Migdal et al, Equation (51)

Percentage Accurate: 99.4% → 99.4%
Time: 6.9s
Alternatives: 7
Speedup: 1.2×

Specification

?
\[\frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot \pi\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)} \]
(FPCore (k n)
  :precision binary64
  (* (/ 1.0 (sqrt k)) (pow (* (* 2.0 PI) n) (/ (- 1.0 k) 2.0))))
double code(double k, double n) {
	return (1.0 / sqrt(k)) * pow(((2.0 * ((double) M_PI)) * n), ((1.0 - k) / 2.0));
}
public static double code(double k, double n) {
	return (1.0 / Math.sqrt(k)) * Math.pow(((2.0 * Math.PI) * n), ((1.0 - k) / 2.0));
}
def code(k, n):
	return (1.0 / math.sqrt(k)) * math.pow(((2.0 * math.pi) * n), ((1.0 - k) / 2.0))
function code(k, n)
	return Float64(Float64(1.0 / sqrt(k)) * (Float64(Float64(2.0 * pi) * n) ^ Float64(Float64(1.0 - k) / 2.0)))
end
function tmp = code(k, n)
	tmp = (1.0 / sqrt(k)) * (((2.0 * pi) * n) ^ ((1.0 - k) / 2.0));
end
code[k_, n_] := N[(N[(1.0 / N[Sqrt[k], $MachinePrecision]), $MachinePrecision] * N[Power[N[(N[(2.0 * Pi), $MachinePrecision] * n), $MachinePrecision], N[(N[(1.0 - k), $MachinePrecision] / 2.0), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]
\frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot \pi\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)}

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 7 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?

\[\frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot \pi\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)} \]
(FPCore (k n)
  :precision binary64
  (* (/ 1.0 (sqrt k)) (pow (* (* 2.0 PI) n) (/ (- 1.0 k) 2.0))))
double code(double k, double n) {
	return (1.0 / sqrt(k)) * pow(((2.0 * ((double) M_PI)) * n), ((1.0 - k) / 2.0));
}
public static double code(double k, double n) {
	return (1.0 / Math.sqrt(k)) * Math.pow(((2.0 * Math.PI) * n), ((1.0 - k) / 2.0));
}
def code(k, n):
	return (1.0 / math.sqrt(k)) * math.pow(((2.0 * math.pi) * n), ((1.0 - k) / 2.0))
function code(k, n)
	return Float64(Float64(1.0 / sqrt(k)) * (Float64(Float64(2.0 * pi) * n) ^ Float64(Float64(1.0 - k) / 2.0)))
end
function tmp = code(k, n)
	tmp = (1.0 / sqrt(k)) * (((2.0 * pi) * n) ^ ((1.0 - k) / 2.0));
end
code[k_, n_] := N[(N[(1.0 / N[Sqrt[k], $MachinePrecision]), $MachinePrecision] * N[Power[N[(N[(2.0 * Pi), $MachinePrecision] * n), $MachinePrecision], N[(N[(1.0 - k), $MachinePrecision] / 2.0), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]
\frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot \pi\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)}

Alternative 1: 99.4% accurate, 1.2× speedup?

\[\frac{{\left(n \cdot 6.283185307179586\right)}^{\left(\mathsf{fma}\left(k, -0.5, 0.5\right)\right)}}{\sqrt{k}} \]
(FPCore (k n)
  :precision binary64
  (/ (pow (* n 6.283185307179586) (fma k -0.5 0.5)) (sqrt k)))
double code(double k, double n) {
	return pow((n * 6.283185307179586), fma(k, -0.5, 0.5)) / sqrt(k);
}
function code(k, n)
	return Float64((Float64(n * 6.283185307179586) ^ fma(k, -0.5, 0.5)) / sqrt(k))
end
code[k_, n_] := N[(N[Power[N[(n * 6.283185307179586), $MachinePrecision], N[(k * -0.5 + 0.5), $MachinePrecision]], $MachinePrecision] / N[Sqrt[k], $MachinePrecision]), $MachinePrecision]
\frac{{\left(n \cdot 6.283185307179586\right)}^{\left(\mathsf{fma}\left(k, -0.5, 0.5\right)\right)}}{\sqrt{k}}
Derivation
  1. Initial program 99.4%

    \[\frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot \pi\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)} \]
  2. Step-by-step derivation
    1. lift-*.f64N/A

