This is a follow up to alog 47604, mostly this gif.

LHO is currently locking the SRCL degree of freedom with 50 cts of offset.  Detuning in the SRCL dof causes an optical spring to appear in DARM. (see e.g. alog 52981, host to this plot).
From the many DARM TFs, 50 cts in the SRCL loop corresponds to about 0.5 degrees of detuning from perfect resonance.

Why should the SRCL loop lock in a detuned state?  I am testing the idea that higher order modes existing in the SRC spoils the SRCL error signal.
The SR3 heater alters the radius of curvature of the SR3, changing the mode content in the SRC.
The gif above shows the effect of a SR3 heater off-on-off test has on the DARM plant.  The DARM plant TF changes from anti-spring to spring to anti-spring again.
This suggests that the mode content of the SRC has a strong effect on the SRCL error signal.

Using Finesse, I have modeled the LHO higher order mode content in each cavity while changing the SR3 radius of curvature (via "lock dragging"). 
If I change the SR3 RoC by ~10 mm, we start to see a number of effects on the model:
1) The 02 and 20 HOM content in the SRC starts to become significantly high. (Plot 1, [here "omc" means incident on omc, and "as" means OMC DCPDs])
2) The SRCL error signal becomes distorted on the upper side (Plot 2)
3) The SRCL error signal zero-crossing moves away from 90 degrees (Plot 3)
4) SRM tuning moves away from 90 degrees to keep the IFO locked (Plot 4)
5) As SRCL becomes detuned, we see the optical spring appear in the DARM TF (Plot 5)

At the moment, this model cannot totally explain the DARM optical spring.  This Finesse model is fairly optimized, with no astigmatism in the arms or mode mismatch from IMC to IFO.  But if we take the DARM plant MCMC fits from the gif to be accurate enough, the SR3 heater caused the optical spring frequency to flip from around -3 Hz to 3 Hz and back again.  This model is good to about that order of magnitude.