Erf function in LaTeX

Based on this answer.

\documentclass{standalone}
\usepackage{tikz}
\makeatletter
\pgfmathdeclarefunction{erf}{1}{%
  \begingroup
    \pgfmathparse{#1 > 0 ? 1 : -1}%
    \edef\sign{\pgfmathresult}%
    \pgfmathparse{abs(#1)}%
    \edef\x{\pgfmathresult}%
    \pgfmathparse{1/(1+0.3275911*\x)}%
    \edef\t{\pgfmathresult}%
    \pgfmathparse{%
      1 - (((((1.061405429*\t -1.453152027)*\t) + 1.421413741)*\t 
      -0.284496736)*\t + 0.254829592)*\t*exp(-(\x*\x))}%
    \edef\y{\pgfmathresult}%
    \pgfmathparse{(\sign)*\y}%
    \pgfmath@smuggleone\pgfmathresult%
  \endgroup
}
\makeatother
\begin{document}
\begin{tikzpicture}[yscale = 3]
  \draw[very thick,->] (-5,0) -- node[at end,below] {$x$}(5,0);
  \draw[very thick,->] (0,-1) -- node[below left] {$0$} node[at end,
  left] {$erf(x)$} (0,1);
  \draw[red,thick] plot[domain=-5:5,samples=200] (\x,{erf(\x)});
\end{tikzpicture}
\end{document}

Using the same approximation idea as cjorssen (I tried the Taylor series as Qrrbrbirlbel suggested but it's pretty hopeless to get a decent approximation this way) I rewrote the function without using low-level PGF. Because we have so many 2D plots here already, I'll just use my 3D plot that I had already.

\documentclass{standalone}
\usepackage{pgfplots}
\usepackage{tikz}
\pgfplotsset{
colormap={bluewhite}{ color(0cm)=(rgb:red,18;green,64;blue,171); color(1cm)=(white)}
}
\begin{document}
\begin{tikzpicture}[
    declare function={erf(\x)=%
      (1+(e^(-(\x*\x))*(-265.057+abs(\x)*(-135.065+abs(\x)%
      *(-59.646+(-6.84727-0.777889*abs(\x))*abs(\x)))))%
      /(3.05259+abs(\x))^5)*(\x>0?1:-1);},
    declare function={erf2(\x,\y)=erf(\x)+erf(\y);}
]
\begin{axis}[
    small,
    colormap name=bluewhite,
    width=\textwidth,
    enlargelimits=false,
    grid=major,
    domain=-3:3,
    y domain=-3:3,
    samples=33,
    unit vector ratio*=1 1 1,
    view={20}{20},
    colorbar,
    colorbar style={
        at={(1,-.15)},
        anchor=south west,
        height=0.25*\pgfkeysvalueof{/pgfplots/parent axis height},
    }
]
\addplot3 [surf,shader=faceted] {erf2(x,y)};
\end{axis}
\end{tikzpicture}
\end{document}

The approximation has a maximum error of 1.5·10^-7 (source).

Thanks to Jake for spotting and fixing the wrong syntax I first had in this code.

For precise values, I recommend externalizing the calculation, here gnuplot is used.

Code (needs --shell-escape enabled)

\documentclass{article}
\usepackage{amsmath,pgfmath,pgffor}
\makeatletter
\def\qrr@split@result#1 #2\@qrr@split@result{\edef\erfInput{#1}\edef\erfResult{#2}}
\newcommand*{\gnuplotErf}[2][\jobname.eval]{%
    \immediate\write18{gnuplot -e "set print '#1'; print #2, erf(#2);"}%
    \everyeof{\noexpand}
    \edef\qrr@temp{\@@input #1 }%
    \expandafter\qrr@split@result\qrr@temp\@qrr@split@result
}
\makeatother
\DeclareMathOperator{\erf}{erf}
\begin{document}
\foreach \x in {-50,...,50}{%
\pgfmathparse{\x/50}%
\gnuplotErf{\x/50.}%
$ x = \pgfmathresult = \erfInput, \erf(x) = \erfResult$\par
}
\end{document}

Output