Typo/bug in ar.yw.default, ar.ols, ar.burg.default

Martin Maechler 
    on Thu, 12 Mar 2026 12:12:25 +0100 writes:

Simone Giannerini 
    on Mon, 9 Mar 2026 12:43:07 +0100 writes:

>> Dear all,
    >> I think there is a (small) bug in

    >> ar.yw.default   (line 142 of the source file ar.R)
    >> ar.ols          (line 429 of the source file ar.R)
    >> ar.burg.default (line 570 of the source file ar.R)

    >> maic <- min(aic)

    >> should be

    >> maic <- min(xaic)

    >> Note that, apparently, this typo/bug does little harm since the
    >> subsequent line of code

    >> xaic <- setNames(if(is.finite(maic)) xaic - min(xaic) else
    >>                  ifelse(xaic == maic, 0, Inf), 0L:order.max)

    >> does not use maic but min(xaic) directly. 

but that is _NOT_ true (as I only see now, when looking more closely):
`maic` is even used *twice* in the above (2nd) construction of `xaic`.
Note that the "bad" code,  maic <- min(aic)   would typically
set maic to 1 (aic = TRUE is the default) or then 0 (for
aic=FALSE), hence always finite.

Consequently the case where some 'xaic' where *not* finite (but
NA, NaN, or +/-Inf) has never been dealt with "properly" till
now (and I think we not know if it *is* "good" now, as we
probably never had test code for such boundary cases ...)

I still think the proposed change should happen, as  min(aic)
never made any sense.
Can anyone think of / create artificial time series data where
the changed code would make a difference?

Martin

    >> does not use maic but min(xaic) directly. In any case, after fixing,
    >> one could replace min(xaic) with maic to avoid recomputing min(xaic).
    >> In absence of comments I will file a bug report.

    > I'm looking into this, thank you, Simone.
    > This is not really a bug, e.g., in the sense of
    > https://www.R-project.org/bugs.html ,
    > as you mention yourself that the code works (flawlessly).

    > Still, of course,  (if you are right which I assume currently)
    > it is a code infelicity we will be happy to fix.

    > Best,
    > Martin


    >> Kind regards
    >> Simone

    >> Tested against R-devel 2026-03-06

    >> ___________________________________________________

    >> Simone Giannerini
    >> Dipartimento di Scienze Economiche e Statistiche
    >> Universita' di Udine
    >> Via Tomadini 30/A - 33100 Udine,  ITALY
    >> Tel: +39 0432 249577
    >> https://simonegiannerini.net/

    > ______________________________________________
    > R-devel at r-project.org mailing list
    > https://stat.ethz.ch/mailman/listinfo/r-devel