YukonModule.ArkLib.ProofSystem.ToyProblem.Impl.FRS.part0
DefinitionYukon_f9f75c203e12e22a1b2b52a1better-codes
Source module ArkLib.ProofSystem.ToyProblem.Impl.FRS.
Definition code
/-
Copyright (c) 2026 ArkLib Contributors. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Authors: Alexander Hicks
-/
import Definitions.Def_Yukon_5ee536a65c1a8881b97bd895
import Definitions.Def_Yukon_7bdfb5c7976bc55dd3e2bf3e
import Mathlib.Analysis.SpecialFunctions.Pow.NNReal
import Mathlib.Tactic.ComputeDegree
import Mathlib.FieldTheory.Finite.Basic
import Mathlib.RingTheory.RootsOfUnity.PrimitiveRoots
import Mathlib.Algebra.Order.Ring.Star
import Mathlib.NumberTheory.LucasPrimality
import Mathlib.Tactic.ReduceModChar
import Init
import Mathlib.Tactic.LinearCombination
import Mathlib.Tactic.FieldSimp
import Mathlib.Algebra.Polynomial.FieldDivision
import Mathlib.Data.ZMod.Basic
import Mathlib.NumberTheory.Zsqrtd.GaussianInt
import Mathlib.Algebra.EuclideanDomain.Int
import Mathlib.Algebra.Lie.OfAssociative
import Mathlib.RingTheory.PrincipalIdealDomain
import Mathlib.RingTheory.UniqueFactorizationDomain.Defs
import Mathlib.RingTheory.Henselian
import Mathlib.Data.Nat.GCD.Basic
import Mathlib.Data.ENNReal.Inv
import Mathlib.Data.ENat.Basic
import Mathlib.Data.ENat.Defs
import Mathlib.Data.Nat.Cast.Order.Field
import Mathlib.Algebra.CharP.Defs
import Mathlib.Data.NNReal.Basic
import Mathlib.Data.NNReal.Defs
import Mathlib.Algebra.BigOperators.Fin
import Mathlib.Algebra.Order.BigOperators.Group.Finset
import Mathlib.Data.Finsupp.Basic
import Mathlib.Data.Nat.Digits.Defs
import Mathlib.Data.Nat.Bitwise
import Mathlib.Algebra.BigOperators.Ring.Finset
import Mathlib.Tactic.IntervalCases
import Mathlib.Order.Interval.Finset.Nat
import Mathlib.Data.Fintype.BigOperators
import Mathlib.Algebra.Ring.Regular
import Mathlib.Algebra.Order.Star.Basic
import Mathlib.RingTheory.AdjoinRoot
import Mathlib.LinearAlgebra.Lagrange
import Mathlib.LinearAlgebra.StdBasis
import Mathlib.Order.CompletePartialOrder
import Mathlib.LinearAlgebra.FreeModule.StrongRankCondition
import Mathlib.LinearAlgebra.FiniteDimensional.Defs
import Mathlib.Algebra.BigOperators.Finsupp.Fin
import Mathlib.Data.Finsupp.Fin
import Mathlib.Algebra.MvPolynomial.Equiv
import Mathlib.RingTheory.MvPolynomial.Basic
import Mathlib.RingTheory.Polynomial.Basic
import Mathlib.Tactic.DepRewrite
import Mathlib.Data.Fin.Basic
import Batteries.Data.Fin.Fold
import Mathlib.Tactic.FinCases
import Mathlib.Data.Fin.Tuple.Take
import Mathlib.Algebra.Polynomial.Eval.Defs
import Mathlib.Algebra.Order.Sub.Basic
import Mathlib.Algebra.Order.Ring.Nat
import Mathlib.Algebra.MvPolynomial.Eval
import Mathlib.Data.FinEnum
import Mathlib.Algebra.Group.Action.Pointwise.Finset
import Mathlib.Algebra.Polynomial.Roots
import Mathlib.Algebra.MvPolynomial.Degrees
import Mathlib.Algebra.MvPolynomial.SchwartzZippel
import Mathlib.RingTheory.SimpleRing.Principal
import Mathlib.RingTheory.RegularLocalRing.Defs
import Mathlib.RingTheory.PicardGroup
import Mathlib.LinearAlgebra.FreeModule.PID
import Mathlib.FieldTheory.Finiteness
import Mathlib.Algebra.Order.Chebyshev
import Mathlib.Data.Real.ENatENNReal
import Mathlib.Topology.MetricSpace.Infsep
import Mathlib.Tactic.Qify
import Mathlib.InformationTheory.Hamming
import Mathlib.Data.ENat.Lattice
import Mathlib.Analysis.InnerProductSpace.PiL2
import Mathlib.Algebra.Order.Floor.Semifield
import Mathlib.LinearAlgebra.AffineSpace.Combination
import Mathlib.LinearAlgebra.AffineSpace.Pointwise
import Mathlib.LinearAlgebra.Matrix.Rank
set_option backward.isDefEq.respectTransparency.types false
/-!
