Excel LP Sensitivity Analysis

Linear Programming Sensitivity Analysis
Engineering Systems Analysis for Design Massachusetts Institute of Technology
Richard de Neufville, Joel Clark and Frank R. Field LP Sensitivity Analysis Slide 1 of 22
Sensitivity Analysis
Rationale Shadow Prices
Definition Use Sign Range of Validity
Opportunity Costs
Definition Use
Engineering Systems Analysis for Design Massachusetts Institute of Technology Richard de Neufville, Joel Clark and Frank R. Field LP Sensitivity Analysis Slide 2 of 22
Rationale for Sensitivity Analysis

Math problem is an approximation
optimum is an approximation we need to check
Constraints often artificial

Designer should question them Should we have different specifications?
Situations always probabilistic

Prices change Need to assess risk
Shadow Price Definition

Recall from Constrained Optimization:
Shadow price = (objective function)/ (constraint) at the optimum Complementary Slackness: Either (Slack variable) or (shadow price) = 0
Shadow Price Illustration

Max: Y = X1 + 4X2 s.t. X1 + X2 < 5 = b1 X1 > 3 = b2 X2 < 3 = b3 X1 , X2 > 0
Notes: a) X1* = 3; X2* = 2; Y* = 11 b) when b1 = +1 X1* = 0 ; X2* = +1 Y*= +4 ; SP1 = 4 c) SP3* = 0; slack3 = 1 d) when b1 > 6 slack3 = 0; SP3 0; SP1 = 1 < 4 X1 + X2 < 7
X1 + X2 < 6
Proactive Use of Shadow Prices

Identify constraints with high S.P See if they can be changed for better solutions Example: New York water supply
Original Design for Third City Tunnel ($1 billion plus) pressure < 40 psi at curb (some point in Brooklyn) No allowance for local tanks, pumps Shadow price in millions of dollars!
Reactive Use of Shadow Prices

Respond to new opportunities Example: client changes specifications Respond to proposals for new constraints Example: trace chemicals
Sign of Shadow Prices

"Obvious Rule" (+SP with +b) not correct Correct Reasoning:
What makes the optimum better? Expansion of feasible region => "Relaxation of constraints" What changes will increase the feasible region? Increase upper bound Decrease lower bound j aijXj < bi k akjXj > bk
i.e., "Raise the roof, lower the floor."

Shadow Price Illustration

Max: Y = X1 + 4X2 s.t. X1 + X2 < 5 = b1 X1 > 3 = b2 X2 < 3 = b3 X1 , X2 > 0
Notes: a) X1* = 3; X2* = 2; Y* = 11 b) when b1 = +1 X1* = 0 ; X2* = +1 Y*= +4 ; SP1 = 4 c) SP3* = 0; slack3 = 1 d) when b1 > 6 slack3 = 0; SP3 0; SP1 = 1 < 4 X1 + X2 < 7
X1 + X2 < 6
Shadow Prices As Constraints Change

increase an upper bound ("raise the roof")
decrease a lower bound ("lower the floor") b2: 3 --> 2 new X* = [2,3] new Y* = 14 DY* = 3
Range of Shadow Prices

In Linear Programming, Shadow prices are constant Until a constraint changes enough so that a new constraint is binding Results given as: SPK = constant for rL < bK < rU Outside the range:
Shadow prices decrease as constraint is relaxed Shadow prices increase as constraint is tightened
High Point Direction of Change in Constraint New Constraint Binding
Shadow Price Ranges for Example

Max: Y = X1 + 4X2 s.t. X1 + X2 < 5 = b1 X1 > 3 = b2 X2 < 3 = b3 X1 , X2 > 0 Shadow Prices SP1 = 4 3 < b1 < 6 SP2 = 3 2 < b2 < 5 SP3 = 0 2 < b3
X1 + X2 < 7
X1 + X2 < 6
Opportunity Cost - Definition

Objective Function = ci Xi Opportunity costs associated with ci -- the coefficients of design/decision variables At optimum, some decision variables = 0
These are non-optimal decision variables
Opportunity cost is:

Degradation of optimum per unit of non-optimal variable introduced into design A "cost" in that it is a worsening of optimum. Units may be almost anything; equal to whatever units are being optimized.
Meaning of Opportunity Costs

Opportunity cost defines design trigger "price" The value of the coefficient of the decision variable for which that variable should be in the design Suppose: Obj.Function = ... + cK XK + ... and XK not optimal with an opportunity cost = OCK Then, as cK changes for the better, (greater for maximization, lesser for minimization) OCK lower OCK = 0 at cK' = cK - OCK cK' is trigger price; defines the limit of best design
Illustration of Opportunity Cost

