Hyperbolic arc-(co)tangent

Average Error: 58.6 → 0.0

Time: 6.4s

Precision: binary64

Cost: 13184

Math

\[\frac{1}{2} \cdot \log \left(\frac{1 + x}{1 - x}\right) \]

↓

\[0.5 \cdot \left(\mathsf{log1p}\left(x\right) - \mathsf{log1p}\left(-x\right)\right) \]

FPCore

(FPCore (x) :precision binary64 (* (/ 1.0 2.0) (log (/ (+ 1.0 x) (- 1.0 x)))))

↓

(FPCore (x) :precision binary64 (* 0.5 (- (log1p x) (log1p (- x)))))

C

double code(double x) {
	return (1.0 / 2.0) * log(((1.0 + x) / (1.0 - x)));
}

↓

double code(double x) {
	return 0.5 * (log1p(x) - log1p(-x));
}

Java

public static double code(double x) {
	return (1.0 / 2.0) * Math.log(((1.0 + x) / (1.0 - x)));
}

↓

public static double code(double x) {
	return 0.5 * (Math.log1p(x) - Math.log1p(-x));
}

Python

def code(x):
	return (1.0 / 2.0) * math.log(((1.0 + x) / (1.0 - x)))

↓

def code(x):
	return 0.5 * (math.log1p(x) - math.log1p(-x))

Julia

function code(x)
	return Float64(Float64(1.0 / 2.0) * log(Float64(Float64(1.0 + x) / Float64(1.0 - x))))
end

↓

function code(x)
	return Float64(0.5 * Float64(log1p(x) - log1p(Float64(-x))))
end

Wolfram

code[x_] := N[(N[(1.0 / 2.0), $MachinePrecision] * N[Log[N[(N[(1.0 + x), $MachinePrecision] / N[(1.0 - x), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]

↓

code[x_] := N[(0.5 * N[(N[Log[1 + x], $MachinePrecision] - N[Log[1 + (-x)], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

Derivation

1. Initial program 58.6

   \[\frac{1}{2} \cdot \log \left(\frac{1 + x}{1 - x}\right) \]

2. Simplified 0.0

   \[\leadsto \color{blue}{0.5 \cdot \left(\mathsf{log1p}\left(x\right) - \mathsf{log1p}\left(-x\right)\right)} \]
   Proof
   (*.f64 1/2 (-.f64 (log1p.f64 x) (log1p.f64 (neg.f64 x)))): 0 points increase in error, 0 points decrease in error
   (*.f64 (Rewrite<= metadata-eval (/.f64 1 2)) (-.f64 (log1p.f64 x) (log1p.f64 (neg.f64 x)))): 0 points increase in error, 0 points decrease in error
   (*.f64 (/.f64 1 2) (-.f64 (Rewrite<= log1p-def_binary64 (log.f64 (+.f64 1 x))) (log1p.f64 (neg.f64 x)))): 256 points increase in error, 0 points decrease in error
   (*.f64 (/.f64 1 2) (-.f64 (log.f64 (+.f64 1 x)) (Rewrite<= log1p-def_binary64 (log.f64 (+.f64 1 (neg.f64 x)))))): 250 points increase in error, 5 points decrease in error
   (*.f64 (/.f64 1 2) (-.f64 (log.f64 (+.f64 1 x)) (log.f64 (Rewrite<= sub-neg_binary64 (-.f64 1 x))))): 0 points increase in error, 0 points decrease in error
   (*.f64 (/.f64 1 2) (Rewrite<= log-div_binary64 (log.f64 (/.f64 (+.f64 1 x) (-.f64 1 x))))): 10 points increase in error, 8 points decrease in error

3. Final simplification 0.0

   \[\leadsto 0.5 \cdot \left(\mathsf{log1p}\left(x\right) - \mathsf{log1p}\left(-x\right)\right) \]

Alternatives

Alternative 1
Error	0.3
Cost	832

\[0.5 \cdot \left(x \cdot \left(x \cdot \left(x \cdot 0.6666666666666666\right)\right) + x \cdot 2\right) \]

Alternative 2
Error	0.6
Cost	320

\[0.5 \cdot \left(x \cdot 2\right) \]

Reproduce

herbie shell --seed 2022313 
(FPCore (x)
  :name "Hyperbolic arc-(co)tangent"
  :precision binary64
  (* (/ 1.0 2.0) (log (/ (+ 1.0 x) (- 1.0 x)))))