Richard showed me that there is a liner coupling from any of the electrodes to the OPLEV even when there's no bias anywhere, and the bias-less coupling is on the same order as the coupling with bias.
I started flipping the bias of the common ("DC") electrode while keeping all four quadrants at zero count output, and indeed there's a linear coupling, and the effect is not subtle.
In the attached, I changed the DC bias from 125000 to 0 to -125000, then to 0 and finally back to +125000. The ETM points up by 0.3urad in the first step, then further up 0.5urad in the second step (instead of going back to the original position). It's 0.8 urad peak to peak!
OTOH, when I set the four quadrants such that top two outputs -50900 counts and the bottom +50900 to tilt the optic legitimately while the DC bias is kept at 125000 counts, I can only move it by -0.5 urad peak to peak (second attachment).
This could still be electronics offset, so I looked at the ESD readback. The readback is not calibrated, but four quadrants always read from -75 to -200 counts when I'm outputing zero, the 300volt-ish thing reads 30000-ish counts, so the voltage offset for the quadrants should be 2V-ish at most. That's not large enough, I think.
I disconnected the ESD drive cables between the current limiting resistor and the flange. Put large drive signals on the cables and did not see a response in the oplev. So it is not a direct coupling in the rack or cables up to the chamber.
At LASTI we saw static charges on the test mass high enough to mimic a bias voltage of 125 V (very large!). John Miller put together a procedure and MATLAB script for measuring what this effective bias was. So the problems you are seeing with a non-symmetric response are within the limits of what has been observed before. See the LASTI ilog http://emvogil-3.mit.edu/ilog/pub/ilog.cgi?group=lasti&task=view&date_to_view=12/18/2009&anchor_to_scroll_to=2009:12:18:16:36:37-brett u: reader p: readonly This script happens to be on the svn at /ligo/svncommon/SusSVN/sus/trunk/QUAD/Common/MatlabTools/ChargeScript It will need some modification to work at the sites, since it was written to work in the old LASTI file structure. Basically though, it automatically runs DTT from MATLAB, and measures the response at 11 Hz with a range of bias voltages. Note, Rai has put together a gizmo to discharge the test masses with ionized nitrogen. Not sure of its status. Venting to some torr might also help, not sure.
The discharging rig Brett is referring to is described in LIGO T1100332. It currently resides at MIT under clean conditions. It and the associated electronics can be shipped to LHO. It needs a 2 3/4 inch conflat gatevalve attached to one of the ten inch blank flanges on the BSC chamber at the charged mirror height. The gas injected into the deionizer needs to be clean nitrogen such as the boil off from liquid nitrogen passed through another liquid nitrogen trap and warmed to room temperature before injection into the BSC. RW