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

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

      \[\leadsto {\left(\left(2 \cdot \pi\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)} \cdot \color{blue}{\frac{1}{\sqrt{k}}} \]
    4. mult-flip-revN/A

      \[\leadsto \color{blue}{\frac{{\left(\left(2 \cdot \pi\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)}}{\sqrt{k}}} \]
    5. lower-/.f6499.4%

      \[\leadsto \color{blue}{\frac{{\left(\left(2 \cdot \pi\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)}}{\sqrt{k}}} \]
  3. Applied rewrites99.4%

    \[\leadsto \color{blue}{\frac{{\left(n \cdot \left(\pi + \pi\right)\right)}^{\left(\mathsf{fma}\left(k, -0.5, 0.5\right)\right)}}{\sqrt{k}}} \]
  4. Evaluated real constant99.4%

    \[\leadsto \frac{{\left(n \cdot \color{blue}{6.283185307179586}\right)}^{\left(\mathsf{fma}\left(k, -0.5, 0.5\right)\right)}}{\sqrt{k}} \]
  5. Add Preprocessing

Alternative 2: 98.0% accurate, 1.2× speedup?

\[\begin{array}{l} \mathbf{if}\;k \leq 0.38:\\ \;\;\;\;\sqrt{n + n} \cdot \sqrt{\frac{\pi}{k}}\\ \mathbf{else}:\\ \;\;\;\;\frac{{\left(n \cdot 6.283185307179586\right)}^{\left(-0.5 \cdot k\right)}}{\sqrt{k}}\\ \end{array} \]
(FPCore (k n)
  :precision binary64
  (if (<= k 0.38)
  (* (sqrt (+ n n)) (sqrt (/ PI k)))
  (/ (pow (* n 6.283185307179586) (* -0.5 k)) (sqrt k))))
double code(double k, double n) {
	double tmp;
	if (k <= 0.38) {
		tmp = sqrt((n + n)) * sqrt((((double) M_PI) / k));
	} else {
		tmp = pow((n * 6.283185307179586), (-0.5 * k)) / sqrt(k);
	}
	return tmp;
}
public static double code(double k, double n) {
	double tmp;
	if (k <= 0.38) {
		tmp = Math.sqrt((n + n)) * Math.sqrt((Math.PI / k));
	} else {
		tmp = Math.pow((n * 6.283185307179586), (-0.5 * k)) / Math.sqrt(k);
	}
	return tmp;
}
def code(k, n):
	tmp = 0
	if k <= 0.38:
		tmp = math.sqrt((n + n)) * math.sqrt((math.pi / k))
	else:
		tmp = math.pow((n * 6.283185307179586), (-0.5 * k)) / math.sqrt(k)
	return tmp
function code(k, n)
	tmp = 0.0
	if (k <= 0.38)
		tmp = Float64(sqrt(Float64(n + n)) * sqrt(Float64(pi / k)));
	else
		tmp = Float64((Float64(n * 6.283185307179586) ^ Float64(-0.5 * k)) / sqrt(k));
	end
	return tmp
end
function tmp_2 = code(k, n)
	tmp = 0.0;
	if (k <= 0.38)
		tmp = sqrt((n + n)) * sqrt((pi / k));
	else
		tmp = ((n * 6.283185307179586) ^ (-0.5 * k)) / sqrt(k);
	end
	tmp_2 = tmp;
end
code[k_, n_] := If[LessEqual[k, 0.38], N[(N[Sqrt[N[(n + n), $MachinePrecision]], $MachinePrecision] * N[Sqrt[N[(Pi / k), $MachinePrecision]], $MachinePrecision]), $MachinePrecision], N[(N[Power[N[(n * 6.283185307179586), $MachinePrecision], N[(-0.5 * k), $MachinePrecision]], $MachinePrecision] / N[Sqrt[k], $MachinePrecision]), $MachinePrecision]]
\begin{array}{l}
\mathbf{if}\;k \leq 0.38:\\
\;\;\;\;\sqrt{n + n} \cdot \sqrt{\frac{\pi}{k}}\\

\mathbf{else}:\\
\;\;\;\;\frac{{\left(n \cdot 6.283185307179586\right)}^{\left(-0.5 \cdot k\right)}}{\sqrt{k}}\\


\end{array}
Derivation
  1. Split input into 2 regimes
  2. if k < 0.38

    1. Initial program 99.4%

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

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

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

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

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

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

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\sqrt{k}} \]
      6. lower-sqrt.f6449.6%

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{0.5}}{\sqrt{k}} \]
    4. Applied rewrites49.6%

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

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\color{blue}{\sqrt{k}}} \]
    6. Applied rewrites37.7%

      \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
    7. Step-by-step derivation
      1. lift-/.f64N/A

