Package org.apfloat

package org.apfloat
The apfloat Application Programming Interface (API).

All application code using apfloat generally needs to only call the classes in this package.

A sample apfloat program might look like this:

import org.apfloat.Apfloat;
import org.apfloat.ApfloatMath;

public class ApfloatTest
    public static void main(String[] args)
        Apfloat x = new Apfloat(2, 1000);   // Value 2, precision 1000 digits

        Apfloat y = ApfloatMath.sqrt(x);    // Square root of 2, to 1000 digits

As apfloats are immutable, they can be easily passed by reference. Also the mantissa data of numbers can be efficiently shared in various situations.

An inherent property of an Apfloat is the radix. The radix is specified at the time an apfloat is created. Due to the way the default implementation works, there is no real performance difference in using radix 2 or some other radix in the internal calculations. While it's generally not possible to use numbers in different radixes in operations, it's possible to convert a number to a different radix using the Apfloat.toRadix(int) method.

The rounding mode for apfloat calculations is undefined. Thus, it's not guaranteed that rounding happens to an optimal direction and more often than not it doesn't. This should be carefully considered when designing numerical algorithms. Round-off errors can accumulate faster than expected, and loss of precision (as returned by Apfloat.precision()) can happen quickly. This bad behaviour is further accelerated by using a radix bigger than two, e.g. base 10, which is the default. Note that precision is defined as the number of digits in the number's radix. If numbers need to be rounded in a specific way then the ApfloatMath.round(Apfloat,long,RoundingMode) method can be invoked explicitly.

Generally, the result of various mathematical operations is accurate to the second last digit in the resulting number. This means roughly that the last significant digit of the result can be inaccurate. For example, the number 12345, with precision 5, should be considered 12345 ± 10. This only applies to elementary mathematical operations. More complicated functions may have slightly larger errors due to error accumulation. This should generally not be a problem, as you should typically be using apfloats for calculations with a precision of thousands or millions of digits.

There is no concept of an infinity or Not-a-Number with apfloats. Whenever the result of an operation would be infinite or undefined, an exception is thrown (usually an ArithmeticException).

All of the apfloat-specific exceptions being thrown by the apfloat library extend the base class ApfloatRuntimeException. This exception, or various subclasses can be thrown in different situations, for example:

  • InfiniteExpansionException - The result of an operation would have infinite size. For example, new Apfloat(2).divide(new Apfloat(3)), in radix 10.
  • OverflowException - Overflow. If the exponent is too large to fit in a long, the situation can't be handled. Also, there is no "infinity" apfloat value that could be returned as the result.
  • LossOfPrecisionException - Total loss of precision. For example, ApfloatMath.sin(new Apfloat(1e100)). If the magnitude (100) is far greater than the precision (1) then the value of the sin() function can't be determined to any accuracy.
The exception is a RuntimeException, because it should "never happen", and in general the cases where it is thrown are irrecoverable with the current implementation. Also any of the situations mentioned above may be relaxed in the future, so this exception handling strategy should be more future-proof than others, even if it has its limitations currently.

The Apfloat class is the basic building block of all the objects used in the apfloat package. An Apcomplex simply consists of two apfloats, the real part and the imaginary part. An Apint is implemented with an apfloat and all its operations just guarantee that the number never gets a fractional part. Last, an Aprational is an aggregate of two apints, the numerator and the denominator. The relations of these classes are shown in a class diagram format below:

Class diagram