Introduction

Operations research has been particularly successful in two areas of decision analysis:

(i) Deterministic optimization: optimization of problems involving many variables when the
outcome of the decisions can be predicted with certainty, and

(ii) Stochastic optimization: the analysis of situations involving a few variables when the
outcome of the decisions cannot be predicted with certainty.

Deterministic Solution:

The optimization algorithm determines the best solution given the parameters of the model. The
modeling and decision analysis process is illustrated in Fig. 1 where the shape labeled situation
represents the real problem under consideration. Various assumptions and abstractions are applied,
including the assumption of deterministic information, to obtain a mathematical model. The model
is input to a computer where an algorithm determines the optimal decision, represented by the
vector x.

SITUATION Model Decision = x

Algorithm
23-Dec-20 Dr. Mohammad Babul Hasan
Figure 1: The deterministic approach to decision making
Deterministic Model: LP

Max / Min Z  cT x

subject to

Ax  B
x0

 a11 a12 ....... a1 n   b1   x1 

     
 a 21 a 22 ..... a2n 
 b2   x2 
Here A   . . . .  B .  , x . 
     
 . . . .   .   . 
   bm   xn 
 a m1 a m 2 ....... a mn 

23-Dec-20 Dr. Mohammad Babul Hasan

 Disadvantage of Deterministic model:

• The fault with this approach is that the decision x is optimum for the model and not the
situation.
• It is usually readily apparent to the manager with the task of implementing the decision, that
x is not at all appropriate for application to the situation.
• The primary reason for this lies in the assumption of deterministic parameters.
• When the situation involves uncertainty or risk, the kind of decision taken is quite different
than if it does not.
• Real decision makers hedge against various possible futures.

23-Dec-20 Dr. Mohammad Babul Hasan

 Example:

Imagine a company that provides energy to households.

The company is responsible for delivering energy to households based on their demand.

The problem could be solved as an LP with constraints based on demand from households.
But
future demand of households is not always known and is likely dependent on factors such
as the weather and time of year.

Therefore, there is uncertainty and the basic LP model will not be sufficient.

23-Dec-20 Dr. Mohammad Babul Hasan

 Stochastic program (SP):
Stochastic programs (linear, integer, Nonlinear) are mathematical programs where some of
the data incorporated into the objective function or constraints is uncertain.
Uncertainty is usually characterized by a probability distribution on the parameters.
Although the uncertainty is rigorously defined, in practice it can range in detail from a few
scenarios to specific and precise joint probability distributions.
The outcomes are generally described in terms of elements x of a set X. Where X can be,
for example, the set of possible demands over the next few months.
When some of the data is random, then solutions and the optimal objective value to the
optimization problem are themselves random.
A distribution of optimal decisions is generally un implementable. Ideally, we would like
one decision and one optimal objective value.

23-Dec-20 Dr. Mohammad Babul Hasan

Scenario based Stochastic Models:

Max / Min Z  cT x

subject to

Ax  B
T   x  H  
x0

 a11 a12 ....... a1 n   b1   a11   a12   ....... a1n     x1 

       
 a 21 a 22 ..... a2n 
 b2   a 21   a 22   ..... a 2 n   
 x2 
Here A   . . . .  B   .  , T     . . . .  , x . 
       
 . . . .   .   . . . .   . 
   
 a m1 am2 ....... a mn   bm   a l 1   a l 2   ....... a ln     xn 

 h1   
 
 h 2   
H     . 
 
 . 
 hl   
23-Dec-20 Dr. Mohammad Babul Hasan
This problem can be formulated according to:
Max / Min Z  c T x  Q x 

subject to

Ax  B
T   x  H  
x  x1 , x 2 , .... x n   0

Here Q  x   E  Q  x , 

Q  x,   min q   y   where W   y    H    T  x and  is a scenario or possible

T

outcome or random variable.

23-Dec-20 Dr. Mohammad Babul Hasan

 General Stochastic Programming

Max/ Min Z = Deterministic part + Stochastic part

s.t. Deterministic constraints

Stochastic constraints

23-Dec-20 Dr. Mohammad Babul Hasan

Numerical Example

Suppose we have the following optimization problem:

This is a simple linear optimization problem with optimal solution

set .
Now assume that variables and are uncertain and that there are three different
scenarios, for the values of and , each occurring with a probability of
1/3. This new problem involves uncertainty and is thus considered a stochastic problem. We
must now partition and into and respectively.

