Академический Документы
Профессиональный Документы
Культура Документы
Row reduction (or Gauss-Jordan elimination) is the process of using row operations to reduce a
matrix to row reduced echelon form. This procedure is used to solve systems of linear equations, invert
matrices, compute determinants, and do many other things.
There are three kinds of row operations. (Actually, there is some redundancy here --- you can get away
with two of them.)
(a) You may swap two rows.
Here is a swap of rows 2 and 3. I'll denote it by
Notice that row 1 was not affected by this operation. Likewise, if you do
changes and row 31 does not.
Example.
, row 17
In each case, tell whether the operation is a valid row operation. If it is, say what it does (in
words).
(a)
This isn't a valid row operation. You can't add or subtract a number from the elements in a row.
(c)
This adds
(d)
.
times row 17 to row 3 (and replaces row 3 with the result). Row 17 is not changed.
.
This isn't a valid row operation, though you could accomplish it using two row operations: First, multiply
row 6 by 5; next, add 11 times row 2 to the new row 6.
(e)
and
This isn't a valid row operation. The only row operation that changes two rows at once is swapping two
rows. And in general, you can't accomplish both
and
at once using a sequence
of valid row operations.
Matrices can be used to represent systems of linear equations. Row operations are intended to mimic the
algebraic operations you use to solve a system. Row-reduced echelon form corresponds to the "solved
form" of a system.
A matrix is in row reduced echelon form if the following conditions are satisfied:
(a) The first nonzero element in each row (if any) is a "1" (a leading coefficient).
(b) Each leading coefficient is the only nonzero element in its column.
(c) All the all-zero rows (if any) are at the bottom of the matrix.
(d) The leading coefficients form a "stairstep pattern" from northwest to southeast:
, ... .
Here some more matrices in row-reduced echelon form. Notice the "stairstep pattern" made by the leading
coefficients. The *'s indicate that the numbers in those positions can be anything.
Example.
( Row-reduced echelon form) This matrix is not in row reduced echelon form:
The leading coefficient in row 3 is not the only nonzero element in its column.
This matrix is not in row reduced echelon form:
The leading coefficient in row 2 is not the only nonzero element in its column.
This matrix is not in row reduced echelon form:
The leading coefficients do not form a "stairstep pattern" from northwest to southeast.
The following matrices are in row-reduced echelon form. You should go through the definition and check
that all the properties are satisfied.
Row reduction is the process of using row operations to transform a matrix into a row reduced echelon
matrix. As the algorithm proceeds, you move in stairstep fashion through different positions in the matrix.
In the description below, when I say that the current position is
, I mean that your current location
is in row i and column j. The current position refers to a location, not the element at that location (which
I'll sometimes call the current element). The current row means the row of the matrix containing the
current position and the current column means the column of the matrix containing the current position.
Trying to learn to row reduce by following the steps below is pretty tedious, and most people will want to
learn by looking at examples. Whatever you do, you have to get your hands dirty by row reducing some
matrices to get a feel for the algorithm.
Algorithm: Row Reducing a Matrix
Step 2. Test the element at the current position. If it's nonzero, go to Step 2(a); if it's 0, go to Step 2(b).
(a)
The element in the current position is nonzero. So I divide the first row by 2:
Next, I subtract 3 times row 1 from row 2, and I add row 1 to row 3. This makes the other elements in the
first column equal to 0.
Finally, I move the current position to the next row and the next column and return to the start of Step 2:
(b)
The element in the current position is 0. I look below it and see a nonzero element in the same column in
row 3. So I swap row 1 and row 3; the current position remains the same, and I return to the start of Step
2.
(c)
The element in the current position is 0. There are no nonzero elements below it in the same column. I
don't perform any row operations; I just move the current position to the next column (in the same row)
and return to the start of Step 2:
The cases in Step 2 cover all the possibilities, and in each case, you perform a finite number of row
operations (no larger than the number of rows in the matrix, plus one) before you move the current
position. Since you're always moving the current position to the right or to the right and down, and since
the matrix has only finitely many rows and columns, you must eventually reach the edge of the matrix and
the algorithm will terminate.
As for the second question, I'll give an informal argument using the matrix with the "stairstep" path
pictured above.
First, if you moved the current position down and to the right, the previous current element was a 1, and
every other element in its column must be 0. In the matrix with the "stairstep" path I gave above, this
means that each spot where a curved arrow starts must be a 1, and all the other elements in the column
with a 1 must be 0. Hence, the matrix must look like this:
Notice that the first operation --- swapping the first two rows --- isn't really in accord with the algorithm.
So I cheated a little, but not too much: It accomplished the aim of creating a "1" in the
position, and
following the algorithm by dividing the given first row by 3 would have created a lot of ugly fractions.
Example.
Row reduction is used to solve systems of linear equations. Consider the following system of
linear equations over
:
Example.
Thus,
and
It's possible for a system of linear equations to have no solutions. Such a system is said to be
inconsistent. Consider the following system of equations over
:
Example.
Consider the case of a system over the real numbers with matrix
a and c correspond to leading coefficients. b, d, and e are called free variables. In this case, you get a
parametrized solution:
Each assignment of numbers to the parameters b, d, and e produces a solution. For example, if
, and
,
Set
):
A system of linear equations over the real numbers can have no solutions, a unique solution, or
infinitely many solutions. (That is, such a system can't have exactly 3 solutions.)
Remark.
( Inverting a matrix) I'll discuss matrix inversion in more detail later. However, it's easy to
describe how row reduction provides a systematic way to find the inverse of a matrix.
Example.
To invert a matrix, adjoin a copy of the identity matrix and row reduce the augmented matrix. When the
block corresponding the original matrix becomes the identity, the block corresponding to the identity will
have become the inverse.
To invert
The inverse is
Think of the equations as a system. Write down the coefficient matrix and row reduce:
The solution is