Fortran矩阵分配出现无效内存引用错误的技术求助

核心错误点 #

• 非法内存访问：代码先执行deallocate(final_spot)释放矩阵内存，之后才调用minval(final_spot)读取矩阵值并打印，访问已释放的内存是直接导致运行报错的原因。
• 整数隐式截断风险：n_spot是整数类型，直接将浮点运算结果n_spot - (n_spot * 0.018)赋值给整数变量，小数部分会被默认截断，逻辑不明确。
• 矩阵维度不匹配：初始分配final_spot时用的是初始值n_spot=125作为列数，后续日循环中n_spot不断缩小，如果需要矩阵尺寸跟随存活点位数量动态变化，需要在每次循环中重分配矩阵。
• 补充：如果编译阶段提示truncated_normal未定义，需要确保你已经实现了该截断正态分布生成子程序，或者链接了包含该子程序的源码/库文件。

修复后参考代码 #

PROGRAM Rice_model
IMPLICIT NONE
real, allocatable, dimension (:,:) :: final_spot
real, dimension(125) :: size_small_spot, size_large_spot
integer :: days, i, j, n_spot, initial_spot
! 建议用Fortran2008标准的kind定义，避免real(kind=8)的移植性问题
integer, parameter :: dp = kind(1.d0)
real(dp) :: mu_small_spot,lower_small_spot,upper_small_spot, size_of_spot
real(dp) :: mu_large_spot,lower_large_spot,upper_large_spot
integer :: seed
real(dp) :: sigma_spot
real(dp) :: min, max

! defining variables for the truncated normal distribution for sizes
mu_small_spot = 42.8_dp
lower_small_spot = 33.0_dp
upper_small_spot = 49.0_dp

mu_large_spot = 63.7_dp
lower_large_spot = 54.0_dp
upper_large_spot = 78.0_dp
sigma_spot = 4.0_dp
seed = 123456789

initial_spot = 125
n_spot = initial_spot
days = 20

! SMALL SPOT
do i = 1,initial_spot
    call truncated_normal(mu_small_spot,sigma_spot, &
        lower_small_spot,upper_small_spot,seed,size_of_spot)
    WRITE (*,*) i, size_of_spot
    size_small_spot(i) = real(size_of_spot)
end do
min = minval(size_small_spot, dim = 1)
max = maxval(size_small_spot, dim = 1)
WRITE(*,*) "Min Size = ", min
WRITE(*,*) "Max Size = ", max

! LARGE SPOT
do i = 1,initial_spot
    call truncated_normal(mu_large_spot,sigma_spot, &
        lower_large_spot,upper_large_spot,seed,size_of_spot)
    WRITE (*,*) i, size_of_spot
    size_large_spot(i) = real(size_of_spot)
end do
min = minval(size_large_spot, dim = 1)
max = maxval(size_large_spot, dim = 1)
WRITE(*,*) "Min Size = ", min
WRITE(*,*) "Max Size = ", max

allocate(final_spot(1:3,initial_spot))
do i = 1,days
    ! 显式取整，避免隐式截断
    n_spot = n_spot - int(n_spot * 0.018_dp)
    do j = 1, n_spot
        final_spot(1,j) = real(j)
        final_spot(2,j) = size_small_spot(j)
        final_spot(3,j) = size_large_spot(j)
    end do
end do

! 打印逻辑移到内存释放之前，仅读取已赋值的前n_spot列
WRITE(*,*) "The min values are", minval(final_spot(:,1:n_spot), dim = 2)
WRITE(*,*) "The max values are", maxval(final_spot(:,1:n_spot), dim = 2)
! 内存释放放在所有使用矩阵的逻辑最后
deallocate(final_spot)

END PROGRAM Rice_model

内容的提问来源于stack exchange，提问作者Kelsey Spencer