Library prosa.util.minmax

Additional lemmas about BigMax.
Section ExtraLemmas.

Given a function F, a predicate P, and a sequence xs, we show that F x max { F i | i xs, P i} for any x in xs.
  Lemma leq_bigmax_cond_seq :
     {X : eqType} (F : X nat) (P : pred X) (xs : seq X) (x : X),
      x \in xs P x F x \max_(i <- xs | P i) F i.
  Proof. by intros × IN Px; rewrite (big_rem x) //= Px leq_maxl. Qed.

Next, we show that the fact max { F i | i xs, P i} m for some m is equivalent to the fact that x xs, P x F x m.
  Lemma bigmax_leq_seqP :
     {X : eqType} (F : X nat) (P : pred X) (xs : seq X) (m : nat),
      reflect ( x, x \in xs P x F x m)
              (\max_(x <- xs | P x) F x m).
  Proof.
    intros ×.
    apply: (iffP idP) ⇒ leFm ⇒ [i IINR Pi|].
    - by apply: leq_trans leFm; apply leq_bigmax_cond_seq.
    - rewrite big_seq_cond; elim/big_ind: _ ⇒ // m1 m2.
      + by intros; rewrite geq_max; apply/andP; split.
      + by move: m2 ⇒ /andP [M1IN Pm1]; apply: leFm.
  Qed.

Given two functions F1 and F2, a predicate P, and sequence xs, we show that if F1 x F2 x for any x \in xs such that P x, then max of {F1 x | x xs, P x} is bounded by the max of {F2 x | x xs, P x}.
  Lemma leq_big_max :
     {X : eqType} (F1 F2 : X nat) (P : pred X) (xs : seq X),
      ( x, x \in xs P x F1 x F2 x)
      \max_(x <- xs | P x) F1 x \max_(x <- xs | P x) F2 x.
  Proof.
    intros × ALL; apply /bigmax_leq_seqP; intros × IN Px.
    specialize (ALL x); feed_n 2 ALL; try done.
    rewrite (big_rem x) //=; rewrite Px.
    by apply leq_trans with (F2 x); [ | rewrite leq_maxl].
  Qed.

We show that for a positive n, max of {0, …, n-1} is smaller than n.
  Lemma bigmax_ord_ltn_identity :
     (n : nat),
      n > 0
      \max_(i < n) i < n.
  Proof.
    intros [ | n] POS; first by rewrite ltn0 in POS.
    clear POS; induction n; first by rewrite big_ord_recr /= big_ord0 maxn0.
    rewrite big_ord_recr /=.
    by rewrite /maxn IHn.
  Qed.

We state the next lemma in terms of ordinals. Given a natural number n, a predicate P, and an ordinal i0 {0, …, n-1} satisfying P, we show that max {i | P i} < n. Note that the element satisfying P is needed to prove that {i | P i} is not empty.
  Lemma bigmax_ltn_ord :
     (n : nat) (P : pred nat) (i0 : 'I_n),
    P i0
    \max_(i < n | P i) i < n.
  Proof.
    intros n P i0 Pi.
    destruct n; first by destruct i0 as [i0 P0]; move: (P0) ⇒ P0'; rewrite ltn0 in P0'.
    rewrite big_mkcond.
    apply leq_ltn_trans with (n := \max_(i < n.+1) i).
    - apply/bigmax_leqP; ins.
      destruct (P i); last by done.
      by apply leq_bigmax_cond.
    - by apply bigmax_ord_ltn_identity.
  Qed.

Next, we show that, given a natural number n, a predicate P, and an ordinal i0 {0, …, n-1} satisfying P, max {i | P i} < n also satisfies P.
  Lemma bigmax_pred :
     (n : nat) (P : pred nat) (i0 : 'I_n),
      P i0
      P (\max_(i < n | P i) i).
  Proof.
    intros × Pi0.
    induction n.
    { by destruct i0 as [i0 P0]; move: (P0) ⇒ P1; rewrite ltn0 in P1. }
    { rewrite big_mkcond big_ord_recr /=.
      destruct (P n) eqn:Pn.
      { destruct n; first by rewrite big_ord0 maxn0.
        unfold maxn at 1.
        destruct (\max_(i < n.+1) (match P (@nat_of_ord (S n) i) return nat with
                                   | true ⇒ @nat_of_ord (S n) i
                                   | falseO
                                   end) < n.+1) eqn:Pi; first by rewrite Pi.
        exfalso.
        apply negbT in Pi; move: Pi ⇒ /negP BUG; apply: BUG.
        apply leq_ltn_trans with (n := \max_(i < n.+1) i).
        { apply/bigmax_leqP; rewrite //= ⇒ i _.
          by destruct (P i); first apply leq_bigmax_cond.
        }
        { by apply bigmax_ord_ltn_identity. }
      }
      { rewrite maxn0 -big_mkcond /=.
        have LT: i0 < n; last by rewrite (IHn (Ordinal LT)).
        rewrite ltn_neqAle; apply/andP; split;
          last by rewrite -ltnS; apply ltn_ord.
        apply/negP; move ⇒ /eqP BUG.
        by rewrite -BUG Pi0 in Pn.
      }
    }
  Qed.

