Download lean/OAI/Combinatorics/LowSpaceSubsetSum/Word/WordDecisionRound.lean from roseking/openAI-math-dataset: direct link, hf CLI and curl.
- Browser
- Download file 5.22 kB
-
https://huggingface.co/datasets/roseking/openAI-math-dataset/resolve/main/lean/OAI/Combinatorics/LowSpaceSubsetSum/Word/WordDecisionRound.lean
- Command line
-
hf download hf://datasets/roseking/openAI-math-dataset/lean/OAI/Combinatorics/LowSpaceSubsetSum/Word/WordDecisionRound.lean
-
curl -L -o WordDecisionRound.lean https://huggingface.co/datasets/roseking/openAI-math-dataset/resolve/main/lean/OAI/Combinatorics/LowSpaceSubsetSum/Word/WordDecisionRound.lean
5.22 kB
| import OAI.Combinatorics.LowSpaceSubsetSum.Word.WordDecisionBranches | |
| import OAI.Combinatorics.LowSpaceSubsetSum.Search.DecisionRound | |
| namespace OAI | |
| /-! # Independent fair-word rounds and counter-executed amplification -/ | |
| noncomputable section | |
| namespace LowSpaceSubsetSum.WordDecisionRound | |
| variable {n : ℕ} | |
| abbrev Words (I : Instance n) := | |
| WordDecisionBranches.CompressionWords I × WordDecisionBranches.RepresentationWords I | |
| def law (I : Instance n) : FiniteLaw (Words I) := | |
| (FiniteLaw.uniform : FiniteLaw (WordDecisionBranches.CompressionWords I)).prod | |
| (FiniteLaw.uniform : FiniteLaw (WordDecisionBranches.RepresentationWords I)) | |
| def combine (c r : (Bool × ℕ) × ℕ) : (Bool × ℕ) × ℕ := | |
| ((c.1.1 || r.1.1,c.1.2+r.1.2+3),max c.2 (r.2+8)+16) | |
| def run (I : Instance n) (hn : 180000000 ≤ n) | |
| (hlog : Real.logb 2 ((mixerSize n : ℝ)+1)+1 ≤ (n : ℝ)/1000) | |
| (words : Words I) : (Bool × ℕ) × ℕ := | |
| combine (WordDecisionBranches.compression I (by omega) words.1) | |
| (WordDecisionBranches.representation I (by omega) hlog words.2) | |
| theorem value (I : Instance n) (hn : 180000000 ≤ n) | |
| (hlog : Real.logb 2 ((mixerSize n : ℝ)+1)+1 ≤ (n : ℝ)/1000) | |
| (words : Words I) : | |
| (run I hn hlog words).1.1 = | |
| ((WordDecisionBranches.compression I (by omega) words.1).1.1 || | |
| (WordDecisionBranches.representation I (by omega) hlog words.2).1.1) := rfl | |
| theorem sound (I : Instance n) (hn : 180000000 ≤ n) | |
| (hlog : Real.logb 2 ((mixerSize n : ℝ)+1)+1 ≤ (n : ℝ)/1000) | |
| (words : Words I) (hy : (run I hn hlog words).1.1 = true) : Solvable I := by | |
| rw [value,Bool.or_eq_true] at hy | |
| rcases hy with hc | hr | |
| · exact WordDecisionBranches.compression_sound I (by omega) words.1 hc | |
| · exact WordDecisionBranches.representation_sound I (by omega) hlog words.2 hr | |
| theorem probability (I : Instance n) (hn : 180000000 ≤ n) | |
| (hlog : Real.logb 2 ((mixerSize n : ℝ)+1)+1 ≤ (n : ℝ)/1000) | |
| (hs : Solvable I) | |
| (hden : ((mixerSize n/2 : ℕ) : ℝ)+1 < | |
| (2 : ℝ)^((1/1500 : ℝ)*(mixerSize n/2 : ℕ))) | |
| (hscale : Real.logb 2 (inputScale I : ℝ) ≤ (n : ℝ)/1000000000) : | |
| 1/(inputScale I : ℝ)^110 ≤ | |
| (law I).probability (fun words => (run I hn hlog words).1.1 = true) := by | |
