math / data /lean /OAI /Analysis /AllenCahn /PoissonEstimates.lean
introvoyz045's picture
Migrated from GitHub
4a56cd6 verified
Raw History Blame Contribute Delete
17.8 kB
import OAI.Analysis.AllenCahn.PoissonMaximum
namespace OAI
noncomputable section
namespace PoissonSupport.DeGiorgi
open MeasureTheory
open scoped ENNReal
open Filter Topology
lemma laplacian_poisson_at_max {n : β„•} {f : Space n β†’ ℝ}
(hf : ContDiff ℝ 2 f) (A Ξ΅ : ℝ) {x y : Space n}
(hxy : y β‰  x) (hsmall : β€–y - xβ€– < 1)
(hm : βˆ€ a b : Space n,
f b - f a - poissonPenalty A β€–b - aβ€– - Ξ΅ * (β€–aβ€– ^ 2 + β€–bβ€– ^ 2) ≀
f y - f x - poissonPenalty A β€–y - xβ€– - Ξ΅ * (β€–xβ€– ^ 2 + β€–yβ€– ^ 2)) :
Laplacian.laplacian f y - Laplacian.laplacian f x ≀ -8 * A + 4 * n * Ξ΅ := by
let r := β€–y - xβ€–
have hr : 0 < r := norm_pos_iff.mpr (sub_ne_zero.mpr hxy)
let e : Space n := r⁻¹ β€’ (y - x)
have he : β€–eβ€– = 1 := by simp [e, norm_smul, r, hr.ne']
have her : r β€’ e = y - x := by dsimp [e]; rw [smul_smul, mul_inv_cancelβ‚€ hr.ne', one_smul]
let J := reflect e
let a : Space n := (-2 : ℝ) β€’ e
let g : Space n β†’ ℝ := fun z => r + inner ℝ a z
have hg : ContDiff ℝ 2 g := contDiff_const.add (innerSL ℝ a).contDiff
have hg0 : g 0 = r := by simp [g]
have hdistance : βˆ€ z : Space n, y + J z - (x + z) = g z β€’ e := by
intro z
have hai : inner ℝ a z = (-2 : ℝ) * inner ℝ e z := by
dsimp only [a]
rw [real_inner_smul_left]
change y + reflect e z - (x + z) = (r + inner ℝ a z) β€’ e
rw [reflect_apply he, hai, add_smul, her, neg_mul, neg_smul]
abel
have hnear : βˆ€αΆ  z in 𝓝 (0 : Space n), 0 < g z ∧ g z < 1 :=
((hg.continuous.isOpen_preimage _ isOpen_Ioo).mem_nhds (by simpa [hg0] using And.intro hr hsmall))
let F : Space n β†’ ℝ := fun z => f (y + J z) - f (x + z) -
A * (2 * g z - (g z) ^ 2) - Ξ΅ * (β€–x + zβ€– ^ 2 + β€–y + J zβ€– ^ 2)
have hfy : ContDiff ℝ 2 (fun z => f (y + J z)) :=
hf.comp (contDiff_const.add J.contDiff)
have hfx : ContDiff ℝ 2 (fun z => f (x + z)) := hf.comp (contDiff_const.add contDiff_id)
have hnx : ContDiff ℝ 2 (fun z : Space n => β€–x + zβ€– ^ 2) :=
(contDiff_const.add contDiff_id).norm_sq ℝ
have hny : ContDiff ℝ 2 (fun z : Space n => β€–y + J zβ€– ^ 2) :=
(contDiff_const.add J.contDiff).norm_sq ℝ
have hp : ContDiff ℝ 2 (fun z => A * (2 * g z - (g z) ^ 2)) := by fun_prop
