As observed by Edwige, a general increase of the floor is the biggest effect in the sensitivity(fig 2).
The noise budget is able to explain very well the origin of the problem: comparing the two conditions (dF/F=-0.03, fig 3; dF/F=0.034, fig 4), we see that the contribution of B5 shot noise is much larger when the finesse asymmetry is positive. The contribution to the horizon is about 1 Mpc both in the actual sensitivity and in the predicted noise.
Probably the coupling of common noises (Common Mode Rejection Factor) is lower when the finesse asymmetry is negative because there is a compensation effect between the contribution from finesse aymmetry and the one from asymmetry of losses.
is
F_asym = ABS( (dF/F) 1/(1+i f/f_rec)
+ (dzeta/zeta) f_p/f_rec (1+i f/f_p)/(1+i f/f_p) )
does predict a compensation in low/medium band on the detector bandwidth.
In this formula,
- F is the finesse, dF the finesse asymmetry;
- f_p is the arm cavity pole (500 Hz) and f_rec the recycling cavity pole (8 Hz)
- zeta is the arm amplitude reflectivity at resonance, and dzeta its asymmetry, so that dzeta/zeta = dR/4 = (+/-) sqrt((1-C)/2), where the sign depends where the losses are higher, and the contrast defect the one for the TEM00 after OMC.