1: \begin{abstract}

2: We develop a new interior-point method for symmetric-cone optimization, a

3: common generalization of linear, second-order-cone, and semidefinite

4: programming.  Our key idea is updating iterates with a \emph{geodesic} of the

5: \emph{cone} instead of the \emph{kernel} of the \emph{linear} constraints.

6: This approach yields a primal-dual-symmetric, scale-invariant, and

7: line-search-free algorithm that uses just half the variables of a standard

8: primal-dual method.  With elementary arguments, we establish polynomial-time

9: convergence  matching the standard $\bigO(\sqrt{n})$ bound.  Finally, we

10: prove global convergence of a long-step variant and compare the approaches

11: computationally. For linear programming, our algorithms reduce to

12: central-path tracking in the log domain.

13: \end{abstract}

14: