7

$$\large \pi^3 \gt 31$$

Using a calculator, $\pi^3/31 \approx 1.0002$, so I thought this may be challenging to do by hand.

It is extremely easy with the use of any calculator, so I was wondering now:

Can you prove the above inequality without the use of calculator or advanced computation in an elegant manner?

  • 11
    3.1415^3>31 (by hand) – sinbadh Jan 12 '16 at 23:14
  • 2
    There are all sorts of questions that begin with "prove without using a calculator." In my opinion, posing the question for the sake of the puzzle is reason enough, and in any case the question is clearly of some relevance to the OP. – pancini Jan 12 '16 at 23:41
  • 2
    I edited. It's a fair question plagued by some red herrings, like the fact that none of us care that 31 is prime or find it especially provocative. Edited out. – djechlin Jan 13 '16 at 01:25
  • @djechlin Great edit. Thank you –  Jan 13 '16 at 03:42

6 Answers6

21

The simplest way is to use the series:

$$\frac{1}{1^6}+\frac{1}{3^6}+\frac{1}{5^6}+...=\sum\limits_{k=1}^{\infty}\frac{1}{(2k-1)^6}=\frac{\pi^6}{960}$$

Now we want to prove that $$\frac{\pi^6}{960}>\frac{31^2}{960}$$ which means that we need to prove

$$960\sum\limits_{k=1}^{\infty}\frac{1}{(2k-1)^6}>31^2=961$$

However it is

$$960\sum\limits_{k=1}^{\infty}\frac{1}{(2k-1)^6}>960\left(\frac{1}{1^6}+\frac{1}{3^6}\right)=960+\frac{960}{3^6}$$

And now $\frac{960}{3^6}>1$ meaning $960>3^6$ because $320>3^5=9^2 \cdot 3=81 \cdot 3 = 243$

(Very minimal calculations involved, just to prove the point, although it was almost obvious once we got there.)

