curve doesn't follow the point

Question

I created this graph using pgfplot package. Here it is the code I used:

\begin{figure}[htp]
\centering
\begin{tikzpicture}
    \begin{semilogxaxis}[xmin=1e-06,xmax=1e-01,xlabel=packet loss,ylabel=valore atteso,
                         grid=both,
                         legend pos = north west]
    \addplot [smooth,thick,blue]
        file {./MATLAB/grafici/valore_atteso/pure_network_coding/var_pkts_loss_probability/valore_atteso_20nodes_rlnc.txt};
\end{semilogxaxis}
\end{tikzpicture}
\end{figure}

here it is the result I got:

but unfortunately this is not the right graph. This because the file contains those values:

1e-06 1.0022
1e-05 1.0095
0.0001 1.0299
0.001 1.135
0.01 2.1753
0.1 46.922

The curve around 1e-3 and 1e-2 goes down. Why?

Answer 1

smooth option enables a path morphing mechanism such that the sharp corners are eliminated via some Bézier curve construction. As the name implies the curve is morphed into a smoother one. The only relation with the original curve is the given fixed points that define the curvature of the path while passing through those points.

\documentclass[tikz]{standalone}
\begin{document}
\begin{tikzpicture}
\begin{axis}
\addplot+[] coordinates {(0,0) (1,1) (2,2) (3,0) (3.3,0.5)};
\addplot+[smooth,ultra thick] coordinates {(0,0) (1,1) (2,2) (3,0) (3.3,0.5)};
\end{axis}
\end{tikzpicture}
\end{document}

In drawings and schematics this is no big deal and can be used safely. However as you have found out, scientific plots should be given without any modification. Especially in the logarithmic plot cases the smallest deviations mean a lot in the linear scale.