Bourbaki

3. Infinite sums and products of complex numbers

1. Infinite sums of complex numbers

Since the additive group of the field C\mathbf{C} is the same as the additive group R2\mathbf{R}^2, it is not necessary to make a special study of summable families and series in C\mathbf{C}, since this is included in the general theory of Chapter VII, § 3; we leave to the reader the exercise of translating the results of this theory into the language of the theory of complex numbers. We state only the following proposition, which is a corollary to Proposition 3 of Chapter VII, § 3, no. 1:

Proposition 1

If* (uλ)λ∈L(u_\lambda)_{\lambda \in L} and (vμ)μ∈M(v_\mu)_{\mu \in M} are two summable families of complex numbers, then the family (uλvμ)(λ,μ)∈L×M(u_\lambda v_\mu)_{(\lambda, \mu) \in L \times M} *is summable and we have

∑(λ,μ)∈L×Muλvμ=(∑λ∈Luλ)(∑μ∈Mvμ). \sum_{(\lambda, \mu) \in L \times M} u_\lambda v_\mu = \left( \sum_{\lambda \in L} u_\lambda \right) \left( \sum_{\mu \in M} v_\mu \right).

We leave it to the reader to state the corresponding result for quaternions.

2. Multipliable families in C∗\mathbf{C}^*

In the multiplicative group C∗\mathbf{C}^* of non-zero complex numbers a family (zi)i∈I(z_i)_{i \in I} cannot be multipliable unless lim⁡zi=1\lim z_i = 1 with respect to the filter of complements of finite subsets of II (Chapter III, § 5, no. 2, Proposition 1); furthermore, since every point of C∗\mathbf{C}^* has a countable fundamental system of neighbourhoods, the set of indices ii such that zi≠1z_i \neq 1 is countable if the family (zi)(z_i) is multipliable (Chapter III, § 5, no. 2, Corollary to Proposition 1).

Proposition 2

A family* (zi)(z_i) of complex numbers zi=ri(cos⁡θi+isin⁡θi)z_i = r_i (\cos \theta_i + i \sin \theta_i) is multipliable if and only if the family (ri)(r_i) of absolute values of the ziz_i is multipliable in R+∗\mathbf{R}_+^* and the family (θi)(\theta_i) of amplitudes of the ziz_i is summable in the group of angles A\mathfrak{A}.

In view of the structure of the group C∗\mathbf{C}^* (§\S 1, no. 3, Proposition 1), the proposition is an immediate consequence of Proposition 4 of Chapter III, §\S 5, no. 4.

If we map each angle θ\theta to that one of its measures (to any given base aa) which belongs to the interval ]−12a,12a]] - \frac{1}{2} a, \frac{1}{2} a ], we have a local isomorphism of A\mathfrak{A} with R\mathbf{R} (§\S 2, no. 2); since lim⁡θi=0\lim \theta_i = 0 with respect to the filter of complements of finite subsets of II, we may replace the condition (in the statement of Proposition 2) that the family θi\theta_i should be summable in A\mathfrak{A} by the condition that the family (ti)(t_i) of measures of the angles θi\theta_i which belong to ]−12a,12a]] - \frac{1}{2} a, \frac{1}{2} a ] should be summable in R\mathbf{R}.

The following theorem gives another criterion for a family of complex numbers, put in the form (1+ui)(1 + u_i), to be multipliable in C∗\mathbf{C}^*. (It generalizes Theorem 4 of Chapter IV, §\S 7, no. 4; see also Chapter IX, Appendix, no. 2, Proposition 1):

Theorem 1

The family* (1+ui)i∈I(1 + u_i)_{i \in I} is multipliable in C∗\mathbf{C}^* if and only if the family (∣ui∣)(|u_i|) is summable in R\mathbf{R}.

For each finite subset JJ of II, put

pJ=∏i∈J(1+ai),sJ=∑i∈Jai,σJ=∑i∈J∣ai∣. p_J = \prod_{i \in J} (1 + a_i), \quad s_J = \sum_{i \in J} a_i, \quad \sigma_J = \sum_{i \in J} |a_i|.

Lemma 1

For each finite subset* JJ of II, let φ(J)=sup⁡L⊂J(pL−1)\varphi(J) = \sup_{L \subset J} (p_L - 1). Then for each subset LL of JJ *we have

∣pL−1−sL∣≤φ(J)σL. |p_L - 1 - s_L| \leq \varphi(J) \sigma_L.

This is clear if LL is empty. We proceed by induction on card (L)(L). Let L=K∪{λ}L = K \cup \{ \lambda \}, where λ∉K\lambda \notin K; then pL=pK(1+aλ)p_L = p_K (1 + a_\lambda) and sL=sK+aλs_L = s_K + a_\lambda, so that pL−1−sL=(pK−1−sK)+(pK−1)aλp_L - 1 - s_L = (p_K - 1 - s_K) + (p_K - 1)a_\lambda; hence, by the inductive hypothesis and the definition of φ(J)\varphi(J), we have

∣pL−1−sL∣≤φ(J)σK+φ(J)∣aλ∣=φ(J)σL, |p_L - 1 - s_L| \leq \varphi(J) \sigma_K + \varphi(J)|a_\lambda| = \varphi(J) \sigma_L,

which proves the lemma.

