Why are the same parameters in the same formula different from those outside cvx

My code is as follows. I can ensure that my parameters are absolutely correct
global B
global J
global N
global N_0
global D_k
global sum_gr
global h_k_u
global h_u_b
global h_j_m_max
global h_j_e
global b_jn
p_j1 = ones(J,N);

    cvx_begin

     variable   p_j(J,N) ;
     expression An1(J,N) ;
     expression An4(J,N) ;
     expression Bn2(J,N) ;
     expression obj(J,N) ;
     expression obj1 ;
     expression w(J,N) ;
     

            for j = 1:J
                for n =1:N
                    An1(j,n) =  b_jn(j,n).*B.*( log( 1+(h_j_m_max(j,n).*p_j(j,n)./(B.*N_0.*b_jn(j,n))) )/log(2) );
                    An4(j,n) = (abs(h_j_e(j,n))^2)./(sum_gr(1,1,n)+B.*N_0.*b_jn(j,n));
                    Bn2(j,n) = b_jn(j,n).*B.*( An4(j,n).*p_j(j,n)./( (1+An4(j,n).*p_j1(j,n)).*log(2) ) + log(1+An4(j,n).*p_j1(j,n))/log(2) - An4(j,n).*p_j1(j,n)./((1+An4(j,n).*p_j1(j,n)).*log(2)) );
                    obj(j,n) = 1e-6.*(0.4.*p_j(j,n));
                    obj1 = sum(sum(obj,2),1);
                    w(j,n) = An1(j,n)-Bn2(j,n);
                end
            end

            minimize(obj1)

            subject to
            p_j<1;
            p_j>0.3;

    cvx_end

The independent variable pj is solved as 0.3, and the parameter An1 is solved as a fully sparse matrix (abnormal) in cvx, while the same parameter An1 is normal outside cvx，can you tell me why？

That’s just the way it is. if you really need An1 to be full, you can set An1 = full(An1); after cvx_end.

Actually,because An1 is a CVX expression and not a CVX variable, if you want to be assured of having the “optimal” value of An1, you have to recompute it after cvx_end, starting from CVX variable values. I.e., after cvx_end, place the code:

    An1 = zeros(J,N);
    for j = 1:J
        for n =1:N
            An1(j,n) =  b_jn(j,n).*B.*( log( 1+(h_j_m_max(j,n).*p_j(j,n)./(B.*N_0.*b_jn(j,n))) )/log(2) );
        end
    end

In this situation, An1 follows the standard MATLAB double precision variable rules.