23-Dec-20 Dr. Mohammad Babul Hasan

 3-Scenarios represents 3 independent deterministic LP Models

Scenario-1 Max Z  1 .5 * 5 x1  10 x 2   4 y1  6 y 2 OBJECTIVE FUNCTION VALUE

subject to 1) 30.00000
y 1  y 2  10 , VARIABLE VALUE RC
X1 9.333333 0.000000
1 .5 *  x1  3 x 2   14 X2 0.000000 7.500000
Y1 10.000000 0.000000
x1 , x 2 , y 1 , y 2  0
Y2 0.000000 2.000000
Scenario-2 Max Z  1 * 5 x1  10 x 2   4 y1  6 y 2 OBJECTIVE FUNCTION VALUE
subject to 1) 30.00000

y 1  y 2  10 , VARIABLE VALUE RC
1 *  x1  3 x 2   14 X1 14.0000 0.000
X2 0.000000 5.000000
x1 , x 2 , y 1 , y 2  0 Y1 10.000000 0.000000
Y2 0.000000 2.000000
Scenario-3 Max Z  0 .7 * 5 x1  10 x 2   4 y 1  6 y 2 OBJECTIVE FUNCTION VALUE
subject to 1) 30.00000
y1  y 2  10 ,
VARIABLE VALUE RC
0 .7 *  x1  3 x 2   14 X1 20.000000 0.000000
X2 0.000000 3.500000
x1 , x 2 , y 1 , y 2  0
Y1 10.000000 0.000000
Y2 0.000000 2.000000

23-Dec-20 Dr. Mohammad Babul Hasan

 Deterministic equivalent of the General stochastic LP problem

1 1 1
Max Z  1 .5 * 5 x11  10 x 21   1 * 5 x12  10 x 22   0 .7 *  x13  x 23   4 y1  6 y 2
3 3 3

subject to

y 1  y 2  10 ,
1 .5 *  x11  3 x 21   14
1 *  x12  3 x 22   14
0 .7 *  x13  3 x 23   14
x11 , x12 , x13 , x 21 , x 22 , x 23 , y 1 , y 2  0

23-Dec-20 Dr. Mohammad Babul Hasan

Once turned into the discrete version, the problem is reformulated as shown below and can
be solved once again using linear programming.

LP OPTIMUM FOUND AT STEP 3

OBJECTIVE FUNCTION VALUE

1) 30.00533

VARIABLE VALUE REDUCED COST

The new optimal solution is . X11 9.333333 0.000000
X21 0.000000 2.500000
X12 14.000000 0.000000
X22 0.000000 1.671000
X13 20.000000 0.000000
X23 0.000000 1.166100
Y1 10.000000 0.000000
Y2 0.000000 2.000000

23-Dec-20 Dr. Mohammad Babul Hasan

 What’s it worth?: expected value of perfect information(EVPI)
With perfect information,
* Scenario-1: (x1 = 9.33, y1 = 10), Profit: $30
* Scenario-2: (x1 = 14, y1 = 10), Profit: $30
* Scenario-3: (x1 = 20, y1 = 10), Profit: $30
Assuming each of these scenarios occurs with probability 1/3, his long run average profit would be
(1/3)(30) + (1/3)(30) + (1/3)(30) = 30

With his (optimal) “here-and-now” decision of (y1 = 10, y2 = 0), the long run profit is $ 30.005
This difference (30-30.005) = - 0.005 is the expected value of perfect information(EVPI)

EVPI: Expected value of perfect information(EVPI) measures the value of knowing the future with certainty.

In some situations, where more information might be available through more extensive forecasting, sampling, or
exploration. In these cases, EVPI would be useful for deciding whether to undertake additional efforts.

23-Dec-20 Dr. Mohammad Babul Hasan

 Example 2: The farmer’s problem

Consider a farmer specializes in raising wheat, corn, and sugar beets on his 500 acres of land. During the winter, he wants to
decide how much land to devote to each crop. The farmer knows that at least 200 tons (T) of wheat and 240 T of corn are
needed for cattle feed. These amounts can be raised on the farm or bought from a wholesaler. Any production in excess of the
feeding requirement would be sold. Over the last decade, mean selling prices have been $170 and $150 per ton of wheat and
corn, respectively. The purchase prices are 40% more than this due to the wholesaler’s margin and transportation costs. Another
profitable crop is sugar beet, which he expects to sell at $36/T; however, there is a quota restriction on sugar beet production.
Any amount in excess of the quota can be sold only at $10/T. The farmer’s quota for next year is 6000 T. Based on past
experience, the farmer knows that the mean yield on his land is roughly 2.5 T, 3 T, and 20 T per acre for wheat, corn, and sugar
beets, respectively.
Table 1 summarizes these data and the planting costs for these crops.
Wheat Corn Sugar Beets
Yield (T/acre) 2.5 3 20
Planting cost ($/acre) 150 230 260
Selling price ($/T) 170 150 36 under 6000 T
Purchase price ($/T) 238 210 10 above 6000 T
Minimum 200 240 -
requirement (T) -

23-Dec-20 Dr. Mohammad Babul Hasan

A farmer raises wheat, corn, and sugar beets on 500 acres of land. Before the planting season he wants to
decide how much land to devote to each crop.
• At least 200 tons of wheat and 240 tons of corn are needed for cattle feed, which can be purchased from a
wholesaler if not raised on the farm.
• Any grain in excess of the cattle feed requirement can be sold at $170 and $150/ton of wheat and corn,
respectively.
• The wholesaler sells the grain for 40% more (namely $238 and $210 per ton, respectively.)
• Up to 6000 tons of sugar beets can be sold for $36 per ton; any additional amounts can be sold for $10/ton.

