Average Error: 58.7 → 0.2
Time: 35.4s
Precision: 64
\[\log \left(\frac{1 - \varepsilon}{1 + \varepsilon}\right)\]
↓
\[{\varepsilon}^{5} \cdot \frac{-2}{5} - \frac{\left(\left(\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon\right) \cdot \left(\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon\right) - 4\right) \cdot \varepsilon}{\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon - 2}\]
double f(double eps) {
        double r5871572 = 1.0;
        double r5871573 = eps;
        double r5871574 = r5871572 - r5871573;
        double r5871575 = r5871572 + r5871573;
        double r5871576 = r5871574 / r5871575;
        double r5871577 = log(r5871576);
        return r5871577;
}
↓
double f(double eps) {
        double r5871578 = eps;
        double r5871579 = 5.0;
        double r5871580 = pow(r5871578, r5871579);
        double r5871581 = -0.4;
        double r5871582 = r5871580 * r5871581;
        double r5871583 = 0.6666666666666666;
        double r5871584 = r5871583 * r5871578;
        double r5871585 = r5871584 * r5871578;
        double r5871586 = r5871585 * r5871585;
        double r5871587 = 4.0;
        double r5871588 = r5871586 - r5871587;
        double r5871589 = r5871588 * r5871578;
        double r5871590 = 2.0;
        double r5871591 = r5871585 - r5871590;
        double r5871592 = r5871589 / r5871591;
        double r5871593 = r5871582 - r5871592;
        return r5871593;
}


\log \left(\frac{1 - \varepsilon}{1 + \varepsilon}\right)
↓
{\varepsilon}^{5} \cdot \frac{-2}{5} - \frac{\left(\left(\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon\right) \cdot \left(\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon\right) - 4\right) \cdot \varepsilon}{\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon - 2}

Error

Bits error versus eps

Target

Original58.7
Target0.2
Herbie0.2
\[-2 \cdot \left(\left(\varepsilon + \frac{{\varepsilon}^{3}}{3}\right) + \frac{{\varepsilon}^{5}}{5}\right)\]

Derivation

  1. Initial program 58.7

    \[\log \left(\frac{1 - \varepsilon}{1 + \varepsilon}\right)\]

  2. Taylor expanded around 0 0.2

    \[\leadsto \color{blue}{-\left(\frac{2}{3} \cdot {\varepsilon}^{3} + \left(\frac{2}{5} \cdot {\varepsilon}^{5} + 2 \cdot \varepsilon\right)\right)}\]

  3. Simplified0.2

    \[\leadsto \color{blue}{\frac{-2}{5} \cdot {\varepsilon}^{5} - \left(\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon + 2\right) \cdot \varepsilon}\]
  4. Using strategy rm

  5. Applied flip-+0.2

    \[\leadsto \frac{-2}{5} \cdot {\varepsilon}^{5} - \color{blue}{\frac{\left(\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon\right) \cdot \left(\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon\right) - 2 \cdot 2}{\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon - 2}} \cdot \varepsilon\]

  6. Applied associate-*l/0.2

    \[\leadsto \frac{-2}{5} \cdot {\varepsilon}^{5} - \color{blue}{\frac{\left(\left(\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon\right) \cdot \left(\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon\right) - 2 \cdot 2\right) \cdot \varepsilon}{\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon - 2}}\]

  7. Final simplification0.2

    \[\leadsto {\varepsilon}^{5} \cdot \frac{-2}{5} - \frac{\left(\left(\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon\right) \cdot \left(\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon\right) - 4\right) \cdot \varepsilon}{\left(\frac{2}{3} \cdot \varepsilon\right) \cdot \varepsilon - 2}\]

Reproduce

herbie shell --seed 2019102 
(FPCore (eps)
  :name "logq (problem 3.4.3)"

  :herbie-target
  (* -2 (+ (+ eps (/ (pow eps 3) 3)) (/ (pow eps 5) 5)))

  (log (/ (- 1 eps) (+ 1 eps))))