        \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
      2. mult-flipN/A

        \[\leadsto \sqrt{\left(\left(\pi + \pi\right) \cdot n\right) \cdot \frac{1}{k}} \]
      3. lift-*.f64N/A

        \[\leadsto \sqrt{\left(\left(\pi + \pi\right) \cdot n\right) \cdot \frac{1}{k}} \]
      4. *-commutativeN/A

        \[\leadsto \sqrt{\left(n \cdot \left(\pi + \pi\right)\right) \cdot \frac{1}{k}} \]
      5. lift-/.f64N/A

        \[\leadsto \sqrt{\left(n \cdot \left(\pi + \pi\right)\right) \cdot \frac{1}{k}} \]
      6. associate-*l*N/A

        \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
      7. lower-*.f64N/A

        \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
      8. lift-/.f64N/A

        \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
      9. mult-flip-revN/A

        \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
      10. lower-/.f6437.6%

        \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
    8. Applied rewrites37.6%

      \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
    9. Step-by-step derivation
      1. lift-sqrt.f64N/A

        \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
      2. lift-*.f64N/A

        \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
      3. lift-/.f64N/A

        \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
      4. associate-*r/N/A

        \[\leadsto \sqrt{\frac{n \cdot \left(\pi + \pi\right)}{k}} \]
      5. lift-+.f64N/A

        \[\leadsto \sqrt{\frac{n \cdot \left(\pi + \pi\right)}{k}} \]
      6. count-2-revN/A

        \[\leadsto \sqrt{\frac{n \cdot \left(2 \cdot \pi\right)}{k}} \]
      7. associate-*r*N/A

        \[\leadsto \sqrt{\frac{\left(n \cdot 2\right) \cdot \pi}{k}} \]
      8. *-commutativeN/A

        \[\leadsto \sqrt{\frac{\left(2 \cdot n\right) \cdot \pi}{k}} \]
      9. associate-/l*N/A

        \[\leadsto \sqrt{\left(2 \cdot n\right) \cdot \frac{\pi}{k}} \]
      10. sqrt-prodN/A

        \[\leadsto \sqrt{2 \cdot n} \cdot \color{blue}{\sqrt{\frac{\pi}{k}}} \]
      11. lower-unsound-*.f64N/A

        \[\leadsto \sqrt{2 \cdot n} \cdot \color{blue}{\sqrt{\frac{\pi}{k}}} \]
      12. lower-unsound-sqrt.f64N/A

        \[\leadsto \sqrt{2 \cdot n} \cdot \sqrt{\color{blue}{\frac{\pi}{k}}} \]
      13. count-2-revN/A

        \[\leadsto \sqrt{n + n} \cdot \sqrt{\frac{\color{blue}{\pi}}{k}} \]
      14. lower-+.f64N/A

        \[\leadsto \sqrt{n + n} \cdot \sqrt{\frac{\color{blue}{\pi}}{k}} \]
      15. lower-unsound-sqrt.f64N/A

        \[\leadsto \sqrt{n + n} \cdot \sqrt{\frac{\pi}{k}} \]
      16. lower-/.f6449.6%

        \[\leadsto \sqrt{n + n} \cdot \sqrt{\frac{\pi}{k}} \]
    10. Applied rewrites49.6%

      \[\leadsto \sqrt{n + n} \cdot \color{blue}{\sqrt{\frac{\pi}{k}}} \]

    if 0.38 < k

    1. Initial program 99.4%

      \[\frac{1}{\sqrt{k}} \cdot {\left(\left(2 \cdot \pi\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)} \]
    2. Step-by-step derivation
      1. lift-*.f64N/A

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

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

        \[\leadsto {\left(\left(2 \cdot \pi\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)} \cdot \color{blue}{\frac{1}{\sqrt{k}}} \]
      4. mult-flip-revN/A

        \[\leadsto \color{blue}{\frac{{\left(\left(2 \cdot \pi\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)}}{\sqrt{k}}} \]
      5. lower-/.f6499.4%

        \[\leadsto \color{blue}{\frac{{\left(\left(2 \cdot \pi\right) \cdot n\right)}^{\left(\frac{1 - k}{2}\right)}}{\sqrt{k}}} \]
    3. Applied rewrites99.4%

      \[\leadsto \color{blue}{\frac{{\left(n \cdot \left(\pi + \pi\right)\right)}^{\left(\mathsf{fma}\left(k, -0.5, 0.5\right)\right)}}{\sqrt{k}}} \]
    4. Evaluated real constant99.4%

      \[\leadsto \frac{{\left(n \cdot \color{blue}{6.283185307179586}\right)}^{\left(\mathsf{fma}\left(k, -0.5, 0.5\right)\right)}}{\sqrt{k}} \]
    5. Taylor expanded in k around inf

      \[\leadsto \frac{{\left(n \cdot 6.283185307179586\right)}^{\color{blue}{\left(\frac{-1}{2} \cdot k\right)}}}{\sqrt{k}} \]
    6. Step-by-step derivation
      1. lower-*.f6453.2%

        \[\leadsto \frac{{\left(n \cdot 6.283185307179586\right)}^{\left(-0.5 \cdot \color{blue}{k}\right)}}{\sqrt{k}} \]
    7. Applied rewrites53.2%

      \[\leadsto \frac{{\left(n \cdot 6.283185307179586\right)}^{\color{blue}{\left(-0.5 \cdot k\right)}}}{\sqrt{k}} \]
  3. Recombined 2 regimes into one program.
  4. Add Preprocessing

