You are on page 1of 3

Elementary matrices and SLn (R)

Let R be a commutative ring. Recall that SLn (R) is the subgroup of GLn (R)
of matrices with determinant 1. A matrix M is an elementary matrix if it is
of the form I +, where I is the identity matrix, and is a matrix with precisely
one nonzero entry, where that entry does not show up on the diagonal. Check
that these elementary matrices generate a group, written En (R). We now prove
Theorem. Let R be a euclidean ring. Then SLn (R) = En (R).
Proof. One direction of this equality is easy. We know that det(I + ) = 1, and
since det(AB) = det(A)det(B), we have En (R) SLn (R).
Observation. Multiplying a matrix by (I + ) is equivalent to adding a multiple
of one column to another column, or adding a multiple of one row to another
row, depending on whether the multiplication is on the left or right. For the rest
of the proof, we will make use of this interpretation of multiplying by elementary
matrices.
We deal first with the case n = 2 (the case n = 1 is trivial). Let M SL2 (R)
be the matrix


a b
c d
Then a = x1 c + a0 , so that we can add x1 c times the bottom row to the
top row, to obtain
 0 0 
a b
c d
Naturally, we can now write c = x2 a0 + c0 , and so add x2 a0 times the top
row to the bottom row, to obtain
 0

a b0
c0 d0
Now, iterate this procedure (call this procedure A). Since we are in a
euclidean ring, this process must terminate after finitely many steps, so that we
either get a 0 in the top left or bottom left positions. We will assume that the 0
is in the top right position. To see that we do not lose generality, assume that
it is in the form


0 y
x z
we can add the bottom row to the top row to obtain


x y+z
x
z

and then subtract the top row from the bottom row


x y+z
0
y
to obtain a 0 in the bottom left column. Thus, we lose nothing by assuming
the zero is in the bottom left column. For future use, call this procedure of
interaching the rows with the zeros in them Procedure B.
For simplicitys sake, write our new matrix as


a b
0 c
Since ac b 0 = 1, we have c = 1/a, so that we can write M as


a
b
0 1/a
Add 1/a times the top row to the bottom row


a
b
1 (1 + b)/a
Now subtract from the right column (1
obtain


a a1
1
1

1+b
a )

times the left column, to

Now subtract one times the right column from the left column, and we get


1 a1
0
1
this matrix is in the form 1 + . Note that we obtained this matrix from our
original


a b
c d
by operations that were equivalent to multiplying by matrices in En (R).
Since all these matrices are invertible, M itself is the product of matrices in the
group En (R), so that M En (R).
We now use induction for the case n 2. For n = 2, we are done. Assume
we have proven it for k 1, and let M SLk (R). By repeating Procedure A,
we can get all but of the entries in the left column to be 0, and by Procedure B,
we can assume the nonzero entry to occur in the diagonal. Iterating, we get an
upper-triangular matrix. By hypothesis, the lower-right k 1 k 1 submatrix
can be put in the form I + via multiplication with elementary matrices, and
2

this only requires adding or subtracting rows and columns indexed 2 k. Thus,
this procedure wil leave the left column unmolested. We obtain


a

[0] (I + )
where [0] is a k 1 column matrix of zeros. Expanding by minors down the
left column, we find that the determinant of this matrix, which must be 1, is a
multiplied by the diagonal entries of I + , which are just 1. Hence a = 1.
What remains to be done is to cancel out the nonzero, nondiagonal elements
in the top row. Firstly, look at the row j in which appears. Let x be the
element in the top row that is in the same column as the 1 element of row j, as
demonstrated by the following sample matrix

1 x y
0 1
0 0 1
Then multiply the top row by x times row j. This cancels out x, but
may alter the other nondiagonal entries in the top row. Now, every other nondiagonal entry y in the top row shares a column with a 1 in the row i, and that
row is empty besides for that 1 element. Thus, subtracting the top row by y
times the ith row cancels out y and changes nothing else. After doing this k 1
times, we are done: the resulting matrix is in the form I + . Moreover, at each
step, we only used operations equivalent to multiplication by elements of the
form (1 + ), which shows that a general element in SLn (R) is in En (R), and
hence (for euclidean rings R),
SLn (R) = En (R)

You might also like