NMSE Section 6.1 mentioned, B

Percentage Accurate: 78.1% → 99.6%
Time: 10.4s
Alternatives: 11
Speedup: 1.7×

Specification

?
\[\begin{array}{l} \\ \left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \end{array} \]
(FPCore (a b)
 :precision binary64
 (* (* (/ (PI) 2.0) (/ 1.0 (- (* b b) (* a a)))) (- (/ 1.0 a) (/ 1.0 b))))
\begin{array}{l}

\\
\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right)
\end{array}

Sampling outcomes in binary64 precision:

Local Percentage Accuracy vs ?

The average percentage accuracy by input value. Horizontal axis shows value of an input variable; the variable is choosen in the title. Vertical axis is accuracy; higher is better. Red represent the original program, while blue represents Herbie's suggestion. These can be toggled with buttons below the plot. The line is an average while dots represent individual samples.

Accuracy vs Speed?

Herbie found 11 alternatives:

AlternativeAccuracySpeedup
The accuracy (vertical axis) and speed (horizontal axis) of each alternatives. Up and to the right is better. The red square shows the initial program, and each blue circle shows an alternative.The line shows the best available speed-accuracy tradeoffs.

Initial Program: 78.1% accurate, 1.0× speedup?

\[\begin{array}{l} \\ \left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \end{array} \]
(FPCore (a b)
 :precision binary64
 (* (* (/ (PI) 2.0) (/ 1.0 (- (* b b) (* a a)))) (- (/ 1.0 a) (/ 1.0 b))))
\begin{array}{l}

\\
\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right)
\end{array}

Alternative 1: 99.6% accurate, 1.2× speedup?

\[\begin{array}{l} \\ \frac{\left(\frac{0.5}{b \cdot a} \cdot \left(b - a\right)\right) \cdot \frac{\mathsf{PI}\left(\right)}{b + a}}{b - a} \end{array} \]
(FPCore (a b)
 :precision binary64
 (/ (* (* (/ 0.5 (* b a)) (- b a)) (/ (PI) (+ b a))) (- b a)))
\begin{array}{l}

\\
\frac{\left(\frac{0.5}{b \cdot a} \cdot \left(b - a\right)\right) \cdot \frac{\mathsf{PI}\left(\right)}{b + a}}{b - a}
\end{array}
Derivation
  1. Initial program 79.2%

    \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
  2. Add Preprocessing
  3. Applied rewrites90.2%

    \[\leadsto \color{blue}{\mathsf{fma}\left(\frac{\mathsf{PI}\left(\right)}{a + b}, \frac{0.5}{b - a} \cdot {a}^{-1}, \frac{\frac{0.5}{a + b} \cdot \frac{\mathsf{PI}\left(\right)}{b - a}}{-b}\right)} \]
  4. Applied rewrites99.6%

    \[\leadsto \color{blue}{\frac{\frac{\mathsf{PI}\left(\right)}{a + b} \cdot \left(\left(b - a\right) \cdot \frac{0.5}{a \cdot b}\right)}{b - a}} \]
  5. Final simplification99.6%

    \[\leadsto \frac{\left(\frac{0.5}{b \cdot a} \cdot \left(b - a\right)\right) \cdot \frac{\mathsf{PI}\left(\right)}{b + a}}{b - a} \]
  6. Add Preprocessing

Alternative 2: 99.0% accurate, 1.7× speedup?

\[\begin{array}{l} \\ \frac{1}{\frac{b \cdot a}{0.5 \cdot \mathsf{PI}\left(\right)} \cdot \left(b + a\right)} \end{array} \]
(FPCore (a b) :precision binary64 (/ 1.0 (* (/ (* b a) (* 0.5 (PI))) (+ b a))))
\begin{array}{l}

\\
\frac{1}{\frac{b \cdot a}{0.5 \cdot \mathsf{PI}\left(\right)} \cdot \left(b + a\right)}
\end{array}
Derivation
  1. Initial program 79.2%

    \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
  2. Add Preprocessing
  3. Applied rewrites90.2%

    \[\leadsto \color{blue}{\mathsf{fma}\left(\frac{\mathsf{PI}\left(\right)}{a + b}, \frac{0.5}{b - a} \cdot {a}^{-1}, \frac{\frac{0.5}{a + b} \cdot \frac{\mathsf{PI}\left(\right)}{b - a}}{-b}\right)} \]
  4. Applied rewrites98.8%

    \[\leadsto \color{blue}{\frac{\frac{\mathsf{PI}\left(\right)}{a + b} \cdot \left(b - a\right)}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)}} \]
  5. Applied rewrites98.8%

    \[\leadsto \color{blue}{\frac{1}{\frac{b \cdot a}{0.5 \cdot \mathsf{PI}\left(\right)} \cdot \left(b + a\right)}} \]
  6. Add Preprocessing

Alternative 3: 73.3% accurate, 1.8× speedup?