Crop yields are uncertain, depending upon weather conditions during the growing season.

Three scenarios have been identified ("good", "fair", and "bad"), each equally likely.

In this data only the yield are scenario-dependent, while in the reality the purchase prices and sales revenues
from gain would be higher in year with poor yield.

23-Dec-20 Dr. Mohammad Babul Hasan

To help the farmer make up his mind, we can set up the following model. Let
x1 = acres of land devoted to wheat,
x2 = acres of land devoted to corn,
x3 = acres of land devoted to sugar beets,

w1 = tons of wheat sold,

y1 = tons of wheat purchased,
w2 = tons of corn sold,
y2 = tons of corn purchased,
w3 = tons of sugar beets sold at the favorable price.
w4 = tons of sugar beets sold at the lower price.
Deterministic LP model is
Min Z  150 x1  230 x 2  260 x 3  238 y1  170 w1   210 y 2  150 w 2   36 w3  10 w 4

subject to

Land constraint: x1  x 2  x 3  500

Wheat constraint: 2 .5 x1  y1  w1  200
Corn constraint: 3 x 2  y 2  w 2  240
Sugar beet constraint: w3  w 4  20 x 3
Quota constraint: w3  6000
23-Dec-20 Dr. Mohammad Babul Hasan
x,y
i j , wk  0
 Deterministic solution:

After solving the above problem, the farmer obtains an optimal solution, as in Table below.
OBJECTIVE FUNCTION VALUE
Culture Wheat Corn Sugar Beets
1) -118600.0
Surface (acres) 120 80 300
Yield (T) 300 240 6000 VARIABLE VALUE REDUCED COST
X1 120.000000 0.000000
Sales (T) 100 - 6000 X2 80.000000 0.000000
Purchase (T) - - - X3 300.000000 0.000000
Y1 0.000000 68.000000
Overall profit: $118,600 W1 100.000000 0.000000
Y2 0.000000 41.666668
W2 0.000000 18.333334
W3 6000.000000 0.000000
Yields: Wheat =120*2.5=300 T W4 0.000000 16.750000

Corn = 80*3= 240 T

Beet = 300* 20 = 6000 T

Does this mean that the farmer's expected revenues will actually be 118600?

23-Dec-20 Dr. Mohammad Babul Hasan

 A scenario representation (Stochastic approach with Discrete variables)

A first possibility is to assume some correlation among the yields of the different crops as good, fair, or bad for all crops,
resulting in above average, average, or below average yields for all crops.

Let “above” and “below” average indicate a yield 20% above or below the mean yield.

The farmer wishes to know whether the optimal solution is sensitive to variations in yields.

Again, the solutions in Tables 3 and 4 seem quite natural. The optimal solution is very sensitive to changes in yields.

• The optimal surfaces devoted to wheat range from 100 acres to 183.33 acres.
• Those devoted to corn range from 25 acres to 80 acres and
• those devoted to sugar beets from 250 acres to 375 acres.
• The overall profit ranges from $59,950 to $167,667.

Long-term weather forecasts would be very helpful here. Unfortunately, as even meteorologists agree, weather conditions
cannot be accurately predicted six months ahead.
The farmer must make up his mind without perfect information on yields.

23-Dec-20 Dr. Mohammad Babul Hasan

 Scenario Wheat yield Corn yield Beet yield
(Tons/Acre) (Tons/Acre) (Tons/Acre)
1. Good (Above average) 3 3.6 24

