expq3 (problem 3.4.2)

Average Error: 60.3 → 0.3

Time: 14.0s

Precision: binary64

\[-1 < \varepsilon \land \varepsilon < 1\]

\[[a, b] = \mathsf{sort}([a, b]) \\]

Math

\[\frac{\varepsilon \cdot \left(e^{\left(a + b\right) \cdot \varepsilon} - 1\right)}{\left(e^{a \cdot \varepsilon} - 1\right) \cdot \left(e^{b \cdot \varepsilon} - 1\right)} \]

↓

\[\begin{array}{l} t_0 := \varepsilon \cdot \left(a + b\right)\\ t_1 := \frac{\varepsilon \cdot \left(e^{t_0} - 1\right)}{\left(e^{\varepsilon \cdot a} - 1\right) \cdot \left(e^{\varepsilon \cdot b} - 1\right)}\\ \mathbf{if}\;t_1 \leq -\infty:\\ \;\;\;\;\frac{-\frac{a + b}{-b}}{a}\\ \mathbf{elif}\;t_1 \leq 207261.66977473313:\\ \;\;\;\;\frac{\varepsilon}{\mathsf{expm1}\left(\varepsilon \cdot b\right)} \cdot \frac{\mathsf{expm1}\left(t_0\right)}{\mathsf{expm1}\left(\varepsilon \cdot a\right)}\\ \mathbf{else}:\\ \;\;\;\;\frac{1}{b} + \frac{1}{a}\\ \end{array} \]

FPCore

(FPCore (a b eps)
 :precision binary64
 (/
  (* eps (- (exp (* (+ a b) eps)) 1.0))
  (* (- (exp (* a eps)) 1.0) (- (exp (* b eps)) 1.0))))

↓

(FPCore (a b eps)
 :precision binary64
 (let* ((t_0 (* eps (+ a b)))
        (t_1
         (/
          (* eps (- (exp t_0) 1.0))
          (* (- (exp (* eps a)) 1.0) (- (exp (* eps b)) 1.0)))))
   (if (<= t_1 (- INFINITY))
     (/ (- (/ (+ a b) (- b))) a)
     (if (<= t_1 207261.66977473313)
       (* (/ eps (expm1 (* eps b))) (/ (expm1 t_0) (expm1 (* eps a))))
       (+ (/ 1.0 b) (/ 1.0 a))))))

C

double code(double a, double b, double eps) {
	return (eps * (exp(((a + b) * eps)) - 1.0)) / ((exp((a * eps)) - 1.0) * (exp((b * eps)) - 1.0));
}

↓

double code(double a, double b, double eps) {
	double t_0 = eps * (a + b);
	double t_1 = (eps * (exp(t_0) - 1.0)) / ((exp((eps * a)) - 1.0) * (exp((eps * b)) - 1.0));
	double tmp;
	if (t_1 <= -((double) INFINITY)) {
		tmp = -((a + b) / -b) / a;
	} else if (t_1 <= 207261.66977473313) {
		tmp = (eps / expm1((eps * b))) * (expm1(t_0) / expm1((eps * a)));
	} else {
		tmp = (1.0 / b) + (1.0 / a);
	}
	return tmp;
}

Java

public static double code(double a, double b, double eps) {
	return (eps * (Math.exp(((a + b) * eps)) - 1.0)) / ((Math.exp((a * eps)) - 1.0) * (Math.exp((b * eps)) - 1.0));
}

↓

public static double code(double a, double b, double eps) {
	double t_0 = eps * (a + b);
	double t_1 = (eps * (Math.exp(t_0) - 1.0)) / ((Math.exp((eps * a)) - 1.0) * (Math.exp((eps * b)) - 1.0));
	double tmp;
	if (t_1 <= -Double.POSITIVE_INFINITY) {
		tmp = -((a + b) / -b) / a;
	} else if (t_1 <= 207261.66977473313) {
		tmp = (eps / Math.expm1((eps * b))) * (Math.expm1(t_0) / Math.expm1((eps * a)));
	} else {
		tmp = (1.0 / b) + (1.0 / a);
	}
	return tmp;
}

Python

def code(a, b, eps):
	return (eps * (math.exp(((a + b) * eps)) - 1.0)) / ((math.exp((a * eps)) - 1.0) * (math.exp((b * eps)) - 1.0))

↓

def code(a, b, eps):
	t_0 = eps * (a + b)
	t_1 = (eps * (math.exp(t_0) - 1.0)) / ((math.exp((eps * a)) - 1.0) * (math.exp((eps * b)) - 1.0))
	tmp = 0
	if t_1 <= -math.inf:
		tmp = -((a + b) / -b) / a
	elif t_1 <= 207261.66977473313:
		tmp = (eps / math.expm1((eps * b))) * (math.expm1(t_0) / math.expm1((eps * a)))
	else:
		tmp = (1.0 / b) + (1.0 / a)
	return tmp

