In this study, I started the optimization with a randomly generated wing to see the effectiveness of the optimization. The initial wing is shown below.
This figure shows the optimized results from a random initial starting geometry. The optimization is performed on the L2 grid. We can see that the performance of the initial design is extremely poor. This is no surprise, since the airfoil shapes are unlike anything one would design: they exhibit oscillations and sharp edges, resulting in wildly varying Cp distribution. In spite of these wild shapes, the optimizer is able to smooth out the airfoils and achieve a shock-free wing similar to the original single-point design.