# Simplex ¶

Go back

This method is quite well-known so you may look for some videos on the net since that might help you understand. We are making a table, and updating it until we have our solution.

## Standard form ¶

You need to rewrite your constraints, you can only have equations (no inequalities). If you don't have an equation, you will have to introduce what we call slack variables (variables artificielles/d'écart/de bruit in French)

• > (or equals): - slack variables
• < (or equals): + slack variables

For instance $x < 5$ becomes $x + S_1 = 5$ with $S_1$ a slack variable.

Note: if you added a negative slack variable, because of >, then you must use the 2-phases algorithm.

Create a matrix Ax = b, with b the results of your equations and A the coefficients before your variables in each equation.

base X Y ... $S_1$ ... $S_n$ b
$S_1$ $a_{11}$ $a_{12}$ ... 1 0 0 $b_1$
... ... ... ... ... ... ... ...
$S_n$ $a_{n1}$ $a_{n2}$ ... 0 0 1 $b_n$
$c_{1}$ $c_{2}$ ... ... ... ... $R=0$

The last line is the coefficient of each variable in the function f. And 0 is the result of the optimization since we haven't started yet.

## Minimization ¶

If you are asked to minimize then

• take the column with the smallest c
• if this column only have negatives values then end
• we want the row having the lowest ratio $S_i = b_i / a_{ij}$ so evaluate all the ratios for your column and find the row.

Now that you got your column and row, you will have to put 1 inside and 0 in all the others values of the diagonal. Since that's a matrix, simply use GAUSS.

Once you did, if set a 1 in $a_{11}$ then replace $S_{1}$ (i=1) in the base column by the variable in the first (j=1) column so X.

Stop? When all the values in the last line (reduced costs) are positives. The result is -R.

## Maximization ¶

Same a minimization, but take the column with the biggest c.

Stop? When all the values in the last line (reduced costs) are negatives. The result is -R.

## 2-phases ¶

You will have to do 2 simplexes. You need to add slack variables (that I'm calling A this time) on each equation with a negative value. In your simplex table, the A variables are in the base and you need to remove them.

Once you did remove them, then you can start using the table you got as the starting table.

## Exercise ¶

Use the simplex method to solve

$\max z = 2x + 3y\ \ s.c. \begin{cases} x + y \le 1\\ x + 4y \le 2\\ x \ge 0\\ y \ge 0\\ \end{cases}$

We rewrite our constraints, so we have the following standard form

$\begin{cases} x + y + e_1 = 1\\ x + 4y + e_2 = 2\\ x \ge 0\\ y \ge 0\\ \end{cases}$

base x y $e_1$ $e_2$ b
$e_1$ 1 1 1 0 1
$e_2$ 1 4 0 1 2
2 3 $R=0$

And we are starting,

• The highest coefficient is 3 (second column)
• The highest row is $min(1/1, 2/4)=2/4$ (second line)
• we are clearing the second column
• and we will replace e2 in the base by y
base x y $e_1$ $e_2$ b
$e_1$ 1 1 1 0 1
$e_2$ 1/4 1 0 1/4 2/4
2 3 $R=0$
base x y $e_1$ $e_2$ b
$e_1$ 3/4 0 1 -1/4 2/4
y 1/4 1 0 1/4 2/4
5/4 0 0 -3/4 $R=-6/4$

Then again

• The highest coefficient is 5/4 (first column)
• The highest row is $min((2/4)/(3/4), (2/4)/(1/4))=(2/4)/(3/4)$ (first line)
• we are clearing the first column
• and we will replace e1 in the base by x
base x y $e_1$ $e_2$ b
$e_1$ 1 0 4/3 -1/3 2/3
y 1/4 1 0 1/4 2/4
5/4 0 0 -3/4 $R=-6/4$
base x y $e_1$ $e_2$ b
x 1 0 4/3 -1/3 2/3
y 0 1 -4/12 4/12 4/12
0 0 -20/12 -4/12 $R=-28/12$

And with a bit of cleaning

base x y $e_1$ $e_2$ b
x 1 0 4/3 -1/3 2/3
y 0 1 -1/3 1/3 1/3
0 0 -5/3 -1/3 $R=-7/3$

All of our slack variables are negatives so we are good. The solution is

$\begin{cases} x = 2/3\\ y = 1/3\\ e_1 = 0\\ e_2 = 0\\ z = 7/3\\ \end{cases}$