In radial compressors or blowers, a low solidity circular cascade diffuser (LSD) is one of the effective devices to improve the pressure recovery at design flow rate while guaranteeing a wide operating range. The improvement is mainly attributed to the so called secondary flow effect, which reduces the flow separation on the LSD blade at small flow rates. However, it is very difficult to find out the effective shape of the blade in order to promote this secondary flow effect. In this paper, a multipoint and multi-objective optimization technique is applied to design the LSD blade of a centrifugal blower. The optimization method has been developed at the von Karman Institute for Fluid Dynamics (VKI), which makes use of an evolutionary algorithm, a metamodel as a rapid exploration tool, and a high fidelity 3D Navier-Stokes solver. The optimization is aiming at improving the static pressure coefficient at design point and at low flow rate condition while constraining the slope of the lift coefficient curve. Seven detailed design parameters describing the shape and position of the LSD vane were introduced, e.g. the radial spacing between impeller exit and the LSD leading edge, the radial chord length and the mean camber angle distribution of the LSD blade with five control points. Moreover, a small tip clearance of the LSD blade was applied in order to activate and to stabilize the secondary flow effect at small flow rate condition. The optimized LSD blade has an extended operating range of 114 % towards smaller flow rate as compared to the baseline design without deteriorating the diffuser pressure recovery at design point. The diffuser pressure rise and operating flow range of the optimized LSD blade are experimentally verified. It is found that the optimized LSD blade shows good improvement of the blade loading in the whole operating range, while at small flow rate the flow separation on the LSD blade has been successfully suppressed by the secondary flow effect. This is fully corresponding to the CFD predictions and demonstrates the effectiveness of the optimization methodology, by limiting the experimental testing to only two geometries.