Matrix Dimension Error with Parfor

I am running the following code

function output = replicateerror2(network,sample)

global stoich_matrix ngas

load(network,'stoich_matrix','BE_DFT','ngas');
load(sample,'sample_bound_low','sample_bound_up');
load('work\lhsfile','lhs');

[nspecies,nreact] = size(stoich_matrix);
[nparam,ndtst] = size(lhs);
range = sample_bound_up - sample_bound_low;
BE_DFT = BE_DFT;
lhs = lhs;
sample_bound_low = sample_bound_low;
output = zeros(nreact,ndtst);

parfor i = 1:ndtst
    param = sample_bound_low + range.*lhs(:,i);
    BE = [zeros(ngas,1); param(1:(nspecies-ngas-1)); 0];
    BE_fit = BE - BE_DFT;
    output(:,i) = microkin(BE_fit);
end

end

function test = microkin(BE_fit
global stoich_matrix
test = stoich_matrix'*BE_fit;
end

The code runs fine with a for loop or if there is no matlabpool open. However, when I open a matlab pool I get the following error:

Inner matrix dimensions must agree.

associated with the line

test = stoich_matrix'*BE_fit;

Answer 1

[Edric Ellis]

You cannot use GLOBAL data together with PARFOR in this way. The MATLAB workers in your pool are separate processes and therefore they have different GLOBAL workspaces. More details here. In particular, MATLAB's LOAD function is not "transparent" -that is, it causes variables to come into existence in a way that PARFOR cannot understand. I would suggest the following: place your PARFOR loop in another sub-function that takes as arguments 'stoich_matrix', 'ngas' etc., and stop using GLOBAL data.