LU with Partial Pivoting¶

MIT License

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

import numpy as np
import numpy.linalg as la

np.set_printoptions(precision=3, suppress=True)

Set-up¶

Let's grab a (admittedly well-chosen) sample matrix A:

n = 4

np.random.seed(235)
A = np.round(5*np.random.randn(n, n))
A[0,0] = 0
A[2,1] = 17
A[0,2] = 19
A

array([[  0.,   4.,  19.,  -7.],
       [ -1.,  -2., -10.,  -0.],
       [  1.,  17.,   1.,  -4.],
       [ -5.,  -8.,  -6.,  -2.]])

Permutation matrices¶

Now define a function row_swap_mat(i, j) that returns a permutation matrix that swaps row i and j:

def row_swap_mat(i, j):
    P = np.eye(n)
    P[i] = 0
    P[j] = 0
    P[i, j] = 1
    P[j, i] = 1
    return P

What do these matrices look like?

row_swap_mat(0,1)

array([[ 0.,  1.,  0.,  0.],
       [ 1.,  0.,  0.,  0.],
       [ 0.,  0.,  1.,  0.],
       [ 0.,  0.,  0.,  1.]])

Do they work?

row_swap_mat(0,1).dot(A)

array([[ -1.,  -2., -10.,   0.],
       [  0.,   4.,  19.,  -7.],
       [  1.,  17.,   1.,  -4.],
       [ -5.,  -8.,  -6.,  -2.]])

Part I¶

U is the copy of A that we'll modify:

U = A.copy()

First column¶

Create P1 to swap up the right row:

P1 = row_swap_mat(0, 3)
U = P1.dot(U)
U

array([[ -5.,  -8.,  -6.,  -2.],
       [ -1.,  -2., -10.,   0.],
       [  1.,  17.,   1.,  -4.],
       [  0.,   4.,  19.,  -7.]])

M1 = np.eye(n)
M1[1,0] = -U[1,0]/U[0,0]
M1[2,0] = -U[2,0]/U[0,0]
M1

array([[ 1. ,  0. ,  0. ,  0. ],
       [-0.2,  1. ,  0. ,  0. ],
       [ 0.2,  0. ,  1. ,  0. ],
       [ 0. ,  0. ,  0. ,  1. ]])

U = M1.dot(U)
U

array([[ -5. ,  -8. ,  -6. ,  -2. ],
       [  0. ,  -0.4,  -8.8,   0.4],
       [  0. ,  15.4,  -0.2,  -4.4],
       [  0. ,   4. ,  19. ,  -7. ]])

Second column¶

Create P2 to swap up the right row:

P2 = row_swap_mat(2,1)
U = P2.dot(U)
U

array([[ -5. ,  -8. ,  -6. ,  -2. ],
       [  0. ,  15.4,  -0.2,  -4.4],
       [  0. ,  -0.4,  -8.8,   0.4],
       [  0. ,   4. ,  19. ,  -7. ]])

Make the second-column elimination matrix M2:

M2 = np.eye(n)
M2[2,1] = -U[2,1]/U[1,1]
M2[3,1] = -U[3,1]/U[1,1]
M2

array([[ 1.   ,  0.   ,  0.   ,  0.   ],
       [ 0.   ,  1.   ,  0.   ,  0.   ],
       [ 0.   ,  0.026,  1.   ,  0.   ],
       [ 0.   , -0.26 ,  0.   ,  1.   ]])

U = M2.dot(U)
U

array([[ -5.   ,  -8.   ,  -6.   ,  -2.   ],
       [  0.   ,  15.4  ,  -0.2  ,  -4.4  ],
       [  0.   ,   0.   ,  -8.805,   0.286],
       [  0.   ,   0.   ,  19.052,  -5.857]])

Third column¶

Create P3 to swap up the right entry:

P3 = row_swap_mat(3, 2)
U = P3.dot(U)
U

array([[ -5.   ,  -8.   ,  -6.   ,  -2.   ],
       [  0.   ,  15.4  ,  -0.2  ,  -4.4  ],
       [  0.   ,   0.   ,  19.052,  -5.857],
       [  0.   ,   0.   ,  -8.805,   0.286]])

Make the third-column elimination matrix M3:

M3 = np.eye(n)
M3[3,2] = -U[3,2]/U[2,2]
M3

array([[ 1.   ,  0.   ,  0.   ,  0.   ],
       [ 0.   ,  1.   ,  0.   ,  0.   ],
       [ 0.   ,  0.   ,  1.   ,  0.   ],
       [ 0.   ,  0.   ,  0.462,  1.   ]])

U = M3.dot(U)
U

array([[ -5.   ,  -8.   ,  -6.   ,  -2.   ],
       [  0.   ,  15.4  ,  -0.2  ,  -4.4  ],
       [  0.   ,   0.   ,  19.052,  -5.857],
       [  0.   ,   0.   ,   0.   ,  -2.421]])

Wrap-up¶

So we've built $M3P_3M_2P_2M_1P_1A=U$.

M3.dot(P3).dot(M2).dot(P2).dot(M1).dot(P1).dot(A)

array([[ -5.   ,  -8.   ,  -6.   ,  -2.   ],
       [  0.   ,  15.4  ,  -0.2  ,  -4.4  ],
       [  0.   ,   0.   ,  19.052,  -5.857],
       [  0.   ,   0.   ,   0.   ,  -2.421]])

That left factor is anything but lower triangular:

M3.dot(P3).dot(M2).dot(P2).dot(M1).dot(P1)

array([[ 0.   ,  0.   ,  0.   ,  1.   ],
       [ 0.   ,  0.   ,  1.   ,  0.2  ],
       [ 1.   ,  0.   , -0.26 , -0.052],
       [ 0.462,  1.   , -0.094, -0.219]])

Part II¶

Now try the reordering trick:

L3 = M3
L2 = P3.dot(M2).dot(la.inv(P3))
L1 = P3.dot(P2).dot(M1).dot(la.inv(P2)).dot(la.inv(P3))

L3.dot(L2).dot(L1).dot(P3).dot(P2).dot(P1)

array([[ 0.   ,  0.   ,  0.   ,  1.   ],
       [ 0.   ,  0.026,  1.   ,  0.   ],
       [ 1.   , -0.266, -0.26 ,  0.   ],
       [ 0.462,  0.878, -0.094,  0.   ]])

We were promised that all of the Ln are still lower-triangular:

print(L1)
print(L2)
print(L3)

[[ 1.   0.   0.   0. ]
 [ 0.2  1.   0.   0. ]
 [ 0.   0.   1.   0. ]
 [-0.2  0.   0.   1. ]]
[[ 1.     0.     0.     0.   ]
 [ 0.     1.     0.     0.   ]
 [ 0.    -0.26   1.     0.   ]
 [ 0.     0.026  0.     1.   ]]
[[ 1.     0.     0.     0.   ]
 [ 0.     1.     0.     0.   ]
 [ 0.     0.     1.     0.   ]
 [ 0.     0.     0.462  1.   ]]

So their product is, too:

Ltemp = L3.dot(L2).dot(L1)
Ltemp

array([[ 1.   ,  0.   ,  0.   ,  0.   ],
       [ 0.2  ,  1.   ,  0.   ,  0.   ],
       [-0.052, -0.26 ,  1.   ,  0.   ],
       [-0.219, -0.094,  0.462,  1.   ]])

P is still a permutation matrix (but a more complicated one):

P = P3.dot(P2).dot(P1)
P

array([[ 0.,  0.,  0.,  1.],
       [ 0.,  0.,  1.,  0.],
       [ 1.,  0.,  0.,  0.],
       [ 0.,  1.,  0.,  0.]])

So to sum up, we've made:

Ltemp.dot(P).dot(A) - U

array([[ 0.,  0.,  0.,  0.],
       [ 0.,  0.,  0.,  0.],
       [ 0.,  0.,  0.,  0.],
       [ 0.,  0.,  0.,  0.]])

Multiply from the left by Ltemp${}^{-1}$, which is also lower triangular:

L = la.inv(Ltemp)
L

array([[ 1.   ,  0.   ,  0.   ,  0.   ],
       [-0.2  ,  1.   ,  0.   ,  0.   ],
       [ 0.   ,  0.26 ,  1.   ,  0.   ],
       [ 0.2  , -0.026, -0.462,  1.   ]])

P.dot(A) - L.dot(U)

array([[ 0.,  0.,  0.,  0.],
       [ 0.,  0.,  0.,  0.],
       [ 0.,  0.,  0.,  0.],
       [ 0., -0.,  0.,  0.]])