have hq : ContDiff ℝ 2 (fun z => Ξ΅ * (β€–x + zβ€– ^ 2 + β€–y + J zβ€– ^ 2)) := by fun_prop
have hF : ContDiff ℝ 2 F := ((hfy.sub hfx).sub hp).sub hq
have hmax : IsLocalMax F 0 := by
filter_upwards [hnear] with z hz
have hnorm : β€–y + J z - (x + z)β€– = g z := by
rw [hdistance, norm_smul, he, mul_one, Real.norm_eq_abs, abs_of_pos hz.1]
have hh := hm (x + z) (y + J z)
rw [hnorm, poissonPenalty_of_le A hz.2.le, poissonPenalty_of_le A hsmall.le] at hh
simpa [F, hg0, r] using hh
have hl := laplacian_nonpos_of_local_max hF hmax
have hly : Laplacian.laplacian (fun z => f (y + J z)) 0 = Laplacian.laplacian f y := by
have ht : ContDiff ℝ 2 (fun u => f (y + u)) := hf.comp (contDiff_const.add contDiff_id)
simpa only [map_zero, laplacian_translate hf, add_zero] using laplacian_comp_isometry ht J 0
have hns : ContDiff ℝ 2 (fun z : Space n => β€–zβ€– ^ 2) := contDiff_id.norm_sq ℝ
have hlny : Laplacian.laplacian (fun z => β€–y + J zβ€– ^ 2) 0 = 2 * n := by
have ht : ContDiff ℝ 2 (fun u : Space n => β€–y + uβ€– ^ 2) :=
(contDiff_const.add contDiff_id).norm_sq ℝ
simpa only [map_zero, laplacian_translate hns, laplacian_norm_sq] using
laplacian_comp_isometry ht J 0
have hal : β€–aβ€– ^ 2 = 4 := by simp [a, norm_smul, he]; norm_num
have hlg : Laplacian.laplacian g 0 = 0 := laplacian_affine a r 0
have hlgs : Laplacian.laplacian (fun z => g z ^ 2) 0 = 8 := by
rw [show (fun z => g z ^ 2) = (fun z => (r + inner ℝ a z) ^ 2) from rfl,
laplacian_affine_sq, hal]
norm_num
have hg2 : ContDiff ℝ 2 (fun z => 2 * g z) := contDiff_const.mul hg
dsimp only [F] at hl
rw [laplacian_fun_sub ((hfy.sub hfx).sub hp) hq,
laplacian_fun_sub (hfy.sub hfx) hp, laplacian_fun_sub hfy hfx,
hly, laplacian_translate hf,
laplacian_fun_const_mul (hg2.sub (hg.pow 2)),
laplacian_fun_sub hg2 (hg.pow 2),
laplacian_fun_const_mul hg, hlg, hlgs,
laplacian_fun_const_mul (hnx.add hny), laplacian_fun_add hnx hny,
laplacian_translate hns, laplacian_norm_sq, hlny] at hl
simp only [add_zero] at hl
linarith
lemma bounded_poisson_lipschitz {n : β„•} {f : Space n β†’ ℝ}
(hf : ContDiff ℝ 2 f) {M C : ℝ} (hM : 0 ≀ M) (hC : 0 ≀ C)
(hbound : βˆ€ x, |f x| ≀ M) (hlap : βˆ€ x, |Laplacian.laplacian f x| ≀ C) :
βˆ€ x y, |f y - f x| ≀ 2 * (2 * M + C + 1) * β€–y - xβ€– := by
let A := 2 * M + C + 1
have hA : 0 < A := by dsimp [A]; positivity
have hlarge : 2 * M < A := by dsimp [A]; linarith
have honeway : βˆ€ x y, f y - f x ≀ 2 * A * β€–y - xβ€– := by
