-- | Partitions of integers. -- Integer partitions are nonincreasing sequences of positive integers. -- -- See: -- -- * Donald E. Knuth: The Art of Computer Programming, vol 4, pre-fascicle 3B. -- -- * -- -- For example the partition -- -- > Partition [8,6,3,3,1] -- -- can be represented by the (English notation) Ferrers diagram: -- -- <> -- {-# LANGUAGE CPP, BangPatterns #-} module Math.Combinat.Partitions.Integer where -------------------------------------------------------------------------------- import Data.List -- import Data.Map (Map) -- import qualified Data.Map as Map import Math.Combinat.Helper import Math.Combinat.ASCII as ASCII import Math.Combinat.Numbers (factorial,binomial,multinomial) -------------------------------------------------------------------------------- -- * Type and basic stuff -- | A partition of an integer. The additional invariant enforced here is that partitions -- are monotone decreasing sequences of /positive/ integers. The @Ord@ instance is lexicographical. newtype Partition = Partition [Int] deriving (Eq,Ord,Show,Read) --------------------------------------------------------------------------------- class HasNumberOfParts p where numberOfParts :: p -> Int instance HasNumberOfParts Partition where numberOfParts (Partition p) = length p --------------------------------------------------------------------------------- -- | Sorts the input, and cuts the nonpositive elements. mkPartition :: [Int] -> Partition mkPartition xs = Partition $ sortBy (reverseCompare) $ filter (>0) xs -- | Assumes that the input is decreasing. toPartitionUnsafe :: [Int] -> Partition toPartitionUnsafe = Partition -- | Checks whether the input is an integer partition. See the note at 'isPartition'! toPartition :: [Int] -> Partition toPartition xs = if isPartition xs then toPartitionUnsafe xs else error "toPartition: not a partition" -- | This returns @True@ if the input is non-increasing sequence of -- /positive/ integers (possibly empty); @False@ otherwise. -- isPartition :: [Int] -> Bool isPartition [] = True isPartition [x] = x > 0 isPartition (x:xs@(y:_)) = (x >= y) && isPartition xs fromPartition :: Partition -> [Int] fromPartition (Partition part) = part -- | The first element of the sequence. height :: Partition -> Int height (Partition part) = case part of (p:_) -> p [] -> 0 -- | The length of the sequence (that is, the number of parts). width :: Partition -> Int width (Partition part) = length part heightWidth :: Partition -> (Int,Int) heightWidth part = (height part, width part) -- | The weight of the partition -- (that is, the sum of the corresponding sequence). weight :: Partition -> Int weight (Partition part) = sum' part -- | The dual (or conjugate) partition. dualPartition :: Partition -> Partition dualPartition (Partition part) = Partition (_dualPartition part) data Pair = Pair !Int !Int _dualPartition :: [Int] -> [Int] _dualPartition [] = [] _dualPartition xs = go 0 (diffSequence xs) [] where go !i (d:ds) acc = go (i+1) ds (d:acc) go n [] acc = finish n acc finish !j (k:ks) = replicate k j ++ finish (j-1) ks finish _ [] = [] {- -- more variations: _dualPartition_b :: [Int] -> [Int] _dualPartition_b [] = [] _dualPartition_b xs = go 1 (diffSequence xs) [] where go !i (d:ds) acc = go (i+1) ds ((d,i):acc) go _ [] acc = concatMap (\(d,i) -> replicate d i) acc _dualPartition_c :: [Int] -> [Int] _dualPartition_c [] = [] _dualPartition_c xs = reverse $ concat $ zipWith f [1..] (diffSequence xs) where f _ 0 = [] f k d = replicate d k -} -- | A simpler, but bit slower (about twice?) implementation of dual partition _dualPartitionNaive :: [Int] -> [Int] _dualPartitionNaive [] = [] _dualPartitionNaive xs@(k:_) = [ length $ filter (>=i) xs | i <- [1..k] ] -- | From