Object clustering is a fundamental task in many data analysis and pattern understanding applications by providing insights into detecting the underlying structures of a large collection of samples. In this paper, we present our work on a novel spectral clustering algorithm that partitions a collection of objects using the spectrum of adistance matrix. If the nodes in a metric space can be associated with a well defined distance, the distance matrix is almost negative definite, implying that the eigenvector for the smallest eigenvalues of this matrix can be used as an approximation of the solution to a quadratic form partition problem. It is proved that this smallest eigenvalue is equivalent to the second largest singular value. Therefore Lanczos iterative algorithm can be applied to findingthe eigenvalues efficiently. We adapted this algorithm to the distributed network community detection problem using a decentralized multi-agent framework, and tested the effectiveness of the proposed approach with simulations.