Solving Negative Infinity in Fortran Calculation

[nguyenthanhlam]

Dear all,

I can not figure out why I got negative infinity in my output, so please help. Here is my calculation for FAMAX:

FAMAG=SQRT(FX**2+FY**2+FZ**2)

FAMAX=MAXVAL(FAMAG(1:NATOM))

and for FPBPMAX:

FMAGPBP(1:NSYN)=SQRT(FXGTASE(1:NSYN)**2+FYGTASE(1:NSYN)**2+FZGTASE(1:NSYN)**2)

FPBPMAX=MAXVAL(FMAGPBP(1:NSYN))

FMAGPBP(1:NSYN)=SQRT(FXTPASE(1:NSYN)**2+FYTPASE(1:NSYN)**2+FZTPASE(1:NSYN)**2)

FPBPMAX=MAX(MAXVAL(FMAGPBP(1:NSYN)),FPBPMAX)

FMAGPBP(1:NSYN)=SQRT(FXEDASE(1:NSYN)**2+FYEDASE(1:NSYN)**2+FZEDASE(1:NSYN)**2)

FPBPMAX=MAX(MAXVAL(FMAGPBP(1:NSYN)),FPBPMAX)

I expected it crash and got either NaN or Infinity for FAMAX and FPBPMAX at some point. But what I did not expect was FAMAX and FPBPMAX got NEGATIVE infinity when it crashed. Does anyone know why?

Thanks,

Lam

 

[SteamKing]

In the first line, FAMAG is used as a single variable; in the next line, FAMAG all of a sudden has become an array variable. This sudden change can be very confusing. Was FAMAG originally dimensioned as an array variable?

 

[nguyenthanhlam]

In the first line, FAMAG is used as a single variable; in the next line, FAMAG all of a sudden has become an array variable. This sudden change can be very confusing. Was FAMAG originally dimensioned as an array variable?

FX,FY,FZ, FAMAG are arrays, but FAMAX is dimensionless. I know I should have used clearly different names to avoid confusion.

Again, FMAGPBP is an array, FPBPMAX is dimensionless.

Lam

 

[SteamKing]

I think when you assign a value to FAMAG in the first line, you have to assign it to a particular location in the array. You can't write FAMAG = some expression, when you need to write FAMAG (array_index) = some expression.

 

[AlephZero]

I think when you assign a value to FAMAG in the first line, you have to assign it to a particular location in the array.

A statement like FAMAG =SQRT(FX**2+FY**2+FZ**2) should work in Fortran 90/95 , so long as FAMAG, FX, FY, FZ are all arrays with the same dimensions. it is equivalent to a loop (or nested loops fur multi-dimension arrays) to operate on each element of the arrays.

The only thing I can think of is something weird like:
Your implementation of the MAXvAL function starts by setting its return value to negative infinity, and then scans through the array to find larger numbers. But because of the curiosities of IEEE floating point arithmetic, comparisons involving NaN tend to always return "false" (in fact NaN is not even equal to NaN!) So the maximum value of an array consisting entirely of NaNs would be returned as negative infinity. (But if that is the case, it sounds like a bug to me).

 

[nguyenthanhlam]

A statement like FAMAG =SQRT(FX**2+FY**2+FZ**2) should work in Fortran 90/95 , so long as FAMAG, FX, FY, FZ are all arrays with the same dimensions. it is equivalent to a loop (or nested loops fur multi-dimension arrays) to operate on each element of the arrays.

The only thing I can think of is something weird like:
Your implementation of the MAXvAL function starts by setting its return value to negative infinity, and then scans through the array to find larger numbers. But because of the curiosities of IEEE floating point arithmetic, comparisons involving NaN tend to always return "false" (in fact NaN is not even equal to NaN!) So the maximum value of an array consisting entirely of NaNs would be returned as negative infinity. (But if that is the case, it sounds like a bug to me).

Yes FAMAG, FX, FY, FZ are in the same dimensions.

And your guessing is right. Maxval(array(1:n)) gives -Infinity if all the elements are NaN. I just did a simple test to confirm this. Thanks a lot.

Lam