7
$\begingroup$

I have a sequence given by an explicit formula for n-th term:

seq[n_] := FullSimplify[LerchPhi[1/2, 2, -n] - 2^(-2-n) (π^2 - 6 Log[2]^2)/3];
Array[seq, 10]
(* {1, 3/4, 35/72, 11/36, 347/1800, 149/1200, 9701/117600, 209/3675, 8093/198450, 6031/198450} *)

I am trying to find a recurrence relation for this sequence (e.g. in a form of DifferenceRoot object). An invocation of DifferenceRootReduce does not produce a desired result:

DifferenceRootReduce[LerchPhi[1/2, 2, -n] - 2^(-2-n) (π^2 - 6 Log[2]^2)/3, n]
(* LerchPhi[1/2, 2, -n] - 2^(-2-n) (π^2 - 6 Log[2]^2)/3 *)

Are there any other ways to find a recurrence relation for a sequence?

$\endgroup$
0

1 Answer 1

5
$\begingroup$

Using the identity:

$$\Phi (z,s,a)=\frac{\Phi (z,s,a-1)-\left((a-1)^2\right)^{-s/2}}{z}$$

First we define two recursive functions:

rGexp[0] = -1/4 (Pi^2 - 6 Log[2]^2)/3;
rGexp[n_] := 1/2 rGexp[n - 1]

rlp[0] = LerchPhi[1/2, 2, 0];
rlp[n_] := 1/2 rlp[n + 1] + 1/(n)^2

Then we add them:

rSeq[n_] := rlp[-n] + rGexp[n]

Simplify@Table[rSeq[n], {n, 1, 10}]

(* {1, 3/4, 35/72, 11/36, 347/1800, 149/1200, 9701/117600, 209/3675, 8093/198450} *)

Now, since

LerchPhi[1/2, 2, 0]
(* π^2/12 - Log[2]^2/2 *)

The above can be reduced to (Thanks to @ybeltukov for noting it!):

f[0] = 0
f[n_] := f[n - 1]/2 + 1/n^2
f /@ Range@9
(* {1, 3/4, 35/72, 11/36, 347/1800, 149/1200, 9701/117600, 209/3675, 8093/198450} *)
$\endgroup$
7
  • $\begingroup$ I'd rather say that I found it by using Mathematics, not Mathematica. Anyway ... $\endgroup$ Commented Sep 3, 2014 at 2:42
  • $\begingroup$ belisarius, your result is quite simple: f[0] = 0; f[n_] := f[n - 1]/2 + 1/n^2; :) May be you add it to the answer? $\endgroup$
    – ybeltukov
    Commented Sep 3, 2014 at 6:50
  • $\begingroup$ @ybeltukov: you're right. But how did you find this simple recurrence relation? We could formulate a modified problem as follows: given a numerical sequence of (say) rational numbers, is there (1) some "NDifferenceRootReduce" to find the recurrence relation or (2) a kind of "RFindFit" for a recurrence relation like f(n) = a f(n-1) + b(n), where a is a numerical parameter to be found and b(n) could be e.g. a polynomial with parameters to be found as well? $\endgroup$ Commented Sep 3, 2014 at 11:34
  • $\begingroup$ @Dr.WolfgangHintze I just looked at the belisarius's answer and saw that these two recurrence relations reduced to f[n_] := f[n - 1]/2 + 1/n^2. DifferenceRootReduce should find this formula, but LerchPhi is too difficult for it. For example, it works fine with Fibonacci[n] + n (see options here). $\endgroup$
    – ybeltukov
    Commented Sep 3, 2014 at 11:51
  • 1
    $\begingroup$ @Dr.WolfgangHintze It is interestiong, that Mathematica can do it in the opposite way. RSolve[{f[n] == f[n - 1]/2 + 1/n^2, f[0] == 0}, f[n], n] gives an equivalent formula with LerchPhi. $\endgroup$
    – ybeltukov
    Commented Sep 3, 2014 at 11:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.