We investigated further the 144 kHz oscillation. The line is not visible in the beat note power spectrum (between N and S lasers). However, when we block the beam at the REFL PD, there is a -120 dBV/rtHz peak at 144 kHz. Because the full range of the PDH signal is about 0.5 V, but the line is about 1000x lower (~40 uV), we're not too worried about it for now.
Next, we checked out the cavity lock. The gain was set too high, such that the servo frequently locked onto a weakly resonant mode, while in the primary mode the laser maintained lock but the PDH signal saturated its range. When we lowered the gain, the 41 Hz cantilever resonance appeared not only in the control signal but in the PDH error signal, indicating that the loop gain is insufficient at low frequency for the laser to track the cavity. We need to change the loop shape to give us enough phase margin to increase the gain at low frequency.
With the gain knob set to 5.08, the cavity maintained a stable lock despite insufficient low frequency gain (at least near 40 Hz). While locked, both the PDH signal and the cavity transmission had a 41 Hz oscillation. This indicates that there is a DC offset in our lock setpoint. A fluctuation at frequency f_0 in the error signal should cause an f_0^2 fluctuation in the transmission signal on resonance; off resonance, the leading order term is instead f_0.