The Simplex Method1

The Simplex Method consists of the flowing steps:

1. Find a feasible basis (phase 1)

2. Compute the reduce costs for
𝑢 =𝑐 −𝒂 ∙𝒚
where
𝒚 ≡ (𝑨 )𝑻 . 𝒄
For all 𝑥 not in the basis

3. If 𝑢 ≥ 0, for all k, the solu on is op mal. Otherwise, chose k corresponding to the

most nega ve 𝑢 as the entering column.
4. Choose the leaving column 𝑖 from the minimum ra o test
Compute
𝒙𝑩 = (𝑨 )𝑻 . 𝒃, 𝒘 = (𝑨 )𝑻 . 𝒂𝒌

For each component 𝑤 > 0, compute the ra o 𝑥 ⁄𝑤 . Choose 𝑖 that corresponds to the
smallest such ra o.

5. Exchange columns 𝑖 and 𝑘 and go back to step 2.

Example: Consider the following problem

Objec ve func on

𝑀𝑖𝑛𝑖𝑚𝑖𝑧𝑒 𝑧 = −40𝑥 − 60𝑥 (1)

Constrains: the 𝑥’s are non-nega ve and

2𝑥 + 𝑥 ≤ 70 (2)

𝑥 + 𝑥 ≥ 40 (3)

𝑥 + 3𝑥 = 90 (4)

Note

𝑥 + 𝑥 ≥ 40
Can be converted to

−𝑥 − 𝑥 ≤ −40

1
Numerical Recipes 3rd Edi on: The Art of Scien fic Compu ng, Press, William H.; Teukolsky, Saul A.;
Ve erling, William T.; Flannery, Brian P.
Now

2𝑥 + 𝑥 ≤ 70 (5)

−𝑥 − 𝑥 ≤ −40 (6)

𝑥 + 3𝑥 = 90 7
Or, adding the slack variables

2𝑥 + 𝑥 + 𝑥 = 70 8
−𝑥 − 𝑥 + 𝑥 = −40 (9)

𝑥 + 3𝑥 + 𝑥 = 90 (10)

This can be wri en

𝑥
⎧𝑥 ⎫
2 1 1 0 0 ⎪ ⎪ 70
−1 −1 0 1 0 𝑥 = −40
1 3 0 0 1 ⎨
⎪
𝑥 ⎬
⎪ 90
⎩𝑥 ⎭
2 1 1 0 0
𝒂 = −1 , 𝒂 = −1 , 𝒂 = 0 , 𝒂 = 1 𝒂 = 0
1 3 0 0 1

Or, permuta ng columns

𝑥
⎧ ⎫
1 0 0 2 1 ⎪𝑥 ⎪ 70
0 1 0 −1 −1 𝑥 = −40
0 0 1 1 3 ⎨ 𝑥 ⎬
⎪ ⎪ 90
⎩𝑥 ⎭
or

𝑨𝒙 = 𝒃
Par on A

𝑨 = [𝑨 |𝑨 ]
where
1 0 0 2 1
𝑨 = 0 1 0 , 𝑨 −1 −1
0 0 1 1 3

First, replace 𝑥 coefficients with 𝑥 coefficients (𝑥 coefficients would work just as well)
 1 0 1
𝑨 = 0 1 −1
0 0 3
The inverse is
1 0 −1⁄3
𝑨 = 0 1 1⁄3
0 0 1⁄3
And
𝑥 1 0 −1⁄3 70 40
𝒙 = 𝑥 =𝑨 ∙ 𝒃 = 0 1 1⁄3 −40 = −10
𝑥 0 0 1⁄3 90 30
This solu on is not feasible because 𝑥 ≱ 0.

