I have a dataset of 547 columns with 643 cases, with a large proportion of missingness, so I am attempting to use full-information maximum likelihood in a factor analysis. To do this, I am attempting to use the corFiml function to get a matrix using fiml to pass to the fa() function. However, when I try to use the corFiml function on the dataset, I receive the error:

Error in nlminb(): ! long vectors not supported yet: memory.c:3948

There is roughly a 50% missingness rate, as we used a planned missingness design. This is unlikely to be a memory issue, as I am running the code using 500gb of RAM. I have tried using both a regular R script and rmd (including removing cache=TRUE and cache.lazy=FALSE, as others have suggested).

As to the factor analysis itself, I have tried using

fa(data, fm = "pa", rotate = "none", missing = TRUE, impute = "fiml")

But have received a nonpositive definite correlation matrix. Using multiple imputation for the missingness has proved insurmountably computationally demanding (even using a 1tb of RAM, the imputation has not finished running in half a year).

No solution that I have found online has worked thus far, and I would appreciate any assistance.