Theory Power_Sum_Polynomials

section ‹Power sum polynomials›
(*
  File:     Power_Sum_Polynomials.thy
  Author:   Manuel Eberl, TU München
*)
theory Power_Sum_Polynomials
imports
  "Symmetric_Polynomials.Symmetric_Polynomials"
  "HOL-Computational_Algebra.Field_as_Ring"
  Power_Sum_Polynomials_Library
begin

subsection ‹Definition›

text ‹
  For $n$ indeterminates $X_1,\ldots,X_n$, we define the $k$-th power sum polynomial as
  \[p_k(X_1, \ldots, X_n) = X_1^k + \ldots + X_n^k\ .\]
›
lift_definition powsum_mpoly_aux :: "nat set  nat  (nat 0 nat) 0 'a :: {semiring_1,zero_neq_one}" is
  "λX k mon. if infinite X  k = 0  mon  0 then 0
             else if k = 0  mon = 0 then of_nat (card X)
             else if finite X  (xX. mon = Poly_Mapping.single x k) then 1 else 0"
  by auto

lemma lookup_powsum_mpoly_aux:
  "Poly_Mapping.lookup (powsum_mpoly_aux X k) mon =
     (if infinite X  k = 0  mon  0 then 0
             else if k = 0  mon = 0 then of_nat (card X)
             else if finite X  (xX. mon = Poly_Mapping.single x k) then 1 else 0)"
  by transfer' simp

lemma lookup_sym_mpoly_aux_monom_singleton [simp]:
  assumes "finite X" "x  X" "k > 0"
  shows   "Poly_Mapping.lookup (powsum_mpoly_aux X k) (Poly_Mapping.single x k) = 1"
  using assms by (auto simp: lookup_powsum_mpoly_aux)

lemma lookup_sym_mpoly_aux_monom_singleton':
  assumes "finite X" "k > 0"
  shows   "Poly_Mapping.lookup (powsum_mpoly_aux X k) (Poly_Mapping.single x k) = (if x  X then 1 else 0)"
  using assms by (auto simp: lookup_powsum_mpoly_aux)

lemma keys_powsum_mpoly_aux: "m  keys (powsum_mpoly_aux A k)  keys m  A"
  by transfer' (auto split: if_splits simp: keys_monom_of_set)


lift_definition powsum_mpoly :: "nat set  nat  'a :: {semiring_1,zero_neq_one} mpoly" is
  "powsum_mpoly_aux" .

lemma vars_powsum_mpoly_subset: "vars (powsum_mpoly A k)  A"
  using keys_powsum_mpoly_aux by (auto simp: vars_def powsum_mpoly.rep_eq)

lemma powsum_mpoly_infinite: "¬finite A  powsum_mpoly A k = 0"
  by (transfer, transfer) auto

lemma coeff_powsum_mpoly:
  "MPoly_Type.coeff (powsum_mpoly X k) mon =
     (if infinite X  k = 0  mon  0 then 0
             else if k = 0  mon = 0 then of_nat (card X)
             else if finite X  (xX. mon = Poly_Mapping.single x k) then 1 else 0)"
  by transfer' (simp add: lookup_powsum_mpoly_aux)

lemma coeff_powsum_mpoly_0_right:
  "MPoly_Type.coeff (powsum_mpoly X 0) mon = (if mon = 0 then of_nat (card X) else 0)"
  by transfer' (auto simp add: lookup_powsum_mpoly_aux)

lemma coeff_powsum_mpoly_singleton:
  assumes "finite X" "k > 0"
  shows   "MPoly_Type.coeff (powsum_mpoly X k) (Poly_Mapping.single x k) = (if x  X then 1 else 0)"
  using assms by transfer' (simp add: lookup_powsum_mpoly_aux)

lemma coeff_powsum_mpoly_singleton_eq_1 [simp]:
  assumes "finite X" "x  X" "k > 0"
  shows   "MPoly_Type.coeff (powsum_mpoly X k) (Poly_Mapping.single x k) = 1"
  using assms by (simp add: coeff_powsum_mpoly_singleton)

lemma coeff_powsum_mpoly_singleton_eq_0 [simp]:
  assumes "finite X" "x  X" "k > 0"
  shows   "MPoly_Type.coeff (powsum_mpoly X k) (Poly_Mapping.single x k) = 0"
  using assms by (simp add: coeff_powsum_mpoly_singleton)

lemma powsum_mpoly_0 [simp]: "powsum_mpoly X 0 = of_nat (card X)"
  by (intro mpoly_eqI ext) (auto simp: coeff_powsum_mpoly_0_right of_nat_mpoly_eq mpoly_coeff_Const)