Now we replace the objec ve func on with a so-called auxiliary objec ve func on

𝑧 = 𝒄 . 𝒙 = −𝑥
i.e,

𝒄 = {0 −1 0}
We now compute the reduced costs using

𝑢 =𝑐 −𝒂 ∙𝒚
where

𝒚 ≡ (𝑨 )𝑻 . 𝒄

Here, 𝑘 = 1,

𝑢 =𝑐 −𝒂 ∙𝒚
and
1 0 0 0 0
𝒚= 0 1 0 . −1 = −1
−1⁄3 1⁄3 1⁄3 0 −1⁄3

0
𝑢 = 0 − 𝒂 ∙ 𝒚 = 0 − {2 1 1}. −1 = − 2⁄3
−1⁄3
Because 𝑢 < 0, 𝑥 should enter the computa on and another 𝑥 should leave. To determine which
variable should leave we perform a minimum ra o test using
1 0 −1⁄3 2 5 ⁄3
𝑨 ∙ 𝒂𝟏 = 0 1 1⁄3 −1 = −2⁄3
0 0 1⁄3 1 1 ⁄3
Ra os
40 −10 30
= 24, = 15, = 90
5 ⁄3 − 2⁄3 1 ⁄3
The middle number is associated with 𝑥 and is the minimum ra o so 𝑥 goes out.

Now the basic variables are 𝑥 , 𝑥 , 𝑥

Now we repeat the procedure

1 2 1 1 5⁄2 1 ⁄2
𝑨 = 0 −1 −1 , 𝑨 = 0 −3⁄2 −1⁄2
0 1 3 0 1⁄2 1 ⁄2
𝑥 1 5⁄4 1 ⁄2 70 15
𝒙 = 𝑥 =𝑨 ∙𝒃= 0 −1⁄4 −1⁄2 −40 = 15
𝑥 0 1⁄2 1 ⁄2 90 25
All variables are posi ve, so the solu on is feasible, and we enter phase II

Now
𝑐 0
𝒄 = 𝑐 = −40
𝑐 −60
1 0 0 0 0
𝒚 = (𝑨 )𝑻 .𝒄 = 5 ⁄ 2 −3⁄2 1⁄2 . −40 = 30
1⁄2 −1⁄2 1⁄2 −60 −10
0
𝑢 = 𝑐 − 𝒂 ∙ 𝒚 = 0 − {0 1 0} ∙ 30 = −30
−10
This result says that 𝑥 should reenter.

Find which should leave

1 5⁄4 1 ⁄2 0 5 ⁄2
𝒘=𝑨 ∙ 𝒂 = 0 −1⁄4 −1 2 1 = −3⁄2
⁄
0 1⁄2 1 ⁄2 0 1 ⁄2
Ra os (only consider posi ve 𝑤 )
15 25
= 6, = 50
5 ⁄2 1 ⁄2
Hence, 𝑥 being associated with the smallest ra o, it leaves and new solu on is 𝑥 , 𝑥 , 𝑥

Now we repeat the procedure

0 2 1 0.4 1 0.2
𝑨 = 1 −1 −1 , 𝑨 = 0.6 0 −0.2
0 1 3 −0.2 0 0.4
 𝑥 0.4 1 0.2 70 6
𝒙 = 𝑥 =𝑨 ∙𝒃= 0.6 0 −0.2 −40 = 24
𝑥 −0.2 0 0.4 90 22

𝑐 0
𝒄 = 𝑐 = −40
𝑐 −60
0.4 0.6 −0.2 0 −12
𝒚 = (𝑨 )𝑻 . 𝒄 = 1 0 0 . −40 = 0
0.2 −0.2 0.4 −60 −16
−12
𝑢 = 𝑐 − 𝒂 ∙ 𝒚 = 0 − {1 0 0} ∙ 0 = 12
−16
As 𝑢 > 0, we have found the op mum solu on: 𝑥 = 24, 𝑥 = 22, and 𝑥 = 6

Constrains are sa sfied

2𝑥 + 𝑥 = 70 ≤ 70
𝑥 + 𝑥 = 46 ≥ 40
𝑥 + 3𝑥 = 90 = 90
𝑥 >0
𝑥 >0