Graph documentation [was: Re: Meaning of GF settings]

[Robert C. Helling]

On Thu, 10 Jan 2013, Dirk Hohndel wrote:

Dirk,

> thank you so much for the detailed, competent, in-depth explanation
> of the science and math involved here. I am amazed to realize that
> Subsurface has its own Chief Scientist to help us get things right!

thanks for the praise.

> Your patch does two things:
> a) it changes the minimum depth taken into account (gf_low_position_min) to 10m
> b) it adds the compile time choice of using either of the two interpretations where GFlow should be applied
>
> Is a) a change we really want to make? Looking at the math (and my
> Mathe Diplom was 20 years ago) it seems that this would have a
> negative impact on the deco times for typical shallow dives (i.e., we
> might end up showing a deco obligations on dives that most
> divecomputers wouldn't) - while this is clearly mostly targeted at
> deeper divers, I don't want to end up in a situation where we suddenly
> seem overly conservative for a 60 minute dive to 12m (because as
> we have already seen here on this mailing list, it's possible to make
> recreational divers unnecessarily nervous by doing so)
>
> And as for b), would it make sense to make this a UI controlled runtime
> setting, or are the variations truly so small that it doesn't matter (I haven't
> played with this for my dives, so I can't tell, yet). My gut feeling would be
> that for a short, very deep dive switching GFlow to apply to the first ceiling
> might once again make deco quite a bit longer.

re b) I implemented it with a compiler option as I think, ultimately you 
want to set it to FALSE and remove the other code but did not do that so 
you can easily compare the two possibilities. As I said, I don't think it 
makes a big difference but from Baker's text it is clear he means "FALSE". 
But as the gradient factor thing is an ad hoc hack and has no real 
physiological basis (apart than giving similar profiles as bubble models) 
"TRUE" is likely to be a similarly valid deco model (assuming good 
parameters), it's just not Baker's.

a) I did as otherwise the computation of the moving ceiling as reference 
for GF_low has no consequence at all for dives with a ceiling that never 
gets deeper than 20m (like my test case). But you are right, you want it 
deeper. This whole gradient factor business is there to add deep stops and 
you only want those for sufficiently deep dives. So at least 20m is 
probably a realistic value.

> Well - the actual depth fluctuations should be taken into account by both
> his divecomputer and Subsurface - but the sampling rate that his
> divecomputer uses for its deco calculations may be shorter than the one
> it uses when storing samples, and that alone could cause this to be off
> (since people want divecomputers to correctly reflect their max depth
> on a dive they may record the max depth during a sampling interval
> while using the correct depth at multiple points during that interval for their
> internal deco algorithm - that could easily explain why Subsurface tends
> to show slightly longer deco requirements on Jan's rather extreme dives.
> But on the other hand for 99.99% of our users this won't make any
> difference.
> But this leads me straight to the next point.

What I ment was: There is no point in worrying in differences in ceilings 
that you cannot (or will not) resolve in your practical diving.

>
>> The mouse over value ("1m") is actually rounded to integer multiples of 
>> meters. If you want more precise values, find the word "Calculated" in 
>> profile.c and in the snprintf replace depthvalue by entry->ceiling. 
>> Then the ceiling is reported in mm and you will see that at the end of 
>> your dive it is even less than a meter I would expect (just don't 
>> commit that change).
>
> The reason I decided to give the ceiling in multiples of meters (and sadly of
> ft for imperial users, but I am considering switching this to multiples of 5 ft)
> is that you otherwise end up in a 'fake precision' situation. There is no such
> thing as 10cm precision of knowing where exactly you risk getting bent.

100% agree. I only said that as I had the impression that Jan was worried 
that his computer had cleared the deco while the mouse over still hat a 1m 
ceiling. I wanted to say that the real difference is likely to be even 
smaller but again, even a 1m difference does not have much of a practical 
significance.

> Finally, hey, Robert… would you like to help us implement an alternative
> deco algorithm as well? I think there are quite a few people who would be
> interested in comparing the profiles between Buehlmann with GF and say
> VPM (which I think is the only bubble algorithm that is sufficiently
> documented that we could try an implementation, anyway).

Hmm. I don't have a strong urge to do that I have to admit. Of course, I 
would be intererested in the results. The problem is only that the prose 
descriptions of VPM are far from being precise/clear enough for me to be 
able to turn that into a (or better: the correct) program. There is, of 
course, the FORTRAN (in all capital letters) source code (or what f2c made 
of that) on http://www.decompression.org/maiken/VPM/Links.htm . It was bad 
enough for me to write C (which I had not done in about 10 years given I 
earn my living in other ways), but 176k of FORTRAN? I had printed that a 
couple of years ago and starred at it for a significant amount of time but 
I have not been able to completely grasp the logic.

Now might be a good time for me to try that again. But doing that and then 
reimplementing this in C is certainly not something that I can do in a 
day. So, no promises.

In addition, I don't believe that VPM(B) (or RGMB for that matter) does 
any better than gradient factors as the additional parameters (like 
physical properties of the bubbles) don't have a good empirical basis 
(like varying those, producing a number of different tables, running real 
decompression experiments at least with goats and from those determining 
the values that get you out of the water in a save way as fast as possible 
for a large number of dive profiles). Everything else is just varying the 
parameters to obtain those deco profiles you knew before were reasonable. 
And that you can do with gradient factors as well. It boils down to the 
fact that it is very hard to sample a parameter space even of realistic 
dimension with a workable number of samples.

Best
Robert

-- 
.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oOo.oO
Robert C. Helling     Elite Master Course Theoretical and Mathematical Physics
                       Scientific Coordinator
 		      Ludwig Maximilians Universitaet Muenchen, Dept. Physik
print "Just another   Phone: +49 89 2180-4523  Theresienstr. 39, rm. B339
     stupid .sig\n";   http://www.atdotde.de