10

It is well-known that if $\sum_{n=1}^\infty a_n$ is an absolutely convergent complex series and $\mathbb N$ is partitioned as $J_1,J_2,\dots$, then the series $\sum_{j\in J_n}a_j$ for all $n$ and $\sum_{n=1}^\infty\sum_{j\in J_n}a_j$ are both absolutely convergent, and $\sum_{n=1}^\infty a_n=\sum_{n=1}^\infty\sum_{j\in J_n}a_j$.

I am trying to make sense of the same identity for infinite products. The convention I'm using is the following

The product $\prod_{n=1}^\infty z_n$ is called convergent if there exists an $n_0$ such that the partial products $\prod_{j=n_0}^nz_j$ converge to a non-zero number as $n\to\infty$. Write $z_n=1+w_n$, the product is said to converge absolutely if $\prod_n(1+|w_n|)$ converges. It is a theorem that an absolutely convergent product is convergent and that $\prod_n(1+|w_n|)$ converges if and only if $\sum_n|w_n|$ does.

Suppose $\prod_{n=1}^\infty z_n$ converges absolutely and $\Bbb N=\bigcup_{n=1}^\infty J_n$ is a partition, I want to show that $$\prod_{n=1}^\infty z_n=\prod_{n=1}^\infty\prod_{j\in J_n}z_j,$$ with the RHS converging absolutely. By the characterisation of absolute convergence, we know that $\sum_{n=1}^\infty|z_n-1|$ converges, so the series $\sum_{j\in J_n}|z_j-1|$ converge for all $n$ and hence all the products $\prod_{j\in J_n}z_j$ converge absolutely. What I'm having trouble with is showing $\prod_{n=1}^\infty\prod_{j\in J_n}z_j$ converges (absolutely or not) and showing it is equal to the original product. By the characterisation of absolute convergence, it would be enough to show that $\sum_{n=1}^\infty|(\prod_{j\in J_n}z_j)-1|$ converges, but this has no discernible relation to the series $\sum_{n=1}^\infty\sum_{j\in J_n}|z_j-1|$, which we do know converges by the "associativity" for absolutely convergent series.

St. Barth
  • 1,642
  • 4
    It would be commutativity rather than associativity right ? – Statistic Dean May 16 '24 at 12:55
  • can you write your questions in full? what are you assuming about the $z_j$'s? it's not clear to me what the assumptions are – mathworker21 Jun 23 '24 at 02:05
  • Multiplication of complex numbers is associative as is their real norms. – mick Jun 23 '24 at 02:21
  • @mathworker21 there are no assumptions on the $z_j$'s, these are simply complex numbers. The assumption is simply that the product $\prod_jz_j$ converge absolutely, according to the specified definitions. – St. Barth Jun 23 '24 at 09:32

2 Answers2

2

Take logs so you are only working with sums. At the end you can exponentiate to get the results you need in terms of products. Therefore, in place of products $\prod_{i=1}^\infty (1+w_i)$ you are working with sums $\sum_{i=1}^\infty \log(1+w_i)$. We don't yet know that this sum $\sum_{i=1}^\infty \log(1+w_i)$ converges absolutely. But if we did, then you would have that $\sum_{i=1}^\infty\sum_{j\in J_i}\log(1+w_j)=\sum_{i=1}^\infty \log(1+w_i)$ where everything converges absolutely as sums. At this point, after exponentiating everything, it would be a short argument to show the result you are after.


It turns out that if the product converges absolutely, then the sum does too:

One has

$$\log(1+|w|)\le |w|$$

$$\implies \log\left(\prod_{i=1}^n(1+|w_i|)\right)\le \sum_{i=1}^n |w_i|\le \sum_{i=1}^\infty |w_i|$$

so that if $\sum w_i$ converges absolutely, $\prod (1+w_i)$ converges absolutely.