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;a \leq -9.2 \cdot 10^{-99}:\\ \;\;\;\;\frac{\frac{0.5 \cdot \mathsf{PI}\left(\right)}{b \cdot a}}{a}\\ \mathbf{else}:\\ \;\;\;\;\frac{\frac{\mathsf{PI}\left(\right)}{b}}{2 \cdot \left(b \cdot a\right)}\\ \end{array} \end{array} \]
(FPCore (a b)
 :precision binary64
 (if (<= a -9.2e-99)
   (/ (/ (* 0.5 (PI)) (* b a)) a)
   (/ (/ (PI) b) (* 2.0 (* b a)))))
\begin{array}{l}

\\
\begin{array}{l}
\mathbf{if}\;a \leq -9.2 \cdot 10^{-99}:\\
\;\;\;\;\frac{\frac{0.5 \cdot \mathsf{PI}\left(\right)}{b \cdot a}}{a}\\

\mathbf{else}:\\
\;\;\;\;\frac{\frac{\mathsf{PI}\left(\right)}{b}}{2 \cdot \left(b \cdot a\right)}\\


\end{array}
\end{array}
Derivation
  1. Split input into 2 regimes
  2. if a < -9.1999999999999994e-99

    1. Initial program 79.4%

      \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
    2. Add Preprocessing
    3. Taylor expanded in b around 0

      \[\leadsto \color{blue}{\frac{1}{2} \cdot \frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \]
    4. Step-by-step derivation
      1. *-commutativeN/A

        \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
      2. lower-*.f64N/A

        \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
      3. lower-/.f64N/A

        \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \cdot \frac{1}{2} \]
      4. lower-PI.f64N/A

        \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{{a}^{2} \cdot b} \cdot \frac{1}{2} \]
      5. *-commutativeN/A

        \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
      6. lower-*.f64N/A

        \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
      7. unpow2N/A

        \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot \frac{1}{2} \]
      8. lower-*.f6473.8

        \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot 0.5 \]
    5. Applied rewrites73.8%

      \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{b \cdot \left(a \cdot a\right)} \cdot 0.5} \]
    6. Step-by-step derivation
      1. Applied rewrites82.9%

        \[\leadsto \frac{\frac{\mathsf{PI}\left(\right) \cdot 0.5}{a \cdot b}}{\color{blue}{a}} \]

      if -9.1999999999999994e-99 < a

      1. Initial program 79.2%

        \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
      2. Add Preprocessing
      3. Step-by-step derivation
        1. lift-*.f64N/A

          \[\leadsto \color{blue}{\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right)} \]
        2. lift-*.f64N/A

          \[\leadsto \color{blue}{\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right)} \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
        3. lift-/.f64N/A

          \[\leadsto \left(\color{blue}{\frac{\mathsf{PI}\left(\right)}{2}} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
        4. associate-*l/N/A

          \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right) \cdot \frac{1}{b \cdot b - a \cdot a}}{2}} \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
        5. lift--.f64N/A

          \[\leadsto \frac{\mathsf{PI}\left(\right) \cdot \frac{1}{b \cdot b - a \cdot a}}{2} \cdot \color{blue}{\left(\frac{1}{a} - \frac{1}{b}\right)} \]
        6. lift-/.f64N/A

          \[\leadsto \frac{\mathsf{PI}\left(\right) \cdot \frac{1}{b \cdot b - a \cdot a}}{2} \cdot \left(\color{blue}{\frac{1}{a}} - \frac{1}{b}\right) \]
        7. lift-/.f64N/A

          \[\leadsto \frac{\mathsf{PI}\left(\right) \cdot \frac{1}{b \cdot b - a \cdot a}}{2} \cdot \left(\frac{1}{a} - \color{blue}{\frac{1}{b}}\right) \]
        8. frac-subN/A

          \[\leadsto \frac{\mathsf{PI}\left(\right) \cdot \frac{1}{b \cdot b - a \cdot a}}{2} \cdot \color{blue}{\frac{1 \cdot b - a \cdot 1}{a \cdot b}} \]
        9. frac-timesN/A

          \[\leadsto \color{blue}{\frac{\left(\mathsf{PI}\left(\right) \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(1 \cdot b - a \cdot 1\right)}{2 \cdot \left(a \cdot b\right)}} \]
        10. lower-/.f64N/A

          \[\leadsto \color{blue}{\frac{\left(\mathsf{PI}\left(\right) \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(1 \cdot b - a \cdot 1\right)}{2 \cdot \left(a \cdot b\right)}} \]
      4. Applied rewrites86.3%

        \[\leadsto \color{blue}{\frac{\frac{\mathsf{PI}\left(\right)}{\left(a + b\right) \cdot \left(b - a\right)} \cdot \left(b - a\right)}{2 \cdot \left(a \cdot b\right)}} \]
      5. Taylor expanded in b around inf

        \[\leadsto \frac{\color{blue}{\frac{\mathsf{PI}\left(\right)}{b}}}{2 \cdot \left(a \cdot b\right)} \]
      6. Step-by-step derivation
        1. lower-/.f64N/A