Julia

function code(a, b, eps)
	return Float64(Float64(eps * Float64(exp(Float64(Float64(a + b) * eps)) - 1.0)) / Float64(Float64(exp(Float64(a * eps)) - 1.0) * Float64(exp(Float64(b * eps)) - 1.0)))
end

↓

function code(a, b, eps)
	t_0 = Float64(eps * Float64(a + b))
	t_1 = Float64(Float64(eps * Float64(exp(t_0) - 1.0)) / Float64(Float64(exp(Float64(eps * a)) - 1.0) * Float64(exp(Float64(eps * b)) - 1.0)))
	tmp = 0.0
	if (t_1 <= Float64(-Inf))
		tmp = Float64(Float64(-Float64(Float64(a + b) / Float64(-b))) / a);
	elseif (t_1 <= 207261.66977473313)
		tmp = Float64(Float64(eps / expm1(Float64(eps * b))) * Float64(expm1(t_0) / expm1(Float64(eps * a))));
	else
		tmp = Float64(Float64(1.0 / b) + Float64(1.0 / a));
	end
	return tmp
end

Wolfram

code[a_, b_, eps_] := N[(N[(eps * N[(N[Exp[N[(N[(a + b), $MachinePrecision] * eps), $MachinePrecision]], $MachinePrecision] - 1.0), $MachinePrecision]), $MachinePrecision] / N[(N[(N[Exp[N[(a * eps), $MachinePrecision]], $MachinePrecision] - 1.0), $MachinePrecision] * N[(N[Exp[N[(b * eps), $MachinePrecision]], $MachinePrecision] - 1.0), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

↓

code[a_, b_, eps_] := Block[{t$95$0 = N[(eps * N[(a + b), $MachinePrecision]), $MachinePrecision]}, Block[{t$95$1 = N[(N[(eps * N[(N[Exp[t$95$0], $MachinePrecision] - 1.0), $MachinePrecision]), $MachinePrecision] / N[(N[(N[Exp[N[(eps * a), $MachinePrecision]], $MachinePrecision] - 1.0), $MachinePrecision] * N[(N[Exp[N[(eps * b), $MachinePrecision]], $MachinePrecision] - 1.0), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]}, If[LessEqual[t$95$1, (-Infinity)], N[((-N[(N[(a + b), $MachinePrecision] / (-b)), $MachinePrecision]) / a), $MachinePrecision], If[LessEqual[t$95$1, 207261.66977473313], N[(N[(eps / N[(Exp[N[(eps * b), $MachinePrecision]] - 1), $MachinePrecision]), $MachinePrecision] * N[(N[(Exp[t$95$0] - 1), $MachinePrecision] / N[(Exp[N[(eps * a), $MachinePrecision]] - 1), $MachinePrecision]), $MachinePrecision]), $MachinePrecision], N[(N[(1.0 / b), $MachinePrecision] + N[(1.0 / a), $MachinePrecision]), $MachinePrecision]]]]]

Error

Bits error versus a

Bits error versus b

Bits error versus eps

Target

Original	60.3
Target	14.7
Herbie	0.3

\[\frac{a + b}{a \cdot b} \]

Derivation

1. Split input into 3 regimes

2. if (/.f64 (*.f64 eps (-.f64 (exp.f64 (*.f64 (+.f64 a b) eps)) 1)) (*.f64 (-.f64 (exp.f64 (*.f64 a eps)) 1) (-.f64 (exp.f64 (*.f64 b eps)) 1))) < -inf.0

   1. Initial program 64.0

      \[\frac{\varepsilon \cdot \left(e^{\left(a + b\right) \cdot \varepsilon} - 1\right)}{\left(e^{a \cdot \varepsilon} - 1\right) \cdot \left(e^{b \cdot \varepsilon} - 1\right)} \]

   2. Simplified 17.4

      \[\leadsto \color{blue}{\frac{\varepsilon \cdot \mathsf{expm1}\left(\varepsilon \cdot \left(a + b\right)\right)}{\mathsf{expm1}\left(\varepsilon \cdot a\right) \cdot \mathsf{expm1}\left(\varepsilon \cdot b\right)}} \]

   3. Taylor expanded in eps around 0 8.7

      \[\leadsto \color{blue}{0.08333333333333333 \cdot \left({\varepsilon}^{2} \cdot a\right) + \left(\frac{1}{b} + \left(\frac{1}{a} + 0.08333333333333333 \cdot \left({\varepsilon}^{2} \cdot b\right)\right)\right)} \]

   4. Simplified 8.7

      \[\leadsto \color{blue}{\mathsf{fma}\left(0.08333333333333333, a \cdot \left(\varepsilon \cdot \varepsilon\right), \frac{1}{b} + \mathsf{fma}\left(0.08333333333333333, b \cdot \left(\varepsilon \cdot \varepsilon\right), \frac{1}{a}\right)\right)} \]

   5. Taylor expanded in eps around 0 0.0

      \[\leadsto \color{blue}{\frac{1}{b} + \frac{1}{a}} \]

   6. Applied egg-rr 2.3

      \[\leadsto \color{blue}{\frac{\frac{b + a}{b}}{a}} \]

   7. Applied egg-rr 2.3

      \[\leadsto \frac{\color{blue}{-\frac{b + a}{-b}}}{a} \]

   if -inf.0 < (/.f64 (*.f64 eps (-.f64 (exp.f64 (*.f64 (+.f64 a b) eps)) 1)) (*.f64 (-.f64 (exp.f64 (*.f64 a eps)) 1) (-.f64 (exp.f64 (*.f64 b eps)) 1))) < 207261.66977473313

   1. Initial program 4.0

      \[\frac{\varepsilon \cdot \left(e^{\left(a + b\right) \cdot \varepsilon} - 1\right)}{\left(e^{a \cdot \varepsilon} - 1\right) \cdot \left(e^{b \cdot \varepsilon} - 1\right)} \]

   2. Simplified 0.1

      \[\leadsto \color{blue}{\frac{\varepsilon \cdot \mathsf{expm1}\left(\varepsilon \cdot \left(a + b\right)\right)}{\mathsf{expm1}\left(\varepsilon \cdot a\right) \cdot \mathsf{expm1}\left(\varepsilon \cdot b\right)}} \]

   3. Applied egg-rr 0.1

      \[\leadsto \color{blue}{\frac{\varepsilon}{\mathsf{expm1}\left(\varepsilon \cdot b\right)} \cdot \frac{\mathsf{expm1}\left(\varepsilon \cdot \left(a + b\right)\right)}{\mathsf{expm1}\left(\varepsilon \cdot a\right)}} \]

   if 207261.66977473313 < (/.f64 (*.f64 eps (-.f64 (exp.f64 (*.f64 (+.f64 a b) eps)) 1)) (*.f64 (-.f64 (exp.f64 (*.f64 a eps)) 1) (-.f64 (exp.f64 (*.f64 b eps)) 1)))

   1. Initial program 64.0

      \[\frac{\varepsilon \cdot \left(e^{\left(a + b\right) \cdot \varepsilon} - 1\right)}{\left(e^{a \cdot \varepsilon} - 1\right) \cdot \left(e^{b \cdot \varepsilon} - 1\right)} \]

   2. Simplified 48.2

      \[\leadsto \color{blue}{\frac{\varepsilon \cdot \mathsf{expm1}\left(\varepsilon \cdot \left(a + b\right)\right)}{\mathsf{expm1}\left(\varepsilon \cdot a\right) \cdot \mathsf{expm1}\left(\varepsilon \cdot b\right)}} \]

   3. Taylor expanded in eps around 0 1.6

      \[\leadsto \color{blue}{0.08333333333333333 \cdot \left({\varepsilon}^{2} \cdot a\right) + \left(\frac{1}{b} + \left(\frac{1}{a} + 0.08333333333333333 \cdot \left({\varepsilon}^{2} \cdot b\right)\right)\right)} \]

   4. Simplified 1.6

      \[\leadsto \color{blue}{\mathsf{fma}\left(0.08333333333333333, a \cdot \left(\varepsilon \cdot \varepsilon\right), \frac{1}{b} + \mathsf{fma}\left(0.08333333333333333, b \cdot \left(\varepsilon \cdot \varepsilon\right), \frac{1}{a}\right)\right)} \]

   5. Taylor expanded in eps around 0 0.0

      \[\leadsto \color{blue}{\frac{1}{b} + \frac{1}{a}} \]
3. Recombined 3 regimes into one program.

4. Final simplification 0.3

   \[\leadsto \begin{array}{l} \mathbf{if}\;\frac{\varepsilon \cdot \left(e^{\varepsilon \cdot \left(a + b\right)} - 1\right)}{\left(e^{\varepsilon \cdot a} - 1\right) \cdot \left(e^{\varepsilon \cdot b} - 1\right)} \leq -\infty:\\ \;\;\;\;\frac{-\frac{a + b}{-b}}{a}\\ \mathbf{elif}\;\frac{\varepsilon \cdot \left(e^{\varepsilon \cdot \left(a + b\right)} - 1\right)}{\left(e^{\varepsilon \cdot a} - 1\right) \cdot \left(e^{\varepsilon \cdot b} - 1\right)} \leq 207261.66977473313:\\ \;\;\;\;\frac{\varepsilon}{\mathsf{expm1}\left(\varepsilon \cdot b\right)} \cdot \frac{\mathsf{expm1}\left(\varepsilon \cdot \left(a + b\right)\right)}{\mathsf{expm1}\left(\varepsilon \cdot a\right)}\\ \mathbf{else}:\\ \;\;\;\;\frac{1}{b} + \frac{1}{a}\\ \end{array} \]

Reproduce

herbie shell --seed 2022133 
(FPCore (a b eps)
  :name "expq3 (problem 3.4.2)"
  :precision binary64
  :pre (and (< -1.0 eps) (< eps 1.0))

  :herbie-target
  (/ (+ a b) (* a b))

  (/ (* eps (- (exp (* (+ a b) eps)) 1.0)) (* (- (exp (* a eps)) 1.0) (- (exp (* b eps)) 1.0))))