Some hints towards potential partial fixes :
- choosing an odd sample size (eg
201 instead of 200) seems to help a bit
- pgfPlots seems to do a bit better for the computations
It also does much better at almost everything else for plotting than raw TikZ.
If the function you want to plot is too complicated, be it :
- hard to compute
- need lots of sample points
- need high float precision
- requires irregular fine-tuned spacing of sample points
then it may be better to precompute the data outside of TeX, eg with python or (Mat-)Scilab/Octave or gnuPlot, and input the data as an external file to pgfPlots
Otherwise, LuaLaTeX is reported to be useful for computations, but examples are a bit scarce to my taste.
The output

The code
\documentclass[12pt,tikz]{standalone}
\usepackage{pgfplots}
\pgfplotsset{compat=1.14}
\begin{document}
\begin{tikzpicture}
\begin{axis}
[
height=10cm,
unit vector ratio = 1 1,
domain=-1:1,samples=41,
axis lines = center,
enlarge y limits=.2,
]
% your function is hard to plot near its vertical tangents
\addplot [line width=1mm,blue,samples=201] {sqrt(1-x*x)} node [pos=.5, anchor= south west] {201 sample points};
\addplot [thick,green!80!black,] {sqrt(1-x*x)} node [pos=.5, anchor= south east] {41 sample points};
% much easier to plot
\addplot [thick,red,domain=-180:0] ({cos(x)},{sin(x)}) node [pos=.5, anchor= north west] {41 sample points};
\end{axis}
\end{tikzpicture}
\end{document}
xslightly greater than 1. The manual explicitly warns about this. Just use\draw (1,0) arc (0:180:1). (manual can be found here: http://ctan.org/pkg/pgf . I can't be bothered to find the exact citation.) – Huang_d Jun 19 '17 at 10:37plot, have a look at the dedicated libraries such asdatavisualizationorpgfplots. – Huang_d Jun 19 '17 at 10:38