26

Is there any representation of the exponential function as an infinite product (where there is no maximal factor in the series of terms which essentially contributes)? I.e.

$$\mathrm e^x=\prod_{n=0}^\infty a_n,$$

and by the sentence in brackets I mean that the $a_n$'s are not just mostly equal to $1$ or pairwise canceling away. The product is infinite but its factors don't contain a subseqeunce of $1$, if that makes sense.

There is of course the limit definition as powers of $(1+x/n)$., but these are no definite $a_n$'s, which one could e.g. divide out.

user91500
  • 5,817
Nikolaj-K
  • 12,559
  • 5
    cheating: let $a_n=e^{b_nx}$ where $\sum_nb_n=1$ – user8268 Sep 23 '13 at 06:56
  • @user8268: Okay, I like your idea. So for example $a_n=\mathrm e^{2^{-(n+1)}}$ In my mind, the $a_n$'s were of course simpler to compute than $\mathrm e^x$ itself. Like like rationals. – Nikolaj-K Sep 23 '13 at 06:58
  • Unfortunately you can't get it really simpler. That is, if $a_n$'s are entire functions of $x$ then they must be non-0 everywhere (as their product is), so each of $a_n$ is exp(some entire function). Perhaps there is som econtrived formula with non-entire functions. – user8268 Sep 23 '13 at 07:02
  • While in the lhs there is a (variable) $x$ I see only constants on the rhs. Where shall the variability be encoded in the rhs? – Gottfried Helms Jan 27 '14 at 09:21
  • @GottfriedHelms: Is the letter $a$ a constant by default? I intended those to be functions of $x$ - of course, solutions $a_n(x)=c_n^x$ where $c_n$ is an complex number are a slight cop-out, but work too. I came to ask the question because I generally have no idea how one does come up with product representations, and am baffled when I then see things like the Weierstrass factorization theorem. – Nikolaj-K Jan 27 '14 at 09:31
  • Just to link further: "(...) The idea of conversion of a powerseries into an infinite product-representation is not new and not only a game in recreational mathematics; according to H. Gingold/A. Knopfmacher in "Analytic Properties of Power Product Expansions" (1995) (...)" This deals with expansions $(1+ax)(1+bx^2)(1+cx^3)...$ - don't know whether this/this keyword is the right direction for you... In case it is then perhaps you like also my small treatize http://go.helms-net.de/math/musings/dreamofasequence.pdf – Gottfried Helms Jan 27 '14 at 10:45
  • @GottfriedHelms: Cool, thanks. – Nikolaj-K Jan 27 '14 at 10:56

9 Answers9

16

Not sure if this satisfies your assumptions, but this is an interesting infinite product for $|z|<1$, $$e^z=\prod_{k=1}^\infty (1-z^k)^{-\frac{\mu(k)}{k}},$$ where $\mu(k)$ is the Möbius function. See here.

Proof

Let's start with the logarithm of the product, $$-\sum_{k=1}^\infty\frac{\mu(k)}{k}\log\left(1-z^k\right)=\sum_{k=1}^\infty\frac{\mu(k)}{k}\sum_{\ell=1}^\infty\frac{z^{k\ell}}{\ell}.$$ We can rewrite this $$\sum_{n=1}^\infty\frac{z^n}{n}\sum_{d\mid n}\mu(d)=z+\sum_{n=2}^\infty\frac{z^n}{n}\sum_{d\mid n}\mu(d)=z$$ since $\sum_{d\mid n}\mu(d)=0$ for $n\geq 2$.

pshmath0
  • 11,208
  • Why is the right hand side not zero at $z=1$? – Nikolaj-K Feb 06 '17 at 19:53
  • 2
    $z=1$ is not permitted, i.e. $|z|<1$. – pshmath0 Feb 06 '17 at 21:00
  • Could you please provide a reference to the proof? – JKDASF Feb 24 '21 at 13:03
  • @JKDASF I don't have a reference but I will try to prove it. I've already made a start but don't have much time right now so will return to it later. – pshmath0 Feb 24 '21 at 15:51
  • 2
    @Pixel Thanks! Actually I've found a reference for that. See page 410 here: Bellman, Richard, and R. C. Buck. “4072.” The American Mathematical Monthly, vol. 51, no. 7, 1944, pp. 409–410. Your starting proof looks promising and we can just proceed by considering series expansion of log(1-x) and the compare the coefficients. – JKDASF Feb 25 '21 at 08:24
  • Convincing and mind-widening proof! Pragmatically, you make us compute a large many kth roots. Which can be done. But needs not. Thus snoke i on to https://math.stackexchange.com/a/3155156 and https://math.stackexchange.com/a/934788 – Roman Czyborra Jul 29 '21 at 23:14
14

There exists an infinite product for $e$ as follows:

If we define a sequence $\lbrace e_n\rbrace$ by $e_1=1$ and $e_{n+1}=(n+1)(e_n+1)$ for $n=1,2,3,...;$ e.g. $$e_1=1,e_2=4,e_3=15,e_4=64,e_5=325,e_6=1956,...$$ then $$e=\prod_{n=1}^\infty\frac{e_n+1}{e_n}=\frac{2}{1}.\frac{5}{4}.\frac{16}{15}.\frac{65}{64}.\frac{326}{325}.\frac{1957}{1956}. ...$$ For proof, first by induction we can show that if $s_n=\sum_{k=0}^n\frac1 {k!}$, then $e_n=n!s_{n-1}$,for $n\in\mathbb N$. And this immediately follows that $s_n/s_{n-1}=(e_n+1)/e_n$ and $s_n=\prod_{k=1}^n\frac{e_k+1}{e_k}$. Hence, $$e^x=\prod_{n=1}^\infty\left(\frac{e_n+1}{e_n}\right)^x.$$

user91500
  • 5,817
  • 4
    Very cool thanks. I translated it to Mathematica: e[n_] = If[n == 1, 1, n (1 + e[n - 1])]; Table[Product[1 + 1/e[n], {n, 1, N}], {N, 1, 7}] – Nikolaj-K Jan 27 '14 at 08:31
  • Question https://math.stackexchange.com/q/4279145/442 asks where this product expansion first appeared. – GEdgar Oct 17 '21 at 14:16
  • @GEdgar http://numbers.computation.free.fr/Constants/E/e.html – user91500 Oct 17 '21 at 15:17
  • Note: @user91500 is the one who posted this here. That HTML link seems faulty, but the postscript version comes out fine for me. http://numbers.computation.free.fr/Constants/E/e.ps This product is at the end of § 4. – GEdgar Oct 17 '21 at 18:22
10

If $x\geqslant0$ (or $x\ne-2^n$ for every $n\geqslant0$), one can use $$a_0=1+x,\qquad a_{n+1}=\left(1+\frac{x^2}{2^{n+2}(x+2^n)}\right)^{2^n} $$ If $x\leqslant0$ (or $x\ne2^n$ for every $n\geqslant0$), one can use $$a_0=\frac1{1-x},\qquad a_{n+1}=\left(1-\frac{x^2}{(2^{n+1}-x)^2}\right)^{2^n} $$ Where does this come from? From the identity, valid for every $n\geqslant0$, $$ \prod_{k=0}^na_k=\left(1\pm\frac{x}{2^n}\right)^{\pm2^n}. $$ The first identity (when $\pm=+$) yields a nondecreasing sequence of partial products. The second identity (when $\pm=-$) yields a nonincreasing sequence of partial products.

Did
  • 284,245
  • 2
    +1, Spontanously I don't see how to check the convergence, but I see your approach via approaching $(1+x/n)$. You "skip" the entire function thing, but making the coefficients case dependend. I wonder, can one get rid of the $2$'s for an arbitrary rational $r>1$? – Nikolaj-K Sep 23 '13 at 07:16
7

Amazingly, the exponential function can be represented as an infinite product of a product! That result was shown in the 2006 paper "Double Integrals and Infinite Products For Some Classical Constants Via Analytic Continuations of Lerch's Transendent" by Jesus Guillera and Jonathan Sondow.

It is proven in Theorem 5.3 that $$e^x=\prod_{n=1}^\infty \left(\prod_{k=1}^n (kx+1)^{(-1)^{k+1} {{n}\choose{k}}}\right) ^{1/n}$$

I dunno, this was too cool not to show you.

