[mostanad]

hello everybody.
i am calculating a geometric series in fortran 77. in fact i have the totall length of series,it's number of statements an stretching factor. my unknown parameter is first number. then i finding the other numbers and find the total length again.all of my parameters are defined by COMMON blocks. after obtaining the first number and finding another numbers in series, i find that my real total length is 0.4999999 but the total from series is 0.4999998 and this difference make for me errors. is any body say me how converge my calculated length of series to real one in fortran 77? thank a lot.

[DarylMusashi]

Hello mostanad,

maybe it helps to define all dependend variables as double precision (if you dont do it already):

IMPLICIT NONE
REAL*8 :: DOUBLE_PRECISION_VARIABLE

Make sure you alsways take the "D0" at the end when you set values to this variable.

DOUBLE_PRECISION_VARIALBE = 3.1415926D0


If this does not help maybe you can share the source code with the community?

[mostanad]

Originally Posted by darylmusashi

hello mostanad,

maybe it helps to define all dependend variables as double precision (if you dont do it already):

Implicit none
real*8 :: Double_precision_variable

make sure you alsways take the "d0" at the end when you set values to this variable.

Double_precision_varialbe = 3.1415926d0


if this does not help maybe you can share the source code with the community?

thanks. But as i said i am defining parameters in common blocks in fortran 77. So any of parameters don't have specific parameter type such as integer,.... . Is any command that can change type of parameter to double precision in code lines? For example 1. Change one to a float parameter.

[DarylMusashi]

Isn*t it possible to define a variable and set it to a parameter after that? Of course take note of the "D0" at the end again.

REAL*8 = PI,VARIABLE_2, RESULT
PARAMETER(PI=3.141592653589793D0,VARIABLE_2=1.2339 2D0)


Secondly a good advise might be not to mix real and integer operations. This might a problem in your case, too. Good example:

RESULT=PI*VARIABLE_2/18.0D0 <---18 is a double precision number

Bad example

RESULT=PI*VARIABLE_2/18 <--- 18 is an integer number

[adrin]

Using COMMON blocks does not prevent you from assigning the variable type. Based on the little that you said I'm guessing you are not using "implicit none", but using fortran's nuance that variables starting with letters other than I-through-N are real. This is generally considered poor programming practice.

Anyway, considering the above you have two options for setting real*8:

1) You can set all your real variables as real*8 at compile time. All compilers have an option to automatically convert real to real*8 at compile/run time.

2) You can specify real*8 in the code for only those variables that you want to specify as double. For example:

real*8 dbl_var
COMMON /mixed_vars/ int_var, real_var, dbl_var

adrin

[flotus1]

Adding to what has already been said: Using higher precision variables is not a magic bullet to avoid round-off or cancellation errors. Without a suitable algorithm you can end up with errors even larger than in your example despite using real8 variables.

[adrin]

Well said! I didn't bother with this component because the original request/message is not easy to comprehend and does not provide any information of value to make recommendations.

To me, if the implementation were correct there would/should be no practical difference between 0.4999999 and 0.4999998. But, again, there was not sufficient information to make any recommendation; in this case, double precision would be the easiest way out (hopefully!)

adrin

[adrin]

Clarification: my "well said" comment was to Alex (flotus1)

adrin