For any real number $r$ we can find a pair of natural numbers $N$ and $M$, such that $\sqrt{N}-\sqrt{M}$ will approximate $r$ with any given precision (if we choose $N,M$ large enough).

That's why I tried to think of any increasing sequences of integer numbers which would allow to compute $\pi$ in this way. And I mean, without knowing the digits of $\pi$ beforehand. Thus, $a_n, b_n$ have to be defined by recursion themselves. However, I wasn't able to make any progress.

1) The easiest way is to use the known value of $\pi$ to compute various pairs of $N,p$, such that:

$$\sqrt{N+p}-\sqrt{N} \approx \pi$$

$$N=\left[\frac{(p-\pi^2)^2}{4 \pi^2} \right]$$

Here $[]$ is the floor function. For example:


2) If we let the numbers be rational, and not necessarily increasing, we can use the known series to make something resembling what I want:

$$\pi^2=6 \left(1+\frac{1}{2^2}+\frac{1}{3^2}+\cdots \right)$$

$$\pi=\sqrt{6 \left(1+\frac{1}{2^2}+\frac{1}{3^2}+\cdots \right)}-\sqrt{0}$$

3) Another way, is to use the Spiral of Theodorus (spiral of square roots):

$$\rho (n)=\sqrt{n}$$

$$\phi (n)=\sum_{k=1}^{n} \arcsin \frac{1}{\sqrt{n}}-\frac{\pi}{2}$$

enter image description here

The $\pi$ arises as a limit of radii difference at approximately the same angle:

$$\lim_{n \to \infty} \rho(\phi+2\pi)-\rho(\phi)=\pi$$

For example, the first such pair of numbers is:


However, I'm not sure if it's possible to properly define a sequence I want using this definition for a spiral - it's just too complicated (especially, finding pairs of rays with the closest angles).

Can we find sequences $\{a_n\}, \{b_n\}$ with $a_n,b_n \in \mathbb{Q}$, increasing and defined by recursion, such that $$\lim_{n \to \infty} (\sqrt{a_n}-\sqrt{b_n})=\pi$$?

Or can we prove that we can't?


I only consider rational sequences because I don't believe that any such integer sequences exist. However, I would be very happy if someone answers with integer sequences.

The main thing though, I want something different from $$\lim_{n \to \infty} \sqrt{a_n} = \pi+c, ~~~~ \lim_{n \to \infty} \sqrt{b_n}=c$$


Let $\sum \limits _{k=0} ^\infty x_k$ be any series with positive and rational terms that converges to $\pi$. One such example may be

$$\sum _{k=0} ^\infty \frac 1 {16 ^k} \left( \frac 4 {8k+1} - \frac 2 {8k+4} - \frac 1 {8k+5} - \frac 1 {8k+6}\right)$$

but you may find others by considering MacLaurin series of various inverse trigonometric functions and evaluating them at special simple points (for instance the series of $\arcsin$ evaluated at $\frac 1 2$, multiplied by $6$).


$$a_n = \left( \sum _{k=0} ^n x_k + 1 \right)^2, \quad b_n = \left( 1 - \frac 1 n \right) ^2 .$$

Notice that $\sqrt{a_n} - \sqrt{b_n} \to (\pi + 1) - 1 = \pi$. (The $1$ here is arbitrary, you may choose any other strictly positive number.)

To show that the sequences are recursive, notice that

$$a_{n+1} = a_n ^2 + 2 x_{n+1} a_n + x_{n+1} ^2, \quad b_{n+1} = b_n ^2 - \frac 2 {n (n+1)} b_n + \frac 1 {n^2 (n+1)^2} .$$

(Notice, though, that the recursion is not linear - but you have not asked for such a property.)

Finally, the sequences are increasing being squares of other increasing sequences (for $a_n$ here it is where the positivity of $x_n$ matters). It is also obvious that the sequences are made only of rational numbers (here is where the rationality of $x_n$ matters.)

  • $\begingroup$ I knew someone would use a trick like this. I considered it myself. So, no 'interesting' sequences then. $\endgroup$ – Yuriy S Mar 20 '16 at 20:29
  • $\begingroup$ As for the recursion - I can be anything, not necessary linear. +1 for you, but I would like something not so trivial $\endgroup$ – Yuriy S Mar 20 '16 at 20:35

I actually found a method to do what I want (without using a trivial $b_n$ sequence like in Alex' answer).

First we use a continued fraction for $\pi$ which has a regular structure, for example:


Then (instead of taking partial convergents, like we usually do) we define a sequence of quadratic irrationals the following way:




And so on, we get a sequence of quadratic irrationals converging to $\pi$.

It's hard to define an explicit recurrence realtion for $x_n$ in this case, but I never said the recursion has to be explicit. The main point was to get a sequence of approximations in the form of root differences without knowing the digits of $\pi$ beforehand.

In this case we get:

$$4 x_1=2 \sqrt{5}-2=2.472\dots$$

$$4 x_2=2 \sqrt{21}-6=3.165\dots$$

$$4 x_3=\frac{\sqrt{745}}{2}-\frac{21}{2}=3.147\dots$$

$$4 x_4=\frac{\sqrt{49585}}{12}-\frac{185}{12}=3.140\dots$$

Both $a_n$ and $b_n$ are increasing (we can prove it) and defined by recursion (in a very broad sense).

Using other regular continued fractions for $\pi$ we can get other such approximations.

Using the simple continued fraction for $\pi$ we of course can get much better quadratic irrational approximations on the first few steps, such as:


However, we have to know the digits of $\pi$ to get a simple continued fraction.

  • $\begingroup$ How do these $x_n$ form a sequence given by recursion (even with a broader definition of recursion)? $\endgroup$ – Alex M. Mar 30 '16 at 10:12
  • $\begingroup$ In a very convoulted way. Since this continued fraction can be defined recursively, I'm sure it's possible to recover every $x_n$ from $x_{n-1}$ using the general form of the CF. I'm just not sure how. Anyway, I tried to clarify here what I meant in my question (to get an infinite sequence of approximations without knowing $\pi$). Note, that I will not change the question itself, so your answer is of course valid $\endgroup$ – Yuriy S Mar 30 '16 at 11:07
  • $\begingroup$ I aslo wanted both $a_n$ and $b_n$ unbounded, which I did not state in my question, so again, your answer is valid. Perhaps, I should delete this question and create a new one reflecting all the changes $\endgroup$ – Yuriy S Mar 30 '16 at 11:24

$$a_{n+1} = a_n + \frac{24}{(n+1)^2} ,~~~ a_1 =25, ~~~a_2 =6 ,~~~ b_{n+1} = b_n + \frac{6}{(n+1)^2} , ~~~b_1=6 , ~~~b_2=\frac{6}{4}$$

  • 1
    $\begingroup$ You have a first order recurrences, why two initial conditions? They also do not appear to match each other and the relations $\endgroup$ – Yuriy S Mar 20 '16 at 18:42
  • 1
    $\begingroup$ I can't undertand what you mean by your expressions. Why $a_2$ and $b_2$ do not match the recurrence? If I use your relations with $a_1$ and $b_1$, I get $a_2=31$ and $b_2=15/2$ $\endgroup$ – Yuriy S Mar 20 '16 at 18:51
  • $\begingroup$ Anyway, I tried all four combinations of initial conditions, they don't work $\endgroup$ – Yuriy S Mar 20 '16 at 19:07

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.