Alternative 3: 74.0% accurate, 1.7× speedup?

\[\begin{array}{l} \mathbf{if}\;n \leq 10^{-12}:\\ \;\;\;\;\frac{n \cdot \sqrt{2 \cdot \frac{k \cdot \pi}{n}}}{k}\\ \mathbf{else}:\\ \;\;\;\;n \cdot \sqrt{2 \cdot \frac{\pi}{k \cdot n}}\\ \end{array} \]
(FPCore (k n)
  :precision binary64
  (if (<= n 1e-12)
  (/ (* n (sqrt (* 2.0 (/ (* k PI) n)))) k)
  (* n (sqrt (* 2.0 (/ PI (* k n)))))))
double code(double k, double n) {
	double tmp;
	if (n <= 1e-12) {
		tmp = (n * sqrt((2.0 * ((k * ((double) M_PI)) / n)))) / k;
	} else {
		tmp = n * sqrt((2.0 * (((double) M_PI) / (k * n))));
	}
	return tmp;
}
public static double code(double k, double n) {
	double tmp;
	if (n <= 1e-12) {
		tmp = (n * Math.sqrt((2.0 * ((k * Math.PI) / n)))) / k;
	} else {
		tmp = n * Math.sqrt((2.0 * (Math.PI / (k * n))));
	}
	return tmp;
}
def code(k, n):
	tmp = 0
	if n <= 1e-12:
		tmp = (n * math.sqrt((2.0 * ((k * math.pi) / n)))) / k
	else:
		tmp = n * math.sqrt((2.0 * (math.pi / (k * n))))
	return tmp
function code(k, n)
	tmp = 0.0
	if (n <= 1e-12)
		tmp = Float64(Float64(n * sqrt(Float64(2.0 * Float64(Float64(k * pi) / n)))) / k);
	else
		tmp = Float64(n * sqrt(Float64(2.0 * Float64(pi / Float64(k * n)))));
	end
	return tmp
end
function tmp_2 = code(k, n)
	tmp = 0.0;
	if (n <= 1e-12)
		tmp = (n * sqrt((2.0 * ((k * pi) / n)))) / k;
	else
		tmp = n * sqrt((2.0 * (pi / (k * n))));
	end
	tmp_2 = tmp;
end
code[k_, n_] := If[LessEqual[n, 1e-12], N[(N[(n * N[Sqrt[N[(2.0 * N[(N[(k * Pi), $MachinePrecision] / n), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] / k), $MachinePrecision], N[(n * N[Sqrt[N[(2.0 * N[(Pi / N[(k * n), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]]
\begin{array}{l}
\mathbf{if}\;n \leq 10^{-12}:\\
\;\;\;\;\frac{n \cdot \sqrt{2 \cdot \frac{k \cdot \pi}{n}}}{k}\\

\mathbf{else}:\\
\;\;\;\;n \cdot \sqrt{2 \cdot \frac{\pi}{k \cdot n}}\\


\end{array}
Derivation
  1. Split input into 2 regimes
  2. if n < 9.9999999999999998e-13

    1. Initial program 99.4%

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

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

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

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

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

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

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\sqrt{k}} \]
      6. lower-sqrt.f6449.6%

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{0.5}}{\sqrt{k}} \]
    4. Applied rewrites49.6%

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

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\color{blue}{\sqrt{k}}} \]
    6. Applied rewrites37.7%

      \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
    7. Taylor expanded in k around 0

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

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

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

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

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

        \[\leadsto \frac{\sqrt{2 \cdot \left(k \cdot \left(n \cdot \mathsf{PI}\left(\right)\right)\right)}}{k} \]
      6. lower-PI.f6438.2%

        \[\leadsto \frac{\sqrt{2 \cdot \left(k \cdot \left(n \cdot \pi\right)\right)}}{k} \]
    9. Applied rewrites38.2%

      \[\leadsto \frac{\sqrt{2 \cdot \left(k \cdot \left(n \cdot \pi\right)\right)}}{\color{blue}{k}} \]
    10. Taylor expanded in n around inf

      \[\leadsto \frac{n \cdot \sqrt{2 \cdot \frac{k \cdot \pi}{n}}}{k} \]
    11. Step-by-step derivation
      1. lower-*.f64N/A