5

This function is related to the one Pixel posted, but it is not the same: $$e=\prod\limits_{n=1}^\infty\left(1-\frac{1}{\tau^n}\right)^{\frac{\mu(n)-\phi(n)}{n}}$$ where $\tau$ denotes the golden ratio, $\mu(n)$ denotes the Möbius Function and $\phi(n)$ denotes Euler's Totient Function.
("A Golden Product Identity for e" by Robert P. Schneider)

pshmath0
  • 11,208
Thorgott
  • 17,265
5

For any $x\in \mathbb{C}/{\mathbb{N}^{-}}$, we have : $$e^{x}=(1+x)\prod_{n=1}^{\infty}\left(1+\frac{x}{n}\right)^{-n}\left(1+\frac{x}{n+1}\right)^{n+1}$$

Mohammad Al Jamal
  • 575
  • 1
  • 16
  • 28
3

From Euler's $$\sin[2x]=\Im[e^{xi+xi}]=\Im[[\cos x+i\sin x][\cos x+i\sin x]]=2\sin[x]\cos[x]$$ $$\frac{\sin[πx]}{πx} =\prod_{n=1}^\infty\left[1^2-\left[\frac{x}n\right]^2\right]$$ we get $$\cos\left[\frac\pi2x\right]=\frac{\sin[πx]}{2\sin[\frac\pi2x]} =\frac{πx\prod_{n=1}^\infty\left[1^2-[\frac{x}{n}]^2\right]}{2\frac\pi2x\prod_{n=1}^\infty\left[1^2-[\frac{x}{2n}]^2\right]} \\=\frac{ [\frac{1-x²}{1}][\frac{4-x²}{4}] [\frac{9-x²}{9}][\frac{16-x²}{16}]\cdots} {[\frac{4-x²}{4}][\frac{16-x²}{16}] [\frac{36-x²}{36}][\frac{64-x²}{64}]\cdots} =\prod_{n=1,3,5}^\infty\left[1^2-\left[\frac{x}{n}\right]^2 \right]$$

which together with$$\sin\left[\fracπ2x\right]=\cos\left[\fracπ2[x-1]\right]$$

yields $$i^x=\exp\left[i\fracπ2x\right] =\cos\left[\frac\pi2x\right]+i\sin\left[\frac\pi2x\right] \\=\left[\prod_{n=1,3,5}^\infty\left[1^2-\left[\frac{x}{n}\right]^2 \right]\right] +i\left[\prod_{n=1,3,5}^\infty\left[1^2-\left[\frac{x-1}{n}\right]^2 \right]\right] $$ consisting of two infinite products. Maybe can we factor this expression in some way?

Mark
  • 7,702
  • 6
  • 41
  • 80
2

We have$$\mathrm e^{-x}=\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad$$$$(1-x)(1+\tfrac12x^2)(1+\tfrac13x^3)(1+\tfrac38x^4)(1+\tfrac15x^5)(1+\tfrac{13}{72}x^6)(1+\tfrac17x^7)(1+\tfrac{27}{128}x^8)(1+\tfrac{8}{81}x^9)\cdots$$for $|x|<1$. Some coefficients for these product terms are listed in OEIS as A170910 and A170911. See the paper on power product expansions by Gingold et al. (1988). The result is due to O. Kolberg (1960), who showed that the coefficient of $x^n$ in the $n$th factor is $1/n$ if and only if $n$ is prime.

John Bentin
  • 20,004
1

OP here. I just realized that the following should hold in general:

$$\lim_{n\to\infty}a_n=a_1+\sum_{n=1}^\infty(a_{n+1}-{a_n}),$$

and for finite $a_n$, similarly

$$\lim_{n\to\infty}a_n=a_1\cdot\prod_{n=1}^\infty\frac{a_{n+1}}{a_n}.$$

Hence, with $a_1=\prod_{n=1}^\infty (a_1)^{2^{-n}}$ and for $x$ that aren't negative integers,

$$\mathrm {exp}(x)=\prod_{n=1}^\infty\ (1+x)^{2^{-n}}\left(1+\frac{x}{n+1}\right)^{n+1}\left(1+\frac{x}{n}\right)^{-n}.$$

Nikolaj-K
  • 12,559
  • Good to know that for $z\in(-1,1)$, we have $${\mathrm e}^z = \prod_{n=1}^\infty (1-z^n)^{-\mu(n)/n}$$ Series[Product[(1-z^n)^(-MoebiusMu[n]/n), {n,1,5}], {z,0,5}] – Nikolaj-K Dec 11 '20 at 22:44
  • 1
    I just added a proof in my old answer above. – pshmath0 Feb 25 '21 at 21:39