# Folded Reed--Solomon reference implementations for the toy problem
This module supplies two concrete KoalaBear sextic folded-RS encoders. It proves
admissibility of their geometric-progression domains, injectivity, exact encoder
range, minimum relative distance, and small exact numerical reference facts.
Application-specific parameter policy is intentionally outside this module.
## References
* [Arnon, G., Boneh, D., and Fenzi, G., *Open Problems in List Decoding and
Correlated Agreement*][ABF26], §6.4 (the folded-RS instantiation whose
parameter shape these reference points mirror).
-/
namespace ToyProblem
namespace Impl.FRS
open scoped NNReal ENNReal
open Polynomial ReedSolomon.Folded Code
set_option maxRecDepth 8000 in
/-- **An exact-order field element.** There exists `γ : KoalaBear.Ext6`
with `γ ≠ 0` and multiplicative order exactly `2^21`. KoalaBear's prime
`q = 2^31 - 2^24 + 1` has `2^24 | q - 1`, so the multiplicative group of
`KoalaBear.Ext6 = 𝔽_{q^6}` (cyclic of order `q^6 - 1`, divisible by `q - 1`) contains an
element of order exactly `2^21`. The proof is pure finite-group theory plus `ℕ`
arithmetic: a generator of `KoalaBear.Ext6ˣ` has order `q^6 - 1`, and the power
`g ^ ((q^6 - 1) / 2^21)` has order `2^21` (`orderOf_pow'` + `Nat.div_div_self`); its unit
coercion is the witness. Ext6 arithmetic is computable; only the selected element and the
definitions depending on that choice are noncomputable.
**This single fact is the structural keystone behind both geometric-progression FRS reference
rows.** From it, domain injectivity, `(L, s)`-admissibility (`admissible_domain` /
`admissible_domain12`), encoder injectivity (`encoder_injective` /
`encoder12_injective`), and the folded minimum distance (`minRelHammingDistCode_range_encoder` /
`minRelHammingDistCode_range_encoder12`) all derive from this theorem. Both `s = 32` and
`s = 2^12` use the
*same* `γ` (each needs only order `≥ s · |L| = 2^21`). -/
theorem gamma_exists : ∃ γ : KoalaBear.Ext6, γ ≠ 0 ∧ orderOf γ = 2 ^ 21 := by
-- Abstract group theory: the multiplicative group `KoalaBear.Ext6ˣ` of the finite field
-- `𝔽_{q^6}` is cyclic of order `q^6 - 1`. Since `q - 1 = 2^24 · 127` and
-- `(q - 1) ∣ (q^6 - 1)`, we have `2^21 ∣ q^6 - 1`, so a generator's appropriate power
-- has order exactly `2^21`; its unit coercion is the desired nonzero element.
classical
obtain ⟨g, hg⟩ := IsCyclic.exists_generator (α := KoalaBear.Ext6ˣ)
set n : ℕ := orderOf g with hn_def
have hcardK : Fintype.card KoalaBear.Ext6 = KoalaBear.fieldSize ^ 6 := by
exact KoalaBear.card_ext6
-- A generator's order is the cardinality of the unit group, `q^6 - 1`.
have hg_ord : n = KoalaBear.fieldSize ^ 6 - 1 := by
have h1 : orderOf g = Nat.card KoalaBear.Ext6ˣ :=
orderOf_eq_card_of_forall_mem_zpowers hg
rw [hn_def, h1, Nat.card_eq_fintype_card, Fintype.card_units, hcardK]
-- `2^21 ∣ q - 1` (concretely `q - 1 = 2^21 · 1016`) and `(q - 1) ∣ (q^6 - 1)`.
have hdvd_q : (2 ^ 21 : ℕ) ∣ KoalaBear.fieldSize - 1 := by
norm_num [KoalaBear.fieldSize]
have hdvd_pow : (KoalaBear.fieldSize - 1) ∣ (KoalaBear.fieldSize ^ 6 - 1) := by
simpa using Nat.sub_one_dvd_pow_sub_one KoalaBear.fieldSize 6
have hdvd : (2 ^ 21 : ℕ) ∣ n := by rw [hg_ord]; exact hdvd_q.trans hdvd_pow
have hn_pos : n ≠ 0 := by rw [hg_ord]; norm_num [KoalaBear.fieldSize]
-- `u := g ^ (n / 2^21)` has order `n / gcd(n, n/2^21) = n / (n/2^21) = 2^21`.
