1-1 Intro to LPs 1-3 Modeling with Algebra

Variable: each variable represents a single quantitative decision made - Target Cell  Objective Function
Regular Constraint: all constraints besides the non-negativity ones - Decision Cells  Variables Hay Farm Example
Binding Constraint: when amount used = amount available (LHS = RHS) - Constraints  Constraints (equations)
Non-Binding Constraint: not all the resource is used (LHS ≠ RHS) - “assume non-negative”  Non-negativity constraints
Redundant Constraint: constraints that don’t impact the feasible region
Yellow Fill: values given by the problem Translating from Excel to Algebra
Red Outline: variable or decision cells  non-negativity constraints 1) Start with variable declarations
Blue Outline: all other constraints 2) Then objective function
Green Fill: target or objective cell  total profit or cost 3) Then constraints
Total # of constraints = non-negativity + regular + blending Constraints 4) Finish with non-negativity constraints

1-2 LPs in Excel 1-4 Graphical Formulation

How to setup Excel Solver: 1) Assume that: 𝑥-axis = first given variable and 𝑦-axis = second given variable
1) Select Objective max or min 2) Draw the constraints as linear equations ( 𝑦 = 𝑚𝑥 + 𝑏 )
2) Select Variable Cells  Turn all the inequality signs into equality signs (=)
3) Input Constraints  Find the 𝑥 and 𝑦 intercepts of each constraint equation to graph the line
4) Click non-negative and “Simplex LP”  Shade the feasible region using inequality values
 Relative Reference: A1 (moves relative to you) 3) Find the Optimal Solution through Enumeration
 Absolute Reference: $A$1 (stuck, Cmd+T) (usually variables)  Find all corner points of the feasible region
 Mixed Reference: $A1 or A$1 (one moves the other doesn’t)  Use systems of equations if needed (2 equations used = 2 constraints that intersect)
Plug in the 𝑥 and 𝑦 values of each point into the objective function to find
2-1 Isoprofit Lines profit/cost
- 3D graphs require a 𝑧-axis for profit (topographical map)  The point with highest profit or lowest cost is the optimal solution
- Isoprofit Lines: allow us to “graph” the objective function and
identify which feasible region corner is the optimal solution
 Profit increases as you move further away from the origin
 The optimal isoprofit line will always pass through a
corner point (since they are linear programs)
Find Optimal Solution by Isoprofit Line
1) Draw the isoprofit line and keep moving your ruler up
 Slope of the isoprofit line is −

2-2 Sensitivity Analysis

- For answering the question: How sensitive is the solution to changes in inputs?
The extent to which changes in the slope of the isoprofit line impacts the optimal
values for our variables and objective function
 Typically change one input at a time (objective coefficient or a RHS value)
2-3 Shadow Price  If we change more than 1 variable at a time, we don’t know how much the target
- Change in the value of the objective function that happens as a cell and optimal solution will change
result of increasing the RHS of a given constraint by 1 unit - Allowable Increase/Decrease for 𝑪𝟏 and 𝑪𝟐 : Largest value we can increase/decrease
 new 𝜋 − old 𝜋 (𝜋 = value of objective function – new profit) 𝐶 /𝐶 so that the optimal solution remains the same and the shadow price remains constant
Binding Constraints slope of binding constraint #1 ≤ slope of an isoprofit line ≤ slope of binding constraint #2
- Change  within allowable range  Allowable increase = Maximum Increase – Current Value
 Changes: Target Cell & Optimal Solution  Allowable decrease = Current Value – Minimum Decrease
 Constant: Shadow Price - If an increase/decrease in an objective coefficient is within the allowable range, then the
- Change  outside allowable range optimal solution does not change, and will lead to a proportional change in target cell
 Changes: Target cell & Optimal Solution - If an increase/decrease in an objective coefficient violates the allowable range, then the
 Re-calculated: Shadow Price (either increase or decrease) optimal solution/point changes
Non-Binding Constraints Calculating Allowable Increase/Decrease
- Change  within allowable range
 Changes: Target Cell - Change in objective function = ∆𝐶 /𝐶 ×
 Constant: Optimal Solution & Shadow Price value you got for variable of 𝐶 /𝐶
*Shadow Price of Non-Binding Constraints = 0*
- Change  outside allowable range
 Changes: Target Cell, Optimal Solution & Shadow Price
Calculating Shadow Price
1) Add +1 to RHS of a selected constraint (constraint in question)
2) Find new optimal solution (at intersection of the new binding
constraint and the other constraint)
3) Compute value of the objective function at the new optimal
point
4) Shadow Price = new 𝜋 − old 𝜋
Calculating Allowable  for Non-Binding Constraints
1) Substitute optimum solution into LHS of constraint
2) Allowable : Original RHS value – answer step 1
3) Allowable : Infinity / 1E + 30
3-1 Blending Models 3-2 Scheduling and Transportation Models
Within the Changing RHS of a Changing Objective Scheduling Models: How many people/resources in each shift (Ladner Police)
allowable  or  binding constraint Coefficient  Objective coefficient will never increase from 1 to 2 suddenly
Changes Optimal YES NO Transportation Models: How much to ship from location to location
Solution?  Demand and Supply: Supply is allowed to be > Demand, however
Changes Target YES YES Demand can never be > Supply
Cell Value?  PowerCo: If plant 1 cannot ship to city 2, you can say:
- To determine # of variables in a blending model, find every possible 1) Add an additional constraint of 0 = 0
outcome there is (2 demand, 3 supply = 6 variables) 2) Set the cost of shipping from plant 1 to city 2 = a big number (999)
Profit = Blending Models: multiple variables and outcomes are mixed in a single model
revenue – cost  How much of each input goes into each output

IMPORTANT NOTES
 “twice as many 𝑟 as 𝑠” means 𝑟 is 2 times more than 𝑠, or equivalently 𝑟 = 2𝑠.
Thus, “no more than twice as many 𝑟 as 𝑠” can be written as 𝑟 ≤ 2𝑠
 If any of the allowable  or  in the top part of the sensitivity analysis report
(variable cells) is ZERO, then there are multiple optima. This is because the isoprofit
line is exactly on one of the binding constraints
 Shadow Price × RHS increase in question = how much profit increases

Infeasible problem: when constraints do not form a feasible region

 no optimal solution (“solver could not find a feasible solution”)
Unbounded problem: when feasible region can continue infinitely
 a solution can’t be given (“objective cell values do not converge”)
*If LP is unbounded, LP can’t be infeasible*