Require Import CoRN.algebra.RSetoid. Require Import CoRN.metric2.Metric. Require Import CoRN.metric2.UniformContinuity. From Coq Require Import QArith. Require Import CoRN.model.totalorder.QposMinMax CoRN.model.metric2.Qmetric CoRN.reals.fast.CRArith. (** The approximation function for CRplus results distributes a given error evenly among its two operands. This is a perfectly reasonable implementation choice, but it is conceptually arbitrary: any ratio works fine. Furthermore, when reasoning about particular uses of CRplus, different ratios can be more natural fits to the proof at hand. For instance, in the additivity proof for the Riemann integral implementation, having the error for the sum of two integrals distributed proportionally to the widths of the two contiguous ranges makes it nicely match the error for the full integral on both sides. For situations like these, we now do a very ad-hoc redefinition of addition with a user-specified error distribution ratio, and show that it is equivalent to the normal CRplus. *) Section uneven_CRplus. Variables (l r: Qpos) (* These are the error weights for x and y. *) (x y: CR). Let ll: Qpos := (l * Qpos_inv (l + r))%Qpos. Let rr: Qpos := (r * Qpos_inv (l + r))%Qpos. Let llrr: QposEq (ll + rr) (1#1). Proof. unfold ll, rr. simpl. unfold QposEq; simpl; field. intro. apply (Qpos_nonzero (l+r)%Qpos). assumption. Qed. Definition uneven_CRplus_approx (e: Qpos): Q_as_MetricSpace := approximate x (Qpos2QposInf (e * ll)) + approximate y (Qpos2QposInf (e * rr)). Lemma uneven_CRplus_is_RegularFunction: is_RegularFunction_noInf _ uneven_CRplus_approx. Proof with auto. intros e1 e2. unfold uneven_CRplus_approx. simpl. assert (QposEq (e1 + e2) ((e1 * ll + e2 * ll) + (e1 * rr + e2 * rr))). { unfold QposEq. transitivity (proj1_sig (e1 + e2)%Qpos * proj1_sig (ll + rr)%Qpos). unfold QposEq in llrr. rewrite llrr. simpl. ring. simpl. ring. } apply (ball_wd _ H _ _ (reflexivity _) _ _ (reflexivity _)). clear H. apply Qball_plus. apply (regFun_prf x (e1*ll)%Qpos (e2*ll)%Qpos). apply (regFun_prf y (e1*rr)%Qpos (e2*rr)%Qpos). Qed. Definition uneven_CRplus: CR := @mkRegularFunction Q_as_MetricSpace 0 _ uneven_CRplus_is_RegularFunction. Lemma uneven_CRplus_correct: (uneven_CRplus == x + y)%CR. Proof. simpl. apply regFunEq_equiv, regFunEq_e. intro e. unfold ball, Q_as_MetricSpace, Qball, QAbsSmall. rewrite approximate_CRplus... unfold uneven_CRplus_approx. setoid_replace (proj1_sig e + proj1_sig e) with (proj1_sig ((e * ll + (1#2) * e) + (e * rr + (1#2) * e)))%Qpos. apply Qball_plus; apply regFun_prf. simpl in llrr. transitivity (proj1_sig e + proj1_sig e * proj1_sig (ll + rr)%Qpos)... unfold QposEq in llrr. simpl in llrr. simpl. rewrite llrr. unfold canonical_names.equiv, stdlib_rationals.Q_eq. ring. unfold canonical_names.equiv, stdlib_rationals.Q_eq. simpl. ring. Qed. End uneven_CRplus.