2
$\begingroup$

Let $m,n,x,y,z$ be positive integers such that $y=m(z-1)+1$ and $yz-1=n(y^2-x^2)$.

If you fix $m$ and $n$ then this is a generalized Pell equation in $x$ and $z$ (i.e., a quadratic Diophantine equation of the sort that can be solved by this website).


I have 2 specific conjectures, which I have checked computationally for $m,n\leq100$, $z\leq10^7$.

Conjecture 1: If $n\geq2$ then $x,y,z$ are all odd.

Conjecture 2: If $(m,n)\neq(1,1)$ then $z$ is indivisible by $4$.

A proof or counterexample of either of these conjectures would be much appreciated.


The table below gives the first four $(x,y,z)$ solutions for fixed $(m,n)$:

$$\begin{array}{c|c|c|c}&n=1&n=2&n=3\\\hline m=1& \begin{array}{c}(1,1,1)\\(1,2,2)\\(1,3,3)\\(1,4,4)\end{array}& \begin{array}{c}(1,1,1)\\(5,7,7)\\(29,41,41)\\(169,239,239)\end{array}& \begin{array}{c}(1,1,1)\\(9,11,11)\\(89,109,109)\\(881,1079,1079)\end{array}\\\hline m=2& \begin{array}{c}(1,1,1)\\(2,3,2)\\(23,33,17)\\(64,91,46)\end{array}& \begin{array}{c}(1,1,1)\\(167,193,97)\\(32397,37409,18705)\\(6284851,7257121,3628561)\end{array}& \begin{array}{c}(1,1,1)\\(439,481,241)\\(211597,231793,115897)\\(101989315,111723697,55861849)\end{array}\\\hline m=3& \begin{array}{c}(1,1,1)\\(3,4,2)\\(69,85,29)\\(287,352,118)\end{array}& \begin{array}{c}(1,1,1)\\(417,457,153)\\(200993,220177,73393)\\(96878209,106124761,35374921)\end{array}& \begin{array}{c}(1,1,1)\\(1053,1117,373)\\(1215161,1288873,429625)\\(1402294741,1487358181,495786061)\end{array} \end{array}$$

In the comments, Will Jagy observed that if $n\geq2$ then the $x$ values satisfy a linear recurrence $x_{n+2}=Cx_{n+1}-x_n$ (and you can do something similar for $y$ and $z$). The table below gives the values of $C$:

$$\begin{array}{c|c|c|c|c|c|c|c|c|c}&n=2&n=3&n=4&n=5&n=6&n=7&n=8&n=9&n=10\\\hline m=1&6&10&14&18&22&26&30&34&38\\\hline m=2&194&482&898&1442&2114&2914&3842&4898&6082\\\hline m=3&482&1154&2114&3362&4898&6722&8834&11234&13922\\\hline m=4&898&2114&3842&6082&8834&12098&15874&20162&24962\\\hline m=5&1442&3362&6082&9602&13922&19042&24962&31682&39202\\\hline m=6&2114&4898&8834&13922&20162&27554&36098&45794&\\\hline m=7&2914&6722&12098&19042&27554&37634&49282&&\\\hline m=8&3842&8834&15874&24962&36098&49282&&&\\\hline m=9&4898&11234&20162&31682&45794&&&&\\\hline m=10&6082&13922&24962&39202&&&&&\end{array}$$

Notice the symmetry!

Gottfried Helms observed the formula $C=4(2mn-1)^2-2$ for $m,n\geq2$.

When $m=1$, it looks like $C=4n-2$ for $n\geq2$.


This answer proves the $m=1$ case of conjecture 2, using methods from Diophantine approximation.

