-- | -- Module: Math.NumberTheory.Zeta.Hurwitz -- Copyright: (c) 2018 Alexandre Rodrigues Baldé -- Licence: MIT -- Maintainer: Alexandre Rodrigues Baldé -- -- Hurwitz zeta function. {-# LANGUAGE ScopedTypeVariables #-} {-# OPTIONS_HADDOCK hide #-} module Math.NumberTheory.Zeta.Hurwitz ( zetaHurwitz ) where import Math.NumberTheory.Recurrences (bernoulli, factorial) import Math.NumberTheory.Zeta.Utils (skipEvens, skipOdds) -- | Values of Hurwitz zeta function evaluated at @ζ(s, a)@ for @s ∈ [0, 1 ..]@. -- -- The algorithm used was based on the Euler-Maclaurin formula and was derived -- from -- by F. Johansson, chapter 4.8, formula 4.8.5. -- The error for each value in this recurrence is given in formula 4.8.9 as an -- indefinite integral, and in formula 4.8.12 as a closed form formula. -- -- It is the __user's responsibility__ to provide an appropriate precision for -- the type chosen. -- -- For instance, when using @Double@s, it does not make sense -- to provide a number @ε < 1e-53@ as the desired precision. For @Float@s, -- providing an @ε < 1e-24@ also does not make sense. -- Example of how to call the function: -- -- >>> zetaHurwitz 1e-15 0.25 !! 5 -- 1024.3489745265808 zetaHurwitz :: forall a . (Floating a, Ord a) => a -> a -> [a] zetaHurwitz eps a = zipWith3 (\s i t -> s + i + t) ss is ts where -- When given @1e-14@ as the @eps@ argument, this'll be -- @div (33 * (length . takeWhile (>= 1) . iterate (/ 10) . recip) 1e-14) 10 == div (33 * 14) 10@ -- @div (33 * 14) 10 == 46. -- meaning @N,M@ in formula 4.8.5 will be @46@. -- Multiplying by 33 and dividing by 10 is because asking for @14@ digits -- of decimal precision equals asking for @(log 10 / log 2) * 14 ~ 3.3 * 14 ~ 46@ -- bits of precision. digitsOfPrecision :: Integer digitsOfPrecision = let magnitude = toInteger . length . takeWhile (>= 1) . iterate (/ 10) . recip $ eps in div (magnitude * 33) 10 -- @a + n@ aPlusN :: a aPlusN = a + fromIntegral digitsOfPrecision -- @[(a + n)^s | s <- [0, 1, 2 ..]]@ powsOfAPlusN :: [a] powsOfAPlusN = iterate (aPlusN *) 1 -- [ [ 1 ] | ] -- | \sum_{k=0}^\(n-1) | ----------- | | s <- [0, 1, 2 ..] | -- [ [ (a + k) ^ s ] | ] -- @S@ value in 4.8.5 formula. ss :: [a] ss = let numbers = map ((a +) . fromInteger) [0..digitsOfPrecision-1] denoms = replicate (fromInteger digitsOfPrecision) 1 : iterate (zipWith (*) numbers) numbers in map (sum . map recip) denoms -- [ (a + n) ^ (1 - s) a + n | ] -- | ----------------- = ---------------------- | s <- [0, 1, 2 ..] | -- [ s - 1 (a + n) ^ s * (s - 1) | ] -- @I@ value in 4.8.5 formula. is :: [a] is = let denoms = zipWith (\powOfA int -> powOfA * fromInteger int) powsOfAPlusN [-1, 0..] in zipWith (/) (repeat aPlusN) denoms -- [ 1 | ] -- [ ----------- | s <- [0 ..] ] -- [ (a + n) ^ s | ] constants2 :: [a] constants2 = map recip powsOfAPlusN -- [ [(s)_(2*k - 1) | k <- [1 ..]], s <- [0 ..]], i.e. odd indices of -- infinite rising factorial sequences, each sequence starting at a -- positive integer. pochhammers :: [[Integer]] pochhammers = let -- [ [(s)_k | k <- [1 ..]], s <- [1 ..]] pochhs :: [[Integer]] pochhs = iterate (\(x : xs) -> map (`div` x) xs) (tail factorial) in -- When @s@ is @0@, the infinite sequence of rising -- factorials starting at @s@ is @[0,0,0,0..]@. repeat 0 : map skipOdds pochhs -- [ B_2k | ] -- | ------------------------- | k <- [1 ..] | -- [ (2k)! (a + n) ^ (2*k - 1) | ] second :: [a] second = take (fromInteger digitsOfPrecision) $ zipWith3 (\bern evenFac denom -> fromRational bern / (denom * fromInteger evenFac)) (tail $ skipOdds bernoulli) (tail $ skipOdds factorial) -- Recall that @powsOfAPlusN = [(a + n) ^ s | s <- [0 ..]]@, so this -- is @[(a + n) ^ (2 * s - 1) | s <- [1 ..]]@ (skipEvens powsOfAPlusN) fracs :: [a] fracs = zipWith (\sec pochh -> sum $ zipWith (\s p -> s * fromInteger p) sec pochh) (repeat second) pochhammers -- Infinite list of @T@ values in 4.8.5 formula, for every @s@ in -- @[0, 1, 2 ..]@. ts :: [a] ts = zipWith (\constant2 frac -> constant2 * (0.5 + frac)) constants2 fracs