33

How do you evaluate $\sqrt { 2+\sqrt { 3+\sqrt { 5+\sqrt { 7+\sqrt { 11+ \dots } } } } } $ ?

This question appears to be rather difficult as there is no way to perfectly know what $p_{ n }$ is , if $p_{ n }$ denotes the $n$th prime.

It is simple to show that the value above is convergent. Bertrand`s Postulate implies that $p_{ n } \le 2^n$, which implies that $\sqrt { 2+\sqrt { 3+\sqrt { 5+\sqrt { 7+\sqrt { 11+ \dots } } } } } \le \sqrt { 2+\sqrt { 4+\sqrt { 8+\sqrt { 16+\sqrt { 32+\dots } } } } } $, which is convergent, as seen here. So it is pretty clear that $\sqrt { 2+\sqrt { 3+\sqrt { 5+\sqrt { 7+\sqrt { 11+ \dots } } } } } $ is convergent.

However, in which fashion can you evaluate the value above? If there is no exact way to evaluate it, is it irrational or rational?

The value seems to be about $2.10359749633989726261993..$ as seen here.

Any help would be appreciated.

Chad Shin
  • 2,132
  • 2
    More likely than not this is a coincidence, but the Inverse Symbolic Calculator gives that the given numerical approximation agrees with E/(TwinPrim+Li4(1/2))+1/2 to the number of decimal places given. – Travis Willse Jan 08 '16 at 01:12
  • 4
    Decimal expansion and some discussion here: https://oeis.org/A105546 – Chris Culter Jan 08 '16 at 01:32
  • 1
    Can we squeeze it between $\sqrt{2+\sqrt{3+\sqrt{4+\sqrt{5+...}}}}$ and $\sqrt{2+\sqrt{4+\sqrt{8...}}}$ – ShankRam May 08 '16 at 17:33
  • The former comes to around 2.090327066201934000204755281140588912195926277376769642782 after 10 terms – ShankRam May 08 '16 at 17:35
  • @Travis A number so close to the given number ? I can barely believe that this should be a coincidence. The given number has decimal expansion $$2.10359749633989726261993964968532544404216228824\cdots $$ – Peter Apr 08 '17 at 21:46
  • The algdep-routine in PARI/GP with a preicision of about $5,000$ digits gives a polynomial with coefficients about $25,000$ digits long, so the given number is very probably irrational. – Peter Apr 08 '17 at 21:58
  • @Travis: The relation you gave C= E/(TwinPrim+Li4(1/2))+1/2, can you give a link to the value of TwinPrime? I can't get your relation to equal $C =2.103597496339\dots$ – Tito Piezas III Apr 11 '17 at 11:06
  • @Peter: Have you verified that the relation given by Travis works to the given number of decimal places? – Tito Piezas III Apr 11 '17 at 11:09
  • @TitoPiezasIII To be sure that my result is correct to my accuarcy, I calculated the value for a far larger prime $p$ and got the same result within the given accuracy. But no, I did not check Travis's formula. – Peter Apr 11 '17 at 11:17
  • If Travis's formula should actually hold, can we have the hope that we can solve the twin-prime-problem using this limit ? – Peter Apr 11 '17 at 11:21
  • @Peter: I tried $C=2.103597496339\dots$ in this version of the ISC http://wayback.cecm.sfu.ca/projects/ISC/ISCmain.html but it did not give a hit, so I was wondering where Travis got that relation. – Tito Piezas III Apr 11 '17 at 11:26
  • @Peter: IF it holds. By plugging in the Twin Prime Constant or the related Brun's Constant, I can't get Travis' relation to work. And I assume Li4(1/2) is PolyLog[4,1/2[. – Tito Piezas III Apr 11 '17 at 11:31
  • @Peter When I made that comment, the question provided a much shorter decimal expansion: $2.103592623\ldots$---according to the edit history the number was revised a few hours later. NB that this number and the number currently given in the post disagree in the 8th decimal place after the post. – Travis Willse Apr 11 '17 at 12:04
  • @TitoPiezasIII This was now a long time ago, but I seem to recall using ISC+: https://isc.carma.newcastle.edu.au/index . Unfortunately trying to load the page now gives me a proxy error. – Travis Willse Apr 11 '17 at 12:06
  • @Travis: Ah, so it was much shorter before. That explains it. By the way, I get the same proxy error. – Tito Piezas III Apr 11 '17 at 12:22
  • S = √2+√3+√5+√7+ ... S = √+√+√+√+ ... = (1+√1+4 )/2 a=2→ S2 = 2 a=3→ S3=2.3 Suppose √3+√3+√5+√7+ ... ≈√3+√3+√3+√3+ ... =S3 = 2.3… then: S2 <√2+√3+√5+√7+ ... < S3 → 2 <√2+√3+√5+√7+ ...<2.3 – sirous Apr 12 '17 at 06:55

2 Answers2

4

For the primes $(p_k):=(2,3,5,\dots)$ we have

$$\tag{1}0\le\dfrac{p_{n+1}}{p_1p_2\cdots p_n}\le2.$$

The inequality is trivial if $p_n\lt p_{n+1}\lt p_1p_2\cdots p_n$. Otherwise the number $1+p_1p_2\cdots p_n$ is a prime and the inequality holds.