$\endgroup$
15
  • $\begingroup$ That's an interesting idea. I tinkered around with the equations a bit earlier today, and was able to write it as $ax^2+cz^2=d$, but the process of getting rid of the linear $z$ term seemed to make the constant term rather large. $\endgroup$ – Thomas Browning Dec 11 '20 at 3:39
  • $\begingroup$ Here it is, in all it's glory: $4mn(mn-1)x^2-(2m(mn-1)z-(m-1)(2mn-1))^2=4m(mn-1)-(m-1)^2$. Maybe it can be transformed into a nicer looking expression. $\endgroup$ – Thomas Browning Dec 11 '20 at 5:04
  • 1
    $\begingroup$ Not quite: it applies to the sequence $5, 1157, 224453,...$ from which we add $1$ then divide by $6$ to get $y.$ so $1158/6 = 193, $ $ 224454/6 = 37409$ The presence of modulus conditions on $y$ says that the recurrence needs a constant term to work properly. Yes, for that one $$y_{j+2} = 194 y_{j+1} - y_j - 32 $$ $\endgroup$ – Will Jagy Dec 13 '20 at 23:09
  • 1
    $\begingroup$ The conjecture 1 is easy to prove, given that a) the recursivity and b) the table of $C$ is a correct fact. Because, if $C_{m,n}=2c$ is even, then of course, if only $x_1=1$, the initialization of the $x_3=2cx_2-x_1$ gives an odd value, and this follows immediately for all $x_{k+2}=2cx_{k+1}-x_k$ with $x_k$ odd. Since your table shows the initial values for all $(x,y,z)=(1,1,1)$ when $n>1$ and all $C_{m,n}$ are even, it must follow that all $(x,y,z)_k$ are odd for that $n$. $\endgroup$ – Gottfried Helms Dec 16 '20 at 19:22
  • 1
    $\begingroup$ @WillJagy - what happened? Why deleted all your answers? Please leave a small note to let us understand... $\endgroup$ – Gottfried Helms Dec 18 '20 at 9:54
1
+50
$\begingroup$

This is not an answer, just a comment to supply more data.

I used Pari/GP to detect the polynomial-structure in your C-table, and find the following table containing additional info for the column $n=1$:

  m \ n|    1     2     3      4      5      6      7      8  |
    -  +    -     -     -      -      -      -      -      -  +
    1  |    2    34    98    194    322    482    674    898  |   // ????? see note
    2  |   34   194   482    898   1442   2114   2914   3842  |
    3  |   98   482  1154   2114   3362   4898   6722   8834  |
    4  |  194   898  2114   3842   6082   8834  12098  15874  |
    5  |  322  1442  3362   6082   9602  13922  19042  24962  |
    6  |  482  2114  4898   8834  13922  20162  27554  36098  |
    7  |  674  2914  6722  12098  19042  27554  37634  49282  |
    8  |  898  3842  8834  15874  24962  36098  49282  64514  |
    -  +  ...   ...   ...    ...    ...    ...    ...    ...  +
          ???

This table is produced by the small Pari/GP function

Cpol(m,n='x)= (4*m*n)^2-16*m*n+2 \\ reproduces OP's table for m,n>=2
or, a bit more symmetric,
Cpol(m,n='x)= 4*(2*m*n-1)^2 - 2 \\ reproduces OP's table for m,n>=2
$\qquad \qquad$ (leaving n in the function-call indeterminate gives the $m$'th polynomial in $x$)

Surprisingly, the first row, (for $m=1$) does not agree with your first row; and the same might occur, if you would supply a first column with $n=1$ of your data (however looking numerically for solutions when $n=1$ agrees with the first column here).
I'm checking further to see what's going on with this.


update1
I have looked at the solutions for $n=1$ and $m \gt 1$ .

First, with a chosen $m$, I take, from sequential search, the first 6 solutions for $(x,y,z)$. We find, that assuming we have 2 interlaced subsequences, that subsequences $x_a$ and $x_b$ have a simple recursion rule; for $x_a$ and $x_b$ we have $$x_{a,k+2}=C_{m,1}\cdot x_{a,k+1} - 1 \cdot x_{a,k} \\ x_{b,k+2}=C_{m,1}\cdot x_{b,k+1} - 1 \cdot x_{b,k} \\\tag {1.x}$$

The empirical data agree with the $C_{m,n}$-values from my proposal.