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

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

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

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

        \[\leadsto \frac{n \cdot \sqrt{2 \cdot \frac{k \cdot \mathsf{PI}\left(\right)}{n}}}{k} \]
      6. lower-PI.f6450.2%

        \[\leadsto \frac{n \cdot \sqrt{2 \cdot \frac{k \cdot \pi}{n}}}{k} \]
    12. Applied rewrites50.2%

      \[\leadsto \frac{n \cdot \sqrt{2 \cdot \frac{k \cdot \pi}{n}}}{k} \]

    if 9.9999999999999998e-13 < n

    1. Initial program 99.4%

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

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

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

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

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

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

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\sqrt{k}} \]
      6. lower-sqrt.f6449.6%

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{0.5}}{\sqrt{k}} \]
    4. Applied rewrites49.6%

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

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\color{blue}{\sqrt{k}}} \]
    6. Applied rewrites37.7%

      \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
    7. Taylor expanded in n around inf

      \[\leadsto n \cdot \color{blue}{\sqrt{2 \cdot \frac{\pi}{k \cdot n}}} \]
    8. Step-by-step derivation
      1. lower-*.f64N/A

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

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

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

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

        \[\leadsto n \cdot \sqrt{2 \cdot \frac{\pi}{k \cdot n}} \]
      6. lower-*.f6449.8%

        \[\leadsto n \cdot \sqrt{2 \cdot \frac{\pi}{k \cdot n}} \]
    9. Applied rewrites49.8%

      \[\leadsto n \cdot \color{blue}{\sqrt{2 \cdot \frac{\pi}{k \cdot n}}} \]
  3. Recombined 2 regimes into one program.
  4. Add Preprocessing

Alternative 4: 61.4% accurate, 2.0× speedup?

\[\begin{array}{l} \mathbf{if}\;n \leq 1.85 \cdot 10^{-50}:\\ \;\;\;\;\sqrt{n \cdot \frac{\pi + \pi}{k}}\\ \mathbf{else}:\\ \;\;\;\;n \cdot \sqrt{2 \cdot \frac{\pi}{k \cdot n}}\\ \end{array} \]
(FPCore (k n)
  :precision binary64
  (if (<= n 1.85e-50)
  (sqrt (* n (/ (+ PI PI) k)))
  (* n (sqrt (* 2.0 (/ PI (* k n)))))))
double code(double k, double n) {
	double tmp;
	if (n <= 1.85e-50) {
		tmp = sqrt((n * ((((double) M_PI) + ((double) M_PI)) / k)));
	} else {
		tmp = n * sqrt((2.0 * (((double) M_PI) / (k * n))));
	}
	return tmp;
}
public static double code(double k, double n) {
	double tmp;
	if (n <= 1.85e-50) {
		tmp = Math.sqrt((n * ((Math.PI + Math.PI) / k)));
	} else {
		tmp = n * Math.sqrt((2.0 * (Math.PI / (k * n))));
	}
	return tmp;
}
def code(k, n):
	tmp = 0
	if n <= 1.85e-50:
		tmp = math.sqrt((n * ((math.pi + math.pi) / k)))
	else:
		tmp = n * math.sqrt((2.0 * (math.pi / (k * n))))
	return tmp
function code(k, n)
	tmp = 0.0
	if (n <= 1.85e-50)
		tmp = sqrt(Float64(n * Float64(Float64(pi + pi) / k)));
	else
		tmp = Float64(n * sqrt(Float64(2.0 * Float64(pi / Float64(k * n)))));
	end
	return tmp
end
function tmp_2 = code(k, n)
	tmp = 0.0;
	if (n <= 1.85e-50)
		tmp = sqrt((n * ((pi + pi) / k)));
	else
		tmp = n * sqrt((2.0 * (pi / (k * n))));
	end
	tmp_2 = tmp;
end
code[k_, n_] := If[LessEqual[n, 1.85e-50], N[Sqrt[N[(n * N[(N[(Pi + Pi), $MachinePrecision] / k), $MachinePrecision]), $MachinePrecision]], $MachinePrecision], N[(n * N[Sqrt[N[(2.0 * N[(Pi / N[(k * n), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]]
\begin{array}{l}
\mathbf{if}\;n \leq 1.85 \cdot 10^{-50}:\\
\;\;\;\;\sqrt{n \cdot \frac{\pi + \pi}{k}}\\

\mathbf{else}:\\
\;\;\;\;n \cdot \sqrt{2 \cdot \frac{\pi}{k \cdot n}}\\


\end{array}
Derivation
  1. Split input into 2 regimes
  2. if n < 1.85e-50

    1. Initial program 99.4%

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

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

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

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

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

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

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\sqrt{k}} \]
      6. lower-sqrt.f6449.6%

