math / data /lean /OAI /Combinatorics /EditApproximation /Refinement /ComputedKernelWork.lean
introvoyz045's picture
Migrated from GitHub
4a56cd6 verified
Raw History Blame Contribute Delete
1.88 kB
import OAI.Combinatorics.EditApproximation.Geometry.RankedMemoKernelWork
import OAI.Combinatorics.EditApproximation.Refinement.ComputedLocalKernelWork
import OAI.Combinatorics.EditApproximation.Definitions
namespace OAI
namespace EditApproximation
open Filter
/-- Local arithmetic annotations follow the actual memo run, hence are
incurred once per cache miss, with the original program and draw values. -/
theorem computedFamilyKernelWork_le (source target : List β„•) (N : β„•)
(input : ComputedFamilyInput source target N)
(cost : ComputedFamilyMemoKey source target N β†’ β„•) (C : β„•)
(hcost : βˆ€ key, cost key ≀ C) :
computedFamilyKernelWork source target N input cost ≀
C * (computedFamilyResult source target N input).requests :=
rankedMemoKernelWork_le _ _ _ _ cost C hcost _ _
theorem computedFamilyKernelWork_near_linear (D : β„•) (epsilon : ℝ) (hepsilon : 0 < epsilon) :
βˆ€αΆ  N in atTop, βˆ€ source target : List β„•, source.length + target.length ≀ N β†’
βˆ€ (input : ComputedFamilyInput source target N)
(cost : ComputedFamilyMemoKey source target N β†’ β„•),
(βˆ€ key, cost key ≀ computedLocalKernelBudget D N) β†’
(computedFamilyKernelWork source target N input cost : ℝ) ≀
(N + 2 : ℝ) ^ (1 + epsilon) := by
filter_upwards [computedLocalKernelBudget_requests_near_linear D epsilon hepsilon] with N hN
intro source target hsize input cost hcost
have h := computedFamilyKernelWork_le source target N input cost
(computedLocalKernelBudget D N) hcost
have hr : (computedFamilyKernelWork source target N input cost : ℝ) ≀
(computedLocalKernelBudget D N : ℝ) * (computedFamilyResult source target N input).requests := by
exact_mod_cast h
exact hr.trans (by simpa only [mul_comm] using hN source target hsize input)
end EditApproximation
end OAI