Loomis.loomis_value_eq_aux
theoremverifiedformalizationimported Lean
Loomis induction: lamB0 A B = muB0 A B for any finite positive-B matrix pair with 2 ≤ |I| + |J|, by strong induction on the total dimension.
Statement
set_option autoImplicit true in set_option relaxedAutoImplicit true in open Loomis in open Finset BigOperators in set_option linter.unusedSectionVars false in variable {I J : Type*} [Fintype I] [Fintype J] [Nonempty I] [Nonempty J] in theorem Loomis.loomis_value_eq_aux : ∀ (n : ℕ), 2 ≤ n → ∀ {I J : Type*} [Fintype I] [Fintype J] [Nonempty I] [Nonempty J], n = Fintype.card I + Fintype.card J → ∀ (A B : I → J → ℝ), IsPositive B → lamB0 A B = muB0 A B
- Formalizes
- EconCSLib, `Loomis.loomis_value_eq_aux`
- Lean proof
- By xbei (from the file's git history), from github.com/gametheoryinlean/EconCSLib at
1a88f809b538(Copyright (c) 2026 EconCSLib contributors. All rights reserved, Apache-2.0). Compiled here against leanprover/lean4:v4.33.1, Mathlib v4.33.1; written for leanprover/lean4:v4.30.0.Loomis.loomis_value_eq_aux← Loomis.loomis_value_eq_aux (EconCSLib/Math/Minimax/Loomis.lean:505): adapted
What was changed in
Loomis.loomis_value_eq_aux--- EconCSLib/Math/Minimax/Loomis.lean:505 (Loomis.loomis_value_eq_aux) +++ as compiled here @@ -1,4 +1,8 @@ -/-- **Loomis induction**: `lamB0 A B = muB0 A B` for any finite positive-`B` -matrix pair with `2 ≤ |I| + |J|`, by strong induction on the total dimension. -/ -private theorem loomis_value_eq_aux : +set_option autoImplicit true in +set_option relaxedAutoImplicit true in +open Loomis in +open Finset BigOperators in +set_option linter.unusedSectionVars false in +variable {I J : Type*} [Fintype I] [Fintype J] [Nonempty I] [Nonempty J] in +theorem Loomis.loomis_value_eq_aux : ∀ (n : ℕ), 2 ≤ n → @@ -82,3 +86,3 @@ intro hcardJ - obtain ⟨j, hj⟩ := Finset.card_eq_one.1 (by simpa using hcardJ) + obtain ⟨j, hj⟩ := Finset.card_eq_one.1 (Finset.card_univ.trans hcardJ) have hj0_eq : j₀ = j := by @@ -197,3 +201,3 @@ intro hcardI - obtain ⟨i, hi⟩ := Finset.card_eq_one.1 (by simpa using hcardI) + obtain ⟨i, hi⟩ := Finset.card_eq_one.1 (Finset.card_univ.trans hcardI) have hi0_eq : i₀ = i := by - Verified
- 08 Oct 2026
- Axioms
Classical.choiceQuot.soundpropext- Built on 45
- Loomis.IsPositiveA matrix
B : I → J → ℝis entrywise positive if every entry is> 0. This is the hypothesis driving the general Loomis theorem, and… - Loomis.lamB0Maxmin Loomis scalar
λ₀ = sup_x λ_aux(x). - Loomis.muB0Minmax Loomis scalar
μ₀ = inf_y μ_aux(y). - Loomis.loomis_value_IJ_2Base case of the Loomis induction: a 1×1 matrix pair has the single ratio
A i₀ j₀ / B i₀ j₀as the common Loomis value. - Loomis.lamB0_le_muB0Weak duality for the Loomis scalars:
lamB0 ≤ muB0. - Loomis.xBRow-vector product
(xB)_j = ∑ᵢ xᵢ Bᵢⱼ. - Loomis.xARow-vector product
(xA)_j = ∑ᵢ xᵢ Aᵢⱼ. - Loomis.exists_xx_lamB0Attainment of
lamB0: there exists a mixed strategyxxwith(xA xx)_j ≥ lamB0 · (xB xx)_jfor every column. - Loomis.AyColumn-vector product
(Ay)_i = ∑ⱼ Aᵢⱼ yⱼ. - Loomis.ByColumn-vector product
(By)_i = ∑ⱼ Bᵢⱼ yⱼ. - Loomis.exists_yy_muB0Attainment of
muB0: there exists a mixed strategyyywith(Ay yy)_i ≤ muB0 · (By yy)_ifor every row. - Loomis.colOffsetLinearised column constraint:
colOffset A B λ x j = (xA)_j - λ · (xB)_j. Note that… - Loomis.rowOffsetLinearised row constraint:
rowOffset A B μ y i = μ · (By)_i - (Ay)_i. - wsumWeighted sum of
fwith weights from a simplex elementx. Thinabbrevover Mathlib's finite dot product⬝ᵥ: definitionally equal,… - Loomis.xBy_posPositivity of the bilinear pairing
xBy = ∑ᵢⱼ xᵢ Bᵢⱼ yⱼ. - wsum_wsum_commExchange order of double weighted sums.
