Is this a new bug?

A'*A, its eigenvalues' calculated results has a negetive value:

C =

0.0215 -0.0014 -0.2572

-0.0014 0.0006 0.1029

-0.2572 0.1029 18.0000

eig(C'*C)

ans =

-1.22819777351878e-17

0.000316768541555439

324.153592202518

[Walter Roberson]

which release and OS are you using?

      1.31720860254795e-10
      0.000317607460000373
          324.153629422408

[Bruno Luong]

Before asking for release make sure you guys exchange data not from screen display.

[Walter Roberson]

num2hex can show us the exact values you used

Answer 1

[Bruno Luong]

Numerical inaccuracy limitation. Nothing new in this front and no one consider them as BUG.

Sometime symmetrize the matrix help

D = C'*C;
eig(0.5*(D + D'))

Further reading if the above is not enough to solve the "bug": nearest SPD