Loading [MathJax]/jax/output/HTML-CSS/jax.js

Sunday, August 23, 2015

How not to prove Bertrand's postulate

This is a first post motivated by Chebyshev's paper "Mémoire sur les nombres premiers" from 1852. In this paper Chebyshev proves Bertrand's postulate that there is always a prime between a and 2a for all a2. Chebyshev bases his proof on inequalities for the function T(x)T(x2)T(x3)T(x5)+T(x30) with T(x)=1nxlogn Chebyshev's paper is quite easy to read and can be found at this link. In this post I will follow Chebyshev's reasoning on an easier version of (1).
A good motivation why Chebyshev uses (1) can be found in this master thesis. Namely, if one defines ψ(x)=nxΛ(n) with Λ the von Mangoldt function, then T(x)=n=1ψ(xn) and thus ψ(x)=n=1μ(n) T(xn) See Murty's book for example for these formulas. Because T(x)xlogxx, ψ(x) can be approximated as ψ(x)nxμ(n)(xnlogxnxn)nxμ(n)n(xlogxx)(nxμ(n)nlogn)x The prime number theorem, which was not proved yet at the time of Chebyshev, states that ψ(x)x. It thus seems natural to use some kind of approximation to the Möbius function ˜μμ so that ˜μ(n)=0 for large n and ˜μ(n)n=0 in order to remove the term xlogx. Chebyshev took ˜μ(1)=1, ˜μ(2)=1, ˜μ(3)=1, ˜μ(5)=1, ˜μ(30)=1, and other ˜μ(n)=0. For this choice, it follows indeed that 1121315+130=0 To get close to the prime number theorem, one also would like that ˜μ(n)nlogn1. Chebyshev's choice gives ˜μ(n)nlogn0.92 which is quite close to 1. In the following, I will take instead ˜μ(1)=1, ˜μ(2)=2, and other ˜μ(n)=0. For this simpler choice, ˜μ(n)n=122=0 and ˜μ(n)nlogn0.69 Because 0.69 is farther from 1 than Chebyshev's value 0.92, one thus expects this simpler choice for ˜μ to give weaker results than Chebyshev's results. I now follow Chebyshev's reasoning from his paper but on this simpler choice of ˜μ. Additionally, I use Landau's large O notation whereas Chebyshev's uses explicit inequalities. I thus study the function T(x)2 T(x2) Because T(x)=n=1ψ(xn), one has T(x)2 T(x2)=ψ(x)ψ(x2)+ψ(x3)ψ(x4)+ This series is alternating and decreasing, thus ψ(x)ψ(x2)T(x)2 T(x2)ψ(x) Because T(x)=xlogxx+O(logx), it follows that T(x)2 T(x2)=xlogxx+O(logx)2(x2logx2x2+O(logx2))=x+xlog2+x+O(logx)=log2x+O(logx) The terms involving xlogx have cancelled. This is not a coincidence, but a consequence of the choice of ˜μ. The function ψ(x) has thus the following lower bound log2x+O(logx)ψ(x) To obtain an upper bound , one can proceed as follows (Chebyshev uses a different argument) ψ(x)ψ(x2)log2x+O(logx)ψ(x2)ψ(x4)log2x2+O(logx)ψ(x4)ψ(x8)log2x4+O(logx) Because ψ(x2m)=0 if m>logxlog2, adding the above inequalities gives the upper bound ψ(x)2log2x+O(log2x)
Comments

  1. From (3) and (4) it follows that log2lim infxψ(x)xlim supxψ(x)x2log2 This is weaker than the prime number theorem limxψ(x)x=1
  2. The inequalities (3) and (4) are not strong enough to prove Bertrand's postulate. Indeed, I can only deduce that ψ(x)ψ(x2)log2x2log2x2+O(log2x)O(log2x) because the coefficient of the factor x has cancelled. Ramanujan has modified the reasoning above so that Bertrand's postulate does follow, see Murty p 38.
  3. In the next post, I will repeat the argument using explicit inequalities instead of using the big O notation.
Other posts about Bertrand's postulate

No comments:

Post a Comment