4

B4 exercise: Express $$x=\sqrt[8]{2207-\cfrac{1}{2207-\cfrac{1}{2207-\cfrac{1}{2207-\cfrac{1}{\ddots}}}}}$$ as $\dfrac{a+b\sqrt{c}}{d}$, where $a,b,c,d$ are integers.

In the following solution https://kskedlaya.org/putnam-archive/1995s.pdf what is the idea behind computing

$$(x^2 - ax + 1)(x^2 + ax + 1) = x^4 - (a^2-2)x^2 + 1,$$

and why is the conclusion "the positive square roots of the quadratic $x^2 − bx + 1$ satisfy the quadratic $x^2 − x\sqrt{b^2 +2} + 1 = 0$"?

ladca
  • 45
  • 1
    You should include the question B-4 too... – Integral Nov 27 '23 at 23:10
  • 1
    @Integral added – ladca Nov 27 '23 at 23:17
  • @CalvinLin But 2 questions: If they want to square the polynomial why they do $(x^2 - ax + 1)(x^2 + ax + 1)$ instead of $(x^2 - ax + 1)^2$? And why do they conclude that the square root will have the same degree, e.g, $x^2$. To me it would make sense that the square root of a polynomial of degree $x^2$ would be $x$, and the square root of the later would be $x^{1/2}$ – ladca Nov 28 '23 at 16:17
  • 1
    @CalvinLin If you have the time, I would encourage you to delete your comments, consolidating them into an answer. Also, I would include the OP's interaction with you in the comments as part of your answer. – user2661923 Jan 04 '24 at 14:43
  • @user2661923 But OP didn't provide much context / show their effort, so I'm not supposed to post an answer by community standards. – Calvin Lin Jan 04 '24 at 21:33
  • 1
    @CalvinLin Ordinarily, I would agree with you. However, I examined the corresponding PDF. In my opinion, the OP's question : "what is the idea behind computing..." is very well taken. That is, in this instance, it isn't really reasonable to ask the OP to show work. The problem solver's thinking is not at all transparent. The OP isn't asking for the solution. Instead, the OP (in effect) asked: what is the problem solver thinking? – user2661923 Jan 04 '24 at 22:21
  • 1
  • @user2661923 FWIW This question simply being closed as a duplicate is part of the reason for my hesitation in answering. – Calvin Lin Feb 26 '24 at 18:58
  • Why has my answer been accepted even though it is practically incomplete? I am wracking my brains here trying to find a proof for $(12)$ and it is stated that the solution will be worked on until it's complete. – Yajat Shamji Feb 27 '24 at 17:04
  • @YajatShamji not to worry. I reversed my vote so that Calvin's answer is now on top. Let me know when you're done and I'll change my vote back. – Benjamin Wang Feb 27 '24 at 23:59
  • @BenjaminWang, I was not talking about the vote. I was talking about why my solution has been accepted by the OP. – Yajat Shamji Feb 28 '24 at 14:17

3 Answers3

7

The inner expression $L$ satisfies the quadratic equation $L^2 - 2207L + 1 = 0 $.
The goal is to determine $ x = \sqrt[8]{L}$.
The approach we take is to find quadratic equations satisfied by $ \sqrt{L}$, then $\sqrt[4]{L}$, then $\sqrt[8]{L}$, which would allow us to calculate the value of $x$.

Claim: For $ a \geq 2$, suppose the roots of $ x^ 2 - ax + 1 = 0 $ are $ x_1, x_2 $ (which will be positive reals), then the roots of $ y^2 - \sqrt{a + 2 } y + 1 = 0 $ are $ \sqrt{ x_1}, \sqrt{ x_2} $.

My Proof: (Presenting this first as I believe it's easier to follow and hence understand why it is true.)

$\begin{align} & x_1 - \sqrt{ a + 2} \sqrt{ x_1} + 1 = 0 \\ \Leftrightarrow \quad & x_1 + 1 = \sqrt{a+2} \sqrt{x_1} \\ \Leftrightarrow \quad & x_1^2 + 2x_1 + 1 = (a+2)x_1 \\ \Leftrightarrow \quad & x_1^2 - ax_1 + 1 = 0 \\ \end{align}$

Note that the condition that $ a \geq 2$ gives us that the roots are positive reals, and so $ \sqrt{x_i}$ is unambiguous. A similar statement holds when $ a < 2$. (I encourage you to investigate this.) $_\square$

Kiran's proof: Using the transformation of roots $ \sqrt{x} = y \Leftrightarrow x = y^2$, we have $y^4 - ay^2 + 1 = 0 $.