          \[\leadsto \frac{\color{blue}{\frac{\mathsf{PI}\left(\right)}{b}}}{2 \cdot \left(a \cdot b\right)} \]
        2. lower-PI.f6472.5

          \[\leadsto \frac{\frac{\color{blue}{\mathsf{PI}\left(\right)}}{b}}{2 \cdot \left(a \cdot b\right)} \]
      7. Applied rewrites72.5%

        \[\leadsto \frac{\color{blue}{\frac{\mathsf{PI}\left(\right)}{b}}}{2 \cdot \left(a \cdot b\right)} \]
    7. Recombined 2 regimes into one program.
    8. Final simplification76.1%

      \[\leadsto \begin{array}{l} \mathbf{if}\;a \leq -9.2 \cdot 10^{-99}:\\ \;\;\;\;\frac{\frac{0.5 \cdot \mathsf{PI}\left(\right)}{b \cdot a}}{a}\\ \mathbf{else}:\\ \;\;\;\;\frac{\frac{\mathsf{PI}\left(\right)}{b}}{2 \cdot \left(b \cdot a\right)}\\ \end{array} \]
    9. Add Preprocessing

    Alternative 4: 73.3% accurate, 1.8× speedup?

    \[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;a \leq -9.2 \cdot 10^{-99}:\\ \;\;\;\;\frac{\frac{0.5 \cdot \mathsf{PI}\left(\right)}{b \cdot a}}{a}\\ \mathbf{else}:\\ \;\;\;\;\frac{\frac{\mathsf{PI}\left(\right)}{b \cdot a}}{b} \cdot 0.5\\ \end{array} \end{array} \]
    (FPCore (a b)
     :precision binary64
     (if (<= a -9.2e-99)
       (/ (/ (* 0.5 (PI)) (* b a)) a)
       (* (/ (/ (PI) (* b a)) b) 0.5)))
    \begin{array}{l}
    
    \\
    \begin{array}{l}
    \mathbf{if}\;a \leq -9.2 \cdot 10^{-99}:\\
    \;\;\;\;\frac{\frac{0.5 \cdot \mathsf{PI}\left(\right)}{b \cdot a}}{a}\\
    
    \mathbf{else}:\\
    \;\;\;\;\frac{\frac{\mathsf{PI}\left(\right)}{b \cdot a}}{b} \cdot 0.5\\
    
    
    \end{array}
    \end{array}
    
    Derivation
    1. Split input into 2 regimes
    2. if a < -9.1999999999999994e-99

      1. Initial program 79.4%

        \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
      2. Add Preprocessing
      3. Taylor expanded in b around 0

        \[\leadsto \color{blue}{\frac{1}{2} \cdot \frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \]
      4. Step-by-step derivation
        1. *-commutativeN/A

          \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
        2. lower-*.f64N/A

          \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
        3. lower-/.f64N/A

          \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \cdot \frac{1}{2} \]
        4. lower-PI.f64N/A

          \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{{a}^{2} \cdot b} \cdot \frac{1}{2} \]
        5. *-commutativeN/A

          \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
        6. lower-*.f64N/A

          \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
        7. unpow2N/A

          \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot \frac{1}{2} \]
        8. lower-*.f6473.8

          \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot 0.5 \]
      5. Applied rewrites73.8%

        \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{b \cdot \left(a \cdot a\right)} \cdot 0.5} \]
      6. Step-by-step derivation
        1. Applied rewrites82.9%

          \[\leadsto \frac{\frac{\mathsf{PI}\left(\right) \cdot 0.5}{a \cdot b}}{\color{blue}{a}} \]

        if -9.1999999999999994e-99 < a

        1. Initial program 79.2%

          \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
        2. Add Preprocessing
        3. Applied rewrites89.5%

          \[\leadsto \color{blue}{\mathsf{fma}\left(\frac{\mathsf{PI}\left(\right)}{a + b}, \frac{0.5}{b - a} \cdot {a}^{-1}, \frac{\frac{0.5}{a + b} \cdot \frac{\mathsf{PI}\left(\right)}{b - a}}{-b}\right)} \]
        4. Applied rewrites99.6%

          \[\leadsto \color{blue}{\frac{\frac{\mathsf{PI}\left(\right)}{a + b} \cdot \left(\left(b - a\right) \cdot \frac{0.5}{a \cdot b}\right)}{b - a}} \]
        5. Taylor expanded in b around inf

          \[\leadsto \color{blue}{\frac{1}{2} \cdot \frac{\mathsf{PI}\left(\right)}{a \cdot {b}^{2}}} \]
        6. Step-by-step derivation
          1. *-commutativeN/A

            \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{a \cdot {b}^{2}} \cdot \frac{1}{2}} \]
          2. lower-*.f64N/A

