Saturday, 17 October 2015

summation - Symmetry relation for finite sums of generalized harmonic numbers



EDITs
- extension to alternating sums (6)
- extension to general sums with parameter x (8)
- extension to general sums with two parameters (11)



Extended post




The relation



nk=1(H(p)kkq+H(q)kkp)=H(p)nH(q)n+H(p+q)n



where H(p)k=kj=1jp, is a nice and important symmetry relation with various applications. Notice the finite upper index n. The proof is not hard.



Corollary 1



If q=p the relation simplifies to




nk=1H(p)kkp=12((H(p)n)2+H(2p)n)



Corollary 2



In the limit n (1) turns into



k=1(H(p)kkq+H(q)kkp)=ζ(p)ζ(q)+ζ(p+q)



where




ζ(s)=limnH(s)n=k=1ks



is the Riemann zeta function.



For q=p formula (3) gives the analogue to (2)



k=1H(p)kkp=12(ζ(p)2+ζ(2p))



This formula was also given in [1], and there it was claimed that it is valid for real p.




Alternating sums



A similar relation can be derived for alternating sums.



Define



A(p)n=nk=1(1)k1kp



then the basic relation is




nk=1(1)kH(p)kkq+nk=1A(q)kkp=H(p)nA(q)n+A(p+q)n



From (6) other corollaries and formulas can be derived in a manner similar to the one for the non-alternating sums described above.



Generalized relation with parameter x



It can be shown with similar methods as before that a more general relation holds from which the two formulas discussed before are special cases.



Let




H(p)n(x)=nk=1xkkp



H(p)n is understood as H(p)n(+1) and we have H(p)n(1)=A(p)n (c.f. (5)).



Then the relation is



nk=1(xkH(p)k(1)kq+H(q)k(x)kp)=H(p)n(1)H(q)n(x)+H(p+q)n(x)



Notice that this relation has lost the symmetry.




For x=1 and x=1 we recover the relations (1) and (6), respectively.



In the limit n we encounter on the right hand side the polylog function



limnH(p)n(x)=k=1xkkp=Lip(x)



so that we have



k=1(xkH(p)k(1)kq+H(q)k(x)kp)=Lip(1)Liq(x)+Lip+q(x)




Note "added in proof"



The special case x=1 of relation (10) can be uncovered (it is there more or less conceiled) from [2], page 33, where it is called "shuffle relation". We now have here an elementary proof of a generalization it.



Generalized relation with two parameters x and y



We can even go one final step further and write down this two-parametric symmetry relation



nk=1xkkqkm=1ymmp+nk=1ykkpkm=1xmmq=(nk=1xkkq)(nk=1ykkp)+nk=1(xy)kkp+q




This relation includes the various cases of alternating and non-alternating sums for apropriate choices of x and y.



In the limit n we obtain



k=1xkkqkm=1ymmp+k=1ykkpkm=1xmmq=Liq(x)Lip(y)+Lip+q(xy)



The zeta functions of (3) have been generalized to polylog functions.



Question




Prove (1), (6), (8), (11) and (12)



References



[1] Identities For Generalized Harmonic Number



[2] http://algo.inria.fr/flajolet/Publications/FlSa98.pdf


Answer



To prove (1) I will use summation by parts in the form given by

nk=1fkgk=fnGnn1k=1Gk(fk+1fk),whereGn=nk=1gk.



Now consider the sum
nk=1H(p)kkq.


Let fk=H(p)k and gk=1/kq. So
Gn=nk=11kq=H(q)n,

and
fk+1fk=H(p)k+1H(p)k=(H(p)k+1(k+1)p)H(p)k=1(k+1)p,

where we have made use of the following result for the generalised harmonic numbers
H(a)n+1=H(a)n+1(n+1)a.




On applying the summation by parts result to our sum we have
nk=1H(p)kkq=H(p)nH(q)nn1k=1H(q)k(k+1)p.


Shifting the index is the sum appearing to the right gives
nk=1H(p)kkq=H(p)nH(q)nnk=2H(q)k1kp =H(p)nH(q)nnk=21kp(H(q)k1kq)=H(p)nH(q)nnk=2H(q)kkp+nk=21kp+q=H(p)nH(q)nnk=1H(q)kkp+nk=11kp+q=H(p)nH(q)nnk=1H(q)kkp+Hp+q)n

or after arranging
nk=1(H(p)kkq+H(q)kp)=H(p)nH(q)n+H(p+q)n,

as required to show.



The result given by (6) can be proved in a similar fashion as to what was done above using summation by parts.


No comments:

Post a Comment

real analysis - How to find limhrightarrow0fracsin(ha)h

How to find limh0sin(ha)h without lhopital rule? I know when I use lhopital I easy get $$ \lim_{h\rightarrow 0}...