refine ⟨(g ^ (n / 2 ^ 21) : KoalaBear.Ext6ˣ), Units.ne_zero _, ?_⟩
rw [orderOf_units]
have hm_ne : n / 2 ^ 21 ≠ 0 := by
have := Nat.div_pos (Nat.le_of_dvd (Nat.pos_of_ne_zero hn_pos) hdvd) (by norm_num)
omega
rw [orderOf_pow' g hm_ne, ← hn_def, Nat.gcd_eq_right (Nat.div_dvd_of_dvd hdvd),
Nat.div_div_self hdvd hn_pos]
/-- The shared high-order generator `γ` of `gamma_exists`. -/
noncomputable def gamma : KoalaBear.Ext6 := gamma_exists.choose
lemma gamma_ne_zero : gamma ≠ 0 := gamma_exists.choose_spec.1
lemma orderOf_gamma : orderOf gamma = 2 ^ 21 := gamma_exists.choose_spec.2
/-- Powers of `γ` below its exact order `2^21` are pinned by their exponent: this is the
`pow`-injectivity on `Set.Iio (orderOf γ)` that turns every distinctness side condition of
the progression construction into a `ℕ`-arithmetic fact (dischargeable by `omega`). -/
lemma gamma_pow_left_inj {a b : ℕ} (ha : a < 2 ^ 21) (hb : b < 2 ^ 21)
(h : gamma ^ a = gamma ^ b) : a = b :=
pow_injOn_Iio_orderOf
(Set.mem_Iio.mpr (by simpa [orderOf_gamma] using ha))
(Set.mem_Iio.mpr (by simpa [orderOf_gamma] using hb)) h
/-- The folding multiplier `ω := γ` for the `s = 32` folded RS code — the shared
exact-order generator (`gamma`). With the geometric-progression domain
`domain j = γ^(32·j)`, the folded points are `γ^(32·j) · γ^i = γ^(32·j + i)`
over `32·j + i < orderOf γ = 2^21`, so `(L, 32)`-admissibility holds
(`admissible_domain`) rather than being a documentary placeholder. -/
noncomputable def foldOmega : KoalaBear.Ext6 := gamma
/-- A neutral `2^16`-point folded-RS evaluation domain in Ext6, given by the geometric
progression `j ↦ γ^(32·j)`. Injectivity is `pow`-injectivity of `γ` below its
exact order: `32·j < 32·2^16 = orderOf γ` (`gamma_pow_left_inj`). The
progression is zero-free (`γ ≠ 0`, so every power is a unit), so the
admissibility intra-orbit clause `α · ω^i ≠ α` is not vacuously false at `0`;
here it holds outright via the order bound.
This proof-only reference point does not claim the smooth base-field-domain provenance required
by ABF26's concrete folded-RS profile; that protected profile is intentionally downstream. -/
noncomputable def domain : Fin (2 ^ 16) ↪ KoalaBear.Ext6 where
toFun j := gamma ^ (32 * j.val)
inj' i j hij := by
have hi : 32 * (i : ℕ) < 2 ^ 21 := by have := i.isLt; omega
have hj : 32 * (j : ℕ) < 2 ^ 21 := by have := j.isLt; omega
exact Fin.val_injective (by have := gamma_pow_left_inj hi hj hij; omega)
open Classical in
/-- **`(L, 32)`-admissibility of the progression domain.** The `32 · 2^16 = 2^21` folded points
`γ^(32·a) · γ^i = γ^(32·a + i)` are pairwise distinct because all exponents lie below
`orderOf γ = 2^21`: both `Admissible` conjuncts reduce, via `gamma_pow_left_inj`, to
`ℕ`-arithmetic facts about `32·a + i` (`omega`) and the order bound
`orderOf_gamma`. -/
lemma admissible_domain :
ReedSolomon.Folded.Admissible (Finset.univ.map domain) 32 foldOmega := by
refine ⟨?_, ?_⟩
· intro α hα β hβ hαβ i hi hcontra
simp only [Finset.mem_map, Finset.mem_univ, true_and] at hα hβ
obtain ⟨a, rfl⟩ := hα
obtain ⟨b, rfl⟩ := hβ
simp only [domain, foldOmega, Function.Embedding.coeFn_mk, ← pow_add] at hαβ hcontra
have ha := a.isLt; have hb := b.isLt
have hab : (a : ℕ) ≠ (b : ℕ) := fun h => hαβ (by rw [h])
have := gamma_pow_left_inj (a := 32 * a.val + i) (b := 32 * b.val) (by omega) (by omega)
hcontra
omega
· intro α hα i hipos hi hcontra
simp only [Finset.mem_map, Finset.mem_univ, true_and] at hα
obtain ⟨a, rfl⟩ := hα
simp only [domain, foldOmega, Function.Embedding.coeFn_mk, ← pow_add] at hcontra
have ha := a.isLt
have := gamma_pow_left_inj (a := 32 * a.val + i) (b := 32 * a.val) (by omega) (by omega)
hcontra
omega
/-- The neutral `s = 32` folded encoder: the degree-`< 2^20` folded Reed–Solomon
evaluation map on the `2^16` points of `domain` with folding `s = 32`
(`k = 2^20`, `|L| = 2^16`, `s = 2^5`, rate `ρ = 1/2`), as an `F`-linear map
`(Fin 2^20 → F) →ₗ (Fin 2^16 → Fin 32 → F)`. Built as
`frsEvalOnPoints ∘ (degreeLTEquiv).symm`, with
`ReedSolomon.Folded.frsEvalOnPoints` in place of the scalar `s = 1`
`evalOnPoints`. The codeword alphabet is `A = Fin 32 → KoalaBear.Ext6`. -/
noncomputable def encoder :
(Fin (2 ^ 20) → KoalaBear.Ext6) →ₗ[KoalaBear.Ext6] (Fin (2 ^ 16) → Fin 32 → KoalaBear.Ext6) :=
(frsEvalOnPoints domain 32 foldOmega).domRestrict
(Polynomial.degreeLT KoalaBear.Ext6 (2 ^ 20))
∘ₗ (Polynomial.degreeLTEquiv KoalaBear.Ext6 (2 ^ 20)).symm.toLinearMap
open Classical in
/-- **Injectivity of the folded encoder** (the "code as the
injective map"). Mathematically this would follow from `(L, s)`-admissibility of
`foldOmega` (`ReedSolomon.Folded.Admissible`, the GR08 condition that the `s·|L|`
folded evaluation points `{α · ω^i}` are pairwise distinct) together with
`k ≤ s·|L|` (here `2^20 ≤ 32·2^16 = 2^21`): a degree-`< k` polynomial vanishing
on `s·|L| ≥ k` distinct points is zero, so the unfolded evaluation — hence
`frsEvalOnPoints` on `degreeLT k` — would be injective. `domain` is
zero-free precisely so `Admissible foldOmega` is not *provably false* (its
intra-orbit clause fails at `0`; see `domain`).
This is a direct derivation through the in-tree bridge
`ReedSolomon.Folded.frsEvalOnPoints_domRestrict_injective` (the `Admissible ω → injective`
bridge that also backs the FRS dimension formula `dim_frsCode`): the encoder is
`(injective domRestrict) ∘ (injective degreeLTEquiv.symm)`. The `domRestrict` injectivity
consumes `admissible_domain`, `gamma_ne_zero`, and `k = 2^20 ≤ 32 · 2^16 =
2^21 = s · |ι|`. The order witness `gamma_exists` is proved using abstract cyclicity. -/
theorem encoder_injective : Function.Injective encoder := by
haveI : NeZero (2 ^ 20 : ℕ) := ⟨by norm_num⟩
simp only [encoder, LinearMap.coe_comp, LinearEquiv.coe_toLinearMap]
refine (ReedSolomon.Folded.frsEvalOnPoints_domRestrict_injective (k := 2 ^ 20) (s := 32)
domain foldOmega admissible_domain gamma_ne_zero ?_).comp
(LinearEquiv.injective _)
rw [Fintype.card_fin]; norm_num
open Classical in
set_option maxRecDepth 8000 in
/-- **The folded encoder's image is exactly the folded RS code** `FRS[domain, 2^20,
32, foldOmega]`. The FRS counterpart of `koalaEnc_range`: `encoder = frsEvalOnPoints ∘
(degreeLTEquiv).symm`, and as the latter ranges over all degree-`< 2^20` polynomials its
image is `(degreeLT 2^20).map (frsEvalOnPoints …) = frsCode …`. This identifies
`Set.range ⇑encoder` with `frsCode`, unlocking the folded MDS distance below. -/
theorem encoder_range :
Set.range ⇑encoder =
(↑(ReedSolomon.Folded.frsCode domain (2 ^ 20) 32 foldOmega) :
Set (Fin (2 ^ 16) → Fin 32 → KoalaBear.Ext6)) := by
ext y
rw [SetLike.mem_coe, ReedSolomon.Folded.frsCode, Submodule.mem_map]
simp only [Set.mem_range]
constructor
· rintro ⟨m, rfl⟩
refine ⟨↑((Polynomial.degreeLTEquiv KoalaBear.Ext6 (2 ^ 20)).symm m), Submodule.coe_mem _, ?_⟩
simp only [encoder, LinearMap.coe_comp, LinearEquiv.coe_toLinearMap, Function.comp_apply,
LinearMap.domRestrict_apply]
· rintro ⟨p, hp, rfl⟩
refine ⟨Polynomial.degreeLTEquiv KoalaBear.Ext6 (2 ^ 20) ⟨p, hp⟩, ?_⟩
simp only [encoder, LinearMap.coe_comp, LinearEquiv.coe_toLinearMap, Function.comp_apply,
LinearEquiv.symm_apply_apply, LinearMap.domRestrict_apply]
open Classical in
/-- **Folded-RS minimum relative distance**, derived through the in-tree folded-distance
bridge `ReedSolomon.Folded.minDist_frsCode`.
