File size: 3,334 Bytes
4a56cd6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
import OAI.Combinatorics.EditApproximation.Binary.IntegerBinaryAccuracy
import OAI.Combinatorics.EditApproximation.Binary.IntegerBinarySpace

namespace OAI

/-! An edit-distance approximation scheme conditional on finite scalar
Bernstein concentration (Tropp, ACM217, corrected 2023, Theorem 4.16).
The same uniform binary algorithm is used in all four clauses. -/

namespace EditApproximation
open MeasureTheory Filter

/-- For each fixed binary accuracy in `(0,1)`, the concrete uniform algorithm
has success probability at least `2/3`, deterministic zero on equal inputs,
worst-case expected almost-linear time including preprocessing and sampling,
and polynomial space on every outcome. Alphabet bounds are analysis parameters,
not inputs to the algorithm. The finite initial range uses exact suffix DP. -/
theorem almostLinearEditDistance_conditional (epsilon : BinaryFraction)
    (hepsilon : 0 < epsilon.value) (_hepsilonOne : epsilon.value < 1)
    (bernstein : βˆ€ N, ComputedBernsteinInputs N) :
    (βˆ€ (source target : List β„€) (N : β„•), source.length + target.length ≀ N β†’
      (2 / 3 : ℝ) ≀ (computedFullLaw (source.map integerSymbolCode) (target.map integerSymbolCode) N epsilon).real
        {outcome | EditDistortion.edit source target ≀ integerBinaryAnswer source target N epsilon outcome ∧
          (integerBinaryAnswer source target N epsilon outcome : ℝ) ≀
            (1 + (epsilon.value : ℝ)) * EditDistortion.edit source target}) ∧
    (βˆ€ (source target : List β„€) (N : β„•), source = target β†’ βˆ€ outcome,
      integerBinaryAnswer source target N epsilon outcome = 0) ∧
    (βˆ€ (C : β„•) (eta : ℝ), 0 < eta β†’ βˆ€αΆ  N : β„• in atTop,
      βˆ€ source target : List β„€, source.length + target.length ≀ N β†’
        (βˆ€ z ∈ source ++ target, z.natAbs ≀ (N + 2) ^ C) β†’
        (∫ outcome, integerBinaryWork source target N epsilon outcome
          βˆ‚computedFullLaw (source.map integerSymbolCode) (target.map integerSymbolCode) N epsilon) ≀
            (N + 2 : ℝ) ^ (1 + eta)) ∧
    (βˆ€ C : β„•, βˆ€αΆ  N : β„• in atTop, βˆ€ source target : List β„€,
      source.length + target.length ≀ N β†’
        (βˆ€ z ∈ source ++ target, z.natAbs ≀ (N + 2) ^ C) β†’
        βˆ€ outcome, integerBinarySpace source target N epsilon outcome ≀ (N + 2) ^ (C + 65564)) := by
  refine ⟨?_, ?_, ?_, ?_⟩
  Β· intro source target N hsize
    exact integerBinaryAnswer_conditional_accuracy source target N epsilon hepsilon hsize (bernstein N)
  Β· intro source target N heq outcome
    exact integerBinaryAnswer_equal source target N epsilon heq outcome
  Β· intro C eta heta
    exact integerBinaryWork_expected_near_linear C epsilon hepsilon eta heta
  Β· intro C
    exact integerBinarySpace_polynomial C epsilon hepsilon

/-- The cutoff selects an exact answer, rather than leaving the small-input
regime outside the uniform algorithm's correctness statement. -/
theorem integerBinaryAnswer_fallback (source target : List β„€) (N : β„•) (epsilon : BinaryFraction)
    (hsmall : Β¬ physicalLargeInput N epsilon) (outcome) :
    integerBinaryAnswer source target N epsilon outcome = EditDistortion.edit source target := by
  simp only [integerBinaryAnswer, hsmall, ite_false, exactSuffixAnswer_eq_edit, integerSymbols_edit]

end EditApproximation

end OAI