FindMinimum convergence criteria
- To: mathgroup at smc.vnet.net
- Subject: [mg130776] FindMinimum convergence criteria
- From: Ray Koopman <koopman at sfu.ca>
- Date: Tue, 14 May 2013 03:14:47 -0400 (EDT)
- Delivered-to: l-mathgroup@mail-archive0.wolfram.com
- Delivered-to: l-mathgroup@wolfram.com
- Delivered-to: mathgroup-outx@smc.vnet.net
- Delivered-to: mathgroup-newsendx@smc.vnet.net
FindMinimum has four convergence tests, controlled by two parameters: AccuracyGoal and PrecisionGoal. The documentation says: * The settings for AccuracyGoal and PrecisionGoal specify the number of digits to seek in both the value of the position of the minimum, and the value of the function at the minimum. * FindMinimum continues until either of the goals specified by AccuracyGoal or PrecisionGoal is achieved. The second statement is clear: FindMinimum will quit if either accuracy or precision is adequate. But does the first statement mean that both the position and the function must be sufficiently accurate/ precise, or does it mean only that both are examined but either one will suffice? In other words, is the convergence criterion ((position is accurate) AND (function is accurate)) OR ((position is precise ) AND (function is precise )), or is it ((position is accurate) OR (function is accurate)) OR ((position is precise ) OR (function is precise ))? Setting one of AccuracyGoal or PrecisionGoal to Infinity will make that goal unattainable, forcing the iterations to continue until the other goal is reached. Is there any way to similarly ignore the error in one of the position of the minimum or the value of the function, so that the iterations will continue until the error in the other is small enough?