For the interlaced $y$- and $z$- sequences we have to include a constant value $D_{y,m}$ and $D_{z,m}$ in the recursion formula: $$ y_{a,k+2}=C_{m,1}\cdot y_{a,k+1} - 1 \cdot y_{a,k} + D_{y,m}\\ y_{b,k+2}=C_{m,1}\cdot y_{b,k+1} - 1 \cdot y_{b,k} + D_{y,m}\\ \tag {1.y}$$ and $$ z_{a,k+2}=C_{m,1}\cdot z_{a,k+1} - 1 \cdot z_{a,k} + D_{z,m}\\ z_{b,k+2}=C_{m,1}\cdot z_{b,k+1} - 1 \cdot z_{b,k} + D_{z,m}\\ \tag {1.z}$$

The complete recursion can be done in a matrix-formula, where the recursion-depth can be formalized by an initial matrix $I$ and the multiplicatormatrix $M$ such that $$ S_{h+1} = I \cdot M^h $$ gives the $h$'th iteration.

Solutions:

  • Let $m=2$. The first few empirically found solutions for $(x,y,z)$ are empirically the following sequences, obviously by 2 recursive intertwined subsequences:

    x=  1  2  23  64   781  2174  26531   73852
    y=  1  3  33  91  1105  3075  37521  104443
    z=  1  2  17  46   553  1538  18761   52222
    

The matrix $M$ is $$ M = \small{ \begin{bmatrix} 1&0&1 \\ 0&0&-1\\0&1&C_{m,1}\end{bmatrix} = \begin{bmatrix} 1&0&1 \\ 0&0&-1\\0&1&34\end{bmatrix} }$$

The initmatrix $I$ is composed of the first two empirical values $[x_a,y_a,z_a]$ and $[x_b,y_b,z_b]$, and additionally prepended by the -at the moment- unknown constants $D$ taken as zero: $$ I_{\text{init}}=\small \begin{bmatrix} 0&x_{a,1}&x_{a,2} \\ D_y&y_{a,1}&y_{a,2}\\D_z&z_{a,1}&z_{a,2} \\ \hline 0&x_{b,1}&x_{b,2} \\ D_y&y_{b,1}&y_{b,2}\\D_z&z_{b,1}&z_{b,2} \end{bmatrix} = \begin{bmatrix} 0&1&23\\0&1&33\\0&1&17 \\ \hline 0&2&64\\ 0&3&91\\0&2&46 \end{bmatrix} $$ If we compute $S_2 = I \cdot M$ we get a wrong result for $y_{a,3},z_{a,3},y_{b,3},z_{a,3}$ - but this differences are then simply the new constant values for $D$ and this gives then $$ \small I= \begin{bmatrix} 0&1&23\\-16&1&33\\-24&1&17 \\ \hline 0&2&64\\ -16&3&91\\-24&2&46 \end{bmatrix}$$

With this we can compute the set of solutions using $S_{h+1}=I \cdot M^h$ for $h=0..7$ (or arbitrarily more); I show this having the two subsets as submatrices: $$\small \begin{bmatrix} o: & 0 & 1 & 2 & 3 & 4 & 5 & 6 & 7 & \cdots \\ \hline x_a: & 1 & 23 & 781 & 26531 & 901273 & 30616751 & 1040068261 & 35331704123 & \cdots \\ y_a: & 1 & 33 & 1105 & 37521 & 1274593 & 43298625 & 1470878641 & 49966575153 & \cdots \\ z_a: & 1 & 17 & 553 & 18761 & 637297 & 21649313 & 735439321 & 24983287577 & \cdots \\ x_b: & 2 & 64 & 2174 & 73852 & 2508794 & 85225144 & 2895146102 & 98349742324 & \cdots \\ y_b: & 3 & 91 & 3075 & 104443 & 3547971 & 120526555 & 4094354883 & 139087539451 & \cdots \\ z_b: & 2 & 46 & 1538 & 52222 & 1773986 & 60263278 & 2047177442 & 69543769726 & \cdots \end{bmatrix}$$

Note that we could even insert negative values for $h$ to continue the matrix of solutions to the left side.


