Migdal et al, Equation (51)

Percentage Accurate: 99.4% → 99.4%
Time: 22.0s
Alternatives: 8
Speedup: 1.1×

Specification

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

\\
\frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\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 8 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: 99.4% accurate, 1.0× speedup?

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

\\
\frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)}
\end{array}

Alternative 1: 99.4% accurate, 0.6× speedup?

\[\begin{array}{l} \\ \begin{array}{l} t_0 := \mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)\\ \sqrt{{k}^{-1}} \cdot \frac{\sqrt{t\_0}}{{t\_0}^{\left(0.5 \cdot k\right)}} \end{array} \end{array} \]
(FPCore (k n)
 :precision binary64
 (let* ((t_0 (* (PI) (* n 2.0))))
   (* (sqrt (pow k -1.0)) (/ (sqrt t_0) (pow t_0 (* 0.5 k))))))
\begin{array}{l}

\\
\begin{array}{l}
t_0 := \mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)\\
\sqrt{{k}^{-1}} \cdot \frac{\sqrt{t\_0}}{{t\_0}^{\left(0.5 \cdot k\right)}}
\end{array}
\end{array}
Derivation
  1. Initial program 99.5%

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

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

      \[\leadsto \frac{1}{\sqrt{k}} \cdot e^{\frac{1}{2} \cdot \color{blue}{\left(\left(1 - k\right) \cdot \log \left(2 \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right)\right)}} \]
    2. associate-*r*N/A

      \[\leadsto \frac{1}{\sqrt{k}} \cdot e^{\color{blue}{\left(\frac{1}{2} \cdot \left(1 - k\right)\right) \cdot \log \left(2 \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right)}} \]
    3. *-commutativeN/A

      \[\leadsto \frac{1}{\sqrt{k}} \cdot e^{\color{blue}{\log \left(2 \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right) \cdot \left(\frac{1}{2} \cdot \left(1 - k\right)\right)}} \]
    4. exp-prodN/A

      \[\leadsto \frac{1}{\sqrt{k}} \cdot \color{blue}{{\left(e^{\log \left(2 \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right)}\right)}^{\left(\frac{1}{2} \cdot \left(1 - k\right)\right)}} \]
    5. lower-pow.f64N/A

      \[\leadsto \frac{1}{\sqrt{k}} \cdot \color{blue}{{\left(e^{\log \left(2 \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right)}\right)}^{\left(\frac{1}{2} \cdot \left(1 - k\right)\right)}} \]
    6. rem-exp-logN/A

      \[\leadsto \frac{1}{\sqrt{k}} \cdot {\color{blue}{\left(2 \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right)}}^{\left(\frac{1}{2} \cdot \left(1 - k\right)\right)} \]
    7. associate-*r*N/A

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

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

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

      \[\leadsto \frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot n\right) \cdot \color{blue}{\mathsf{PI}\left(\right)}\right)}^{\left(\frac{1}{2} \cdot \left(1 - k\right)\right)} \]
    11. *-lft-identityN/A

      \[\leadsto \frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\frac{1}{2} \cdot \left(1 - \color{blue}{1 \cdot k}\right)\right)} \]
    12. metadata-evalN/A

      \[\leadsto \frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\frac{1}{2} \cdot \left(1 - \color{blue}{\left(\mathsf{neg}\left(-1\right)\right)} \cdot k\right)\right)} \]
    13. fp-cancel-sign-sub-invN/A

      \[\leadsto \frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\frac{1}{2} \cdot \color{blue}{\left(1 + -1 \cdot k\right)}\right)} \]
    14. +-commutativeN/A

      \[\leadsto \frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\frac{1}{2} \cdot \color{blue}{\left(-1 \cdot k + 1\right)}\right)} \]
    15. distribute-lft-inN/A

