A Common Approach to Three Open Problems in Number Theory

1. Composite numbers of the form $2^{2^n}+1$

Let $\mathcal{A}$ denote the following system of equations:

$$\left\{x_i·x_j=x_k:i,j,k\in \{1,\dots ,7\}\right\}\cup \left\{2^{2^{x_j}}=x_k:j,k\in \{1,\dots ,7\}\right\}$$

The following subsystem of $\mathcal{A}$has exactly one solution in ${(\mathbb{N}\setminus \{0,1\})}^7$, namely $(2,4,16,2,2,2,2)$.

Hypothesis 1.

If a system of equations $\mathcal{S}\subseteq \mathcal{A}$ has at most five equations and at most finitely many solutions in ${(\mathbb{N}\setminus \{0,1\})}^7$, then each such solution $(x_1,\dots ,x_7)$ satisfies $x_1,\dots ,x_7⩽16$.

Lemma 1.

([7] (p. 109)). For every non-negative integers x and y, $x+1=y$ if and only if $2^{2^x}·2^{2^x}=2^{2^y}$.

Theorem 1.

Hypothesis 1 implies that $2^{2^{x_1}}+1$ is composite for infinitely many integers $x_1$ greater than 1.

Proof. 

Assume, on the contrary, that Hypothesis 1 holds and $2^{2^{x_1}}+1$ is composite for at most finitely many integers $x_1$ greater than 1. Then, the equation

$$x_2·x_3=2^{2^{x_1}}+1$$

has at most finitely many solutions in ${(\mathbb{N}\setminus \{0,1\})}^3$. By Lemma 1, in positive integers greater than 1, the following subsystem of $\mathcal{A}$ has at most finitely many solutions in ${(\mathbb{N}\setminus \{0,1\})}^7$ and expresses that $x_2·x_3=2^{2^{x_1}}+1$. Since $641·6700417=2^{2^5}+1>16$, we get a contradiction.    □

Most mathematicians believe that $2^{2^n}+1$ is composite for every integer $n⩾5$, see [2] (p. 23).

Open Problem 1.

([3] (p. 159)). Are there infinitely many composite numbers of the form $2^{2^n}+1$?

Primes of the form $2^{2^n}+1$ are called Fermat primes, as Fermat conjectured that every integer of the form $2^{2^n}+1$ is prime, see [3] (p. 1). Fermat remarked that $2^{2^0}+1=3$, $2^{2^1}+1=5$, $2^{2^2}+1=17$, $2^{2^3}+1=257$, and $2^{2^4}+1=65537$ are all prime, see [3, p. 1].

Open Problem 2.

([3] (p. 158)). Are there infinitely many prime numbers of the form $2^{2^n}+1$?

2. The Brocard-Ramanujan equation $x!+1=y^2$

Let $\mathcal{B}$ denote the following system of equations:

$$\{x_i·x_j=x_k:i,j,k\in \{1,\dots ,6\}\}\cup \{x_j!=x_k:(j,k\in \{1,\dots ,6\})\wedge (j\ne k)\}$$

The following subsystem of $\mathcal{B}$ has exactly two solutions in positive integers, namely $(1,\dots ,1)$ and $(2,2,4,24,24!,(24!)!)$.

Hypothesis 2.

If a system of equations $\mathcal{S}\subseteq \mathcal{B}$ has at most finitely many solutions in positive integers $x_1,\dots ,x_6$, then each such solution $(x_1,\dots ,x_6)$ satisfies $x_1,\dots ,x_6⩽(24!)!$.

Lemma 2.

For every positive integers x and y, $x!·y=y!$ if and only if

$$(x+1=y)\vee (x=y=1)$$

Theorem 2.

Hypothesis 2 implies that if the equation $x_1!+1=x_2^2$ has at most finitely many solutions in positive integers $x_1$ and $x_2$, then each such solution $(x_1,x_2)$ belongs to the set $\left\{\right(4,5),(5,11),(7,71\left)\right\}$.

Proof. 

The following system of equations ${\mathcal{B}}_1$is a subsystem of $\mathcal{B}$. By Lemma 2, the system ${\mathcal{B}}_1$ expresses that

$$(x_1!+1=x_2^2)\vee (x_1=\dots =x_6=1)$$

If the equation $x_1!+1=x_2^2$ has at most finitely many solutions in positive integers $x_1$ and $x_2$, then ${\mathcal{B}}_1$ has at most finitely many solutions in positive integers $x_1,\dots ,x_6$ and Hypothesis 2 implies that every tuple $(x_1,\dots ,x_6)$ of positive integers that solves ${\mathcal{B}}_1$ satisfies $x_6⩽(24!)!$. Hence, $x_1\in \{1,\dots ,23\}$. If $x_1\in \{1,\dots ,23\}$, then $x_1!+1$ is a square only for $x_1\in \{4,5,7\}$.    □

It is conjectured that $x!+1$ is a square only for $x\in \{4,5,7\}$, see [10] (p. 297). A weak form of Szpiro’s conjecture implies that the equation $x!+1=y^2$ has only finitely many solutions in positive integers, see [6].

3. Erdös’ equation $x(x+1)=y!$

Let $\mathcal{C}$ denote the following system of equations:

$$\{x_i·x_j=x_k:(i,j,k\in \{1,\dots ,6\})\wedge (i\ne j)\}\cup \{x_j!=x_k:(j,k\in \{1,\dots ,6\})\wedge (j\ne k)\}$$

The following subsystem of $\mathcal{C}$has exactly three solutions in positive integers, namely $(1,\dots ,1)$, $(1,1,2,2,2,2)$, and $(2,2,3,6,720,720!)$.

Hypothesis 3.

If a system of equations $\mathcal{S}\subseteq \mathcal{C}$ has at most finitely many solutions in positive integers $x_1,\dots ,x_6$, then each such solution $(x_1,\dots ,x_6$) satisfies $x_1,\dots ,x_6⩽720!$.

Theorem 3.

Hypothesis 3 implies that if the equation $x_1(x_1+1)=x_2!$ has at most finitely many solutions in positive integers $x_1$ and $x_2$, then each such solution $(x_1,x_2)$ belongs to the set $\left\{\right(1,2),(2,3\left)\right\}$.

Proof. 

The following system of equations ${\mathcal{C}}_1$is a subsystem of $\mathcal{C}$. By Lemma 2, the system ${\mathcal{C}}_1$ expresses that

$$(x_1(x_1+1)=x_2!)\vee (x_1=\dots =x_6=1)$$

If the equation $x_1(x_1+1)=x_2!$ has at most finitely many solutions in positive integers $x_1$ and $x_2$, then ${\mathcal{C}}_1$ has at most finitely many solutions in positive integers $x_1,\dots ,x_6$ and Hypothesis 3 implies that every tuple $(x_1,\dots ,x_6)$ of positive integers that solves ${\mathcal{C}}_1$ satisfies $x_3⩽720!$. Hence, $x_2\in \{1,\dots ,720\}$. If $x_2\in \{1,\dots ,720\}$, then $x_2!$ is a product of two consecutive positive integers only for $x_2\in \{2,3\}$ because the following MuPAD program

• for x2 from 1 to 720 do
• x1:=round(sqrt(x2!+(1/4))-(1/2)):
• if x1∗(x1+1)=x2! then print(x2) end_if:
• end_for:

returns 2 and 3. □

The question of solving the equation $x(x+1)=y!$ was posed by P. Erdös, see [1]. F. Luca proved that the $abc$ conjecture implies that the equation $x(x+1)=y!$ has only finitely many solutions in positive integers, see [4].

4. There is no hope for similar hypotheses with an arbitrary number of variables

Let $f\left(1\right)=2$, $f\left(2\right)=4$, and let $f(n+1)=f\left(n\right)!$ for every integer $n⩾2$. Let ${\mathcal{U}}_1$ denote the system of equations $\{x_1!=x_1$. For an integer $n⩾2$, let ${\mathcal{U}}_n$ denote the following system of equations:

For every positive integer n, the system ${\mathcal{U}}_n$ has exactly two solutions in positive integers $x_1,\dots ,x_n$, namely $(1,\dots ,1)$ and $\left(f\right(1),\dots ,f(n\left)\right)$. For a positive integer n, let ${\mathsf{\Psi }}_n$ denote the following statement: if a system of equations

$$\mathcal{S}\subseteq \{x_i·x_j=x_k:i,j,k\in \{1,\dots ,n\}\}\cup \{x_j!=x_k:j,k\in \{1,\dots ,n\}\}$$

has at most finitely many solutions in positive integers $x_1,\dots ,x_n$, then each such solution $(x_1,\dots ,x_n)$ satisfies $x_1,\dots ,x_n⩽f\left(n\right)$. The statements ${\mathsf{\Psi }}_n$ are discussed in [8,9].

Theorem 4.

Every factorial Diophantine equation can be algorithmically transformed into an equivalent system of equations of the forms $x_i·x_j=x_k$ and $x_j!=x_k$.

Proof. 

It follows from Lemmas 2–4 in [7] and Lemma 2. □

The statement $\forall n\in \mathbb{N}\setminus \left\{0\right\}{\mathsf{\Psi }}_n$ is dubious. By Theorem 4, this statement implies that there is an algorithm which takes as input a factorial Diophantine equation and returns an integer which is greater than the solutions in positive integers, if these solutions form a finite set. This conclusion is strange because properties of factorial Diophantine equations are similar to properties of exponential Diophantine equations and a computable upper bound on non-negative integer solutions does not exist for exponential Diophantine equations with a finite number of solutions, see [5].