Add option for high frequency cutoff in pycbc_inspiral

[yi-fan-wang]

Adding options for high frequency cutoff in pybc_insiral. The motivation here is that: when downsampling, the antialising filter would only work well until ~0.85 Nyquist Frequency. This is not a problem when the gravitational wave frequency doesn't approach to that high, but may be an issue if so.

Also fix a typo in the pycbc_insiral: args.require_valid_checkpoint --> opt.require_valid_checkpoint, because args doesn't exist.

Tested with the default search example:

• The example search with this PR: https://hypatia.aei.mpg.de/~yiwang/pub/ssm/fupper_search/html/
• Without this PR: https://hypatia.aei.mpg.de/~yiwang/pub/ssm/search/html/
  It turns out the ranking statistic changed a little bit

[yi-fan-wang]

Hi @ahnitz @tdent I'll add more descriptions, more test runs are ongoing, but let me tag you here to let you know this.

[tdent]

Concerning the example test search, there are very small changes in the single detector triggers, see the binned triggers and binned histogram plots.

[ahnitz]

@yi-fan-wang I don't think we can approve this. In particular, the lalsim internal conversions don't preserve phase / timing in as exact a way. This has caused several problems in the past with inconsistencies between templates and injections. If you want to be able to select between the path, I think having a way to do that would be fine, but please do not change the default here.

[yi-fan-wang]

@ahnitz Got it. I'll revert to remove this. (BTW I haven't understood the purpose of this. It's cherry-picked from v23_RELEASE_BRANCH and may be useful for some runs in some specific settings, though it's never in the main branch probably for the reasons you mentioned).

[ahnitz]

@yi-fan-wang Yeah, I think addresses if you want to allow either path is probably outside the scope of this PR, so if needed consider what to do separately.

[ahnitz]

@yi-fan-wang One thing broadly here, is how does this compare to setting the f_final in the template bank? That should be similar, no? If not, you may want to check to see if there are places where the high_frequency_cutoff is not currently getting supplied that it would need to be. Naively I think setting f_final is more guaranteed to affect all the right places (think all the signal consistency tests, etc, not just MF).

[spxiwh]

@yi-fan-wang The discussion of the purpose of the lalsimulation change (and why it musn't be included on the main branch) is had on the PR that the commit links to #4530

[tdent]

@ahnitz considering your question: In the FilterBank code, it looks like there is no way to specify a high frequency cutoff directly. So if the MatchedFilterControl can take f_upper, why can't FilterBank as well? (Or vice versa, why does MatchedFilterControl advertise an option if setting it doesn't consistently affect the template?)

I'll also note that power_chisq_bins() has a kwarg high_frequency_cutoff=None ... but this isn't set by any of the other functions in vetoes that call it.

So at present it looks like neither the bank f_final nor MatchedFilterControl can influence the actual setup of power chisq ... A bit of a mess no?

[tdent]

I just added a commit to pass through a template end_frequency value to the power chisq bins calculation.
However, it seems like nontrivial extra work is needed in bank.py to make the template end_frequency settable in a nicely controlled way analogous to f_lower (see ensure_standard_filter_columns()). Well, the actual code may be straightforward but several extra lines will be needed.
@ahnitz would you agree with that?