While gene or genome duplications have provided raw genetic materials for evolutionary innovations, these events have also generated massive duplicate genes, resulting in a tremendous increase to the genetic robustness of organism. Duplicate compensation indicate functional redundancies generated by gene duplications, which are widespread in all known genomes. However, the fitness trade-offs of their mutational compensation (genetic robustness) and their role in evolutionary innovation remains largely obscure. In this paper, we discuss how we can utilize the mathematical modeling approach to predicting under which condition duplicate compensation may occur. After a critical review for the models about expression dosage, compensation, and long-term survival of duplicate genes, we highlight the importance to distinguish between Function (F)-triggered and Expression (E)-triggered mechanism of duplicate compensation. Moreover, we address three fundamental questions: (i) Why a backup duplicate can be effectively activated by any silence mutation of the dominant duplicate, but hardly by any coding mutation resulting in impaired protein function? (ii) Why a dispensable duplicate gene, i.e., knockout leads to virtually no phenotype, still remains a great deal of selective constraints in the coding region? And (iii) under which condition expression subfunctionalization between duplicates is reversible (dosage-sharing) or irreversible (long-term survival)?