  • Could be streamlined, all you really need is $3^6=729<960$ so$$\pi^6>960\Bigl(1+\frac1{960}\Bigr)=31^2\ .$$Very nice all the same, +1. – David Jan 13 '16 at 01:18
  • Elegant, concise, and in the spirit of the question asked. – A.S. Jan 13 '16 at 01:18
  • I tend to be detailed slightly above just marking the solution path or giving general instructions even when all details are plotted. You just do not know who is reading. Like a student, every line with its purpose. Anyway it is just 960 .vs. 961 to compare. One step off and I fail. (Not that it did not happen.) –  Jan 13 '16 at 01:24
  • 5
    This is like using a jackhammer to pound in a framing nail. The facts relied on here need significant more computation than multiplication. You are just skipping over these details in the guise of known facts(but how do you know them? Because you did the computation originally). You might as well just say $\pi^3>31$ because I know that $\pi^3>31.00000001$. No computation necessary.@AlexPeter this is a complaint about the nature of the question, not about your answer, it is great, but just hides computation. – Sean English Jan 13 '16 at 03:03
  • @SE318 What is the alternative solution? How do you actually approximate $\pi$ instead? "Remembering" $3.1415\dots$ is not good enough. – A.S. Jan 13 '16 at 03:47
  • @A.S. Approximating pi isn't difficult and has been done well past the precision necessary for over 2000 years. One method that can be done using elementary facts about Euclidean geometry is to look at the area of regular polygons inscribed inside a unit circle. I'm not saying there isn't some computation needing to be done here, but I am saying that if we are not accepting $\pi^3>3.1415^3>31$ because it is too computationally complex, then we really can't accept any answers. Any viable answer "without" a computation will just wrap the computation up in some "known fact" so we don't see it. – Sean English Jan 13 '16 at 04:37
  • @SE318 Proving "known fact" used in the above answer will likely involve very little computation. Proving first $\pi>3.1415$ and raising the latter to the 6th power will involve more computation (i.e. elementary operations). You wouldn't claim that computing $\sum_1^n i$ term-by-term doesn't differ from $n(n+1)/2$ in terms of elementary operations, would you? Or that all series for $\pi$ require the same amount of computation for a given precision? The above answer is REALLY well suited for a given problem. – A.S. Jan 13 '16 at 05:40
  • @A.S. Again, I am not saying that proving $\pi^3>3.1415^3>31$ doesn't require some computation. I am saying that any solution is going to require some computation. This solution just hides it in known facts. Also, that reference you posted showing a proof of the result used here has huge computations in it and again involves non-elementary techniques that hide further computations(somehow you are okay with using an integral and Fourier series but not cubing?). I think this is a case where we just aren't going to agree. – Sean English Jan 13 '16 at 12:27
  • @SE3 The problem is not so much cubing - it is showing that $\pi>3.1415$ (even though cubing with needed accuracy will take a human quite a few elementary operations). Of course any solution will require some computation - computing $n(n+1)/2$ also requires 3 operations, but it's a vast improvements over alternative $(n-1)$ operations - so we are looking for a minimum. For a mathematician, (simple) integration, Parseval's identity, and Besel's problem are "built-in" and take a very low constant time ($~1$). $\pi>3.14$ could be considered "build-in", but $\pi>3.1415$ is not. – A.S. Jan 13 '16 at 17:37
  • @A.S. I am not sure why you think proving $\pi>3.1415$ is so difficult. As I said, you can quickly and easily get this estimate by calculating areas of regular polygons inscribed inside a circle. All you need to do is choose a large n so the area exceeds 3.1415. The derivation of this would be easy and not computationally complex. There are faster methods, but I prefer the polygons because the idea is elementary, using only facts about Euclidean geometry.The groundwork that would be necessary to even talk about an integral would take more time than to prove this entire result from axioms. – Sean English Jan 13 '16 at 17:44
  • @SE3 Let me repeat myself - in "mathematician" architecture, (simple) integrals are build-in and computing $\int x^n=x^{n+1}/(n+1)$ takes time $1$. The solution I linked is incredibly easy to follow (doesn't take many elementary operations to verify). Please include (computational) details of using regular polygons to show $\pi>3.1415$ - I'd like a direct comparison. The question wasn't about using the most elementary methods. It was about using the least computationally intensive method. Computation is based on a "mathematician" architecture - not 10th grader. – A.S. Jan 13 '16 at 18:02
  • @A.S. The "least computationally intensive method" is not well-defined when we allow for the use of "known facts". As I have said, if I "know" that $\pi>3.1415$ and I "know" that $3.1415^3>31$ then I need no computation to do this problem. The only way I can think of to choose what we "know" and don't "know" non-arbitrarily is to choose that we only "know" the relevant axioms. So "computationally intensive" and "elementary" are intrinsically intertwined. Unless you can show that this method is "computationally easy" from basic principles, all you are doing is committing sleight of hand. – Sean English Jan 13 '16 at 18:27
16

We know that $\zeta(6)=\frac{\pi^6}{945}$ and $\sum_{k=1}^\infty \frac1{k^6}$ converges quite quickly. It turns out we only need $k=1,2,3$ to prove the inequality.

$$\zeta(6)=\frac{\pi^6}{945}\Leftrightarrow \pi^3=\sqrt{945\zeta(6)}\Rightarrow \pi^3>\sqrt{945\sum_{k=1}^3 \frac1{k^6}}=\frac{\sqrt{4982145}}{72}>31.$$

Calculating the last square root is not very easy but it proves the inequality.

  • 1
    That's so awkward. I'm sure! Knowing the value of $\zeta(6)$ is far easier than the first few numbers of $\pi$! –  Jan 13 '16 at 01:08
  • @menag that's kind of a fair point but you can get $\zeta(2n)$ in a few pages of hand-written math proof, so perhaps if a few more digits of pi were required this would seem less of a stretch. – djechlin Jan 13 '16 at 01:22
  • @menag conversely if you can get $\zeta(2n+1)$ in any number of pages of proof for any $n > 0$ you will enjoy fame and fortune. – djechlin Jan 13 '16 at 01:23
  • 1
    You can avoid the last square root $4982145>(31·72)^2=4981824$ – Jaume Oliver Lafont Feb 09 '16 at 00:13
13

