Summarizing techniques are widely used in the reasoning of unbounded data structures. These techniques prohibit strong update unless certain restricted safety conditions are satisfied. We find that by setting and enforcing the analysis boundaries to a limited scope of program identifiers, called targets in this paper, more cases of strong update can be shown sound, not with regard to the entire heap, but with regard to the targets. We have implemented the analysis for inferring numeric properties in Java programs. The experimental results show a tangible precision enhancement compared with classical approaches while preserving high scalability.