[Libreoffice-bugs] [Bug 117049] Severe bug in LibreOffice (Calc Speadsheet)

[bugzilla-daemon at bugs.documentfoundation.org]

https://bugs.documentfoundation.org/show_bug.cgi?id=117049

--- Comment #14 from Mike Kaganski <mikekaganski at hotmail.com> ---
I will ask one more time, trying to rephrase my question, since the rude
behaviour of OP shows not only that s/he is not well-mannered person, but also
that s/he cannot understand the idea behind the prior questions, in the hope
that OP would choose to return to a constructive dialogue.

Your idea is based on assumption that the initial value of B1 in the iterative
calculation must be zero. Only then the single iteration of formula "=B1+A1"
might bring the 100.

But the actual value in the B1 at the moment of the first calculation is not 0,
but (a formula) "=B1+A1", which is simply *uncertainty* initially, since it's a
self-referencing formula. There's *no* number in the universe that may
represent that uncertainty; and any initial number that we might choose to use
as its initial value is just a guess, which will be improved iteratively. If we
happen to blindly guess a value that would happen to be the limit of the
iterative calculations, then the very first iteration would give a delta that
is less than our minimum change, and thus, the iterative calculation will have
the minimal length.

Just to make things clear: the value in the B2 in the course of iterative
calculations is not a single static value, but a series of values, which are
expected to converge to prescribed maximum error in the given number of
iterations.

Let's put the series in a hypothetical B1 here in some abstract series for some
unspecified formulas:

> i0  i1   i2    i3     i4 ...
>  0 -10 -100 -1000 -10000 ...
>  1 1.1 1.11 1.111 1.1111 ...

Here you may see that some formula might give a diverging results if the
initial guess is 0, but converging result when the initial guess is 1. And if
the initial guess would magically happen to be 1.1111111... from the very
beginning, then the following iterations would not change the value in that B1,
i.e., the initial guess would be the final value.

Of course, it's unrealistic to expect that we would be able to guess the
correct final value in that cell each time; but *generally*, the initial guess
does affect the number of iterations before the delta becomes small enough; and
in some cases, it even affects the very possibility to come to a converging
iterations. And I repeat once more: the very idea of the iterative calculations
is to make sure we come to a converging result series, not to misuse the
feature in obscure ways.

And again: the initial value in the cell with iterative calculations is not
something that necessarily must be decided once for all; it's just an initial
guess of the final result, and the quality of the guess affects the quality
(time to get, and ability to get close enough to, the theoretical limit of the
infinite calculation series)!

There exist different approaches to guessing the initial value that starts the
series; of course, *one of them* (the dumbest one) is just to take it to be
zero. But this dumbest initial guess is by no means the best out there in
existence.

===================================================
That is why I question the validity of your assumption that it must be like
that. And that is why I ask you to provide an evidence in form of documentation
that states that *in the beginning of the series of iterative self-referencing
calculations, the initial guess of the value in that cell must be equal to
zero*!
===================================================

I set the status to NEEDINFO once again. The status here is not for you, Carlo
Sarti; it's for us, who decide what to do next with the issue: if we need to
proceed and declare it confirmed (and thus a candidate to be fixed eventually),
or if we need more details on this; or if an issue should be closed. We ask you
the necessary information, because it's you who is interested in "fixing" this,
so it's in your interests to provide as much information as it's required.
Doing this does not mean any hostile behaviour towards you; if someone wanted,
one could just close this, and not ask you in the hope that you could make the
issue clear (and so, we tell you that we hope that you can convince us!).
Please only change the status back to UNCONFIRMED when you have provided the
requested evidence in form of references to documentation. Thank you.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/libreoffice-bugs/attachments/20180620/107adf04/attachment-0001.html>