      \[\leadsto \frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\color{blue}{\left(\frac{1}{2} \cdot \left(-1 \cdot k\right) + \frac{1}{2} \cdot 1\right)}} \]
  5. Applied rewrites99.5%

    \[\leadsto \frac{1}{\sqrt{k}} \cdot \color{blue}{{\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(-0.5, k, 0.5\right)\right)}} \]
  6. Step-by-step derivation
    1. lift-/.f64N/A

      \[\leadsto \color{blue}{\frac{1}{\sqrt{k}}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(\frac{-1}{2}, k, \frac{1}{2}\right)\right)} \]
    2. lift-sqrt.f64N/A

      \[\leadsto \frac{1}{\color{blue}{\sqrt{k}}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(\frac{-1}{2}, k, \frac{1}{2}\right)\right)} \]
    3. pow1/2N/A

      \[\leadsto \frac{1}{\color{blue}{{k}^{\frac{1}{2}}}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(\frac{-1}{2}, k, \frac{1}{2}\right)\right)} \]
    4. pow-flipN/A

      \[\leadsto \color{blue}{{k}^{\left(\mathsf{neg}\left(\frac{1}{2}\right)\right)}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(\frac{-1}{2}, k, \frac{1}{2}\right)\right)} \]
    5. metadata-evalN/A

      \[\leadsto {k}^{\color{blue}{\frac{-1}{2}}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(\frac{-1}{2}, k, \frac{1}{2}\right)\right)} \]
    6. lift-pow.f6499.5

      \[\leadsto \color{blue}{{k}^{-0.5}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(-0.5, k, 0.5\right)\right)} \]
  7. Applied rewrites99.5%

    \[\leadsto \color{blue}{{k}^{-0.5}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(-0.5, k, 0.5\right)\right)} \]
  8. Step-by-step derivation
    1. Applied rewrites99.7%

      \[\leadsto {k}^{-0.5} \cdot \frac{\sqrt{\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)}}{\color{blue}{{\left(\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)\right)}^{\left(0.5 \cdot k\right)}}} \]
    2. Taylor expanded in k around 0

      \[\leadsto \color{blue}{\sqrt{\frac{1}{k}}} \cdot \frac{\sqrt{\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)}}{{\left(\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)\right)}^{\left(\frac{1}{2} \cdot k\right)}} \]
    3. Step-by-step derivation
      1. lower-sqrt.f64N/A

        \[\leadsto \color{blue}{\sqrt{\frac{1}{k}}} \cdot \frac{\sqrt{\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)}}{{\left(\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)\right)}^{\left(\frac{1}{2} \cdot k\right)}} \]
      2. lower-/.f6499.7

        \[\leadsto \sqrt{\color{blue}{\frac{1}{k}}} \cdot \frac{\sqrt{\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)}}{{\left(\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)\right)}^{\left(0.5 \cdot k\right)}} \]
    4. Applied rewrites99.7%

      \[\leadsto \color{blue}{\sqrt{\frac{1}{k}}} \cdot \frac{\sqrt{\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)}}{{\left(\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)\right)}^{\left(0.5 \cdot k\right)}} \]
    5. Final simplification99.7%

      \[\leadsto \sqrt{{k}^{-1}} \cdot \frac{\sqrt{\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)}}{{\left(\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)\right)}^{\left(0.5 \cdot k\right)}} \]
    6. Add Preprocessing

    Alternative 2: 61.9% accurate, 0.4× speedup?

    \[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;{\left(\sqrt{k}\right)}^{-1} \cdot {\left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)} \leq 0:\\ \;\;\;\;\frac{\sqrt{\mathsf{PI}\left(\right) \cdot \left(\mathsf{PI}\left(\right) \cdot 2\right)}}{{\left(k \cdot k\right)}^{0.25}}\\ \mathbf{else}:\\ \;\;\;\;\sqrt{\mathsf{PI}\left(\right) \cdot n} \cdot \sqrt{\frac{2}{k}}\\ \end{array} \end{array} \]
    (FPCore (k n)
     :precision binary64
     (if (<=
          (* (pow (sqrt k) -1.0) (pow (* (* 2.0 (PI)) n) (/ (- 1.0 k) 2.0)))
          0.0)
       (/ (sqrt (* (PI) (* (PI) 2.0))) (pow (* k k) 0.25))
       (* (sqrt (* (PI) n)) (sqrt (/ 2.0 k)))))
    \begin{array}{l}
    
