This paper proposes gridless sparse direction-of-arrival (DOA) refinement using gradient-based optimization. The objective function minimizes the fit between the sample covariance matrix (SCM) and a reconstructed covariance matrix. The latter is constrained to contain only a few atoms. but otherwise maximally matches the SCM. https://tonerparts.shop/product-category/maintenance-kit-fuser/