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{0.5}}{\sqrt{k}} \]
    4. Applied rewrites49.6%

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

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\color{blue}{\sqrt{k}}} \]
    6. Applied rewrites37.7%

      \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
    7. Step-by-step derivation
      1. lift-/.f64N/A

        \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
      2. mult-flipN/A

        \[\leadsto \sqrt{\left(\left(\pi + \pi\right) \cdot n\right) \cdot \frac{1}{k}} \]
      3. lift-*.f64N/A

        \[\leadsto \sqrt{\left(\left(\pi + \pi\right) \cdot n\right) \cdot \frac{1}{k}} \]
      4. *-commutativeN/A

        \[\leadsto \sqrt{\left(n \cdot \left(\pi + \pi\right)\right) \cdot \frac{1}{k}} \]
      5. lift-/.f64N/A

        \[\leadsto \sqrt{\left(n \cdot \left(\pi + \pi\right)\right) \cdot \frac{1}{k}} \]
      6. associate-*l*N/A

        \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
      7. lower-*.f64N/A

        \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
      8. lift-/.f64N/A

        \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
      9. mult-flip-revN/A

        \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
      10. lower-/.f6437.6%

        \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
    8. Applied rewrites37.6%

      \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]

    if 1.85e-50 < n

    1. Initial program 99.4%

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

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

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

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

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

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

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\sqrt{k}} \]
      6. lower-sqrt.f6449.6%

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{0.5}}{\sqrt{k}} \]
    4. Applied rewrites49.6%

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

        \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\color{blue}{\sqrt{k}}} \]
    6. Applied rewrites37.7%

      \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
    7. Taylor expanded in n around inf

      \[\leadsto n \cdot \color{blue}{\sqrt{2 \cdot \frac{\pi}{k \cdot n}}} \]
    8. Step-by-step derivation
      1. lower-*.f64N/A

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

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

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

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

        \[\leadsto n \cdot \sqrt{2 \cdot \frac{\pi}{k \cdot n}} \]
      6. lower-*.f6449.8%

        \[\leadsto n \cdot \sqrt{2 \cdot \frac{\pi}{k \cdot n}} \]
    9. Applied rewrites49.8%

      \[\leadsto n \cdot \color{blue}{\sqrt{2 \cdot \frac{\pi}{k \cdot n}}} \]
  3. Recombined 2 regimes into one program.
  4. Add Preprocessing

Alternative 5: 49.6% accurate, 2.7× speedup?

\[\sqrt{n + n} \cdot \sqrt{\frac{\pi}{k}} \]
(FPCore (k n)
  :precision binary64
  (* (sqrt (+ n n)) (sqrt (/ PI k))))
double code(double k, double n) {
	return sqrt((n + n)) * sqrt((((double) M_PI) / k));
}
public static double code(double k, double n) {
	return Math.sqrt((n + n)) * Math.sqrt((Math.PI / k));
}
def code(k, n):
	return math.sqrt((n + n)) * math.sqrt((math.pi / k))
function code(k, n)
	return Float64(sqrt(Float64(n + n)) * sqrt(Float64(pi / k)))
end
function tmp = code(k, n)
	tmp = sqrt((n + n)) * sqrt((pi / k));
end
code[k_, n_] := N[(N[Sqrt[N[(n + n), $MachinePrecision]], $MachinePrecision] * N[Sqrt[N[(Pi / k), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]
\sqrt{n + n} \cdot \sqrt{\frac{\pi}{k}}
Derivation
  1. Initial program 99.4%

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

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

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

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

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

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

      \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\sqrt{k}} \]
    6. lower-sqrt.f6449.6%

      \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{0.5}}{\sqrt{k}} \]
  4. Applied rewrites49.6%

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

      \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\color{blue}{\sqrt{k}}} \]
  6. Applied rewrites37.7%

    \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
  7. Step-by-step derivation
    1. lift-/.f64N/A

      \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
    2. mult-flipN/A

      \[\leadsto \sqrt{\left(\left(\pi + \pi\right) \cdot n\right) \cdot \frac{1}{k}} \]
    3. lift-*.f64N/A

      \[\leadsto \sqrt{\left(\left(\pi + \pi\right) \cdot n\right) \cdot \frac{1}{k}} \]
    4. *-commutativeN/A

      \[\leadsto \sqrt{\left(n \cdot \left(\pi + \pi\right)\right) \cdot \frac{1}{k}} \]
    5. lift-/.f64N/A

      \[\leadsto \sqrt{\left(n \cdot \left(\pi + \pi\right)\right) \cdot \frac{1}{k}} \]
    6. associate-*l*N/A