intro xβ‚€ yβ‚€
by_contra hbad
have hbad' : 2 * A * β€–yβ‚€ - xβ‚€β€– < f yβ‚€ - f xβ‚€ := lt_of_not_ge hbad
let Ξ΄ := f yβ‚€ - f xβ‚€ - poissonPenalty A β€–yβ‚€ - xβ‚€β€–
have hΞ΄ : 0 < Ξ΄ := by
dsimp [Ξ΄]
have hh := poissonPenalty_le_linear hA.le (norm_nonneg (yβ‚€ - xβ‚€))
linarith
let S := β€–xβ‚€β€– ^ 2 + β€–yβ‚€β€– ^ 2
have hS : 0 ≀ S := by dsimp [S]; positivity
let Ξ΅ := min (Ξ΄ / (2 * (S + 1))) (1 / (n + 1 : ℝ))
have hΞ΅ : 0 < Ξ΅ := lt_min (by positivity) (by positivity)
have hΞ΅S : Ξ΅ * (S + 1) ≀ Ξ΄ / 2 := by
have hh := min_le_left (Ξ΄ / (2 * (S + 1))) (1 / (n + 1 : ℝ))
have hh' := (le_div_iffβ‚€ (show 0 < 2 * (S + 1) by positivity)).mp hh
change Ξ΅ * (2 * (S + 1)) ≀ Ξ΄ at hh'
linarith
have hΞ΅n : (n : ℝ) * Ξ΅ ≀ 1 := by
have hh := min_le_right (Ξ΄ / (2 * (S + 1))) (1 / (n + 1 : ℝ))
have hh' := (le_div_iffβ‚€ (show 0 < (n : ℝ) + 1 by positivity)).mp hh
change Ξ΅ * ((n : ℝ) + 1) ≀ 1 at hh'
nlinarith
have hposβ‚€ : 0 < f yβ‚€ - f xβ‚€ - poissonPenalty A β€–yβ‚€ - xβ‚€β€– -
Ξ΅ * (β€–xβ‚€β€– ^ 2 + β€–yβ‚€β€– ^ 2) := by
change 0 < Ξ΄ - Ξ΅ * S
nlinarith
obtain ⟨x, y, hm⟩ := poisson_maximum hf.continuous hM hbound hA.le hΡ
have hp : 0 < f y - f x - poissonPenalty A β€–y - xβ€– - Ξ΅ * (β€–xβ€– ^ 2 + β€–yβ€– ^ 2) :=
hposβ‚€.trans_le (hm xβ‚€ yβ‚€)
have hpen : 0 ≀ Ξ΅ * (β€–xβ€– ^ 2 + β€–yβ€– ^ 2) := by positivity
have hxy : y β‰  x := by
rintro rfl
simp only [sub_self, norm_zero, poissonPenalty, sub_zero, max_eq_left zero_le_one,
one_pow, sub_self, mul_zero] at hp
linarith
have hs : β€–y - xβ€– < 1 := by
by_contra h
rw [poissonPenalty_of_ge A (le_of_not_gt h)] at hp
have hb₁ := (abs_le.mp (hbound x)).1
have hbβ‚‚ := (abs_le.mp (hbound y)).2
linarith
have hl := laplacian_poisson_at_max hf A Ξ΅ hxy hs hm
have hb₁ := (abs_le.mp (hlap y)).1
have hbβ‚‚ := (abs_le.mp (hlap x)).2
dsimp [A] at hl
linarith
intro x y
apply abs_le.mpr
constructor
Β· have h := honeway y x
rw [norm_sub_rev x y] at h
linarith
Β· exact honeway x y
lemma gradient_poisson_at_max_y {n : β„•} {f : Space n β†’ ℝ}
(hf : Differentiable ℝ f) (A Ξ΅ : ℝ) {x y : Space n}
(hxy : y β‰  x) (hs : β€–y - xβ€– < 1)
(hm : βˆ€ b : Space n,
f b - f x - poissonPenalty A β€–b - xβ€– - Ξ΅ * (β€–xβ€– ^ 2 + β€–bβ€– ^ 2) ≀
f y - f x - poissonPenalty A β€–y - xβ€– - Ξ΅ * (β€–xβ€– ^ 2 + β€–yβ€– ^ 2)) :
