I always hear that monolithic coupling is the only way to go, as partitioned coupling introduces error. Is this correct?
In general yes, but it depends on the application whether this additional error is problematic. For some applications, e.g. FSI with a compressible fluid, the error can be controlled by the timestep size. For other applications, e.g. FSI with an incompressible fluid and strong coupling, the error can lead to instabilities (the so-called Added-Mass Effects), which in turn can be controlled by the implicit coupling algorithms preCICE offers.
Moved from our FAQ on precice.org.