Professional Documents
Culture Documents
A Translation Approach To Teaching Linear
A Translation Approach To Teaching Linear
A Translation Approach To Teaching Linear
This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@informs.org.
The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.
With 12,500 members from nearly 90 countries, INFORMS is the largest international association of operations research (O.R.)
and analytics professionals and students. INFORMS provides unique networking and learning opportunities for individual
professionals, and organizations of all types and sizes, to better understand and use O.R. and analytics tools and methods to
transform strategic visions and achieve better outcomes.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
STEVENS & PALOCSAY
A Translation Approach To Teaching Linear Program Formulation
Abstract
While there have been many improvements in the teaching of operations research/management science (OR/MS)
in recent years, students continue to have great difficulty with the process of constructing a linear programming
model. We propose addressing this issue with a translation approach that breaks the process down into a series
of small, well-defined steps. The underlying idea is to develop a representation of the problem in terms of
measurable quantities that can then be translated, via application of an explicit rule, into the proper algebraic
form and/or equivalent spreadsheet formulas. As background, related research on word problem solving from
cognitive psychology and mathematical education is reviewed in the paper. We illustrate the translation approach
with several examples and explain how using it can improve students' modeling skills. Preliminary data on
the effectiveness of the methodology for undergraduate business students in an introductory MS course is also
presented.
Editor's note: This is a pdf copy of an html document which resides at http://ite.pubs.informs.org/Vo4No3/
StevensPalocsay/ (Volume 4, Number 3, May 2004)
1. students have difficulty in verbally recognizing The translation approach has two stages: rephrasing
model constraints, and the word problem and constructing the mathematical
program. In the first stage we focus on semantic con-
2. students have difficulty translating a verbal con-
tent only, restating the original problem in terms of
cept into the corresponding algebraic expression.
measurable quantities. This corresponds to the general
steps (1) through (4) listed above. We essentially use
Both the translation and interpretation stages of LP
a re-ordered version of the "helpful questions" posed
modeling require an understanding of the meaning of
by Camm and Evans (2000) to help identify the major
an equation and an operational process for converting
components of the LP model:
a word problem into a mathematical model and stating
its solution in words. Based on our experience, we
1. What is the objective to be maximized or mini-
believe that these translation difficulties are the prima-
mized?
ry cause of our students' struggles when attempting
to construct their own LP models. Grossman (2002) 2. What limitations or requirements restrict the values
has identified three "keys" that students need if they of the decision variables?
are to effectively use management science: model cre-
INFORMS Transactions on Education 4:3(38-54) 42 © INFORMS ISSN: 1532-0545
STEVENS & PALOCSAY
A Translation Approach To Teaching Linear Program Formulation
3. What am I trying to decide? than the second, or that the first is no less than the
second. We associate these three mathematical possi-
However, our approach differs in that the answers are bilities with three semantic categories of constraints:
formally expressed as measurable quantities. Because limited resource, minimum performance, and conser-
no numbers or variables are involved at this stage, vation.
students can and must focus on meaning and "math
phobic" reactions are minimized. The quantities iden-
tified are natural choices for calculated cells in an Excel
representation, and the relationships identified among
them form the constraints of the program. We note
that it is often appropriate to use graphical diagrams
in this stage, to show structural relationships for more
complex problems. In the second stage, a simple pro- Figure 1: Categories of Constraints.
cedure is used to transform each measurable quantity
into proper mathematical form. This stage corresponds Each constraint category has an equivalent in common
to the general steps (5) and (6) but augments them English. These English forms help students to identify
with a rule that generalizes the mechanics for building and correctly frame the problem's requirements. The
mathematical expressions in the model. result is the measurable quantity representation of the
problem. We present the following example to illus-
In the following sections of the paper, we detail both trate this process.
the steps in applying the translation approach and the
rationale for these steps, using several example prob- 5.1. Example 1
lems to discuss the process from the perspective of the
student. The development of this approach is based A bookstore receives 40 paperback copies and 65
on our own work in modeling real world problems hardback copies of a new novel from its shipping
(e.g., Bopp, et al., 1996) as well as our teaching experi- warehouse, but due to preorders taken from cus-
ence. A summary of the steps in the form of a class tomers, it needs at least 80 copies of each type before
handout that can be used by the instructor is provided the store opens tomorrow. The bookstore must have
as an appendix to the paper. these 160 books, but would like to have as many copies
of the book available for sale tomorrow as possible.
Given the short time horizon, the shipping warehouse
5. Measurable Quantities and Categories does its best. It has facilities to rush deliver up to ten
of Constraints boxes of books. The "paperback boxes" each hold 6
copies of the paperback. The warehouse also has 7
A measurable quantity is any physically meaningful "hardback boxes" on hand, but these boxes actually
quantity that can be identified as "the number of units contain 5 hardbacks and 2 paperbacks each. What
of something". The number of hours spent doing a job should the warehouse ship in order to best address
or the number employees working a given shift are the store's needs?
examples of measurable quantities. Sometimes an im-
portant quantity is best thought of as a percentage of The student begins in normal English, summarizing
another measurable quantity, such as "20% of total the problem as: "maximize books in the store tomor-
revenue". In those instances, we can broaden the term row"; "can't ship too many hardback boxes", "can't ship
"measurable quantity" to mean "measurable quantity more than 10 boxes" and "store has to have at least 80
or percentage thereof". of each book type". Most students will be able to reach
this point with relative ease. If not, students are encour-
Measurable quantities are natural building blocks for aged to look for key "constraint words" in the problem
LPs. The objective function of an LP represents a description, such as at most, at least, no more than, no
measurable quantity, as does each decision variable less than, up to, exactly, required, limited, restricted, must,
in its formulation. Each constraint in an LP represents cannot, budget, and demand. The objective is relatively
a relation between two measurable quantities, asserting easy to identify; it is the quantity to be made as large
either that they are equal, that the first is no greater as possible or as small as possible. The student then
5.3. Example 1 (continued) H is the number of hardback boxes, one more box gives
2 more paperbacks, and the term 2H in the linear ex-
The bookstore problem has two decision variables, P pression. The error is thus avoided.
and H, so students know that every measurable
quantity in the problem can be written as __P + __H + We have seen improvements in students' ability to
__. They must simply fill in the blanks with the appro- develop an LP spreadsheet model when they express
priate numbers. Informally, the question becomes "if the problem initially in terms of measurable quantities,
the variable goes up by one, by how much does the whether or not they write an algebraic model. The
measurable quantity go up?" For example, consider measurable quantity analysis suggests to the student
the objective, # of books in store, total: __ P + __ H + a sensible template for entering the program into Excel.
__. If P increases by 1, then the warehouse shipped The decision variables (changing cells) and the objec-
one more box of paperbacks. The measurable quantity tive (target cell) have been identified, and each pro-
- the total number of books obtained - increases by 6. gram requirement has been explicitly stated as a direct
On the other hand, if H increases by 1, then one more comparison between two measurable quantities. It
hardback box is shipped and the bookstore gains five would then be reasonable to have the values of these
hardbacks and two paperbacks. The measurable two quantities be reflected by two proximate cells in
quantity is # of books in store, total, and this increases the spreadsheet. The application of the Coefficient
by 5 + 2 = 7 as a result of the additional hardback box. Rule from Figure 3 in this context identifies the proper
Thus, the coefficient of H is 7. To finish, the constant multiplier for each changing cell in the constraint and
term is found by setting P and H equal to zero, which objective function formulas. Figure 5 shows one possi-
means that the warehouse does not ship at all. Then ble arrangement of the model elements, completed
how many books does the bookstore have, total? Only with the optimal solution to demonstrate the appropri-
what they already have on hand: 65 + 40 = 105 books. ate calculations. We follow a convention of red borders
Hence, the objective is 6P + 7H + 105. Applying the for changing cells, gray cells for data provided in the
same techniques to the rest of the measurable quanti- problem, yellow for the objective value, and white for
ties, one at a time, yields the LP formulation shown in computed values.
Figure 4. The program is completed by adding any
appropriate nonnegativity constraints ( P ≥ 0, H ≥
0).