`minRelHammingDistCode (Set.range ⇑encoder) = 32769/65536`, the folded-Singleton (MDS-type)
distance for `FRS[F, L, 2^20, 32, ω]`: a nonzero degree-`< 2^20` polynomial has `< 2^20`
roots, so it vanishes on `≤ ⌊(2^20-1)/32⌋ = 32767` *whole* folded symbols, hence the folded
Hamming distance is `D = |L| − 32767 = 65536 − 32767 = 32769` and `δ_min = D/|L| =
32769/65536`. Via `encoder_range` (the code *is* `frsCode`), `minDist_frsCode`, and the
absolute-to-relative bridge `minDist_div_card_eq_minRelHammingDistCode`. -/
theorem minRelHammingDistCode_range_encoder :
minRelHammingDistCode (Set.range ⇑encoder) = (32769 / 65536 : ℚ≥0) := by
haveI : Nonempty (Fin (2 ^ 16)) := inferInstance
haveI : NeZero (2 ^ 20 : ℕ) := ⟨by norm_num⟩
have hcode : Set.range ⇑encoder =
(↑(ReedSolomon.Folded.frsCode domain (2 ^ 20) 32 foldOmega) :
Set (Fin (2 ^ 16) → Fin 32 → KoalaBear.Ext6)) := encoder_range
have hmin : Code.minDist (Set.range ⇑encoder) = 32769 := by
have h := ReedSolomon.Folded.minDist_frsCode (k := 2 ^ 20) (s := 32) (by norm_num)
domain foldOmega admissible_domain gamma_ne_zero
(by rw [Fintype.card_fin]; norm_num)
simp only [Fintype.card_fin] at h
norm_num at h
rw [hcode]; exact h
have hbridge := minDist_div_card_eq_minRelHammingDistCode (Set.range ⇑encoder)
have hcardι : Fintype.card (Fin (2 ^ 16)) = 65536 := by
rw [Fintype.card_fin]; norm_num
rw [hmin, hcardι] at hbridge
have hQ : ((minRelHammingDistCode (Set.range ⇑encoder) : ℚ≥0) : ℚ) =
((32769 / 65536 : ℚ≥0) : ℚ) := by
rw [← hbridge]; push_cast; norm_num
exact_mod_cast hQ
/-- **Exact numerical lower-bound reference:** `2^(-128.01) ≤
(32767/65536)^128`. The crux of the folded MDS sweep-floor. Split off `2^(-128)`:
since `32767/65536 = (32767/32768)·(1/2)`, this reduces to `2^(-0.01) ≤
(32767/32768)^128`. Bernoulli (`one_add_mul_le_pow`) gives `(32767/32768)^128 =
(1 - 1/32768)^128 ≥ 1 - 128/32768 = 255/256`, and `2^(-0.01) ≤ 255/256` is the
**proven integer inequality** `256^100 ≤ 2·255^100` (`log₁₀`: `100·2.4082 =
240.82 ≤ 0.301 + 240.65`). No float `#eval`. (True value `(32767/65536)^128 ≈
2^(-128.006)`, comfortably above the `2^(-128.01)` ceiling.) -/
theorem two_rpow_le_spotcheckPow :
(2 : ℝ≥0) ^ (-(128.01 : ℝ)) ≤ ((32767 : ℝ≥0) / 65536) ^ (128 : ℕ) := by
rw [← NNReal.coe_le_coe]
push_cast [NNReal.coe_rpow]
rw [show (-128.01 : ℝ) = (-0.01 : ℝ) + (-(128 : ℝ)) by norm_num]
rw [Real.rpow_add (by norm_num : (0:ℝ) < 2)]
rw [show (32767:ℝ)/65536 = (32767/32768) * (1/2) by ring]
rw [mul_pow]
rw [show ((1:ℝ)/2)^128 = 2^(-(128:ℝ)) by
rw [Real.rpow_neg (by norm_num), show (128:ℝ) = ((128:ℕ):ℝ) by norm_num,
Real.rpow_natCast]; norm_num]
apply mul_le_mul_of_nonneg_right _ (by positivity)
have hbern : (255 : ℝ) / 256 ≤ (32767/32768 : ℝ)^128 := by
have := one_add_mul_le_pow (a := (-1/32768:ℝ)) (by norm_num) 128
