourhood of oo; f‾(Rn)\overline{f}(\mathbf{R}^n) contains, by hypothesis, a neighbourhood of the identity element of GG and therefore (Chapter III, § 2, no. 1, Corollary to Proposition 4) is an open subgroup of GG; moreover f‾\overline{f} is a strict morphism of Rn\mathbf{R}^n onto f‾(Rn)\overline{f}(\mathbf{R}^n), by Chapter III, § 2, no. 8, Proposition 24.

Theorem 1

Every connected group GG which is locally isomorphic to Rn\mathbf{R}^n is isomorphic to a group Rp×Tn−p\mathbf{R}^p \times \mathbf{T}^{n-p} (0⩽p⩽n0 \leqslant p \leqslant n).

A suitably chosen local isomorphism ff of Rn\mathbf{R}^n with GG extends to a strict morphism of Rn\mathbf{R}^n onto an open subgroup of GG (Proposition 2, Corollary 2) and therefore onto GG itself, since GG is connected. It follows that GG is isomorphic to a quotient group Rn/H\mathbf{R}^n/H of Rn\mathbf{R}^n: HH is discrete, otherwise there would exist points x≠ox \neq o of HH arbitrarily close to oo and such that f(x)=f(o)f(x) = f(o), contrary to the hypothesis that ff is a local isomorphism. The theorem is therefore a consequence of Theorem 1 of § 1, no. 1.

3. Continuous homomorphisms of Rm\mathbf{R}^m into Tn\mathbf{T}^n

Proposition 3

Every continuous homomorphism of Rm\mathbf{R}^m into Tn\mathbf{T}^n is of the form x→φ(u(x))x \to \varphi(u(x)), where φ\varphi is the canonical homomorphism of Rn\mathbf{R}^n onto Tn\mathbf{T}^n (identified with Rn/Zn\mathbf{R}^n/\mathbf{Z}^n) and uu is a linear mapping of Rm\mathbf{R}^m into Rn\mathbf{R}^n.

Let ff be a continuous homomorphism of Rm\mathbf{R}^m into Tn\mathbf{T}^n. We shall show that there is a linear mapping uu of Rm\mathbf{R}^m into Rn\mathbf{R}^n such that the homomorphisms x→f(x)x \to f(x) and x→φ(u(x))x \to \varphi(u(x)) coincide at all points of a neighbourhood of oo in Rm\mathbf{R}^m; the proposition will then follow from Corollary 1 to Proposition 2 of no. 2. Now let VV be a neighbourhood of oo in Rn\mathbf{R}^n such that φ\varphi, restricted to VV, is a local isomorphism of Rn\mathbf{R}^n with Tn\mathbf{T}^n, and let ψ\psi be the inverse of φ\varphi, defined on φ(V)\varphi(V). Since ff is continuous, V′=f‾−1(φ(V))V' = \overline{f}^{-1}(\varphi(V)) is a neighbourhood of oo in Rm\mathbf{R}^m; the mapping

x→ψ(f(x)), x \to \psi(f(x)),

restricted to V′V', is a continuous mapping of V′V' into Rn\mathbf{R}^n such that ψ(f(x+y))=ψ(f(x))+ψ(f(y))\psi(f(x+y)) = \psi(f(x)) + \psi(f(y)) for each pair of points x,yx, y in Rm\mathbf{R}^m such that x∈V′,y∈V′x \in V', y \in V' and x+y∈V′x + y \in V'; hence (no. 2, Corollary 1 to Proposition 2) this mapping coincides with a well-determined continuous homomorphism uu of Rm\mathbf{R}^m into Rn\mathbf{R}^n at all points of a neighbourhood WW of o in Rm\mathbf{R}^m. By Proposition 1 (no. 1) uu is a linear mapping of Rm\mathbf{R}^m into Rn\mathbf{R}^n; for all x∈Wx \in W we have therefore f(x)=φ(u(x))f(x) = \varphi(u(x)), which completes the proof.

Remark

The same argument shows, more generally, that if φ\varphi is a strict morphism of Rn\mathbf{R}^n into a group GG, whose restriction to a suitable neighbourhood of o is a local isomorphism of Rn\mathbf{R}^n with GG, then every continuous homomorphism of Rn\mathbf{R}^n into GG is of the form x↦φ(u(x))x \mapsto \varphi(u(x)), where uu is a linear mapping of Rm\mathbf{R}^m into Rn\mathbf{R}^n.

In the case m=n=1m = n = 1, Proposition 3 gives:

Proposition 4

If φ\varphi is the canonical homomorphism of R\mathbf{R} onto T\mathbf{T}, every continuous homomorphism of R\mathbf{R} into T\mathbf{T} is of the form x↦φ(ax)x \mapsto \varphi(ax) where a∈Ra \in \mathbf{R}; and it is a strict morphism of R\mathbf{R} onto T\mathbf{T} if a≠0a \neq 0.

4. Automorphisms of Tn\mathbf{T}^n

Let HH be a closed subgroup of Rn\mathbf{R}^n and let φ\varphi be the canonical homomorphism of Rn\mathbf{R}^n onto the quotient group Rn/H\mathbf{R}^n/H. If ff is a continuous homomorphism of Rn/H\mathbf{R}^n/H into a topological group GG, then f=f∘φf = f \circ \varphi is a continuous homomorphism of Rn\mathbf{R}^n into GG which is periodic and has a group of periods containing HH; conversely every periodic continuous homomorphism of Rn\mathbf{R}^n into GG, whose group of periods contains HH, is of this form.

