Optimization Codes by ChatGPT

The prompts given to ChatGPT-4 are shown in the codes. Sometimes cleanup was required. This is an experiment in code generation by ChatGPT -- the codes should not be relied upon.

Compile with

gfortran optim_util.f90 hooke_jeeves.f90 simulated_annealing.f90 adam.f90 nelder_mead.f90 conjugate_gradient_min.f90 bfgs_min.f90 luus_jaakola.f90 gradient_descent.f90 xminimize.f90

Works with gfortran 12.0.1 20220213 and ifort version 2021.6.0.

Some ChatGPT-4 quirks

• When ChatGPT-4 generates code, it often forgets the import statement that is needed in the interface below.

    interface
        function funk(x)
            import :: dp
            real(kind=dp), dimension(:), intent(in) :: x
            real(kind=dp) :: funk
        end function funk
    end interface`

• It can produce code with implicit none and undeclared variables that does not compile. It tends to declare a variable i as integer even in a procedure that does not use the variable i.

• When ChatGPT-4 translated some algorithms from the literature with g*g' or equivalently g*transpose(g), it erroneously translated this to g**2 in Fortran instead of sum(g**2).

• ChatGPT-4 sometimes forgets that the random_number() intrinsic is a subroutine, not a function, so that you cannot write

x = random_number()

• ChatGPT-4 sometimes declares as intent(in) an argument that needs to be intent(in out), since its initial value in changed in the procedure.

• ChatGPT-4 can incorrectly use a procedure argument in what is supposed to be a constant expression, as in

module sv_mod
use kind_mod, only: dp
use random_mod, only: random_normal
implicit none
contains
subroutine ret(n, kappa, theta, eta, y, x)
integer      , intent(in)    :: n
real(kind=dp), intent(in)    :: kappa, theta, eta
real(kind=dp), intent(out)   :: y(n), x(n)
! for line below, gfortran says 'Error: Parameter 'n' at (1) has not been declared or is a variable, which does not reduce to a constant expression'
real(kind=dp)                :: dt = 1.0_dp / n 
integer                      :: i
real(kind=dp)                :: dW, dV
y(1) = theta
do i=2,n
   dW = sqrt(dt) * random_normal()
   dV = sqrt(dt) * random_normal()
   y(i) = y(i-1) + kappa * (theta - y(i-1)) * dt + eta * sqrt(max(0.0_dp, y(i-1))) * dV
   x(i) = exp(y(i-1)) * dW
end do
end subroutine ret
end module sv_mod