lemma div_power_2_int_float[simp]: "x ∈ float ==> x / (2::int)^d ∈ float" by simp
lemma div_numeral_Bit0_float[simp]: assumes"x / numeral n ∈ float" shows"x / (numeral (Num.Bit0 n)) ∈ float" proof - have"(x / numeral n) / 2^1 ∈ float" by (intro assms div_power_2_float) alsohave"(x / numeral n) / 2^1 = x / (numeral (Num.Bit0 n))" by (induct n) auto finallyshow ?thesis . qed
lemma div_neg_numeral_Bit0_float[simp]: assumes"x / numeral n ∈ float" shows"x / (- numeral (Num.Bit0 n)) ∈ float" using assms by force
lemma power_float[simp]: assumes"a ∈ float" shows"a ^ b ∈ float" proof - from assms obtain m e :: int where"a = m * 2 powr e" by (auto simp: float_def) thenshow ?thesis by (intro floatI[where m="m^b"and e = "e*b"])
(auto simp: powr_powr power_mult_distrib simp flip: powr_realpow) qed
lift_definition Float :: "int ==> int ==> float"is"λ(m::int) (e::int). m * 2 powr e" by simp declare Float.rep_eq[simp]
code_datatype Float
lemma compute_real_of_float[code]: "real_of_float (Float m e) = (if e ≥ 0 then m * 2 ^ nat e else m / 2 ^ (nat (-e)))" by (simp add: powr_int)
subsection‹Arithmetic operations on floating point numbers›
instantiation float :: "{ring_1,linorder,linordered_ring,linordered_idom,numeral,equal}" begin
lemma real_of_float_of_int_eq [simp]: "real_of_float (of_int z) = of_int z" by (cases z rule: int_diff_cases) (simp_all add: of_rat_diff)
lemma Float_0_eq_0[simp]: "Float 0 e = 0" by transfer simp
lemma real_of_float_power[simp]: "real_of_float (f^n) = real_of_float f^n" for f :: float by (induct n) simp_all
lemma real_of_float_min: "real_of_float (min x y) = min (real_of_float x) (real_of_float y)" and real_of_float_max: "real_of_float (max x y) = max (real_of_float x) (real_of_float y)" for x y :: float by (simp_all add: min_def max_def)
instance float :: unbounded_dense_linorder proof fix a b :: float show "∃c. a < c" by (metis Float.real_of_float less_float.rep_eq reals_Archimedean2) show "∃c. c < a" by (metis add_0 add_strict_right_mono neg_less_0_iff_less zero_less_one) show "∃c. a < c ∧ c < b" if "a < b" apply (rule exI[of _ "(a + b) * Float 1 (- 1)"]) using that apply transfer apply (simp add: powr_minus) done qed
instantiation float :: lattice_ab_group_add begin
definition inf_float :: "float ==> float ==> float" where "inf_float a b = min a b"
definition sup_float :: "float ==> float ==> float" where "sup_float a b = max a b"
instance by standard (transfer; simp add: inf_float_def sup_float_def real_of_float_min real_of_float_max)+
end
lemma float_numeral[simp]: "real_of_float (numeral x :: float) = numeral x" by (metis of_int_numeral real_of_float_of_int_eq)
lemma float_of_numeral: "numeral k = float_of (numeral k)" and float_of_neg_numeral: "- numeral k = float_of (- numeral k)" unfolding real_of_float_eq by simp_all
subsection ‹Quickcheck›
instantiation float :: exhaustive begin
definition exhaustive_float where "exhaustive_float f d =
Quickcheck_Exhaustive.exhaustive (λx. Quickcheck_Exhaustive.exhaustive (λy. f (Float x y)) d) d"
instance ..
end
context includes term_syntax begin
definition [code_unfold]: "valtermify_float x y = Code_Evaluation.valtermify Float {⋅} x {⋅} y"
end
instantiation float :: full_exhaustive begin
definition "full_exhaustive_float f d =
Quickcheck_Exhaustive.full_exhaustive
(λx. Quickcheck_Exhaustive.full_exhaustive (λy. f (valtermify_float x y)) d) d"
instance ..
end
instantiation float :: random begin
definition "Quickcheck_Random.random i =
scomp (Quickcheck_Random.random (2 ^ nat_of_natural i))
(λman. scomp (Quickcheck_Random.random i) (λexp. Pair (valtermify_float man exp)))"
instance ..
end
subsection ‹Represent floats as unique mantissa and exponent›
lemma int_induct_abs[case_names less]: fixes j :: int assumes H: "∧n. (∧i. ∣i∣ < ∣n∣==> P i) ==> P n" shows "P j" proof (induct "nat ∣j∣" arbitrary: j rule: less_induct) case less show ?case by (rule H[OF less]) simp qed
lemma int_cancel_factors: fixes n :: int assumes "1 < r" shows "n = 0∨ (∃k i. n = k * r ^ i ∧¬ r dvd k)" proof (induct n rule: int_induct_abs) case (less n) have "∃k i. n = k * r ^ Suc i ∧¬ r dvd k" if "n ≠0" "n = m * r" for m proof - from that have "∣m ∣ < ∣n∣" using ‹1 < r› by (simp add: abs_mult) from less[OF this] that show ?thesis by auto qed then show ?case by (metis dvd_def monoid_mult_class.mult.right_neutral mult.commute power_0) qed
lemma mult_powr_eq_mult_powr_iff_asym: fixes m1 m2 e1 e2 :: int assumes m1: "¬2 dvd m1" and "e1 ≤ e2" shows "m1 * 2 powr e1 = m2 * 2 powr e2 ⟷ m1 = m2 ∧ e1 = e2" (is "?lhs ⟷ ?rhs") proof show ?rhs if eq: ?lhs proof - have "m1 ≠0" using m1 unfolding dvd_def by auto from ‹e1 ≤ e2› eq have "m1 = m2 * 2 powr nat (e2 - e1)" by (simp add: powr_diff field_simps) also have "… = m2 * 2^nat (e2 - e1)" by (simp add: powr_realpow) finally have m1_eq: "m1 = m2 * 2^nat (e2 - e1)" by linarith with m1 have "m1 = m2" by (cases "nat (e2 - e1)") (auto simp add: dvd_def) then show ?thesis using eq ‹m1 ≠ 0› by (simp add: powr_inj) qed show ?lhs if ?rhs using that by simp qed
lemma mult_powr_eq_mult_powr_iff: "¬2 dvd m1 ==>¬2 dvd m2 ==> m1 * 2 powr e1 = m2 * 2 powr e2 ⟷ m1 = m2 ∧ e1 = e2" for m1 m2 e1 e2 :: int using mult_powr_eq_mult_powr_iff_asym[of m1 e1 e2 m2] using mult_powr_eq_mult_powr_iff_asym[of m2 e2 e1 m1] by (cases e1 e2 rule: linorder_le_cases) auto
lemma floatE_normed: assumes x: "x ∈ float" obtains (zero) "x = 0" | (powr) m e :: int where "x = m * 2 powr e" "¬2 dvd m" "x ≠0" proof - have "∃(m::int) (e::int). x = m * 2 powr e ∧¬ (2::int) dvd m" if "x ≠0" proof - from x obtain m e :: int where x: "x = m * 2 powr e" by (auto simp: float_def) with ‹x ≠ 0› int_cancel_factors[of 2 m] obtain k i where "m = k * 2 ^ i" "¬2 dvd k" by auto with ‹¬ 2 dvd k› x have "x = real_of_int k * 2 powr real_of_int (e + int i) ∧ odd k" by (simp add: powr_add powr_realpow) then show ?thesis by blast qed with that show thesis by blast qed
lemma float_normed_cases: fixes f :: float obtains (zero) "f = 0" | (powr) m e :: int where "real_of_float f = m * 2 powr e" "¬2 dvd m" "f ≠0" proof (atomize_elim, induct f) case (float_of y) then show ?case by (cases rule: floatE_normed) (auto simp: zero_float_def) qed
definition mantissa :: "float ==> int" where "mantissa f =
fst (SOME p::int × int. (f = 0∧ fst p = 0∧ snd p = 0) ∨
(f ≠0∧ real_of_float f = real_of_int (fst p) * 2 powr real_of_int (snd p) ∧¬2 dvd fst p))"
definition exponent :: "float ==> int" where "exponent f =
snd (SOME p::int × int. (f = 0∧ fst p = 0∧ snd p = 0) ∨
(f ≠0∧ real_of_float f = real_of_int (fst p) * 2 powr real_of_int (snd p) ∧¬2 dvd fst p))"
lemma exponent_0[simp]: "exponent 0 = 0" (is ?E) and mantissa_0[simp]: "mantissa 0 = 0" (is ?M) proof - have "∧p::int × int. fst p = 0∧ snd p = 0⟷ p = (0, 0)" by auto then show ?E ?M by (auto simp add: mantissa_def exponent_def zero_float_def) qed
lemma mantissa_exponent: "real_of_float f = mantissa f * 2 powr exponent f" (is ?E) and mantissa_not_dvd: "f ≠0==>¬2 dvd mantissa f" (is "_ ==> ?D") proof cases assume [simp]: "f ≠0" have "f = mantissa f * 2 powr exponent f ∧¬2 dvd mantissa f" proof (cases f rule: float_normed_cases) case zero then show ?thesis by simp next case (powr m e) then have "∃p::int × int. (f = 0∧ fst p = 0∧ snd p = 0) ∨
(f ≠0∧ real_of_float f = real_of_int (fst p) * 2 powr real_of_int (snd p) ∧¬2 dvd fst p)" by auto then show ?thesis unfolding exponent_def mantissa_def by (rule someI2_ex) simp qed then show ?E ?D by auto qed simp
lemma mantissa_noteq_0: "f ≠0==> mantissa f ≠0" using mantissa_not_dvd[of f] by auto
lemma mantissa_eq_zero_iff: "mantissa x = 0⟷ x = 0" (is "?lhs ⟷ ?rhs") proof show ?rhs if ?lhs proof - from that have z: "0 = real_of_float x" using mantissa_exponent by simp show ?thesis by (simp add: zero_float_def z) qed show ?lhs if ?rhs using that by simp qed
lemma mantissa_pos_iff: "0 < mantissa x ⟷0 < x" by (auto simp: mantissa_exponent algebra_split_simps)
lemma mantissa_nonneg_iff: "0≤ mantissa x ⟷0≤ x" by (auto simp: mantissa_exponent algebra_split_simps)
lemma mantissa_neg_iff: "0 > mantissa x ⟷0 > x" by (auto simp: mantissa_exponent algebra_split_simps)
lemma fixes m e :: int defines "f ≡ float_of (m * 2 powr e)" assumes dvd: "¬2 dvd m" shows mantissa_float: "mantissa f = m" (is "?M") and exponent_float: "m ≠0==> exponent f = e" (is "_ ==> ?E") proof cases assume "m = 0" with dvd show "mantissa f = m" by auto next assume "m ≠0" then have f_not_0: "f ≠0" by (simp add: f_def zero_float_def) from mantissa_exponent[of f] have "m * 2 powr e = mantissa f * 2 powr exponent f" by (auto simp add: f_def) then show ?M ?E using mantissa_not_dvd[OF f_not_0] dvd by (auto simp: mult_powr_eq_mult_powr_iff) qed
lemma Float_cases [cases type: float]: fixes f :: float obtains (Float) m e :: int where "f = Float m e" using Float_mantissa_exponent[symmetric] by (atomize_elim) auto
lemma denormalize_shift: assumes f_def: "f = Float m e" and not_0: "f ≠0" obtains i where "m = mantissa f * 2 ^ i" "e = exponent f - i" proof from mantissa_exponent[of f] f_def have "m * 2 powr e = mantissa f * 2 powr exponent f" by simp then have eq: "m = mantissa f * 2 powr (exponent f - e)" by (simp add: powr_diff field_simps) moreover have "e ≤ exponent f" proof (rule ccontr) assume "¬ e ≤ exponent f" then have pos: "exponent f < e" by simp then have "2 powr (exponent f - e) = 2 powr - real_of_int (e - exponent f)" by simp also have "… = 1 / 2^nat (e - exponent f)" using pos by (simp flip: powr_realpow add: powr_diff) finally have "m * 2^nat (e - exponent f) = real_of_int (mantissa f)" using eq by simp then have "mantissa f = m * 2^nat (e - exponent f)" by linarith with ‹exponent f < e› have "2 dvd mantissa f" by (force intro: dvdI[where k="m * 2^(nat (e-exponent f)) div 2"]) then show False using mantissa_not_dvd[OF not_0] by simp qed ultimately have "real_of_int m = mantissa f * 2^nat (exponent f - e)" by (simp flip: powr_realpow) with ‹e ≤ exponent f› show "m = mantissa f * 2 ^ nat (exponent f - e)" by linarith show "e = exponent f - nat (exponent f - e)" using ‹e ≤ exponent f› by auto qed
context begin
qualified lemma compute_float_zero[code_unfold, code]: "0 = Float 00" by transfer simp
qualified lemma compute_float_one[code_unfold, code]: "1 = Float 10" by transfer simp
lift_definition normfloat :: "float ==> float" is "λx. x" . lemma normloat_id[simp]: "normfloat x = x" by transfer rule
qualified lemma compute_normfloat[code]: "normfloat (Float m e) =
(if m mod 2 = 0∧ m ≠0then normfloat (Float (m div 2) (e + 1))
else if m = 0then0 else Float m e)" by transfer (auto simp add: powr_add zmod_eq_0_iff)
qualified lemma compute_float_numeral[code_abbrev]: "Float (numeral k) 0 = numeral k" by transfer simp
qualified lemma compute_float_neg_numeral[code_abbrev]: "Float (- numeral k) 0 = - numeral k" by transfer simp
qualified lemma compute_float_uminus[code]: "- Float m1 e1 = Float (- m1) e1" by transfer simp
qualified lemma compute_float_minus[code]: "f - g = f + (-g)" for f g :: float by simp
qualified lemma compute_float_sgn[code]: "sgn (Float m1 e1) = (if0 < m1 then1 else if m1 < 0then -1 else 0)" by transfer (simp add: sgn_mult)
lift_definition is_float_pos :: "float ==> bool" is "(<) 0 :: real ==> bool" .
