1: \begin{abstract}

2: Many classical and modern machine learning algorithms

3: % , such as principal component analysis, canonical correlation analysis and dictionary learning, all 

4: require solving optimization tasks under orthogonal constraints. 

5: Solving these tasks often require calculating retraction-based gradient descent updates on the corresponding Riemannian manifold, which can be computationally expensive.

6: Recently \cite{ablin2022fast} proposed an infeasible retraction-free algorithm, which is significantly more efficient.

7: In this paper, we study the decentralized non-convex optimization task over a network of agents on the Stiefel manifold with retraction-free updates.

8: We propose \textbf{D}ecentralized \textbf{R}etraction-\textbf{F}ree \textbf{G}radient \textbf{T}racking (DRFGT) algorithm, and show that DRFGT exhibits ergodic $\mathcal{O}(1/K)$ convergence rate, the same rate of convergence as the centralized, retraction-based methods.

9: We also provide numerical experiments demonstrating that DRFGT performs on par with the state-of-the-art retraction based methods with substantially reduced computational overhead.

10: \end{abstract}

11: