1: \begin{abstract}

2:   %% Text of abstract

3:   Efficient approximation of geodesics is crucial for practical algorithms on manifolds.

4:   Here we introduce a class of retractions on submanifolds,

5:   induced by a foliation of the ambient manifold.

6:   They match the projective retraction to the third order

7:   and thus match the exponential map to the second order.

8:   In particular, we show that Newton retraction (NR) is always stabler

9:   than the popular approach known as oblique projection or orthographic retraction:

10:   per Kantorovich-type convergence theorems,

11:   the superlinear convergence regions of NR include those of the latter.

12:   We also show that NR always has a lower computational cost.

13:   The preferable properties of NR are useful for

14:   optimization, sampling, and many other statistical problems on manifolds.

15: \end{abstract}

16: