Processing math: 100%

Friday, September 25, 2015

A proof of Bertrand's postulate by Sylvester

In his paper "On Tchebycheff's Theory of the Totality of the Prime Numbers Comprised within Given Limits" from 1881, Sylvester gives an outline of a proof of Bertrand's postulate. Namely, he writes "the formulae above indicated would suffice to prove M. Bertrand's postulate, and would lead to an equation somewhat simpler in form than that led to by M. Tchebycheff's process". In this post, I provide more detail on this proof.
The proof is based on inequalities for the function V(x)=T(x)T(x2)T(x3)T(x6)
where T(x)=1nxlogn
If ψ(x)=nxΛ(n) with Λ the von Mangoldt function, then T(x)=n=1ψ(xn)
and thus V(x)=ψ(x)+ψ(x5)2ψ(x6)+ψ(x7)+ψ(x11)2ψ(x12)+
The coefficients of ψ(xn) in this expression have period 6. In contrast to a similar equation in Chebyshev's Mémoire, the signs in (2) are not alternating, this makes it less straightforward to use (2) to obtain bounds on ψ(x).

For all real x1, one has xlogxxlogx+1T(x)xlogxx+logx+1
From (1) and (3) one immediately deduces that for all real x6 V(x)Bx+4logx22log6
and Bx4logx2+2log6V(x)
where B=12log2+13log3+13log3.

Calculation of upper bound on ψ(x)
Because nψ(xn) is decreasing, it follows that ψ(x7)+ψ(x11)2ψ(x12)0
and ψ(x13)+ψ(x17)2ψ(x18)0
and so on. Therefore from (2) one obtains V(x)ψ(x)+ψ(x5)2ψ(x6)
Because ψ(x5)ψ(x6)0, it follows also that V(x)ψ(x)ψ(x6)
One can obtain an upper bound on ψ(x) with a standard calculation: combining (4) and (6) and using 22log6<5, one has for all real x6 that ψ(x)ψ(x6)Bx+4logx5
Take mN such that 6x6m<62. Applying (7) several times gives ψ(x6)ψ(x62)Bx6+4logx65ψ(x62)ψ(x63)Bx62+4logx625ψ(x6m)ψ(x6m+1)Bx6m+4logx6m5
Adding the inequalities gives ψ(x)ψ(x6m+1)+65B+4(m+1)logx(m+1)5
The inequality 6m+1x gives m+1logxlog6. Also, 1x6m+1<6, thus ψ(x6m+1)3. Therefore, for all x6 one has ψ(x)65Bx+4log2xlog6
One can verify in Mathematica that this inequality is also valid for x1

Calculation of lower bound on ψ(x)
One has 2ψ(x6)+ψ(x7)+ψ(x11)0
and 2ψ(x12)+ψ(x13)+ψ(x17)0
and so on. Therefore from (2) one obtains V(x)ψ(x)+ψ(x5)
Combining this inequality with (5) and (8) then gives for x5 Bx4logx2+2log665Bx54log2x/5log6ψ(x)
I simplify this expression slightly by using 2+2log6>0, then 1925Bx4logx4log2x/5log6ψ(x)


Proof of Bertrand's postulate
Because 6/5B19/25B<2, the upper and lower bound are sufficiently close and one can now deduce Bertrand's postulate in a standard way. I do not do so here, because the post is already quite long. Details can for example be found in Murty's book.

Other posts about Bertrand's postulate

No comments:

Post a Comment