Strange *CLOAD error

1

Hi,

I’ve encountered a strange error. CalculiX fails with the *ERROR reading *CLOAD message pointing to this line:

*CLOAD
2000,2,1.7302535485473385e-06

Just removing the last decimal works:

*CLOAD
2000,2,1.730253548547338e-06

But it’s not only about the number of decimals because this works too:

*CLOAD
2000,2,1.73025354854733853545345345e-06

Now this one fails:

*CLOAD
2000,2,1.73025354854733854e6

But this works:

*CLOAD
2000,2,1.7302535485473385434234243e6

What’s up with that ? I don’t see any pattern here. Is it just random and somehow related to number precision ?

Lines like this were generated by FreeCAD FEM. I need to know what causes the error to figure out how to prevent FreeCAD from creating such problematic lines. Unless it’s just a nasty bug that should be fixed in CalculiX and there are no particular workarounds to avoid it.

P.S. It happened in a large model but I reduced it to a basic beam test where you can easily check which lines work and which don’t:

*NODE
1000,0,0,0
2000,10,0,0
*ELEMENT, TYPE=B31, ELSET=EALL
1,1000,2000
*MATERIAL, NAME=STEEL
*ELASTIC
210000,0.3
*BEAM SECTION, ELSET=EALL, MATERIAL=STEEL, SECTION=RECT
1,2
*BOUNDARY
1000,1,6
*STEP
*STATIC
*CLOAD
2000,1,1.7302535485473385e-06
*NODE FILE
U
*EL FILE
S
*END STEP
2

Ccx reads real numbers using Fortran format f20.0 which is very particular, so check this format rules to understand the behaviour. Not random. I think already said something like this in a previous post.

1 Like
3

@Calc_em
In fact the real problem is a combination that only 20 characters are read for the cload value together with the digit for the float is define to f20, which cause for your input the number will look like this

2000,2,1.7302535485473385e-06
txt 1.7302535485473385E-
error

2000,2,1.730253548547338e-06
txt 1.730253548547338E-0
error

2000,2,1.73025354854733853545345345e-06
txt 1.730253548547338535
wrong number

2000,2,1.73025354854733854e6
txt 1.73025354854733854E
error

2000,2,1.7302535485473385434234243e6
txt 1.730253548547338543
wrong number

reading 30 character and 30 digits instead of 20 example
read(textpart(3)(1:30),‘(f30.0)’,iostat=istat) forcval
will give you this result
2000,2,1.7302535485473385434234243e6
txt 1.7302535485473385434234243E6
float 1730253.5485473385
which except for the minor digits would be better than a complete far away value

1 Like
4

Thank you for the explanation.

But CalculiX requires F20.0, right ? So this number can have a maximum precision like this:

1.73025354854734e-06

Or if 0 after minus is omitted:

1.730253548547339e-6

Is that correct ?

5

@Calc_em
ccx defines reals as real*8 and as far as I know the fortran follows the IEEE 754 standard where you can read exactly how the bits are being used.
Double-precision floating-point format - Wikipedia

which for real 8 gives 17 significant digits no matter how many digits you put in your test string. Looking as my last example where I defined a text string of 30 characters and even that I ask for reading 30 digit I only gets 17 significant digits since the variable is defined as real*8

Fortran format specifier can be quite confusing but “F20.0” will only read 20 characters from the text string no matter how long the text string. The following example explains one way the format specifier could be used.

A modified example from Oracle’s Fortran 77 language reference gives this output and just for fun I defined the variables as real*16 which actual is supported by GFortran and give 36 significant digits:

      implicit none
      real*16 r, s, t
      character line*24 / '12345678 23.5678 .345678' / 
      write(*,*) 'This is what you have as characters'
      write(*,'(a/)') line
      read( line, '( f8.3, f8.3, f8.3 )') r, s, t 
      write(*,*) 'This is what you print'
      print '( f10.3, f11.4, f13.6 ,/)', r, s, t 
      write(*,'(/a)') 'This is what you have internal'
      write (*,*) r, s, t 
      end

the output will be this:

 This is what you have as characters
12345678 23.5678 .345678

 This is what you print
 12345.678    23.5678     0.345678

This is what you have internal
   12345.6779999999999999999999999999993         23.5677999999999999999999999999999995        0.345677999999999999999999999999999982

So f8.3 means that exact 8 characters will be read and used for the value but what also should be noticed about the read format specifier is, if no dot is found in the number, then a dot will be placed 3 digits from right most number, but in case the character string contains a dot then this dot will override number 3 in the format. specifier.

2 Likes