            \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{a \cdot {b}^{2}} \cdot \frac{1}{2}} \]
          3. associate-/r*N/A

            \[\leadsto \color{blue}{\frac{\frac{\mathsf{PI}\left(\right)}{a}}{{b}^{2}}} \cdot \frac{1}{2} \]
          4. unpow2N/A

            \[\leadsto \frac{\frac{\mathsf{PI}\left(\right)}{a}}{\color{blue}{b \cdot b}} \cdot \frac{1}{2} \]
          5. associate-/r*N/A

            \[\leadsto \color{blue}{\frac{\frac{\frac{\mathsf{PI}\left(\right)}{a}}{b}}{b}} \cdot \frac{1}{2} \]
          6. associate-/r*N/A

            \[\leadsto \frac{\color{blue}{\frac{\mathsf{PI}\left(\right)}{a \cdot b}}}{b} \cdot \frac{1}{2} \]
          7. lower-/.f64N/A

            \[\leadsto \color{blue}{\frac{\frac{\mathsf{PI}\left(\right)}{a \cdot b}}{b}} \cdot \frac{1}{2} \]
          8. lower-/.f64N/A

            \[\leadsto \frac{\color{blue}{\frac{\mathsf{PI}\left(\right)}{a \cdot b}}}{b} \cdot \frac{1}{2} \]
          9. lower-PI.f64N/A

            \[\leadsto \frac{\frac{\color{blue}{\mathsf{PI}\left(\right)}}{a \cdot b}}{b} \cdot \frac{1}{2} \]
          10. *-commutativeN/A

            \[\leadsto \frac{\frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot a}}}{b} \cdot \frac{1}{2} \]
          11. lower-*.f6472.5

            \[\leadsto \frac{\frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot a}}}{b} \cdot 0.5 \]
        7. Applied rewrites72.5%

          \[\leadsto \color{blue}{\frac{\frac{\mathsf{PI}\left(\right)}{b \cdot a}}{b} \cdot 0.5} \]
      7. Recombined 2 regimes into one program.
      8. Final simplification76.1%

        \[\leadsto \begin{array}{l} \mathbf{if}\;a \leq -9.2 \cdot 10^{-99}:\\ \;\;\;\;\frac{\frac{0.5 \cdot \mathsf{PI}\left(\right)}{b \cdot a}}{a}\\ \mathbf{else}:\\ \;\;\;\;\frac{\frac{\mathsf{PI}\left(\right)}{b \cdot a}}{b} \cdot 0.5\\ \end{array} \]
      9. Add Preprocessing

      Alternative 5: 70.5% accurate, 1.8× speedup?

      \[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;a \leq -3.2 \cdot 10^{-184}:\\ \;\;\;\;\frac{\frac{0.5 \cdot \mathsf{PI}\left(\right)}{b \cdot a}}{a}\\ \mathbf{else}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(2 \cdot \left(b \cdot a\right)\right) \cdot \left(b - a\right)}\\ \end{array} \end{array} \]
      (FPCore (a b)
       :precision binary64
       (if (<= a -3.2e-184)
         (/ (/ (* 0.5 (PI)) (* b a)) a)
         (/ (PI) (* (* 2.0 (* b a)) (- b a)))))
      \begin{array}{l}
      
      \\
      \begin{array}{l}
      \mathbf{if}\;a \leq -3.2 \cdot 10^{-184}:\\
      \;\;\;\;\frac{\frac{0.5 \cdot \mathsf{PI}\left(\right)}{b \cdot a}}{a}\\
      
      \mathbf{else}:\\
      \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(2 \cdot \left(b \cdot a\right)\right) \cdot \left(b - a\right)}\\
      
      
      \end{array}
      \end{array}
      
      Derivation
      1. Split input into 2 regimes
      2. if a < -3.2e-184

        1. Initial program 81.2%

          \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
        2. Add Preprocessing
        3. Taylor expanded in b around 0

          \[\leadsto \color{blue}{\frac{1}{2} \cdot \frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \]
        4. Step-by-step derivation
          1. *-commutativeN/A

            \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
          2. lower-*.f64N/A

            \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
          3. lower-/.f64N/A

            \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \cdot \frac{1}{2} \]
          4. lower-PI.f64N/A

            \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{{a}^{2} \cdot b} \cdot \frac{1}{2} \]
          5. *-commutativeN/A

            \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
          6. lower-*.f64N/A

            \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
          7. unpow2N/A

            \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot \frac{1}{2} \]
          8. lower-*.f6467.0

            \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot 0.5 \]
        5. Applied rewrites67.0%

