The (x1,x3) contour plot of RP33 is wrong.
the contour plot of the g function is:
Let us focus on the (x1,x3) projection:
Below is my attempt to reproduce this with OpenTURNS:
The difference is that the function seems to depend only on x3 in the center of the domain in my graphics, while the reference graphics shows that the function depends only on x1 in the center of the domain.
In order to get a third version, I used Mathematica with x2=0:
I claim that Mathematica is correct and here is why. Setting x2=0, we get:
g = min(-x1 - x3 + 3 * sqrt(3), -x3 + 3)
g = -x3 + 3 if x1 < 3 * sqrt(3) - 3 = 2.96
g = -x1 - x3 + 3 * sqrt(3) otherwise.
Hence, in the center of the domain, the function only depends on x3.