Processing math: 100%

Tuesday, September 12, 2017

Euler's transformation - series acceleration

Theorem. Assume that (an)n0 satisfies lim supn|an|1/n1. Then for x[0,1) we have n=0(1)nanxn=11+xn=0(1)nΔna0(x1+x)n. Here, both sides converges absolutely and Δn is the n-fold forward difference defined by Δnaj=nk=0(nk)(1)nkaj+k.

Proof. There is nothing to prove if x=0, so we assume throughout that 0<x<1. First, absolute convergence of the LHS of (*) is straightforward. To verify absolute convergence of the RHS, we first note that there exist C and β such that 1<β<x+12x and |an|Cβn. Then 12n|Δnaj|max0kn|aj+n|Cβj+k. Then absolute convergence of the RHS of (*) follows from comparison together with the following estimate |(1)nΔna0(x1+x)n|C(2βx1+x)n. Next we show the equality (*). We write n=0(1)nanxn=n=0(1)na0xn+n=0(1)n(ana0)xn=a01+xxn=0(1)n(an+1a0)xn. The latter sum can be computed as n=0(1)n(an+1a0)xn=n=0(1)n(l=0Δal)xn=l=0Δal(n=l(1)nxn)=11+xl=0(1)lΔalxl. Here, interchanging the order of summations is justified by the absolute convergence. So we obtain n=0(1)nanxn=a01+xx1+xl=0(1)lΔalxl. Since we have lim supn|Δan|1/n1, we can recursively apply (2) to obtain n=0(1)nanxn=11+xN1n=0(1)nΔna0(x1+x)n+(x1+x)Nl=0(1)lΔNalxl. But by the estimate (1), we have |(x1+x)Nl=0(1)lΔNalxl|(2βx1+x)Nl=0C(βx)l. Since βx<2βx1+x<1, this bound vanishes as N. Therefore we obtain the desired identity. ////

Example. Let f:[0,)R be a completely monotone function and set an=f(n). We know that f admits the representation f(s)=[0,)estμ(dt) for some finite Borel measure μ on [0,). Then a simple computation shows that (1)nΔna0=[0,)(1et)nμ(dt). This sequence is non-negative and decreases to 0 as n. So the limiting form of (*) is available and we obtain n=0(1)nan=n=0(1)nΔna02n+1. Notice that the resulting transformation converges exponentially fast regardless of how slow the original series was. As a useful example, consider μ(dt)=tp1et/Γ(p) for p>0. Then we have the following series acceleration for the Dirichlet eta function: n=0(1)n(n+1)p=n=012n+1nk=0(nk)(1)k(k+1)p. Of course this can be used to compute the Riemann zeta function.

No comments:

Post a Comment