a sequence @[a1,a2,..,an]@ computes the sequence of differences -- @[a1-a2,a2-a3,...,an-0]@ diffSequence :: [Int] -> [Int] diffSequence = go where go (x:ys@(y:_)) = (x-y) : go ys go [x] = [x] go [] = [] -- | Example: -- -- > elements (toPartition [5,4,1]) == -- > [ (1,1), (1,2), (1,3), (1,4), (1,5) -- > , (2,1), (2,2), (2,3), (2,4) -- > , (3,1) -- > ] -- elements :: Partition -> [(Int,Int)] elements (Partition part) = _elements part _elements :: [Int] -> [(Int,Int)] _elements shape = [ (i,j) | (i,l) <- zip [1..] shape, j<-[1..l] ] --------------------------------------------------------------------------------- -- * Exponential form -- | We convert a partition to exponential form. -- @(i,e)@ mean @(i^e)@; for example @[(1,4),(2,3)]@ corresponds to @(1^4)(2^3) = [2,2,2,1,1,1,1]@. Another example: -- -- > toExponentialForm (Partition [5,5,3,2,2,2,2,1,1]) == [(1,2),(2,4),(3,1),(5,2)] -- toExponentialForm :: Partition -> [(Int,Int)] toExponentialForm = _toExponentialForm . fromPartition _toExponentialForm :: [Int] -> [(Int,Int)] _toExponentialForm = reverse . map (\xs -> (head xs,length xs)) . group fromExponentialFrom :: [(Int,Int)] -> Partition fromExponentialFrom = Partition . sortBy reverseCompare . go where go ((j,e):rest) = replicate e j ++ go rest go [] = [] --------------------------------------------------------------------------------- -- * Automorphisms -- | Computes the number of \"automorphisms\" of a given integer partition. countAutomorphisms :: Partition -> Integer countAutomorphisms = _countAutomorphisms . fromPartition _countAutomorphisms :: [Int] -> Integer _countAutomorphisms = multinomial . map length . group --------------------------------------------------------------------------------- -- * Generating partitions -- | Partitions of @d@. partitions :: Int -> [Partition] partitions = map Partition . _partitions -- | Partitions of @d@, as lists _partitions :: Int -> [[Int]] _partitions d = go d d where go _ 0 = [[]] go !h !n = [ a:as | a<-[1..min n h], as <- go a (n-a) ] countPartitions :: Int -> Integer countPartitions d = countPartitions' (d,d) d -- | All integer partitions up to a given degree (that is, all integer partitions whose sum is less or equal to @d@) allPartitions :: Int -> [Partition] allPartitions d = concat [ partitions i | i <- [0..d] ] -- | All integer partitions up to a given degree (that is, all integer partitions whose sum is less or equal to @d@), -- grouped by weight allPartitionsGrouped :: Int -> [[Partition]] allPartitionsGrouped d = [ partitions i | i <- [0..d] ] -- | All integer partitions fitting into a given rectangle. allPartitions' :: (Int,Int) -- ^ (height,width) -> [Partition] allPartitions' (h,w) = concat [ partitions' (h,w) i | i <- [0..d] ] where d = h*w -- | All integer partitions fitting into a given rectangle, grouped by weight. allPartitionsGrouped' :: (Int,Int) -- ^ (height,width) -> [[Partition]] allPartitionsGrouped' (h,w) = [ partitions' (h,w) i | i <- [0..d] ] where d = h*w -- | # = \\binom { h+w } { h } countAllPartitions' :: (Int,Int) -> Integer countAllPartitions' (h,w) = binomial (h+w) (min h w) --sum [ countPartitions' (h,w) i | i <- [0..d] ] where d = h*w countAllPartitions :: Int -> Integer countAllPartitions d = sum' [ countPartitions i | i <- [0..d] ] -- | Integer partitions of @d@, fitting into a given rectangle, as lists. _partitions' :: (Int,Int) -- ^ (height,width) -> Int -- ^ d -> [[Int]] _partitions' _ 0 = [[]] _partitions' ( 0 , _) d = if d==0 then [[]] else [] _partitions' ( _ , 0) d = if d==0 then [[]] else [] _partitions' (!h ,!w) d = [ i:xs | i <- [1..min d h] , xs <- _partitions' (i,w-1) (d-i) ] -- | Partitions of d, fitting into a given rectangle. The order is again lexicographic. partitions' :: (Int,Int) -- ^ (height,width) -> Int -- ^ d -> [Partition] partitions' hw d = map toPartitionUnsafe $ _partitions' hw d countPartitions' :: (Int,Int) -> Int -> Integer countPartitions' _ 0 = 1 countPartitions' (0,_) d = if d==0 then 1 else 0 countPartitions' (_,0) d = if d==0 then 1 else 0 countPartitions' (h,w) d = sum [ countPartitions' (i,w-1) (d-i) | i <- [1..min d h] ] --------------------------------------------------------------------------------- -- * Dominance order -- | @q \`dominates\` p@ returns @True@ if @q >= p@ in the dominance order of partitions -- (this is partial ordering on the set of partitions of @n@). -- -- See -- dominates :: Partition -> Partition -> Bool dominates (Partition qs) (Partition ps) = and $ zipWith (>=) (sums (qs ++ repeat 0)) (sums ps) where sums = scanl (+) 0 -- | Lists all partitions of the same weight as @lambda@ and also dominated by @lambda@ -- (that is, all partial sums are less or equal): -- -- > dominatedPartitions lam == [ mu | mu <- partitions (weight lam), lam `dominates` mu ] -- dominatedPartitions :: Partition -> [Partition] dominatedPartitions (Partition lambda) = map Partition (_dominatedPartitions lambda) _dominatedPartitions :: [Int] -> [[Int]] _dominatedPartitions [] = [[]] _dominatedPartitions lambda = go (head lambda) w dsums 0 where n = length lambda w = sum lambda dsums = scanl1 (+) (lambda ++ repeat 0) go _ 0 _ _ = [[]] go !h !w (!d:ds) !e | w > 0 = [ (a:as) | a <- [1..min h (d-e)] , as <- go a (w-a) ds (e+a) ] | w == 0 = [[]] | w < 0 = error "_dominatedPartitions: fatal error; shouldn't happen" -- | Lists all partitions of the sime weight as @mu@ and also dominating @mu@ -- (that is, all partial sums are greater or equal): -- -- > dominatingPartitions mu == [ lam | lam <- partitions (weight mu), lam `dominates` mu ] -- dominatingPartitions :: Partition -> [Partition] dominatingPartitions (Partition mu) = map Partition (_dominatingPartitions mu) _dominatingPartitions :: [Int] -> [[Int]] _dominatingPartitions [] = [[]] _dominatingPartitions mu = go w w dsums 0 where n = length mu w = sum mu dsums = scanl1 (+) (mu ++ repeat 0) go _ 0 _ _ = [[]] go !h !w (!d:ds) !e | w > 0 = [ (a:as) | a <- [max 0 (d-e)..min h w] , as <- go a (w-a) ds (e+a) ] | w == 0 = [[]] | w < 0 = error "_dominatingPartitions: fatal error; shouldn't happen" -------------------------------------------------------------------------------- -- * Partitions with given number of parts -- | Lists partitions of @n@ into @k@ parts. -- -- > sort (partitionsWithKParts k n) == sort [ p | p <- partitions n , numberOfParts p == k ] -- -- Naive recursive algorithm. -- partitionsWithKParts :: Int -- ^ @k@ = number of parts -> Int -- ^ @n@ = the integer we partition -> [Partition] partitionsWithKParts k n = map Partition $ go n k n where {- h = max height k = number of parts n = integer -} go !h !k !n | k < 0 = [] | k == 0 = if h>=0 && n==0 then [[] ] else [] | k == 1 = if h>=n && n>=1 then [[n]] else [] | otherwise = [ a:p | a <- [1..(min h (n-k+1))] , p <- go a (k-1) (n-a) ] countPartitionsWithKParts :: Int -- ^ @k@ = number of parts -> Int -- ^ @n@ = the integer we partition -> Integer countPartitionsWithKParts k n = go n k n where go !h !k !n | k < 0 = 0 | k == 0 = if h>=0 && n==0 then 1 else 0 | k == 1 = if h>=n && n>=1 then 1 else 0 | otherwise = sum' [ go a (k-1) (n-a) | a<-[1..(min h (n-k+1))] ] -------------------------------------------------------------------------------- -- * Partitions with only odd\/distinct parts -- | Partitions of @n@ with only odd parts partitionsWithOddParts :: Int -> [Partition] partitionsWithOddParts d = map Partition (go d d) where go _ 0 = [[]] go !h !n = [ a:as | a<-[1,3..min n h], as <- go a (n-a) ] {- -- | Partitions of @n@ with only even parts -- -- Note: this is not very interesting, it's just @(map.map) (2*) $ _partitions (div n 2)@ -- partitionsWithEvenParts :: Int -> [Partition] partitionsWithEvenParts d = map Partition (go d d) where go _ 0 = [[]] go !h !n = [ a:as | a<-[2,4..min n h], as <- go a (n-a) ] -} -- | Partitions of @n@ with distinct parts. -- -- Note: -- -- > length (partitionsWithDistinctParts d) == length (partitionsWithOddParts d) -- partitionsWithDistinctParts :: Int -> [Partition] partitionsWithDistinctParts d = map Partition (go d d) where go _ 0 = [[]] go !h !n = [ a:as | a<-[1..min n h], as <- go (a-1) (n-a) ] -------------------------------------------------------------------------------- -- * Sub-partitions of a given partition -- | Returns @True@ of the first partition is a subpartition (that is, fit inside) of the second. -- This includes equality isSubPartitionOf :: Partition -> Partition -> Bool isSubPartitionOf (Partition ps) (Partition qs) = and $ zipWith (<=) ps (qs ++ repeat 0) -- | Sub-partitions of a given partition with the given weight: -- -- > sort (subPartitions d q) == sort [ p | p <- partitions d, isSubPartitionOf p q ] -- subPartitions :: Int -> Partition -> [Partition] subPartitions d (Partition ps) = map Partition (_subPartitions d ps) _subPartitions :: Int -> [Int] -> [[Int]] _subPartitions d big | null big = if d==0 then [[]] else [] | d > sum' big = [] | d < 0 = [] | otherwise = go d (head big) big where go :: Int -> Int -> [Int] -> [[Int]] go !k !h [] = if k==0 then [[]] else [] go !k !h (b:bs) | k<0 || h<0 = [] | k==0 = [[]] | h==0 = [] | otherwise = [ this:rest | this <- [1..min h b] , rest <- go (k-this) this bs ] ---------------------------------------- -- | All sub-partitions of a given partition allSubPartitions :: Partition -> [Partition] allSubPartitions (Partition ps) = map Partition (_allSubPartitions ps) _allSubPartitions :: [Int] -> [[Int]] _allSubPartitions big | null big = [[]] | otherwise = go (head big) big where go _ [] = [[]] go !h (b:bs) | h==0 = [] | otherwise = [] : [ this:rest | this <- [1..min h b] , rest <- go this bs ] -------------------------------------------------------------------------------- -- * The Pieri rule -- | The Pieri rule computes @s[lambda]*h[n]@ as a sum of @s[mu]@-s (each with coefficient 1). -- -- See for example -- pieriRule :: Partition -> Int -> [Partition] pieriRule (Partition lambda) n = map Partition (_pieriRule lambda n) where -- | We assume here that @lambda@ is a partition (non-increasing sequence of /positive/ integers)! _pieriRule :: [Int] -> Int -> [[Int]] _pieriRule lambda n | n == 0 = [lambda] | n < 0 = [] | otherwise = go n diffs dsums (lambda++[0]) where diffs = n : diffSequence lambda -- maximum we can add to a given row dsums = reverse $ scanl1 (+) (reverse diffs) -- partial sums of remaining total we can add go !k (d:ds) (p:ps@(q:_)) (l:ls) | k > p = [] | otherwise = [ h:tl | a <- [ max 0 (k-q) .. min d k ] , let h = l+a , tl <- go (k-a) ds ps ls ] go !k [d] _ [l] = if k <= d then if l+k>0 then [[l+k]] else [[]] else [] go !k [] _ _ = if k==0 then [[]] else [] -- | The dual Pieri rule computes @s[lambda]*e[n]@ as a sum of @s[mu]@-s (each with coefficient 1) dualPieriRule :: Partition -> Int -> [Partition] dualPieriRule lam n = map dualPartition $ pieriRule (dualPartition lam) n {- -- | Computes the Schur expansion of @h[n1]*h[n2]*h[n3]*...*h[nk]@ via iterating the Pieri rule iteratedPieriRule :: Num coeff => [Int] -> Map Partition coeff iteratedPieriRule = iteratedPieriRule' (Partition []) -- | Iterating the Pieri rule, we can compute the Schur expansion of -- @h[lambda]*h[n1]*h[n2]*h[n3]*...