qualified lemma compute_is_float_pos[code]: "is_float_pos (Float m e) ⟷0 < m" by transfer (auto simp add: zero_less_mult_iff not_le[symmetric, of _ 0])
lift_definition is_float_nonneg :: "float ==> bool" is "(≤) 0 :: real ==> bool" .
qualified lemma compute_is_float_nonneg[code]: "is_float_nonneg (Float m e) ⟷0≤ m" by transfer (auto simp add: zero_le_mult_iff not_less[symmetric, of _ 0])
lift_definition is_float_zero :: "float ==> bool" is "(=) 0 :: real ==> bool" .
qualified lemma compute_is_float_zero[code]: "is_float_zero (Float m e) ⟷0 = m" by transfer (auto simp add: is_float_zero_def)
qualified lemma compute_float_abs[code]: "∣Float m e∣ = Float ∣m∣ e" by transfer (simp add: abs_mult)
qualified lemma compute_float_eq[code]: "equal_class.equal f g = is_float_zero (f - g)" by transfer simp
end
subsection ‹Lemmas for types 🍋‹real›, 🍋‹nat›, 🍋‹int››
lemma round_up_uminus_eq: "round_up p (-x) = - round_down p x" and round_down_uminus_eq: "round_down p (-x) = - round_up p x" by (auto simp: round_up_def round_down_def ceiling_def)
lemma round_up_mono: "x ≤ y ==> round_up p x ≤ round_up p y" by (auto intro!: ceiling_mono simp: round_up_def)
lemma round_up_le1: assumes "x ≤1" "prec ≥0" shows "round_up prec x ≤1" proof - have "real_of_int ⌈x * 2 powr prec⌉≤ real_of_int ⌈2 powr real_of_int prec⌉" using assms by (auto intro!: ceiling_mono) also have "… = 2 powr prec" using assms by (auto simp: powr_int intro!: exI[where x="2^nat prec"]) finally show ?thesis by (simp add: round_up_def) (simp add: powr_minus inverse_eq_divide) qed
lemma round_up_less1: assumes "x < 1 / 2" "p > 0" shows "round_up p x < 1" proof - have "x * 2 powr p < 1 / 2 * 2 powr p" using assms by simp also have "…≤2 powr p - 1" using ‹p > 0› by (auto simp: powr_diff powr_int field_simps self_le_power) finally show ?thesis using ‹p > 0› by (simp add: round_up_def field_simps powr_minus powr_int ceiling_less_iff) qed
lemma round_down_ge1: assumes x: "x ≥1" assumes prec: "p ≥ - log 2 x" shows "1≤ round_down p x" proof cases assume nonneg: "0≤ p" have "2 powr p = real_of_int ⌊2 powr real_of_int p⌋" using nonneg by (auto simp: powr_int) also have "…≤ real_of_int ⌊x * 2 powr p⌋" using assms by (auto intro!: floor_mono) finally show ?thesis by (simp add: round_down_def) (simp add: powr_minus inverse_eq_divide) next assume neg: "¬0≤ p" have "x = 2 powr (log 2 x)" using x by simp also have "2 powr (log 2 x) ≥2 powr - p" using prec by auto finally have x_le: "x ≥2 powr -p" .
from neg have "2 powr real_of_int p ≤2 powr 0" by (intro powr_mono) auto also have "…≤⌊2 powr 0::real⌋" by simp also have "…≤⌊x * 2 powr (real_of_int p)⌋" unfolding of_int_le_iff using x x_le by (intro floor_mono) (simp add: powr_minus_divide field_simps) finally show ?thesis using prec x by (simp add: round_down_def powr_minus_divide pos_le_divide_eq) qed
lemma round_up_le0: "x ≤0==> round_up p x ≤0" unfolding round_up_def by (auto simp: field_simps mult_le_0_iff zero_le_mult_iff)
subsection ‹Rounding Floats›
definition div_twopow :: "int ==> nat ==> int" where [simp]: "div_twopow x n = x div (2 ^ n)"
definition mod_twopow :: "int ==> nat ==> int" where [simp]: "mod_twopow x n = x mod (2 ^ n)"
lemma compute_div_twopow[code]: "div_twopow x n = (if x = 0∨ x = -1∨ n = 0then x else div_twopow (x div 2) (n - 1))" by (cases n) (auto simp: zdiv_zmult2_eq div_eq_minus1)
lemma compute_mod_twopow[code]: "mod_twopow x n = (if n = 0then0 else x mod 2 + 2 * mod_twopow (x div 2) (n - 1))" by (cases n) (auto simp: zmod_zmult2_eq)
lift_definition float_up :: "int ==> float ==> float" is round_up by simp declare float_up.rep_eq[simp]
lemma round_up_correct: "round_up e f - f ∈ {0..2 powr -e}" unfolding atLeastAtMost_iff proof have "round_up e f - f ≤ round_up e f - round_down e f" using round_down by simp also have "…≤2 powr -e" using round_up_diff_round_down by simp finally show "round_up e f - f ≤2 powr - (real_of_int e)" by simp qed (simp add: algebra_simps round_up)
lemma float_up_correct: "real_of_float (float_up e f) - real_of_float f ∈ {0..2 powr -e}" by transfer (rule round_up_correct)
lift_definition float_down :: "int ==> float ==> float" is round_down by simp declare float_down.rep_eq[simp]
lemma round_down_correct: "f - (round_down e f) ∈ {0..2 powr -e}" unfolding atLeastAtMost_iff proof have "f - round_down e f ≤ round_up e f - round_down e f" using round_up by simp also have "…≤2 powr -e" using round_up_diff_round_down by simp finally show "f - round_down e f ≤2 powr - (real_of_int e)" by simp qed (simp add: algebra_simps round_down)
lemma float_down_correct: "real_of_float f - real_of_float (float_down e f) ∈ {0..2 powr -e}" by transfer (rule round_down_correct)
context begin
qualified lemma compute_float_down[code]: "float_down p (Float m e) =
(if p + e < 0then Float (div_twopow m (nat (-(p + e)))) (-p) else Float m e)" proof (cases "p + e < 0") case True then have "real_of_int ((2::int) ^ nat (-(p + e))) = 2 powr (-(p + e))" using powr_realpow[of 2 "nat (-(p + e))"] by simp also have "… = 1 / 2 powr p / 2 powr e" unfolding powr_minus_divide of_int_minus by (simp add: powr_add) finally show ?thesis using ‹p + e < 0› apply transfer apply (simp add: round_down_def field_simps flip: floor_divide_of_int_eq powr_add) apply (metis (no_types, opaque_lifting) Float.rep_eq add.inverse_inverse compute_real_of_float diff_minus_eq_add floor_divide_of_int_eq int_of_reals(1) linorder_not_le minus_add_distrib of_int_eq_numeral_power_cancel_iff) done next case False then have r: "real_of_int e + real_of_int p = real (nat (e + p))" by simp have r: "⌊(m * 2 powr e) * 2 powr real_of_int p⌋ = (m * 2 powr e) * 2 powr real_of_int p" by (auto intro: exI[where x="m*2^nat (e+p)"] simp add: ac_simps powr_add[symmetric] r powr_realpow) with ‹¬ p + e < 0› show ?thesis by transfer (auto simp add: round_down_def field_simps powr_add powr_minus) qed
lemma abs_round_down_le: "∣f - (round_down e f)∣≤2 powr -e" using round_down_correct[of f e] by simp
lemma abs_round_up_le: "∣f - (round_up e f)∣≤2 powr -e" using round_up_correct[of e f] by simp
lemma round_down_nonneg: "0≤ s ==>0≤ round_down p s" by (auto simp: round_down_def)
lemma ceil_divide_floor_conv: assumes "b ≠0" shows "⌈real_of_int a / real_of_int b⌉ =
(if b dvd a then a div b else ⌊real_of_int a / real_of_int b⌋ + 1)" proof (cases "b dvd a") case True then show ?thesis by (simp add: ceiling_def floor_divide_of_int_eq dvd_neg_div flip: of_int_minus divide_minus_left) next case False then have "a mod b ≠0" by auto then have ne: "real_of_int (a mod b) / real_of_int b ≠0" using ‹b ≠ 0› by auto have "⌈real_of_int a / real_of_int b⌉ = ⌊real_of_int a / real_of_int b⌋ + 1" by (metis add_cancel_left_right ceiling_altdef floor_divide_of_int_eq ne of_int_div_aux) then show ?thesis using ‹¬ b dvd a› by simp qed
qualified lemma compute_float_up[code]: "float_up p x = - float_down p (-x)" by transfer (simp add: round_down_uminus_eq)
end
lemma bitlen_Float: fixes m e defines [THEN meta_eq_to_obj_eq]: "f ≡ Float m e" shows "bitlen ∣mantissa f∣ + exponent f = (if m = 0then0 else bitlen ∣m∣ + e)" proof (cases "m = 0") case True then show ?thesis by (simp add: f_def bitlen_alt_def) next case False then have "f ≠0" unfolding real_of_float_eq by (simp add: f_def) then have "mantissa f ≠0" by (simp add: mantissa_eq_zero_iff) moreover obtain i where "m = mantissa f * 2 ^ i" "e = exponent f - int i" by (rule f_def[THEN denormalize_shift, OF ‹f ≠ 0›]) ultimately show ?thesis by (simp add: abs_mult) qed