The converse is also true: suppose that $\prod (1+w_i)$ converges absolutely. Then $\prod(1+|w_i|)$ converges, and hence, for sufficiently large $N$, $|w_i|<1$ for all $i>N$. If $|w|<1$ then

$$\frac{1}{2}|w|\le \log (1+|w|).$$ Therefore,

\begin{align*} \sum_{i>N} |w_i|&\le 2\sum_{i>N}\log(1+|w_i|)\\ &=2\log\left(\prod_{i>N}(1+|w_i|)\right). \end{align*}

So, $\sum |w_i|$ is convergent, so $\sum w_i$ is absolutely convergent.

Now that you have this, you can leverage all the theorems for absolutely convergent sums and apply them to absolutely convergent products. In particular, the results in your question.


Suppose that $\prod (1+w_i)$ converges absolutely. Take the logarithm of the partial products of $\prod_{i=1}^\infty\prod_{j\in J_i}(1+|w_i|)$ to get

$$\log\left(\prod_{i=1}^N\prod_{j\in J_i}(1+|w_i|)\right)=\sum_{i=1}^N\sum_{j\in J_i}\log(1+|w_i|)\le \sum_{i=1}^\infty\sum_{j\in J_i}|w_i|$$

the RHS of which is finite by well-known results cited in your question. Therefore,

$$\prod_{i=1}^N\prod_{j\in J_i}(1+|w_i|)$$

is bounded, hence $\prod_{i=1}^\infty\prod_{j\in J_i}(1+|w_i|)$ is convergent. So, $\prod_{i=1}^\infty\prod_{j\in J_i}(1+|w_i|)$ is absolutely convergent.

Joshua Tilley
  • 10,106
  • I am aware of this criterion for convergence, I mention it in the question, and it's easily applied to show that each of the products $\prod_{j\in J_n}z_j$ converges absolutely. My problem, as outlined in the question, is showing that the entire product $\prod_{n=1}^\infty\prod_{j\in J_n}z_j$ converges. Its absolute convergence is, by the theorem you mention, equivalent to the convergence of the series $\sum_{n=1}^\infty|(\prod_{j\in J_n}z_j)-1|$, no? My problem is I don't see why this series should converge. – St. Barth Jun 23 '24 at 11:37
  • Thank you! I was so fixated on proving convergence directly from the series criterion, I completely missed this. – St. Barth Jun 24 '24 at 08:08
  • You're welcome! The rough strategy for infinite products I would suggest is to take logs so you are always thinking in terms of sums only. Then you need to estimate $\log(1+w_i)$ which I have addressed in my answer. – Joshua Tilley Jun 24 '24 at 15:17
  • I believe I was overhasty in accepting your answer: by definition $\prod_n(1+w_n)$ is absolutely convergent if $\prod_n(1+|w_n|)$ convergers, right? Well in our case the product is $\prod_n(\prod_{j\in J_n}(1+w_j))$, so you'd have to show that $\prod_n(1+|(\prod_{j\in J_n}(1+w_n))-1|)$ converges, no? – St. Barth Jun 25 '24 at 08:11
  • I don't agree, can you explain why? That is a different product. – Joshua Tilley Jun 25 '24 at 08:20
  • I'm saying that if $\prod_nu_n$ is your product, then to show it converges absolutely you have to show $\prod_n(1+|u_n-1|)$ converges, right? Since if $u_n=1+w_n$, then $w_n=u_n-1$. Well in our case the individual factors are $u_n=\prod_{j\in J_n}(1+w_j)$, so you'd have to show $\prod_n(1+|(\prod_{j\in J_n}(1+w_j))-1|)$ converges to show $\prod_n\prod_{j\in J_n}(1+w_j)$ converges absolutely, no? Or do I have it wrong somehow? – St. Barth Jun 25 '24 at 09:43
1

Summary: absolute convergence of the infinite product of a sequence $(z_n)_{n\in\mathbb{N}}$ of non-zero complex numbers is equivalent to multipliability of that sequence in $\mathbb{C}^*;$ and, as a special case of the general concept of summability in complete Hausdorff commutative topological groups, this concept behaves 'nicely', including obeying the infinite associative law.