2. Fair (Average) 2.5 3 20

3. Bad (Below Average) 2 2.4 16

23-Dec-20 Dr. Mohammad Babul Hasan

 Scenario-1 Min
AVERAGE Z  150 x1  230 x 2  260 x 3  238 y1  170 w1   210 y 2  150 w 2   36 w3  10 w 4
subject to
Land constraint: x1  x 2  x 3  500
Wheat constraint: 2 . 5 x1  y1  w1  200
Corn constraint: 3 x 2  y 2  w 2  240
Sugar beet constraint: w3  w 4  20 x 3
Quota constraint: w3  6000
xi , y j , wk  0
Scenario-2 Min
ABOVE Z  150 x1  230 x 2  260 x 3  238 y1  170 w1   210 y 2  150 w 2   36 w3  10 w 4
AVERAGE
subject to
Land constraint: x1  x 2  x 3  500
Wheat constraint: 3 x1  y 1  w1  200
Corn constraint: 3 . 6 x 2  y 2  w 2  240
Sugar beet constraint: w 3  w 4  24 x 3
Quota constraint: w3  6000
xi , y j , wk  0
Scenario-3 Min
BELOW Z  150 x1  230 x 2  260 x 3  238 y1  170 w1   210 y 2  150 w 2   36 w3  10 w 4
AVERAGE
subject to
Land constraint: x1  x 2  x 3  500
Wheat constraint: 2 . x1  y 1  w1  200
Corn constraint: 2 . 4 x 2  y 2  w 2  240
Sugar beet constraint: w 3  w 4  16 x 3
Quota constraint: w3  6000
xi , y j , wk  0

23-Dec-20 Dr. Mohammad Babul Hasan

 Table 3 Optimal solution based on above average yields (+ 20%).

Culture Wheat Corn Sugar Beets

Surface (acres) 183.33 66.67 250
Yield (T) 550 240 6000
Sales (T) 350 - 6000
Purchase (T) - - -
Overall profit: $167,667

Table 4 Optimal solution based on below average yields ( −20% ).

Culture Wheat Corn Sugar Beets
Surface (acres) 100 25 375
Yield (T) 200 60 6000
Sales (T) - - 6000
Purchase (T) - 180 -
Overall profit: $59,950

The main issue here is clearly on sugar beet production. Planting large surfaces would make it certain to produce and sell the
quota, but would also make it likely to sell some sugar beets at the unfavorable price. Planting small surfaces would make it
likely to miss the opportunity to sell the full quota at the favorable price.

23-Dec-20 Dr. Mohammad Babul Hasan

 Assuming "Perfect Information", i.e., assuming that the farmer has advance knowledge of
the quality of the yield and can base his decision upon that knowledge

Solution for scenario #1 "Good" yield

Optimal cost: 167666.6667

Stage One Variables:

i X[i]
1 183.33 Wheat Acres
2 66.67 Corn Acres
3 250.00 Beet Acres
4 0.00

Second-stage: nonzero variables

i Y[i]
3 350.00 Sales of wheat
5 6000.00 Sales of Beets

23-Dec-20 Dr. Mohammad Babul Hasan

Solution for scenario #2 "Fair" yield
Solution for scenario #3 "Bad" yield
Optimal cost: 118600
Optimal cost: 59950
Stage One Variables: Stage One Variables:
i X[i] i X[i]
1 120.00 Wheat Acres 1 100.00 Wheat Acres
2 80.00 Corn Acres 2 25.00 Corn Acres
3 300.00 Beet Acres 3 375.00 Beet Acres
4 0.00 4 0.00
Second-stage: nonzero variables
Second-stage: nonzero variables
i Y[i]
i Y[i]
2 180.00 Purchase of Corn
3 100.00 Sales of Wheat
5 6000.00 Sales of Beets 5 6000.00 Sales of Beets

23-Dec-20 Dr. Mohammad Babul Hasan

 Expected Value problem (EVP) : (General Stochastic Programming problem)

The farmer now realizes that he is unable to make a perfect decision that would be best in all circumstances.
He would, therefore, want to assess the benefits and losses of each decision in each situation.
Decisions on land assignment (x1, x2, x3) have to be taken now,
but sales and purchases (wi, i = 1, . . . ,4, y j, j = 1,2) depend on the yields.
Use a scenario index s = 1,2,3 corresponding to above average, average, or below average yields, respectively.
This creates a new set of variables of the form wis, i = 1,2,3,4 , s = 1,2,3 and yjs , j = 1,2 , s = 1,2,3 .
As an example, w32 represents the amount of sugar beets sold at the favorable price if yields are average.
Assuming the farmer wants to maximize long-run profit, it is reasonable to seek a solution that maximizes his expected profit.
This assumption means that the farmer is neutral about risk.
If the three scenarios have an equal probability of 1/3 , the farmer’s problem reads as follows:

23-Dec-20 Dr. Mohammad Babul Hasan

Min Deterministic Equivalent of SP problem directly, without decomposition
1
Z  150 x1  230 x 2  260 x 3  238 y11  170 w11  210 y 21  150 w 21  36 w31  10 w 41 
3


