Operations Required in Matrix Elimination
Kundan Chintamaneni
September 23, 2015
1 Introduction
The amount of time needed to solve Ax = b can be estimated by the number of elemen-
tary operations performed: addition/subtractions, multiplications and divisions. For some
matrices, Ax = b is easy to solve. If A is triangular we can find x by substitution, and if
A is the identity, then we know x = b without performing any operations! However, for
practical applications, we are not likely to be so lucky. Given an arbitrary nxn matrix A,
how many operations are required to solve Ax = b?
2 At a Glance
Before we do any calculations, it is helpful to understand where these operations arise.
When are additions, subtractions, multiplications and divisions actually performed?
Ax = b can be solved in two steps, elimination and back substitution. During
elimination, we perform row operations where we replace a row vector ri with ri + crj .
This requires one division to calculate c and one multiplication and one addition for
each nonzero element of crj . During back substitution, we solve equations of the form
ai xi + ai+1 xi+1 + ... + an−1 xn−1 + an xn = bi for xi . This requires one multiplication for
each term ak xk , k 6= i, one subtraction to move it to the right hand side, and finally one
division to isolate xi .
One surprising result is that the total number of addition/subtractions must be the
same as the total number of multiplications. During elimination and back substitution,
whenever a multiplication occurs, it is immediately followed by an addition or a subtraction.
Thus, their final counts must be the same.
1
3 Explicit Calculation
Below we tackle an nxn matrix, let us count the number of operations for a 4x4 matrix:
a1,1 a1,2 a1,3 a1,4 x1 b1
a2,1 a2,2 a2,3 a2,4 x2 b2
a3,1 a3,2 a3,3 a3,4 x3 = b3
a4,1 a4,2 a4,3 a4,4 x4 b4
3.1 Elimination
The first step is to turn A into an upper triangular matrix. We perform row operations
on the augmented matrix:
a1,1 a1,2 a1,3 a1,4 b1
a2,1 a2,2 a2,3 a2,4 b2
M= a3,1 a3,2 a3,3 a3,4 b3
a4,1 a4,2 a4,3 a4,4 b4
To eliminate A2,1 , we replace row r2 with r2 − a2,1 /a1,1 ∗ r1 . Looking at each element of
this row vector, we replace M2,i with M2,i − a2,1 /a1,1 ∗ M1,i for 1 ≤ i ≤ 5. This requires 1
division to calculate (a2,1 /a1,1 ), 1 multiplication per element to calculate (a2,1 /a1,1 ) ∗ M1,i ,
and 1 subtraction per element to calculate M2,i − (a2,1 /a1,1 ∗ M1,i ). In total, this takes
1 division, 5 multiplications and 5 additions. However, we can improve this. By the
way we choose to eliminate, we know that M2,1 will be 0 without having to calculate
M2,1 − (a2,1 /a1,1 ∗ M1,1 ). Thus, this takes 1 division, 4 multiplications and 4 additions
Eliminating A3,1 and A4,1 also requires the same number of operations. These three
eliminations then take 3 divisions, 12 multiplications and 12 additions and create the
matrix:
a1,1 a1,2 a1,3 a1,4 b1
0 a02,2 a02,3 a02,4 b02
M0 = 0 a03,2 a03,3 a03,4 b03
0 a04,2 a04,3 a04,4 b04
Continuing elimination on this matrix is the same as performing elimination on the 3x3
matrix: 0
a2,2 a02,3 a02,4 b02
a03,2 a03,3 a03,4 b03
a04,2 a04,3 a04,4 b04
By the same reasoning as above, this will take 2 divisions, 6 multiplications and 6
additions and will leave the 2x2 matrix:
00
a3,3 a003,4 b003
a004,3 a004,4 b004
2
Finally, eliminating this matrix takes 1 division, 2 multiplications and 2 additions. All
together, it takes 6 divisions, 20 multiplications and 20 additions to get the upper triangular
augmented matrix:
a1,1 a1,2 a1,3 a1,4 b1
0 a02,2 a02,3 a02,4 b02
U=
0 a003,3 a003,4 b003
0
0 0 0 a000 000
4,4 b4
3.2 Back Substitution
To finish solving Ax = b, we need to calculate x. Starting from the bottom row, we have:
a000 000
4,4 x4 = b4
This can be solved with 1 division. The next row is:
a003,3 x3 + a003,4 x4 = b003
This requires 1 multiplication to calculate a003,4 x4 , 1 subtraction to move that term to right,
and 1 division to isolate x3 . Similarly, the second equation:
a02,2 x2 + a02,3 x3 + a02,4 x4 = b02
requires 2 multiplications, 2 subtractions and 1 division, and the first equation:
a1,1 x1 + a1,2 x2 + a1,3 x3 + a1,4 x4 = b1
requires 3 multiplications, 3 subtractions and 1 division for a total of 4 divisions, 6 multi-
plications and 6 addition/subtractions.
Bringing it all together, solving Ax = b for a 4x4 matrix takes a total of 10 divisions,
26 multiplications and 26 additions.
3.3 General Case
We can apply the same reasoning for an nxn matrix A. During elimination, the first row
will require [n − 1, n(n − 1), n(n − 1)] divisions, multiplications and addition/subtractions
respectively. The second row will require [n − 2, (n − 1)(n − 2), (n − 1)(n − 2)], the third
row [n − 2, (n − 2)(n − 3), (n − 2)(n − 3)], and in general, the ith row requires [n − i, (n +
1 − i)(n − i), (n + 1 − i)(n − i)] operations.
During back substitution, the last row requires [1,0,0] operations, the second to last
row requires [1, 1, 1] operations, the third to last requires [1, 2, 2] operations, etc. The ith
row from the bottom requires [1, i − 1, i − 1] operations.
3
There are n rows to the matrix, so i can range from 1 to n. The total number of
operations is the sum:
n
X n
X
[n − i, (n + 1 − i)(n − i), (n + 1 − i)(n − i)] + [1, i − 1, i − 1]
i=1 i=1
n
X
= [n − i + 1, (n + 1 − i)(n − i) + (i − 1), (n + 1 − i)(n − i) + (i − 1)]
i=1
n
X
= [−i + (n + 1), i2 − (2n)i + (n2 + n − 1), i2 − (2n)i + (n2 + n − 1)]
i=1
=[−n(n + 1)/2 + n(n + 1), n(n + 1)(2n + 1)/6 − (2n)(n(n + 1)/2) + n(n2 + n − 1),
n(n + 1)(2n + 1)/6 − (2n)(n(n + 1)/2) + n(n2 + n − 1)]
=[n(n + 1)/2, (2n3 + 3n2 − 5n)/6, (2n3 + 3n2 − 5n)/6]
In conclusion, solving Ax = b takes at most n(n + 1)/2 divisions, (2n3 + 3n2 − 5n)/6
multiplications, and (2n3 + 3n2 − 5n)/6 addition/subtractions.