In the case where H=ZnH = \mathbf{Z}^n, the quotient group Rn/Zn=Tn\mathbf{R}^n/\mathbf{Z}^n = \mathbf{T}^n is compact, and therefore every continuous homomorphism ff of Tn\mathbf{T}^n into a topological group GG is a strict morphism of Tn\mathbf{T}^n into GG, provided that GG is Hausdorff (Chapter III, § 2, no. 8, Remark 1); f=f∘φf = f \circ \varphi is a strict morphism of Rn\mathbf{R}^n into GG; moreover, f(Tn)=f(Rn)f(\mathbf{T}^n) = f(\mathbf{R}^n) is a compact subgroup of GG, isomorphic to a group Tp\mathbf{T}^p (0≤p≤n0 \leq p \leq n).

In particular, we see that the only continuous homomorphism of Tn\mathbf{T}^n into a group Rm\mathbf{R}^m is the zero mapping, since {o}\{ o \} is the only compact subgroup of Rm\mathbf{R}^m.

Let us apply this to continuous homomorphisms of Tn\mathbf{T}^n into a group Tp\mathbf{T}^p; if ff is such a homomorphism, φ\varphi the canonical homomorphism of Rn\mathbf{R}^n onto Tn\mathbf{T}^n, then f∘φf \circ \varphi is a continuous homomorphism of Rn\mathbf{R}^n into Tp\mathbf{T}^p; hence (no. 3, Proposition 3) if ψ\psi denotes the canonical homomorphism of Rp\mathbf{R}^p onto Tp\mathbf{T}^p, there exists a linear mapping uu of Rn\mathbf{R}^n into Rp\mathbf{R}^p such that f∘φ=ψ∘uf \circ \varphi = \psi \circ u. If x∈Znx \in \mathbf{Z}^n, f(φ(x))f(\varphi(x)) is the identity element of Tp\mathbf{T}^p, so that we must have u(x)∈Zpu(x) \in \mathbf{Z}^p, i.e., u(Zn)⊂Zpu(\mathbf{Z}^n) \subset \mathbf{Z}^p. Conversely, if uu is any linear mapping of Rn\mathbf{R}^n into Rp\mathbf{R}^p which satisfies this condition, then ψ∘u\psi \circ u is a periodic continuous homomorphism of Rn\mathbf{R}^n into Tp\mathbf{T}^p, whose group of periods contains Zn\mathbf{Z}^n, and therefore defines a continuous homomorphism of Tn\mathbf{T}^n into Tp\mathbf{T}^p.

Consider under what conditions ff is an isomorphism of Tn\mathbf{T}^n onto a subgroup of Tp\mathbf{T}^p. First of all, uu must be an injective mapping of Rn\mathbf{R}^n into Rp\mathbf{R}^p, otherwise the vector subspace u‾(0)\overline{u}(0) would contain points x≠0x \neq 0 arbitrarily close to the origin, and at such a point we should have

f(φ(x))=f(φ(0))andφ(x)≠φ(0), f(\varphi(x)) = f(\varphi(0)) \quad \text{and} \quad \varphi(x) \neq \varphi(0),
contrary to hypothesis. This condition implies that p≥np \geq n. The image u(Zn)u(\mathbf{Z}^n) is then a discrete subgroup of rank nn of the group Zp\mathbf{Z}^p; the invariant factors of u(Zn)u(\mathbf{Z}^n) with respect to Zp\mathbf{Z}^p (§ 1, no. 1) must all be equal to 1, otherwise there would exist a point x∈Znx \in \mathbf{Z}^n and an integer k>1k > 1 such that u(k−1x)∈Znu(k^{-1}x) \in \mathbf{Z}^n and k−1∉Znk^{-1} \notin \mathbf{Z}^n, hence f(φ(k−1x))=f(φ(0))f(\varphi(k^{-1}x)) = f(\varphi(0)) and φ(k−1x)≠φ(0)\varphi(k^{-1}x) \neq \varphi(0), contrary to hypothesis. Conversely, if this condition is satisfied, u(Rn)∩Zn=u(Zn)u(\mathbf{R}^n) \cap \mathbf{Z}^n = u(\mathbf{Z}^n), and ff is an isomorphism of Tn\mathbf{T}^n onto u(Rn)/u(Zn)u(\mathbf{R}^n)/u(\mathbf{Z}^n).

If we apply this argument to the case p=np = n, we have the following proposition:

Proposition 5

Every isomorphism of the topological group Tn\mathbf{T}^n onto one of its subgroups is an automorphism of Tn\mathbf{T}^n which is obtained by passing to the quotient from a linear mapping uu of Rn\mathbf{R}^n onto itself which, restricted to Zn\mathbf{Z}^n, is an automorphism of Zn\mathbf{Z}^n.

Equivalently (§ 1, no. 1), if u(ei)=∑j=1naijeju(e_i) = \sum_{j=1}^n a_{ij} e_j, the aija_{ij} must be integers such that det⁡(aij)=±1\det(a_{ij}) = \pm 1. In particular, for n=1n = 1:

Proposition 6

The only isomorphisms of the topological group T\mathbf{T} onto one of its subgroups are the identity mapping and the symmetry x→−xx \to -x.

Exercises

See the exercises for § 2.

Transcribed from the printed volume. Not translated.