1
238 y12  170 w12  210 y 22  150 w 22  36 w32  10 w 42 
This model of stochastic decision program is known as the
3 extensive form of the stochastic program because it explicitly
1 describes the second-stage decision variables for all scenarios.
 238 y13  170 w13  210 y 23  150 w 23  36 w33  10 w 43 
3
subject to
Scenario 1: Land constraint: x1  x 2  x 3  500
Wheat constraint: 3 x1  y11  w11  200
Corn constraint: 3 .6 x 2  y 21  w21  240
Sugar beet constraint: w31  w 41  24 x 3
Quota constraint: w 31  6000
Scenario 2: Wheat constraint: 2 .5 x1  y12  w12  200
Corn constraint: 3 x 2  y 22  w 22  240
Sugar beet constraint: w 32  w 42  20 x 3
Quota constraint: w 32  6000
Scenario 3: Wheat constraint: 2 x1  y13  w13  200
Corn constraint: 2 . 4 x 2  y 23  w 23  240
Sugar beet constraint: w33  w 43  16 x 3
Quota constraint: w 33  6000
23-Dec-20 w k  0 Babul Hasan
x i , yDr.j ,Mohammad
 LP OPTIMUM FOUND AT STEP 0

OBJECTIVE FUNCTION VALUE

1) -108390.2

VARIABLE VALUE REDUCED COST

X1 170.000000 0.000000
X2 80.000000 0.000000
X3 250.000000 0.000000
Y11 0.000000 22.666000
W11 310.000000 0.000000
Y21 0.000000 20.000000
W21 48.000000 0.000000
W31 6000.000000 0.000000
W41 0.000000 0.961771
Y12 0.000000 22.663002
W12 225.000000 0.000000
Y22 0.000000 17.665834
W22 0.000000 2.334166
W32 5000.000000 0.000000
W42 0.000000 8.670000
Y13 0.000000 22.663002
W13 140.000000 0.000000
Y23 48.000000 0.000000
W23 0.000000 20.000000
W33 4000.000000 0.000000
W43 0.000000 8.670000

23-Dec-20 Dr. Mohammad Babul Hasan

 Optimal Solution

(Found by solving deterministic equivalent problem directly,

without decomposition)
----------------------------------------
Total cost: 108390
Stage One Variables:
i variable value
1 X[1] 170 Wheat Acres
2 X[2] 80 Corn Acres
3 X[3] 250 Beets Acres
4 slack 1 0
----------------------------------------

23-Dec-20 Dr. Mohammad Babul Hasan

The top line of Table 5 gives the planting areas, which must be determined before realizing the weather and crop yields. This
decision is called the first stage. The other lines describe the yields, sales, and purchases in the three scenarios. They are called
the second stage. The bottom line shows the overall expected profit.

Table 5 Optimal solution based on the stochastic model (1.2).

Culture Wheat Corn Sugar Beets
First stage Area (Acres) 170 80 250

s =1 Above Yield (T) 510 288 6000

Sales (T) 310 48 6000
Purchase (T) - - -
s=2 Yield (T) 425 240 5000
Average Sales (T) 225 - 5000
Purchase (T) - - -
s = 3 Below Yield (T) 340 192 4000
Sales (T) 140 - 4000
Purchase (T) - 48 -
Overall profit: $108,390

23-Dec-20 Dr. Mohammad Babul Hasan

 Second Stage
For each scenario, the optimal recourse variables are computed:

Scenario #1 "Good" yield Scenario #2 "Fair" yield

i variable value i variable value
1 Y[1] 0 1 Y[1] 0
2 Y[2] 0 2 Y[2] 0
3 W1 310 Sales of wheat 3 W1 225 Sales of wheat
4 W2 48 Sales of corn 4 W2 0
5 W3 6000 Sales of beets 5 W3 5000 Sales of beets
6 W4 0 6 W4 0
---------------------------------------- ----------------------------------------

Scenario #3 "Bad" yield

i variable value
1 Y[1] 0
2 Y[2] 48 Purchase of corn
3 W1 140 Sales of wheat
4 W2 0
5 W3 4000 Sales of beets
6 W4 0

23-Dec-20 Dr. Mohammad Babul Hasan

 What’s it worth?: expected value of perfect information(EVPI)

With perfect information, Farmer Ted’s would plant (wheat, corn, beans).
* Good yield: (183.33, 66.67, 250), Profit: $167,667
* Average yield: (120, 80, 300), Profit: $118,600
* Bad yield: (100, 25, 375), Profit: $59,950

Assuming each of these scenarios occurs with probability 1/3, his long run average profit would be
(1/3)(167667) + (1/3)(118600) + (1/3)(59950) = 115406

With his (optimal) “here-and-now” decision of (170, 80, 250), he would make a long run profit of 108390
The difference (115406-108390) = 7016 is the expected value of perfect information(EVPI)

23-Dec-20 Dr. Mohammad Babul Hasan

 Example 3
Pop-Donuts company makes two products, donuts and cakes. It has bottlenecks in its capital used to purchase flour
that is required in making both donuts and cakes, and also in labor hours used to make these two products. The goal of
Pop-Donuts is to make as much profits as possible within its capacities. Currently contribution margins for a dozen
donuts and a cake are $3 and $9, respectively. Pop-Donuts will adjust prices to keep same contribution margins when
variable costs change. Required direct labor hours are 0.25 h per a dozen donuts and 1 h per cake. Pop-Donuts has
business uncertainties and several possible scenarios in its business operations are summarized in the following:

Scenario 1: Pop-Donuts has a tight capital budget for only $15,000 of flour costs and 10,000 labor hours annually. The
flour for donuts costs $0.5 per a dozen donuts and $1.2 per cake.