The expression that OP asks about is $$(m^2 - \alpha m + 1 ) ( m^2 + \alpha m + 1) = m^4 - ( \alpha^2 - 2 ) m^2 + 1 $$ where I replaced the variables to reduce confusion within the context of proving this claim.

With this expression, letting $m = y, \alpha^2 - 2 = a$, we can see the factorization $$ (y^2 - \sqrt{a+2} y + 1) ( y^2 + \sqrt{a+2}y + 1) = y^4 - ay^2 + 1 = 0 .$$

Now, since the 2 square roots are positive, we have to reject $(y^2 + \sqrt{a+2} y + 1) = 0 $ (which has no positive roots), and hence they must satisfy $(y^2 - \sqrt{a+2} y + 1) = 0 $.

Note that the negative square roots satisfy $(y^2 + \sqrt{a+2} y + 1) = 0 $. $_\square$

Corollary:

  • Since $ \sqrt{ 2207 + 2 } = 47$, thus $\sqrt{L}$ satisfies the quadratic equation $y^2 - 47y + 1 = 0 $.
  • Since $ \sqrt{ 47 + 2 } = 7$, thus $\sqrt[4]{L}$ satisfies the quadratic equation $y^2 - 7y + 1 = 0 $.
  • Since $ \sqrt{ 7 + 2 } = 3$, thus $\sqrt[8]{L}$ satisfies the quadratic equation $y^2 - 3y + 1 = 0$.
  • Hence, $ \sqrt[8]{L} = \frac{ 3 + \sqrt{5} } { 2 } $ (reject negative root, why?).
Calvin Lin
  • 68,864
6

Here is my take on the question:

$$x=\sqrt[8]{2207-\cfrac{1}{2207-\cfrac{1}{2207-\cfrac{1}{2207-\cfrac{1}{\ddots}}}}}$$

By raising everything to the power of $8$, we get:

$$x^8=2207-\cfrac{1}{2207-\cfrac{1}{2207-\cfrac{1}{2207-\cfrac{1}{\ddots}}}}$$

Manipulating the above expression, we get:

$$2207-x^8=\cfrac{1}{2207-\cfrac{1}{2207-\cfrac{1}{2207-\cfrac{1}{\ddots}}}}$$

Raising everything to the power of $-1$, we get:

$$\frac{1}{2207-x^8}=2207-\cfrac{1}{2207-\cfrac{1}{2207-\cfrac{1}{\ddots}}}$$

However, since the RHS of this expression is the same as the RHS of the expression after raising everything to the power of $8$:

$$\frac{1}{2207-x^8}=x^8$$

Manipulating the above, we are left with the following:

$$x^{16}-2207x^8-1=0$$

Letting $y=x^8$, we have: $$y^2-2207y+1=0$$

Using the quadratic formula to solve the above quadratic, we have:

$$y=\frac{2207 \pm \sqrt{2207^2-2^2}}{2}$$

Using the difference of two squares identity and prime factorisation:

$$2207^2-2^2=2205*2209=21^2*47^2*5$$

Therefore: $$y=\frac{2207 \pm 987\sqrt{5}}{2}$$

Calculating both potential values of $y$, we find that: $$\fbox{$y_1, y_2 > 0$}$$

But $y=x^8$.

Therefore: $$x=\sqrt[8]{\frac{2207 \pm 987\sqrt{5}}{2}}$$

However, the fact that if the first number of the continued fraction is $0$, then the number it represents is less than $1$ (and given that the first number of our continued fraction is $2207$), we find that:

$$\fbox{$x> 1$}$$

Calculating both potential values of $x$ and discarding the value that is below $1$, we find that:

$$x=\sqrt[8]{\frac{2207+987\sqrt{5}}{2}}$$

Now, we need to get $x$ to be in the form of $\frac{a+b\sqrt{c}}{d}$.

Therefore:

$$(a+b\sqrt{c})^8 = \frac{d^8}{2}(2207 + 987\sqrt{5})$$

Expanding the LHS and splitting the equation, we get the following:

$$a^8+28a^6b^2c+70a^4b^4c^2+28a^2b^6c^3+b^8c^4=\frac{2207d^8}{2}$$

$$(8a^7b+56a^5b^3c+56a^3b^5c^2+8ab^7c^3)\sqrt{c}=\frac{987d^8}{2}\sqrt{5}$$

From the second equation, we can deduce that:

$$\fbox{$c=5$}$$

Therefore, the equations now become:

$$a^8+140a^6b^2+1750a^4b^4+3500a^2b^6+625b^8=\frac{2207d^8}{2}$$

$$8a^7b+280a^5b^3+1400a^3b^5+1000ab^7=\frac{987d^8}{2}$$

This is where the solution ends.

Post-Solution Thoughts:

Since WolframAlpha says that:

$$\fbox{$\sqrt[8]{\frac{2207 + 987\sqrt{5}}{2}} \equiv \frac{3+\sqrt{5}}{2}$}$$

and given my approach to finding $x$, I was looking into different ways to find $a,b,d$. My attempts so far have included the following:

$$(a+b\sqrt{5})^8-(a-b\sqrt{5})^8=16ab\sqrt{5}(a^2+5b^2)(a^4+30a^2b^2+25b^4)=2207d^8$$

(this was found using the second-to-last set of equations and adding and subtracting to get $(a+b\sqrt{5})^8$ and $(a-b\sqrt{5})^8$ respectively then subtracting them and simplifying the expression on the LHS.)

I am unsure how the above helps at all. I have also tried adding the last two equations and seeing if the LHS of the resulting equation can simplify into the form of $(a+kb)^n$ but it doesn't look like that will be the case. The same applies for subtraction.

Given that I am struggling to see any methods to find $a,b,d$, I turn to the viewers of this solution to help me find any viable approaches that follow on from my solution to finding the rest of the values of $x$ .

Update to Post-Solution Thoughts: I have been recently thinking about $\frac{3+\sqrt{5}}{2}$ and noticed that $$\fbox{$d=a-b$}$$

However, I would need to think of some proof of it going from $d=a-b$ to an equation or result I have previously obtained. At the time of posting this update, it looked like a promising avenue of attack and would've certainly helped but after investigating it, it led me nowhere. Plus, the posting of Update 2 (based off a commenter's method that was looking a lot more promising (at the time of posting the first iteration of Update 2)) effectively ended the usability of this method.

Update 2 to Post-Solution Thoughts: Having received a comment talking about how $k^{\frac{1}{8}}$ is the same as doing $\sqrt{k}$ $3$ times, I decided to try it out.

First, we will do the following:

$$\sqrt{\frac{2207+987\sqrt{5}}{2}}=\frac{a_1+b_1\sqrt{c_1}}{d_1} \tag1$$ ($a_m, b_m, c_m, d_m$ are all constants to be figured out.)

Rearranging the above, we get

$$\frac{d_1^2}{2}(2207+987\sqrt{5})=(a_1+b_1\sqrt{c_1})^2 \tag2$$

Expanding and simplifying the RHS before breaking the terms into a set of equations, we get

$$a_1^2+c_1b_1^2=\frac{2207d_1^2}{2} \tag3$$

$$2a_1b_1\sqrt{c_1}=\frac{987d_1^2\sqrt{5}}{2} \tag4$$

$(4)$ implies that $$\fbox{$c_1=5$}$$

This therefore means that $(3)$, $(4)$ becomes

$$a_1^2+5b_1^2=\frac{2207d_1^2}{2} \tag5$$

$$2a_1b_1=\frac{987d_1^2}{2} \tag6$$

Solving for $d_1^2$ then merging the $2$ equations, we get

$$a_1b_1=\frac{987(2a_1^2+10b_1^2)}{4(2207)} \tag7$$

Rearranging and simplifying, we get

$$0=987a_1^2-4414a_1b_1+4935b_1^2 \tag8$$

Letting the roots of the equation be $k_1,k_2$, we find (using Viete's formulae for the roots of a polynomial) that

$$k_1+k_2=\frac{4414}{987} \tag9$$

$$k_1k_2=5 \tag{10}$$

Rearranging the second equation and substituting into the first equation gives us

$$\frac{5}{k_f}+k_f=\frac{4414}{987} \tag{11}$$

(where $f$ can be $1$ or $2$)

Manipulating, rearranging and solving $(11)$ gives us the following pairs of solutions to $(8)$:

$(k_1,k_2)$ $=$ $(\frac{105}{47}$ , $\frac{47}{21})$ , $(\frac{47}{21}$ , $\frac{105}{47})$

A comment explained that my roots are equivalent to $\frac{a_1}{b_1}$.