    \\
    \begin{array}{l}
    \mathbf{if}\;{\left(\sqrt{k}\right)}^{-1} \cdot {\left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)} \leq 0:\\
    \;\;\;\;\frac{\sqrt{\mathsf{PI}\left(\right) \cdot \left(\mathsf{PI}\left(\right) \cdot 2\right)}}{{\left(k \cdot k\right)}^{0.25}}\\
    
    \mathbf{else}:\\
    \;\;\;\;\sqrt{\mathsf{PI}\left(\right) \cdot n} \cdot \sqrt{\frac{2}{k}}\\
    
    
    \end{array}
    \end{array}
    
    Derivation
    1. Split input into 2 regimes
    2. if (*.f64 (/.f64 #s(literal 1 binary64) (sqrt.f64 k)) (pow.f64 (*.f64 (*.f64 #s(literal 2 binary64) (PI.f64)) n) (/.f64 (-.f64 #s(literal 1 binary64) k) #s(literal 2 binary64)))) < 0.0

      1. Initial program 100.0%

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

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

          \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
        2. lower-*.f64N/A

          \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
        3. lower-sqrt.f64N/A

          \[\leadsto \color{blue}{\sqrt{2}} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}} \]
        4. lower-sqrt.f64N/A

          \[\leadsto \sqrt{2} \cdot \color{blue}{\sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
        5. lower-/.f64N/A

          \[\leadsto \sqrt{2} \cdot \sqrt{\color{blue}{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
        6. *-commutativeN/A

          \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
        7. lower-*.f64N/A

          \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
        8. lower-PI.f643.1

          \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right)} \cdot n}{k}} \]
      5. Applied rewrites3.1%

        \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k}}} \]
      6. Step-by-step derivation
        1. Applied rewrites3.8%

          \[\leadsto \frac{\sqrt{\mathsf{PI}\left(\right) \cdot \left(\mathsf{PI}\left(\right) \cdot 2\right)}}{\color{blue}{\sqrt{k}}} \]
        2. Step-by-step derivation
          1. Applied rewrites54.8%

            \[\leadsto \frac{\sqrt{\mathsf{PI}\left(\right) \cdot \left(\mathsf{PI}\left(\right) \cdot 2\right)}}{{\left(k \cdot k\right)}^{\color{blue}{0.25}}} \]

          if 0.0 < (*.f64 (/.f64 #s(literal 1 binary64) (sqrt.f64 k)) (pow.f64 (*.f64 (*.f64 #s(literal 2 binary64) (PI.f64)) n) (/.f64 (-.f64 #s(literal 1 binary64) k) #s(literal 2 binary64))))

          1. Initial program 99.4%

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

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

              \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
            2. lower-*.f64N/A

              \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
            3. lower-sqrt.f64N/A

              \[\leadsto \color{blue}{\sqrt{2}} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}} \]
            4. lower-sqrt.f64N/A

              \[\leadsto \sqrt{2} \cdot \color{blue}{\sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
            5. lower-/.f64N/A

              \[\leadsto \sqrt{2} \cdot \sqrt{\color{blue}{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
            6. *-commutativeN/A

              \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
            7. lower-*.f64N/A

              \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
            8. lower-PI.f6445.0

              \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right)} \cdot n}{k}} \]
          5. Applied rewrites45.0%