linarith
have h2neg : (2:ℝ)^(-0.01:ℝ) ≤ 255/256 := by
apply le_of_pow_le_pow_left₀ (n := 100) (by norm_num) (by positivity)
rw [← Real.rpow_natCast ((2:ℝ)^(-0.01:ℝ)) 100]
rw [← Real.rpow_mul (by norm_num : (0:ℝ) ≤ 2)]
norm_num
linarith [hbern, h2neg]
/-! ## Large-folding reference at `s = 2^12 = 4096`
This second concrete parameter point keeps the message dimension `k = 2^20`
and unfolded length `s · |L| = 2^21`, so `s = 2^12` gives `|L| = 2^9 = 512`.
The folded MDS distance is
`δ_min = (|L| − ⌊(k−1)/s⌋)/|L| = (512 − 255)/512 = 257/512` (a degree-`< 2^20`
polynomial has `< 2^20` roots, so at most `⌊(2^20−1)/2^12⌋ = 255` *whole*
folded symbols can vanish — the same count that gives the `s = 32` value
`32769/65536`). -/
/-- The `2^9 = 512`-point geometric-progression Ext6 domain for the neutral `s = 2^12`
reference row, `j ↦ γ^(2^12·j)`, exactly as `domain` but at folding `2^12`.
Injectivity is `pow`-injectivity below
`orderOf γ`: `2^12·j < 2^12·2^9 = orderOf γ` (`gamma_pow_left_inj`). -/
noncomputable def domain12 : Fin (2 ^ 9) ↪ KoalaBear.Ext6 where
toFun j := gamma ^ (2 ^ 12 * j.val)
inj' i j hij := by
have hi : 2 ^ 12 * (i : ℕ) < 2 ^ 21 := by have := i.isLt; omega
have hj : 2 ^ 12 * (j : ℕ) < 2 ^ 21 := by have := j.isLt; omega
exact Fin.val_injective (by have := gamma_pow_left_inj hi hj hij; omega)
open Classical in
/-- **`(L, 2^12)`-admissibility of the `s = 2^12` progression domain.** The `2^12 · 2^9 = 2^21`
folded points `γ^(2^12·a) · γ^i = γ^(2^12·a + i)` are pairwise distinct, all exponents
below `orderOf γ = 2^21`; both `Admissible` conjuncts reduce to `ℕ`-arithmetic via
`gamma_pow_left_inj`. Same (now-proven) order bound as `admissible_domain`. -/
lemma admissible_domain12 :
ReedSolomon.Folded.Admissible (Finset.univ.map domain12) (2 ^ 12) foldOmega := by
refine ⟨?_, ?_⟩
· intro α hα β hβ hαβ i hi hcontra
simp only [Finset.mem_map, Finset.mem_univ, true_and] at hα hβ
obtain ⟨a, rfl⟩ := hα
obtain ⟨b, rfl⟩ := hβ
simp only [domain12, foldOmega, Function.Embedding.coeFn_mk, ← pow_add] at hαβ hcontra
have ha := a.isLt; have hb := b.isLt
have hab : (a : ℕ) ≠ (b : ℕ) := fun h => hαβ (by rw [h])
have := gamma_pow_left_inj (a := 2 ^ 12 * a.val + i) (b := 2 ^ 12 * b.val)
(by omega) (by omega) hcontra
omega
· intro α hα i hipos hi hcontra
simp only [Finset.mem_map, Finset.mem_univ, true_and] at hα
obtain ⟨a, rfl⟩ := hα
simp only [domain12, foldOmega, Function.Embedding.coeFn_mk, ← pow_add] at hcontra
have ha := a.isLt
have := gamma_pow_left_inj (a := 2 ^ 12 * a.val + i) (b := 2 ^ 12 * a.val)
(by omega) (by omega) hcontra
omega
/-- The `s = 2^12 = 4096` folded encoder: degree-`< 2^20` folded RS evaluation on
the `2^9` points of `domain12` with folding `s = 4096` (`k = 2^20`,
`|L| = 2^9`, rate `ρ = 1/2`), as `(Fin 2^20 → F) →ₗ (Fin 2^9 → Fin 4096 → F)`.
