Rational Periodic Points of x^d+c and Fermat-Catalan Equations

Let x_i=X_i/Z be rational periodic points of exact period n of f_{d,c}(x) for i=1,\dots,n. Observe that

where x_{n+1}=x_1. Thus, \displaystyle \prod_{i=1}^{n}\dfrac{(x_i^{2k}-x_{i+1}^{2k})}{x_i-x_{i+1}}=1, or equivalently,

Assume that there is q\in \mathbb{N} such that q\mid \gcd(X_1,X_2).
Since q\mid \gcd(X_1,X_2) and

we have q\mid Z^{n(2k-1)}. However, X_1 and Z are coprime. Thus, q=1.
For period n=2, we have that

Since \gcd(X_1,X_2)=1 and

we have that \gcd(X_1^k+X_2^k,X_1^k-X_2^k)\mid 2.
We consider 2 cases.
Case 1. \gcd(X_1^k+X_2^k,\dfrac{X_1^k-X_2^k}{X_1-X_2})=1
If \gcd(X_1^k+X_2^k,\dfrac{X_1^k-X_2^k}{X_1-X_2})=1, then we easily have that X_1^k+X_2^k=Z_1^{2k-1} for some integer Z_1 such that Z_1\mid Z.
Case 2. \gcd(X_1^k+X_2^k,\dfrac{X_1^k-X_2^k}{X_1-X_2})=2
If \gcd(X_1^k+X_2^k,\dfrac{X_1^k-X_2^k}{X_1-X_2})=2, then we have that Z is even. Therefore, X_1 and X_2 are odd.
If 4\mid X_1^k+X_2^k, then 2\mathrel\Vert X_1^k-X_2^k. Thus, \dfrac{X_1^k-X_2^k}{X_1-X_2} is odd and

a contradiction. Therefore, 2 \mathrel\Vert X_1^k+X_2^k. Thus,

for some integers Z_1 and Z_2 such that \gcd(Z_1,Z_2)=1.
If k is an odd integer, then lifting the exponent lemma (see, for example, [lte, titu]) implies that \dfrac{X_1^k-X_2^k}{X_1-X_2} is odd. Thus,

a contradiction.
From 2 cases, we can conclude that if k is odd, then X_1^k+X_2^k= Z_1^{2k-1} for some Z_1\in \mathbb{Z}.

( Theorem 3) Let n be an integer greater than 1. Let x_i’s be periodic points of exact period n such that f_{d,c}(x_i)=x_{i+1} and f_{d,c}(x_n)=x_{n+1}=x_1 for i=1,2,\dots,n. By Theorem 1, we can assume that d is even.

Without loss of generality, assume that \displaystyle |x_2|=\max_{1\leq i\leq n}(|x_i|). By Corollary 1 and Theorem 8, we can write x_i=\dfrac{X_i}{Z} such that \gcd(X_i,X_{i+1})=1=\gcd(X_i,Z) for i=1,2,\dots,n. We have that f_{d,c}(x_2)-f_{d,c}(x_1)=x_3-x_2 or, equivalently, x_2^d-x_1^d=x_3-x_2. Thus, \dfrac{X_2^d-X_1^d}{X_3-X_2}=Z^{d-1}, i.e.,

If \max(|X_2|,|Z|)=1, then Corollary 1 implies that c\in \mathbb{Z}. Thus, f_{d,c}(x)=x^d+c\in\mathbb{Z}[x] and X_1=x_1,X_2=x_2,X_3=x_3\in\{0,1,-1\}. Since f_{d,c}(x)\in\mathbb{Z}[x], we have n\leq 2 and c\in\{0,-1,-2\}. Otherwise, f_{d,c}(a) converges to \infty for every a\in \mathbb{R}. If c=0 or -2, we can only have fixed points x\in \{0,1\} or x=-1, respectively. Thus, period n=1, a contradiction. If c=-1 and d even, then period n=2. This is also a trivial case.
For the case \max(|X_2|,|Z|)>1, we assume the abc-conjecture. From Eq. ?, and properties of X_1,X_2,X_3 and Z, we have the following system:

The desired result follows directly from Lemma 2 for k=d=l and m=d-1. Therefore, if d is sufficiently large and c\neq -1, then f_{d,c} cannot have a rational point of period greater than 1. Therefore, f_{d,c} can have at most 4 fixed points (with no preperiodic points) if d is odd, and at most 2 fixed points if d is even (including \infty). By Lemma 5, \#\operatorname{PrePer}(f_{d,c},\mathbb{P}^1(\mathbb{Q}))\leq 4. For c= -1 and d even, it is easy to see that f_{d,-1} can only have points of exact period 2: x_1=0,x_2=-1 with the only one preperiodic point x_0=1, and \infty as a fixed point.
For the rest of the proof, we assume explicit abc-conjecture. By Lemma 2, the result holds for d\geq 10.
Consider Eq. ? for the case d= 8.
Let W=\max(|X_1|,|X_2|,|X_3|,|Z|). Theorem 11 implies that for \epsilon=0.62991, we can choose K=10 so that, by Lemma 2,

If W>1300, then

a contradiction.
If W\leq 1300 and period n=2, then we can exhaust all possible values of X_1,X_2,X_3 and Z. The only possible value of c is c=-1 (The computation was done on MS Windows 10 running Julia 1.4.1 and AMD Ryzen 7 2700X 3.7 Ghz processor and 16 Gb RAM.). For W\leq 1300 and period n>2, it was examined by Hutz [MR3266961] in table 1 with the height bound 1000000 that, for 3\leq d\leq 11, f_{d,c}(x)=x^d+c has no periodic points with exact period greater than 2.
For odd d, by Theorem 1, f_{d,c} has at most 4 preperiodic (fixed) points.

( Theorem 6) Let K=\mathbb{Q}(\theta), \theta^5=3, with the real value \theta_0\sim 1.24573. Let O_K be the ring of integers of K. The class number of K is 1, and there are two units, normalized to have norm +1, which we may take as

We have the ideal factorizations

Let \phi be the evaluation homomorphism \theta\rightarrow \theta_0. Then

We set (u,v)=(x/z^2,y/z^5), with (x,z)=1, and the equation becomes

A common (ideal) divisor of the two factors on the left divides (x+\theta^4z^2) and (5\theta^{16}), so can only contain factors (\theta) and (p_5). Suppose p_5 is a common divisor. Then x+16z^2\equiv 0\mod 5, so x+z^2=5M, say, with

forcing 5\mid y,5\mid z, 5\mid x, a contradiction. Thus, any common divisor can only be a power of (\theta). Further, when (\theta) is a common divisor, then \theta\mid x, so that 3\mid x, which implies that the greatest common divisor is precisely (\theta)^4. In summary, the greatest common divisor of the two factors is (g), where g=1,\theta^4.
We thus deduce element equations

for some unit u, and elements a,b\in O_K with gab=y. Without loss of generality, u=\pm \epsilon_1^{i_1}\epsilon_2^{i_2}, where i_1,i_2=0,1. The first equation above implies \phi(g)\phi(u)\phi(a)^2=x+\theta_0^4z^2>0, so we must have \phi(u)>0, and the negative sign is impossible.
It suffices to find all x,z\in \mathbb{Z} satisfying (2); and since g=1,\theta^4 is a perfect square, it suffices to consider solely the case g=1. Let

for u=1,\epsilon_1,\epsilon_2,\epsilon_1\epsilon_2.
Case u=1. The curve C_1 is birationally equivalent to the curve

The elliptic Chabauty technique developed by Bruin and implemented into Magma is able to treat this curve without problem (the auxiliary prime used being 7), and the result is that the only points X=x/z^2 on C_1 for x,y\in\mathbb{Z}, occur for (x,z)=(1,0),(0,1), corresponding to the point at infinity, and (u,v)=(0,9) on the original curve.
Case u=\epsilon_1. The curve C_{\epsilon_1} contains the point

allowing construction of a birational map to the curve

Again, elliptic Chabauty techniques using Magma, with auxiliary prime 11, show that the only point X=x/z^2 on C_{\epsilon_1} for x,z\in \mathbb{Z}, is given by (x,z)=(-2,1), corresponding to (u,v)=(-2,7) on the original curve.
Case u=\epsilon_2. The point on C_{\epsilon_2}

allows construction of a birational map to the curve

Elliptic Chabauty techniques using Magma, with auxiliary prime 11, show that there are no points X/z^2 on C_{\epsilon_2} for x,z\in\mathbb{Z}.
Case u=\epsilon_1 \epsilon_2. The point on C_{\epsilon_1\epsilon_2}

leads to a birational map to the curve

Elliptic Chabauty techniques in Magma, with auxiliary prime 11, show that the only points X=x/z^2 on C_{\epsilon_1\epsilon_2} for x,z\in \mathbb{Z}, are given by (x,z)=(-1,1),(3,1). The former is “ spurious" in that equation (1) is not satisfied for (x,z)=(-1,1); and the latter corresponds to (u,v)=(3,18) on the original curve.

( Lemma 4) Let c=\dfrac{C}{Z^d}\in \mathbb{Q}. For c\in \{-2,-1,0,1,2\}, we can verify directly that only f_{d,0}(x)=x^d has 2 rational fixed points 0 and 1 (with only one preperiodic point -1) for even d\geq 4, and 3 rational fixed points -1,0,1 for odd d\geq 3. For every other c’s, f_{d,c} has at most 1 rational fixed point.

For c\in\mathbb{Q}\backslash\{-2,1,0,1,2\}, we have that 0 is not a preperiodic point of f_{d,c}(x)=x^d+c. Otherwise, if c\in \mathbb{Q}\backslash \mathbb{Z}, then for some prime p, f_{d,c}^n(0) diverges to infinity p-adically. For c\in \mathbb{Z}\backslash \{-2,1,0,1,2\}, f_{d,c}^n(0) diverges to infinity.
If c \in\mathbb{Z}\backslash \{-2,1,0,1,2\}, then f_{d,c}(x) has at most one fixed point. Otherwise, if we have 2 integral fixed points X\neq Y, then X^d-X=c=Y^d-Y, equivalently, \dfrac{X^d-Y^d}{X-Y}=1. Note that \left| \dfrac{X^d-Y^d}{X-Y}\right|\geq 2, for \max(|X|,|Y|)\geq 2 and d\geq 3. Since \max(|X|,|Y|)\leq 1, we have X,Y\in\{-1,1\} and d is odd. Therefore, c=\pm 2, a contradiction.
Consider the case c\not\in \mathbb{Z}, i.e., |Z|>1. Assume that there are 2 fixed points

where \dfrac{X_1}{Z} and \dfrac{X_2}{Z} are written in the lowest forms. Without loss of generality, assume |X_1|\geq |X_2|. Since x_1 and x_2 are fixed points,

Therefore, we have

If \gcd(X_1,X_2)\neq 1, then \gcd(X_1,Z)>1, a contradiction.
Thus, \gcd(X_1,X_2)=1 and \max(|X_1|,|X_2|,|Z|)>1. By Lemma 2, the equation has no integer solution for sufficiently large d. Moreover, if we assume the explicit abc-conjecture, then it is true for d\geq 10.
Consider for the case 8\leq d<10.
For d=8, by Lemma 3, there is f_{8,c'}(x)=x^8+c'\in\mathbb{Q}[x] with 2-periodic points, say, y_1=-x_1=\dfrac{-X_1}{Z} and y_2=-x_2=\dfrac{-X_2}{Z}. Since

we have X_1^8-X_2^8=(X_1-X_2)Z^7.
Theorem 11 implies that for \epsilon=0.62991, we can choose K_{\epsilon}=10.
Let W=\max(|X_1|,|X_2|,|Z|). Applying the same argument in Lemma 2, we have

or equivalently,

If W>1300, then

a contradiction.
If W\leq 1300, then we can exhaust all possible values of X_1,X_2, and Z. None of the values can be points of period 2, except c'=-1. By Lemma 3, this yields the value of c=0, a contradiction.

( Lemma 5) Assume to the contrary that there is r\in \mathbb{Q} such that r is a 1_2-type point of f_{d,c}. Thus, f_{d,c}(r)=-x_0 for a fixed point x_0\in \mathbb{Q}. Since c=x_0-x_0^d, we have

Thus, r^d+2x_0-x_0^d=0. By Corollary 1, write r=\dfrac{R}{Z} and x_0=\dfrac{X_0}{Z} in the lowest forms. Thus, \gcd(X_0,Z)=1=\gcd(R,Z). Then

Since X_0 and Z are integers, X_0\mid R^d. We consider 2 cases.
Case 1. X_0 is odd.
Then R is odd (otherwise, 2Z^{d-1} is odd).
Since

we have \gcd(X_0^{d-1},\dfrac{R^d}{X_0})=1. Thus, X_0^{d-1},\dfrac{R^d}{X_0} and 2Z^{d-1} are coprime. Assume the abc-conjecture with \epsilon =1. Then there is a constant K_1>0 such that

If |X_0|=\max(|X_0|,|R|,|Z|)>1, then

Therefore, |X_0^{d-7}|< 4K_1. Thus, for sufficiently large d, there are no integers solution to Eq. Proof. Similarly, we can yield the same result when |R| and |Z| are the maximum values. Moreover, if we assume the explicit abc-conjecture with \epsilon=1 and K_1<1, the result holds when d\geq 9. For d=8, we check all values |X_0|,|R|,|Z|\leq 3. None of the values, except X_0=R=0, yields the solution to the equation.
Case 2. X_0 is even.
Since X_0\mid R^d, R is even. Consider

Since \dfrac{X_0^{d-1}}{2} is an integer, so is \dfrac{R^d}{2X_0}. Since \gcd(X_0,Z)=1, \gcd(\dfrac{X_0^{d-1}}{2},\dfrac{R^d}{2X_0})=1. From the previous argument in Case 1 for the tuple (\dfrac{X_0^{d-1}}{2},\dfrac{R^d}{2X_0},Z^{d-1}), we can yield the same result in this case.