*h[nk]@ iteratedPieriRule' :: Num coeff => Partition -> [Int] -> Map Partition coeff iteratedPieriRule' plambda ns = iteratedPieriRule'' (plambda,1) ns {-# SPECIALIZE iteratedPieriRule'' :: (Partition,Int ) -> [Int] -> Map Partition Int #-} {-# SPECIALIZE iteratedPieriRule'' :: (Partition,Integer) -> [Int] -> Map Partition Integer #-} iteratedPieriRule'' :: Num coeff => (Partition,coeff) -> [Int] -> Map Partition coeff iteratedPieriRule'' (plambda,coeff0) ns = worker (Map.singleton plambda coeff0) ns where worker old [] = old worker old (n:ns) = worker new ns where stuff = [ (coeff, pieriRule lam n) | (lam,coeff) <- Map.toList old ] new = foldl' f Map.empty stuff f t0 (c,ps) = foldl' (\t p -> Map.insertWith (+) p c t) t0 ps -} -------------------------------------------------------------------------------- -- * ASCII Ferrers diagrams -- | Which orientation to draw the Ferrers diagrams. -- For example, the partition [5,4,1] corrsponds to: -- -- In standard English notation: -- -- > @@@@@ -- > @@@@ -- > @ -- -- -- In English notation rotated by 90 degrees counter-clockwise: -- -- > @ -- > @@ -- > @@ -- > @@ -- > @@@ -- -- -- And in French notation: -- -- -- > @ -- > @@@@ -- > @@@@@ -- -- data PartitionConvention = EnglishNotation -- ^ English notation | EnglishNotationCCW -- ^ English notation rotated by 90 degrees counterclockwise | FrenchNotation -- ^ French notation (mirror of English notation to the x axis) deriving (Eq,Show) -- | Synonym for @asciiFerrersDiagram\' EnglishNotation \'\@\'@ -- -- Try for example: -- -- > autoTabulate RowMajor (Right 8) (map asciiFerrersDiagram $ partitions 9) -- asciiFerrersDiagram :: Partition -> ASCII asciiFerrersDiagram = asciiFerrersDiagram' EnglishNotation '@' asciiFerrersDiagram' :: PartitionConvention -> Char -> Partition -> ASCII asciiFerrersDiagram' conv ch part = ASCII.asciiFromLines (map f ys) where f n = replicate n ch ys = case conv of EnglishNotation -> fromPartition part EnglishNotationCCW -> reverse $ fromPartition $ dualPartition part FrenchNotation -> reverse $ fromPartition $ part instance DrawASCII Partition where ascii = asciiFerrersDiagram -------------------------------------------------------------------------------- {- -- * Tests -- we need some custom types for quickcheck generating not too large partitions... newtype PartitionWeight = PartitionWeight Int data PartitionWeightPair = PartitionWeightPair Int Int data PartitionPair = PartitionPair Partition Int prop_partitions_in_bigbox :: PartitionWeight -> Bool prop_partitions_in_bigbox (PartitionWeight n) = (partitions n == partitions' (n,n) n) prop_kparts :: PartitionWeightPair -> Bool prop_kparts (PartitionWeightPair n k) = (partitionsWithKParts k n == [ mu | mu <- partitions n, numberOfParts mu == k ]) prop_odd_partitions :: PartitionWeight -> Bool prop_odd_partitions (PartitionWeight n) = (partitionsWithOddParts n == [ mu | mu <- partitions n, and (map odd (fromPartition mu)) ]) prop_distinct_partitions :: PartitionWeight -> Bool prop_distinct_partitions (PartitionWeight n) = (partitionsWithDistinctParts n == [ mu | mu <- partitions n, let xs = fromPartition mu, xs == nub xs ]) prop_subparts :: PartitionPair -> Bool prop_subparts (PartitionPair lam d) = (subPartitions d lam) == sort [ p | p <- partitions d, isSubPartitionOf p lam ] prop_dual_dual :: Partition -> Bool prop_dual_dual lam = (lam == dualPartition (dualPartition lam)) prop_dominated_list :: Partition -> Bool prop_dominated_list lam = (dominatedPartitions lam == [ mu | mu <- partitions (weight lam), lam `dominates` mu ]) prop_dominating_list :: Partition -> Bool prop_dominating_list mu = (dominatingPartitions mu == [ lam | lam <- partitions (weight mu ), lam `dominates` mu ]) -} --------------------------------------------------------------------------------