Lemma 2

If* JJ is a finite subset of II such that φ(J)<1/4\varphi(J) < 1/4, *then

∣σJ∣≤4φ(J)/(1−4φ(J)). |\sigma_J| \leq 4 \varphi(J)/(1 - 4 \varphi(J)).

For since σL≤σJ\sigma_L \leq \sigma_J for all subsets LL of JJ, it follows from (2) that ∣sL∣≤φ(J)σJ+∣pL−1∣≤(1+σJ)φ(J)|s_L| \leq \varphi(J) \sigma_J + |p_L - 1| \leq (1 + \sigma_J)\varphi(J); but by virtue of Chapter VII, §\S 3, no. 1, Proposition 2, we have ∣σJ∣≤4sup⁡L⊂J∣sL∣|\sigma_J| \leq 4 \sup_{L \subset J} |s_L|, hence σJ≤4φ(J)(1+σJ)\sigma_J \leq 4 \varphi(J)(1 + \sigma_J), and the result follows.

Now let us show that the condition stated in Theorem 1 is sufficient. The hypothesis that the family (∣ui∣)(|u_i|) is summable in R\mathbf{R} implies that the family (1+∣ui∣)(1 + |u_i|) is multipliable in R+∗\mathbf{R}_+^* (Chapter IV, § 7, no. 4, Theorem 4); hence, for each ε>0\varepsilon > 0, there is a finite subset J0J_0 of II such that, for each finite subset LL of II which does not meet J0J_0, we have ∏i∈L(1+∣ui∣)−1≤ε\prod_{i \in L} (1 + |u_i|) - 1 \leq \varepsilon. But we can write ∏i∈L(1+ui)−1\prod_{i \in L} (1 + u_i) - 1 in the form ∑M(∏i∈Mui)\sum_M \left( \prod_{i \in M} u_i \right) where MM runs through all non-empty subsets of LL; and since ∣∏i∈Mui∣=∏i∈M∣ui∣\left| \prod_{i \in M} u_i \right| = \prod_{i \in M} |u_i|, we have

∣∏i∈L(1+ui)−1∣≤∑M(∏i∈M∣ui∣)=∏i∈L(1+∣ui∣)−1≤ε. \left| \prod_{i \in L} (1 + u_i) - 1 \right| \leq \sum_M \left( \prod_{i \in M} |u_i| \right) = \prod_{i \in L} (1 + |u_i|) - 1 \leq \varepsilon.

This proves our assertion, by virtue of Cauchy’s criterion, since C∗\mathbf{C}^* is a complete group.

We still have to show that the condition of Theorem 1 is necessary. If (1+ui)i∈I(1 + u_i)_{i \in I} is a multipliable family in C∗\mathbf{C}^*, there exists a finite subset JJ of II such that, for every finite subset HH of II which does not meet JJ, we have ∣∏i∈H(1+ui)−1∣≤1/8\left| \prod_{i \in H} (1 + u_i) - 1 \right| \leq 1/8. By Lemma 2, it follows that ∑i∈H∣ui∣≤1\sum_{i \in H} |u_i| \leq 1 for every finite subset HH of II which does not meet JJ, and hence the family (∣ui∣)(|u_i|) is summable in R\mathbf{R} (Chapter IV, § 7, no. 1, Theorem 1).

The proof above applies also, mutatis mutandis, to (ordered) infinite products in certain non-commutative division rings and algebras (see Exercise 6, and Chapter IX, Appendix).

3. Infinite products of complex numbers

For an infinite product of non-zero complex numbers with general factor zn=rn(cos⁡θn+isin⁡θn)z_n = r_n (\cos \theta_n + i \sin \theta_n) to be convergent in C∗\mathbf{C}^*, it is necessary and sufficient, from the structure of the group C∗\mathbf{C}^*, that the product with general factor rnr_n converge in R+∗\mathbf{R}_+^* and the series with general term tnt_n (the measure of θn\theta_n which lies in ]−12a-\frac{1}{2}a, 12a\frac{1}{2}a]) converge in R\mathbf{R}.

Definition 1

An infinite product of complex numbers, with general factor* 1+un1 + u_n, is said to be absolutely convergent if the product with general factor 1+∣un∣1 + |u_n| is convergent (or, equivalently, if the series with general term ∣un∣|u_n| is convergent).

Proposition 3

An infinite product of complex numbers is commutatively convergent if and only if it is absolutely convergent.

Remark 1

The product with general factor ∣1+un∣|1 + u_n| can be convergent, and indeed absolutely convergent in Ri∗\mathbf{R}_i^*, without the convergence of the product with general factor 1+∣un∣1 + |u_n| (see Exercise 4); of course, this cannot happen if all the unu_n are real and >0> 0 from a certain index onwards.

Remark 2

As already remarked for products of factors >0> 0, the convergence of the series with general term unu_n is neither necessary nor sufficient for the convergence of the product with general factor 1+un1 + u_n.

Exercises

See the exercises for § 3.

Transcribed from the printed volume. Not translated.