          \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{b \cdot \left(a \cdot a\right)} \cdot 0.5} \]
        6. Step-by-step derivation
          1. Applied rewrites74.9%

            \[\leadsto \frac{\frac{\mathsf{PI}\left(\right) \cdot 0.5}{a \cdot b}}{\color{blue}{a}} \]

          if -3.2e-184 < a

          1. Initial program 77.9%

            \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
          2. Add Preprocessing
          3. Applied rewrites89.2%

            \[\leadsto \color{blue}{\mathsf{fma}\left(\frac{\mathsf{PI}\left(\right)}{a + b}, \frac{0.5}{b - a} \cdot {a}^{-1}, \frac{\frac{0.5}{a + b} \cdot \frac{\mathsf{PI}\left(\right)}{b - a}}{-b}\right)} \]
          4. Applied rewrites98.5%

            \[\leadsto \color{blue}{\frac{\frac{\mathsf{PI}\left(\right)}{a + b} \cdot \left(b - a\right)}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)}} \]
          5. Taylor expanded in b around inf

            \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)} \]
          6. Step-by-step derivation
            1. lower-PI.f6472.9

              \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)} \]
          7. Applied rewrites72.9%

            \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)} \]
        7. Recombined 2 regimes into one program.
        8. Final simplification73.7%

          \[\leadsto \begin{array}{l} \mathbf{if}\;a \leq -3.2 \cdot 10^{-184}:\\ \;\;\;\;\frac{\frac{0.5 \cdot \mathsf{PI}\left(\right)}{b \cdot a}}{a}\\ \mathbf{else}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(2 \cdot \left(b \cdot a\right)\right) \cdot \left(b - a\right)}\\ \end{array} \]
        9. Add Preprocessing

        Alternative 6: 70.4% accurate, 1.8× speedup?

        \[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;a \leq -3.2 \cdot 10^{-184}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{a} \cdot \frac{0.5}{b \cdot a}\\ \mathbf{else}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(2 \cdot \left(b \cdot a\right)\right) \cdot \left(b - a\right)}\\ \end{array} \end{array} \]
        (FPCore (a b)
         :precision binary64
         (if (<= a -3.2e-184)
           (* (/ (PI) a) (/ 0.5 (* b a)))
           (/ (PI) (* (* 2.0 (* b a)) (- b a)))))
        \begin{array}{l}
        
        \\
        \begin{array}{l}
        \mathbf{if}\;a \leq -3.2 \cdot 10^{-184}:\\
        \;\;\;\;\frac{\mathsf{PI}\left(\right)}{a} \cdot \frac{0.5}{b \cdot a}\\
        
        \mathbf{else}:\\
        \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(2 \cdot \left(b \cdot a\right)\right) \cdot \left(b - a\right)}\\
        
        
        \end{array}
        \end{array}
        
        Derivation
        1. Split input into 2 regimes
        2. if a < -3.2e-184

          1. Initial program 81.2%

            \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
          2. Add Preprocessing
          3. Taylor expanded in b around 0

            \[\leadsto \color{blue}{\frac{1}{2} \cdot \frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \]
          4. Step-by-step derivation
            1. *-commutativeN/A

              \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
            2. lower-*.f64N/A

              \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
            3. lower-/.f64N/A

              \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \cdot \frac{1}{2} \]
            4. lower-PI.f64N/A

              \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{{a}^{2} \cdot b} \cdot \frac{1}{2} \]
            5. *-commutativeN/A

              \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
            6. lower-*.f64N/A

              \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
            7. unpow2N/A

              \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot \frac{1}{2} \]
            8. lower-*.f6467.0

              \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot 0.5 \]
          5. Applied rewrites67.0%

            \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{b \cdot \left(a \cdot a\right)} \cdot 0.5} \]
          6. Step-by-step derivation
            1. Applied rewrites74.9%

              \[\leadsto \frac{0.5}{a \cdot b} \cdot \color{blue}{\frac{\mathsf{PI}\left(\right)}{a}} \]

            if -3.2e-184 < a

            1. Initial program 77.9%

              \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
            2. Add Preprocessing
            3. Applied rewrites89.2%

              \[\leadsto \color{blue}{\mathsf{fma}\left(\frac{\mathsf{PI}\left(\right)}{a + b}, \frac{0.5}{b - a} \cdot {a}^{-1}, \frac{\frac{0.5}{a + b} \cdot \frac{\mathsf{PI}\left(\right)}{b - a}}{-b}\right)} \]
            4. Applied rewrites98.5%

              \[\leadsto \color{blue}{\frac{\frac{\mathsf{PI}\left(\right)}{a + b} \cdot \left(b - a\right)}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)}} \]
            5. Taylor expanded in b around inf

              \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)} \]
            6. Step-by-step derivation
              1. lower-PI.f6472.9

                \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)} \]
            7. Applied rewrites72.9%

              \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)} \]
          7. Recombined 2 regimes into one program.
          8. Final simplification73.7%

            \[\leadsto \begin{array}{l} \mathbf{if}\;a \leq -3.2 \cdot 10^{-184}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{a} \cdot \frac{0.5}{b \cdot a}\\ \mathbf{else}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(2 \cdot \left(b \cdot a\right)\right) \cdot \left(b - a\right)}\\ \end{array} \]
          9. Add Preprocessing

          Alternative 7: 70.4% accurate, 2.0× speedup?