In more detail (but without copying out the whole of Bourbaki!):

[1, Chapter III, $\S5$]

Proposition 2. In a complete group $G,$ every subfamily of a summable family is summable.

Theorem 2 (Associativity of the sum). Let $(x_\iota)_{\iota\in I}$ be a summable family in a complete group $G,$ and let $(I_\lambda)_{\lambda\in L}$ be any partition of $I.$ If $s_\lambda$ denotes $\sum_{\iota\in I_\lambda}x_\iota,$ then the family $(s_\lambda)_{\lambda\in L}$ is summable and has the same sum as the family $(x_\iota)_{\iota\in I}.$

We can thus write the associativity formula for sums: $$ \sum_{\lambda\in L}\bigg(\sum_{\iota\in I_\lambda}x_\iota\bigg) = \sum_{\substack{\iota\in\bigcup I_\lambda \\ \lambda\in L}}x_\iota, $$ which is valid whenever the family $(I_\lambda)$ is a partition of its union and the right-hand side is defined.

[2, Chapter VIII, $\S3$]

Theorem 1. The family $(1+u_\iota)_{\iota\in I}$ is multipliable in $\mathbb{C}^*$ if and only if the family $(|u_\iota|)$ is summable in $\mathbb{R}.$

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

In the notation of the question: suppose $\prod_jz_j$ is absolutely convergent. By Theorem 1 and Definition 1, the family $(z_j)_{j\in\mathbb{N}}$ is multipliable in $\mathbb{C}^*.$ For all $n,$ by Proposition 2, the family $(z_j)_{j\in J_n}$ is multipliable in $\mathbb{C}^*.$ By Theorem 2, the family $(\prod_{j\in J_n}z_j)_{n\in\mathbb{N}}$ is multipliable in $\mathbb{C}^*,$ and $$ \prod_{n=1}^\infty\bigg(\prod_{j\in J_n}z_j\bigg) = \prod_{j\in\mathbb{N}}z_j = \prod_{j=1}^\infty z_j. $$

Bibliography:

[1] Nicolas Bourbaki, Elements of Mathematics: General Topology, Part 1 (Springer, 1989; original French edition published by Masson, Paris, 1971).

[2] Nicolas Bourbaki, Elements of Mathematics: General Topology, Part 2 (Springer, 1989; original French edition published by Masson, Paris, 1974).

