You are looking for a set of arithmetic sequences. We'll consider your example
ee = {0, 4, 6, 10, 12, 16, 18, 22};
which has two such sequences, and an example with four of them.
ff = {0, 3, 7, 11, 17, 20, 24, 28, 34, 37, 41, 45};
In this second one we start with {0,3,7,11} and then increase by 17. So what is the general way to get from the nth term to the n+1th? If the set has k sequences (k=2 for ee and 4 for ff) then add the modulus to the n-k+1th term. What is the modulus? It is the difference between the nth and n-kth terms.
Putting this together and assuming we know k (we don't in general, but we'll get to that) we have a recurrence of the form f(n+1)=f(n-k+1) + (f(n)-f(n-k)). So we need to find a recurrence (if one exists), check that it is of the correct form, and post-process if so.
Here is code to do all this. Note that it in effect solves for k.
findArithmeticSequences[ll : {_Integer ..}] := With[
{rec = FindLinearRecurrence[ll]},
{Take[ll, Length[rec] - 1], ll[[Length[rec]]]} /;
ListQ[rec] &&
(rec === {1, 1, -1} || MatchQ[rec, {1, 0 .., 1, -1}])
]
(Afficionados of pure functions might prefer the variant below. Failure cases are handled a bit differently, for no compelling reason.)
findArithmeticSequences2[ll : {_Integer ..}] :=
If[ListQ[#] &&
(# === {1, 1, -1} || MatchQ[#, {1, 0 .., 1, -1}]), {Take[ll,
Length[#] - 1], ll[[Length[#]]]}, $Failed] &[
FindLinearRecurrence[ll]]
Tests:
In[115]:= findArithmeticSequences[ee]
Out[115]= {{0, 4}, 6}
In[116]:= findArithmeticSequences[ff]
Out[116]= {{0, 3, 7, 11}, 17}
Note that one can "almost" do such problems by polynomial factorization (if the input has no partial sequences at the end). For example, the polynomial
In[117]:= poly = Plus @@ (x^ee)
Out[117]= 1 + x^4 + x^6 + x^10 + x^12 + x^16 + x^18 + x^22
factors into
(1+x^4)*(1+x^6+x^12+x^18)
which contains the needed information in a way that is easy to see. Unfortunately for this particular purpose, Factor will factor beyond this point, and obscure the information in so doing.
I keep wondering if there might be a signal processing way to go about this sort of thing, e.g. via DFTs. But I've not come up with anything.
Daniel Lichtblau