Therefore, the above roots now become (after rearranging)

$$\fbox{$b_1=\frac{21a_1}{47}$, $b_1=\frac{47a_1}{105}$}$$

Now, substituting these roots back into $(6)$ and solving for $d_1$ , we get the following pairs of solutions to $(5)$ , $(6)$

$$\fbox{$b_1=\frac{21a_1}{47}$, $d_1=\pm\frac{2a_1}{47}$}$$

$$\fbox{$b_1=\frac{47a_1}{105}$, $d_1=\pm\frac{2a_1}{21\sqrt{5}}$}$$

Whilst investigating ways to find $a_1, b_1, d_1$, it occurred to me that if we proved that

$$\sqrt{\frac{e_1 + f_1\sqrt{g_1}}{h_1}} \equiv \frac{i_1+j_1\sqrt{l_1}}{m_1} \tag{12}$$

and that $e_1, f_1, g_1, h_1, i_1, j_1, l_1, m_1$ are all positive integers, then the first root pair would be the only valid root pair since in the second root pair, the denominator of $d_1$ is irrational. This would also mean that $47 \vert a_1$ and $2\vert d_1$.

Proof of $(12)$:

$$\sqrt{\frac{e_1 + f_1\sqrt{g_1}}{h_1}} = \frac{i_1+j_1\sqrt{l_1}}{m_1}$$

Rearranging, expanding and simplifying $(12)$, we get:

$$\frac{m_1^2}{h_1}(e_1 + f_1\sqrt{g_1})=i_1^2+2i_1j_1\sqrt{l_1}+g_1j_1^2$$

Splitting the equation into a pair of simultaneous equations:

$$\frac{m_1^2}{h_1}e_1=i_1^2+g_1j_1^2$$

$$\frac{m_1^2}{h_1}f_1\sqrt{g_1}=2i_1j_1\sqrt{l_1}$$

Splitting the second equation into another pair of simultaneous equations:

$$\frac{m_1^2}{h_1}f_1=2i_1j_1$$

$$\sqrt{g_1}=\sqrt{l_1}$$

Solving the second equation, we get:

$$\fbox{$g_1=l_1$}$$

Therefore, the original pair of simultaneous equations becomes:

$$\frac{m_1^2}{h_1}e_1=i_1^2+g_1j_1^2$$

$$\frac{m_1^2}{h_1}f_1=2i_1j_1$$

Manipulating the above equations such that the RHS of each equation is equal to $\frac{m_1^2}{h_1}$ , we find that

$$\frac{2i_1j_1}{f_1}=\frac{i_1^2+g_1j_1^2}{e_1}$$

Rearranging the above, we find that it is equivalent to

$$0=f_1i_1^2-2e_1i_1j_1+f_1g_1j_1^2$$

Letting the solutions of the above polynomial be $A_1, A_2$ , and using Viete's roots of polynomials formulae for a quadratic, we find that

$$A_1 + A_2=\frac{2e_1}{f_1}$$ $$A_1A_2=g_1$$

Manipulating, rearranging and solving the first equation (using the second equation as a substitution), we get

$$A_C = \frac{e_1}{f_1} \pm \sqrt{\frac{e_1^2}{f_1^2}-g_1}$$

(where $C$ can equal $1$ or $2$ .)

Therefore, the solution to the set of Viete's polynomial root equations (and hence, the original polynomial) are

$(A_1, A_2) = (\frac{e_1}{f_1} + \sqrt{\frac{e_1^2}{f_1^2}-g_1}, \frac{e_1}{f_1} - \sqrt{\frac{e_1^2}{f_1^2}-g_1}), (\frac{e_1}{f_1} - \sqrt{\frac{e_1^2}{f_1^2}-g_1}, \frac{e_1}{f_1} + \sqrt{\frac{e_1^2}{f_1^2}-g_1})$

The roots given above are of the form $\frac{i_1}{j_1}$ .