| let P : WordDecisionBranches.CompressionWords I → Prop := fun words => | |
| (WordDecisionBranches.compression I (by omega) words).1.1 = true | |
| let Q : WordDecisionBranches.RepresentationWords I → Prop := fun words => | |
| (WordDecisionBranches.representation I (by omega) hlog words).1.1 = true | |
| by_cases hm : (153/20000 : ℝ)*n ≤ | |
| meanDeficiency (CompressionData.mk I).weights Finset.univ (compressionHalfSize n) | |
| · have hp := WordDecisionBranches.compression_probability I (by omega) hs hm hscale | |
| have hu : (2 : ℝ) ≤ inputScale I := by | |
| exact_mod_cast (show 2 ≤ inputScale I by unfold inputScale; omega) | |
| have hd : (8 : ℝ) ≤ (inputScale I : ℝ)^110 := | |
| (by norm_num : (8 : ℝ) ≤ 2^110).trans (pow_le_pow_left₀ (by norm_num) hu _) | |
| apply ((one_div_le_one_div_of_le (by norm_num) hd).trans hp).trans | |
| simpa only [law,value,Bool.or_eq_true,P,Q] using | |
| FiniteLaw.prod_or_ge_left | |
| (FiniteLaw.uniform : FiniteLaw (WordDecisionBranches.CompressionWords I)) | |
| (FiniteLaw.uniform : FiniteLaw (WordDecisionBranches.RepresentationWords I)) P Q | |
| · have hm' : meanDeficiency (canonicalRepresentationData I).zeroSetup.originalWeights | |
| Finset.univ (mixerSize n/2) < (153/20000 : ℝ)*n := lt_of_not_ge hm | |
| apply (WordDecisionBranches.representation_probability I hn hlog hs hm' hden hscale).trans | |
| simpa only [law,value,Bool.or_eq_true,P,Q] using | |
| FiniteLaw.prod_or_ge_right | |
| (FiniteLaw.uniform : FiniteLaw (WordDecisionBranches.CompressionWords I)) | |
| (FiniteLaw.uniform : FiniteLaw (WordDecisionBranches.RepresentationWords I)) P Q | |
| def repeated (I : Instance n) (hn : 180000000 ≤ n) | |
| (hscale : Real.logb 2 (inputScale I : ℝ) ≤ (n : ℝ)/1000000000) | |
| (words : Fin (inputScale I^120) → Words I) : (Bool × ℕ) × ℕ := | |
| DecisionRepetitionCells.run (run I hn (roundLogBudget I (by omega) hscale)) words | |
| theorem repeated_sound (I : Instance n) (hn : 180000000 ≤ n) | |
| (hscale : Real.logb 2 (inputScale I : ℝ) ≤ (n : ℝ)/1000000000) | |
| (words : Fin (inputScale I^120) → Words I) | |
| (hy : (repeated I hn hscale words).1.1 = true) : Solvable I := by | |
| rw [repeated,DecisionRepetitionCells.value,DecisionRepetition.value] at hy | |
| exact repeatAccept_sound _ _ (sound I hn _) words hy | |
| theorem repeated_probability (I : Instance n) (hn : 180000000 ≤ n) | |
| (hscale : Real.logb 2 (inputScale I : ℝ) ≤ (n : ℝ)/1000000000) | |
| (hden : ((mixerSize n/2 : ℕ) : ℝ)+1 < | |
| (2 : ℝ)^((1/1500 : ℝ)*(mixerSize n/2 : ℕ))) (hs : Solvable I) : | |
| (2/3 : ℝ) < ((law I).iid (inputScale I^120)).probability | |
| (fun words => (repeated I hn hscale words).1.1 = true) := by | |
| simp only [repeated,DecisionRepetitionCells.value,DecisionRepetition.value] | |
| exact repetition_acceptance_probability (law I) _ (by unfold inputScale; omega) | |
| (probability I hn _ hs hden hscale) | |
| end LowSpaceSubsetSum.WordDecisionRound | |
| end | |
| end OAI | |