Scenario 2: Pop-Donuts has a modest capital budget for $17,000 of flour costs and 10,000 labor hours annually. The
flour for donuts costs $0.6 per a dozen donuts and $1.3 per cake.

Scenario 3: Pop-Donuts has a very good capital budget for $20,000 of flour costs and 10,000 labor hours annually.
The flour for donuts costs $0.7 per a dozen donuts and $1.4 per cake.

23-Dec-20 Dr. Mohammad Babul Hasan

For Discrete Distribution:
When discrete distributions are used, the problem will have a finite number of scenarios.
The probabilities of these three scenarios are 1/3 each. Then the problem can be formulated as
follows:

Scenarios 1: Scenarios 2: Scenarios 3:

Max Z 1  3 x1  9 x 2 Max Z 1  3 x1  9 x 2 Max Z 1  3 x1  9 x 2
subject to subject to subject to
0 .5 x1  1 .2 x 2  15000 0 .6 x1  1 .3 x 2  17000 0 .7 x1  1 .4 x 2  20000
2 .5 x1  x 2  10000 2 .5 x1  x 2  10000 2 .5 x1  x 2  10000
x1 , x 2  0 . x1 , x 2  0 . x1 , x 2  0 .

23-Dec-20 Dr. Mohammad Babul Hasan

LP OPTIMUM FOUND AT STEP 0 LP OPTIMUM FOUND AT STEP 1
OBJECTIVE FUNCTION VALUE OBJECTIVE FUNCTION VALUE
1) 101250.0
1) 100909.1
VARIABLE VALUE REDUCED COST
X1 15000.000000 0.000000 VARIABLE VALUE REDUCED COST
X2 6250.000000 0.000000 X1 14545.454102 0.000000
X2 6363.636230 0.000000

ROW SLACK OR SURPLUS DUAL PRICES ROW SLACK OR SURPLUS DUAL PRICES
2) 0.000000 3.750000 2) 0.000000 2.727273
3) 0.000000 4.500000
3) 0.000000 5.454545
4) 15000.000000 0.000000
5) 6250.000000 0.000000 4) 14545.454102 0.000000
5) 6363.636230 0.000000
NO. ITERATIONS= 0
NO. ITERATIONS= 1

LP OPTIMUM FOUND AT STEP 0

OBJECTIVE FUNCTION VALUE

1) 102857.1
VARIABLE VALUE REDUCED COST
X1 17142.857422 0.000000
X2 5714.285645 0.000000

ROW SLACK OR SURPLUS DUAL PRICES

2) 0.000000 2.142857
3) 0.000000 6.000000
4) 17142.857422 0.000000
5) 5714.285645 0.000000

NO. ITERATIONS=
23-Dec-20 0 Dr. Mohammad Babul Hasan
 Solutions to these problems are shown bellow.

Scenario 1 Scenario 2 Scenario 3

Unit contribution margin Unit contribution margin Donuts: Unit contribution margin
Donuts: 15,000 units at $3 14,545 units at $3 Donuts: 17,142 units at $3
Cake: 6,250 units at $9 Cake: 6,366 units at $9 Cake: 5,714 units at $9
Contribution margin $101,250 Contribution margin $100,902 Contribution margin $102,852
Fixed expenses Fixed expenses Fixed expenses
Manufacturing overhead Manufacturing overhead $10,000 Manufacturing overhead $10,000
$10,000 Sales and administrative $8,000 Sales and administrative $8,000
Sales and administrative Operating income $82,902 Operating income $84,852
$8,000
Operating income $83,250

Mean value of operating income $83,668

23-Dec-20 Dr. Mohammad Babul Hasan

The model can be further simplified as the following:

Q  x ,    E  Q  x ,     p  Q  x ,   , the problem can be converted to a deterministic equivalent


problem with more variables and inequalities,

Z  c T x   p   q   y  
T
Max / Min


subject to

Ax  B
W   y    H    T   x
x  x1 , x 2 , .... x n   0, y  0

23-Dec-20 Dr. Mohammad Babul Hasan

 Stochastic formulation:
1
(SP 1) Max Z 1  3 x1  9 x 2   y1 y2  y3 
3
subject to 0 .5 x1  1 .2 x 2  y1  15000
0 .6 x1  1 .3 x 2  y 2  17000
0 .7 x1  1 .4 x 2  y 3  20000
2 .5 x1  x 2  10000
x1 , x 2  0
y1 , y 2 , y 3  0

This discrete SLP model is hence converted to a larger DLP model. LP methods can then be
applied to solve this question.