  • I don't see how the multipliability of $(\prod_{j\in J_n}z_j)n$ follows from the summability of the original family. The problem is that to go from a product to a sum you need to pass from $(z_n)_n$ to $(|z_n-1|)_n$ and in this case I don't see how $(\prod{j\in J_n}z_j)n$ relates to $(\sum{j\in J_n}|z_j-1|)_n$, which is indeed summable. – St. Barth Jun 11 '24 at 21:40
  • Which of the logical steps in the last paragraph of my answer do you suspect may be invalid? (I'm sorry if I've made an obvious goof; but I just can't see it.) – Calum Gilhooley Jun 12 '24 at 17:03
  • I fully agree that $(z_j){j\in\Bbb N}$ absolutely multipliable $\implies(|z_j-1|){j\in\Bbb N}$ summable $\implies(\sum_{j\in J_n}|z_j-1|){n\in\Bbb N}$ summable, but this implication chain has no (discernible, to me) relation to $(\prod{j\in J_n}z_j){n\in\Bbb N}$ being absolutely multipliable, since the corresponding family whose summability we'd have to investigate is $(|\prod{j\in J_n}z_j-1|)_{n\in\Bbb N}$. – St. Barth Jun 12 '24 at 20:10
  • Are you saying, then, that Theorem 2 does not imply that the family $(\prod_{j\in J_n}z_j)_{n\in\mathbb{N}}$ is multipliable in $\mathbb{C}^*$? (I'm sorry if I'm missing your point; but in order to try to understand, I asked you to specify, if possible, precisely which part of the last paragraph of my answer is incorrect. You haven't done that yet, so I'm trying to guess - with little sense that I'm guessing correctly.) – Calum Gilhooley Jun 12 '24 at 20:41
  • I'm saying that Theorem 2 isn't applicable to $(\prod_{j\in J_n}z_j){n\in\mathbb{N}}$ in the way you seem to apply it. In order for $(z_n){n\in\mathbb N}$ to be multipliable, the family $(|z_n-1|){n\in\mathbb N}$ needs to be summable, yes? So the additive family corresponding to $(\prod{j\in J_n}z_j){n\in\mathbb{N}}$ would be $(|\prod{j\in J_n}z_j-1|){n\in\mathbb{N}}$, no? How do you relate this to the family $(\sum{j\in J_n}|z_j-1|){n\in\Bbb N}$? This family is indeed summable by Theorem 2, but it has no discernible relation to $(\prod{j\in J_n}z_j)_{n\in\mathbb{N}}$ that I can see. – St. Barth Jun 13 '24 at 10:06
  • You correctly assert that the additive family corresponding to $(\prod_{j\in J_n}z_j){n\in\mathbb{N}}$ is $(|\prod{j\in J_n}z_j-1|)_{n\in\mathbb{N}},$ but you go astray in seeking a second proof of the summability of this family. It would be nice to have a proof which does not require wading through two volumes of Bourbaki (replete with distractions such as "commutative convergence"); indeed, I delayed posting my answer, in the hope that someone would post such a proof. Nevertheless, Theorem 2 does apply. If you maintain that it does not, please say which of its hypotheses are not satisfied. – Calum Gilhooley Jun 13 '24 at 14:11
  • Please edit your answer with a concrete proof that $(|\prod_{j\in J_n}z_j-1|){n\in\mathbb{N}}$ is summable (not just "we're done by Theorem 2"), because I see absolutely no reason why it should be or why Theorem 2 is applicable to it. Our original additive family is $(|z_j-1|){j\in\Bbb N}$ (which is summable), from which we pass to the additive family $(\sum_{j\in J_n}|z_j-1|){n\in\Bbb N}$ (which is summable by Theorem 2). As I've already said several times, I don't see how this last additive family relates to the multiplicative family $(\prod{j\in J_n}z_j)_{n\in\mathbb{N}}$. – St. Barth Jun 13 '24 at 14:26
  • Theorem 2 is a reason why! I think we are done here. I certainly have nothing more to add. OK, one thing: I'm probably not much happier than you are with having to take such a lot on trust from Bourbaki. I most certainly have not checked all the details of all the proofs, or even of all the definitions! But I trust Bourbaki to have got their sums right. Theorems 1 and 2, no matter what their internal complexity, can be applied to the present problem like "black boxes"; and the structure of the argument as a whole, as given in the last paragraph of my answer, is blessedly simple, is it not? – Calum Gilhooley Jun 13 '24 at 14:45
  • You don't seem to understand my concern: I have no doubts as to the veracity of Theorem 2, having proved it for myself in the past, what I don't understand is how you're applying it to this situation. – St. Barth Jun 13 '24 at 19:12
  • I have made it clear that I don't wish to continue this conversation, so please don't force me to do so against my will, by twisting my words. In the last two sentences of my last comment but one, I addressed your criticism of my application of Theorem 2 by making a simple request, which you simply ignored (as you ignored a similar request in my very first comment). By all means ignore my comment about Theorems 1 and 2! On all previous occasions I have resisted the temptation to speculate about what is bothering you, and I wish I had done so on that occasion too, although I was trying to help. – Calum Gilhooley Jun 13 '24 at 20:04