          \[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;a \leq -3.2 \cdot 10^{-184}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(b \cdot a\right) \cdot a} \cdot 0.5\\ \mathbf{else}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(2 \cdot \left(b \cdot a\right)\right) \cdot \left(b - a\right)}\\ \end{array} \end{array} \]
          (FPCore (a b)
           :precision binary64
           (if (<= a -3.2e-184)
             (* (/ (PI) (* (* b a) a)) 0.5)
             (/ (PI) (* (* 2.0 (* b a)) (- b a)))))
          \begin{array}{l}
          
          \\
          \begin{array}{l}
          \mathbf{if}\;a \leq -3.2 \cdot 10^{-184}:\\
          \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(b \cdot a\right) \cdot a} \cdot 0.5\\
          
          \mathbf{else}:\\
          \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(2 \cdot \left(b \cdot a\right)\right) \cdot \left(b - a\right)}\\
          
          
          \end{array}
          \end{array}
          
          Derivation
          1. Split input into 2 regimes
          2. if a < -3.2e-184

            1. Initial program 81.2%

              \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
            2. Add Preprocessing
            3. Taylor expanded in b around 0

              \[\leadsto \color{blue}{\frac{1}{2} \cdot \frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \]
            4. Step-by-step derivation
              1. *-commutativeN/A

                \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
              2. lower-*.f64N/A

                \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
              3. lower-/.f64N/A

                \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \cdot \frac{1}{2} \]
              4. lower-PI.f64N/A

                \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{{a}^{2} \cdot b} \cdot \frac{1}{2} \]
              5. *-commutativeN/A

                \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
              6. lower-*.f64N/A

                \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
              7. unpow2N/A

                \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot \frac{1}{2} \]
              8. lower-*.f6467.0

                \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot 0.5 \]
            5. Applied rewrites67.0%

              \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{b \cdot \left(a \cdot a\right)} \cdot 0.5} \]
            6. Step-by-step derivation
              1. Applied rewrites74.4%

                \[\leadsto \frac{\mathsf{PI}\left(\right)}{\left(a \cdot b\right) \cdot a} \cdot 0.5 \]

              if -3.2e-184 < a

              1. Initial program 77.9%

                \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
              2. Add Preprocessing
              3. Applied rewrites89.2%

                \[\leadsto \color{blue}{\mathsf{fma}\left(\frac{\mathsf{PI}\left(\right)}{a + b}, \frac{0.5}{b - a} \cdot {a}^{-1}, \frac{\frac{0.5}{a + b} \cdot \frac{\mathsf{PI}\left(\right)}{b - a}}{-b}\right)} \]
              4. Applied rewrites98.5%

                \[\leadsto \color{blue}{\frac{\frac{\mathsf{PI}\left(\right)}{a + b} \cdot \left(b - a\right)}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)}} \]
              5. Taylor expanded in b around inf

                \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)} \]
              6. Step-by-step derivation
                1. lower-PI.f6472.9

                  \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)} \]
              7. Applied rewrites72.9%

                \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{\left(b - a\right) \cdot \left(\left(a \cdot b\right) \cdot 2\right)} \]
            7. Recombined 2 regimes into one program.
            8. Final simplification73.5%

              \[\leadsto \begin{array}{l} \mathbf{if}\;a \leq -3.2 \cdot 10^{-184}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(b \cdot a\right) \cdot a} \cdot 0.5\\ \mathbf{else}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(2 \cdot \left(b \cdot a\right)\right) \cdot \left(b - a\right)}\\ \end{array} \]
            9. Add Preprocessing

            Alternative 8: 67.5% accurate, 2.2× speedup?

            \[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;a \leq -9.2 \cdot 10^{-99}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(b \cdot a\right) \cdot a} \cdot 0.5\\ \mathbf{else}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(b \cdot b\right) \cdot a} \cdot 0.5\\ \end{array} \end{array} \]
            (FPCore (a b)
             :precision binary64
             (if (<= a -9.2e-99)
               (* (/ (PI) (* (* b a) a)) 0.5)
               (* (/ (PI) (* (* b b) a)) 0.5)))
            \begin{array}{l}
            
            \\
            \begin{array}{l}
            \mathbf{if}\;a \leq -9.2 \cdot 10^{-99}:\\
            \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(b \cdot a\right) \cdot a} \cdot 0.5\\
            
            \mathbf{else}:\\
            \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(b \cdot b\right) \cdot a} \cdot 0.5\\
            
            
            \end{array}
            \end{array}
            
            Derivation
            1. Split input into 2 regimes
            2. if a < -9.1999999999999994e-99

              1. Initial program 79.4%

                \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
              2. Add Preprocessing
              3. Taylor expanded in b around 0

                \[\leadsto \color{blue}{\frac{1}{2} \cdot \frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \]
              4. Step-by-step derivation
                1. *-commutativeN/A

                  \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
                2. lower-*.f64N/A

