Gradient Covariance

Search

More Tech Stuff:

Indexing Books: Lessons in Language Computations

Client-Side Frame Manipulation Inside the Microsoft Internet Explorer Object Model with Visual Basic .NET

Replacing a PC power supply

Constructing a Mandelbrot Set Based Logo with Visual Basic.NET and Fireworks


page 3 of 4

Because I want to use some standard matrix algebra formulas, the above equations can be recast in general form. This introduces the notation a sub i j for constants arranged in matrices. The exercise could be redone for the general case, using the a sub i j, an exercise left to the reader.

x1 bar = ax's

a2 bar = ax's

For the a sub i j arranged in a matrix, i corresponds to the rows and j to the columns. We call this the (coordinate) transformation matrix T:

T = [a's] = transformation example

We will later need to express the x sub i's in terms of  the x bar sub i's. This can be done with high school algebra techniques, but matrix techniques were designed to make this easier, especially when problems involve three or more dimensions. We will need to invert T, which we call T inverse. We’ll need the determinant of the matrix, which is easy for a 2 x 2 matrix:

determinate of T

Then we will use the known formula for inverting 2 x 2 matrices [See also Inverting a Matrix]

inverse of T calculated

Don’t believe me. The first time I did this I made a subtraction error. The way I found there was an error was testing to see if T inverse gets T back for us (the a sub i j are not from T inverse, not the originals).

T from inverse of T

This encapsulates the process of transformations between coordinate systems, which are bidirectional.

Using T inverse we can now also write out the x sub i in terms of the x bar i's

x sub 1 = bar x's

x sub 2 = bar x's

By subtracting the second equation from the first you can obtain x2 bar = x's. Then substitute that back in and you get x1 bar = x's, verifying that the transformation equations are consistent.

The Gradient in the New Coordinate System

Our scalar values are expressed as a function F of the x sub i. Let us substitute in the x bar i's to get F bar

F bar original

F bar = substituted

F bar = all terms

F bar = in terms of x bars

Now by taking partial derivatives we can get the gradient components in the new (barred) coordinate system:

Del F bar 1

del F bar 2

Next up: Covariance

continued Gradient Covariance page 4