Precision Error in logical evaluation?

Question

Nathan 2014-5-14

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/129647-precision-error-in-logical-evaluation

编辑： Star Strider 2014-5-14

In debugging some code I found the following results confusing. Anyone know how to further investigate this? Or do I simply need to avoid dividing a number by itself? I thought MatLab was "smart" enough to use 1 for the fraction of CF/CF. BTW I'm using R2014a student version on a mid 2009 Mac book pro with OSX 10.9.2

Here is the actual code that causes problems only some of the time as this is inside a function that is called 1000's of times but this error only happens a handful of them.

PL=fzero(@(y)(...
        (s.State(6).H-s.State(4).H)*M_Dot/eta_pump...
        -y*CVTModel_NV(RPM_Pump/RPM_Exp,y/CF)),[0.2*CF CF])/CF;

This is what I did at the command prompt in debugger mode at the line.

CVTModel_NV(RPM_Pump/RPM_Exp,0.2*CF/CF)
ans =
   NaN
K>> CVTModel_NV(RPM_Pump/RPM_Exp,0.2)
ans =
   0.828663580586898
K>> 0.2*CF/CF
ans =
   0.200000000000000
K>> CVTModel_NV(RPM_Pump/RPM_Exp,ans)
ans =
   NaN
K>> CF
CF =
     2.669084425186238e+03
K>> CVTModel_NV(RPM_Pump/RPM_Exp,0.200000000000000000000)
ans =
   0.828663580586898

At this point I commented this line out of the function CVTModel_NV:

Eff(SR>2|SR<0.5|PL<0.2|PL>1)=NaN; % to limit results to applicable area

And then got the following results:

K>> CVTModel_NV(RPM_Pump/RPM_Exp,0.2*CF/CF)
ans =
   0.828663580586898

Then further testing reviled the problem.

K>> CF/CF
ans =
     1
K>> 0.2*CF/CF<0.2
ans =
     1
K>> 0.2*CF/CF==0.2
ans =
     0

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Nathan 2014-5-14

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/129647-precision-error-in-logical-evaluation#answer_136828

在 MATLAB Online 中打开

Adjusting the code to avoid the division repaired the problem. Goes to show that programing in MatLab while quite user friendly, it still requires some training in numerics....

PL=fzero(@(y)(...
        (s.State(6).H-s.State(4).H)*M_Dot/eta_pump...
        -y*CF*CVTModel_NV(RPM_Pump/RPM_Exp,y)),[0.2 1]);

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Star Strider 2014-5-14

编辑：Star Strider 2014-5-14

Vectorizing them may also have solved it. If it only occurred a handful of times, those times could have been when the elements were vectors rather than scalars, and required element-wise operations.

You don’t tell us what the error was, so we’ll never know.

请先登录，再进行评论。