            \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k}}} \]
          6. Step-by-step derivation
            1. Applied rewrites45.1%

              \[\leadsto \color{blue}{\sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k} \cdot 2}} \]
            2. Step-by-step derivation
              1. Applied rewrites61.9%

                \[\leadsto \sqrt{\mathsf{PI}\left(\right) \cdot n} \cdot \color{blue}{\sqrt{\frac{2}{k}}} \]
            3. Recombined 2 regimes into one program.
            4. Final simplification60.2%

              \[\leadsto \begin{array}{l} \mathbf{if}\;{\left(\sqrt{k}\right)}^{-1} \cdot {\left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)} \leq 0:\\ \;\;\;\;\frac{\sqrt{\mathsf{PI}\left(\right) \cdot \left(\mathsf{PI}\left(\right) \cdot 2\right)}}{{\left(k \cdot k\right)}^{0.25}}\\ \mathbf{else}:\\ \;\;\;\;\sqrt{\mathsf{PI}\left(\right) \cdot n} \cdot \sqrt{\frac{2}{k}}\\ \end{array} \]
            5. Add Preprocessing

            Alternative 3: 99.4% accurate, 1.1× speedup?

            \[\begin{array}{l} \\ \frac{{\left(\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)\right)}^{\left(\mathsf{fma}\left(-0.5, k, 0.5\right)\right)}}{\sqrt{k}} \end{array} \]
            (FPCore (k n)
             :precision binary64
             (/ (pow (* (PI) (* n 2.0)) (fma -0.5 k 0.5)) (sqrt k)))
            \begin{array}{l}
            
            \\
            \frac{{\left(\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)\right)}^{\left(\mathsf{fma}\left(-0.5, k, 0.5\right)\right)}}{\sqrt{k}}
            \end{array}
            
            Derivation
            1. Initial program 99.5%

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

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

                \[\leadsto \frac{1}{\sqrt{k}} \cdot e^{\frac{1}{2} \cdot \color{blue}{\left(\left(1 - k\right) \cdot \log \left(2 \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right)\right)}} \]
              2. associate-*r*N/A

                \[\leadsto \frac{1}{\sqrt{k}} \cdot e^{\color{blue}{\left(\frac{1}{2} \cdot \left(1 - k\right)\right) \cdot \log \left(2 \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right)}} \]
              3. *-commutativeN/A

                \[\leadsto \frac{1}{\sqrt{k}} \cdot e^{\color{blue}{\log \left(2 \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right) \cdot \left(\frac{1}{2} \cdot \left(1 - k\right)\right)}} \]
              4. exp-prodN/A

                \[\leadsto \frac{1}{\sqrt{k}} \cdot \color{blue}{{\left(e^{\log \left(2 \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right)}\right)}^{\left(\frac{1}{2} \cdot \left(1 - k\right)\right)}} \]
              5. lower-pow.f64N/A

                \[\leadsto \frac{1}{\sqrt{k}} \cdot \color{blue}{{\left(e^{\log \left(2 \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right)}\right)}^{\left(\frac{1}{2} \cdot \left(1 - k\right)\right)}} \]
              6. rem-exp-logN/A

                \[\leadsto \frac{1}{\sqrt{k}} \cdot {\color{blue}{\left(2 \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right)}}^{\left(\frac{1}{2} \cdot \left(1 - k\right)\right)} \]
              7. associate-*r*N/A

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

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

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

                \[\leadsto \frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot n\right) \cdot \color{blue}{\mathsf{PI}\left(\right)}\right)}^{\left(\frac{1}{2} \cdot \left(1 - k\right)\right)} \]
              11. *-lft-identityN/A

                \[\leadsto \frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\frac{1}{2} \cdot \left(1 - \color{blue}{1 \cdot k}\right)\right)} \]
              12. metadata-evalN/A

                \[\leadsto \frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\frac{1}{2} \cdot \left(1 - \color{blue}{\left(\mathsf{neg}\left(-1\right)\right)} \cdot k\right)\right)} \]
              13. fp-cancel-sign-sub-invN/A