lemma float_gt1_scale: assumes "1≤ Float m e" shows "0≤ e + (bitlen m - 1)" proof - have "0 < Float m e" using assms by auto then have "0 < m" using powr_gt_zero[of 2 e] by (auto simp: zero_less_mult_iff) then have "m ≠0" by auto show ?thesis proof (cases "0≤ e") case True then show ?thesis using ‹0 < m› by (simp add: bitlen_alt_def) next case False have "(1::int) < 2" by simp let ?S = "2^(nat (-e))" have "inverse (2 ^ nat (- e)) = 2 powr e" using assms False powr_realpow[of 2 "nat (-e)"] by (auto simp: powr_minus field_simps) then have "1≤ real_of_int m * inverse ?S" using assms False powr_realpow[of 2 "nat (-e)"] by (auto simp: powr_minus) then have "1 * ?S ≤ real_of_int m * inverse ?S * ?S" by (rule mult_right_mono) auto then have "?S ≤ real_of_int m" unfolding mult.assoc by auto then have "?S ≤ m" unfolding of_int_le_iff[symmetric] by auto from this bitlen_bounds[OF ‹0 < m›, THEN conjunct2] have "nat (-e) < (nat (bitlen m))" unfolding power_strict_increasing_iff[OF ‹1 < 2›, symmetric] by (rule order_le_less_trans) then have "-e < bitlen m" using False by auto then show ?thesis by auto qed qed
subsection ‹Truncating Real Numbers›
definition truncate_down::"nat ==> real ==> real" where "truncate_down prec x = round_down (prec - ⌊log 2∣x∣⌋) x"
lemma truncate_down: "truncate_down prec x ≤ x" using round_down by (simp add: truncate_down_def)
lemma truncate_down_le: "x ≤ y ==> truncate_down prec x ≤ y" by (rule order_trans[OF truncate_down])
lemma truncate_up_float[simp]: "truncate_up p x ∈ float" by (auto simp: truncate_up_def)
lemma mult_powr_eq: "0 < b ==> b ≠1==>0 < x ==> x * b powr y = b powr (y + log b x)" by (simp_all add: powr_add)
lemma truncate_down_pos: assumes "x > 0" shows "truncate_down p x > 0" proof - have "0≤ log 2 x - real_of_int ⌊log 2 x⌋" by (simp add: algebra_simps) moreover have "0≤ real p - real_of_int ⌊log 2 x⌋ + log 2 x" by linarith ultimately show ?thesis using assms by (auto simp: truncate_down_def round_down_def mult_powr_eq intro!: ge_one_powr_ge_zero mult_pos_pos) qed
lemma truncate_down_nonneg: "0≤ y ==>0≤ truncate_down prec y" by (auto simp: truncate_down_def round_down_def)
lemma truncate_down_ge1: "1≤ x ==>1≤ truncate_down p x" apply (auto simp: truncate_down_def algebra_simps intro!: round_down_ge1) apply linarith done
lemma truncate_up_nonpos: "x ≤0==> truncate_up prec x ≤0" by (auto simp: truncate_up_def round_up_def intro!: mult_nonpos_nonneg)
lemma truncate_up_le1: assumes "x ≤1" shows "truncate_up p x ≤1" proof - consider "x ≤0" | "x > 0" by arith then show ?thesis proof cases case 1 with truncate_up_nonpos[OF this, of p] show ?thesis by simp next case 2 then have le: "⌊log 2∣x∣⌋≤0" using assms by (auto simp: log_less_iff) from assms have "0≤ int p" by simp from add_mono[OF this le] show ?thesis using assms by (simp add: truncate_up_def round_up_le1 add_mono) qed qed
lemma truncate_down_shift_int: "truncate_down p (x * 2 powr real_of_int k) = truncate_down p x * 2 powr k" by (cases "x = 0") (simp_all add: algebra_simps abs_mult log_mult truncate_down_def round_down_shift[of _ _ k, simplified])
lemma truncate_down_shift_nat: "truncate_down p (x * 2 powr real k) = truncate_down p x * 2 powr k" by (metis of_int_of_nat_eq truncate_down_shift_int)
lemma truncate_up_shift_int: "truncate_up p (x * 2 powr real_of_int k) = truncate_up p x * 2 powr k" by (cases "x = 0") (simp_all add: algebra_simps abs_mult log_mult truncate_up_def round_up_shift[of _ _ k, simplified])
lemma truncate_up_shift_nat: "truncate_up p (x * 2 powr real k) = truncate_up p x * 2 powr k" by (metis of_int_of_nat_eq truncate_up_shift_int)
subsection ‹Truncating Floats›
lift_definition float_round_up :: "nat ==> float ==> float" is truncate_up by (simp add: truncate_up_def)
lemma float_round_up: "real_of_float x ≤ real_of_float (float_round_up prec x)" using truncate_up by transfer simp
lemma float_round_up_zero[simp]: "float_round_up prec 0 = 0" by transfer simp
lift_definition float_round_down :: "nat ==> float ==> float" is truncate_down by (simp add: truncate_down_def)
lemma float_round_down: "real_of_float (float_round_down prec x) ≤ real_of_float x" using truncate_down by transfer simp
lemma float_round_down_zero[simp]: "float_round_down prec 0 = 0" by transfer simp
lemma minus_float_round_up_eq: "- float_round_up prec x = float_round_down prec (- x)" and minus_float_round_down_eq: "- float_round_down prec x = float_round_up prec (- x)" by (transfer; simp add: truncate_down_uminus_eq truncate_up_uminus_eq)+
context begin
qualified lemma compute_float_round_down[code]: "float_round_down prec (Float m e) =
(let d = bitlen ∣m∣ - int prec - 1in if0 < d then Float (div_twopow m (nat d)) (e + d)
else Float m e)" using Float.compute_float_down[of "Suc prec - bitlen ∣m∣ - e" m e, symmetric] by transfer (simp add: field_simps abs_mult log_mult bitlen_alt_def truncate_down_def cong del: if_weak_cong)
qualified lemma compute_float_round_up[code]: "float_round_up prec x = - float_round_down prec (-x)" by transfer (simp add: truncate_down_uminus_eq)
end
lemma truncate_up_nonneg_mono: assumes "0≤ x" "x ≤ y" shows "truncate_up prec x ≤ truncate_up prec y" proof - consider "⌊log 2 x⌋ = ⌊log 2 y⌋" | "⌊log 2 x⌋≠⌊log 2 y⌋" "0 < x" | "x ≤0" by arith then show ?thesis proof cases case 1 then show ?thesis using assms by (auto simp: truncate_up_def round_up_def intro!: ceiling_mono) next case 2 from assms ‹0 < x› have "log 2 x ≤ log 2 y" by auto with ‹⌊log 2 x⌋≠⌊log 2 y⌋› have logless: "log 2 x < log 2 y" by linarith have flogless: "⌊log 2 x⌋ < ⌊log 2 y⌋" using ‹⌊log 2 x⌋≠⌊log 2 y⌋›‹log 2 x ≤ log 2 y› by linarith have "truncate_up prec x =
real_of_int ⌈x * 2 powr real_of_int (int prec - ⌊log 2 x⌋ )⌉ * 2 powr - real_of_int (int prec - ⌊log 2 x⌋)" using assms by (simp add: truncate_up_def round_up_def) also have "⌈x * 2 powr real_of_int (int prec - ⌊log 2 x⌋)⌉≤ (2 ^ (Suc prec))" proof (simp only: ceiling_le_iff) have "x * 2 powr real_of_int (int prec - ⌊log 2 x⌋) ≤
x * (2 powr real (Suc prec) / (2 powr log 2 x))" using real_of_int_floor_add_one_ge[of "log 2 x"] assms by (auto simp: algebra_simps simp flip: powr_diff intro!: mult_left_mono) then show "x * 2 powr real_of_int (int prec - ⌊log 2 x⌋) ≤ real_of_int ((2::int) ^ (Suc prec))" using ‹0 < x› by (simp add: powr_realpow powr_add) qed then have "real_of_int ⌈x * 2 powr real_of_int (int prec - ⌊log 2 x⌋)⌉≤2 powr int (Suc prec)" by (auto simp: powr_realpow powr_add) (metis power_Suc of_int_le_numeral_power_cancel_iff) also have "2 powr - real_of_int (int prec - ⌊log 2 x⌋) ≤2 powr - real_of_int (int prec - ⌊log 2 y⌋ + 1)" using logless flogless by (auto intro!: floor_mono) also have "2 powr real_of_int (int (Suc prec)) ≤ 2 powr (log 2 y + real_of_int (int prec - ⌊log 2 y⌋ + 1))" using assms ‹0 < x› by (auto simp: algebra_simps) finally have "truncate_up prec x ≤ 2 powr (log 2 y + real_of_int (int prec - ⌊log 2 y⌋ + 1)) * 2 powr - real_of_int (int prec - ⌊log 2 y⌋ + 1)" by simp also have "… = 2 powr (log 2 y + real_of_int (int prec - ⌊log 2 y⌋) - real_of_int (int prec - ⌊log 2 y⌋))" by (subst powr_add[symmetric]) simp also have "… = y" using ‹0 < x› assms by (simp add: powr_add) also have "…≤ truncate_up prec y" by (rule truncate_up) finally show ?thesis . next case 3 then show ?thesis using assms by (auto intro!: truncate_up_le) qed qed
lemma truncate_up_switch_sign_mono: assumes "x ≤0" "0≤ y" shows "truncate_up prec x ≤ truncate_up prec y" proof - note truncate_up_nonpos[OF ‹x ≤ 0›] also note truncate_up_le[OF ‹0 ≤ y›] finally show ?thesis . qed
lemma truncate_down_switch_sign_mono: assumes "x ≤0" and "0≤ y" and "x ≤ y" shows "truncate_down prec x ≤ truncate_down prec y" proof - note truncate_down_le[OF ‹x ≤ 0›] also note truncate_down_nonneg[OF ‹0 ≤ y›] finally show ?thesis . qed