gradient f y = (2 * A * (1 - β€–y - xβ€–) / β€–y - xβ€–) β€’ (y - x) +
(2 * Ξ΅) β€’ y := by
have hn : HasFDerivAt (fun b : Space n => β€–b - xβ€–)
(β€–y - x‖⁻¹ β€’ innerSL ℝ (y - x)) y := by
simpa only [Function.comp_def, ContinuousLinearMap.comp_id, id_eq] using
(hasFDerivAt_norm_nonzero (sub_ne_zero.mpr hxy)).comp y
((hasFDerivAt_id y).sub_const x)
have hq : HasFDerivAt (fun b : Space n => β€–xβ€– ^ 2 + β€–bβ€– ^ 2)
((2 : ℝ) β€’ innerSL ℝ y) y := by
have hh := (hasStrictFDerivAt_norm_sq y).hasFDerivAt.const_add (β€–xβ€– ^ 2)
rw [← Nat.cast_smul_eq_nsmul ℝ 2, Nat.cast_ofNat] at hh
exact hh
have hp := ((hn.const_mul 2).sub (hn.pow 2)).const_mul A
have hd := (((hf y).hasFDerivAt.sub_const (f x)).fun_sub hp).fun_sub (hq.const_mul Ξ΅)
have hnear : βˆ€αΆ  b : Space n in 𝓝 y, β€–b - xβ€– < 1 :=
(isOpen_lt (continuous_id.sub continuous_const).norm continuous_const).mem_nhds hs
have hmax : IsLocalMax (fun b : Space n => f b - f x -
A * (2 * β€–b - xβ€– - β€–b - xβ€– ^ 2) - Ξ΅ * (β€–xβ€– ^ 2 + β€–bβ€– ^ 2)) y := by
filter_upwards [hnear] with b hb
simpa only [poissonPenalty_of_le A hb.le, poissonPenalty_of_le A hs.le] using hm b
have hz := hmax.hasFDerivAt_eq_zero hd
apply (InnerProductSpace.toDual ℝ (Space n)).injective
ext z
have he := congrArg (fun L : Space n β†’L[ℝ] ℝ => L z) hz
simp only [sub_apply, smul_apply, smul_eq_mul, zero_apply, pow_one,
Nat.cast_ofNat, Nat.reduceSub, nsmul_eq_mul, innerSL_apply_apply] at he
simp only [toDual_gradient, map_add, map_smul, add_apply, smul_apply,
InnerProductSpace.toDual_apply_apply, smul_eq_mul]
change (fderiv ℝ f y) z = _
change (fderiv ℝ f y) z - A * (2 * (β€–y - x‖⁻¹ * inner ℝ (y - x) z) -
(2 * β€–y - xβ€–) * (β€–y - x‖⁻¹ * inner ℝ (y - x) z)) -
Ξ΅ * (2 * inner ℝ y z) = 0 at he
rw [div_eq_mul_inv]
nlinarith
lemma poisson_gradient_bounds {n : β„•} {f : Space n β†’ ℝ}
(hf : Differentiable ℝ f) {M A Ξ΅ : ℝ} (hM : 0 ≀ M)
(hbound : βˆ€ x, |f x| ≀ M) (hA : 0 ≀ A) (hΞ΅ : 0 < Ξ΅)
{x y : Space n} (hxy : y β‰  x) (hs : β€–y - xβ€– < 1)
(hpos : 0 < f y - f x - poissonPenalty A β€–y - xβ€– - Ξ΅ * (β€–xβ€– ^ 2 + β€–yβ€– ^ 2))
(hm : βˆ€ a b : Space n,
f b - f a - poissonPenalty A β€–b - aβ€– - Ξ΅ * (β€–aβ€– ^ 2 + β€–bβ€– ^ 2) ≀
f y - f x - poissonPenalty A β€–y - xβ€– - Ξ΅ * (β€–xβ€– ^ 2 + β€–yβ€– ^ 2)) :
β€–gradient f xβ€– ≀ 2 * A + 2 * Real.sqrt (2 * M * Ξ΅) ∧
β€–gradient f yβ€– ≀ 2 * A + 2 * Real.sqrt (2 * M * Ξ΅) := by
have hpen : Ξ΅ * (β€–xβ€– ^ 2 + β€–yβ€– ^ 2) ≀ 2 * M := by
have hx := (abs_le.mp (hbound x)).1
have hy := (abs_le.mp (hbound y)).2
have hz := poissonPenalty_nonneg hA (norm_nonneg (y - x))
linarith
have hsmall : βˆ€ z : Space n, Ξ΅ * β€–zβ€– ^ 2 ≀ 2 * M β†’
β€–(2 * Ξ΅) β€’ zβ€– ≀ 2 * Real.sqrt (2 * M * Ξ΅) := by
intro z hz
rw [norm_smul, Real.norm_eq_abs, abs_of_nonneg (by positivity : 0 ≀ 2 * Ξ΅)]
have hsq := Real.sq_sqrt (show 0 ≀ 2 * M * Ξ΅ by positivity)
have hc := mul_le_mul_of_nonneg_right hz hΞ΅.le
nlinarith [Real.sqrt_nonneg (2 * M * Ξ΅), norm_nonneg z]
have hxsmall := hsmall x (by nlinarith [mul_nonneg hΞ΅.le (sq_nonneg β€–yβ€–)])
have hysmall := hsmall y (by nlinarith [mul_nonneg hΞ΅.le (sq_nonneg β€–xβ€–)])
have hpbound {a b : Space n} (hab : b β‰  a) (hr : β€–b - aβ€– < 1) :
β€–(2 * A * (1 - β€–b - aβ€–) / β€–b - aβ€–) β€’ (b - a)β€– ≀ 2 * A := by
rw [norm_smul, Real.norm_eq_abs, abs_of_nonneg (by positivity),
div_mul_cancelβ‚€ _ (norm_ne_zero_iff.mpr (sub_ne_zero.mpr hab))]
nlinarith [norm_nonneg (b - a)]
have hgy := gradient_poisson_at_max_y hf A Ξ΅ hxy hs (hm x)
have hnegmax : βˆ€ a b : Space n,
-f b - -f a - poissonPenalty A β€–b - aβ€– - Ξ΅ * (β€–aβ€– ^ 2 + β€–bβ€– ^ 2) ≀
-f x - -f y - poissonPenalty A β€–x - yβ€– - Ξ΅ * (β€–yβ€– ^ 2 + β€–xβ€– ^ 2) := by
intro a b
simpa only [neg_sub_neg, norm_sub_rev, add_comm] using hm b a
have hgx := gradient_poisson_at_max_y hf.neg A Ξ΅ (Ne.symm hxy)
(by simpa only [norm_sub_rev] using hs) (hnegmax y)
have hng : gradient (fun z => -f z) x = -gradient f x := by
simp only [gradient, fderiv_fun_neg, map_neg]
change gradient (fun z => -f z) x = _ at hgx
rw [hng] at hgx
constructor
Β· calc β€–gradient f xβ€– = β€–-gradient f xβ€– := (norm_neg _).symm
_ = β€–(2 * A * (1 - β€–x - yβ€–) / β€–x - yβ€–) β€’ (x - y) + (2 * Ξ΅) β€’ xβ€– := congrArg norm hgx
_ ≀ _ := (norm_add_le _ _).trans (add_le_add
(hpbound (Ne.symm hxy) (by simpa only [norm_sub_rev] using hs)) hxsmall)
Β· rw [hgy]
exact (norm_add_le _ _).trans (add_le_add (hpbound hxy hs) hysmall)
lemma bounded_unit_drift_lipschitz {n : β„•} {f : Space n β†’ ℝ}
(hf : ContDiff ℝ 2 f) (b : Space n β†’ Space n) {M C : ℝ}
(hM : 0 ≀ M) (hC : 0 ≀ C) (hbound : βˆ€ x, |f x| ≀ M)
(hb : βˆ€ x, β€–b xβ€– ≀ 1)
(hpde : βˆ€ x, |Laplacian.laplacian f x - fderiv ℝ f x (b x)| ≀ C) :
βˆ€ x y, |f y - f x| ≀ 2 * (2 * M + C + 3) * β€–y - xβ€– := by
let A := 2 * M + C + 3
have hA : 0 < A := by dsimp [A]; positivity
have hlarge : 2 * M < A := by dsimp [A]; linarith
have honeway : βˆ€ x y, f y - f x ≀ 2 * A * β€–y - xβ€– := by
intro xβ‚€ yβ‚€
by_contra hbad
have hbad' : 2 * A * β€–yβ‚€ - xβ‚€β€– < f yβ‚€ - f xβ‚€ := lt_of_not_ge hbad
let Ξ΄ := f yβ‚€ - f xβ‚€ - poissonPenalty A β€–yβ‚€ - xβ‚€β€–
have hΞ΄ : 0 < Ξ΄ := by
dsimp [Ξ΄]
have hh := poissonPenalty_le_linear hA.le (norm_nonneg (yβ‚€ - xβ‚€))
linarith
let S := β€–xβ‚€β€– ^ 2 + β€–yβ‚€β€– ^ 2
have hS : 0 ≀ S := by dsimp [S]; positivity
let Ξ΅ := min (Ξ΄ / (2 * (S + 1))) (min (1 / (n + 1 : ℝ)) (1 / (8 * M + 1)))
have hΞ΅ : 0 < Ξ΅ := lt_min (by positivity) (lt_min (by positivity) (by positivity))
have hΞ΅S : Ξ΅ * (S + 1) ≀ Ξ΄ / 2 := by
have hh := min_le_left (Ξ΄ / (2 * (S + 1))) (min (1 / (n + 1 : ℝ)) (1 / (8 * M + 1)))
have hh' := (le_div_iffβ‚€ (show 0 < 2 * (S + 1) by positivity)).mp hh
change Ξ΅ * (2 * (S + 1)) ≀ Ξ΄ at hh'
linarith
have hΞ΅n : (n : ℝ) * Ξ΅ ≀ 1 := by
have hh : Ξ΅ ≀ 1 / (n + 1 : ℝ) := (min_le_right _ _).trans (min_le_left _ _)
have hh' := (le_div_iffβ‚€ (show 0 < (n : ℝ) + 1 by positivity)).mp hh
nlinarith
have hΞ΅M : 2 * Real.sqrt (2 * M * Ξ΅) ≀ 1 := by
have hh : Ξ΅ ≀ 1 / (8 * M + 1) := (min_le_right _ _).trans (min_le_right _ _)
have hh' := (le_div_iffβ‚€ (show 0 < 8 * M + 1 by positivity)).mp hh
have hs := Real.sq_sqrt (show 0 ≀ 2 * M * Ξ΅ by positivity)
nlinarith [Real.sqrt_nonneg (2 * M * Ξ΅)]
have hposβ‚€ : 0 < f yβ‚€ - f xβ‚€ - poissonPenalty A β€–yβ‚€ - xβ‚€β€– -
Ξ΅ * (β€–xβ‚€β€– ^ 2 + β€–yβ‚€β€– ^ 2) := by
change 0 < Ξ΄ - Ξ΅ * S
nlinarith
obtain ⟨x, y, hm⟩ := poisson_maximum hf.continuous hM hbound hA.le hΡ
have hp : 0 < f y - f x - poissonPenalty A β€–y - xβ€– - Ξ΅ * (β€–xβ€– ^ 2 + β€–yβ€– ^ 2) :=
hposβ‚€.trans_le (hm xβ‚€ yβ‚€)
have hpen : 0 ≀ Ξ΅ * (β€–xβ€– ^ 2 + β€–yβ€– ^ 2) := by positivity
have hxy : y β‰  x := by
rintro rfl
simp only [sub_self, norm_zero, poissonPenalty, sub_zero, max_eq_left zero_le_one,
one_pow, sub_self, mul_zero] at hp
linarith
have hs : β€–y - xβ€– < 1 := by
by_contra h
rw [poissonPenalty_of_ge A (le_of_not_gt h)] at hp
have hb₁ := (abs_le.mp (hbound x)).1
have hbβ‚‚ := (abs_le.mp (hbound y)).2
linarith
have hl := laplacian_poisson_at_max hf A Ξ΅ hxy hs hm