                \[\leadsto \frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\frac{1}{2} \cdot \color{blue}{\left(1 + -1 \cdot k\right)}\right)} \]
              14. +-commutativeN/A

                \[\leadsto \frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\frac{1}{2} \cdot \color{blue}{\left(-1 \cdot k + 1\right)}\right)} \]
              15. distribute-lft-inN/A

                \[\leadsto \frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\color{blue}{\left(\frac{1}{2} \cdot \left(-1 \cdot k\right) + \frac{1}{2} \cdot 1\right)}} \]
            5. Applied rewrites99.5%

              \[\leadsto \frac{1}{\sqrt{k}} \cdot \color{blue}{{\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(-0.5, k, 0.5\right)\right)}} \]
            6. Step-by-step derivation
              1. lift-/.f64N/A

                \[\leadsto \color{blue}{\frac{1}{\sqrt{k}}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(\frac{-1}{2}, k, \frac{1}{2}\right)\right)} \]
              2. lift-sqrt.f64N/A

                \[\leadsto \frac{1}{\color{blue}{\sqrt{k}}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(\frac{-1}{2}, k, \frac{1}{2}\right)\right)} \]
              3. pow1/2N/A

                \[\leadsto \frac{1}{\color{blue}{{k}^{\frac{1}{2}}}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(\frac{-1}{2}, k, \frac{1}{2}\right)\right)} \]
              4. pow-flipN/A

                \[\leadsto \color{blue}{{k}^{\left(\mathsf{neg}\left(\frac{1}{2}\right)\right)}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(\frac{-1}{2}, k, \frac{1}{2}\right)\right)} \]
              5. metadata-evalN/A

                \[\leadsto {k}^{\color{blue}{\frac{-1}{2}}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(\frac{-1}{2}, k, \frac{1}{2}\right)\right)} \]
              6. lift-pow.f6499.5

                \[\leadsto \color{blue}{{k}^{-0.5}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(-0.5, k, 0.5\right)\right)} \]
            7. Applied rewrites99.5%

              \[\leadsto \color{blue}{{k}^{-0.5}} \cdot {\left(\left(2 \cdot n\right) \cdot \mathsf{PI}\left(\right)\right)}^{\left(\mathsf{fma}\left(-0.5, k, 0.5\right)\right)} \]
            8. Applied rewrites99.6%

              \[\leadsto \color{blue}{\frac{{\left(\mathsf{PI}\left(\right) \cdot \left(n \cdot 2\right)\right)}^{\left(\mathsf{fma}\left(-0.5, k, 0.5\right)\right)}}{\sqrt{k}}} \]
            9. Add Preprocessing

            Alternative 4: 49.4% accurate, 3.6× speedup?

            \[\begin{array}{l} \\ \sqrt{\mathsf{PI}\left(\right) \cdot n} \cdot \sqrt{\frac{2}{k}} \end{array} \]
            (FPCore (k n) :precision binary64 (* (sqrt (* (PI) n)) (sqrt (/ 2.0 k))))
            \begin{array}{l}
            
            \\
            \sqrt{\mathsf{PI}\left(\right) \cdot n} \cdot \sqrt{\frac{2}{k}}
            \end{array}
            
            Derivation
            1. Initial program 99.5%

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

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

                \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
              2. lower-*.f64N/A

                \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
              3. lower-sqrt.f64N/A

                \[\leadsto \color{blue}{\sqrt{2}} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}} \]
              4. lower-sqrt.f64N/A

                \[\leadsto \sqrt{2} \cdot \color{blue}{\sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
              5. lower-/.f64N/A

                \[\leadsto \sqrt{2} \cdot \sqrt{\color{blue}{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
              6. *-commutativeN/A

                \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
              7. lower-*.f64N/A

                \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
              8. lower-PI.f6434.9

                \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right)} \cdot n}{k}} \]
            5. Applied rewrites34.9%

              \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k}}} \]
            6. Step-by-step derivation
              1. Applied rewrites35.0%

                \[\leadsto \color{blue}{\sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k} \cdot 2}} \]
              2. Step-by-step derivation
                1. Applied rewrites47.7%

                  \[\leadsto \sqrt{\mathsf{PI}\left(\right) \cdot n} \cdot \color{blue}{\sqrt{\frac{2}{k}}} \]
                2. Add Preprocessing

                Alternative 5: 38.0% accurate, 4.8× speedup?

                \[\begin{array}{l} \\ \sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k} \cdot 2} \end{array} \]
                (FPCore (k n) :precision binary64 (sqrt (* (/ (* (PI) n) k) 2.0)))
                \begin{array}{l}
                
                \\
                \sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k} \cdot 2}
                \end{array}
                
                Derivation
                1. Initial program 99.5%

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

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

                    \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                  2. lower-*.f64N/A

                    \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                  3. lower-sqrt.f64N/A

                    \[\leadsto \color{blue}{\sqrt{2}} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}} \]
                  4. lower-sqrt.f64N/A

                    \[\leadsto \sqrt{2} \cdot \color{blue}{\sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                  5. lower-/.f64N/A

                    \[\leadsto \sqrt{2} \cdot \sqrt{\color{blue}{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                  6. *-commutativeN/A

                    \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
                  7. lower-*.f64N/A

                    \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
                  8. lower-PI.f6434.9

                    \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right)} \cdot n}{k}} \]
                5. Applied rewrites34.9%

                  \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k}}} \]
                6. Step-by-step derivation
                  1. Applied rewrites35.0%

                    \[\leadsto \color{blue}{\sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k} \cdot 2}} \]
                  2. Add Preprocessing

                  Alternative 6: 38.0% accurate, 4.8× speedup?

                  \[\begin{array}{l} \\ \sqrt{\left(\mathsf{PI}\left(\right) \cdot 2\right) \cdot \frac{n}{k}} \end{array} \]
                  (FPCore (k n) :precision binary64 (sqrt (* (* (PI) 2.0) (/ n k))))
                  \begin{array}{l}
                  
                  \\
                  \sqrt{\left(\mathsf{PI}\left(\right) \cdot 2\right) \cdot \frac{n}{k}}
                  \end{array}
                  
                  Derivation
                  1. Initial program 99.5%

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

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

                      \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                    2. lower-*.f64N/A

                      \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                    3. lower-sqrt.f64N/A

                      \[\leadsto \color{blue}{\sqrt{2}} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}} \]
                    4. lower-sqrt.f64N/A

                      \[\leadsto \sqrt{2} \cdot \color{blue}{\sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                    5. lower-/.f64N/A

                      \[\leadsto \sqrt{2} \cdot \sqrt{\color{blue}{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                    6. *-commutativeN/A

                      \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
                    7. lower-*.f64N/A

                      \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
                    8. lower-PI.f6434.9

                      \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right)} \cdot n}{k}} \]
                  5. Applied rewrites34.9%

                    \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k}}} \]
                  6. Step-by-step derivation
                    1. Applied rewrites35.0%

                      \[\leadsto \color{blue}{\sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k} \cdot 2}} \]
                    2. Step-by-step derivation
                      1. Applied rewrites35.0%

                        \[\leadsto \sqrt{\left(\mathsf{PI}\left(\right) \cdot 2\right) \cdot \frac{n}{k}} \]
                      2. Add Preprocessing

                      Alternative 7: 38.0% accurate, 4.8× speedup?

                      \[\begin{array}{l} \\ \sqrt{\left(n \cdot \frac{\mathsf{PI}\left(\right)}{k}\right) \cdot 2} \end{array} \]
                      (FPCore (k n) :precision binary64 (sqrt (* (* n (/ (PI) k)) 2.0)))
                      \begin{array}{l}
                      