      \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
    7. lower-*.f64N/A

      \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
    8. lift-/.f64N/A

      \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
    9. mult-flip-revN/A

      \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
    10. lower-/.f6437.6%

      \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
  8. Applied rewrites37.6%

    \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
  9. Step-by-step derivation
    1. lift-sqrt.f64N/A

      \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
    2. lift-*.f64N/A

      \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
    3. lift-/.f64N/A

      \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
    4. associate-*r/N/A

      \[\leadsto \sqrt{\frac{n \cdot \left(\pi + \pi\right)}{k}} \]
    5. lift-+.f64N/A

      \[\leadsto \sqrt{\frac{n \cdot \left(\pi + \pi\right)}{k}} \]
    6. count-2-revN/A

      \[\leadsto \sqrt{\frac{n \cdot \left(2 \cdot \pi\right)}{k}} \]
    7. associate-*r*N/A

      \[\leadsto \sqrt{\frac{\left(n \cdot 2\right) \cdot \pi}{k}} \]
    8. *-commutativeN/A

      \[\leadsto \sqrt{\frac{\left(2 \cdot n\right) \cdot \pi}{k}} \]
    9. associate-/l*N/A

      \[\leadsto \sqrt{\left(2 \cdot n\right) \cdot \frac{\pi}{k}} \]
    10. sqrt-prodN/A

      \[\leadsto \sqrt{2 \cdot n} \cdot \color{blue}{\sqrt{\frac{\pi}{k}}} \]
    11. lower-unsound-*.f64N/A

      \[\leadsto \sqrt{2 \cdot n} \cdot \color{blue}{\sqrt{\frac{\pi}{k}}} \]
    12. lower-unsound-sqrt.f64N/A

      \[\leadsto \sqrt{2 \cdot n} \cdot \sqrt{\color{blue}{\frac{\pi}{k}}} \]
    13. count-2-revN/A

      \[\leadsto \sqrt{n + n} \cdot \sqrt{\frac{\color{blue}{\pi}}{k}} \]
    14. lower-+.f64N/A

      \[\leadsto \sqrt{n + n} \cdot \sqrt{\frac{\color{blue}{\pi}}{k}} \]
    15. lower-unsound-sqrt.f64N/A

      \[\leadsto \sqrt{n + n} \cdot \sqrt{\frac{\pi}{k}} \]
    16. lower-/.f6449.6%

      \[\leadsto \sqrt{n + n} \cdot \sqrt{\frac{\pi}{k}} \]
  10. Applied rewrites49.6%

    \[\leadsto \sqrt{n + n} \cdot \color{blue}{\sqrt{\frac{\pi}{k}}} \]
  11. Add Preprocessing

Alternative 6: 49.6% accurate, 2.7× speedup?

\[\sqrt{n} \cdot \sqrt{\frac{\pi + \pi}{k}} \]
(FPCore (k n)
  :precision binary64
  (* (sqrt n) (sqrt (/ (+ PI PI) k))))
double code(double k, double n) {
	return sqrt(n) * sqrt(((((double) M_PI) + ((double) M_PI)) / k));
}
public static double code(double k, double n) {
	return Math.sqrt(n) * Math.sqrt(((Math.PI + Math.PI) / k));
}
def code(k, n):
	return math.sqrt(n) * math.sqrt(((math.pi + math.pi) / k))
function code(k, n)
	return Float64(sqrt(n) * sqrt(Float64(Float64(pi + pi) / k)))
end
function tmp = code(k, n)
	tmp = sqrt(n) * sqrt(((pi + pi) / k));
end
code[k_, n_] := N[(N[Sqrt[n], $MachinePrecision] * N[Sqrt[N[(N[(Pi + Pi), $MachinePrecision] / k), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]
\sqrt{n} \cdot \sqrt{\frac{\pi + \pi}{k}}
Derivation
  1. Initial program 99.4%

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

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

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

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

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

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

      \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\sqrt{k}} \]
    6. lower-sqrt.f6449.6%

      \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{0.5}}{\sqrt{k}} \]
  4. Applied rewrites49.6%

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

      \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\color{blue}{\sqrt{k}}} \]
  6. Applied rewrites37.7%

    \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
  7. Step-by-step derivation
    1. lift-sqrt.f64N/A

      \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
    2. lift-/.f64N/A

      \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
    3. mult-flipN/A

      \[\leadsto \sqrt{\left(\left(\pi + \pi\right) \cdot n\right) \cdot \frac{1}{k}} \]
    4. lift-*.f64N/A

      \[\leadsto \sqrt{\left(\left(\pi + \pi\right) \cdot n\right) \cdot \frac{1}{k}} \]
    5. *-commutativeN/A

      \[\leadsto \sqrt{\left(n \cdot \left(\pi + \pi\right)\right) \cdot \frac{1}{k}} \]
    6. lift-/.f64N/A

      \[\leadsto \sqrt{\left(n \cdot \left(\pi + \pi\right)\right) \cdot \frac{1}{k}} \]
    7. associate-*l*N/A

