逐步推導
S001 · implies
操作前
source roles = [qf-mod50-numerr-affine-propagation, [domain, input_error_bound, operation, slope]]
操作後
derived = [numerr-affine-propagation, Exact_numerical_error_density]
條件:無
原bank的operation是LOCAL hyphen,keys與domain各自精確匹配。欄位順序及數字/Boolean/字串角色依來源,所有有理數文字整數省略/1;不改原題與英文、MKO、AST或狀態。
步驟識別與依賴
規則:R66_EXACT_source_bank_keys_local_operation_domain · 依賴:G001
S002 · implies
操作前
source roles = [propagation ORIGINAL slope, 100000000, 1, 1]
操作後
derived = [[1, 1], 1]
條件:無
先驗來源傳入rational的原num/den範圍與非零den,之後才sign/gcd約分。MOD50界1億且naive運算不先cross-cancel;Wave4此均值界100萬。超界是來源切片不合,不是數學為假。
步驟識別與依賴
規則:R66_SOURCE_PREreduction_components_then_sign_gcd · 依賴:G001, S001
S003 · calculation
操作前
propagation ORIGINAL slope ORIGINAL numerator over denominator := div(1, 1)
操作後
propagation ORIGINAL slope ORIGINAL numerator over denominator = 1
條件:無
按本步原操作數與次序精確運算;來源有理數的pre-reduction界另行保留,不用浮點或偷偷跨bank放寬。
步驟識別與依賴
規則:exact_rational_arithmetic · 依賴:G001, S002
S004 · implies
操作前
source roles = [propagation ORIGINAL input bound, 100000000, 12, 3]
操作後
derived = [[4, 1], 4]
條件:無
先驗來源傳入rational的原num/den範圍與非零den,之後才sign/gcd約分。MOD50界1億且naive運算不先cross-cancel;Wave4此均值界100萬。超界是來源切片不合,不是數學為假。
步驟識別與依賴
規則:R66_SOURCE_PREreduction_components_then_sign_gcd · 依賴:G001, S003
S005 · calculation
操作前
propagation ORIGINAL input bound ORIGINAL numerator over denominator := div(12, 3)
操作後
propagation ORIGINAL input bound ORIGINAL numerator over denominator = 4
條件:無
按本步原操作數與次序精確運算;來源有理數的pre-reduction界另行保留,不用浮點或偷偷跨bank放寬。
步驟識別與依賴
規則:exact_rational_arithmetic · 依賴:G001, S004
S006 · implies
操作前
source roles = [Affine output error bound SOURCE naive mul, 100000000, 4, 1]
操作後
derived = [[4, 1], 4]
條件:無
先驗來源傳入rational的原num/den範圍與非零den,之後才sign/gcd約分。MOD50界1億且naive運算不先cross-cancel;Wave4此均值界100萬。超界是來源切片不合,不是數學為假。
步驟識別與依賴
規則:R66_SOURCE_PREreduction_components_then_sign_gcd · 依賴:G001, S005
S007 · calculation
操作前
Affine output error bound SOURCE naive mul ORIGINAL numerator over denominator := div(4, 1)
操作後
Affine output error bound SOURCE naive mul ORIGINAL numerator over denominator = 4
條件:無
按本步原操作數與次序精確運算;來源有理數的pre-reduction界另行保留,不用浮點或偷偷跨bank放寬。
步驟識別與依賴
規則:exact_rational_arithmetic · 依賴:G001, S006
S008 · calculation
操作前
Affine output error bound ORIGINAL rational mul := mul(1, 4)
操作後
Affine output error bound ORIGINAL rational mul = 4
條件:無
按本步原操作數與次序精確運算;來源有理數的pre-reduction界另行保留,不用浮點或偷偷跨bank放寬。
步驟識別與依賴
規則:exact_rational_arithmetic · 依賴:G001, S007
S009 · implies
操作前
source roles = [1, 4]
操作後
derived = [4]
條件:無
由|m(x+δ)+b−(mx+b)|=|m|*|δ|≤|m|*bound;bound非負,m或bound為0合法。這是此affine model的absolute bound,不辨識rounding/truncation/measurement來源或證convergence。
步驟識別與依賴
規則:Error_AFFINE_propagation_absolute_bound_INCLUDES_zero · 依賴:G001, S008