Inexacterror when inverting matrix of Int64

[Atul Shrivastava]

When projecting onto a vector space, I use of inv(A'*A) often in deriving the projection matrix. However, when A = [1 0 0; 0 1 0; 0 0 1; 0 0 0] and B = 2*A (example used from Strang's Linear Algebra book 4th ed, pg 215), I get an InexactError() when invoking inv(B'*B). However inv(A'*A) works fine (A'*A is the 3x3 identity matrix).

When B is array of Float64 (using B = 2.0 *A), the matrix inversion works - the returned array is approx. 0.25 * I (identity matrix).

Saw another post where type conversion (Int64-> Float64) was required in some matrix operation context (not inversion though). Is this type conversion a required step before matrix inversion?

Thanks

[Andreas Noack Jensen]

Hi Atul

Thank you for reporting. Generally, Julia takes care of the promotion and she should have done so in your case also. The bug is now fixed. 

Regarding the projection, it might be better to use the QR or SVD of A instead of explicitly calculating inv(A'A).

--
Med venlig hilsen

Andreas Noack Jensen