Attempts to find correlations between these glitches and those in other channels did not produce any clear results for some witnesses.

I then tried to calculate the time series of the rate of omicron triggers (trigger clusters at 0.1 sec) and do a brute force correlation with all the trend _mean and _rms channels, excluding those excluded by BruCo: /data/dev/detchar/online/bruco/share/virgo_excluded_channels_Hrec_hoft.txt . The same analysis could have been done with some BRMS channels or the BNS range, but I preferred focussing on the glitches. The resulting series shows a step, without particular variations, during the lock between October 14 and 15.

(As one might expect) A very large number of channels were found to be correlated with the glitch rate series. I have attached a text file with the most correlated ones and some plots. I leave it to the experts to decide if someone can be informative or if the correlated channels are obvious reactions to the excess energy in the strain channel.

As alluded by Francesco in his comment, UPV running on a 5-hour GPS range between 0030 and 0530 UTC did not find any correlation between the h(t) glitches and the glitches on the standard DetChar channels for which Omicron triggers are computed online. With the help of Florent I made some crosschecks to make sure UPV was processing those triggers correctly and we found no problem.

The same conclusion is true for BruCo: no coherence worth mentioning were found using the sets of channels used as inputs by the daily BruCo runs reported in VIM.

These results are helpful, unfortunately they turn out not to be physically useful correlations.

The correlation with the 50Hz FeedForward channel is expected, as these channels are based on DARM, so glitches on DARM will as a consequence cause glitches in these FF channels

The ALS NEB Corr correlation shouldn't be a symptom, as the ALS system is turned-off once the interferometer is locked, it serves only during the lock acquisition.

Figure 1. The phase step looked very interesting at first glance, but looking at the figure we have different steps separated by 2*pi at each lock, which are all equivalent. So it is a coincidence than in this shorter time period the good BNS range and bad BNS range locks correspond to two distinc phase choice. It is not the case when looking at a week of data. But this shows that the issue of the phase jumps should be fixed also on B4 and the RFC, as it was done recently on B1s, to avoid these 2*pi phase jumps which are distracting and confusing.

This afternoon I made a test to understand the possible relation between this excess noise and DAC noise from the IM suspensions. In the Figure the results, where:

• red trace: 15/10 22:00 UTC: no excess noise, reference curve;
• purple trace: 15/10 02:00 UTC: excess 1/f^2.5 noise;
• black trace: 16/10 16:30 UTC: both WI and NI relays left closed in HighPower;
• green trace: 16/10 16:37 UTC: NI relays open, WI relays still in HighPower;
• blue trace: 16/10 16:49 UTC: both WI and NI open in standard condition.

None of the configurations seem able to reproduce the noise we observed, as they are all quite noisier than what we observed the other night. The only configuration which is missing is the one with the NI relays in HighPower and the WI ones open; in principle it should be more or less identical to the one explored, but given that part of the WI actuators are off altogether because of the faulty magnet I am not sure how different this could be in reality.

Looking a bit more into it, if one rescales the black curve (all coils in HighPower) by 1/8, this fits well the observed noise (Figure 1); so this is still compatible with one single arms coil not being switched to "open" for some reason. As the commands sent are only one per tower, with a bytecode comprising all coils at once, it looks more likely that the issue is not software, at least not from VPM to the switchbox.

I looked at two transitions in time domain, and they are reported in Figure 2 (downsampled at 10 Hz), where the purple trace is a good transition (16/10 10:37 UTC) and the blue one is a bad transition (15/10 00:09:35 UTC). It has to be noted that anything one can see implies that there is a little DC change still happening during the transition, which is something we already reduced some time ago; so, it's not guarantee that not seeing one is and indication of something.

In any case, six coils are easily dismissed from being suspects (both DLs, ULs and URs); the two DRs are left, but if I am not wrong WI_DR is the disconnected magnet, so maybe it is normal that no change is observed. NI_DR also shows no step in both occasions, so it's unclear what conclusion to drive.

This correlation with WE correction is very interesting.

Figure 1. Looking at spectra of the corrections, the WE correction have 10 times more noise below 0.1Hz than the NE corrections, both during this glitchy time, and at normal times. It is a noise that is added in the DSP, and is not present on the LSC signals as received by the DSP. It needs to be investigated what signal is being added. That signal added to WE is making the WE correction RMS clearly higher when looking in time domain compared to the NE correction.