Well $\pi^3>3.1415^3>31$. Cubing 3.1415 is not pleasant by hand but most certainly can be done. Or are you looking for a more elegant solution?

siegehalver
  • 1,286
  • 5
    I am looking for a more elegant solution, yes –  Jan 12 '16 at 23:18
  • 1
    Any "elegant" proof is going to involve far more computation than just multiplying twice. – DanielV Jan 12 '16 at 23:26
  • Clearly the easiest way to prove this is by direct computation. That is why the OP requested a proof without computing the result directly. – pancini Jan 12 '16 at 23:27
  • 1
    @DanielV Showing $\pi>3.1415$ requires some ingenuity not to involve a lot of computation. – A.S. Jan 13 '16 at 09:00
7

It is well known, at least, I know it well :-) that $$\pi>3+\frac1{7+\dfrac1{15}}=3+\frac{15}{106}\ .$$ If you are prepared to accept this, then the rest can certainly be done with $\color{red}{\hbox{mental arithmetic}}$ (or pencil and paper at worst). By the binomial theorem we have $$\pi^3>3^3+3\frac{3^2\times15}{106}+3\frac{3\times15^2}{106^2} =27+\frac{405}{106}+\frac{45^2}{106^2}\ .$$ Now by a familiar trick, $$45^2=2025>1900+120>1900+6\times19=106\times19$$ and so $$\pi^3>27+\frac{405}{106}+\frac{19}{106}=31\ .$$

Comment. The initial inequality is in fact well known to those who know about $\color{red}{\hbox{continued fractions}}$. If it's not familiar in that context, one can quite easily divide mentally to confirm that $\frac{15}{106}<0.14151<\pi-3$.

David
  • 82,662
5

The answer is given for this question

Is there an integral for $\pi^4-\frac{2143}{22}$?

Linked to this paper.

http://www.jstor.org/stable/27642693

Extract of what you want:

For $\pi^3$

The truncated continued fractions are $31, 4930/159, 14821/478, \dots$.

\begin{align} \pi^3-31 &= \int_0^1 \frac{8 \, x^5 \, (1-x)^2 \, \left(324889-120736 \, x^2\right)} {445625 \, (1 + x^2) } \log^2 x \, dx\\ \frac{4930}{159}-\pi^3 &= \int_0^1 \frac{4 \, x^{10} \, (1-x)^4 \, \left(695774836+470936528857 \, x^2\right)} {470240754021 \, (1 + x^2) } \log^2 x \, dx. \end{align}