                      \\
                      \sqrt{\left(n \cdot \frac{\mathsf{PI}\left(\right)}{k}\right) \cdot 2}
                      \end{array}
                      
                      Derivation
                      1. Initial program 99.5%

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

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

                          \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                        2. lower-*.f64N/A

                          \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                        3. lower-sqrt.f64N/A

                          \[\leadsto \color{blue}{\sqrt{2}} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}} \]
                        4. lower-sqrt.f64N/A

                          \[\leadsto \sqrt{2} \cdot \color{blue}{\sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                        5. lower-/.f64N/A

                          \[\leadsto \sqrt{2} \cdot \sqrt{\color{blue}{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                        6. *-commutativeN/A

                          \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
                        7. lower-*.f64N/A

                          \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
                        8. lower-PI.f6434.9

                          \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right)} \cdot n}{k}} \]
                      5. Applied rewrites34.9%

                        \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k}}} \]
                      6. Step-by-step derivation
                        1. Applied rewrites35.0%

                          \[\leadsto \color{blue}{\sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k} \cdot 2}} \]
                        2. Step-by-step derivation
                          1. Applied rewrites34.9%

                            \[\leadsto \sqrt{\left(n \cdot \frac{\mathsf{PI}\left(\right)}{k}\right) \cdot 2} \]
                          2. Add Preprocessing

                          Alternative 8: 5.1% accurate, 5.6× speedup?

                          \[\begin{array}{l} \\ \mathsf{PI}\left(\right) \cdot \sqrt{\frac{2}{k}} \end{array} \]
                          (FPCore (k n) :precision binary64 (* (PI) (sqrt (/ 2.0 k))))
                          \begin{array}{l}
                          
                          \\
                          \mathsf{PI}\left(\right) \cdot \sqrt{\frac{2}{k}}
                          \end{array}
                          
                          Derivation
                          1. Initial program 99.5%

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

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

                              \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                            2. lower-*.f64N/A

                              \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                            3. lower-sqrt.f64N/A

                              \[\leadsto \color{blue}{\sqrt{2}} \cdot \sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}} \]
                            4. lower-sqrt.f64N/A

                              \[\leadsto \sqrt{2} \cdot \color{blue}{\sqrt{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                            5. lower-/.f64N/A

                              \[\leadsto \sqrt{2} \cdot \sqrt{\color{blue}{\frac{n \cdot \mathsf{PI}\left(\right)}{k}}} \]
                            6. *-commutativeN/A

                              \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
                            7. lower-*.f64N/A

                              \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right) \cdot n}}{k}} \]
                            8. lower-PI.f6434.9

                              \[\leadsto \sqrt{2} \cdot \sqrt{\frac{\color{blue}{\mathsf{PI}\left(\right)} \cdot n}{k}} \]
                          5. Applied rewrites34.9%

                            \[\leadsto \color{blue}{\sqrt{2} \cdot \sqrt{\frac{\mathsf{PI}\left(\right) \cdot n}{k}}} \]
                          6. Step-by-step derivation
                            1. Applied rewrites5.0%

                              \[\leadsto \frac{\sqrt{\mathsf{PI}\left(\right) \cdot \left(\mathsf{PI}\left(\right) \cdot 2\right)}}{\color{blue}{\sqrt{k}}} \]
                            2. Step-by-step derivation
                              1. Applied rewrites5.0%

                                \[\leadsto \mathsf{PI}\left(\right) \cdot \color{blue}{\sqrt{\frac{2}{k}}} \]
                              2. Add Preprocessing

                              Reproduce

                              ?
                              herbie shell --seed 2024331 
                              (FPCore (k n)
                                :name "Migdal et al, Equation (51)"
                                :precision binary64
                                (* (/ 1.0 (sqrt k)) (pow (* (* 2.0 (PI)) n) (/ (- 1.0 k) 2.0))))