Fig. 6. Results for the three-dimensional flow around a cylinder for the IGA with third order velocity and second order pressure and
for the finite element pairs Q2/Q1, Q3/Q2, P2/P1, and P3/P2.
flow around a cylinder. For brevity, results for a parametrization of type Ħ2 will not be included in the
presentation, since Ħ2 is constructed very similarly to Ħ6 and the latter provided more accurate results for
the two-dimensional example. For computing the drag and lift coefficients with the volume integrals, one
has to use vector-valued test functions that take the value 1 in some component at the cylinder and 0 at all
other boundaries. In the simulations, the value 1 was also prescribed at the intersection of the cylinder and
the wall. The obtained results are compared with results from [15] for finite element simulations with the
Taylor.Hood pairs of spaces Pk/Pk.1 and Qk/Qk.1, k ¸ {2, 3}.
Reference values for the drag coefficient, the lift coefficient, and the pressure difference are provided in [15].
Relative errors to these values are presented in Fig. 6. Concerning the different parametrizations used in the
IGA, Ħ6 performed best. All finite element methods compute drag coefficients of a similar accuracy as the
best IGA approach. Concerning the lift coefficient, only Q2/Q1 and Q3/Q2 give similarly accurate results
as IGA with Ħ6. And with respect to the pressure difference, the higher order finite element methods P3/P2
and Q3/Q2 are clearly more accurate than their lower order counterparts. But only the result obtained with
Q3/Q2 is of a similar order of accuracy as the result of the IGA with Ħ6.
Since the IGA and the finite element methods are implemented in different codes, a comparison of
computing times is not meaningful. It should only be mentioned that the computing times for the IGA were
reasonable.