SabaPivot's picture
download
raw
8.45 kB
{
"claim": "Corollary 3.7 (2^Omega(d) ERM queries)",
"combined_regime": {
"eps_prime": 0.014666666666666666,
"note": "Corollary 3.7 as stated says rho <= 1 + O_eps(eps); the reduction gives rho <= 1 + eps/eps' * L, which is O(eps) for fixed eps' and L.",
"rho_at_eps_1_over_12_L_1": 6.681818181818182,
"rho_bound": "1 + (eps/eps') L"
},
"erm_query_drives_the_search": {
"conclusion": "every ERM query of the constructed performative instance yields T(x) exactly, so the query counts above are simultaneously ERM query counts.",
"max_error_recovering_T_from_ERM": 1.1102230246251565e-15
},
"proposition_3_2": [
{
"d": 2,
"empirical_rho": 0.9651904169835266,
"eps": 0.08333333333333333,
"lambda": 1.0,
"max_T_recovery_error_from_one_ERM_query": 0.0,
"max_fixed_point_gap_identity_error": 0.0,
"rho_bound_1_plus_lambda_L": 2.0,
"rho_within_bound": true
},
{
"d": 2,
"empirical_rho": 0.7667395491970505,
"eps": 0.041666666666666664,
"lambda": 0.5,
"max_T_recovery_error_from_one_ERM_query": 1.1102230246251565e-16,
"max_fixed_point_gap_identity_error": 1.1102230246251565e-16,
"rho_bound_1_plus_lambda_L": 1.5,
"rho_within_bound": true
},
{
"d": 2,
"empirical_rho": 0.9623367010017474,
"eps": 0.008333333333333333,
"lambda": 0.1,
"max_T_recovery_error_from_one_ERM_query": 6.106226635438361e-16,
"max_fixed_point_gap_identity_error": 1.249000902703301e-16,
"rho_bound_1_plus_lambda_L": 1.1,
"rho_within_bound": true
},
{
"d": 2,
"empirical_rho": 0.9987650411022788,
"eps": 0.0008333333333333333,
"lambda": 0.01,
"max_T_recovery_error_from_one_ERM_query": 5.551115123125783e-15,
"max_fixed_point_gap_identity_error": 1.4051260155412137e-16,
"rho_bound_1_plus_lambda_L": 1.01,
"rho_within_bound": true
},
{
"d": 5,
"empirical_rho": 0.8726205536485927,
"eps": 0.08333333333333333,
"lambda": 1.0,
"max_T_recovery_error_from_one_ERM_query": 0.0,
"max_fixed_point_gap_identity_error": 0.0,
"rho_bound_1_plus_lambda_L": 2.0,
"rho_within_bound": true
},
{
"d": 5,
"empirical_rho": 0.8257986738803997,
"eps": 0.041666666666666664,
"lambda": 0.5,
"max_T_recovery_error_from_one_ERM_query": 1.1102230246251565e-16,
"max_fixed_point_gap_identity_error": 1.1102230246251565e-16,
"rho_bound_1_plus_lambda_L": 1.5,
"rho_within_bound": true
},
{
"d": 5,
"empirical_rho": 0.9606677098850223,
"eps": 0.008333333333333333,
"lambda": 0.1,
"max_T_recovery_error_from_one_ERM_query": 6.106226635438361e-16,
"max_fixed_point_gap_identity_error": 1.3877787807814457e-16,
"rho_bound_1_plus_lambda_L": 1.1,
"rho_within_bound": true
},
{
"d": 5,
"empirical_rho": 0.9976924785537874,
"eps": 0.0008333333333333333,
"lambda": 0.01,
"max_T_recovery_error_from_one_ERM_query": 5.551115123125783e-15,
"max_fixed_point_gap_identity_error": 1.1102230246251565e-16,
"rho_bound_1_plus_lambda_L": 1.01,
"rho_within_bound": true
},
{
"d": 20,
"empirical_rho": 0.8602434218272195,
"eps": 0.08333333333333333,
"lambda": 1.0,
"max_T_recovery_error_from_one_ERM_query": 0.0,
"max_fixed_point_gap_identity_error": 0.0,
"rho_bound_1_plus_lambda_L": 2.0,
"rho_within_bound": true
},
{
"d": 20,
"empirical_rho": 0.6776069518694584,
"eps": 0.041666666666666664,
"lambda": 0.5,
"max_T_recovery_error_from_one_ERM_query": 5.551115123125783e-17,
"max_fixed_point_gap_identity_error": 1.1102230246251565e-16,
"rho_bound_1_plus_lambda_L": 1.5,
"rho_within_bound": true
},
{
"d": 20,
"empirical_rho": 0.9454915757570909,
"eps": 0.008333333333333333,
"lambda": 0.1,
"max_T_recovery_error_from_one_ERM_query": 3.3306690738754696e-16,
"max_fixed_point_gap_identity_error": 8.326672684688674e-17,
"rho_bound_1_plus_lambda_L": 1.1,
"rho_within_bound": true
},
{
"d": 20,
"empirical_rho": 0.993858969845494,
"eps": 0.0008333333333333333,
"lambda": 0.01,
"max_T_recovery_error_from_one_ERM_query": 3.941291737419306e-15,
"max_fixed_point_gap_identity_error": 6.245004513516506e-17,
"rho_bound_1_plus_lambda_L": 1.01,
"rho_within_bound": true
}
],
"query_equivalence": {
"max_recovery_error": 5.551115123125783e-15,
"statement": "G(x) = (1-lambda)x + lambda T(x) with lambda and x known, so T(x) = (G(x) - (1-lambda)x)/lambda: the ERM oracle and the T-oracle are information-theoretically equivalent, and the Hirsch et al. bound transfers with no loss."
},
"seed": 20260725,
"theorem_3_6_arithmetic": {
"bound": "c * ((1/eps - 10) L)^(d-2)",
"conclusion": "for any constant eps <= 1/12 the bound is at least c*2^(d-2) = 2^Omega(d); for eps > 1/11 the base drops below 1 and the bound is vacuous, so 'even when eps is a constant' means a constant below 1/11.",
"eps_threshold_for_base_1": "[1/11]",
"eps_threshold_for_base_2": "[1/12]",
"rows": [
{
"base_at_L_1": "1",
"base_float": 1.0,
"base_ge_2_gives_2_to_the_d": false,
"base_gt_1_nontrivial": false,
"eps": "1/11"
},
{
"base_at_L_1": "2",
"base_float": 2.0,
"base_ge_2_gives_2_to_the_d": true,
"base_gt_1_nontrivial": true,
"eps": "1/12"
},
{
"base_at_L_1": "10",
"base_float": 10.0,
"base_ge_2_gives_2_to_the_d": true,
"base_gt_1_nontrivial": true,
"eps": "1/20"
},
{
"base_at_L_1": "90",
"base_float": 90.0,
"base_ge_2_gives_2_to_the_d": true,
"base_gt_1_nontrivial": true,
"eps": "1/100"
},
{
"base_at_L_1": "0",
"base_float": 0.0,
"base_ge_2_gives_2_to_the_d": false,
"base_gt_1_nontrivial": false,
"eps": "1/10"
}
]
},
"toy_hidden_path": {
"R2": 0.9975602300281715,
"blocker": "the true Hirsch-Papadimitriou-Vavasis instance is a continuous Brouwer map on [0,1]^d whose construction we do not implement; this is a discrete hidden-path family with the same information-hiding mechanism, run at m=3 and d<=9.",
"implied_base": 1.737024441318484,
"label": "toy",
"log_queries_slope_per_dimension": 0.5521735581550697,
"measurements": [
{
"cells": 9,
"d": 2,
"grid_side": 3,
"mean_queries": 4.2,
"median_queries": 4.0,
"path_length": 3,
"trials": 40
},
{
"cells": 27,
"d": 3,
"grid_side": 3,
"mean_queries": 6.925,
"median_queries": 6.0,
"path_length": 5,
"trials": 40
},
{
"cells": 81,
"d": 4,
"grid_side": 3,
"mean_queries": 12.875,
"median_queries": 10.0,
"path_length": 9,
"trials": 40
},
{
"cells": 243,
"d": 5,
"grid_side": 3,
"mean_queries": 21.25,
"median_queries": 19.0,
"path_length": 16,
"trials": 40
},
{
"cells": 729,
"d": 6,
"grid_side": 3,
"mean_queries": 33.075,
"median_queries": 32.5,
"path_length": 27,
"trials": 40
},
{
"cells": 2187,
"d": 7,
"grid_side": 3,
"mean_queries": 58.725,
"median_queries": 58.0,
"path_length": 47,
"trials": 40
},
{
"cells": 6561,
"d": 8,
"grid_side": 3,
"mean_queries": 122.2,
"median_queries": 92.5,
"path_length": 81,
"trials": 40
},
{
"cells": 19683,
"d": 9,
"grid_side": 3,
"mean_queries": 199.05,
"median_queries": 186.5,
"path_length": 140,
"trials": 40
}
],
"predicted_base_sqrt_m": 1.7320508075688772,
"predicted_slope_half_log_m": 0.5493061443340549
},
"verdict": {
"empirical_family": "toy",
"exponential_arithmetic_verified": true,
"reduction_and_query_equivalence_verified": true
}
}

Xet Storage Details

Size:
8.45 kB
·
Xet hash:
7f002545fc47eb0c98f04e1595f591532db786143fb6b0343afb114c966dbf36

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.