> I don't use smart methods like Newton's (mentioned by Dirk).
> My implementation is dumb and straightforward.
> I solve trivially constructed system of equations to find Tailor coefficients of din().
> (I simply substitute one Tailor series into another.)
> If n first coefficients are already known, the formula for the (n+1)-th one is not very
> complex.
Thank you for the code.
Does not look dumb / straightforward to me !
Still don't understand what d table is used for ...
Anyway, I think I also did it ... uh, by guessing :-)
First 20 din(x) coefficients matched your version (ignoring rounding errors).
A bonus is that coefficients generation is almost twice as fast.
Using your code as template, just replace f() and maclaurin_of_din():
local function f(coef)
local r = 0
for j = 1, #c do
r = r + coef[j] * g(2*j+1, #c+1-j, c)
end
return r
end
local d2 = {[0] = 1}
function maclaurin_of_din(k)
for n = #c + 1, k do
a = {}
local r, r2, r3 = f(c), f(d), f(d2)
s, a = -(2*n)*(2*n+1)*s
local t = (1/s-r-r2-r3)/4
assert(t*t < 1)
c[n], d[n], d2[n] = t, r + 3*t, r2 + 2*t
end
return c[k]
end
