2sin (example 3.3)

Average Error: 57.53% → 0.57%

Time: 13.9s

Precision: binary64

Cost: 39168

Math

\[\sin \left(x + \varepsilon\right) - \sin x \]

↓

\[\cos x \cdot \sin \varepsilon - \frac{\sin x \cdot {\sin \varepsilon}^{2}}{1 + \cos \varepsilon} \]

FPCore

(FPCore (x eps) :precision binary64 (- (sin (+ x eps)) (sin x)))

↓

(FPCore (x eps)
 :precision binary64
 (-
  (* (cos x) (sin eps))
  (/ (* (sin x) (pow (sin eps) 2.0)) (+ 1.0 (cos eps)))))

C

double code(double x, double eps) {
	return sin((x + eps)) - sin(x);
}

↓

double code(double x, double eps) {
	return (cos(x) * sin(eps)) - ((sin(x) * pow(sin(eps), 2.0)) / (1.0 + cos(eps)));
}

Fortran

real(8) function code(x, eps)
    real(8), intent (in) :: x
    real(8), intent (in) :: eps
    code = sin((x + eps)) - sin(x)
end function

↓

real(8) function code(x, eps)
    real(8), intent (in) :: x
    real(8), intent (in) :: eps
    code = (cos(x) * sin(eps)) - ((sin(x) * (sin(eps) ** 2.0d0)) / (1.0d0 + cos(eps)))
end function

Java

public static double code(double x, double eps) {
	return Math.sin((x + eps)) - Math.sin(x);
}

↓

public static double code(double x, double eps) {
	return (Math.cos(x) * Math.sin(eps)) - ((Math.sin(x) * Math.pow(Math.sin(eps), 2.0)) / (1.0 + Math.cos(eps)));
}

Python

def code(x, eps):
	return math.sin((x + eps)) - math.sin(x)

↓

def code(x, eps):
	return (math.cos(x) * math.sin(eps)) - ((math.sin(x) * math.pow(math.sin(eps), 2.0)) / (1.0 + math.cos(eps)))

Julia

function code(x, eps)
	return Float64(sin(Float64(x + eps)) - sin(x))
end

↓

function code(x, eps)
	return Float64(Float64(cos(x) * sin(eps)) - Float64(Float64(sin(x) * (sin(eps) ^ 2.0)) / Float64(1.0 + cos(eps))))
end

MATLAB

function tmp = code(x, eps)
	tmp = sin((x + eps)) - sin(x);
end

↓

function tmp = code(x, eps)
	tmp = (cos(x) * sin(eps)) - ((sin(x) * (sin(eps) ^ 2.0)) / (1.0 + cos(eps)));
end

Wolfram

code[x_, eps_] := N[(N[Sin[N[(x + eps), $MachinePrecision]], $MachinePrecision] - N[Sin[x], $MachinePrecision]), $MachinePrecision]

↓

code[x_, eps_] := N[(N[(N[Cos[x], $MachinePrecision] * N[Sin[eps], $MachinePrecision]), $MachinePrecision] - N[(N[(N[Sin[x], $MachinePrecision] * N[Power[N[Sin[eps], $MachinePrecision], 2.0], $MachinePrecision]), $MachinePrecision] / N[(1.0 + N[Cos[eps], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

Target

Original	57.53%
Target	23.65%
Herbie	0.57%

\[2 \cdot \left(\cos \left(x + \frac{\varepsilon}{2}\right) \cdot \sin \left(\frac{\varepsilon}{2}\right)\right) \]

Derivation

1. Initial program 57.53

   \[\sin \left(x + \varepsilon\right) - \sin x \]

2. Applied egg-rr 33.75

   \[\leadsto \color{blue}{\sin x \cdot \cos \varepsilon + \left(\left(-\sin x\right) + \cos x \cdot \sin \varepsilon\right)} \]

3. Simplified 0.61

   \[\leadsto \color{blue}{\mathsf{fma}\left(\sin \varepsilon, \cos x, \sin x \cdot \left(\cos \varepsilon + -1\right)\right)} \]
   Proof

