feedstocks corn stover and generic perennial lead to floating point error in smoothed NPV deviation

[rjhanes]

this bug appeared after solving issue #23 in branch feedstockarray

[rjhanes]

in feedstockarray branch, prior to merging pull request

• setting the corn stover starting price to $61 (default price + $1) resolves the error, but $60.90 and below still has the error
• Identically, setting the perennial starting price to $137 (default price + $1) resolves the error, but $136.90 and below has the error

[rjhanes]

in master branch after merging pull request

• selecting the commodity sugar feedstock does not give an error even when sugar starting price is set to $136 (default value for perennial) or $60 (default value for corn stover)
• same error for stover, perennial at slightly different time steps, all before 2016.0, when integration type is changed from Euler (default) to RK4 Auto, Difference, RK2 Auto, with the same time step
• no error for stover, perennial but many warnings when integrator is RK2 Fixed or RK4 Fixed, with the same time step
• no error for stover, perennial, but new error for soybeans, and fewer warnings (BS balance only) when integrator is Euler and time step is increased to 0.0625

[rjhanes]

this bug first appeared in commit d6ace32

• no changes were made to either the corn stover price model or the perennial price model, beyond moving and rearranging the visual representation
• the monthly change in prices for all price models are not consistent across commits even when no parameters except feedstock choice are altered - this suggests that the RANDOM NORMAL() number generators in the monthly price changes are not providing identical numbers from one run to the next
• changing the random stream to 1 (default is 0) resolves this bug
• this is leading me to think this bug comes from some behind-the-scenes Vensim functioning rather than a change to the model itself
• unclear how to confirm or disprove this theory
• also the values for smoothed NPV deviation and its component factors look normal up until the floating point error - nothing apparent is increasing/decreasing to the point of causing an error

sample results: latest-commit is with commodity sugar feedstock, latest-commit-local is with corn stover feedstock (smoothed NPV deviation is the one with the floating point error)

latest-commit is commodity sugar, latest-commit-local is perennial (floating point error occurs earlier than with corn stover):

calculation structure:

[bwbush]

@rjhanes, this was caused by the argument of the SQRT function in smoothed NPV deviation becoming very slightly negative due to round-of errors. In 3283715757cf34dba74d5a4def8e9075b8fbaea3, I added a guard against this. I also reviewed the code and found no other uses of SQRT.

Please retest a couple of the problematic cases to verify that this change fixes all of them. (Because some of the round-off errors didn't appear on my Windows VM, I couldn't test every case you mentioned as being problematical. I mostly focused on testing the perennial case with RK4 auto.)

[rjhanes]

I verified that under default model settings, corn stover and generic perennial are both error-free