lemma truncate_down_nonneg_mono: assumes "0≤ x" "x ≤ y" shows "truncate_down prec x ≤ truncate_down prec y" proof - consider "x ≤0" | "⌊log 2∣x∣⌋ = ⌊log 2∣y∣⌋" | "0 < x" "⌊log 2∣x∣⌋≠⌊log 2∣y∣⌋" by arith then show ?thesis proof cases case 1 with assms have "x = 0" "0≤ y" by simp_all then show ?thesis by (auto intro!: truncate_down_nonneg) next case 2 then show ?thesis using assms by (auto simp: truncate_down_def round_down_def intro!: floor_mono) next case 3 from ‹0 < x› have "log 2 x ≤ log 2 y" "0 < y" "0≤ y" using assms by auto with ‹⌊log 2 ∣x∣⌋≠⌊log 2 ∣y∣⌋› have logless: "log 2 x < log 2 y" and flogless: "⌊log 2 x⌋ < ⌊log 2 y⌋" unfolding atomize_conj abs_of_pos[OF ‹0 < x›] abs_of_pos[OF ‹0 < y›] by (metis floor_less_cancel linorder_cases not_le) have "2 powr prec ≤ y * 2 powr real prec / (2 powr log 2 y)" using ‹0 < y› by simp also have "…≤ y * 2 powr real (Suc prec) / (2 powr (real_of_int ⌊log 2 y⌋ + 1))" using ‹0 ≤ y›‹0 ≤ x› assms(2) by (auto intro!: powr_mono divide_left_mono simp: of_nat_diff powr_add powr_diff) also have "… = y * 2 powr real (Suc prec) / (2 powr real_of_int ⌊log 2 y⌋ * 2)" by (auto simp: powr_add) finally have "(2 ^ prec) ≤⌊y * 2 powr real_of_int (int (Suc prec) - ⌊log 2∣y∣⌋ - 1)⌋" using ‹0 ≤ y› by (auto simp: powr_diff le_floor_iff powr_realpow powr_add) then have "(2 ^ (prec)) * 2 powr - real_of_int (int prec - ⌊log 2∣y∣⌋) ≤ truncate_down prec y" by (auto simp: truncate_down_def round_down_def) moreover have "x ≤ (2 ^ prec) * 2 powr - real_of_int (int prec - ⌊log 2∣y∣⌋)" proof - have "x = 2 powr (log 2∣x∣)" using ‹0 < x› by simp also have "…≤ (2 ^ (Suc prec )) * 2 powr - real_of_int (int prec - ⌊log 2∣x∣⌋)" using real_of_int_floor_add_one_ge[of "log 2∣x∣"] ‹0 < x› by (auto simp flip: powr_realpow powr_add simp: algebra_simps powr_mult_base le_powr_iff) also have "2 powr - real_of_int (int prec - ⌊log 2∣x∣⌋) ≤2 powr - real_of_int (int prec - ⌊log 2∣y∣⌋ + 1)" using logless flogless ‹x > 0›‹y > 0› by (auto intro!: floor_mono) finally show ?thesis by (auto simp flip: powr_realpow simp: powr_diff assms) qed ultimately show ?thesis by (metis dual_order.trans truncate_down) qed qed
lemma truncate_down_eq_truncate_up: "truncate_down p x = - truncate_up p (-x)" and truncate_up_eq_truncate_down: "truncate_up p x = - truncate_down p (-x)" by (auto simp: truncate_up_uminus_eq truncate_down_uminus_eq)
lemma truncate_down_mono: "x ≤ y ==> truncate_down p x ≤ truncate_down p y" by (smt (verit) truncate_down_nonneg_mono truncate_up_nonneg_mono truncate_up_uminus_eq)
lemma truncate_up_mono: "x ≤ y ==> truncate_up p x ≤ truncate_up p y" by (simp add: truncate_up_eq_truncate_down truncate_down_mono)
lemma truncate_up_nonneg: "0≤ truncate_up p x" if "0≤ x" by (simp add: that truncate_up_le)
lemma truncate_up_pos: "0 < truncate_up p x" if "0 < x" by (meson less_le_trans that truncate_up)
lemma truncate_up_less_zero_iff[simp]: "truncate_up p x < 0⟷ x < 0" by (smt (verit) truncate_down_pos truncate_down_uminus_eq truncate_up_nonneg)
lemma truncate_up_nonneg_iff[simp]: "truncate_up p x ≥0⟷ x ≥0" using truncate_up_less_zero_iff[of p x] truncate_up_nonneg[of x] by linarith
lemma truncate_down_less_zero_iff[simp]: "truncate_down p x < 0⟷ x < 0" by (metis le_less_trans not_less_iff_gr_or_eq truncate_down truncate_down_pos truncate_down_zero)
lemma truncate_down_nonneg_iff[simp]: "truncate_down p x ≥0⟷ x ≥0" using truncate_down_less_zero_iff[of p x] truncate_down_nonneg[of x p] by linarith
lemma truncate_down_eq_zero_iff[simp]: "truncate_down prec x = 0⟷ x = 0" by (metis not_less_iff_gr_or_eq truncate_down_less_zero_iff truncate_down_pos truncate_down_zero)
lemma truncate_up_eq_zero_iff[simp]: "truncate_up prec x = 0⟷ x = 0" by (metis not_less_iff_gr_or_eq truncate_up_less_zero_iff truncate_up_pos truncate_up_zero)
subsection ‹Approximation of positive rationals›
lemma div_mult_twopow_eq: "a div ((2::nat) ^ n) div b = a div (b * 2 ^ n)" for a b :: nat by (cases "b = 0") (simp_all add: div_mult2_eq[symmetric] ac_simps)
lemma real_div_nat_eq_floor_of_divide: "a div b = real_of_int ⌊a / b⌋" for a b :: nat by (simp add: floor_divide_of_nat_eq [of a b])
definition "rat_precision prec x y =
(let d = bitlen x - bitlen y in int prec - d + (if Float (abs x) 0 < Float (abs y) d then1 else 0))"
lemma floor_log_divide_eq: assumes "i > 0" "j > 0" "p > 1" shows "⌊log p (i / j)⌋ = floor (log p i) - floor (log p j) -
(if i ≥ j * p powr (floor (log p i) - floor (log p j)) then0 else 1)" proof - let ?l = "log p" let ?fl = "λx. floor (?l x)" have "⌊?l (i / j)⌋ = ⌊?l i - ?l j⌋" using assms by (auto simp: log_divide) also have "… = floor (real_of_int (?fl i - ?fl j) + (?l i - ?fl i - (?l j - ?fl j)))" (is "_ = floor (_ + ?r)") by (simp add: algebra_simps) also note floor_add2 also note ‹p > 1› note powr = powr_le_cancel_iff[symmetric, OF ‹1 < p›, THEN iffD2] note powr_strict = powr_less_cancel_iff[symmetric, OF ‹1 < p›, THEN iffD2] have "floor ?r = (if i ≥ j * p powr (?fl i - ?fl j) then0 else -1)" (is "_ = ?if") using assms apply simp by (smt (verit, ccfv_SIG) floor_less_iff floor_uminus_of_int le_log_iff mult_powr_eq of_int_1 real_of_int_floor_add_one_gt zero_le_floor) finally show ?thesis by simp qed
lemma truncate_down_rat_precision: "truncate_down prec (real x / real y) = round_down (rat_precision prec x y) (real x / real y)" and truncate_up_rat_precision: "truncate_up prec (real x / real y) = round_up (rat_precision prec x y) (real x / real y)" unfolding truncate_down_def truncate_up_def rat_precision_def by (cases x; cases y) (auto simp: floor_log_divide_eq algebra_simps bitlen_alt_def)
qualified lemma compute_lapprox_posrat[code]: "lapprox_posrat prec x y =
(let
l = rat_precision prec x y;
d = if0≤ l then x * 2^nat l div y else x div 2^nat (- l) div y in normfloat (Float d (- l)))" unfolding div_mult_twopow_eq by transfer (simp add: round_down_def powr_int real_div_nat_eq_floor_of_divide field_simps Let_def truncate_down_rat_precision del: two_powr_minus_int_float)
qualified lemma compute_rapprox_posrat[code]: fixes prec x y defines "l ≡ rat_precision prec x y" shows "rapprox_posrat prec x y =
(let
l = l;
(r, s) = if0≤ l then (x * 2^nat l, y) else (x, y * 2^nat(-l));
d = r div s;
m = r mod s in normfloat (Float (d + (if m = 0∨ y = 0then0 else 1)) (- l)))" proof (cases "y = 0") assume "y = 0" then show ?thesis by transfer simp next assume "y ≠0" show ?thesis proof (cases "0≤ l") case True define x' where "x' = x * 2 ^ nat l" have "int x * 2 ^ nat l = x'" by (simp add: x'_def) moreover have "real x * 2 powr l = real x'" by (simp flip: powr_realpow add: ‹0 ≤ l› x'_def) ultimately show ?thesis using ceil_divide_floor_conv[of y x'] powr_realpow[of 2 "nat l"] ‹0 ≤ l›‹y ≠ 0› l_def[symmetric, THEN meta_eq_to_obj_eq] apply transfer apply (auto simp add: round_up_def truncate_up_rat_precision) apply (metis floor_divide_of_int_eq of_int_of_nat_eq) done next case False define y' where "y' = y * 2 ^ nat (- l)" from ‹y ≠ 0› have "y' ≠0" by (simp add: y'_def) have "int y * 2 ^ nat (- l) = y'" by (simp add: y'_def) moreover have "real x * real_of_int (2::int) powr real_of_int l / real y = x / real y'" using ‹¬ 0 ≤ l› by (simp flip: powr_realpow add: powr_minus y'_def field_simps) ultimately show ?thesis using ceil_divide_floor_conv[of y' x] ‹¬ 0 ≤ l›‹y' ≠ 0›‹y ≠ 0› l_def[symmetric, THEN meta_eq_to_obj_eq] apply transfer apply (auto simp add: round_up_def ceil_divide_floor_conv truncate_up_rat_precision) apply (metis floor_divide_of_int_eq of_int_of_nat_eq) done qed qed
end
lemma rat_precision_pos: assumes "0≤ x" and "0 < y" and "2 * x < y" shows "rat_precision n (int x) (int y) > 0" proof - have "0 < x ==> log 2 x + 1 = log 2 (2 * x)" by (simp add: log_mult) then have "bitlen (int x) < bitlen (int y)" using assms by (simp add: bitlen_alt_def) (auto intro!: floor_mono simp add: one_add_floor) then show ?thesis using assms by (auto intro!: pos_add_strict simp add: field_simps rat_precision_def) qed
lemma rapprox_posrat_less1: "0≤ x ==>0 < y ==>2 * x < y ==> real_of_float (rapprox_posrat n x y) < 1" by transfer (simp add: rat_precision_pos round_up_less1 truncate_up_rat_precision)
lift_definition lapprox_rat :: "nat ==> int ==> int ==> float" is "λprec (x::int) (y::int). truncate_down prec (x / y)" by simp
context begin
qualified lemma compute_lapprox_rat[code]: "lapprox_rat prec x y =
(if y = 0then0
else if0≤ x then
(if0 < y then lapprox_posrat prec (nat x) (nat y)
else - (rapprox_posrat prec (nat x) (nat (-y))))
else