Furthermore, we observe that, if there is any element that satisfies the predicate, then there exists a witness for the computed maximum.
  Lemma bigmax_witness {T : eqType} :
     {xs : seq T} {P} F,
      has P xs
       x, x \in xs P x (F x = \max_(x <- xs | P x) F x).
  Proof.
    movexs P F.
    elim: xs ⇒ // a xs IH HAS.
    case: (boolP (P a)); last first.
    { move⇒ /negPf NOT.
      move: HAS; rewrite /has NOT //= -/(has P xs) ⇒ HAS.
      move: (IH HAS) ⇒ [x [IN [Px Fx]]].
       x; repeat split ⇒ //;
                            first by rewrite in_cons; apply /orP; right.
      by rewrite big_cons ifF. }
    { movePa.
      case: (boolP (has P xs)) ⇒ HAS'; last first.
      { a; repeat split ⇒ //; first by apply: mem_head.
        by rewrite big_cons ifT // big_hasC. }
      { move: (IH HAS') ⇒ [x [IN [Px MAX]]].
        case: (leqP (\max_(x <- xs | P x) F x) (F a)).
        { moveLEQ. a.
          repeat split ⇒ //; first by rewrite mem_head.
          rewrite big_cons ifT //= [maxn (_) (_)]/maxn ifF //.
          by lia. }
        { moveLT. x.
          repeat split ⇒ //; first by rewrite in_cons; apply /orP; right.
          by rewrite big_cons ifT //= [maxn (_) (_)]/maxn ifT. } } }
  Qed.

Additionally, we observe that, if two predicates yield different maxima, then there must exist a witness that satisfies only one of the predicates.
  Lemma bigmax_witness_diff {T : eqType} :
     {xs : seq T} {P1 P2 F},
      \max_(x <- xs | P1 x) F x < \max_(x <- xs | P2 x) F x
       x, x \in xs ~~ P1 x P2 x.
  Proof.
    movexs P1 P2 F LT.
    case: (boolP (has P1 xs)) ⇒ HP1;
                                  case: (boolP (has P2 xs)) ⇒ HP2.
    { move: (bigmax_witness F HP2) ⇒ [x2 [IN2 [Px2 MAX2]]].
       x2; repeat split ⇒ //.
      apply: contraT ⇒ /negPn Px1; exfalso.
      have BOUNDED: F x2 \max_(x <- xs | P1 x) F x by apply: leq_bigmax_cond_seq.
      by lia. }
    { by exfalso; move: LT; rewrite (big_hasC _ _ _ HP2). }
    { move: (bigmax_witness F HP2) ⇒ [x2 [IN2 [Px2 MAX2]]].
       x2; repeat split ⇒ //.
      move: HP1 ⇒ /hasPn HP1.
      by move: (HP1 x2 IN2). }
    { by exfalso; move: LT; rewrite !big_hasC. }
  Qed.

Conversely, we observe that if one predicates implies another, then the corresponding maxima are related.
  Corollary bigmax_subset {T : eqType} :
     {xs : seq T} {P1 P2 : pred T} {F},
      ( x, x \in xs P1 x P2 x)
      \max_(x <- xs | P1 x) F x \max_(x <- xs | P2 x) F x.
  Proof.
    movexs P1 P2 F IMPL.
    apply: contraT; rewrite -ltnNgeLT.
    exfalso.
    move: (bigmax_witness_diff LT) ⇒ [x [IN [/negP Px2 Px1]]].
    by apply Px2, IMPL.
  Qed.

End ExtraLemmas.