theoretical foundations of gnuplot csplines

Which is the theoretical method employed in the csplines smoothing algorithm available in gnuplot? Is it a piecewise least square fit with polynomials? Like a Savitzky–Golay smoothing filter or, as the name suggests, some modification of bsplines? Also: in gnuplot version 4.4 I observed an oscillatory behaviour in 开发者_如何转开发the interpolating curve if I employ it for double-logarithmic plots. Is that to be expected?

csplines stands for a cubic spline algorithm. You can see this by typing help csplines in gnuplot, or by looking at the source code (see the functions gen_cubic_spline and intp_cubic_spline in contour.c).

About the oscillations in cubic splines (they are to be expected in some circumstances), see this reference, or this if you can access it.