(if0 < y then - (rapprox_posrat prec (nat (-x)) (nat y))
else lapprox_posrat prec (nat (-x)) (nat (-y))))" by transfer (simp add: truncate_up_uminus_eq)
lift_definition rapprox_rat :: "nat ==> int ==> int ==> float" is "λprec (x::int) (y::int). truncate_up prec (x / y)" by simp
lemma "rapprox_rat = rapprox_posrat" by transfer auto
lemma "lapprox_rat = lapprox_posrat" by transfer auto
qualified lemma compute_rapprox_rat[code]: "rapprox_rat prec x y = - lapprox_rat prec (-x) y" by transfer (simp add: truncate_down_uminus_eq)
qualified lemma compute_truncate_down[code]: "truncate_down p (Ratreal r) = (let (a, b) = quotient_of r in lapprox_rat p a b)" by transfer (auto split: prod.split simp: of_rat_divide dest!: quotient_of_div)
qualified lemma compute_truncate_up[code]: "truncate_up p (Ratreal r) = (let (a, b) = quotient_of r in rapprox_rat p a b)" by transfer (auto split: prod.split simp: of_rat_divide dest!: quotient_of_div)
end
subsection ‹Division›
definition "real_divl prec a b = truncate_down prec (a / b)"
definition "real_divr prec a b = truncate_up prec (a / b)"
lift_definition float_divl :: "nat ==> float ==> float ==> float" is real_divl by (simp add: real_divl_def)
lift_definition float_divr :: "nat ==> float ==> float ==> float" is real_divr by (simp add: real_divr_def)
qualified lemma compute_float_divr[code]: "float_divr prec x y = - float_divl prec (-x) y" by transfer (simp add: real_divr_def real_divl_def truncate_down_uminus_eq)
end
subsection ‹Approximate Addition›
definition "plus_down prec x y = truncate_down prec (x + y)"
definition "plus_up prec x y = truncate_up prec (x + y)"
lemma float_plus_down_float[intro, simp]: "x ∈ float ==> y ∈ float ==> plus_down p x y ∈ float" by (simp add: plus_down_def)
lemma float_plus_up_float[intro, simp]: "x ∈ float ==> y ∈ float ==> plus_up p x y ∈ float" by (simp add: plus_up_def)
lemma plus_down: "plus_down prec x y ≤ x + y" and plus_up: "x + y ≤ plus_up prec x y" by (auto simp: plus_down_def truncate_down plus_up_def truncate_up)
lemma float_plus_down: "real_of_float (float_plus_down prec x y) ≤ x + y" and float_plus_up: "x + y ≤ real_of_float (float_plus_up prec x y)" by (transfer; rule plus_down plus_up)+
lemmas plus_down_le = order_trans[OF plus_down] and plus_up_le = order_trans[OF _ plus_up] and float_plus_down_le = order_trans[OF float_plus_down] and float_plus_up_le = order_trans[OF _ float_plus_up]
lemma compute_plus_up[code]: "plus_up p x y = - plus_down p (-x) (-y)" using truncate_down_uminus_eq[of p "x + y"] by (auto simp: plus_down_def plus_up_def)
lemma truncate_down_log2_eqI: assumes "⌊log 2∣x∣⌋ = ⌊log 2∣y∣⌋" assumes "⌊x * 2 powr (p - ⌊log 2∣x∣⌋)⌋ = ⌊y * 2 powr (p - ⌊log 2∣x∣⌋)⌋" shows "truncate_down p x = truncate_down p y" using assms by (auto simp: truncate_down_def round_down_def)
lemma sum_neq_zeroI: "∣a∣≥ k ==>∣b∣ < k ==> a + b ≠0" "∣a∣ > k ==>∣b∣≤ k ==> a + b ≠0" for a k :: real by auto
lemma abs_real_le_2_powr_bitlen[simp]: "∣real_of_int m2∣ < 2 powr real_of_int (bitlen ∣m2∣)" proof (cases "m2 = 0") case True then show ?thesis by simp next case False then have "∣m2∣ < 2 ^ nat (bitlen ∣m2∣)" using bitlen_bounds[of "∣m2∣"] by (auto simp: powr_add bitlen_nonneg) then show ?thesis by (metis bitlen_nonneg powr_int of_int_abs of_int_less_numeral_power_cancel_iff zero_less_numeral) qed
lemma floor_sum_times_2_powr_sgn_eq: fixes ai p q :: int and a b :: real assumes "a * 2 powr p = ai" and b_le_1: "∣b * 2 powr (p + 1)∣≤1" and leqp: "q ≤ p" shows "⌊(a + b) * 2 powr q⌋ = ⌊(2 * ai + sgn b) * 2 powr (q - p - 1)⌋" proof - consider "b = 0" | "b > 0" | "b < 0" by arith then show ?thesis proof cases case 1 then show ?thesis by (simp flip: assms(1) powr_add add: algebra_simps powr_mult_base) next case 2 then have "b * 2 powr p < ∣b * 2 powr (p + 1)∣" by simp also note b_le_1 finally have b_less_1: "b * 2 powr real_of_int p < 1" .
from b_less_1 ‹b > 0› have floor_eq: "⌊b * 2 powr real_of_int p⌋ = 0" "⌊sgn b / 2⌋ = 0" by (simp_all add: floor_eq_iff)
have "⌊(a + b) * 2 powr q⌋ = ⌊(a + b) * 2 powr p * 2 powr (q - p)⌋" by (simp add: algebra_simps flip: powr_realpow powr_add) also have "… = ⌊(ai + b * 2 powr p) * 2 powr (q - p)⌋" by (simp add: assms algebra_simps) also have "… = ⌊(ai + b * 2 powr p) / real_of_int ((2::int) ^ nat (p - q))⌋" using assms by (simp add: algebra_simps divide_powr_uminus flip: powr_realpow powr_add) also have "… = ⌊ai / real_of_int ((2::int) ^ nat (p - q))⌋" by (simp del: of_int_power add: floor_divide_real_eq_div floor_eq) finally have "⌊(a + b) * 2 powr real_of_int q⌋ = ⌊real_of_int ai / real_of_int ((2::int) ^ nat (p - q))⌋" . moreover have "⌊(2 * ai + (sgn b)) * 2 powr (real_of_int (q - p) - 1)⌋ = ⌊real_of_int ai / real_of_int ((2::int) ^ nat (p - q))⌋" proof - have "⌊(2 * ai + sgn b) * 2 powr (real_of_int (q - p) - 1)⌋ = ⌊(ai + sgn b / 2) * 2 powr (q - p)⌋" by (subst powr_diff) (simp add: field_simps) also have "… = ⌊(ai + sgn b / 2) / real_of_int ((2::int) ^ nat (p - q))⌋" using leqp by (simp flip: powr_realpow add: powr_diff) also have "… = ⌊ai / real_of_int ((2::int) ^ nat (p - q))⌋" by (simp del: of_int_power add: floor_divide_real_eq_div floor_eq) finally show ?thesis . qed ultimately show ?thesis by simp next case 3 then have floor_eq: "⌊b * 2 powr (real_of_int p + 1)⌋ = -1" using b_le_1 by (auto simp: floor_eq_iff algebra_simps pos_divide_le_eq[symmetric] abs_if divide_powr_uminus intro!: mult_neg_pos split: if_split_asm) have "⌊(a + b) * 2 powr q⌋ = ⌊(2*a + 2*b) * 2 powr p * 2 powr (q - p - 1)⌋" by (simp add: algebra_simps powr_mult_base flip: powr_realpow powr_add) also have "… = ⌊(2 * (a * 2 powr p) + 2 * b * 2 powr p) * 2 powr (q - p - 1)⌋" by (simp add: algebra_simps) also have "… = ⌊(2 * ai + b * 2 powr (p + 1)) / 2 powr (1 - q + p)⌋" using assms by (simp add: algebra_simps powr_mult_base divide_powr_uminus) also have "… = ⌊(2 * ai + b * 2 powr (p + 1)) / real_of_int ((2::int) ^ nat (p - q + 1))⌋" using assms by (simp add: algebra_simps flip: powr_realpow) also have "… = ⌊(2 * ai - 1) / real_of_int ((2::int) ^ nat (p - q + 1))⌋" using ‹b < 0› assms by (simp add: floor_divide_of_int_eq floor_eq floor_divide_real_eq_div del: of_int_mult of_int_power of_int_diff) also have "… = ⌊(2 * ai - 1) * 2 powr (q - p - 1)⌋" using assms by (simp add: algebra_simps divide_powr_uminus flip: powr_realpow) finally show ?thesis using ‹b < 0› by simp qed qed
lemma log2_abs_int_add_less_half_sgn_eq: fixes ai :: int and b :: real assumes "∣b∣≤1/2" and "ai ≠0" shows "⌊log 2∣real_of_int ai + b∣⌋ = ⌊log 2∣ai + sgn b / 2∣⌋" proof (cases "b = 0") case True then show ?thesis by simp next case False define k where "k = ⌊log 2∣ai∣⌋" then have "⌊log 2∣ai∣⌋ = k" by simp then have k: "2 powr k ≤∣ai∣" "∣ai∣ < 2 powr (k + 1)" by (simp_all add: floor_log_eq_powr_iff ‹ai ≠ 0›) have "k ≥0" using assms by (auto simp: k_def) define r where "r = ∣ai∣ - 2 ^ nat k" have r: "0≤ r" "r < 2 powr k" using ‹k ≥ 0› k by (auto simp: r_def k_def algebra_simps powr_add abs_if powr_int) then have "r ≤ (2::int) ^ nat k - 1" using ‹k ≥ 0› by (auto simp: powr_int) from this[simplified of_int_le_iff[symmetric]] ‹0 ≤ k› have r_le: "r ≤2 powr k - 1" by (auto simp: algebra_simps powr_int) (metis of_int_1 of_int_add of_int_le_numeral_power_cancel_iff)
have "∣ai∣ = 2 powr k + r" using ‹k ≥ 0› by (auto simp: k_def r_def simp flip: powr_realpow)
have pos: "∣b∣ < 1==>0 < 2 powr k + (r + b)" for b :: real using ‹0 ≤ k›‹ai ≠ 0› by (auto simp add: r_def powr_realpow[symmetric] abs_if sgn_if algebra_simps split: if_split_asm) have less: "∣sgn ai * b∣ < 1" and less': "∣sgn (sgn ai * b) / 2∣ < 1" using ‹∣b∣≤ _› by (auto simp: abs_if sgn_if split: if_split_asm)
have floor_eq: "∧b::real. ∣b∣≤1 / 2==> ⌊log 2 (1 + (r + b) / 2 powr k)⌋ = (if r = 0∧ b < 0then -1 else 0)" using ‹k ≥ 0› r r_le by (auto simp: floor_log_eq_powr_iff powr_minus_divide field_simps sgn_if)
