Software Suggestions

[UWSojourner]

Really the ZHL thing is that there are limits of supersaturation and the algorithm works by keeping within those limits. The bodge is to reduce those limits according to where the first stop lies. One part is a tested algorithm, developed by well regarded scientist in the field. The other part is an addon produced by someone else with a view to adding conservatism at depth. Other than lots of opinion, is there really anything to support how that is done? The choice of those numbers is subject to the same flaws as whatever critical bubble radius applies to a bubble model.

GF is deceptively accessible 'dial a profile', they ought to be more conservative than raw ZHL-16C, but without testing how do we know? We are back to the same place as arguing over VPM etc.

I think there's at least reasonable evidence that GF would have faired better than VPM in the NEDU profile (the evidence being similarity in supersaturation patterns and exposure). Also, those patterns are repeatable in other profiles were all more familiar with. So I guess I'd just say GF seems to deemphasize deep stops naturally, whereas VPM is clearly a deep stop model.

I thought the "bodge" you described was more toward problems with implementation, not the fact that live dive trials haven't confirmed any particular GFX/Y.

 

[sunnyboy]

I'm just curious. Does anyone have a working link for the original Baker VPM fortran program? The link I have does not work.

 

[UWSojourner]

ftp://decompression.org/Baker/VPM-B%20Fortran%20Source%20Code.txt
Login: downloadfiles@decompression.org
Password: decompression1

 

[sunnyboy]

ftp://decompression.org/Baker/VPM-B%20Fortran%20Source%20Code.txt
Login: downloadfiles@decompression.org
Password: decompression1

Thank you very much. The link I had contained an older login (same password). I have the program now. Time to try it on my replica PDP8. :-D

 

[leadduck]

Thank you very much. The link I had contained an older login (same password). I have the program now. Time to try it on my replica PDP8. :-D

In order to make it run on Linux today, you need to change a few things:

• change CALL SYSTEMQQ to CALL SYSTEM
• change PAUSE to
  WRITE (*,*) 'Press Enter to Continue'
  READ (*,*)
• change OS_Command = 'CLS' to OS_Command = 'clear'
• change length of OS_Command from 3 to 5 in its declaration
• replace the call to GETTIM at the very end with

INTEGER, dimension(8) :: values
CALL DATE_AND_TIME(VALUES=values)
Year = values(1)
Month = values(2)
Day = values(3)
Clock_Hour = values(5)
Minute = values(6)

Then compile with gfortran.

In order to run it, you need an input file VPMDECO.IN for the profile, and settings file VPMDECO.SET
You find their format description in ftp://decompression.org/Baker/VPM%20Decompression%20Program%20in%20Fortran.pdf

The settings there are for VPM. To get VPM-B, reduce critical radius N2 to 0.55, He to 0.45, and lambda to 6500, as suggested in ftp://decompression.org/Baker/VPM-B%20Program%20Update%20Explanation.pdf

 

[leadduck]

Really the ZHL thing is that there are limits of supersaturation and the algorithm works by keeping within those limits. The bodge is to reduce those limits according to where the first stop lies. One part is a tested algorithm, developed by well regarded scientist in the field. The other part is an addon produced by someone else with a view to adding conservatism at depth. Other than lots of opinion, is there really anything to support how that is done? The choice of those numbers is subject to the same flaws as whatever critical bubble radius applies to a bubble model.

GF is deceptively accessible 'dial a profile', they ought to be more conservative than raw ZHL-16C, but without testing how do we know? We are back to the same place as arguing over VPM etc.

I think the point about using GF was, that with GF<100 the profile is guaranteed to stay within the ceiling of original ZHL16B. Whereas a bubble model may produce profiles that are equivalent to GFhigh>100 and violate ZHL16B ceilings. While we don't know much about how much safer GF70 is compared to GF80, we can be quite confident that GF110 has a high DCS risk. (limiting by two models such as VPM-B/GFS avoids that)

I agree that GF choice is "dial a profile", just like using VPM-B/GFS, or adding Pyle stops, or GVE, or PDIS, or extending or inserting some stops manually as a multilevel dive, ... all you can do is take care that your final profile stays within the ceiling of an experimentally tested model. But you don't know how much safer your profile is compared to a plain GF100/100, or whether there's another profile that is more efficient (i.e. safer with equal runtime) than the one you picked.

Thinking about it, you may not even be safer than plain ZHL. The model limits the maximum supersaturation of compartments, but we know that this is insufficient to limit DCS risk because the time integral of supersaturation (ISS) matters, too. When you add deep stops and recalculate the shallow stops by ZHL, you will end up with a profile that stays within the supersaturation limits but still may have a higher ISS in some compartments, and hence a higher DCS risk, than if you had followed the ceiling with no added stops. Same thing with GF, 60/80 may produce less integral supersaturation than 40/70 in some slower compartments. I suppose the future lies in probabilistic models based on ISS.

 

[KenGordon]

I think there's at least reasonable evidence that GF would have faired better than VPM in the NEDU profile (the evidence being similarity in supersaturation patterns and exposure). Also, those patterns are repeatable in other profiles were all more familiar with. So I guess I'd just say GF seems to deemphasize deep stops naturally, whereas VPM is clearly a deep stop model.

