Release 1.2.0

API changes

New method isDoubleValue()

The new method isDoubleValue(BigDecimal) returns whether the specified BigDecimal value can be represented as double.

The javadoc of isIntValue(BigDecimal) has been adapted to avoid confusion about “loss of precision”.

Bugfixes

log() and sqrt() fail if argument is not a valid double value

The functions log() and sqrt() used for the approximation an initial value that was calculated using Math.log(double) respectively Math.sqrt(double). This would lead to exceptions if the argument was not a valid double value (for example new BigDecimal("1E309")).

The fix uses the new method isDoubleValue() to verify the argument. If it is not a valid double value another initial value for the approximation is taken.

Enhancements

Constants pi, log2, log3, log10 with improved caching

The constants pi, log2, log3, log10 are now cached with a different strategy.

Previously the constants where stored with a precalculated precision of slightly over 1000 digits and rounded down to the desired precision. This increases the size of the jar library and uses unnecessary memory if the full precision is never needed. If more precision is needed the constants where calculated again and a again with the high precision.

The new caching strategy keeps now precalculated values (making the library smaller). Whenever a constant with a higher precision than cached is calculated it will replace the previously cached value. This uses only the minimum memory for the cached values and works well even if very high precision is needed.

Tip: If your application needs the constant with many different precisions it might be more efficient to calculate the necessary constants with the maximum precision in the initialization phase of your application.

Constant e is cached

The constant e is now cached with the same strategy as pi and the other mathematical constants.

Peformance improvements in sqrt()

The sqrt() was optimized so that the square root of square numbers is calculated much faster.

The adaptive precision calculation was slightly optimized, assuming that the precision increases every iteration by a factor of ~ 1.8.

Examples

Note: The example code is available on github, but not part of the big-math library.

added PerformanceRegressionBigDecimalMath

The class PerformanceRegressionBigDecimalMath was added in an attempt to catch performance regression between releases.

The output for all releases has been added as csv files in ch.obermuhlner.math.big.example/docu/benchmarks/regression.

This is still somewhat experimental, for example the files where performance was measured on my laptop without any attempt of standardizing the execution environment (stopping other applications, …).