lemma powsum_mpoly_empty [simp]: "powsum_mpoly {} k = 0"
  by (intro mpoly_eqI) (auto simp: coeff_powsum_mpoly)

lemma powsum_mpoly_altdef: "powsum_mpoly X k = (xX. monom (Poly_Mapping.single x k) 1)"
proof (cases "finite X")
  case [simp]: True
  show ?thesis
  proof (cases "k = 0")
    case True
    thus ?thesis by auto
  next
    case False
    show ?thesis
    proof (intro mpoly_eqI, goal_cases)
      case (1 mon)
      show ?case using False
        by (cases "xX. mon = Poly_Mapping.single x k")
           (auto simp: coeff_powsum_mpoly coeff_monom when_def)
    qed
  qed
qed (auto simp: powsum_mpoly_infinite)

text ‹
  Power sum polynomials are symmetric:
›
lemma symmetric_powsum_mpoly [intro]:
  assumes "A  B"
  shows   "symmetric_mpoly A (powsum_mpoly B k)"
  unfolding powsum_mpoly_altdef
proof (rule symmetric_mpoly_symmetric_sum)
  fix x π
  assume "x  B" "π permutes A"
  thus "mpoly_map_vars π (MPoly_Type.monom (Poly_Mapping.single x k) 1) =
        MPoly_Type.monom (Poly_Mapping.single (π x) k) 1"
    using assms by (auto simp: mpoly_map_vars_monom permutes_bij permutep_single
                               bij_imp_bij_inv permutes_inv_inv)
qed (use assms in auto simp: permutes_subset)

lemma insertion_powsum_mpoly [simp]: "insertion f (powsum_mpoly X k) = (iX. f i ^ k)"
  unfolding powsum_mpoly_altdef insertion_sum insertion_single by simp

lemma powsum_mpoly_nz:
  assumes "finite X" "X  {}" "k > 0"
  shows   "(powsum_mpoly X k :: 'a :: {semiring_1, zero_neq_one} mpoly)  0"
proof -
  from assms obtain x where "x  X" by auto
  hence "coeff (powsum_mpoly X k) (Poly_Mapping.single x k) = (1 :: 'a)"
    using assms by (auto simp: coeff_powsum_mpoly)
  thus ?thesis by auto
qed

lemma powsum_mpoly_eq_0_iff:
  assumes "k > 0"
  shows   "powsum_mpoly X k = 0  infinite X  X = {}"
  using assms powsum_mpoly_nz[of X k] by (auto simp: powsum_mpoly_infinite)


subsection ‹The Girard--Newton Theorem›

text ‹
  The following is a nice combinatorial proof of the Girard--Newton Theorem due to
  Doron Zeilberger~cite"zeilberger".

  The precise statement is this:

  Let $e_k$ denote the $k$-th elementary symmetric polynomial in $X_1,\ldots,X_n$.
  This is the sum of all monomials that can be formed by taking the product of $k$ 
  distinct variables.

  Next, let $p_k = X_1^k + \ldots + X_n^k$ denote that $k$-th symmetric power sum polynomial
  in $X_1,\ldots,X_n$.

  Then the following equality holds:
  \[(-1)^k k e_k + \sum_{i=0}^{k-1} (-1)^i e_i p_{k-i}\]
›
theorem Girard_Newton:
  assumes "finite X"
  shows   "(-1) ^ k * of_nat k * sym_mpoly X k +
           (i<k. (-1) ^ i * sym_mpoly X i * powsum_mpoly X (k - i)) =
             (0 :: 'a :: comm_ring_1 mpoly)"
  (is "?lhs = 0")
proof -
  write Poly_Mapping.single (sng)

  define n where "n = card X"
  define 𝒜 :: "(nat set × nat) set"
    where "𝒜 = {(A, j). A  X  card A  k  j  X  (card A = k  j  A)}"
  define 𝒜1 :: "(nat set × nat) set"
    where "𝒜1 = {APow X. card A < k} × X"
  define 𝒜2 :: "(nat set × nat) set"
    where "𝒜2 = (SIGMA A:{APow X. card A = k}. A)"

  have 𝒜_split: "𝒜 = 𝒜1  𝒜2" "𝒜1  𝒜2 = {}"
    by (auto simp: 𝒜_def 𝒜1_def 𝒜2_def)
  have [intro]: "finite 𝒜1" "finite 𝒜2"
    using assms finite_subset[of _ X] by (auto simp: 𝒜1_def 𝒜2_def intro!: finite_SigmaI)
  have [intro]: "finite 𝒜"
    by (subst 𝒜_split) auto

  ― ‹
    We define a `weight' function w› from 𝒜› to the ring of polynomials as
    \[w(A,j) = (-1)^{|A|} x_j^{k-|A|} \prod_{i\in A} x_i\ .\]
  ›
  define w :: "nat set × nat  'a mpoly"
    where "w = (λ(A, j). monom (monom_of_set A + sng j (k - card A)) ((-1) ^ card A))"

  ― ‹The sum of these weights over all of 𝒜› is precisely the sum that we want to show equals 0:›
  have "?lhs = (x𝒜. w x)"
  proof -
    have "(x𝒜. w x) = (x𝒜1. w x) + (x𝒜2. w x)"
      by (subst 𝒜_split, subst sum.union_disjoint, use 𝒜_split(2) in auto)

    also have "(x𝒜1. w x) = (i<k. (-1) ^ i * sym_mpoly X i * powsum_mpoly X (k - i))"
    proof -
      have "(x𝒜1. w x) = (A | A  X  card A < k. jX. w (A, j))"
        using assms by (subst sum.Sigma) (auto simp: 𝒜1_def)
      also have " = (A | A  X  card A < k. jX.
                        monom (monom_of_set A) ((-1) ^ card A) * monom (sng j (k - card A)) 1)"
        unfolding w_def by (intro sum.cong) (auto simp: mult_monom)
      also have " = (A | A  X  card A < k. monom (monom_of_set A) ((-1) ^ card A) *
                        powsum_mpoly X (k - card A))"
        by (simp add: sum_distrib_left powsum_mpoly_altdef)
      also have " = ((i,A)  (SIGMA i:{..<k}. {A. A  X  card A = i}).
                        monom (monom_of_set A) ((-1) ^ i) * powsum_mpoly X (k - i))"
        by (rule sum.reindex_bij_witness[of _ snd "λA. (card A, A)"]) auto
      also have " = (i<k. A | A  X  card A = i.
                        monom (monom_of_set A) 1 * monom 0 ((-1) ^ i) * powsum_mpoly X (k - i))"
        using assms by (subst sum.Sigma) (auto simp: mult_monom)
      also have " = (i<k. (-1) ^ i * sym_mpoly X i * powsum_mpoly X (k - i))"
        by (simp add: sum_distrib_left sum_distrib_right mpoly_monom_0_eq_Const 
                      mpoly_Const_power mpoly_Const_uminus algebra_simps sym_mpoly_altdef)
      finally show ?thesis .
    qed

    also have "(x𝒜2. w x) = (-1) ^ k * of_nat k * sym_mpoly X k"
    proof -
      have "(x𝒜2. w x) = ((A,j)𝒜2. monom (monom_of_set A) ((- 1) ^ k))"
        by (intro sum.cong) (auto simp: 𝒜2_def w_def mpoly_monom_0_eq_Const intro!: sum.cong)
      also have " = (A | A  X  card A = k. jA. monom (monom_of_set A) ((- 1) ^ k))"
        using assms finite_subset[of _ X] by (subst sum.Sigma) (auto simp: 𝒜2_def)
      also have "(λA. monom (monom_of_set A) ((- 1) ^ k) :: 'a mpoly) =
                   (λA. monom 0 ((-1) ^ k) * monom (monom_of_set A) 1)"
        by (auto simp: fun_eq_iff mult_monom)
      also have "monom 0 ((-1) ^ k) = (-1) ^ k"
        by (auto simp: mpoly_monom_0_eq_Const mpoly_Const_power mpoly_Const_uminus)
      also have "(A | A  X  card A = k. jA. (- 1) ^ k * monom (monom_of_set A) 1) =
                   ((-1) ^ k * of_nat k * sym_mpoly X k :: 'a mpoly)"
        by (auto simp: sum_distrib_left sum_distrib_right mult_ac sym_mpoly_altdef)
      finally show ?thesis .
    qed

