math.abs on complex

Average Error: 32.0 → 17.7

Time: 1.3s

Precision: binary64

Math

\[\sqrt{re \cdot re + im \cdot im}\]

↓

\[\begin{array}{l} \mathbf{if}\;re \le -2.58022773065394553 \cdot 10^{96}:\\ \;\;\;\;-1 \cdot re\\ \mathbf{elif}\;re \le 2.0014054693044192 \cdot 10^{92}:\\ \;\;\;\;\sqrt{re \cdot re + im \cdot im}\\ \mathbf{else}:\\ \;\;\;\;re\\ \end{array}\]

C

double code(double re, double im) {
	return ((double) sqrt(((double) (((double) (re * re)) + ((double) (im * im))))));
}

↓

double code(double re, double im) {
	double VAR;
	if ((re <= -2.5802277306539455e+96)) {
		VAR = ((double) (-1.0 * re));
	} else {
		double VAR_1;
		if ((re <= 2.0014054693044192e+92)) {
			VAR_1 = ((double) sqrt(((double) (((double) (re * re)) + ((double) (im * im))))));
		} else {
			VAR_1 = re;
		}
		VAR = VAR_1;
	}
	return VAR;
}

Error

Bits error versus re

Bits error versus im

Derivation

1. Split input into 3 regimes

2. if re < -2.58022773065394553e96

   1. Initial program 51.3

      \[\sqrt{re \cdot re + im \cdot im}\]

   2. Taylor expanded around -inf 11.1

      \[\leadsto \color{blue}{-1 \cdot re}\]

   if -2.58022773065394553e96 < re < 2.0014054693044192e92

   1. Initial program 21.5

      \[\sqrt{re \cdot re + im \cdot im}\]

   if 2.0014054693044192e92 < re

   1. Initial program 51.1

      \[\sqrt{re \cdot re + im \cdot im}\]

   2. Taylor expanded around inf 10.6

      \[\leadsto \color{blue}{re}\]
3. Recombined 3 regimes into one program.

4. Final simplification 17.7

   \[\leadsto \begin{array}{l} \mathbf{if}\;re \le -2.58022773065394553 \cdot 10^{96}:\\ \;\;\;\;-1 \cdot re\\ \mathbf{elif}\;re \le 2.0014054693044192 \cdot 10^{92}:\\ \;\;\;\;\sqrt{re \cdot re + im \cdot im}\\ \mathbf{else}:\\ \;\;\;\;re\\ \end{array}\]

Reproduce

herbie shell --seed 2020150 
(FPCore (re im)
  :name "math.abs on complex"
  :precision binary64
  (sqrt (+ (* re re) (* im im))))