Precision problem with matrix inversion

[Anders]

It could be that the internal precision is too low when inverting the matrix combined with a multiplication with a vector in RPN mode or simpler: division of a vector with matrix.

"In the good old days...bla bla" for instance with the HP 28s in the 80's, this particular situation was recognized as so common by the HP 28S and HP 48 too I believe (trying to solve a large system of unknowns) so when the user did a division of a vector with matrix from the stack it recognized this situation and used the increased 16 digits precision precisely because this extra precision is needed (if I remember correctly the HP 28s used the standard 12 digits precision for normal calculations).

I do not know if the Prime add extra precision or not for Vector Matrix division, but is should and hopefully this would lead to the right result.

After I wrote this I just had to dig up my HP 28S and guess what - it gets it right!

So I got around to punched in the matrix (let's call it M1) in my Prime too and RPN does not behave as it should with this matrix:
1) V1 M1 / results in the null vector and on top:
2) inverting the matrix: M1 shift / results in the 4x4 null matrix
3) M1 1 / also results in the 4x4 null matrix.
As previously posed
4) M1 inv(1) do result in the correct vector