Hyperbolic arc-(co)tangent

Average Accuracy: 8.6% → 100.0%

Time: 5.6s

Precision: binary64

Cost: 13504

Math

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

↓

\[0.5 \cdot \left(\mathsf{log1p}\left(x \cdot \left(-x\right)\right) + \mathsf{log1p}\left(-x\right) \cdot -2\right) \]

FPCore

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

↓

(FPCore (x)
 :precision binary64
 (* 0.5 (+ (log1p (* x (- x))) (* (log1p (- x)) -2.0))))

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 * -x)) + (log1p(-x) * -2.0));
}

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 * -x)) + (Math.log1p(-x) * -2.0));
}

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 * -x)) + (math.log1p(-x) * -2.0))

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(Float64(x * Float64(-x))) + Float64(log1p(Float64(-x)) * -2.0)))
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 + N[(x * (-x)), $MachinePrecision]], $MachinePrecision] + N[(N[Log[1 + (-x)], $MachinePrecision] * -2.0), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

Derivation

1. Initial program 8.6%

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

2. Simplified 8.6%

   \[\leadsto \color{blue}{0.5 \cdot \log \left(\frac{1 + x}{1 - x}\right)} \]
   Proof

   [Start] 8.6	\[ \frac{1}{2} \cdot \log \left(\frac{1 + x}{1 - x}\right) \]
   metadata-eval [=>] 8.6	\[ \color{blue}{0.5} \cdot \log \left(\frac{1 + x}{1 - x}\right) \]

3. Applied egg-rr 99.1%

   \[\leadsto 0.5 \cdot \color{blue}{\left(\log \left(1 - x \cdot x\right) - \left(\mathsf{log1p}\left(-x\right) + \mathsf{log1p}\left(-x\right)\right)\right)} \]

4. Simplified 100.0%

   \[\leadsto 0.5 \cdot \color{blue}{\left(\mathsf{log1p}\left(x \cdot \left(-x\right)\right) - 2 \cdot \mathsf{log1p}\left(-x\right)\right)} \]
   Proof

   [Start] 99.1	\[ 0.5 \cdot \left(\log \left(1 - x \cdot x\right) - \left(\mathsf{log1p}\left(-x\right) + \mathsf{log1p}\left(-x\right)\right)\right) \]
   cancel-sign-sub-inv [=>] 99.1	\[ 0.5 \cdot \left(\log \color{blue}{\left(1 + \left(-x\right) \cdot x\right)} - \left(\mathsf{log1p}\left(-x\right) + \mathsf{log1p}\left(-x\right)\right)\right) \]
   *-commutative [<=] 99.1	\[ 0.5 \cdot \left(\log \left(1 + \color{blue}{x \cdot \left(-x\right)}\right) - \left(\mathsf{log1p}\left(-x\right) + \mathsf{log1p}\left(-x\right)\right)\right) \]
   log1p-def [=>] 100.0	\[ 0.5 \cdot \left(\color{blue}{\mathsf{log1p}\left(x \cdot \left(-x\right)\right)} - \left(\mathsf{log1p}\left(-x\right) + \mathsf{log1p}\left(-x\right)\right)\right) \]
   count-2 [=>] 100.0	\[ 0.5 \cdot \left(\mathsf{log1p}\left(x \cdot \left(-x\right)\right) - \color{blue}{2 \cdot \mathsf{log1p}\left(-x\right)}\right) \]

5. Final simplification 100.0%

   \[\leadsto 0.5 \cdot \left(\mathsf{log1p}\left(x \cdot \left(-x\right)\right) + \mathsf{log1p}\left(-x\right) \cdot -2\right) \]

Alternatives

Alternative 1
Accuracy	100.0%
Cost	13184

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

Alternative 2
Accuracy	99.5%
Cost	832

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

Alternative 3
Accuracy	99.0%
Cost	320

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

Reproduce

herbie shell --seed 2023129 
(FPCore (x)
  :name "Hyperbolic arc-(co)tangent"
  :precision binary64
  (* (/ 1.0 2.0) (log (/ (+ 1.0 x) (- 1.0 x)))))