Smoothed aggregation constructs the smoothed prolongator $P = (I - \omega D^{-1} A) T$ where $T$ is the tentative prolongator and $D$ is the diagonal of $A$.
Eigenvalue estimates (based on a few CG or GMRES iterations) are computed to choose $\omega$ so that this is a stable smoothing operation.
If Chebyshev with Jacobi (diagonal) preconditioning is used for smoothing, then the eigenvalue estimates can be reused.
This option is only used when the smoother uses Jacobi, and should be turned off if a different PCJacobiType is used.
It became default in PETSc 3.17.