    finally show ?thesis by (simp add: algebra_simps)
  qed

  ― ‹Next, we show that the weights sum to 0:›
  also have "(x𝒜. w x) = 0"
  proof -
    ― ‹We define a function T› that is a involutory permutation of 𝒜›.
        To be more precise, it bijectively maps those elements (A,j)› of 𝒜› with j ∈ A›
        to those where j ∉ A› and the other way round. `Involutory' means that T› is its
        own inverse function, i.\,e.\ $T(T(x)) = x$.›
    define T :: "nat set × nat  nat set × nat"
      where "T = (λ(A, j). if j  A then (A - {j}, j) else (insert j A, j))"
    have [simp]: "T (T x) = x" for x
      by (auto simp: T_def split: prod.splits)
    have [simp]: "T x  𝒜" if "x  𝒜" for x
    proof -
      have [simp]: "n  n - Suc 0  n = 0" for n
        by auto
      show ?thesis using that assms finite_subset[of _ X]
        by (auto simp: T_def 𝒜_def split: prod.splits)
    qed
    have "snd (T x)  fst (T x)  snd x  fst x" if "x  𝒜" for x
      by (auto simp: T_def split: prod.splits)
    hence bij: "bij_betw T {x𝒜. snd x  fst x} {x𝒜. snd x  fst x}"
      by (intro bij_betwI[of _ _ _ T]) auto

    ―‹Crucially, we show that termT flips the weight of each element:›
    have [simp]: "w (T x) = -w x" if "x  𝒜" for x
    proof -
      obtain A j where [simp]: "x = (A, j)" by force
      
      ― ‹Since termT is an involution, we can assume w.\,l.\,o.\,g.\ that j ∈ A›:›
      have aux: "w (T (A, j)) = - w (A, j)" if "(A, j)  𝒜" "j  A" for j A
      proof -
        from that have [simp]: "j  A" "A  X" and "k > 0"
          using finite_subset[OF _ assms, of A] by (auto simp: 𝒜_def intro!: Nat.gr0I)
        have [simp]: "finite A"
          using finite_subset[OF _ assms, of A] by auto
        from that have "card A  k"
          by (auto simp: 𝒜_def)

        have card: "card A = Suc (card (A - {j}))"
          using card.remove[of A j] by auto
        hence card_less: "card (A - {j}) < card A" by linarith

        have "w (T (A, j)) = monom (monom_of_set (A - {j}) + sng j (k - card (A - {j})))
                         ((- 1) ^ card (A - {j}))" by (simp add: w_def T_def)
        also have "(- 1) ^ card (A - {j}) = ((- 1) ^ Suc (Suc (card (A - {j}))) :: 'a)"
          by simp
        also have "Suc (card (A - {j})) = card A"
          using card by simp
        also have "k - card (A - {j}) = Suc (k - card A)"
          using k > 0 card A  k card_less by (subst card) auto
        also have "monom_of_set (A - {j}) + sng j (Suc (k - card A)) =
                   monom_of_set A + sng j (k - card A)"
          by (transfer fixing: A j k) (auto simp: fun_eq_iff)
        also have "monom  ((-1)^ Suc (card A)) = -w (A, j)"
          by (simp add: w_def monom_uminus)
        finally show ?thesis .
      qed

      show ?thesis
      proof (cases "j  A")
        case True
        with aux[of A j] that show ?thesis by auto
      next
        case False
        hence "snd (T x)  fst (T x)"
          by (auto simp: T_def split: prod.splits)
        with aux[of "fst (T x)" "snd (T x)"] that show ?thesis by auto
      qed
    qed

    text ‹
      We can now show fairly easily that the sum is equal to zero.
    ›
    have *: "𝒜 = {x𝒜. snd x  fst x}  {x𝒜. snd x  fst x}"
      by auto
    have "(x𝒜. w x) = (x | x  𝒜  snd x  fst x. w x) + (x | x  𝒜  snd x  fst x. w x)"
      using finite 𝒜 by (subst *, subst sum.union_disjoint) auto
    also have "(x | x  𝒜  snd x  fst x. w x) = (x | x  𝒜  snd x  fst x. w (T x))"
      using sum.reindex_bij_betw[OF bij, of w] by simp
    also have " = -(x | x  𝒜  snd x  fst x. w x)"
      by (simp add: sum_negf)
    finally show "(x𝒜. w x) = 0"
      by simp
  qed

  finally show ?thesis .
qed

text ‹
  The following variant of the theorem holds for k > n›. Note that this is now a
  linear recurrence relation with constant coefficients for $p_k$ in terms of
  $e_0, \ldots, e_n$.
›
corollary Girard_Newton':
  assumes "finite X" and "k > card X"
  shows   "(icard X. (-1) ^ i * sym_mpoly X i * powsum_mpoly X (k - i)) =
             (0 :: 'a :: comm_ring_1 mpoly)"
proof -
  have "(0 :: 'a mpoly) = (i<k. (- 1) ^ i * sym_mpoly X i * powsum_mpoly X (k - i))"
    using Girard_Newton[of X k] assms by simp
  also have " = (icard X. (- 1) ^ i * sym_mpoly X i * powsum_mpoly X (k - i))"
    using assms by (intro sum.mono_neutral_right) auto
  finally show ?thesis ..
qed  