  • 14
    So we're not allowed to use multiplication (because it wouldn't be "elegant" enough) but we will allow a definite integral involving a ninth-order polynomial and $\log^2$?? Oh so elegant! – David G. Stork Jan 12 '16 at 23:32
  • 1
    Hey, I was only trying to answer the question based on the restrictions... – Jack Tiger Lam Jan 12 '16 at 23:33
  • I think this is an elegant solution. It relates this fact to a more general and interesting one - and if one accepts that that integral equals the claimed value (which is admittedly much harder than exponentiating 3.1415), it's clear that it's positive. The question might not be very good, but this is a strong answer to it. (If only I hadn't hit my voting cap for today - just gotta wait 20 more minutes) – Milo Brandt Jan 12 '16 at 23:37
  • 1
    But Jack Lam's isn't a solution! – David G. Stork Jan 12 '16 at 23:39
  • Well, given the arbitrary, and quite frankly, in my opinion, absurd, nature of the OP's question, we have no clue whatsoever as to what is and what isn't an acceptable solution. – Jack Tiger Lam Jan 12 '16 at 23:39
  • 2
    @Jack Lam Exactly! That's why this problem should be down voted. – David G. Stork Jan 12 '16 at 23:43
  • I dunno, I'm just here to try and be a mind reader, nothing else... rolls for telepathy – Jack Tiger Lam Jan 12 '16 at 23:43
  • 1
    I don't honestly understand why so many people hate this question. Can you please explain? –  Jan 12 '16 at 23:45
  • 1
    @JohnVine You have placed arbitrary and bizarre restrictions on the question in such a way that the question is impossible to answer. Also, "elegant", is subjective. – Jack Tiger Lam Jan 12 '16 at 23:46
  • I'm with @JohnVine. Incidentally, I'm writing a book How to ask good questions (see my TEDx talk, which focuses on math and sudoku) and this question fails on nearly every criterion. – David G. Stork Jan 13 '16 at 00:01
  • @DavidG.Stork Thank you...? –  Jan 13 '16 at 00:16
  • I think that Jack Lam's answer is ok since we don't have to solve the integral to see that it is greater than zero. – Elsa Jan 13 '16 at 00:19
  • (Ooops... I meant I'm with @JackLam... not John Vine.) My mistake. – David G. Stork Jan 13 '16 at 00:20
  • @DavidG.Stork Sorry for asking such a ridiculously bad question –  Jan 13 '16 at 00:25
  • Try and be specific about what you will allow and won't allow, it helps us figure out means of answering your question faster... Tell us if definite integrals are allowed, geometry, curve comparisons, etc. – Jack Tiger Lam Jan 13 '16 at 00:27
  • I don't know... Which is why I am trying desperately to delete it –  Jan 13 '16 at 00:28
  • The only reason I got to this right away was due to the $\pi^e < e^\pi$ inequality having been done using calculus. – Jack Tiger Lam Jan 13 '16 at 00:29
  • @JohnVine Don't worry about deleting the question, you have an answer now, so just let it be. – Jack Tiger Lam Jan 13 '16 at 00:29
  • @JackLam Ok. I guess I will just consider this a learning experience. If anyone here wants to punish me further, be my guest... –  Jan 13 '16 at 00:31
  • Nah, I'm just gonna try and figure out other means of proving the inequality, perhaps using geometry... – Jack Tiger Lam Jan 13 '16 at 00:34
  • Can a simpler integral for $\pi^3-31$ be built using $$\int_{0}^{1} \frac{192 (\tan^{-1}(x))^2}{1+x^2} dx = \pi^3$$ ? – Jaume Oliver Lafont Feb 10 '16 at 11:26
  • Yes, but that is immediately obvious, and doesn't involve polynomials on the numerator, which is what the paper was after. – Jack Tiger Lam Feb 10 '16 at 12:03
  • The examples are built on $$\int_{0}^{1} \frac{16(\log(x))^2}{1+x^2}dx=\pi^3$$ I was wondering whether changing the function (and then combining with polynomials as well) would lead to smaller numbers. – Jaume Oliver Lafont Feb 10 '16 at 21:31
  • $$\int_0^1 \frac{4x^3(1-x)^2\left(18504-60025x(1-x)^2\right)\log^2(x)}{50773\left(1+x^2\right)}dx=\pi^3-31$$ – Jaume Oliver Lafont May 03 '17 at 17:50
2

From $$\sum_{k=0}^\infty \frac{960}{(2k+1)^6} = \pi^6$$ we have $$\sum_{k=1}^\infty \frac{960}{(2k+1)^6} = \pi^6-960$$ and $$\sum_{k=2}^\infty \frac{960}{(2k+1)^6} = \pi^6-961-\frac{77}{243}$$

Since

$$960<961<961+\frac{77}{243}$$

we can form a new series for $\pi^6-961$ as a weighted sum of the two truncations. Solving the equation

$$\left(\pi^6-960\right)a+\left(\pi^6-961-\frac{77}{243}\right)b=\pi^6-961$$

for rational $a$ and $b$ yields $$a=\frac{77}{320}$$ $$b=\frac{243}{320}$$

Finally, $$\pi^6-961=(\pi^3-31)(\pi^3+31)=3\sum_{k=0}^\infty \left(\frac{77}{(2k+3)^6}+\frac{243}{(2k+5)^6}\right)$$

so

$$\pi^3-31=\frac{3}{\pi^3+31}\sum_{k=0}^\infty \left(\frac{77}{(2k+3)^6}+\frac{243}{(2k+5)^6}\right)$$

is positive because the series contains only positive terms.