1: \begin{abstract}

2: 		The spectral bundle method proposed by Helmberg and Rendl~\cite{helmberg2000spectral} is well established for solving large scale semidefinite programs (SDP) thanks to its low per iteration computational complexity and strong practical performance.

3: 		In this paper, we revisit this classic method showing it achieves sublinear convergence rates in terms of both primal and dual SDPs under merely strong duality. Prior to this work, only limited dual guarantees were known.

4: 		Moreover, we develop a novel variant, called the block spectral bundle method (Block-Spec), which not only enjoys the same convergence rate and low per iteration complexity, but also speeds up to linear convergence when the SDP admits strict complementarity. Numerically, we demonstrate the effectiveness of both methods, confirming our theoretical findings that the block spectral bundle method can substantially speed up convergence. 

5: %		In this paper, we show that the HR spectral bundle method (HR-Spec) originates from Helmberg and Rendl~\cite{helmberg2000spectral},

6: %		achieves sublinear rates in terms of both primal 

7: %		and dual convergence under merely strong duality. Such method enjoys low per iteration computational complexity and 

8: %		is well suited for solving large scale semidefinite programs (SDP).

9: %		Moreover, we develop a novel variant, called 

10: %		block spectral bundle method (Block-Spec), which not only enjoys the same convergence rate and low per iteration complexity, but also speed 

11: %		up to linear convergence when the SDP admits strict complementarity. We demonstrate the effectiveness of both methods and confirm our 

12: %		theoretical findings in our numerical experiments. 

13: 	\end{abstract}