from ‹real_of_int ∣ai∣ = _› have "∣ai + b∣ = 2 powr k + (r + sgn ai * b)" using ‹∣b∣≤ _›‹0 ≤ k› r by (auto simp add: sgn_if abs_if) also have "⌊log 2…⌋ = ⌊log 2 (2 powr k + r + sgn (sgn ai * b) / 2)⌋" proof - have "2 powr k + (r + (sgn ai) * b) = 2 powr k * (1 + (r + sgn ai * b) / 2 powr k)" by (simp add: field_simps) also have "⌊log 2…⌋ = k + ⌊log 2 (1 + (r + sgn ai * b) / 2 powr k)⌋" using pos[OF less] by (subst log_mult) (simp_all add: log_mult powr_mult field_simps) also let ?if = "if r = 0∧ sgn ai * b < 0then -1 else 0" have "⌊log 2 (1 + (r + sgn ai * b) / 2 powr k)⌋ = ?if" using ‹∣b∣≤ _› by (intro floor_eq) (auto simp: abs_mult sgn_if) also have "… = ⌊log 2 (1 + (r + sgn (sgn ai * b) / 2) / 2 powr k)⌋" by (subst floor_eq) (auto simp: sgn_if) also have "k + … = ⌊log 2 (2 powr k * (1 + (r + sgn (sgn ai * b) / 2) / 2 powr k))⌋" unfolding int_add_floor using pos[OF less'] ‹∣b∣≤ _› by (simp add: field_simps add_log_eq_powr del: floor_add2) also have "2 powr k * (1 + (r + sgn (sgn ai * b) / 2) / 2 powr k) = 2 powr k + r + sgn (sgn ai * b) / 2" by (simp add: sgn_if field_simps) finally show ?thesis . qed also have "2 powr k + r + sgn (sgn ai * b) / 2 = ∣ai + sgn b / 2∣" unfolding ‹real_of_int ∣ai∣ = _›[symmetric] using ‹ai ≠ 0› by (auto simp: abs_if sgn_if algebra_simps) finally show ?thesis . qed
context begin
qualified lemma compute_far_float_plus_down: fixes m1 e1 m2 e2 :: int and p :: nat defines "k1 ≡ Suc p - nat (bitlen ∣m1∣)" assumes H: "bitlen ∣m2∣≤ e1 - e2 - k1 - 2" "m1 ≠0" "m2 ≠0" "e1 ≥ e2" shows "float_plus_down p (Float m1 e1) (Float m2 e2) =
float_round_down p (Float (m1 * 2 ^ (Suc (Suc k1)) + sgn m2) (e1 - int k1 - 2))" proof - let ?a = "real_of_float (Float m1 e1)" let ?b = "real_of_float (Float m2 e2)" let ?sum = "?a + ?b" let ?shift = "real_of_int e2 - real_of_int e1 + real k1 + 1" let ?m1 = "m1 * 2 ^ Suc k1" let ?m2 = "m2 * 2 powr ?shift" let ?m2' = "sgn m2 / 2" let ?e = "e1 - int k1 - 1"
have "∣?m2∣ * 2 < 2 powr (bitlen ∣m2∣ + ?shift + 1)" by (auto simp: field_simps powr_add powr_mult_base powr_diff abs_mult) also have "…≤2 powr 0" using H by (intro powr_mono) auto finally have abs_m2_less_half: "∣?m2∣ < 1 / 2" by simp
then have "∣real_of_int m2∣ < 2 powr -(?shift + 1)" unfolding powr_minus_divide by (auto simp: bitlen_alt_def field_simps powr_mult_base abs_mult) also have "…≤2 powr real_of_int (e1 - e2 - 2)" by simp finally have b_less_quarter: "∣?b∣ < 1/4 * 2 powr real_of_int e1" by (simp add: powr_add field_simps powr_diff abs_mult) also have "1/4 < ∣real_of_int m1∣ / 2" using ‹m1 ≠ 0› by simp finally have b_less_half_a: "∣?b∣ < 1/2 * ∣?a∣" by (simp add: algebra_simps powr_mult_base abs_mult) then have a_half_less_sum: "∣?a∣ / 2 < ∣?sum∣" by (auto simp: field_simps abs_if split: if_split_asm)
from b_less_half_a have "∣?b∣ < ∣?a∣" "∣?b∣≤∣?a∣" by simp_all
have "∣real_of_float (Float m1 e1)∣≥1/4 * 2 powr real_of_int e1" using ‹m1 ≠ 0› by (auto simp: powr_add powr_int bitlen_nonneg divide_right_mono abs_mult) then have "?sum ≠0" using b_less_quarter by (rule sum_neq_zeroI) then have "?m1 + ?m2 ≠0" unfolding sum_eq by (simp add: abs_mult zero_less_mult_iff)
have "∣real_of_int ?m1∣≥2 ^ Suc k1" "∣?m2'∣ < 2 ^ Suc k1" using ‹m1 ≠ 0›‹m2 ≠ 0› by (auto simp: sgn_if less_1_mult abs_mult simp del: power.simps) then have sum'_nz: "?m1 + ?m2' ≠0" by (intro sum_neq_zeroI)
have "⌊log 2∣real_of_float (Float m1 e1) + real_of_float (Float m2 e2)∣⌋ = ⌊log 2∣?m1 + ?m2∣⌋ + ?e" using ‹?m1 + ?m2 ≠ 0› unfolding floor_add[symmetric] sum_eq by (simp add: abs_mult log_mult) linarith also have "⌊log 2∣?m1 + ?m2∣⌋ = ⌊log 2∣?m1 + sgn (real_of_int m2 * 2 powr ?shift) / 2∣⌋" using abs_m2_less_half ‹m1 ≠ 0› by (intro log2_abs_int_add_less_half_sgn_eq) (auto simp: abs_mult) also have "sgn (real_of_int m2 * 2 powr ?shift) = sgn m2" by (auto simp: sgn_if zero_less_mult_iff less_not_sym) also have "∣?m1 + ?m2'∣ * 2 powr ?e = ∣?m1 * 2 + sgn m2∣ * 2 powr (?e - 1)" by (auto simp: field_simps powr_minus[symmetric] powr_diff powr_mult_base) then have "⌊log 2∣?m1 + ?m2'∣⌋ + ?e = ⌊log 2∣real_of_float (Float (?m1 * 2 + sgn m2) (?e - 1))∣⌋" using ‹?m1 + ?m2' ≠ 0› unfolding floor_add_int by (simp add: log_add_eq_powr abs_mult_pos del: floor_add2) finally have "⌊log 2∣?sum∣⌋ = ⌊log 2∣real_of_float (Float (?m1*2 + sgn m2) (?e - 1))∣⌋" . then have "plus_down p (Float m1 e1) (Float m2 e2) =
truncate_down p (Float (?m1*2 + sgn m2) (?e - 1))" unfolding plus_down_def proof (rule truncate_down_log2_eqI) let ?f = "(int p - ⌊log 2∣real_of_float (Float m1 e1) + real_of_float (Float m2 e2)∣⌋)" let ?ai = "m1 * 2 ^ (Suc k1)" have "⌊(?a + ?b) * 2 powr real_of_int ?f⌋ = ⌊(real_of_int (2 * ?ai) + sgn ?b) * 2 powr real_of_int (?f - - ?e - 1)⌋" proof (rule floor_sum_times_2_powr_sgn_eq) show "?a * 2 powr real_of_int (-?e) = real_of_int ?ai" by (simp add: powr_add powr_realpow[symmetric] powr_diff) show "∣?b * 2 powr real_of_int (-?e + 1)∣≤1" using abs_m2_less_half by (simp add: abs_mult powr_add[symmetric] algebra_simps powr_mult_base) next have "e1 + ⌊log 2∣real_of_int m1∣⌋ - 1 = ⌊log 2∣?a∣⌋ - 1" using ‹m1 ≠ 0› by (simp add: int_add_floor algebra_simps log_mult abs_mult del: floor_add2) also have "…≤⌊log 2∣?a + ?b∣⌋" using a_half_less_sum ‹m1 ≠ 0›‹?sum ≠ 0› unfolding floor_diff_of_int[symmetric] by (auto simp add: log_minus_eq_powr powr_minus_divide intro!: floor_mono) finally have "int p - ⌊log 2∣?a + ?b∣⌋≤ p - (bitlen ∣m1∣) - e1 + 2" by (auto simp: algebra_simps bitlen_alt_def ‹m1 ≠ 0›) also have "…≤ - ?e" using bitlen_nonneg[of "∣m1∣"] by (simp add: k1_def) finally show "?f ≤ - ?e" by simp qed also have "sgn ?b = sgn m2" using powr_gt_zero[of 2 e2] by (auto simp add: sgn_if zero_less_mult_iff simp del: powr_gt_zero) also have "⌊(real_of_int (2 * ?m1) + real_of_int (sgn m2)) * 2 powr real_of_int (?f - - ?e - 1)⌋ = ⌊Float (?m1 * 2 + sgn m2) (?e - 1) * 2 powr ?f⌋" by (simp flip: powr_add powr_realpow add: algebra_simps) finally show "⌊(?a + ?b) * 2 powr ?f⌋ = ⌊real_of_float (Float (?m1 * 2 + sgn m2) (?e - 1)) * 2 powr ?f⌋" . qed then show ?thesis by transfer (simp add: plus_down_def ac_simps Let_def) qed
lemma compute_float_plus_down_naive: "float_plus_down p x y = float_round_down p (x + y)" by transfer (auto simp: plus_down_def)
qualified lemma compute_float_plus_down[code]: fixes p::nat and m1 e1 m2 e2::int shows "float_plus_down p (Float m1 e1) (Float m2 e2) =
(if m1 = 0then float_round_down p (Float m2 e2)
else if m2 = 0then float_round_down p (Float m1 e1)
else
(if e1 ≥ e2 then
(let k1 = Suc p - nat (bitlen ∣m1∣) in if bitlen ∣m2∣ > e1 - e2 - k1 - 2 then float_round_down p ((Float m1 e1) + (Float m2 e2))
else float_round_down p (Float (m1 * 2 ^ (Suc (Suc k1)) + sgn m2) (e1 - int k1 - 2)))
else float_plus_down p (Float m2 e2) (Float m1 e1)))" proof - { assume "bitlen ∣m2∣≤ e1 - e2 - (Suc p - nat (bitlen ∣m1∣)) - 2" "m1 ≠0" "m2 ≠0" "e1 ≥ e2" note compute_far_float_plus_down[OF this] } then show ?thesis by transfer (simp add: Let_def plus_down_def ac_simps) qed
qualified lemma compute_float_plus_up[code]: "float_plus_up p x y = - float_plus_down p (-x) (-y)" using truncate_down_uminus_eq[of p "x + y"] by transfer (simp add: plus_down_def plus_up_def ac_simps)
lemma mantissa_zero: "mantissa 0 = 0" by (fact mantissa_0)
qualified lemma compute_float_less[code]: "a < b ⟷ is_float_pos (float_plus_down 0 b (- a))" using truncate_down[of 0 "b - a"] truncate_down_pos[of "b - a" 0] by transfer (auto simp: plus_down_def)
qualified lemma compute_float_le[code]: "a ≤ b ⟷ is_float_nonneg (float_plus_down 0 b (- a))" using truncate_down[of 0 "b - a"] truncate_down_nonneg[of "b - a" 0] by transfer (auto simp: plus_down_def)
end
lemma plus_down_mono: "plus_down p a b ≤ plus_down p c d" if "a + b ≤ c + d" by (auto simp: plus_down_def intro!: truncate_down_mono that)
lemma plus_up_mono: "plus_up p a b ≤ plus_up p c d" if "a + b ≤ c + d" by (auto simp: plus_up_def intro!: truncate_up_mono that)
subsection ‹Approximate Multiplication›
lemma mult_mono_nonpos_nonneg: "a * b ≤ c * d" if "a ≤ c" "a ≤0" "0≤ d" "d ≤ b" for a b c d::"'a::ordered_ring" by (meson dual_order.trans mult_left_mono_neg mult_right_mono that)
lemma mult_mono_nonneg_nonpos: "b * a ≤ d * c" if "a ≤ c" "c ≤0" "0≤ d" "d ≤ b" for a b c d::"'a::ordered_ring" by (meson dual_order.trans mult_right_mono_neg mult_left_mono that)
lemma mult_mono_nonpos_nonpos: "a * b ≤ c * d" if "a ≥ c" "a ≤0" "b ≥ d" "d ≤0" for a b c d::real by (meson dual_order.trans mult_left_mono_neg mult_right_mono_neg that)
lemma mult_float_mono1: shows "a ≤ b ==> ab ≤ bb ==>
aa ≤ a ==>
b ≤ ba ==>
ac ≤ ab ==>
bb ≤ bc ==>
plus_down prec (nprt aa * pprt bc)
(plus_down prec (nprt ba * nprt bc)
(plus_down prec (pprt aa * pprt ac)
(pprt ba * nprt ac))) ≤ plus_down prec (nprt a * pprt bb)
(plus_down prec (nprt b * nprt bb)
(plus_down prec (pprt a * pprt ab)
(pprt b * nprt ab)))" by (smt (verit, best) mult_mono plus_down_mono add_mono nprt_mono nprt_le_zero zero_le_pprt pprt_mono mult_mono_nonpos_nonneg mult_mono_nonpos_nonpos mult_mono_nonneg_nonpos)
lemma mult_float_mono2: shows "a ≤ b ==>
ab ≤ bb ==>
aa ≤ a ==>
b ≤ ba ==>
ac ≤ ab ==>
bb ≤ bc ==>
plus_up prec (pprt b * pprt bb)
(plus_up prec (pprt a * nprt bb)
(plus_up prec (nprt b * pprt ab)
(nprt a * nprt ab))) ≤ plus_up prec (pprt ba * pprt bc)
(plus_up prec (pprt aa * nprt bc)
(plus_up prec (nprt ba * pprt ac)
(nprt aa * nprt ac)))" by (smt (verit, best) plus_up_mono add_mono mult_mono nprt_mono nprt_le_zero zero_le_pprt pprt_mono mult_mono_nonpos_nonneg mult_mono_nonpos_nonpos mult_mono_nonneg_nonpos)
subsection ‹Approximate Power›
lemma div2_less_self[termination_simp]: "odd n ==> n div 2 < n" for n :: nat by (simp add: odd_pos)
fun power_down :: "nat ==> real ==> nat ==> real" where "power_down p x 0 = 1" | "power_down p x (Suc n) =
(if odd n then truncate_down (Suc p) ((power_down p x (Suc n div 2))2)
else truncate_down (Suc p) (x * power_down p x n))"
fun power_up :: "nat ==> real ==> nat ==> real" where "power_up p x 0 = 1" | "power_up p x (Suc n) =
(if odd n then truncate_up p ((power_up p x (Suc n div 2))2)
else truncate_up p (x * power_up p x n))"
lift_definition power_up_fl :: "nat ==> float ==> nat ==> float" is power_up by (induct_tac rule: power_up.induct) simp_all
lift_definition power_down_fl :: "nat ==> float ==> nat ==> float" is power_down by (induct_tac rule: power_down.induct) simp_all
lemma compute_power_up_fl[code]: "power_up_fl p x 0 = 1" "power_up_fl p x (Suc n) =
(if odd n then float_round_up p ((power_up_fl p x (Suc n div 2))2)
else float_round_up p (x * power_up_fl p x n))" and compute_power_down_fl[code]: "power_down_fl p x 0 = 1" "power_down_fl p x (Suc n) =
(if odd n then float_round_down (Suc p) ((power_down_fl p x (Suc n div 2))2)
else float_round_down (Suc p) (x * power_down_fl p x n))" unfolding atomize_conj by transfer simp
lemma power_down_pos: "0 < x ==>0 < power_down p x n" by (induct p x n rule: power_down.induct) (auto simp del: odd_Suc_div_two intro!: truncate_down_pos)
lemma power_down_nonneg: "0≤ x ==>0≤ power_down p x n" by (induct p x n rule: power_down.induct) (auto simp del: odd_Suc_div_two intro!: truncate_down_nonneg mult_nonneg_nonneg)
lemma power_down: "0≤ x ==> power_down p x n ≤ x ^ n" proof (induct p x n rule: power_down.induct) case (2 p x n) have ?case if "odd n" proof - from that 2 have "(power_down p x (Suc n div 2)) ^ 2≤ (x ^ (Suc n div 2)) ^ 2" by (auto intro: power_mono power_down_nonneg simp del: odd_Suc_div_two) also have "… = x ^ (Suc n div 2 * 2)" by (simp flip: power_mult) also have "Suc n div 2 * 2 = Suc n" using ‹odd n› by presburger finally show ?thesis using that by (auto intro!: truncate_down_le simp del: odd_Suc_div_two) qed then show ?case by (auto intro!: truncate_down_le mult_left_mono 2 mult_nonneg_nonneg power_down_nonneg) qed simp
lemma power_up: "0≤ x ==> x ^ n ≤ power_up p x n" proof (induct p x n rule: power_up.induct) case (2 p x n) have ?case if "odd n" proof - from that even_Suc have "Suc n = Suc n div 2 * 2" by presburger then have "x ^ Suc n ≤ (x ^ (Suc n div 2))2" by (simp flip: power_mult) also from that 2 have "…≤ (power_up p x (Suc n div 2))2" by (auto intro: power_mono simp del: odd_Suc_div_two) finally show ?thesis using that by (auto intro!: truncate_up_le simp del: odd_Suc_div_two) qed then show ?case by (auto intro!: truncate_up_le mult_left_mono 2) qed simp
lemmas power_up_le = order_trans[OF _ power_up] and power_up_less = less_le_trans[OF _ power_up] and power_down_le = order_trans[OF power_down]
lemma power_down_fl: "0≤ x ==> power_down_fl p x n ≤ x ^ n" by transfer (rule power_down)
lemma power_up_fl: "0≤ x ==> x ^ n ≤ power_up_fl p x n" by transfer (rule power_up)
lemma real_power_up_fl: "real_of_float (power_up_fl p x n) = power_up p x n" by transfer simp
lemma real_power_down_fl: "real_of_float (power_down_fl p x n) = power_down p x n" by transfer simp
lemma power_down_even_nonneg: "even n ==>0≤ power_down p x n" by (induct p x n rule: power_down.induct) (auto simp: power_down_simp simp del: odd_Suc_div_two intro!: truncate_down_nonneg )
lemma power_down_eq_zero_iff[simp]: "power_down prec b n = 0⟷ b = 0∧ n ≠0" proof (induction n arbitrary: b rule: less_induct) case (less x) then show ?case using power_down_simp[of _ _ "x - 1"] by (cases x) (auto simp add: div2_less_self) qed
lemma power_down_nonneg_iff[simp]: "power_down prec b n ≥0⟷ even n ∨ b ≥0" proof (induction n arbitrary: b rule: less_induct) case (less x) show ?case using less(1)[of "x - 1" b] power_down_simp[of _ _ "x - 1"] by (cases x) (auto simp: algebra_split_simps zero_le_mult_iff) qed
lemma power_down_neg_iff[simp]: "power_down prec b n < 0⟷
b < 0∧ odd n" using power_down_nonneg_iff[of prec b n] by (auto simp del: power_down_nonneg_iff)
lemma power_down_nonpos_iff[simp]: notes [simp del] = power_down_neg_iff power_down_eq_zero_iff shows "power_down prec b n ≤0⟷ b < 0∧ odd n ∨ b = 0∧ n ≠0" using power_down_neg_iff[of prec b n] power_down_eq_zero_iff[of prec b n] by auto
lemma power_down_mono: "power_down prec a n ≤ power_down prec b n" if "((0≤ a ∧ a ≤ b)∨(odd n ∧ a ≤ b) ∨ (even n ∧ a ≤0∧ b ≤ a))" using that proof (induction n arbitrary: a b rule: less_induct) case (less i) show ?case proof (cases i) case j: (Suc j) note IH = less[unfolded j even_Suc not_not] note [simp del] = power_down.simps show ?thesis proof cases assume [simp]: "even j" have "a * power_down prec a j ≤ b * power_down prec b j" by (metis IH(1) IH(2) ‹even j› lessI linear mult_mono mult_mono' mult_mono_nonpos_nonneg power_down_even_nonneg) then have "truncate_down (Suc prec) (a * power_down prec a j) ≤ truncate_down (Suc prec) (b * power_down prec b j)" by (auto intro!: truncate_down_mono simp: abs_le_square_iff[symmetric] abs_real_def) then show ?thesis unfolding j by (simp add: power_down_simp) next assume [simp]: "odd j" have "power_down prec 0 (Suc (j div 2)) ≤ - power_down prec b (Suc (j div 2))" if "b < 0" "even (j div 2)" by (metis even_Suc le_minus_iff Suc_neq_Zero neg_equal_zero power_down_eq_zero_iff power_down_nonpos_iff that) then have "truncate_down (Suc prec) ((power_down prec a (Suc (j div 2)))2) ≤ truncate_down (Suc prec) ((power_down prec b (Suc (j div 2)))2)" by (smt (verit) IH Suc_less_eq ‹odd j› div2_less_self mult_mono_nonpos_nonpos Suc_neq_Zero power2_eq_square power_down_neg_iff power_down_nonpos_iff power_mono truncate_down_mono) then show ?thesis unfolding j by (simp add: power_down_simp) qed qed simp qed
lemma power_up_even_nonneg: "even n ==>0≤ power_up p x n" by (induct p x n rule: power_up.induct) (auto simp: power_up.simps simp del: odd_Suc_div_two)
lemma power_up_eq_zero_iff[simp]: "power_up prec b n = 0⟷ b = 0∧ n ≠0" proof (induction n arbitrary: b rule: less_induct) case (less x) then show ?case using power_up_simp[of _ _ "x - 1"] by (cases x) (auto simp: algebra_split_simps zero_le_mult_iff div2_less_self) qed
lemma power_up_nonneg_iff[simp]: "power_up prec b n ≥0⟷ even n ∨ b ≥0" proof (induction n arbitrary: b rule: less_induct) case (less x) show ?case using less(1)[of "x - 1" b] power_up_simp[of _ _ "x - 1"] by (cases x) (auto simp: algebra_split_simps zero_le_mult_iff) qed
lemma power_up_neg_iff[simp]: "power_up prec b n < 0⟷ b < 0∧ odd n" using power_up_nonneg_iff[of prec b n] by (auto simp del: power_up_nonneg_iff)
lemma power_up_nonpos_iff[simp]: notes [simp del] = power_up_neg_iff power_up_eq_zero_iff shows "power_up prec b n ≤0⟷ b < 0∧ odd n ∨ b = 0∧ n ≠0" using power_up_neg_iff[of prec b n] power_up_eq_zero_iff[of prec b n] by auto
lemma power_up_mono: "power_up prec a n ≤ power_up prec b n" if "((0≤ a ∧ a ≤ b)∨(odd n ∧ a ≤ b) ∨ (even n ∧ a ≤0∧ b ≤ a))" using that proof (induction n arbitrary: a b rule: less_induct) case (less i) show ?case proof (cases i) case j: (Suc j) note IH = less[unfolded j even_Suc not_not] note [simp del] = power_up.simps show ?thesis proof cases assume [simp]: "even j" have "a * power_up prec a j ≤ b * power_up prec b j" by (metis IH(1) IH(2) ‹even j› lessI linear mult_mono mult_mono' mult_mono_nonpos_nonneg power_up_even_nonneg) then have "truncate_up prec (a * power_up prec a j) ≤ truncate_up prec (b * power_up prec b j)" by (auto intro!: truncate_up_mono simp: abs_le_square_iff[symmetric] abs_real_def) then show ?thesis unfolding j by (simp add: power_up_simp) next assume [simp]: "odd j" have "power_up prec 0 (Suc (j div 2)) ≤ - power_up prec b (Suc (j div 2))" if "b < 0" "even (j div 2)" by (metis Suc_neq_Zero even_Suc neg_0_le_iff_le power_up_eq_zero_iff power_up_nonpos_iff that) then have "truncate_up prec ((power_up prec a (Suc (j div 2)))2) ≤ truncate_up prec ((power_up prec b (Suc (j div 2)))2)" using IH by (auto intro!: truncate_up_mono intro: order_trans[where y=0] simp: abs_le_square_iff[symmetric] abs_real_def div2_less_self) then show ?thesis unfolding j by (simp add: power_up_simp) qed qed simp qed
lemma real_of_Float_int[simp]: "real_of_float (Float n 0) = real n" by simp
lemma float_zero[simp]: "real_of_float (Float 0 e) = 0" by simp
lemma abs_div_2_less: "a ≠0==> a ≠ -1==>∣(a::int) div 2∣ < ∣a∣" by arith
lemma lapprox_rat: "real_of_float (lapprox_rat prec x y) ≤ real_of_int x / real_of_int y" by (simp add: lapprox_rat.rep_eq truncate_down)
lemma mult_div_le: fixes a b :: int assumes "b > 0" shows "a ≥ b * (a div b)" by (smt (verit, ccfv_threshold) assms minus_div_mult_eq_mod mod_int_pos_iff mult.commute)
lemma lapprox_rat_nonneg: assumes "0≤ x" and "0≤ y" shows "0≤ real_of_float (lapprox_rat n x y)" using assms by transfer (simp add: truncate_down_nonneg)
lemma rapprox_rat: "real_of_int x / real_of_int y ≤ real_of_float (rapprox_rat prec x y)" by (simp add: rapprox_rat.rep_eq truncate_up)
lemma rapprox_rat_le1: assumes "0≤ x" "0 < y" "x ≤ y" shows "real_of_float (rapprox_rat n x y) ≤1" using assms by transfer (simp add: truncate_up_le1)
lemma rapprox_rat_nonneg_nonpos: "0≤ x ==> y ≤0==> real_of_float (rapprox_rat n x y) ≤0" by transfer (simp add: truncate_up_nonpos divide_nonneg_nonpos)
lemma rapprox_rat_nonpos_nonneg: "x ≤0==>0≤ y ==> real_of_float (rapprox_rat n x y) ≤0" by transfer (simp add: truncate_up_nonpos divide_nonpos_nonneg)
lemma real_divl: "real_divl prec x y ≤ x / y" by (simp add: real_divl_def truncate_down)
lemma real_divr: "x / y ≤ real_divr prec x y" by (simp add: real_divr_def truncate_up)
lemma float_divl: "real_of_float (float_divl prec x y) ≤ x / y" by transfer (rule real_divl)
lemma real_divl_lower_bound: "0≤ x ==>0≤ y ==>0≤ real_divl prec x y" by (simp add: real_divl_def truncate_down_nonneg)
lemma float_divl_lower_bound: "0≤ x ==>0≤ y ==>0≤ real_of_float (float_divl prec x y)" by transfer (rule real_divl_lower_bound)
lemma exponent_1: "exponent 1 = 0" using exponent_float[of 1 0] by (simp add: one_float_def)
lemma mantissa_1: "mantissa 1 = 1" using mantissa_float[of 1 0] by (simp add: one_float_def)
lemma bitlen_1: "bitlen 1 = 1" by (simp add: bitlen_alt_def)
lemma float_upper_bound: "x ≤2 powr (bitlen ∣mantissa x∣ + exponent x)" proof (cases "x = 0") case True then show ?thesis by simp next case False then have "mantissa x ≠0" using mantissa_eq_zero_iff by auto have "x = mantissa x * 2 powr (exponent x)" by (rule mantissa_exponent) also have "mantissa x ≤∣mantissa x∣" by simp also have "…≤2 powr (bitlen ∣mantissa x∣)" using bitlen_bounds[of "∣mantissa x∣"] bitlen_nonneg ‹mantissa x ≠ 0› by (auto simp del: of_int_abs simp add: powr_int) finally show ?thesis by (simp add: powr_add) qed
lemma float_divl_pos_less1_bound: "0 < real_of_float x ==> real_of_float x ≤1==> prec ≥1==> 1≤ real_of_float (float_divl prec 1 x)" by transfer (rule real_divl_pos_less1_bound)
lemma float_divr: "real_of_float x / real_of_float y ≤ real_of_float (float_divr prec x y)" by (simp add: float_divr.rep_eq real_divr)
lemma real_divr_pos_less1_lower_bound: assumes "0 < x" and "x ≤1" shows "1≤ real_divr prec 1 x" proof - have "1≤1 / x" using ‹0 < x› and ‹x ≤ 1› by auto also have "…≤ real_divr prec 1 x" using real_divr[where x = 1 and y = x] by auto finally show ?thesis by auto qed
lemma float_divr_pos_less1_lower_bound: "0 < x ==> x ≤1==>1≤ float_divr prec 1 x" by transfer (rule real_divr_pos_less1_lower_bound)
lemma real_divr_nonpos_pos_upper_bound: "x ≤0==>0≤ y ==> real_divr prec x y ≤0" by (simp add: real_divr_def truncate_up_nonpos divide_le_0_iff)
lemma float_divr_nonpos_pos_upper_bound: "real_of_float x ≤0==>0≤ real_of_float y ==> real_of_float (float_divr prec x y) ≤0" by transfer (rule real_divr_nonpos_pos_upper_bound)
lemma real_divr_nonneg_neg_upper_bound: "0≤ x ==> y ≤0==> real_divr prec x y ≤0" by (simp add: real_divr_def truncate_up_nonpos divide_le_0_iff)
lemma float_divr_nonneg_neg_upper_bound: "0≤ real_of_float x ==> real_of_float y ≤0==> real_of_float (float_divr prec x y) ≤0" by transfer (rule real_divr_nonneg_neg_upper_bound)
lemma Float_le_zero_iff: "Float a b ≤0⟷ a ≤0" by (auto simp: zero_float_def mult_le_0_iff)
lemma real_of_float_pprt[simp]: fixes a :: float shows "real_of_float (pprt a) = pprt (real_of_float a)" unfolding pprt_def sup_float_def max_def sup_real_def by auto
lemma real_of_float_nprt[simp]: fixes a :: float shows "real_of_float (nprt a) = nprt (real_of_float a)" unfolding nprt_def inf_float_def min_def inf_real_def by auto
context begin
lift_definition int_floor_fl :: "float ==> int" is floor .
qualified lemma compute_int_floor_fl[code]: "int_floor_fl (Float m e) = (if0≤ e then m * 2 ^ nat e else m div (2 ^ (nat (-e))))" apply transfer by (smt (verit, ccfv_threshold) Float.rep_eq compute_real_of_float floor_divide_of_int_eq floor_of_int of_int_1 of_int_add of_int_mult of_int_power)
lift_definition floor_fl :: "float ==> float" is "λx. real_of_int ⌊x⌋" by simp
qualified lemma compute_floor_fl[code]: "floor_fl (Float m e) = (if0≤ e then Float m e else Float (m div (2 ^ (nat (-e)))) 0)" apply transfer using compute_int_floor_fl int_floor_fl.rep_eq powr_int by auto
end
lemma floor_fl: "real_of_float (floor_fl x) ≤ real_of_float x" by transfer simp
lemma int_floor_fl: "real_of_int (int_floor_fl x) ≤ real_of_float x" by transfer simp
lemma floor_pos_exp: "exponent (floor_fl x) ≥0" proof (cases "floor_fl x = 0") case True then show ?thesis by (simp add: floor_fl_def) next case False have eq: "floor_fl x = Float ⌊real_of_float x⌋0" by transfer simp obtain i where "⌊real_of_float x⌋ = mantissa (floor_fl x) * 2 ^ i" "0 = exponent (floor_fl x) - int i" by (rule denormalize_shift[OF eq False]) then show ?thesis by simp qed
lemma compute_mantissa[code]: "mantissa (Float m e) =
(if m = 0then0 else if2 dvd m then mantissa (normfloat (Float m e)) else m)" by (auto simp: mantissa_float Float.abs_eq simp flip: zero_float_def)
lemma compute_exponent[code]: "exponent (Float m e) =
(if m = 0then0 else if2 dvd m then exponent (normfloat (Float m e)) else e)" by (auto simp: exponent_float Float.abs_eq simp flip: zero_float_def)
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.