I tried to estimate the coupling from seismic to displacement noise. With the common mode rejection taken into account, the coupling from vertical acceleration to differential length between the two cavities is about 6x10^-12.

• With random errors add in the position (+/- 1mm) of the support position, I used COMSOL to find out the coupling from acceleration to strain noise to a single cavity. Note that I did not use Gaussian distribution because sometimes the support position will be place away from the cavity. so I used random error with confined limit to +/- 1mm.
• Then I histogram it. The result is shown in the figure below.

• Take some common mode rejection into account. Since most of the time, the coupling will fall between 0.98 x10^-10 to 1.04 x10^-10  [s^2/m].  We can assume an upperbound that one cavity has the coupling of 1.04x10^-10 and another has 0.98x10^-10. Thus, the effective coupling for the differential strain is (1.04-0.98)x10^-10 = 6x10^-12. 
• I have to admit that the result is very weird. With the ideal support position, the coupling for a single cavity is also about 6x10^-12. I did not expect that by changing the positions slightly, the coupling for each cavity can be worse by 200 times. I suspect that COMSOL might add some errors as well, see psl:1056. The cavity does not bend down symmetrically, even with the perfectly symmetric support position, but I did not think it will be this huge.