   [Start] 33.75	\[ \sin x \cdot \cos \varepsilon + \left(\left(-\sin x\right) + \cos x \cdot \sin \varepsilon\right) \]
   associate-+r+ [=>] 0.62	\[ \color{blue}{\left(\sin x \cdot \cos \varepsilon + \left(-\sin x\right)\right) + \cos x \cdot \sin \varepsilon} \]
   +-commutative [<=] 0.62	\[ \color{blue}{\cos x \cdot \sin \varepsilon + \left(\sin x \cdot \cos \varepsilon + \left(-\sin x\right)\right)} \]
   *-commutative [=>] 0.62	\[ \color{blue}{\sin \varepsilon \cdot \cos x} + \left(\sin x \cdot \cos \varepsilon + \left(-\sin x\right)\right) \]
   fma-def [=>] 0.62	\[ \color{blue}{\mathsf{fma}\left(\sin \varepsilon, \cos x, \sin x \cdot \cos \varepsilon + \left(-\sin x\right)\right)} \]
   *-commutative [=>] 0.62	\[ \mathsf{fma}\left(\sin \varepsilon, \cos x, \color{blue}{\cos \varepsilon \cdot \sin x} + \left(-\sin x\right)\right) \]
   neg-mul-1 [=>] 0.62	\[ \mathsf{fma}\left(\sin \varepsilon, \cos x, \cos \varepsilon \cdot \sin x + \color{blue}{-1 \cdot \sin x}\right) \]
   distribute-rgt-out [=>] 0.61	\[ \mathsf{fma}\left(\sin \varepsilon, \cos x, \color{blue}{\sin x \cdot \left(\cos \varepsilon + -1\right)}\right) \]

4. Applied egg-rr 0.56

   \[\leadsto \mathsf{fma}\left(\sin \varepsilon, \cos x, \color{blue}{\frac{{\sin \varepsilon}^{2} \cdot \sin x}{-1 - \cos \varepsilon}}\right) \]

5. Taylor expanded in eps around inf 0.57

   \[\leadsto \color{blue}{\cos x \cdot \sin \varepsilon + -1 \cdot \frac{\sin x \cdot {\sin \varepsilon}^{2}}{1 + \cos \varepsilon}} \]

6. Final simplification 0.57

   \[\leadsto \cos x \cdot \sin \varepsilon - \frac{\sin x \cdot {\sin \varepsilon}^{2}}{1 + \cos \varepsilon} \]

Alternatives

Alternative 1
Error	0.61%
Cost	32448

\[\mathsf{fma}\left(\sin \varepsilon, \cos x, \sin x \cdot \left(-1 + \cos \varepsilon\right)\right) \]

Alternative 2
Error	0.62%
Cost	26176

\[\cos x \cdot \sin \varepsilon + \sin x \cdot \left(-1 + \cos \varepsilon\right) \]

Alternative 3
Error	22.45%
Cost	26048

\[\cos x \cdot \sin \varepsilon + \left(\sin x - \sin x\right) \]

Alternative 4
Error	23.65%
Cost	13888

\[\sin \left(\left(\varepsilon + \left(x - x\right)\right) \cdot 0.5\right) \cdot \left(2 \cdot \cos \left(0.5 \cdot \left(\varepsilon + \left(x + x\right)\right)\right)\right) \]

Alternative 5
Error	23.03%
Cost	13257

\[\begin{array}{l} \mathbf{if}\;\varepsilon \leq -0.000335 \lor \neg \left(\varepsilon \leq 0.0008\right):\\ \;\;\;\;\sin \varepsilon - \sin x\\ \mathbf{else}:\\ \;\;\;\;\cos x \cdot \varepsilon\\ \end{array} \]

Alternative 6
Error	23.67%
Cost	6856

\[\begin{array}{l} \mathbf{if}\;\varepsilon \leq -3.6 \cdot 10^{-6}:\\ \;\;\;\;\sin \varepsilon\\ \mathbf{elif}\;\varepsilon \leq 7.5 \cdot 10^{-5}:\\ \;\;\;\;\cos x \cdot \varepsilon\\ \mathbf{else}:\\ \;\;\;\;\sin \varepsilon\\ \end{array} \]

Alternative 7
Error	44.76%
Cost	6464

\[\sin \varepsilon \]

Alternative 8
Error	95.76%
Cost	64

\[0 \]

Alternative 9
Error	70.44%
Cost	64

\[\varepsilon \]

Reproduce

herbie shell --seed 2023115 
(FPCore (x eps)
  :name "2sin (example 3.3)"
  :precision binary64

  :herbie-target
  (* 2.0 (* (cos (+ x (/ eps 2.0))) (sin (/ eps 2.0))))

  (- (sin (+ x eps)) (sin x)))