MATH
21 B
Mathematics Math21b Spring 2009
Linear Algebra and Differential Equations
Exhibit: Duality in Linear Algebra
Course Head: Oliver Knill
Office: SciCtr 434


Row and column picture

One of the most important insights of linear algebra is obtained from the role of columns and rows. The columns are helpful to describe the image, the rows helpful to describe the kernel.

The column picture tells that A x = x1 v1 + ... + xn vn is a linear combination of columns. The row picture tells that A x = 0 means that x is perpendicular to the rows of A. One can state this as
 ker(A)T = im (AT)
which relates the transposed matrix AT of the matrix A with the orthogonal complement VT of the column space V.

Data

Duality reflects also in statistics If you have two data sets like x = (1,3,1,-2,3) and y = (2,1,-5,5,1). Then there are two ways to describe these data. The first is to see the data as two vectors in 5 dimensional space. The cosine of the angle between X = x-E[x] and Y = y-E[y] is called the correlation of the data. The second dual picture is to visualize the data points (1,2),(3,1),(1,-5),(-2,5),(3,1) in the plane. The regression line passes through the point E[x],E[y] and has the slope m = cos(alpha) ||Y||/||X|| = (X.Y)/||X||2 because Y-m X is minimized if m X is the projection of Y onto X Again, if the data are written as a matrix, then the two pictures correspond to taking the transpose of the matrix.

A = |  1  2 | 
    |  3  1 |
    |  1 -5 |
    | -2  5 |
    |  3  1 |
AT = 
    | 1 3  1 -2 3 |
    | 2 1 -5  5 1 |


Please send questions and comments to math21b@fas.harvard.edu
Math21b (Exam Group 1)| Oliver Knill | Spring 2009 | Department of Mathematics | Faculty of Art and Sciences | Harvard University