Same `frsEvalOnPoints ∘ (degreeLTEquiv).symm` shape as `encoder`, with the
folding parameter `2^12` and the smaller `2^9`-point domain. -/
noncomputable def encoder12 :
(Fin (2 ^ 20) → KoalaBear.Ext6) →ₗ[KoalaBear.Ext6]
(Fin (2 ^ 9) → Fin (2 ^ 12) → KoalaBear.Ext6) :=
(frsEvalOnPoints domain12 (2 ^ 12) foldOmega).domRestrict
(Polynomial.degreeLT KoalaBear.Ext6 (2 ^ 20))
∘ₗ (Polynomial.degreeLTEquiv KoalaBear.Ext6 (2 ^ 20)).symm.toLinearMap
open Classical in
/-- **Injectivity of the `s = 2^12` folded encoder** — the large-folding counterpart of
`encoder_injective`, derived through the in-tree bridge
`ReedSolomon.Folded.frsEvalOnPoints_domRestrict_injective`, consuming
`admissible_domain12`, `gamma_ne_zero`, and `k = 2^20 ≤ s·|L| = 2^12·2^9 =
2^21`. It uses the same shared order witness `gamma_exists` as the `s = 32` row. -/
theorem encoder12_injective : Function.Injective encoder12 := by
haveI : NeZero (2 ^ 20 : ℕ) := ⟨by norm_num⟩
simp only [encoder12, LinearMap.coe_comp, LinearEquiv.coe_toLinearMap]
refine (ReedSolomon.Folded.frsEvalOnPoints_domRestrict_injective (k := 2 ^ 20) (s := 2 ^ 12)
domain12 foldOmega admissible_domain12 gamma_ne_zero ?_).comp
(LinearEquiv.injective _)
rw [Fintype.card_fin]; norm_num
open Classical in
set_option maxRecDepth 8000 in
/-- **The `s = 2^12` folded encoder's image is exactly `frsCode`** — the large-folding
counterpart of `encoder_range`. -/
theorem encoder12_range :
Set.range ⇑encoder12 =
(↑(ReedSolomon.Folded.frsCode domain12 (2 ^ 20) (2 ^ 12) foldOmega) :
Set (Fin (2 ^ 9) → Fin (2 ^ 12) → KoalaBear.Ext6)) := by
ext y
rw [SetLike.mem_coe, ReedSolomon.Folded.frsCode, Submodule.mem_map]
simp only [Set.mem_range]
constructor
· rintro ⟨m, rfl⟩
refine ⟨↑((Polynomial.degreeLTEquiv KoalaBear.Ext6 (2 ^ 20)).symm m), Submodule.coe_mem _, ?_⟩
simp only [encoder12, LinearMap.coe_comp, LinearEquiv.coe_toLinearMap, Function.comp_apply,
LinearMap.domRestrict_apply]
· rintro ⟨p, hp, rfl⟩
refine ⟨Polynomial.degreeLTEquiv KoalaBear.Ext6 (2 ^ 20) ⟨p, hp⟩, ?_⟩
simp only [encoder12, LinearMap.coe_comp, LinearEquiv.coe_toLinearMap, Function.comp_apply,
LinearEquiv.symm_apply_apply, LinearMap.domRestrict_apply]
open Classical in
set_option maxRecDepth 8000 in
/-- **Folded-RS minimum relative distance at `s = 2^12`**, derived through
`ReedSolomon.Folded.minDist_frsCode` exactly as `minRelHammingDistCode_range_encoder`.