Rearranging the roots, we get

$$\fbox{$i_1 = j_1^2(\frac{e_1}{f_1} + \sqrt{\frac{e_1^2}{f_1^2}-g_1}), j_1^2(\frac{e_1}{f_1} - \sqrt{\frac{e_1^2}{f_1^2}-g_1})$}$$

Putting both solutions into the original set of simultaneous equations derived from $(12)$ , we get the following pair of solutions

$$\fbox{$i_1 = j_1^2(\frac{e_1}{f_1} + \sqrt{\frac{e_1^2}{f_1^2}-g_1}), j_1 = \pm \frac{m\sqrt[4]{(e_1f_1)^2+ g_1)}}{\sqrt{2h(e_1+f_1)\sqrt{\frac{e_1^4}{f_1^4}-g_1^2}}}$}$$

$$\fbox{$i_1 = j_1^2(\frac{e_1}{f_1} - \sqrt{\frac{e_1^2}{f_1^2}-g_1}), j_1 = \pm \frac{m\sqrt[4]{(e_1f_1)^2+ g_1)}}{\sqrt{2h(e_1-f_1)\sqrt{\frac{e_1^4}{f_1^4}-g_1^2}}}$}$$

(Rest of proof TBD at a later date)

Also, noticed the following: if you square both values of $d_1$ from both root pairs and compare the denominator, you get

$$d_1^2=\pm\frac{4a_1^2}{2207\pm2} \tag{13}$$

This means that $(13)$ can be used to rewrite the pairs of roots as the following

$$d_1^2=\pm \begin{cases} \frac{4a_1^2}{2207+2}, & \text{if $b_1=\frac{21a_1}{47}$} \\ \frac{4a_1^2}{2207-2}, & \text{if $b_1=\frac{47a_1}{105}$} \end{cases}$$

$(13)$ also shows that the denominator is of the form $a\pm2$, which indicates some connection to part of the original question the OP was asking - will need to investigate this further.

Edit 1: Despite this question now being labeled as a duplicate, I will still continue to edit and refine this solution until it is complete (i.e. I have solved the equation and have answered the OP's original question.).

Any suggestions for/edits to this solution are welcome.

  • If you square $\ \frac{3+\sqrt{5}}{2}\ $ three times (which I confess I haven't actually done), I presume you'll end up with $\ \frac{2207+987\sqrt{5}}{2}\ .$ – lonza leggiera Jan 01 '24 at 13:40
  • I have now, and indeed I did end up with the expected result. – lonza leggiera Jan 01 '24 at 13:59
  • The best way to find the $8$th root is simply to attempt to square-root it $4$ times: the equations you get when attempting to find the square root of such an expression are far simpler. – Prasiortle Jan 04 '24 at 14:50
  • 1
    @Prasiortle, thanks for suggesting another approach for me to try out! Unfortunately, I will probably start tomorrow. – Yajat Shamji Jan 04 '24 at 14:52
  • @Prasiortle, see Update 2 of my Post-Solution Thoughts. Also, isn't $x^{\frac{1}{8}} \equiv ((x^{\frac{1}{2}})^{\frac{1}{2}})^{\frac{1}{2}}$? – Yajat Shamji Jan 06 '24 at 15:15
  • @YajatShamji On the latter point, yes, you're exactly right, my brainfart! Your working so far is correct: note that by finding the roots of the polynomial, you've effectively calculated $\frac{a_1}{b_1}$, since the equation $987a_1^2-4414a_1b_1+4935b_1^2 = 0$ becomes $987x^2-4414x+4935 = 0$ by substituting $x = \frac{a_1}{b_1}$. Mathematica, by contrast, is giving the solutions as expressions for $b_1$ in terms of $a_1$, but you can check that they're exactly equivalent. – Prasiortle Jan 06 '24 at 17:50
  • 1
    Thanks for the help, @Prasiortle! – Yajat Shamji Jan 07 '24 at 13:28
  • You're very welcome! – Prasiortle Jan 07 '24 at 13:28
-1

Since $x^8=2207-\frac1{x^8}$, $$x^{16}-2207x^8+1=0$$ $$(x^2-3x+1)(x^2+3x+1)(x^4+7x^2+1)(x^8+47x^4+2)=0.$$ The real roots are coming from the first factor.

$$x^2-3x+1=0\implies x=\frac{3+\sqrt5}2$$

since the other root is less than $1$.

Bob Dobbs
  • 10,988