Numeric.Integration.TanhSinh:everywhere from integration-0.2.1

Average Error: 8.04% → 0.13%

Time: 4.1s

Precision: binary64

Cost: 6985

Math

\[x \cdot \left(1 + y \cdot y\right) \]

↓

\[\begin{array}{l} \mathbf{if}\;y \leq -3.8 \cdot 10^{+18} \lor \neg \left(y \leq 7.8 \cdot 10^{+98}\right):\\ \;\;\;\;y \cdot \left(y \cdot x\right)\\ \mathbf{else}:\\ \;\;\;\;\mathsf{fma}\left(x, y \cdot y, x\right)\\ \end{array} \]

FPCore

(FPCore (x y) :precision binary64 (* x (+ 1.0 (* y y))))

↓

(FPCore (x y)
 :precision binary64
 (if (or (<= y -3.8e+18) (not (<= y 7.8e+98)))
   (* y (* y x))
   (fma x (* y y) x)))

C

double code(double x, double y) {
	return x * (1.0 + (y * y));
}

↓

double code(double x, double y) {
	double tmp;
	if ((y <= -3.8e+18) || !(y <= 7.8e+98)) {
		tmp = y * (y * x);
	} else {
		tmp = fma(x, (y * y), x);
	}
	return tmp;
}

Julia

function code(x, y)
	return Float64(x * Float64(1.0 + Float64(y * y)))
end

↓

function code(x, y)
	tmp = 0.0
	if ((y <= -3.8e+18) || !(y <= 7.8e+98))
		tmp = Float64(y * Float64(y * x));
	else
		tmp = fma(x, Float64(y * y), x);
	end
	return tmp
end

Wolfram

code[x_, y_] := N[(x * N[(1.0 + N[(y * y), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

↓

code[x_, y_] := If[Or[LessEqual[y, -3.8e+18], N[Not[LessEqual[y, 7.8e+98]], $MachinePrecision]], N[(y * N[(y * x), $MachinePrecision]), $MachinePrecision], N[(x * N[(y * y), $MachinePrecision] + x), $MachinePrecision]]

Target

Original	8.04%
Target	0.13%
Herbie	0.13%

\[x + \left(x \cdot y\right) \cdot y \]

Derivation

1. Split input into 2 regimes

2. if y < -3.8e18 or 7.7999999999999999e98 < y

   1. Initial program 36.25

      \[x \cdot \left(1 + y \cdot y\right) \]

   2. Taylor expanded in y around inf 36.25

      \[\leadsto \color{blue}{{y}^{2} \cdot x} \]

   3. Simplified 0.38

      \[\leadsto \color{blue}{y \cdot \left(y \cdot x\right)} \]
      Proof

      [Start] 36.25	\[ {y}^{2} \cdot x \]
      unpow2 [=>] 36.25	\[ \color{blue}{\left(y \cdot y\right)} \cdot x \]
      associate-*l* [=>] 0.38	\[ \color{blue}{y \cdot \left(y \cdot x\right)} \]

   if -3.8e18 < y < 7.7999999999999999e98

   1. Initial program 0.07

      \[x \cdot \left(1 + y \cdot y\right) \]

   2. Simplified 0.06

      \[\leadsto \color{blue}{\mathsf{fma}\left(x, y \cdot y, x\right)} \]
      Proof

      [Start] 0.07	\[ x \cdot \left(1 + y \cdot y\right) \]
      distribute-lft-in [=>] 0.06	\[ \color{blue}{x \cdot 1 + x \cdot \left(y \cdot y\right)} \]
      +-commutative [=>] 0.06	\[ \color{blue}{x \cdot \left(y \cdot y\right) + x \cdot 1} \]
      *-rgt-identity [=>] 0.06	\[ x \cdot \left(y \cdot y\right) + \color{blue}{x} \]
      fma-def [=>] 0.06	\[ \color{blue}{\mathsf{fma}\left(x, y \cdot y, x\right)} \]

3. Recombined 2 regimes into one program.

4. Final simplification 0.13

   \[\leadsto \begin{array}{l} \mathbf{if}\;y \leq -3.8 \cdot 10^{+18} \lor \neg \left(y \leq 7.8 \cdot 10^{+98}\right):\\ \;\;\;\;y \cdot \left(y \cdot x\right)\\ \mathbf{else}:\\ \;\;\;\;\mathsf{fma}\left(x, y \cdot y, x\right)\\ \end{array} \]

Alternatives

Alternative 1
Error	0.14%
Cost	713

\[\begin{array}{l} \mathbf{if}\;y \leq -2 \cdot 10^{+20} \lor \neg \left(y \leq 6.8 \cdot 10^{+100}\right):\\ \;\;\;\;y \cdot \left(y \cdot x\right)\\ \mathbf{else}:\\ \;\;\;\;x \cdot \left(y \cdot y + 1\right)\\ \end{array} \]

Alternative 2
Error	9.28%
Cost	580

\[\begin{array}{l} \mathbf{if}\;y \cdot y \leq 1:\\ \;\;\;\;x\\ \mathbf{else}:\\ \;\;\;\;x \cdot \left(y \cdot y\right)\\ \end{array} \]

Alternative 3
Error	1.55%
Cost	580

\[\begin{array}{l} \mathbf{if}\;y \cdot y \leq 5 \cdot 10^{-8}:\\ \;\;\;\;x\\ \mathbf{else}:\\ \;\;\;\;y \cdot \left(y \cdot x\right)\\ \end{array} \]

Alternative 4
Error	31.85%
Cost	64

\[x \]

Reproduce

herbie shell --seed 2023102 
(FPCore (x y)
  :name "Numeric.Integration.TanhSinh:everywhere from integration-0.2.1"
  :precision binary64

  :herbie-target
  (+ x (* (* x y) y))

  (* x (+ 1.0 (* y y))))