forums.silverfrost.com

[Neo]

Dear People,

My question can be distilled to this specific example. I need to write the interface for a subroutine that calculates the gradient of an arbitrary scalar function of a vector argument x. This should work in general for a function of 2 variables , 3 variables, etc. Thus, I am tempted to use an assumed shape for position argument x in the interface. In other words, I want to avoid having to put a specific dimension of argument x to the function being passed - otherwise the subroutine is not general for any dimension of space. The following attempts do not work, but I don't know why. I looked in the corresponding sections of Metcalf and two other books, but they do not cover this. Any comments will be gratefully received.

With many thanks,
Ludwig

PS: The colon in the assumed shape is being turned into an emiticon, but you get the idea.

FIRST EXAMPLE (pass function as argument)

subroutine gradient (func,x,g)
real, dimension(, intent(in) :: x
real, dimension(size(x)), intent(out) :: g
interface
function func (y) result (f)
real, dimension(, intent(in) :: y
real :: f
end function func
end interface
...
end subroutine gradient

SECOND EXAMPLE (pass function as argument)

subroutine gradient (func,x,g)
real, dimension(, intent(in) :: x
real, dimension(size(x)), intent(out) :: g
interface
function func (y) result (f)
real, dimension(size(x)), intent(in) :: y
real :: f
end function func
end interface
...
end subroutine gradient

THIRD EXAMPLE (pass subroutine as argument)

subroutine gradient (sbr,x,g)
real, dimension(, intent(in) :: x
real, dimension(size(x)), intent(out) :: g
interface
subroutine sbr (y,f)
real, dimension(, intent(in) :: y
real, intent(out) :: f
end subroutine sbr
end interface
...
end subroutine gradient

[PaulLaidler]

The following appears to work provided that you don't use CHECKMATE

[Neo]

Dear Paul,

Yes, yours is the solution. I have now tried this out and the code works.

It does mean that I each function must accept an assumed shape argument, even when the mathematical argument is a (/ x,y /) or (/ x,y,z /), etc. But that is really no problem.

With profuse thanks,
Ludwig