Main Content

pinv

Moore-Penrose pseudoinverse

Description

example

B = pinv(A) returns the Moore-Penrose pseudoinverse of matrix A.

B = pinv(A,tol) specifies a value for the tolerance. pinv treats singular values of A that are less than or equal to the tolerance as zero.

Examples

collapse all

Compare solutions to a system of linear equations obtained by backslash (\), pinv, and lsqminnorm.

If a rectangular coefficient matrix A is of low rank, then the least-squares problem of minimizing norm(A*x-b) has infinitely many solutions. Two solutions are returned by x1 = A\b and x2 = pinv(A)*b. The distinguishing properties of these solutions are that x1 has only rank(A) nonzero components, and norm(x2) is smaller than for any other solution.

Create an 8-by-6 matrix that has rank(A) = 3.

A = magic(8); 
A = A(:,1:6) 
A = 8×6

    64     2     3    61    60     6
     9    55    54    12    13    51
    17    47    46    20    21    43
    40    26    27    37    36    30
    32    34    35    29    28    38
    41    23    22    44    45    19
    49    15    14    52    53    11
     8    58    59     5     4    62

Create a vector for the right side of the system of equations.

b = 260*ones(8,1)
b = 8×1

   260
   260
   260
   260
   260
   260
   260
   260

The number chosen for the right side, 260, is the value of the 8-by-8 magic sum for A. If A were still an 8-by-8 matrix, then one solution for x would be a vector of 1s. With only six columns, a solution exists because the equations are still consistent, but the solution is not all 1s. Because the matrix is of low rank, there are infinitely many solutions.

Solve for two of the solutions using backslash and pinv.

x1 = A\b
Warning: Rank deficient, rank = 3, tol =  1.882938e-13.
x1 = 6×1

    3.0000
    4.0000
         0
         0
    1.0000
         0

x2 = pinv(A)*b
x2 = 6×1

    1.1538
    1.4615
    1.3846
    1.3846
    1.4615
    1.1538

Both of these solutions are exact, in the sense that norm(A*x1-b) and norm(A*x2-b) are on the order of round-off error. The solution x1 is special because it has only three nonzero elements. The solution x2 is special because norm(x2) is smaller than it is for any other solution, including norm(x1).

norm(x1)
ans = 5.0990
norm(x2)
ans = 3.2817

Using lsqminnorm to compute the least-squares solution of this problem produces the same solution as using pinv does. lsqminnorm(A,b) is typically more efficient than pinv(A)*b.

x3 = lsqminnorm(A,b)
x3 = 6×1

    1.1538
    1.4615
    1.3846
    1.3846
    1.4615
    1.1538

norm(x3)
ans = 3.2817

Input Arguments

collapse all

Input matrix.

Data Types: single | double
Complex Number Support: Yes

Singular value tolerance, specified as a scalar. pinv treats singular values that are less than or equal to tol as zeros during the computation of the pseudoinverse.

The default tolerance is max(size(A))*eps(norm(A)).

Example: pinv(A,1e-4)

More About

collapse all

Moore-Penrose Pseudoinverse

The Moore-Penrose pseudoinverse is a matrix that can act as a partial replacement for the matrix inverse in cases where it does not exist. This matrix is frequently used to solve a system of linear equations when the system does not have a unique solution or has many solutions.

For any matrix A, the pseudoinverse B exists, is unique, and has the same dimensions as A'. If A is square and not singular, then pinv(A) is simply an expensive way to compute inv(A). However, if A is not square, or is square and singular, then inv(A) does not exist. In these cases, pinv(A) has some (but not all) of the properties of inv(A):

1.ABA=A2.BAB=B3.(AB)*=AB4.(BA)*=BA

Here, AB and BA are Hermitian. The pseudoinverse computation is based on svd(A). The calculation treats singular values less than or equal to tol as zero.

Tips

  • You can replace most uses of pinv applied to a vector b, as in pinv(A)*b, with lsqminnorm(A,b) to get the minimum-norm least-squares solution of a system of linear equations. For example, in Solve System of Linear Equations Using Pseudoinverse, lsqminnorm produces the same solution as pinv does. lsqminnorm is generally more efficient than pinv because lsqminnorm uses the complete orthogonal decomposition of A to find its low-rank approximation and applies its factors to b. In contrast, pinv uses singular value decomposition to explicitly form the pseudoinverse of A that you then must multiply by b. lsqminnorm also supports sparse matrices.

Algorithms

pinv uses singular value decomposition to form the pseudoinverse of A. Singular values along the diagonal of S that are less than or equal to tol are treated as zeros, and the representation of A becomes:

A=USV*=[U1U2][S1000][V1V2]*A=U1S1V1*.

The pseudoinverse of A is then equal to:

B=V1S11U1*.

Extended Capabilities

Version History

Introduced before R2006a

expand all