I thought the "bodge" you described was more toward problems with implementation, not the fact that live dive trials haven't confirmed any particular GFX/Y.

The distribution of stops with GF will depend on the numbers dialled in. Give it a low enough low value and a high enough high value and I expect you could get nasty supersaturation graphs with it by staying deep too long.

I am not at all arguing pros and cons of VPM vs GF, just pointing out that GF is something made up by an engineer, not Buhlmann and not tested. Worse still, the choice of GF value is left to the user who is not very well informed.

 

[KenGordon]

Thinking about it, you may not even be safer than plain ZHL. The model limits the maximum supersaturation of compartments, but we know that this is insufficient to limit DCS risk because the time integral of supersaturation (ISS) matters, too. When you add deep stops and recalculate the shallow stops by ZHL, you will end up with a profile that stays within the supersaturation limits but still may have a higher ISS in some compartments, and hence a higher DCS risk, than if you had followed the ceiling with no added stops. Same thing with GF, 60/80 may produce less integral supersaturation than 40/70 in some slower compartments. I suppose the future lies in probabilistic models based on ISS.

I think this is a good point. We'd have to run the numbers to see, and then make a blind leap as to whether supersaturation time time in compartment n is more or less important than the similar integral in compartment n+1. If it is not a blind leap please point me at why as I would be VERY interested.

My points are, while the science in complicated the code is not. And while I might not trust certain deco software authors with the science I do actually trust their code enough to plan dives based on it.

Babies, bath water etc.

 

[leadduck]

I think this is a good point. We'd have to run the numbers to see, and then make a blind leap as to whether supersaturation time time in compartment n is more or less important than the similar integral in compartment n+1. If it is not a blind leap please point me at why as I would be VERY interested.

Please see this paper:
http://www.diverbelow.it/attachment...diction using linear-exponential kinetics.pdf

The weighted sum of integral supersaturation is in eq.9, with weights Gn. The weights, thresholds and half-times got fitted by experimental data. I wrote a program for the EE1 model and an optimizer to find the profile with shortest runtime for a given maximum DCS risk. The resulting profiles were very shallow. Maybe it's a bug, I have no reference data to compare and test.

 

[UWSojourner]

The distribution of stops with GF will depend on the numbers dialled in. Give it a low enough low value and a high enough high value and I expect you could get nasty supersaturation graphs with it by staying deep too long ... Worse still, the choice of GF value is left to the user who is not very well informed.

As I pointed out in the various deep stop threads, GF in most any configuration shows better supersaturation patterns than VPM (better if you're trying to reflect the current state of research regarding deep stop models). Since you brought up ill-informed users, just contemplate what they might do with GFs and the impact.

I've heard people say GF "can be made to look like VPM". I think what they mean is that you can pick a low GF to give you the same 1st stop as a VPM profile. Then you can pick a high GF to get you out of the water at the same time. Therefore, first stop is the same and exit time is the same. Does that mean GF and VPM would be equivalent? No.

Look at the GF by Depth chart below for a CCR 300ft 25 min 1.2 O2@20 dive. The 3 profiles all have the same run time. The GF16/94 profile also has the same 1st stop depth as the VPM-B+3 profile.

Comments:

1. Notice the sagging VPM line. This is very typical of VPM and is a direct reflection of the deeper stops issue.
2. Notice that even the GF settings that "matched" the VPM schedule (i.e. same 1st stop, same run time) is shallower in the sense that it doesn't linger as long on the deeper stops. This manifests itself in surfacing cleaner (i.e. VPM exits at about GF 107 and GF16/94 at GF of 94). This could have been shown in the heat map as well, but the GF chart above clearly shows the same pattern.
3. Finally, if the diver moved the 1st stop to GF40/89 and kept the same run time, then a little better exit could be achieved. But notice that the larger benefit was just switching to GF even if you kept the same 1st stop and total runtime.

Now look at the ISS chart below for these dives.

The GF setting that "matches" VPM exits the water with 14% less deco stress. Moving to the GF40/89 profile (remember, same run time) reduces your surface deco stress by 19% compared to the same VPM schedule.

So, while I agree with you that there are some divers still on a learning curve regarding how to exactly use the GF settings I would note the following:

1. Just using GF has benefit even if the diver is completely unversed in the current deep stop debate and is still trying to make GF look like VPM.
2. Some divers (perhaps most?) would look at the GF40/89 settings and think it worth considering dropping the high GF further (which can't be a bad thing). I mean, I think surfacing GFs around 90 are seen as a bit on the aggressive side especially when, for example, Shearwater computers use GF30/70 as the default (down from 30/85 when I first bought one).
3. The VPM-B+3 setting exits the water at a GF107. Yet that setting (+3) is still being described as "the normal setting for most divers" and appropriate for "strenuous, cold, a series of multi day dives, extra safety, or a prior history of DCS or symptoms."

If we're worried about divers who are still on the learning curve regarding deco theory and settings, it seems obvious that GF is more forgiving than VPM and I'd much rather have a diver pick settings for GF than VPM, especially given the "guidance" being offered regarding the meaning of VPM's conservatism settings.