Experience has shown that students who approach The pervasive error noted by Sharma (1987), the "re-
the bookstore problem without using this procedure versal of variables" is a special case of a more general
will commonly write the left-hand-side of the paper- error which we call "the recipe error". As an example,
back constraint as 6 P + 2 P + 40, since the two copies suppose that each Quality Control team consists of a
from the "hardback box" are actually paperback books. statistician, two inspectors, and two workers. Let T be
With our procedure, they ask, "if H goes up by 1, how the number of teams and S, I, and W represent the total
much does the # of paperbacks delivered go up?" Since number of statisticians, inspectors and workers on
INFORMS Transactions on Education 4:3(38-54) 45 © INFORMS ISSN: 1532-0545
STEVENS & PALOCSAY
A Translation Approach To Teaching Linear Program Formulation
these teams, respectively. Students commonly write need some partition of the variables into "essential in-
T = S + 2I + 2W. But the Coefficient Rule in Figure 3 formation" and "derived information". Once the parti-
shows this to be absurd. The constraint is measured tion into independent and auxiliary variables has been
in teams, so the term 2W, for example, says that obtain- carried out, correct formulation is assisted by invoking
ing one additional worker increases the number of the Auxiliary Rule given in Figure 7.
teams by two. Most students think of 2W as meaning
"two workers". A student using the Coefficient Rule
sees it properly as "two (of the measurable quantity -
teams) - for every worker". The Coefficient Rule thus
indirectly includes an analysis of units, showing that Figure 7: Constraints for Auxiliary Variables: the Auxiliary
the coefficient of W in this constraint would have to Rule
be measured in teams/worker, not in workers/team.
This constraint can often be sensibly viewed as a con-
The correct formulation, of course, is S = T, I = 2T, W
servation constraint, but it may also be defined as a
= 2T. The first constraint, in measurable quantities,
fourth category of constraint, the auxiliary variable
says "# of statisticians involved = # of statisticians on QC
definition. We use the next problem to demonstrate
teams ", and the other two are similar.
the incorporation of the Auxiliary Rule into the trans-
The recipe error often arises in problems that include lation approach.
decision variables whose values are entirely deter-
mined by the value of other variables, as in the QC 5.5. Example 2
team example. The inclusion of such dependent vari-
ables is not a mistake; indeed, "extra" variables may (From Winston and Albright, 2001) Rylon Corporation
improve program readability and clarify its structure, manufactures Brute and Channelle perfumes from a
as in many multiperiod planning problems. When raw material costing $3/pound. Processing 1 pound
using such variables, though, students often omit of the material takes 1 hour and yields 3 ounces of
needed program constraints. To address this problem, Regular Brute and 4 ounces of Regular Chanelle. This
we introduce the notion of "extra", or auxiliary variables perfume can be sold directly or can be further pro-
as defined in Figure 6. cessed to make luxury versions of the perfumes. One
ounce of Regular Brute can be sold for $7, but 3 hours
of processing will convert it into one ounce of Luxury
Brute, which sells for $18. One ounce of Regular
Chanelle can be sold for $6, but 2 hours of processing
will convert it into one ounce of Luxury Chanelle,
which sells for $14. Rylon has available 4000 pounds
of raw material, 6000 hours of processing time, and
will sell everything it produces. Determine how Rylon
can maximize its profit.
Blessing, S. B. and J. R. Anderson. (1996), "How People Evans, J. R. and J. D. Camm. (1990), "Using Pictorial
Learn to Skip Steps," Journal of Experimental Representations in Teaching Linear Program-
Psychology: Learning, Memory, and Cognition, ming Modeling," IIE Transactions, Vol. 22, No.
Vol. 22, No. 3, pp. 576-598. 2, pp. 191-195.
Bobrow, D. G. (1968), "Natural Language Input for a Gass, S. I. (1964), Linear Programming: Methods and Ap-
Computer Problem Solving System," in Seman- plications, McGraw-Hill Book Company: New
tic Information Processing, M. Minsky (Ed.), pp. York, NY.
132-215, MIT Press: Cambridge, MA. Glover, F., J. D. Klingman, and C. McMillan. (1977),
Booth, L. (1984), "Misconceptions Leading to Error in "The NETFORM Concept: A More Effective
Elementary Algebra," Journal of Structural Model Form and Solution Procedure for Large
Learning, Vol. 8, pp. 125-138. Scale Nonlinear Problems," Annual Proceedings
of the ACM, October 16-19, pp. 283-289.
Bopp, A., V. Kannan, S. Palocsay, S. Stevens. (1996),
"An Optimization Model for Planning Natural Grossman, T. A. (2002), "The Keys to the Vault,"
Gas Purchases, Transportation, Storage, and OR/MS Today, Vol. 29, No. 6, pp. 12-13.
Deliverability," Omega, International Journal of Hillier, F. S. and M. S. