Primes Are Rather Elusive

If we listed all the counting numbers $1,2,3,4,5,\dots$, excluded 1, and then crossed out all the multiples of $2,3,4,\dots$ we’d be left with the primes. This sieving process emphasises that primes are defined more by what they are not, than by what they are.

If there was a simple pattern in the primes, we’d be able to encode it into a simple formula for generating them. For example, the triangle numbers $1,3,6,10,15,\dots$ can be generated by the simple expression $\frac{1}{2}n(n+1)$. The prime numbers, however, have resisted attempts by mathematicians over hundreds of years to find precise and simple patterns in them.

One of the first questions anyone enthusiastic about prime numbers asks is whether a polynomial can generate the $n^{th}$ prime. Polynomials are both simple and rather flexible, and it would be quite pleasing if one could generate primes.

Let’s prove that prime numbers are so elusive that no simple polynomial in $n$ can generate the $n^{th}$ prime.

slides: [pdf]

video: [youtube]

No Simple Polynomial Generates Only Primes

A polynomial in $n$ has the following general form, simple yet flexible.

$$P(n)=a+bn+c{n}^2+d{n}^3+\dots +\alpha n^{\beta }$$

By simple polynomial we mean the coefficients $a,b,c\dots \alpha$ are whole numbers. Let’s also  say that $b,c,d,\dots \alpha$ are not all zero. This way we exclude trivial polynomials like $P(n)=7$ that only generate a single value no matter what $n$ is.

Let’s start our proof by assuming there is indeed a $P(n)$ that generates only primes, given a counting number $n$. When $n=1$, it generates a prime, which we can call $p_1$.

$$p_1=P(1)=a+b+c+d+\dots +\alpha$$

Now let’s try $n=(1+p_1)$.

$$P(1+p_1)=a+b(1+p_1)+c(1+p_1{)}^2+d(1+p_1{)}^3+\dots$$

That looks complicated, but all we need to notice is that if we expand out all the terms, we’ll have two kinds, those with $p_1$ as a factor, and those without. We can collect together all those terms with factor $p_1$ and call them $p_1\cdot X$.

$$P(1+p_1)=(a+b+c+d+e+\dots \alpha )+p_1·X$$

We then notice that $(a+b+c+d+e+\dots \alpha )$ is actually $p_1$.

$$\begin{array}{rl}P(1+p_1)& =p_1+p_1\cdot X\\ \\ & =p_1(1+X)\end{array}$$

Since $X$ is a whole number, this is divisible by $p_1$. It shouldn’t be because $P(1+p_1)$ is supposed to be a prime. This contradiction means the starting assumption that there is a polynomial $P(n)$ that generates only primes is wrong.

We’ve actually proved a stronger statement than we intended. We intended to prove that there is no polynomial $P(n)$ that generates the $n^{th}$ prime. We ended up proving that no simple polynomial $P(n)$ can generate only primes.

Polynomials With Rational Coefficients

Insisting on integer coefficients for polynomials might seem overly restrictive. Let's broaden our definition to allow rational coefficients of the form $\frac{s}{t}$ where $s$ and $t$ are integers. 

We again assume $P(n)$ does indeed generate only primes, and so $p_1=P(1)$ is prime. This time we'll consider $n=(1+k\cdot p_1)$.

$$\begin{array}{rl}P(1+k\cdot p_1)& =a+b(1+k\cdot p_1)+c(1+k\cdot p_1{)}^2+\dots \\ & =p_1+k\cdot p_1\cdot X\\ & =p_1(1+k\cdot X)\end{array}$$

Here $X$ contains terms that are combinations of the rational coefficients $a,b,c\dots \alpha$ multiplied together. We can choose a $k$ which cancels all the denominators of the rational coefficients leaving $k\cdot X$ as an integer. The lowest common multiple of all the denominators is one way to do this.

Our proof by contradiction then continues as before because we've found an example of $P(n)$ that is not prime.

The primes really are rather elusive if even polynomials with rational coefficients can't generate only primes.