What happens when forced to use a non-optimal decision variable? Example: Min Cost = 2X1 + 10X2 + 20 X3 s.t. X1 + X1 , X* = (2, 1, 0); cost* = 14 X2 + X2 X2 , X3 > 3 >1 X3 > 0
If forced to use X3, new X* = (1,1,1); new cost* = 32 Thus: (opportunity cost)3 = Z*/1 = 18
Use of Opportunity Cost

At what price would it be desirable to use X3 ? If X3 is used with no change in its unit cost (= c3), the optimal cost would increase by 18 If the cost of X3 were to fall by an amount equal to the opportunity cost (c3' = c3 - OC3 = 20 -18 = 2). It would then compete with X1 So the answer is: When its unit cost falls by its opportunity cost: 20 - 18 = 2
How do you find SP and OC?

LP optimization programs all calculate shadow prices and opportunity costs routinely and print them out for you Sometimes, programs report this information in special ways. Thus:
Shadow Prices => Shadow prices or dual decision variables Opportunity Costs => Reduced Gradient or dual slack variables More on duality later
A Possible Semantic Confusion

Note that the Phrases shadow price and opportunity cost have somewhat different meanings in LP and Economics literature The opportunity cost of an action in economics can be interpreted as the shadow price of that action on the budget
Example for Assumptions

minimize: Z = 3X1 + 5X2 s.t. X1 + X2 > 5 3X1 + 2X2 < 18 6X1 + 5X2 < 42 -7X1 + 8X2 < 0 0 < X1 < 4 0 < X2 < 6 Solution: X1* = 4 X2* = 1
Example Solution from Excel Solver

X(SUB1) 4 EXPRESSION CONSTRAINT X1 +X2 5 >= 3 X1 + 2 X2 14 <= 6 X1 +5 X2 29 <= (-7 X1) + 8 X2 -20 <= X1 4 >= X1 4 <= X2 1 >= X2 1 <=
X(SUB2) 1 LIMIT 5 18 42 0 0 4 0 6
OBJ FCN 17
Example Sensitivity Report from Excel Solver

Adjustable Cells Cell $B$5 X(SUB1) $C$5 $D$5 X(SUB2) Constraints Cell $B$8 $B$9 $B$10 $B$11 $B$12 $B$13 $B$14 $B$15 Final Lagrange Name Value Multiplier X1 +X2 CONSTRAINT 5 5 3 X1 + 2 X2 CONSTRAINT 14 0 6 X1 +5 X2 CONSTRAINT 29 0 (-7 X1) + 8 X2 CONSTRAINT -20 0 X1 CONSTRAINT 4 0 X1 CONSTRAINT 4 -2 X2 CONSTRAINT 1 0 X2 CONSTRAINT 1 0 => Shadow Price Name Final Reduced Value Gradient 4 0 0 0 1 0 => Opportunity Cost
Summary on LP Sensitivity Analysis

LP Optimization Programs automatically provide important information useful for improving/changing design Shadow prices -- to help redefine constraints Opportunity costs -- to identify critical prices Need to understand these quantities carefully

Excel LP Sensitivity Analysis

Uploaded by

Copyright:

Available Formats

Excel LP Sensitivity Analysis

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Excel LP Sensitivity Analysis

Uploaded by

Copyright:

Available Formats

Linear Programming Sensitivity Analysis

Engineering Systems Analysis for Design Massachusetts Institute of Technology

Rationale for Sensitivity Analysis

Constraints often artificial

Situations always probabilistic

Shadow Price Definition

Engineering Systems Analysis for Design Massachusetts Institute of Technology

Shadow Price Illustration

Proactive Use of Shadow Prices

Reactive Use of Shadow Prices

Engineering Systems Analysis for Design Massachusetts Institute of Technology

Sign of Shadow Prices

i.e., "Raise the roof, lower the floor."

Shadow Price Illustration

Shadow Prices As Constraints Change

Range of Shadow Prices

Shadow Price Ranges for Example

Opportunity Cost - Definition

Opportunity cost is:

Meaning of Opportunity Costs

Illustration of Opportunity Cost

Use of Opportunity Cost

How do you find SP and OC?

A Possible Semantic Confusion

Engineering Systems Analysis for Design Massachusetts Institute of Technology

Example for Assumptions

Example Solution from Excel Solver

Example Sensitivity Report from Excel Solver

Engineering Systems Analysis for Design Massachusetts Institute of Technology

Summary on LP Sensitivity Analysis

Engineering Systems Analysis for Design Massachusetts Institute of Technology

You might also like