      \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
    8. sqrt-prodN/A

      \[\leadsto \sqrt{n} \cdot \color{blue}{\sqrt{\left(\pi + \pi\right) \cdot \frac{1}{k}}} \]
    9. lower-unsound-*.f64N/A

      \[\leadsto \sqrt{n} \cdot \color{blue}{\sqrt{\left(\pi + \pi\right) \cdot \frac{1}{k}}} \]
    10. lower-unsound-sqrt.f64N/A

      \[\leadsto \sqrt{n} \cdot \sqrt{\color{blue}{\left(\pi + \pi\right) \cdot \frac{1}{k}}} \]
    11. lower-unsound-sqrt.f64N/A

      \[\leadsto \sqrt{n} \cdot \sqrt{\left(\pi + \pi\right) \cdot \frac{1}{k}} \]
    12. lift-/.f64N/A

      \[\leadsto \sqrt{n} \cdot \sqrt{\left(\pi + \pi\right) \cdot \frac{1}{k}} \]
    13. mult-flip-revN/A

      \[\leadsto \sqrt{n} \cdot \sqrt{\frac{\pi + \pi}{k}} \]
    14. lower-/.f6449.6%

      \[\leadsto \sqrt{n} \cdot \sqrt{\frac{\pi + \pi}{k}} \]
  8. Applied rewrites49.6%

    \[\leadsto \sqrt{n} \cdot \color{blue}{\sqrt{\frac{\pi + \pi}{k}}} \]
  9. Add Preprocessing

Alternative 7: 37.6% accurate, 3.2× speedup?

\[\sqrt{n \cdot \frac{\pi + \pi}{k}} \]
(FPCore (k n)
  :precision binary64
  (sqrt (* n (/ (+ PI PI) k))))
double code(double k, double n) {
	return sqrt((n * ((((double) M_PI) + ((double) M_PI)) / k)));
}
public static double code(double k, double n) {
	return Math.sqrt((n * ((Math.PI + Math.PI) / k)));
}
def code(k, n):
	return math.sqrt((n * ((math.pi + math.pi) / k)))
function code(k, n)
	return sqrt(Float64(n * Float64(Float64(pi + pi) / k)))
end
function tmp = code(k, n)
	tmp = sqrt((n * ((pi + pi) / k)));
end
code[k_, n_] := N[Sqrt[N[(n * N[(N[(Pi + Pi), $MachinePrecision] / k), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]
\sqrt{n \cdot \frac{\pi + \pi}{k}}
Derivation
  1. Initial program 99.4%

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

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

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

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

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

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

      \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\sqrt{k}} \]
    6. lower-sqrt.f6449.6%

      \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{0.5}}{\sqrt{k}} \]
  4. Applied rewrites49.6%

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

      \[\leadsto \frac{{\left(2 \cdot \left(n \cdot \pi\right)\right)}^{\frac{1}{2}}}{\color{blue}{\sqrt{k}}} \]
  6. Applied rewrites37.7%

    \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
  7. Step-by-step derivation
    1. lift-/.f64N/A

      \[\leadsto \sqrt{\frac{\left(\pi + \pi\right) \cdot n}{k}} \]
    2. mult-flipN/A

      \[\leadsto \sqrt{\left(\left(\pi + \pi\right) \cdot n\right) \cdot \frac{1}{k}} \]
    3. lift-*.f64N/A

      \[\leadsto \sqrt{\left(\left(\pi + \pi\right) \cdot n\right) \cdot \frac{1}{k}} \]
    4. *-commutativeN/A

      \[\leadsto \sqrt{\left(n \cdot \left(\pi + \pi\right)\right) \cdot \frac{1}{k}} \]
    5. lift-/.f64N/A

      \[\leadsto \sqrt{\left(n \cdot \left(\pi + \pi\right)\right) \cdot \frac{1}{k}} \]
    6. associate-*l*N/A

      \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
    7. lower-*.f64N/A

      \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
    8. lift-/.f64N/A

      \[\leadsto \sqrt{n \cdot \left(\left(\pi + \pi\right) \cdot \frac{1}{k}\right)} \]
    9. mult-flip-revN/A

      \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
    10. lower-/.f6437.6%

      \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
  8. Applied rewrites37.6%

    \[\leadsto \sqrt{n \cdot \frac{\pi + \pi}{k}} \]
  9. Add Preprocessing

Reproduce

?
herbie shell --seed 2025212 
(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))))