Need to use eps=sqrt(1e-16) for double precision and sqrt(1e-8) for single precision when computing step size h for numerical differentiation. If the step size is too small, the cancellation error in f(x+h) - f(x) dominates, and if the step size is too large, the numerical derivative error dominates. Bumps mostly assumes double precision.
{
"status": "new",
"changetime": "2016-06-26T17:23:37",
"_ts": "2016-06-26 17:23:37.000382+00:00",
"description": "Need to use eps=sqrt(1e-16) for double precision and sqrt(1e-8) for single precision when computing step size h for numerical differentiation. If the step size is too small, the cancellation error in f(x+h) - f(x) dominates, and if the step size is too large, the numerical derivative error dominates. Bumps mostly assumes double precision.",
"reporter": "pkienzle",
"cc": "",
"resolution": "",
"workpackage": "SasView Bug Fixing",
"time": "2016-04-21T22:39:36",
"component": "SasView",
"summary": "numerical derivative step size is precision dependent",
"priority": "minor",
"keywords": "",
"milestone": "SasView Next Release +1",
"owner": "",
"type": "defect"
}
As agreed on fortnightly call of June 21, am moving all minor priority tickets to next release (+1) given the resources and time available to get this release out.
butler changed milestone from "SasView 4.0.0" to "SasView Next Release +1"
Need to use eps=sqrt(1e-16) for double precision and sqrt(1e-8) for single precision when computing step size h for numerical differentiation. If the step size is too small, the cancellation error in f(x+h) - f(x) dominates, and if the step size is too large, the numerical derivative error dominates. Bumps mostly assumes double precision.
Migrated from http://trac.sasview.org/ticket/563