  • Let $m=3$. The first few empirically found solutions for $(x,y,z)$ are empirically the following sequences, obviously again by 2 recursive intertwined subsequences:

    x=    1  3  69  287  6761  28123
    y=    1  4  85  352  8281  34444
    z=    1  2  29  118  2761  11482
    

The matrix $M$ is $$ M = \begin{bmatrix} 1&0&1 \\ 0&0&-1\\0&1&98\end{bmatrix} $$

The initmatrix $I$ is made as before and this gives then $$ I = \small \begin{bmatrix} 0 & 1 & 69 \\ -48 & 1 & 85 \\ -80 & 1 & 29 \\ \hline 0 & 3 & 287 \\ -48 & 4 & 352 \\ -80 & 2 & 118 \end{bmatrix}$$

With this we can compute the set of solutions using $S_{h+1}=I \cdot M^h$ for $h=0..7$ (or arbitrarily more); again I show this having the two subsets as submatrices: $$\small \begin{bmatrix} o: & 0 & 1 & 2 & 3 & 4 & 5 & 6 & 7 & \cdots \\ \hline x_a: & 1 & 69 & 6761 & 662509 & 64919121 & 6361411349 & 623353393081 & 61082271110589 \\ y_a: & 1 & 85 & 8281 & 811405 & 79509361 & 7791105925 & 763448871241 & 74810198275645 \\ z_a: & 1 & 29 & 2761 & 270469 & 26503121 & 2597035309 & 254482957081 & 24936732758549 \\ \hline x_b: & 3 & 287 & 28123 & 2755767 & 270037043 & 26460874447 & 2592895658763 & 254077313684327 \\ y_b: & 4 & 352 & 34444 & 3375112 & 330726484 & 32407820272 & 3175635660124 & 311179886871832 \\ z_b: &2 & 118 & 11482 & 1125038 & 110242162 & 10802606758 & 1058545220042 & 103726628957278 \end{bmatrix} $$

$\endgroup$
6
  • 1
    $\begingroup$ @Will - yeah, this might be a point. What I got with $n=1$ is $(x,y,z)=(1,1,1),(2,3,2)(23,33,17)(64,91,46)...$ whose two subsets of each second parenthese give two sequences with expected $C$ for recursion. When I had this I didn't notice the distinction into even and odd solutions. $\endgroup$ – Gottfried Helms Dec 14 '20 at 14:36
  • $\begingroup$ Gottfried, $m=1$ is strange. If you just ask, "what value of $C$ makes the recurrence hold", then you get my values $C=4n-2$ for $n\geq2$, which doesn't match the formula for $m,n\geq2$. I'm not sure why $m=1$ would be special in this way. $\endgroup$ – Thomas Browning Dec 14 '20 at 19:03
  • 1
    $\begingroup$ @Will, for whatever reason this came to my mind: "Kids are different today, I hear every mother say, mother needs something today to calm her down..." ;-) $\endgroup$ – Gottfried Helms Dec 18 '20 at 15:54
  • 1
    $\begingroup$ @Will - my most used resource is the Göttingen Digitalizing Center (at University Göttingen). They have real good Journals; the fact that they look a bit "prehistoric" is not so a problem for me, because that's the level where I can mostly follow. But to be not unfair: many modern english articles, online, well explanative I have (and appreciate much) in my "collection". My own "articles" are not much more than explorative, compared to that... $\endgroup$ – Gottfried Helms Dec 18 '20 at 16:40
  • 1
    $\begingroup$ @WillJagy I apologize if my comment came off as dismissive or rude. I have been reading through your answers carefully and was just hoping that you might include a sentence or two at the end explaining how the classification that you obtained implies the parity conjecture. I also had one or two more specific questions, like how you got the congruence $w\equiv R-m\pmod{m(R-1)}$. You have put a lot of time into your answers, and I have learned a lot from them, and I want to reward you for them. $\endgroup$ – Thomas Browning Dec 18 '20 at 20:29
1
$\begingroup$

Phew! Thanks to the help of Will Jagy and Gottfried Helms, I have a complete solution!

Fix positive integers $m$ and $n$. We will study the equation \begin{equation} \tag{1} mn(y-x)(y+x)=(y-1)(y+m). \end{equation} Classification of Small Solutions

We begin by classifying the integer solutions to (1) with small $y$ coordinate.

Case 1: $y>1$

Assume that $y>1$. Then the right hand side of (1) is positive, so the left hand side of (1) is also positive. Then $y^2-x^2$ is positive, so $y^2-x^2\geq2y-1$. This gives $$(y-1)(y+m)=mn(y^2-x^2)\geq mn(2y-1).$$ Dividing through by $2y-1$ gives $$\frac{(y-1)(y+m)}{2y-1}\geq mn.$$ Let $f(y)=(y-1)(y+m)/(2y-1)$. Note that $f$ is increasing on $(\frac{1}{2},\infty)$. Let $y_0=2mn-m$. Then $$f(y_0)=\frac{2mn(y_0-1)}{2y_0-1}<mn\leq f(y)$$ so $y_0<y$. Thus, $y\geq2mn-m+1$.

Case 2: $-m<y<1$

Assume that $-m<y<1$. Then the right hand side of (1) is negative, so the left hand side of (1) is also negative. Then $y^2-x^2$ is negative, so $y^2-x^2\leq2y-1$. This gives $$(y-1)(y+m)=mn(y^2-x^2)\leq mn(2y-1).$$ Dividing through by $2y-1$ gives $$\frac{(y-1)(y+m)}{2y-1}\geq mn.$$ Let $f(y)=(y-1)(y+m)/(2y-1)$. Note that $f$ is increasing on $(-\infty,\frac{1}{2})$. Let $y_0=0$. Then $$f(y_0)=m\leq mn\leq f(y)$$ so $y_0\leq y$. Thus, $y=0$.

Case 3: $y<-m$

Assume that $y<-m$. Then the right hand side of (1) is positive, so the left hand side of (1) is also positive. Then $y^2-x^2$ is positive, so $y^2-x^2\geq-2y-1$. This gives $$(y-1)(y+m)=mn(y^2-x^2)\geq mn(-2y-1).$$ Dividing through by $-2y-1$ gives $$\frac{(y-1)(y+m)}{-2y-1}\geq mn.$$ Let $f(y)=(y-1)(y+m)/(-2y-1)$. Note that $f$ is decreasing on $(-\infty,-\frac{1}{2})$. Let $y_0=-2mn-m+2$. Then $$f(y_0)=\frac{(y_0-1)(y_0+m)}{-2y_0-1}\leq\frac{(y_0+1)(y_0+m-2)}{-2y_0-1}=\frac{-2mn(y_0+1)}{-2y_0-1}<mn\leq f(y)$$ so $y_0>y$. Thus, $y\leq-2mn-m+1$.

Completing the Classification

Theorem 1: If $(x,y)\in\mathbb{Z}^2$ satisfies (1) then exactly one of the following holds:

  • $y\geq2mn-m+2$,
  • $(x,y)=(\pm m,m+1)$ and $n=1$,
  • $(x,y)=(\pm1,1)$,
  • $(x,y)=(\pm1,0)$ and $n=1$,
  • $(x,y)=(\pm m,-m)$,
  • $(x,y)=(\pm(3m-2),-3m+1)$ and $n=1$,
  • $y\leq-2mn-m$.

Proof: We have already shown that either $y\geq2mn-m+1$ or $y=1$ or $y=0$ or $y=-m$ or $y\leq-2mn-m+1$.

  • Assume that $y=2mn-m+1$. Then (1) becomes $$mn(y-x)(y+x)=(2mn-m)(2mn+1).$$ Dividing through by $m$ and reducing modulo $n$ forces $n=1$. Then $y=m+1$ and solving for $x$ gives $x=\pm m$.
  • If $y=1$ then solving for $x$ gives $x=\pm1$.
  • If $y=0$ then solving for $x$ gives $nx^2=1$. This forces $n=1$ and $x=\pm1$.
  • If $y=-m$ then solving for $x$ gives $x=\pm m$.
  • Assume that $y=-2mn-m+1$. Then (1) becomes $$mn(y-x)(y+x)=(2mn+m)(2mn-1).$$ Dividing through by $m$ and reducing modulo $n$ forces $n=1$. Then $y=-3m+1$ and solving for $x$ gives $x=\pm(3m-2)$. $\blacksquare$

Mapping Solutions

We now define functions $f,g\colon\mathbb{R}^2\to\mathbb{R}^2$ that preserve integer solutions to (1). Our goal is to show that every integer solution to (1) can be obtained via $f$ and $g$ from one of the five solutions listed in Theorem 1.

Let $R=2mn-1$. Let $f,g\colon\mathbb{R}^2\to\mathbb{R}^2$ be defined by $$f\left(\begin{bmatrix}x\\y\end{bmatrix}\right)=\begin{bmatrix}(2R^2-1)x+2R(R-1)y-2(m-1)R\\(2R^2-1)y+2R(R+1)x-2(m-1)(R+1)\end{bmatrix},$$ $$g\left(\begin{bmatrix}x\\y\end{bmatrix}\right)=\begin{bmatrix}(2R^2-1)x-2R(R-1)y+2(m-1)R\\(2R^2-1)y-2R(R+1)x-2(m-1)(R+1)\end{bmatrix}.$$

Lemma 2: The functions $f$ and $g$ preserve solutions to (1).

Proof: The algebraic manipulations are rather involved, so I instead wrote up a proof in Lean 3.

import tactic.ring

variables {A : Type*} [comm_ring A]

lemma lem0 {a b c d : A} (h : a = b) : c - a = d - b → c = d :=
by { rw h, exact sub_left_inj.mp }

structure sol (m n : A) :=
(x y : A)
(h : m * n * (y - x) * (y + x) = (y - 1) * (y + m))

def R (m n : A) : A := 2 * m * n - 1

def sol.forward {m n : A} (s : sol m n) : sol m n :=
{ x := (2*(R m n)*(R m n)-1)*s.x+2*(R m n)*((R m n)-1)*s.y-2*(m-1)*(R m n),
  y := (2*(R m n)*(R m n)-1)*s.y+2*(R m n)*((R m n)+1)*s.x-2*(m-1)*((R m n)+1),
  h := by { dsimp only [R], apply lem0 s.h, ring } }

def sol.backward {m n : A} (s : sol m n) : sol m n :=
{ x := (2*(R m n)*(R m n)-1)*s.x-2*(R m n)*((R m n)-1)*s.y+2*(m-1)*(R m n),
  y := (2*(R m n)*(R m n)-1)*s.y-2*(R m n)*((R m n)+1)*s.x-2*(m-1)*((R m n)+1),
  h := by { dsimp only [R], apply lem0 s.h, ring } }

You just need to check that I correctly copied (1) (sol.h), $f$ (sol.forward), and $g$ (sol.backward). $\blacksquare$

We now give a few lemmas describing the relationship between $f$ and $g$.

Lemma 3: Let $\overline{(x,y)}=(-x,y)$. Then $f(\overline{p})=\overline{g(p)}$ for all points $p\in\mathbb{R}^2$.

Proof: This is clear from the definitions of $f$ and $g$. $\blacksquare$

Lemma 4: The functions $f$ and $g$ are inverses to each other.

Proof: We can write $f$ in terms of matrix multiplication as $$f\left(\begin{bmatrix}x\\y\end{bmatrix}\right)=\begin{bmatrix}2R^2-1&2R(R-1)\\2R(R+1)&2R^2-1\end{bmatrix}\begin{bmatrix}x\\y\end{bmatrix}-2(m-1)\begin{bmatrix}R\\R+1\end{bmatrix}.$$ Note that the $2\times2$ matrix has determinant $(2R^2-1)^2-4R^2(R^2-1)=1$. Then the inverse function to $f$ is given by \begin{align*} f^{-1}\left(\begin{bmatrix}x\\y\end{bmatrix}\right)&=\begin{bmatrix}2R^2-1&2R(R-1)\\2R(R+1)&2R^2-1\end{bmatrix}^{-1}\left(\begin{bmatrix}x\\y\end{bmatrix}+2(m-1)\begin{bmatrix}R\\R+1\end{bmatrix}\right)\\ &=\begin{bmatrix}2R^2-1&-2R(R-1)\\-2R(R+1)&2R^2-1\end{bmatrix}\left(\begin{bmatrix}x\\y\end{bmatrix}+2(m-1)\begin{bmatrix}R\\R+1\end{bmatrix}\right)\\ &=\begin{bmatrix}2R^2-1&-2R(R-1)\\-2R(R+1)&2R^2-1\end{bmatrix}\begin{bmatrix}x\\y\end{bmatrix}+2(m-1)\begin{bmatrix}2R^2-1&-2R(R-1)\\-2R(R+1)&2R^2-1\end{bmatrix}\begin{bmatrix}R\\R+1\end{bmatrix}\\ &=\begin{bmatrix}2R^2-1&-2R(R-1)\\-2R(R+1)&2R^2-1\end{bmatrix}\begin{bmatrix}x\\y\end{bmatrix}+2(m-1)\begin{bmatrix}R\\-(R+1)\end{bmatrix}=g\left(\begin{bmatrix}x\\y\end{bmatrix}\right). \end{align*} Thus, $f$ and $g$ are inverses to each other. $\blacksquare$

By Lemma 4, fixed points of $f$ are fixed points of $g$, and vice versa.

Lemma 5: If $(x,y)$ is a fixed point of $f$ and $g$ then $x=0$.

Proof: Suppose that $(x,y)$ is a fixed point of $f$ and $g$. Then $$\begin{bmatrix}2R^2-1&2R(R-1)\\2R(R+1)&2R^2-1\end{bmatrix}\begin{bmatrix}x\\y\end{bmatrix}-2(m-1)\begin{bmatrix}R\\R+1\end{bmatrix}=f\left(\begin{bmatrix}x\\y\end{bmatrix}\right)=\begin{bmatrix}x\\y\end{bmatrix}=\begin{bmatrix}1&0\\0&1\end{bmatrix}\begin{bmatrix}x\\y\end{bmatrix}.$$ Rearranging terms and dividing through by 2 gives \begin{equation}\tag{2}\begin{bmatrix}R^2-1&R(R-1)\\R(R+1)&R^2-1\end{bmatrix}\begin{bmatrix}x\\y\end{bmatrix}=(m-1)\begin{bmatrix}R\\R+1\end{bmatrix}.\end{equation} Note that the $2\times2$ matrix has determinant $(R^2-1)^2-R^2(R^2-1)=1-R^2$. If $R=1$ then $(m,n)=(1,1)$, so (2) becomes $$\begin{bmatrix}0&0\\2&0\end{bmatrix}\begin{bmatrix}x\\y\end{bmatrix}=\begin{bmatrix}0\\0\end{bmatrix}.$$ This forces $x=0$. Now suppose that $R>1$. Then the $2\times2$ matrix in (2) is invertible, so (2) has a unique solution. In other words, $f$ and $g$ have a unique fixed point $p$. However, Lemma 3 shows that $\overline{p}$ is also a fixed point of $f$ and $g$. Then $p=\overline{p}$, so $x=0$. Alternatively, you can solve (2) explicitly and get $(x,y)=(0,(m-1)/(R-1))$. $\blacksquare$

Lemma 6: The functions $f,g\colon\mathbb{R}^2\to\mathbb{R}^2$ have no fixed points satisfying equation (1).

Proof: Let $(x,y)$ be a fixed point of $f$ and $g$. By Lemma 5, $x=0$. However, setting $x=0$ in (1) gives $$mny^2=(y-1)(y+m).$$ The quadratic $(mn-1)y^2-(m-1)y+m$ has discriminant $(m-1)^2-4m(mn-1)<0$. $\blacksquare$

We can now prove the main theorem.

Theorem 7: Any integer solution to (1) can be mapped via $f$ and $g$ to one of the five solutions listed in Theorem 1.

Proof: Since the quadratic $(mn-1)y^2-(m-1)y+m$ has negative discriminant, the graph of (1) consists of the two curves $x=\pm\sqrt{y^2-((y-1)(y+m)/(mn))}$. Since $f(-1,1)=(1-2mR,1-2m(R+1))$ has negative $x$-coordinate, $f$ and $g$ do not swap the two curves. By Lemma 6, $f$ and $g$ do not flip either of the two curves, and each shift each curve either up or down. Since $f(-1,1)=(1-2mR,1-2m(R+1))$ has negative $y$-coordinate, $f$ shifts the left curve downward. Then by Lemma 4, $g$ shifts the left curve upward. Then by Lemma 3, $f$ shifts the right curve upward and $g$ shifts the right curve downward.

It remains to show that $f$ and $g$ can't jump over the gap $-2mn-m+1\leq y\leq2mn-m+1$. By Lemmas 3 and 4, it suffices to show that if $y_0=2mn-m+1$ and $x_0=\sqrt{y_0^2-((y_0-1)(y_0+m)/(mn))}$ then the $y$-coordinate of $g(x_0,y_0)$ is at least $-2mn-m+1$. In other words, we need to prove the inequality $$(2R^2-1)y_0-2R(R+1)x_0-2(m-1)(R+1)\geq-2mn-m+1.$$ Rearranging terms gives the equivalent inequality $$x_0\leq\frac{(2R^2-1)y_0-2(m-1)(R+1)+2mn+m-1}{2R(R+1)}.$$ Simplifying the right hand side gives the equivalent inequality $$x_0\leq2mn-m.$$ To prove this, we directly compute $$y_0^2-x_0^2=\frac{(y_0-1)(y_0+m)}{mn}=\frac{(2mn-m)(2mn+1)}{mn}=4mn-2m+2-\frac{1}{n}\geq4mn-2m+1=y_0^2-(y_0-1)^2.$$ Then $x_0\leq y_0-1$ as desired. $\blacksquare$

Consequences

We can combine Theorems 1 and 7 to make conclusions about all integer solutions to (1).

Theorem 8: Let $(x,y)\in\mathbb{Z}^2$ satisfy (1).

  • $y\equiv0,1\pmod{m}$.
  • If $n\neq1$ and $y\equiv0\pmod{m}$ then $x\equiv y\equiv m\pmod{2}$.
  • If $n\neq1$ and $y\equiv1\pmod{m}$ then $x\equiv y\equiv1\pmod{2}$.
  • Either way, if $n\neq1$ then $x\equiv y\pmod{2}$.

Proof: Theorem 7 states that any integer solution to (1) can be mapped via $f$ and $g$ to one of the five solutions listed in Theorem 1. Note that $f$ and $g$ preserve the value of $y\pmod{m}$, $x\pmod{2}$, and $y\pmod{2}$. Since each of the five solutions listed in Theorem 1 satisfy the theorem statement, all integer solutions to (1) satisfy the theorem statement. $\blacksquare$

Theorem 9: Let $m,n,x,y,z$ be positive integers such that $y=m(z-1)+1$ and $yz-1=n(y^2-x^2)$.

  • If $n\geq2$ then $x,y,z$ are all odd.
  • If $(m,n)\neq(1,1)$ then $z$ is indivisible by 4.

Proof: Note that $$mn(y^2-x^2)=m(yz-1)=y(mz)-m=y(y+m-1)-m=(y-1)(y+m)$$ so our previous work applies. Furthermore, we are in the $y\equiv1\pmod{m}$ case. By Theorem 8, if $n\geq2$ then $x\equiv y\equiv1\pmod{2}$. Then the equation $yz-1=n(y^2-x^2)$ forces $z\equiv1\pmod{2}$.

It remains to show that if $m\geq2$ and $n=1$ then $z$ is indivisible by 4. We know that $(x,y)$ can be mapped via $f$ and $g$ to one of the five solutions listed in Theorem 1. Since we are in the $y\equiv1\pmod{m}$ case, and since $m\geq2$, we know that $(x,y)$ can be mapped via $f$ and $g$ to $(\pm m,m+1)$ or $(\pm1,1)$ or $(\pm(3m-2),-3m+1)$. Since $(x,y)$ is in the first quadrant, we know that $(x,y)$ can be obtained from $(1,1)$ or $(m,m+1)$ via repeated application of $f$ (we can discard $(3m-2,-3m+1)$ since $g(m,m+1)=(3m-2,-3m+1)$). The solution $(1,1)$ has $z=1$. The solution $(m,m+1)$ has $z=2$.

We now check how applying $f$ changes $z$. Applying $f$ has the effect $$y\mapsto(2R^2-1)y+2R(R+1)x-2(m-1)(R+1).$$ In terms of $z$, this is $$z\mapsto\frac{(2R^2-1)(m(z-1)+1)+2R(R+1)x-2(m-1)(R+1)-1}{m}+1$$ which simplifies to $$z\mapsto(2R^2-1)z+4nRx-4(m-1)nR.$$ In particular, the value of $z\pmod{4}$ does not change. $\blacksquare$

$\endgroup$

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.