                  \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
                3. lower-/.f64N/A

                  \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \cdot \frac{1}{2} \]
                4. lower-PI.f64N/A

                  \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{{a}^{2} \cdot b} \cdot \frac{1}{2} \]
                5. *-commutativeN/A

                  \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
                6. lower-*.f64N/A

                  \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
                7. unpow2N/A

                  \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot \frac{1}{2} \]
                8. lower-*.f6473.8

                  \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot 0.5 \]
              5. Applied rewrites73.8%

                \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{b \cdot \left(a \cdot a\right)} \cdot 0.5} \]
              6. Step-by-step derivation
                1. Applied rewrites82.3%

                  \[\leadsto \frac{\mathsf{PI}\left(\right)}{\left(a \cdot b\right) \cdot a} \cdot 0.5 \]

                if -9.1999999999999994e-99 < a

                1. Initial program 79.2%

                  \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
                2. Add Preprocessing
                3. Taylor expanded in b around inf

                  \[\leadsto \color{blue}{\frac{1}{2} \cdot \frac{\mathsf{PI}\left(\right)}{a \cdot {b}^{2}}} \]
                4. Step-by-step derivation
                  1. *-commutativeN/A

                    \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{a \cdot {b}^{2}} \cdot \frac{1}{2}} \]
                  2. lower-*.f64N/A

                    \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{a \cdot {b}^{2}} \cdot \frac{1}{2}} \]
                  3. lower-/.f64N/A

                    \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{a \cdot {b}^{2}}} \cdot \frac{1}{2} \]
                  4. lower-PI.f64N/A

                    \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{a \cdot {b}^{2}} \cdot \frac{1}{2} \]
                  5. *-commutativeN/A

                    \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{{b}^{2} \cdot a}} \cdot \frac{1}{2} \]
                  6. lower-*.f64N/A

                    \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{{b}^{2} \cdot a}} \cdot \frac{1}{2} \]
                  7. unpow2N/A

                    \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{\left(b \cdot b\right)} \cdot a} \cdot \frac{1}{2} \]
                  8. lower-*.f6464.0

                    \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{\left(b \cdot b\right)} \cdot a} \cdot 0.5 \]
                5. Applied rewrites64.0%

                  \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{\left(b \cdot b\right) \cdot a} \cdot 0.5} \]
              7. Recombined 2 regimes into one program.
              8. Final simplification70.4%

                \[\leadsto \begin{array}{l} \mathbf{if}\;a \leq -9.2 \cdot 10^{-99}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(b \cdot a\right) \cdot a} \cdot 0.5\\ \mathbf{else}:\\ \;\;\;\;\frac{\mathsf{PI}\left(\right)}{\left(b \cdot b\right) \cdot a} \cdot 0.5\\ \end{array} \]
              9. Add Preprocessing

              Alternative 9: 62.4% accurate, 2.6× speedup?

              \[\begin{array}{l} \\ \frac{\mathsf{PI}\left(\right)}{\left(b \cdot a\right) \cdot a} \cdot 0.5 \end{array} \]
              (FPCore (a b) :precision binary64 (* (/ (PI) (* (* b a) a)) 0.5))
              \begin{array}{l}
              
              \\
              \frac{\mathsf{PI}\left(\right)}{\left(b \cdot a\right) \cdot a} \cdot 0.5
              \end{array}
              
              Derivation
              1. Initial program 79.2%

                \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
              2. Add Preprocessing
              3. Taylor expanded in b around 0

                \[\leadsto \color{blue}{\frac{1}{2} \cdot \frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \]
              4. Step-by-step derivation
                1. *-commutativeN/A

                  \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
                2. lower-*.f64N/A

                  \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
                3. lower-/.f64N/A

                  \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \cdot \frac{1}{2} \]
                4. lower-PI.f64N/A

                  \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{{a}^{2} \cdot b} \cdot \frac{1}{2} \]
                5. *-commutativeN/A

                  \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
                6. lower-*.f64N/A

                  \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
                7. unpow2N/A

                  \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot \frac{1}{2} \]
                8. lower-*.f6458.4

                  \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot 0.5 \]
              5. Applied rewrites58.4%

                \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{b \cdot \left(a \cdot a\right)} \cdot 0.5} \]
              6. Step-by-step derivation
                1. Applied rewrites64.3%

                  \[\leadsto \frac{\mathsf{PI}\left(\right)}{\left(a \cdot b\right) \cdot a} \cdot 0.5 \]
                2. Final simplification64.3%

                  \[\leadsto \frac{\mathsf{PI}\left(\right)}{\left(b \cdot a\right) \cdot a} \cdot 0.5 \]
                3. Add Preprocessing

                Alternative 10: 62.4% accurate, 2.6× speedup?

                \[\begin{array}{l} \\ \frac{0.5}{\left(b \cdot a\right) \cdot a} \cdot \mathsf{PI}\left(\right) \end{array} \]
                (FPCore (a b) :precision binary64 (* (/ 0.5 (* (* b a) a)) (PI)))
                \begin{array}{l}
                