have hg := poisson_gradient_bounds (hf.differentiable (by norm_num)) hM hbound hA.le hΞ΅ hxy hs hp hm
have hdr (z : Space n) : |fderiv ℝ f z (b z)| ≀ β€–gradient f zβ€– := by
rw [← toDual_gradient]
change |inner ℝ (gradient f z) (b z)| ≀ _
exact (abs_real_inner_le_norm _ _).trans (by nlinarith [norm_nonneg (gradient f z), hb z])
have hx := (abs_le.mp (hpde x)).2
have hy := (abs_le.mp (hpde y)).1
have hd₁ := (abs_le.mp (hdr x)).2
have hdβ‚‚ := (abs_le.mp (hdr y)).1
dsimp [A] at *
linarith
intro x y
apply abs_le.mpr
constructor
Β· have h := honeway y x
rw [norm_sub_rev x y] at h
linarith
Β· exact honeway x y
lemma fderiv_dilation {n : β„•} {f : Space n β†’ ℝ} (hf : Differentiable ℝ f)
(r : ℝ) (x a : Space n) :
fderiv ℝ (fun z => f (r β€’ z)) x a = r * fderiv ℝ f (r β€’ x) a := by
have hd : HasFDerivAt (fun z => f (r β€’ z))
((fderiv ℝ f (r β€’ x)).comp (r β€’ ContinuousLinearMap.id ℝ (Space n))) x := by
simpa only [Function.comp_def, id_eq] using
(hf (r β€’ x)).hasFDerivAt.comp x ((hasFDerivAt_id x).const_smul r)
rw [hd.fderiv]
simp
lemma coordDeriv_dilation {n : β„•} {f : Space n β†’ ℝ} (hf : Differentiable ℝ f)
(r : ℝ) (i : Fin n) (x : Space n) :
coordDeriv (fun z => f (r β€’ z)) i x = r * coordDeriv f i (r β€’ x) :=
fderiv_dilation hf r x (basis n i)
lemma coordDeriv_const_mul {n : β„•} {f : Space n β†’ ℝ} (hf : Differentiable ℝ f)
(r : ℝ) (i : Fin n) (x : Space n) :
coordDeriv (fun z => r * f z) i x = r * coordDeriv f i x := by
dsimp [coordDeriv]
rw [fderiv_const_mul (hf x)]
simp
lemma laplacian_dilation {n : β„•} {f : Space n β†’ ℝ} (hf : ContDiff ℝ 2 f)
(r : ℝ) (x : Space n) :
Laplacian.laplacian (fun z => f (r β€’ z)) x = r ^ 2 * Laplacian.laplacian f (r β€’ x) := by
have hc : ContDiff ℝ 2 (fun z => f (r β€’ z)) := hf.comp (contDiff_id.const_smul r)
rw [laplacian_eq_sum_coordDeriv hc, laplacian_eq_sum_coordDeriv hf, Finset.mul_sum]
apply Finset.sum_congr rfl
intro i _
have he : coordDeriv (fun z => f (r β€’ z)) i = fun z => r * coordDeriv f i (r β€’ z) :=
funext (coordDeriv_dilation (hf.differentiable (by norm_num)) r i)
have hd : Differentiable ℝ (fun z => coordDeriv f i (r β€’ z)) :=
((coordDeriv_contDiff hf i).differentiable (by norm_num)).comp (differentiable_id.const_smul r)
rw [he, coordDeriv_const_mul hd,
coordDeriv_dilation ((coordDeriv_contDiff hf i).differentiable (by norm_num))]
ring
end PoissonSupport.DeGiorgi
end
end OAI