avlemi bug fixed

It turns out that Greg Ward's solution proposed in his message of 31
January 2003 is the solution after all. I have now run a number of tests
with two different models and varying numbers of processes and, hence,
varying numbers of processors sharing processes with other programs
running on the same machine. In no case has the bug reared its ugly head.


A P.S. to Thomas' message -- we corresponded off-line and determined with the help of his sysadmin that he was originally running the old code when he got the avlmemi error, not the fixed version he thought as he thought. I asked him to make sure that the new version was indeed working before he posted a retraction.