Is there any elementary proof that the limit of infinite product $\prod_{k=1}^{\infty}(1-\frac{1}{2^k})$ is not 0?


This is similar to the beginning of Alex's answer, however it proceeds differently.
We express the log of the product as a sum of logarithms
$$\small L=-\log(\prod_{k=1}^\infty (1-{1 \over 2^k})) = -\sum_{k=1}^\infty \log(1-{1 \over 2^k})$$ Next we expand each of the logs according to its power series and write this as a double sum; then we change order of summation and sum up columnwise; this is allowed because both directions provide convergent sums. So $$\small \begin{array} {lllllll} -\log(1-1/2)=&+1/2 & +1/2 \cdot 1/2^2 & +1/3 \cdot 1/2^3 & +1/4 \cdot 1/2^4 & \ldots \\ -\log(1-1/4)=&+1/4 & +1/2 \cdot 1/4^2 & +1/3 \cdot 1/4^3 & +1/4 \cdot 1/4^4 & \ldots \\ -\log(1-1/8)=&+1/8 & +1/2 \cdot 1/8^2 & +1/3 \cdot 1/8^3 & +1/4 \cdot 1/8^4 & \ldots \\ -\log(1-1/16)=&+1/16 & +1/2 \cdot 1/16^2 & +1/3 \cdot 1/16^3 & +1/4 \cdot 1/16^4 & \ldots \\ \cdots & & \cdots &\cdots &\cdots & \\ \hline \\ L = & 1 & + 1/2 \cdot 1/3 & + 1/3 \cdot 1/7 & + 1/4 \cdot 1/15 & \ldots \\ \hline \\ \zeta(2)=&1&+ 1/2\cdot 1/2& + 1/3 \cdot 1/3 & + 1/4 \cdot 1/4 & \dots \\ \end{array} $$ We compare L with $\small \zeta(2) $ and observe, that the absolute value each term in L is smaller than the corresponding term in $\small \zeta(2)$ thus L is a finite/converging sum $\small 0 \lt L \lt \zeta(2) \lt \infty $ and thus $\small \exp(-L) \ne 0 $ . QED .


Using the same "product-to-sum" strategy as @Did, we get
$${1\over 2}\prod_{k=2}^N\left({1-{1\over 2^k}}\right) \geq {1\over 2}\left(1-\sum_{k=2}^N{1\over 2^k}\right) ={1\over 4}+\left({1\over 2}\right)^{N+1}\geq {1\over 4}.$$

  • $\begingroup$ @Didier Thanks for the kind words. $\endgroup$ – user940 Nov 4 '11 at 12:13

We begin with a simple inequality. For every $x$ in $[0,\frac12]$, $\mathrm e^x\leqslant2$ hence $\frac12\leqslant\mathrm e^{-x}$. Integrating this on the interval $[0,x]$ yields $\frac12x\leqslant1-\mathrm e^{-x}$, that is, $1-\frac12x\geqslant\mathrm e^{-x}$.

Applying this to $x=\frac1{2^{k-1}}$ for each $k$ between $2$ and $n$, one gets $$ \prod_{k=1}^n\left(1-\frac1{2^k}\right)\geqslant\frac12\exp\left(-\sum_{k=2}^n\frac1{2^{k-1}}\right). $$ Since the sum in the exponential is less than $1$, all the partial products are $\geqslant\frac1{2\mathrm e}$ and $$ \prod_{k=1}^{+\infty}\left(1-\frac1{2^k}\right)\geqslant\frac1{2\mathrm e}>0. $$

Edit The convexity inequality $1-x\geqslant4^{-x}$, valid for every $x$ in $(0,\frac12]$, yields the lower bound $\frac14$.

  • 2
    $\begingroup$ Of course, it is a general fact that, for any sequence $(a_k)_k\subset[0,1)$, the infinite product $\prod\limits_k(1-a_k)$ is positive if and only if the series $\sum\limits_ka_k$ converges. $\endgroup$ – Did Nov 3 '11 at 19:46

The following requires no tools other than induction. In particular, machinery from the calculus is not used.

Let $f(n)$ be the product up to the term $1-\dfrac{1}{2^n}$. We show by induction that $$f(n)\ge\frac{1}{4} +\frac{1}{2^{n+1}}.$$ The result is true at $n=1$. For the induction step, note that $$f(m+1)=f(m)\left(1-\frac{1}{2^{m+1}}\right)\ge\left(\frac{1}{4}+\frac{1}{2^{m+1}}\right)\left(1-\frac{1}{2^{m+1}}\right).$$ Expand the product on the right. We get $$\frac{1}{4}+\frac{1}{2^{m+1}}-\frac{1}{4}\cdot\frac{1}{2^{m+1}}-\frac{1}{2^{2m+2}}.\qquad(\ast)$$ Rewrite $\dfrac{1}{4}\cdot\dfrac{1}{2^{m+1}}$ as $\dfrac{1}{2^{m+2}}-\dfrac{1}{2^{m+3}}$. Then $(\ast)$ becomes $$\frac{1}{4}+\frac{1}{2^{m+2}}+\frac{1}{2^{m+3}}-\frac{1}{2^{2m+2}}.$$ The term $\dfrac{1}{2^{m+3}}-\dfrac{1}{2^{2m+2}}$ is non-negative. Thus $f(m+1)\ge\dfrac{1}{4}+\dfrac{1}{2^{m+2}}$, which completes the induction step.

  • $\begingroup$ When you substitute $\frac{1}{2^{m+2}}-\frac{1}{2^{m+3}}$ in (*) it should become: $\frac{1}{4} $\endgroup$ – Chun-Yue Nov 4 '11 at 0:02
  • $\begingroup$ The $-(1/4)(1/2^{m+1})$ becomes $-1/2^{m+2}+1/2^{m+3}$. Then the $-1/2^{m+2}$ combines with the $1/2^{m+1}$ to the left of it to give $1/2^{m+2}$ (I skipped a step). $\endgroup$ – André Nicolas Nov 4 '11 at 0:17

I wonder if this works. It suffices to show that $\displaystyle \sum_{k=1}^{\infty}\log\left(1-\frac{1}{2^k}\right)>-\infty$. Note though that $\displaystyle \lim_{k\to\infty}\frac{\log\left(1-\frac{1}{2^k}\right)}{-\frac{1}{2^k}}=1$ and $\displaystyle \sum_{k=1}^{\infty}-\frac{1}{2^k}$ converges easily by the ratio test.

  • 1
    $\begingroup$ I think you have that last formula wrong - as the terms go to $1$ as $k$ goes to infinity, the sum can't possibly converge. Did you mean just $-1\over 2^k$ instead of $1-{1\over 2^k}$? $\endgroup$ – Steven Stadnicki Nov 3 '11 at 20:11
  • $\begingroup$ Of course! Thank you Steven! $\endgroup$ – Alex Youcis Nov 3 '11 at 20:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.