                \\
                \frac{0.5}{\left(b \cdot a\right) \cdot a} \cdot \mathsf{PI}\left(\right)
                \end{array}
                
                Derivation
                1. Initial program 79.2%

                  \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
                2. Add Preprocessing
                3. Taylor expanded in b around 0

                  \[\leadsto \color{blue}{\frac{1}{2} \cdot \frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \]
                4. Step-by-step derivation
                  1. *-commutativeN/A

                    \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
                  2. lower-*.f64N/A

                    \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
                  3. lower-/.f64N/A

                    \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \cdot \frac{1}{2} \]
                  4. lower-PI.f64N/A

                    \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{{a}^{2} \cdot b} \cdot \frac{1}{2} \]
                  5. *-commutativeN/A

                    \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
                  6. lower-*.f64N/A

                    \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
                  7. unpow2N/A

                    \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot \frac{1}{2} \]
                  8. lower-*.f6458.4

                    \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot 0.5 \]
                5. Applied rewrites58.4%

                  \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{b \cdot \left(a \cdot a\right)} \cdot 0.5} \]
                6. Step-by-step derivation
                  1. Applied rewrites58.3%

                    \[\leadsto \mathsf{PI}\left(\right) \cdot \color{blue}{\frac{0.5}{\left(a \cdot a\right) \cdot b}} \]
                  2. Step-by-step derivation
                    1. Applied rewrites64.2%

                      \[\leadsto \mathsf{PI}\left(\right) \cdot \frac{0.5}{\left(b \cdot a\right) \cdot \color{blue}{a}} \]
                    2. Final simplification64.2%

                      \[\leadsto \frac{0.5}{\left(b \cdot a\right) \cdot a} \cdot \mathsf{PI}\left(\right) \]
                    3. Add Preprocessing

                    Alternative 11: 56.5% accurate, 2.6× speedup?

                    \[\begin{array}{l} \\ \frac{0.5}{\left(a \cdot a\right) \cdot b} \cdot \mathsf{PI}\left(\right) \end{array} \]
                    (FPCore (a b) :precision binary64 (* (/ 0.5 (* (* a a) b)) (PI)))
                    \begin{array}{l}
                    
                    \\
                    \frac{0.5}{\left(a \cdot a\right) \cdot b} \cdot \mathsf{PI}\left(\right)
                    \end{array}
                    
                    Derivation
                    1. Initial program 79.2%

                      \[\left(\frac{\mathsf{PI}\left(\right)}{2} \cdot \frac{1}{b \cdot b - a \cdot a}\right) \cdot \left(\frac{1}{a} - \frac{1}{b}\right) \]
                    2. Add Preprocessing
                    3. Taylor expanded in b around 0

                      \[\leadsto \color{blue}{\frac{1}{2} \cdot \frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \]
                    4. Step-by-step derivation
                      1. *-commutativeN/A

                        \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
                      2. lower-*.f64N/A

                        \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b} \cdot \frac{1}{2}} \]
                      3. lower-/.f64N/A

                        \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{{a}^{2} \cdot b}} \cdot \frac{1}{2} \]
                      4. lower-PI.f64N/A

                        \[\leadsto \frac{\color{blue}{\mathsf{PI}\left(\right)}}{{a}^{2} \cdot b} \cdot \frac{1}{2} \]
                      5. *-commutativeN/A

                        \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
                      6. lower-*.f64N/A

                        \[\leadsto \frac{\mathsf{PI}\left(\right)}{\color{blue}{b \cdot {a}^{2}}} \cdot \frac{1}{2} \]
                      7. unpow2N/A

                        \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot \frac{1}{2} \]
                      8. lower-*.f6458.4

                        \[\leadsto \frac{\mathsf{PI}\left(\right)}{b \cdot \color{blue}{\left(a \cdot a\right)}} \cdot 0.5 \]
                    5. Applied rewrites58.4%

                      \[\leadsto \color{blue}{\frac{\mathsf{PI}\left(\right)}{b \cdot \left(a \cdot a\right)} \cdot 0.5} \]
                    6. Step-by-step derivation
                      1. Applied rewrites58.3%

                        \[\leadsto \mathsf{PI}\left(\right) \cdot \color{blue}{\frac{0.5}{\left(a \cdot a\right) \cdot b}} \]
                      2. Final simplification58.3%

                        \[\leadsto \frac{0.5}{\left(a \cdot a\right) \cdot b} \cdot \mathsf{PI}\left(\right) \]
                      3. Add Preprocessing

                      Reproduce

                      ?
                      herbie shell --seed 2024243 
                      (FPCore (a b)
                        :name "NMSE Section 6.1 mentioned, B"
                        :precision binary64
                        (* (* (/ (PI) 2.0) (/ 1.0 (- (* b b) (* a a)))) (- (/ 1.0 a) (/ 1.0 b))))