23-Dec-20 Dr. Mohammad Babul Hasan

 LP OPTIMUM FOUND AT STEP 1

OBJECTIVE FUNCTION VALUE

1) 100579.1

VARIABLE VALUE REDUCED COST

X1 14545.454102 0.000000
X2 6363.636230 0.000000
Y1 90.909088 0.000000
Y2 0.000000 3.717273
Y3 909.090881 0.000000

ROW SLACK OR SURPLUS DUAL PRICES

2) 0.000000 -0.330000
3) 0.000000 3.387273
4) 0.000000 -0.330000
5) 0.000000 5.454545
6) 14545.454102 0.000000
7) 6363.636230 0.000000

NO. ITERATIONS= 1

23-Dec-20 Dr. Mohammad Babul Hasan

 What’s it worth?: expected value of perfect information(EVPI)

With perfect information, Farmer Ted’s would plant (wheat, corn, beans).
* Scenario-1: (15000, 6250), Profit: $101250
* Scenario-2: (14545.45, 6363.63), Profit: $100909.1
* Scenario-3: (17142.85, 5714.28), Profit: $102857.1
Assuming each of these scenarios occurs with probability 1/3, his long run average profit would be
(1/3)(101250) + (1/3)(100909.1) + (1/3)(102857.1) = 101672.07

With his (optimal) “here-and-now” decision, he would make a long run profit of 100579.1
This difference (101672.07 – 100579.1) = 1092.97 is the expected value of perfect information(EVPI)

23-Dec-20 Dr. Mohammad Babul Hasan

Discrete Random Variables:

Uniform: U[1,n]
n 1 n2 1
n  0 with E   
1
P   i   , i  1, 2 ,..., n , and Var   
n 2 12

Binomial: Bi(n, p)

n i
P   i     p 1  p  , 0  p  1 1 with E    np and Var    np 1  p 
n i
i  0,1, 2,..., n ,
i

Poisson: P  
i
P   i   e 
,   0, i  0,1,.... 1 with E     and Var    
i!

23-Dec-20 Dr. Mohammad Babul Hasan

Continuous Random Variables:

Uniform: U 0 , a 
a2
a  0 with E    a and Var   
1
f    , 0    a,
a  12

Exponential: exp  
2
1
  0 with E   
1
f     e  
, 0  , and Var     
  

Normal: N  ,  2

   2

  0 with E     and Var    
1
f    2 2 2
e ,
2 2

Gamma: G  ,  
 

1
f      1
e 
,  ,   0 where     x
 1
e  x dx ,   0 with E     and
   
2
0

Var     2

For a random variable ξ, we define its cumulative distribution F  x   P   x  , or more precisely

F  x   P  |   x  . Two major cases are then considered.
23-Dec-20 Dr. Mohammad Babul Hasan
A discrete random variable takes a finite or countable number of different values. It is best
described by its probability distribution, which is the list of possible values,  k , k  K , with
associated probabilities,
f  k   P    k  s. t.  f  k   1 Σ
k K

Continuous random variables can often be described through a so-called density function f (ξ ) .
b

The probability of ξ being in an interval [a,b] is obtained as P(a ≤ ξ ≤ b) =  f   d 

a
b

or equivalently P(a ≤ ξ ≤ b) =  dF  
a

where F(·) is the cumulative distribution as earlier. Contrary to the discrete case, the probability of a
single value P(ξ = a) is always zero for a continuous random variable. The distribution F(·) must be

such that  d F    1


23-Dec-20 Dr. Mohammad Babul Hasan

The expectation of a random variable in discrete case is computed as     f   .
k k

k K

The expectation of a random variable in continuous case is computed as     d F   .


The variance of a random variable is E[(ξ−μ)2] . The expectation of ξr is called the rth moment
of ξ and is denoted ¯ξ(r) = E[ξr ] . A point η is called the α -quantile of ξ if and only if for 0 <α < 1,
η = min{x | F(x) ≥α} .

