[MATH-280] bug in inverseCumulativeProbability() for Normal Distribution

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.2
• Fix Version/s: 2.0

Environment

Java 1.5

• Assignee: Luc Maisonobe
• Reporter: Michael Borcherds
• Created: 06/Jul/09 21:26
• Updated: 07/Aug/09 09:18
• Resolved: 07/Jul/09 09:21

Description

• @version $Revision: 617953 $ $Date: 2008-02-02 22:54:00 -0700 (Sat, 02 Feb 2008) $
  */
  public class NormalDistributionImpl extends AbstractContinuousDistribution

• @version $Revision: 506600 $ $Date: 2007-02-12 12:35:59 -0700 (Mon, 12 Feb 2007) $
  */
  public abstract class AbstractContinuousDistribution

This code:

DistributionFactory factory = app.getDistributionFactory();
NormalDistribution normal = factory.createNormalDistribution(0,1);
double result = normal.inverseCumulativeProbability(0.9772498680518209);

gives the exception below. It should return (approx) 2.0000...

normal.inverseCumulativeProbability(0.977249868051820); works fine

These also give errors:
0.9986501019683698 (should return 3.0000...)
0.9999683287581673 (should return 4.0000...)

org.apache.commons.math.MathException: Number of iterations=1, maximum iterations=2,147,483,647, initial=1, lower bound=0, upper bound=179,769,313,486,231,570,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000, final a value=0, final b value=2, f(a)=-0.477, f(b)=0
at org.apache.commons.math.distribution.AbstractContinuousDistribution.inverseCumulativeProbability(AbstractContinuousDistribution.java:103)
at org.apache.commons.math.distribution.NormalDistributionImpl.inverseCumulativeProbability(NormalDistributionImpl.java:145)