-- | This module implements the classical operations on ideals used in Hallgren’s -- algorithm (including also classical versions of the quantum operations required). module Quipper.Algorithms.CL.RegulatorClassical where import Data.Maybe import Quipper.Libraries.Arith import Quipper.Libraries.FPReal import Quipper.Algorithms.CL.Auxiliary import Quipper.Algorithms.CL.Types -- ====================================================================== -- * Basic operations on ideals -- | @'unit_ideal' /bigD/ /l/@: the unit ideal /O/, for Δ = /bigD/, with the ideal’s coefficients given as /l/-bit integers. -- ([Jozsa 2003], Prop. 14.) unit_ideal :: CLIntP -> Ideal unit_ideal = forget_reduced . unit_idealRed -- | Like 'unit_ideal', but considered as a reduced ideal. unit_idealRed :: CLIntP -> IdealRed unit_idealRed bigD = assert (is_valid_bigD bigD) ("unit_idealRed: " ++ show bigD ++ " not valid discriminant") $ IdealRed bigD 1 (tau bigD (fromIntegral bigD) 1) -- | The integer constant /c/ of an ideal. -- ([Jozsa 2003], page 14 bottom: \"Since 4/a/ divides /b/[sup 2]-/D/ -- (cf. proposition 16) we introduce the integer /c/ = |/D/ − /b/[sup 2]|\/(4/a/)\") c_of_ideal :: Ideal -> CLInt c_of_ideal i@(Ideal bigD m l a b) = if (num `mod` denom == 0) then num `div` denom else error error_string where num = abs ((fromIntegral bigD) - b*b) denom = 4*a error_string = "rho of ideal [" ++ show i ++ "] produces non-integer c" -- | γ(/I/) = (/b/+√Δ)\/(2/a/) for a given ideal /I/. -- ([Jozsa 2003], Sec. 6.2.) gamma_of_ideal :: Ideal -> AlgNum gamma_of_ideal (Ideal bigD m l a b) = AlgNum a' b' bigD where a' = (fromIntegral b / fromIntegral (2*a)) :: CLRational b' = (fromIntegral 1 / fromIntegral (2*a)) :: CLRational -- Recall: @'AlgNum' u v bigD@ represents (u + v * √bigD). -- | The reduction function ρ on ideals. -- ([Jozsa 2003], Sec. 6.2.) rho :: Ideal -> Ideal rho ii@(Ideal bigD m l a b) = (Ideal bigD m'' l'' a' b') where -- With little algebra, one can derive these: m' = m * a l' = l * a' m'' = m' `div` (gcd m' l') l'' = l' `div` (gcd m' l') -- From [Jozsa 2003], p.15, equation (10) a' = c_of_ideal ii b' = tau bigD (-b) a' -- | The ρ[sup -1] function on ideals. Inverse to 'rho'. -- ([Jozsa 2003], Sec. 6.4.) rho_inv :: Ideal -> Ideal rho_inv (Ideal bigD m l a b) = (Ideal bigD m'' l'' a'' b'') where -- Create m and l m' = m * a l' = l * a'' -- Reduce them to have no common denominator m'' = m' `div` (gcd m' l') l'' = l' `div` (gcd m' l') -- Calculate b* (b' below) b' = tau bigD (-b) a -- Calculate new a and b a'' = ((fromIntegral bigD) - b'*b') `divchk` (4*a) b'' = tau bigD b' a'' -- | The ρ operation on reduced ideals. rho_red :: IdealRed -> IdealRed rho_red = to_reduced . rho . forget_reduced -- | The ρ[sup –1] operation on reduced ideals. rho_inv_red :: IdealRed -> IdealRed rho_inv_red = to_reduced . rho_inv . forget_reduced -- | The ρ operation on ideals-with-distance. rho_d :: IdDist -> IdDist rho_d (ii, del) = (rho ii, del + del_change) where gamma = gamma_of_ideal ii gamma_bar_by_gamma = (floating_of_AlgNum $ conjugate gamma) / (floating_of_AlgNum gamma) del_change = (log $ abs gamma_bar_by_gamma) / 2 -- | The ρ[sup –1] operation on ideals-with-distance. rho_inv_d :: IdDist -> IdDist rho_inv_d (ii, del) = (ii', del - del_change) where ii' = rho_inv ii gamma = gamma_of_ideal ii' gamma_bar_by_gamma = (floating_of_AlgNum $ conjugate gamma) / (floating_of_AlgNum gamma) del_change = (log $ abs gamma_bar_by_gamma) / 2 -- | The ρ operation on ideals-with-generator (i.e. pairs of an ideal /I/ and an 'AlgNum' /x/ such that /I/ is the principal ideal (/x/)). rho_num :: (Ideal,AlgNum) -> (Ideal,AlgNum) rho_num (ii, gen) = (rho ii, gen / (gamma_of_ideal ii)) -- | Apply ρ to an reduced-ideals-with-generator rho_red_num :: (IdealRed,AlgNum) -> (IdealRed,AlgNum) rho_red_num (ii, gen) = (rho_red ii, gen / (gamma_of_ideal $ forget_reduced ii)) -- ====================================================================== -- * Ideal reductions (bounded) -- | Reduce an ideal, by repeatedly applying ρ. reduce :: Ideal -> IdealRed reduce ii = to_reduced $ while (not . is_reduced) rho ii -- | Reduce an ideal within a bounded loop. Applies the ρ function -- until the ideal is reduced. Used in 'star' and 'fJN' algorithms. bounded_reduce :: IdDist -> IdDist bounded_reduce k@(ideal@(Ideal bigD m l a b),dist) = fst $ bounded_while condition bound func (k,0) where -- NOTE: some uncertainty regarding loop bound. bound = ceiling $ bound_log + 1 bound_log = (logBase 2 ((fromIntegral bound_a) / (sqrt $ fromIntegral $ bigD_of_Ideal $ ideal))) bound_a = 2 ^ (fromJust (intm_length $ a)) condition (k,itr) = not $ is_reduced $ fst k func (k,itr) = ((rho_d k),(itr+1)) -- | Apply a function (like ρ,ρ[sup -1],ρ[sup 2]) to an ideal, bounded -- by 3*ln(Δ)\/2*ln(2). Execute while satisfies condition function. bounded_step :: (IdDist -> Bool) -> (IdDist -> IdDist) -> IdDist -> IdDist bounded_step condition step_function ideal = -- Execute a bounded while loop bounded_while condition bound step_function ideal where bound = ceiling $ (3 * (log $ fromIntegral $ bigD_of_Ideal $ fst ideal)) / (2 * log 2) -- | Like 'bounded_step', but the condition is checked against delta of the -- current ideal. bounded_step_delta :: (CLReal -> Bool) -> (IdDist -> IdDist) -> IdDist -> IdDist bounded_step_delta condition step_function ideal = bounded_step new_condition step_function ideal where new_condition = (\k -> condition (delta k)) -- ====================================================================== -- * Products of ideals -- | The ordinary (not necessarily reduced) product of two reduced fractional ideals. -- -- /I/⋅/J/ of [Jozsa 2003], Sec 7.1, following the description -- given in Prop. 34. -- NOTE: assumes I, J reduced. Type should reflect this! dot :: IdDist -> IdDist -> IdDist dot i1@(Ideal bigD1 m1 l1 a1 b1, delta1) i2@(Ideal bigD2 m2 l2 a2 b2, delta2) = assert_reduced (fst i1) $ assert_reduced (fst i2) $ (Ideal bigD1 m l a3 b3, del) where sqrtd :: CLReal sqrtd = sqrt (fromIntegral bigD1) (k', u', v') = extended_euclid a1 a2 (k, x, w ) = extended_euclid k' ((b1 + b2) `divchk` 2) a3 = (a1 * a2) `divchk` (k * k) t1 = x * u' * a1 * b2 t2 = x * v' * a2 * b1 t3 = w*(b1 * b2 + (fromIntegral bigD1)) `divchk` 2 t = (t1 + t2 + t3) `divchk` k b3 = tau bigD1 t a3 m = k l = a3 del = ((delta i1) + (delta i2)) -- | The dot-square /I/⋅/I/ of an ideal-with-distance /I/. dot' :: IdDist -> IdDist dot' i1@(Ideal bigD1 m1 l1 a1 b1, delta1) = assert_reduced (fst i1) $ (Ideal bigD1 m l a3 b3, del) where sqrtd :: CLReal sqrtd = sqrt (fromIntegral bigD1) (k, u, w) = extended_euclid (abs a1) (abs b1) a3 = (a1 * a1) `divchk` (k * k) t1 = u * a1 * b1 t3 = w*(b1 * b1 + (fromIntegral bigD1)) `divchk` 2 t = (t1 + t3) `divchk` k b3 = tau bigD1 t a3 m = k -- l = a3 l = 1 del = ((delta i1) + (delta i1)) -- | The star-product of two ideals-with-distance. -- -- This is /I/*/J/ of [Jozsa 2003], Sec. 7.1, defined as the first reduced -- ideal-with-distance following /I/⋅/J/. -- NOTE: assumes I, J reduced. Type should reflect this! star :: IdDist -> IdDist -> IdDist star i j = if (delta k1 <= delta i_dot_j) then bounded_step_delta (<= delta i_dot_j) rho_d k1 else rho_d $ bounded_step_delta (>= delta i_dot_j) rho_inv_d k1 where i_dot_j = i `dot` j -- FIX: Over bound (infinite loop) in reducing the following: -- <m:1 l:3 a:3 b:2 bigD:28 del:1.1>*<m:1 l:3 a:3 b:4 bigD:28 del:1.6> k1 = bounded_reduce i_dot_j -- ====================================================================== -- * The function f[sub /N/], and variants -- | Compute the expression i\/N + j\/L. compute_injl :: (Integral int) => CLInt -> int -> CLInt -> int -> CLReal compute_injl i nn j ll = (fromIntegral i)/(fromIntegral nn) + (fromIntegral j)/(fromIntegral ll) -- | @'fN' /i/ /j/ /n/ /l/ Δ@: find the minimal ideal-with-distance (/J/,δ[sub /J/]) such that δ[sub /J/] > /x/, where /x/ = /i/\//N/ + /j/\//L/, where /N/ = 2[sup /n/], /L/ = 2[sup /l/]. Return (/i/,/J/,δ[sub /J/]–/x/). Work under the assumption that /R/ < 2[sup /s/]. -- -- This is the function /h/ of [Jozsa 2003, Section 9], discretized with precision 1\//N/ = 2[sup −/n/], -- and perturbed by the jitter parameter /j/\//L/. fN :: (Integral int) => CLInt -> CLInt -> int -> int -> CLIntP -> (Ideal, CLInt) fN i j nn ll bigD = let ((ideal_J, _), diff) = fN_d i j nn ll bigD in (ideal_J, diff) -- | Like 'fN', but returning an ideal-with-distance not just an ideal. fN_d :: (Integral int) => CLInt -> CLInt -> int -> int -> CLIntP -> (IdDist, CLInt) fN_d i j nn ll bigD = (j_star_19, floor $ (fromIntegral nn) * (toRational $ injl - (snd j_star_19))) where -- Expression "i/N + j/L" used repeatedly injl = compute_injl i nn j ll -- Generate J1 j1 = rho_d $ rho_d $ (unit_ideal bigD, 0) -- Generate Jk's (make an infinite list and take only what is needed) jks = takeWhile (\jk -> (delta jk) <= injl) $ bounded_iterate bound_jks (\jk -> jk `star` jk) j1 where -- Bound for jk generation max_i = 2^(fromJust (intm_length $ i)) bound_jks = ceiling $ (log $ fromIntegral max_i) / ((fromIntegral nn) * (delta j1)) -- Apply all Jk's to J* using '*' in reverse (remember that the last -- element is J* itself) j_star_14 = foldr1 applyJkIfConditionHolds jks -- Apply Jk to J* if a condition holds. applyJkIfConditionHolds :: IdDist -> IdDist -> IdDist applyJkIfConditionHolds jk jstar = if (delta (jstar `star` jk) <= injl) then jstar `star` jk else jstar -- Go forward one step at a time as much as needed j_star_17 = bounded_step_delta (< injl) (rho_d.rho_d) j_star_14 -- Go back one step if needed j_star_19 = if (delta (rho_inv_d j_star_17) >= injl) then rho_inv_d j_star_17 else j_star_17 -- | Analogue of 'fN', working within the cycle determined by a given ideal /J/. -- ([Hallgren 2006], Section 5.) fJN :: IdDist -> CLInt -> CLInt -> CLInt -> CLInt -> CLIntP -> (Ideal, CLInt) fJN ideal_J i j nn ll bigD = let ((ideal_J', _), diff) = fJN_d ideal_J i j nn ll bigD in (ideal_J', diff) -- | Like 'fJN', but returning an ideal-with-distance not just an ideal. fJN_d :: IdDist -> CLInt -> CLInt -> CLInt -> CLInt -> CLIntP -> (IdDist, CLInt) fJN_d ideal_J i j nn ll bigD = (ideal_KFinal, floor $ (fromIntegral nn) * (injl - (delta ideal_KFinal))) where -- Expression "i/N + j/L" injl = compute_injl i nn j ll -- Generate I and K ideal_I = fst (fN_d i j nn ll bigD) ideal_K = bounded_reduce (ideal_I `dot` ideal_J) -- Step forward/backward as much as needed ideal_KFinal = if (delta ideal_K <= injl) then bounded_step_delta (< injl) (rho_d) ideal_K else rho_d $ bounded_step_delta (>= injl) (rho_inv_d) ideal_K -- ====================================================================== -- * Classical period-finding -- $ Functions for classically finding the regulator and fundamental unit of a field using the period of /f_N/. -- ====================================================================== -- ** Auxiliary functions -- | Find the order of an endofunction on an argument. That is, @'order' /f/ /x/@ returns the first /n/ > 0 such that /f/[sup /n/]\(/x/) = /x/. -- -- Method: simple brute-force search/comparison. order :: (Eq a) => (a -> a) -> a -> Int order = order_with_projection id -- | Given a function /p/, an endofunction /f/, and an argument /x/, returns the first /n/ > 0 such that /p/(/f/[sup /n/]\(/x/)) = /p/(/x/). -- -- Method: simple brute-force search/comparison. order_with_projection :: (Eq b) => (a -> b) -> (a -> a) -> a -> Int order_with_projection p f x = 1 + (length $ takeWhile (\y -> p y /= p x) (tail $ iterate f x)) -- | Given a function /p/, an endofunction /f/, and an argument /x/, return /f/[sup /n/]\(/x/), for the first /n/ > 0 such that /p/(/f/[sup /n/]\(/x/)) = /p/(/x/). -- -- Method: simple brute-force search/comparison. first_return_with_projection :: (Eq b) => (a -> b) -> (a -> a) -> a -> a first_return_with_projection p f x = head $ dropWhile (\y -> p y /= p x) $ tail $ iterate f x -- | Given a bound /b/, a function /p/, an endofunction /f/, and an argument /x/, return /f/[sup /n/]\(/x/), for the first /n/ > 0 such that /p/(/f/[sup /n/]\(/x/)) = /p/(/x/), if there exists such an /n/ ≤ /b/. -- -- Method: simple brute-force search/comparison. first_return_with_proj_bdd :: (Eq b) => Int -> (a -> b) -> (a -> a) -> a -> Maybe a first_return_with_proj_bdd b p f x = listToMaybe $ dropWhile (\y -> p y /= p x) $ take b $ tail $ iterate f x -- | Find the period of a function on integers, assuming that it is periodic and injective on its period. That is, @'period' /f/@ returns the first /n/ > 0 such that /f/(/n/) = /f/(0). Method: simple brute-force search/comparison. period :: (Eq a, Integral int) => (int -> a) -> int period f = minimum [ n | n <- [1..], f n == f 0 ] -- ====================================================================== -- ** Haskell native arithmetic -- $ The functions of this section use Haskell’s native integer and floating computation. -- | Find the regulator /R/ = log ε[sub 0] of a field, given the discriminant Δ, -- by finding (classically) the order of ρ. -- -- Uses 'IdDist' and 'rho_d'. regulator :: CLIntP -> FPReal regulator bigD = snd $ head $ dropWhile (\(ii,_) -> ii /= calO) $ tail $ iterate rho_d $ (calO,0) where calO = unit_ideal bigD -- | Find the fundamental unit ε[sub 0] of a field, given the discriminant Δ, -- by finding (classically) the order of ρ. -- -- Uses '(Ideal,Number)' and 'rho_num'. fundamental_unit :: CLIntP -> AlgNum fundamental_unit bigD = maximum [eps, -eps, 1/eps, -1/eps] where eps = snd $ first_return_with_projection fst rho_num (calO,1) calO = unit_ideal bigD -- | Find the fundamental solution of Pell’s equation, given /d/. -- -- Solutions of Pell’s equations are integer pairs (/x/,/y/) such that -- /x/,/y/ > 0, and (/x/ + /y/√d)(/x/ – /y/√d) = 1. -- -- In this situation, (/x/ + /y/√d) is a unit of the algebraic integers -- of /K/, and is >1, so we simply search the powers of ε[sub 0] for a -- unit of the desired form. fundamental_solution :: CLIntP -> (Integer,Integer) fundamental_solution d = head pell_solutions where bigD = bigD_of_d d eps0 = fundamental_unit bigD pell_solutions = [ (round x, round y) | n <- [1..], let eps@(AlgNum a b _) = eps0^n, a >= 0, b >= 0, let x = a, let y = if bigD == d then b else 2*b, is_int x, is_int y, eps * (conjugate eps) == 1] -- ====================================================================== -- ** Fixed-precision arithmetic -- $ The functions of this section perform period-finding using fixed-precision arithmetic. -- This should parallel closely (though at present not exactly, due to the implementations -- of floating-point operations) the quantum circuit implementations, and hence allows -- testing of whether the chosen precisions are accurate. -- | Find the regulator /R/ = log ε[sub 0] of a field, given the discriminant Δ, -- by finding (classically) the order of ρ -- using fixed-precision arithmetic: 'fix_sizes_Ideal' for 'Ideal's, -- and given an assumed bound /b/ on log[sub 2] /R/. -- -- Uses 'IdDist' and 'rho_d'. regulator_fixed_prec :: Int -> CLIntP -> Maybe FPReal regulator_fixed_prec b bigD = fmap snd (first_return_with_proj_bdd b' fst rho_d (calO,zero)) where calO = fix_sizes_Ideal $ unit_ideal bigD n = n_of_bigD bigD -- S = 2RN, so this /i/ gives an assumed bound on log_2 S: i = 1 + b + n -- Following precisions are as used in 'approximate_regulator_circuit', 'q_fN': q = 2 + 2 * i l = 4 + q + n p = precision_for_fN bigD n l zero = fprealx (-p) (intm (q - n + p) 0) -- δ(I, ρ^2 (I)) ≥ √2 for any I, so the order of ρ is at most 2R / (√2 / 2): b' = ceiling $ 2 * sqrt(2) * (fromIntegral 2^b) {- Waiting for 'AlgNum' to be rebased using 'IntM' instead of 'Integer'. -- | Find the fundamental unit ε[sub 0] of a field, given the discriminant Δ, -- by finding (classically) the order of ρ, -- using fixed-precision arithmetic: 'fix_sizes_Ideal' for 'Ideal's, -- and a given /l/ for the 'AlgNum's generating them. -- -- Uses '(Ideal,Number)' and 'rho_num'. fundamental_unit_with_fixed :: Int -> CLIntP -> AlgNum fundamental_unit_with_fixed l bigD = maximum [eps, -eps, 1/eps, -1/eps] where eps = snd $ head $ dropWhile (\(ii,_) -> ii /= calO) $ tail $ iterate rho_num $ (calO,one) calO = fix_sizes_Ideal $ unit_ideal bigD one = intm_with_length (Just l) 1 -}