A posteriori error estimators via bubble functions