- Loomis.xBy_swapThe bilinear pairing
xByand its symmetric variants. - Loomis.wsum_const_mulWeight a constant multiple under
wsum:wsum z (c · f) = c · wsum z f. A direct unfold ofwsum_smul, restated here so chained… - Loomis.lamB.auxPlayer I's guaranteed Loomis ratio under mixed strategy
x: infimum over pure columns. - Loomis.colRatioRow player's per-column Loomis ratio
(xA)_j / (xB)_j. - Loomis.xB_posPositivity of the row aggregate when
Bis entrywise positive. - Loomis.lamB.aux.le_lamB0Every
lamB.auxvalue is bounded by the supremumlamB0. - Loomis.dropColDrop the
j₀-column ofAand view as a matrix onI × {j // j ≠ j₀}. - Loomis.dropCol.IsPositiveDropping a column preserves entrywise positivity.
- Loomis.muB.auxPlayer II's Loomis-ratio cap under mixed strategy
y: supremum over pure rows. - MinimaxLoomis.extendDropColumnExtend a mixed strategy on
J' = {j // j ≠ j₀}to a mixed strategy onJby putting zero mass onj₀. - Loomis.Ay_extendDropColumnExtending
y' ∈ Δ(J')toΔ(J)by zero atj₀recovers the same row aggregates fromA(andB). - Loomis.By_extendDropColumnLoomis.By_extendDropColumn
- Loomis.rowRatioColumn player's per-row Loomis ratio
(Ay)_i / (By)_i. - Loomis.muB.aux.ge_muB0Every
muB.auxvalue dominates the infimummuB0. - mix_gt_of_gt_nbhExistence of a strictly interior
tkeepingt·x + (1-t)·y > c, givenc < x. Constructive over any ordered field: picktjust above… - stdSimplex.mixConvex combination of two simplex points:
mix α hα₀ hα₁ x y = α·x + (1-α)·y. This is the basic vocabulary for compound lotteries and for… - Loomis.colOffset_mixConvex combination linearity for
colOffsetin the simplex argument. - linear_comb_gt_of_ge_gtConvex combination of "≥ c" and "> c" stays "> c" (provided
α ≥ 0andα < 1). - Loomis.lamB.aux_gt_of_colOffset_pos
lamB.aux A B xstrictly exceedslamiff every offsetcolOffsetis strictly positive atx. - Loomis.By_posPositivity of the column aggregate when
Bis entrywise positive. - Loomis.dropRowDrop the
i₀-row ofAand view as a matrix on{i // i ≠ i₀} × J. - Loomis.dropRow.IsPositiveDropping a row preserves entrywise positivity.
- MinimaxLoomis.extendDropRowDual: extend a mixed strategy on
I' = {i // i ≠ i₀}to one onI. - Loomis.xA_extendDropRowLoomis.xA_extendDropRow
- Loomis.xB_extendDropRowLoomis.xB_extendDropRow
- Loomis.rowOffset_mixConvex combination linearity for
rowOffsetin the simplex argument. - Loomis.muB.aux_lt_of_rowOffset_pos
muB.aux A B yis strictly belowmuiff every offsetrowOffsetis strictly positive aty. - stdSimplex.mix_applystdSimplex.mix_apply
- wsum_pure_applyWeighted sum at a point mass evaluates the chosen coordinate.
- Loomis.IsPositiveA matrix
- Used by 1
- Loomis.loomis_value_eqLoomis scalar equality: every finite positive-
Bmatrix pair overℝhas equal maxmin and minmax Loomis ratios.
- Loomis.loomis_value_eqLoomis scalar equality: every finite positive-
Lean source view module on GitHub
Choose a tactic — a marked line — to see the goals Lean had before and after it, as in Lean’s infoview. Step with ← → or j k.
set_option autoImplicit true inset_option relaxedAutoImplicit true inopen Loomis inopen Finset BigOperators inset_option linter.unusedSectionVars false invariable {I J : Type*} [Fintype I] [Fintype J] [Nonempty I] [Nonempty J] in/-- **Loomis induction**: `lamB0 A B = muB0 A B` for any finite positive-`B` matrix pair with `2 ≤ |I| + |J|`, by strong induction on the total dimension. -/theorem Loomis.loomis_value_eq_aux : ∀ (n : ℕ), 2 ≤ n → ∀ {I J : Type*} [Fintype I] [Fintype J] [Nonempty I] [Nonempty J], n = Fintype.card I + Fintype.card J → ∀ (A B : I → J → ℝ), IsPositive B → lamB0 A B = muB0 A B := by intro n Hgt induction n, Hgt using Nat.le_induction with | base => intro I J _ _ _ _ Hn A B hB exact loomis_value_IJ_2 Hn hB | succ n _ IH => intro I J _ _ _ _ Hn A B hB classical rcases (lamB0_le_muB0 A B hB).lt_or_eq with hlt | heq swap · exact heq exfalso obtain ⟨xx, Hxx⟩ := exists_xx_lamB0 A B hB obtain ⟨yy, Hyy⟩ := exists_yy_muB0 A B hB -- The Hxx/Hyy say `colOffset ≥ 0` / `rowOffset ≥ 0` everywhere. have HxxOff : ∀ j, 0 ≤ colOffset A B (lamB0 A B) xx j := by intro j; unfold colOffset; linarith [Hxx j] have HyyOff : ∀ i, 0 ≤ rowOffset A B (muB0 A B) yy i := by intro i; unfold rowOffset; linarith [Hyy i] -- If both are identically zero, we get `lamB0 = muB0`, contradiction. have exits_ij : (∃ j : J, 0 < colOffset A B (lamB0 A B) xx j) ∨ (∃ i : I, 0 < rowOffset A B (muB0 A B) yy i) := by by_contra HP push_neg at HP obtain ⟨HP1, HP2⟩ := HP have HxxZero : ∀ j, colOffset A B (lamB0 A B) xx j = 0 := fun j => le_antisymm (HP1 j) (HxxOff j) have HyyZero : ∀ i, rowOffset A B (muB0 A B) yy i = 0 := fun i => le_antisymm (HP2 i) (HyyOff i) -- These give wsum yy (colOffset xx ·) = 0 and wsum xx (rowOffset yy ·) = 0. set xxByy : ℝ := wsum xx (fun i => By B yy i) have hxxByy_pos : 0 < xxByy := xBy_pos hB xx yy -- wsum yy (xA xx ·) = lamB0 * (wsum yy (xB xx ·)) and -- wsum xx (Ay yy ·) = muB0 * (wsum xx (By yy ·)) = muB0 * xxByy have h_pair_swap : wsum xx (fun i => Ay A yy i) = wsum yy (fun j => xA A xx j) := by unfold Ay xA exact wsum_wsum_comm xx yy A have h_xxByy_swap : wsum yy (fun j => xB B xx j) = xxByy := by show wsum yy (fun j => xB B xx j) = wsum xx (fun i => By B yy i) exact (xBy_swap B xx yy).symm -- From HxxZero: wsum yy (xA xx ·) = lamB0 * wsum yy (xB xx ·) have h_lhs : wsum yy (fun j => xA A xx j) = lamB0 A B * xxByy := by rw [← h_xxByy_swap] have : wsum yy (fun j => xA A xx j) = wsum yy (fun j => lamB0 A B * xB B xx j) := by refine congrArg (wsum yy) (funext ?_) intro j have := HxxZero j unfold colOffset at this linarith rw [this, wsum_const_mul] -- From HyyZero: wsum xx (Ay yy ·) = muB0 * xxByy have h_rhs : wsum xx (fun i => Ay A yy i) = muB0 A B * xxByy := by have : wsum xx (fun i => Ay A yy i) = wsum xx (fun i => muB0 A B * By B yy i) := by refine congrArg (wsum xx) (funext ?_) intro i have := HyyZero i unfold rowOffset at this linarith rw [this, wsum_const_mul] have heq2 : lamB0 A B = muB0 A B := by have hcombo : lamB0 A B * xxByy = muB0 A B * xxByy := by linarith [h_lhs, h_pair_swap, h_rhs] exact mul_right_cancel₀ hxxByy_pos.ne' hcombo linarith rcases exits_ij with ⟨j₀, HJ⟩ | ⟨i₀, HI⟩ · -------------------- Column-drop case -------------------- -- |J| ≥ 2 because strict ineq at j₀ ⇒ otherwise lamB.aux xx > lamB0, -- contradicting lamB.aux ≤ lamB0. have cardJ_ne_one : Fintype.card J ≠ 1 := by intro hcardJ obtain ⟨j, hj⟩ := Finset.card_eq_one.1 (Finset.card_univ.trans hcardJ) have hj0_eq : j₀ = j := by have hmem : j₀ ∈ (Finset.univ : Finset J) := Finset.mem_univ _ rw [hj] at hmem exact Finset.mem_singleton.1 hmem -- lamB.aux xx evaluated at the unique column j₀ have hlamB_xx : lamB.aux A B xx = colRatio A B xx j₀ := by simp [lamB.aux, hj0_eq, hj] have hxxB : 0 < xB B xx j₀ := xB_pos hB xx j₀ have hratio_gt : lamB0 A B < colRatio A B xx j₀ := by unfold colRatio rw [lt_div_iff₀ hxxB] have hHJ := HJ; unfold colOffset at hHJ; linarith have hle : lamB.aux A B xx ≤ lamB0 A B := lamB.aux.le_lamB0 hB xx rw [hlamB_xx] at hle linarith have cardJ_ge_two : 2 ≤ Fintype.card J := by have hpos : 1 ≤ Fintype.card J := Fintype.card_pos omega have nonempty_J' : Nonempty {j : J // j ≠ j₀} := by obtain ⟨j, hj⟩ : ∃ j : J, j ≠ j₀ := by by_contra H1 push_neg at H1 have hsubsing : Fintype.card J ≤ 1 := by have hsingle : (Finset.univ : Finset J) = {j₀} := by ext; simp [H1] simpa [← Finset.card_univ, hsingle] omega exact ⟨⟨j, hj⟩⟩ haveI : Nonempty {j : J // j ≠ j₀} := nonempty_J' have cardn : n = Fintype.card I + Fintype.card {j : J // j ≠ j₀} := by have hJ' : Fintype.card {j : J // j ≠ j₀} = Fintype.card J - 1 := by simp [Fintype.card_subtype_compl] have hposJ : 1 ≤ Fintype.card J := Fintype.card_pos omega -- Apply IH to the column-dropped game. let A' : I → {j : J // j ≠ j₀} → ℝ := dropCol A j₀ let B' : I → {j : J // j ≠ j₀} → ℝ := dropCol B j₀ have hB' : IsPositive B' := dropCol.IsPositive hB j₀ have IH' : lamB0 A' B' = muB0 A' B' := IH cardn A' B' hB' -- Show muB0 A B ≤ muB0 A' B' via the extend-by-zero trick. have h_mu_mono : muB0 A B ≤ muB0 A' B' := by apply le_ciInf intro y' have hwsum_Ay : ∀ i, Ay A (MinimaxLoomis.extendDropColumn j₀ y') i = Ay A' y' i := fun i => Ay_extendDropColumn i A j₀ y' have hwsum_By : ∀ i, By B (MinimaxLoomis.extendDropColumn j₀ y') i = By B' y' i := fun i => By_extendDropColumn i B j₀ y' have hmuA' : muB.aux A' B' y' = muB.aux A B (MinimaxLoomis.extendDropColumn j₀ y') := by simp only [muB.aux] congr 1 ext i unfold rowRatio rw [hwsum_Ay i, hwsum_By i] rw [hmuA'] exact muB.aux.ge_muB0 hB (MinimaxLoomis.extendDropColumn j₀ y') have lamB0_lt_lamB0' : lamB0 A B < lamB0 A' B' := by calc lamB0 A B < muB0 A B := hlt _ ≤ muB0 A' B' := h_mu_mono _ = lamB0 A' B' := IH'.symm -- Get the inductive optimiser `xx'` for the restricted game. obtain ⟨xx', Hxx'⟩ := exists_xx_lamB0 A' B' hB' -- On non-j₀ columns, `colOffset A B lamB0 xx' j > 0`. have HxxOff' : ∀ j : J, j ≠ j₀ → 0 < colOffset A B (lamB0 A B) xx' j := by intro j hj have hxx'_j : lamB0 A' B' * xB B' xx' ⟨j, hj⟩ ≤ xA A' xx' ⟨j, hj⟩ := Hxx' ⟨j, hj⟩ have : lamB0 A B * xB B xx' j < xA A xx' j := by have hxB'_xB : xB B' xx' ⟨j, hj⟩ = xB B xx' j := rfl have hxA'_xA : xA A' xx' ⟨j, hj⟩ = xA A xx' j := rfl have hxBpos : 0 < xB B xx' j := xB_pos hB xx' j calc lamB0 A B * xB B xx' j < lamB0 A' B' * xB B xx' j := by exact (mul_lt_mul_iff_of_pos_right hxBpos).mpr lamB0_lt_lamB0' _ ≤ xA A xx' j := by rw [← hxB'_xB, ← hxA'_xA]; exact hxx'_j unfold colOffset; linarith -- For j = j₀ use the neighborhood-of-1 continuity lemma; for j ≠ j₀ -- the convex combination keeps strict positivity. obtain ⟨t, ht0pos, ht1lt, hstrict_j0_raw⟩ : ∃ t : ℝ, 0 < t ∧ t < 1 ∧ 0 < t * colOffset A B (lamB0 A B) xx j₀ + (1 - t) * colOffset A B (lamB0 A B) xx' j₀ := mix_gt_of_gt_nbh _ _ _ HJ have ht₀ : 0 ≤ t := le_of_lt ht0pos have ht₁ : t ≤ 1 := le_of_lt ht1lt have hstrict_j0 : 0 < colOffset A B (lamB0 A B) (stdSimplex.mix t ht₀ ht₁ xx xx') j₀ := by rw [colOffset_mix]; exact hstrict_j0_raw -- Assemble strict on every j. have hAll : ∀ j, 0 < colOffset A B (lamB0 A B) (stdSimplex.mix t ht₀ ht₁ xx xx') j := by intro j by_cases hj : j = j₀ · rw [hj]; exact hstrict_j0 · rw [colOffset_mix] exact linear_comb_gt_of_ge_gt (colOffset A B (lamB0 A B) xx j) (colOffset A B (lamB0 A B) xx' j) 0 (HxxOff j) (HxxOff' j hj) ht₀ ht1lt -- Hence lamB.aux at the combination strictly exceeds lamB0, contradiction. have hgt : lamB0 A B < lamB.aux A B (stdSimplex.mix t ht₀ ht₁ xx xx') := lamB.aux_gt_of_colOffset_pos hB hAll have hle : lamB.aux A B (stdSimplex.mix t ht₀ ht₁ xx xx') ≤ lamB0 A B := lamB.aux.le_lamB0 hB _ linarith · -------------------- Row-drop case -------------------- have cardI_ne_one : Fintype.card I ≠ 1 := by intro hcardI obtain ⟨i, hi⟩ := Finset.card_eq_one.1 (Finset.card_univ.trans hcardI) have hi0_eq : i₀ = i := by have hmem : i₀ ∈ (Finset.univ : Finset I) := Finset.mem_univ _ rw [hi] at hmem exact Finset.mem_singleton.1 hmem have hmuB_yy : muB.aux A B yy = rowRatio A B yy i₀ := by simp [muB.aux, hi0_eq, hi] have hyyB : 0 < By B yy i₀ := By_pos hB yy i₀ have hratio_lt : rowRatio A B yy i₀ < muB0 A B := by unfold rowRatio rw [div_lt_iff₀ hyyB] have hHI := HI; unfold rowOffset at hHI; linarith have hge : muB0 A B ≤ muB.aux A B yy := muB.aux.ge_muB0 hB yy rw [hmuB_yy] at hge linarith have cardI_ge_two : 2 ≤ Fintype.card I := by have hpos : 1 ≤ Fintype.card I := Fintype.card_pos omega have nonempty_I' : Nonempty {i : I // i ≠ i₀} := by obtain ⟨i, hi⟩ : ∃ i : I, i ≠ i₀ := by by_contra H1 push_neg at H1 have hsubsing : Fintype.card I ≤ 1 := by have hsingle : (Finset.univ : Finset I) = {i₀} := by ext; simp [H1] simpa [← Finset.card_univ, hsingle] omega exact ⟨⟨i, hi⟩⟩ haveI : Nonempty {i : I // i ≠ i₀} := nonempty_I' have cardn : n = Fintype.card {i : I // i ≠ i₀} + Fintype.card J := by have hI' : Fintype.card {i : I // i ≠ i₀} = Fintype.card I - 1 := by simp [Fintype.card_subtype_compl] have hposI : 1 ≤ Fintype.card I := Fintype.card_pos omega let A' : {i : I // i ≠ i₀} → J → ℝ := dropRow A i₀ let B' : {i : I // i ≠ i₀} → J → ℝ := dropRow B i₀ have hB' : IsPositive B' := dropRow.IsPositive hB i₀ have IH' : lamB0 A' B' = muB0 A' B' := IH cardn A' B' hB' have h_lam_mono : lamB0 A' B' ≤ lamB0 A B := by apply ciSup_le intro x' have hwsum_xA : ∀ j, xA A (MinimaxLoomis.extendDropRow i₀ x') j = xA A' x' j := fun j => xA_extendDropRow j A i₀ x' have hwsum_xB : ∀ j, xB B (MinimaxLoomis.extendDropRow i₀ x') j = xB B' x' j := fun j => xB_extendDropRow j B i₀ x' have hlamA' : lamB.aux A' B' x' = lamB.aux A B (MinimaxLoomis.extendDropRow i₀ x') := by simp only [lamB.aux] congr 1 ext j unfold colRatio rw [hwsum_xA j, hwsum_xB j] rw [hlamA'] exact lamB.aux.le_lamB0 hB (MinimaxLoomis.extendDropRow i₀ x') have muB0_gt_muB0' : muB0 A' B' < muB0 A B := by calc muB0 A' B' = lamB0 A' B' := IH'.symm _ ≤ lamB0 A B := h_lam_mono _ < muB0 A B := hlt obtain ⟨yy', Hyy'⟩ := exists_yy_muB0 A' B' hB' have HyyOff' : ∀ i : I, i ≠ i₀ → 0 < rowOffset A B (muB0 A B) yy' i := by intro i hi have hyy'_i : Ay A' yy' ⟨i, hi⟩ ≤ muB0 A' B' * By B' yy' ⟨i, hi⟩ := Hyy' ⟨i, hi⟩ have : Ay A yy' i < muB0 A B * By B yy' i := by have hAy'_Ay : Ay A' yy' ⟨i, hi⟩ = Ay A yy' i := rfl have hBy'_By : By B' yy' ⟨i, hi⟩ = By B yy' i := rfl have hBypos : 0 < By B yy' i := By_pos hB yy' i calc Ay A yy' i = Ay A' yy' ⟨i, hi⟩ := hAy'_Ay.symm _ ≤ muB0 A' B' * By B' yy' ⟨i, hi⟩ := hyy'_i _ = muB0 A' B' * By B yy' i := by rw [hBy'_By] _ < muB0 A B * By B yy' i := (mul_lt_mul_iff_of_pos_right hBypos).mpr muB0_gt_muB0' unfold rowOffset; linarith obtain ⟨t, ht0pos, ht1lt, hstrict_i0_raw⟩ : ∃ t : ℝ, 0 < t ∧ t < 1 ∧ 0 < t * rowOffset A B (muB0 A B) yy i₀ + (1 - t) * rowOffset A B (muB0 A B) yy' i₀ := mix_gt_of_gt_nbh _ _ _ HI have ht₀ : 0 ≤ t := le_of_lt ht0pos have ht₁ : t ≤ 1 := le_of_lt ht1lt have hstrict_i0 : 0 < rowOffset A B (muB0 A B) (stdSimplex.mix t ht₀ ht₁ yy yy') i₀ := by rw [rowOffset_mix]; exact hstrict_i0_raw have hAll : ∀ i, 0 < rowOffset A B (muB0 A B) (stdSimplex.mix t ht₀ ht₁ yy yy') i := by intro i by_cases hi : i = i₀ · rw [hi]; exact hstrict_i0 · rw [rowOffset_mix] exact linear_comb_gt_of_ge_gt (rowOffset A B (muB0 A B) yy i) (rowOffset A B (muB0 A B) yy' i) 0 (HyyOff i) (HyyOff' i hi) ht₀ ht1lt have hlt' : muB.aux A B (stdSimplex.mix t ht₀ ht₁ yy yy') < muB0 A B := muB.aux_lt_of_rowOffset_pos hB hAll have hge : muB0 A B ≤ muB.aux A B (stdSimplex.mix t ht₀ ht₁ yy yy') := muB.aux.ge_muB0 hB _ linarith