Compute the polar decomposition.
Returns the factors of the polar decomposition [R72] u and p such that a = up (if side is “right”) or a = pu (if side is “left”), where p is positive semidefinite. Depending on the shape of a, either the rows or columns of u are orthonormal. When a is a square array, u is a square unitary array. When a is not square, the “canonical polar decomposition” [R73] is computed.
Parameters: | a : (m, n) array_like
side : string, optional
|
---|---|
Returns: | u : (m, n) ndarray
p : ndarray
|
References
[R72] | (1, 2) R. A. Horn and C. R. Johnson, “Matrix Analysis”, Cambridge University Press, 1985. |
[R73] | (1, 2) N. J. Higham, “Functions of Matrices: Theory and Computation”, SIAM, 2008. |
Examples
>>> a = np.array([[1, -1], [2, 4]])
>>> u, p = polar(a)
>>> u
array([[ 0.85749293, -0.51449576],
[ 0.51449576, 0.85749293]])
>>> p
array([[ 1.88648444, 1.2004901 ],
[ 1.2004901 , 3.94446746]])
A non-square example, with m < n:
>>> b = np.array([[0.5, 1, 2], [1.5, 3, 4]])
>>> u, p = polar(b)
>>> u
array([[-0.21196618, -0.42393237, 0.88054056],
[ 0.39378971, 0.78757942, 0.4739708 ]])
>>> p
array([[ 0.48470147, 0.96940295, 1.15122648],
[ 0.96940295, 1.9388059 , 2.30245295],
[ 1.15122648, 2.30245295, 3.65696431]])
>>> u.dot(p) # Verify the decomposition.
array([[ 0.5, 1. , 2. ],
[ 1.5, 3. , 4. ]])
>>> u.dot(u.T) # The rows of u are orthonormal.
array([[ 1.00000000e+00, -2.07353665e-17],
[ -2.07353665e-17, 1.00000000e+00]])
Another non-square example, with m > n:
>>> c = b.T
>>> u, p = polar(c)
>>> u
array([[-0.21196618, 0.39378971],
[-0.42393237, 0.78757942],
[ 0.88054056, 0.4739708 ]])
>>> p
array([[ 1.23116567, 1.93241587],
[ 1.93241587, 4.84930602]])
>>> u.dot(p) # Verify the decomposition.
array([[ 0.5, 1.5],
[ 1. , 3. ],
[ 2. , 4. ]])
>>> u.T.dot(u) # The columns of u are orthonormal.
array([[ 1.00000000e+00, -1.26363763e-16],
[ -1.26363763e-16, 1.00000000e+00]])