Bit-burst algorithm version of BigMath.erf and BigMath.erfc

[tompng]

Implement Bit-burst algorithm version of BigMath.erf and BigMath.erfc

Bit-burst algorithm implementation

Repeatedly apply Taylor expansion at increasing x.

1. Split x into x0 + x1 + x2 + ...
2. Calculate f(x0) by Taylor expansion(erf) or asymptotic expansion(erfc)
3. Calculate f(x0 + x1) by Taylor expansion of f at x0
4. Calculate f(x0 + x1 + x2) by Taylor expansion of f at x0 + x1
5. So on...

Each Taylor expansion calculation uses binary-splitting method.
Details are described in source code comment.

Benchmark

# Erf of integer

BigMath.erf(10, 10000)
# processing time: 0.324237s → 0.094490s
BigMath.erf(10, 100000)
# processing time: 13.380194s → 1.044322s

# Erf of full-precision number

BigMath.erf(BigDecimal(2).sqrt(10000), 10000)
# processing time: 0.501813s → 0.268420s
BigMath.erf(BigDecimal(2).sqrt(100000), 100000)
# processing time: 21.532002s → 5.708505s

# Erfc of integer

BigMath.erfc(1234, 10000)
# processing time: 0.216168s → 0.094400s
BigMath.erfc(1234, 100000)
# processing time: 19.402288s → 1.258311s

# Erfc of full-precision number

BigMath.erfc(1234 + BigDecimal(4).div(7, 10000), 10000)
# processing time: 4.616296s → 0.219974s
BigMath.erfc(1234 + BigDecimal(4).div(7, 20000), 20000)
# processing time: 23.335180s → 0.572202s
BigMath.erfc(1234 + BigDecimal(4).div(7, 100000), 100000)
# processing time: 1137.080714s → 5.270239s