variance matrix is nonsymmetric or highly singular problem using the traj command

[Luis Mijares Castaneda]

Hello, I have an issue with the covariance matrix when I use the traj command shown below, I include the variables used here and the covariance matrix as well, I tried exporting the coefficients using the matrix list = () command and changing them to see if the MLE would converge but it did not, I thought death_indicator might be a sparse variable but when I didn't include the death_indicator in the model the results still had a nonsingular matrix. I also use the model without the dropout() and obsmar() subcommands and when then I don't have issues of correlations. Any suggestions would be appreciated

traj, model(cnorm) var(cogtot27_55-cogtot27_90) order(1, 3, 2, 1) indep(zage_55-zage_90) min(-99) max(99) weight(new_weight) obsmar(death_indicator) droput(0, 0, 0,0)