23-Dec-20 Dr. Mohammad Babul Hasan

 For Uniform Distribution:
Probability density function (pdf)
The probability density function of the continuous uniform distribution is:
 1
 a xb
f x   
,
ba
 0, otherwise

The values of f(x) at the two boundaries a and b are usually unimportant because they do not alter the
values of the integrals of f(x) dx over any interval, nor of x f(x) dx or any higher moment. Sometimes
1
they are chosen to be zero, and sometimes chosen to be . The latter is appropriate in the context
ba
of estimation by the method of maximum likelihood. In the context of Fourier analysis, one may take
1
the value of f(a) or f(b) to be ,since then the inverse transform of many integral transforms of
2 b  a 
this uniform function will yield back the function itself, rather than a function which is equal "almost
everywhere", i.e. except on a set of points with zero measure. Also, it is consistent with the sign
function which has no such ambiguity.

23-Dec-20 Dr. Mohammad Babul Hasan

Graphically, the probability density function is portrayed as a rectangle where (b  a ) is the base
1
and is the height. As the distance between a and b increases, the density at any particular value
ba
within the distribution boundaries decreases. Since the probability density function integrates to 1, the
height of the probability density function decreases as the base length increases.
Probability density function

23-Dec-20 Dr. Mohammad Babul Hasan

Example 1. For random variable X, X  U 0, 23  Find P 2  x  18 
1 16
Solution: P 2  x  18   18  2   
23  0 23
In graphical representation of uniform distribution function [f(x) vs x], the area under the curve
within the specified bounds displays the probability (shaded area is depicted as a rectangle). For this
specific example above, the base would be (18-2) and the height would be 1/23.
Example 2. For random variable X, X  U 0, 23  Find P  x  12 | x  8 
1 11
Solution: P  x  12 | x  8   23  12   
23  8 15
The example above is for a conditional probability case for the uniform distribution: given x>8 is
true, what is the probability that x>12. Conditional probability changes the sample space so a new
interval length (b-a) has to be calculated, where b is 23 and a is 8. The graphical representation would
still follow Example 1, where the area under the curve within the specified bounds displays the
probability and the base of the rectangle would be 23-12 and the height 1/15.

23-Dec-20 Dr. Mohammad Babul Hasan

In terms of mean μ and variance σ2, the probability density may be written as:
 1
 ,  3  x    3
f  x    2 3
 0 , otherwise

23-Dec-20 Dr. Mohammad Babul Hasan

 Notation U a , b  or unif a , b 
Parameters   a b
Support x  a , b 
PDF  1
 a xb
f x   
,
ba

 0, otherwise
CDF  0, xa
 1
f x    , a xb
b  a
 1, xb

Mean 1
a  b 
2
Median 1
a  b 
2
Mode any value in a, b 
Variance 1
b  a 2
12
23-Dec-20 Skewness 0 Dr. Mohammad Babul Hasan
 For Exponential distribution
The probability density function (pdf) of an exponential distribution is

  e  x , x0
f  x;    
 0, otherwise
Here λ > 0 is the parameter of the distribution, often called the rate parameter. The distribution is
supported on the interval [0, ∞). If a random variable X has this distribution, we write X ~ Exp(λ).
The exponential distribution exhibits infinite divisibility.

23-Dec-20 Dr. Mohammad Babul Hasan

Cumulative distribution function

The cumulative distribution function (CDF) is given by

 1  e  x , x0
f  x;    
 0, otherwise

23-Dec-20 Dr. Mohammad Babul Hasan

 Alternative parametrizatio
The exponential distribution is sometimes parametrized in terms of the scale parameter β = 1/λ:
 1 
x

 
x0
f  x;    
e ,


 0, otherwise

The mean is the probability mass centre, the first moment. The median is the preimage F−1(1/2).

23-Dec-20 Dr. Mohammad Babul Hasan

The mean or expected value of an exponentially distributed random variable X with rate parameter λ
1
is given by EX  

In light of the examples given below, this makes sense: if you receive phone calls at an average rate
of 2 per hour, then you can expect to wait half an hour for every call.
1
The variance of X is given by V  X  
2
so the standard deviation is equal to the mean.

The moments of X, for n   are given by E  X n  

n!
n

!n n! n
 1k
The central moments of X, for n   are given by  n 
 n

 n  k!
where !n is
k 0

subfactorial of n
ln 2
The median of X is given by m  X    E  X  . Thus the absolute difference between the mean

1  ln 2 1
and median is E X   mX       X  in accordance with the median-mean inequality.
23-Dec-20  
Dr. Mohammad Babul Hasan
 Parameters   0, rate, inverse scale
Support x  0 ,  
PDF   e  x , x0
f  x;    
 0, otherwise
CDF  1  e  x , x0
f  x;    
 0, otherwise
Quantile ln 1  p 


Mean 1
EX  

Median ln 2

Mode 0
Variance 1
V X  
2
Skewness 2
Ex. kurtosis 6
Entropy 1  ln 
MGF 
, t
23-Dec-20  Mohammad
Dr. t Babul Hasan