`minRelHammingDistCode (Set.range ⇑encoder12) = 257/512`, the folded-Singleton distance for
`FRS[F, L, 2^20, 4096, ω]`: a nonzero degree-`< 2^20` polynomial vanishes on
`≤ ⌊(2^20−1)/4096⌋ = 255` whole folded symbols, so `D = |L| − 255 = 512 − 255 = 257` and
`δ_min = 257/512`. -/
theorem minRelHammingDistCode_range_encoder12 :
minRelHammingDistCode (Set.range ⇑encoder12) = (257 / 512 : ℚ≥0) := by
haveI : Nonempty (Fin (2 ^ 9)) := inferInstance
haveI : NeZero (2 ^ 20 : ℕ) := ⟨by norm_num⟩
have hcode : Set.range ⇑encoder12 =
(↑(ReedSolomon.Folded.frsCode domain12 (2 ^ 20) (2 ^ 12) foldOmega) :
Set (Fin (2 ^ 9) → Fin (2 ^ 12) → KoalaBear.Ext6)) := encoder12_range
have hmin : Code.minDist (Set.range ⇑encoder12) = 257 := by
have h := ReedSolomon.Folded.minDist_frsCode (k := 2 ^ 20) (s := 2 ^ 12) (by norm_num)
domain12 foldOmega admissible_domain12 gamma_ne_zero
(by rw [Fintype.card_fin]; norm_num)
simp only [Fintype.card_fin] at h
norm_num at h
rw [hcode]; exact h
have hbridge := minDist_div_card_eq_minRelHammingDistCode (Set.range ⇑encoder12)
have hcardι : Fintype.card (Fin (2 ^ 9)) = 512 := by
rw [Fintype.card_fin]; norm_num
rw [hmin, hcardι] at hbridge
have hQ : ((minRelHammingDistCode (Set.range ⇑encoder12) : ℚ≥0) : ℚ) =
((257 / 512 : ℚ≥0) : ℚ) := by
rw [← hbridge]; push_cast; norm_num
exact_mod_cast hQ
/-- **Exact numerical lower-bound reference:** `2^(-128.75) ≤
(255/512)^128`, the folded MDS sweep-floor for `s = 2^12`. Split `255/512 =
(255/256)·(1/2)`, reduce `(1/2)^128 = 2^(-128)` to `2^(-0.75) ≤ (255/256)^128`.
Unlike the `s = 32` leaf (step `1/256` is far coarser than `1/32768`), Bernoulli
is too weak here, so we **sandwich through `3/5`**: `2^(-0.75) ≤ 3/5` (the small
integer fact `(3/5)^4 = 81/625 ≥ 1/8 = 2^(-3)`) and `3/5 ≤ (255/256)^128` (the
integer fact `3·256^128 ≤ 5·255^128`). True value `(255/512)^128 ≈ 2^(-128.723)`,
comfortably above the `2^(-128.75)` ceiling. No float `#eval`. -/
theorem two_rpow_le_spotcheckPow12 :
(2 : ℝ≥0) ^ (-(128.75 : ℝ)) ≤ ((255 : ℝ≥0) / 512) ^ (128 : ℕ) := by
rw [← NNReal.coe_le_coe]
push_cast [NNReal.coe_rpow]
rw [show (-128.75 : ℝ) = (-0.75 : ℝ) + (-(128 : ℝ)) by norm_num]
rw [Real.rpow_add (by norm_num : (0:ℝ) < 2)]
rw [show (255:ℝ)/512 = (255/256) * (1/2) by ring]
rw [mul_pow]
rw [show ((1:ℝ)/2)^128 = 2^(-(128:ℝ)) by
rw [Real.rpow_neg (by norm_num), show (128:ℝ) = ((128:ℕ):ℝ) by norm_num,
Real.rpow_natCast]; norm_num]
apply mul_le_mul_of_nonneg_right _ (by positivity)
-- ⊢ 2^(-0.75) ≤ (255/256)^128, via the rational sandwich 2^(-0.75) ≤ 3/5 ≤ (255/256)^128
have hsand : (3 : ℝ) / 5 ≤ (255 / 256 : ℝ) ^ 128 := by
rw [div_pow, div_le_div_iff₀ (by norm_num) (by positivity)]
exact_mod_cast (by norm_num : (3:ℕ) * 256 ^ 128 ≤ 255 ^ 128 * 5)
have h2neg : (2:ℝ) ^ (-0.75:ℝ) ≤ 3 / 5 := by
apply le_of_pow_le_pow_left₀ (n := 4) (by norm_num) (by positivity)
rw [← Real.rpow_natCast ((2:ℝ)^(-0.75:ℝ)) 4]
rw [← Real.rpow_mul (by norm_num : (0:ℝ) ≤ 2)]
norm_num
linarith [hsand, h2neg]
end Impl.FRS
end ToyProblem
Source