For $a,x\gt0$ we have

$$0\lt\sqrt{a+x}={\sqrt{a}+\int_a^{a+x}\dfrac{d\xi}{2\sqrt{\xi}}}\le {\sqrt{a}+\int_a^{a+x}\dfrac{d\xi}{2\sqrt{a}}}={\sqrt{a}+\dfrac{x}{2\sqrt{a}}}.$$

We obtain

$$\tag{2}\sqrt{p_1}\lt{\sqrt{p_1+\sqrt{p_2+\sqrt{p_3+\sqrt{p_4+\sqrt{\dots+\sqrt{p_n}}}}}}}\le {\sqrt{p_1}+\dfrac{1}{2\sqrt{p_1}}\sqrt{p_2+\sqrt{p_3+\sqrt{p_4+\sqrt{\dots+\sqrt{p_n}}}}}}\le {\sqrt{p_1}+\dfrac{1}{2}\sqrt{\dfrac{p_2}{p_1}}+\dfrac{1}{2^2\sqrt{p_1p_2}}\sqrt{p_3+\sqrt{p_4+\sqrt{\dots+\sqrt{p_n}}}}}\le {\sqrt{p_1}+\dfrac{1}{2}\sqrt{\dfrac{p_2}{p_1}}+\dfrac{1}{2^2}\sqrt{\dfrac{p_3}{p_1p_2}}+\dfrac{1}{2^3\sqrt{p_1p_2p_3}}\sqrt{p_4+\sqrt{\dots+\sqrt{p_n}}}}\le {\sqrt{2}+\dfrac{1}{2}\sqrt{2}+\dfrac{1}{2^2}\sqrt{2}+\dfrac{1}{2^3}\sqrt{2}+\dots}\le{2\sqrt{2}}$$

and the limit

$$\sqrt{2}\lt\lim_{n\to\infty}{\sqrt{p_1+\sqrt{p_2+\sqrt{p_3+\sqrt{p_4+\sqrt{\dots+\sqrt{p_n}}}}}}}\le{2\sqrt{2}}$$

exists. Though a strictly monotonic bounded sequence converges we prefer to know the error of the sum

$$s_{n-1}:={\sqrt{p_1+\sqrt{p_2+\sqrt{p_3+\sqrt{p_4+\sqrt{\dots+\sqrt{p_{n-1}}}}}}}}\;.$$

Similar to above we have

$${\sqrt{p_n}}\le\sqrt{p_n+\sqrt{p_{n+1}+\sqrt{\dots}}}\le{\sqrt{p_n}+\dfrac{1}{2}\sqrt{\dfrac{p_{n+1}}{p_{n}}}+\dfrac{1}{2^2}\sqrt{\dfrac{p_{n+2}}{p_{n}p_{n+1}}}+\dots}$$

and with the inequality $(1)$ we obtain

$${\dfrac{\sqrt{p_n+\sqrt{p_{n+1}+\sqrt{\dots}}}}{2^{n-1}\sqrt{p_1p_2\cdots p_{n-1}}}}\le {\dfrac{1}{2^{n-1}}\sqrt{\dfrac{p_{n}}{p_1p_2\cdots p_{n-1}}}+\dfrac{1}{2^{n}}\sqrt{\dfrac{p_{n+1}}{p_1p_2\cdots p_{n}}}+\dots}\le {\dfrac{\sqrt{2}}{2^{n-1}}+\dfrac{\sqrt{2}}{2^{n}}+\dots}\le\dfrac{\sqrt{2}}{2^{n-2}}$$

or

$$\sqrt{p_n+\sqrt{p_{n+1}+\sqrt{\dots}}}\le2\sqrt{2p_1p_2\cdots p_{n-1}}.$$

We define

$$\tag{3}{s_{n-1}(\sigma):=\sqrt{p_1+\sqrt{p_2+\sqrt{\dots+\sqrt{p_{n-1}+\sigma}}}}}.$$

and obtain

$$s_{n-1}(0)\le s_{\infty}\le s_{n-1}(2\sqrt{2p_1p_2\cdots p_{n-1}}).$$

We get

$$s_5(0)=2.1028\dots\le s_{\infty}\le2.1046\dots=s_5(135.94\dots)$$

and

$$s_{10}(0)=2.10359748\dots\le s_{\infty}\le2.10359778\dots=s_{10}(227502.84\dots).$$

1

Since convergence has been established, let's note $$\alpha=\sqrt{p_1+\sqrt{p_2+\sqrt{p_3+...}}}$$ A very raw estimation may be obtained from $$\alpha\sqrt{2}=\sqrt{2p_1+2\sqrt{p_2+\sqrt{p_3+...}}}=\sqrt{2p_1+\sqrt{4p_2+\sqrt{16p_3+...}}}>...$$ using Bertrand's postulate $$...>\sqrt{p_2+\sqrt{2p_3+\sqrt{8p_4+...}}}>\sqrt{p_2+\sqrt{p_3+\sqrt{p_4+...}}}=\alpha^2-p_1=\alpha^2-2$$ $$\alpha^2-\alpha\sqrt{2}-2<0$$ this gives an estimate of $$2.0708...=\sqrt{2+\sqrt{3}+\sqrt{5}}<\alpha<\frac{1+\sqrt{5}}{\sqrt{2}}=2.2882...$$

rtybase
  • 16,907