text ‹
  The following variant is the Newton--Girard Theorem solved for $e_k$, giving us
  an explicit way to determine $e_k$ from $e_0, \ldots, e_{k-1}$ and $p_1, \ldots, p_k$:
›
corollary sym_mpoly_recurrence:
  assumes k: "k > 0" and "finite X"
  shows   "(sym_mpoly X k :: 'a :: field_char_0 mpoly) =
             -smult (1 / of_nat k) (i=1..k. (-1) ^ i * sym_mpoly X (k - i) * powsum_mpoly X i)"
proof -
  define e p :: "nat  'a mpoly" where [simp]: "e = sym_mpoly X" "p = powsum_mpoly X"
  have *: "0 = (-1) ^ k * of_nat k * e k +
              (i<k. (- 1) ^ i * e i * p (k - i) :: 'a mpoly)"
    using Girard_Newton[of X k] assms by simp

  have "0 = (-1) ^ k * smult (1 / of_nat k) (0 :: 'a mpoly)"
    by simp
  also have " = smult (1 / of_nat k) (of_nat k) * e k +
                  smult (1 / of_nat k) (i<k. (-1)^(k+i) * e i * p (k - i))"
    unfolding smult_conv_mult
    using k by (subst *) (simp add: power_add sum_distrib_left sum_distrib_right field_simps 
                               del: div_mult_self3 div_mult_self4 div_mult_self2 div_mult_self1)
  also have "smult (1 / of_nat k :: 'a) (of_nat k) = 1"
    using k by (simp add: of_nat_monom smult_conv_mult mult_monom del: monom_of_nat)
  also have "(i<k. (-1) ^ (k+i) * e i * p (k - i)) = (i=1..k. (-1) ^ i * e (k-i) * p i)"
    by (intro sum.reindex_bij_witness[of _ "λi. k - i" "λi. k - i"])
       (auto simp: minus_one_power_iff)
  finally show ?thesis unfolding e_p_def by algebra
qed

text ‹
  Analogously, the following is the theorem solved for $p_k$, giving us a
  way to determine $p_k$ from $e_0, \ldots, e_k$ and $p_1, \ldots, p_{k-1}$:
›
corollary powsum_mpoly_recurrence:
  assumes k: "k > 0" and X: "finite X"
  shows   "(powsum_mpoly X k :: 'a :: comm_ring_1 mpoly) =
             (-1) ^ (k + 1) * of_nat k * sym_mpoly X k -
             (i=1..<k. (-1) ^ i * sym_mpoly X i * powsum_mpoly X (k - i))"
proof -
  define e p :: "nat  'a mpoly" where [simp]: "e = sym_mpoly X" "p = powsum_mpoly X"
  have *: "0 = (-1) ^ k * of_nat k * e k +
                 (i<k. (-1) ^ i * e i * p (k - i) :: 'a mpoly)"
    using Girard_Newton[of X k] assms by simp
  also have "{..<k} = insert 0 {1..<k}"
    using assms by auto
  finally have "(-1) ^ k * of_nat k * e k + (i=1..<k. (-1) ^ i * e i * p (k - i)) + p k = 0"
    using assms by (simp add: algebra_simps)
  from add.inverse_unique[OF this] show ?thesis by simp
qed

text ‹
  Again, if we assume $k > n$, the above takes a much simpler form and is, in fact,
  a linear recurrence with constant coefficients:
›
lemma powsum_mpoly_recurrence':
  assumes k: "k > card X" and X: "finite X"
  shows   "(powsum_mpoly X k :: 'a :: comm_ring_1 mpoly) =
             -(i=1..card X. (-1) ^ i * sym_mpoly X i * powsum_mpoly X (k - i))"
proof -
  define e p :: "nat  'a mpoly" where [simp]: "e = sym_mpoly X" "p = powsum_mpoly X"
  have "p k = (-1) ^ (k + 1) * of_nat k * e k - (i=1..<k. (-1) ^ i * e i * p (k - i))"
    unfolding e_p_def using assms by (intro powsum_mpoly_recurrence) auto
  also have " = -(i=1..<k. (-1) ^ i * e i * p (k - i))"
    using assms by simp
  also have "(i=1..<k. (-1) ^ i * e i * p (k - i)) = (i=1..card X. (-1) ^ i * e i * p (k - i))"
    using assms by (intro sum.mono_neutral_right) auto
  finally show ?thesis by simp
qed

end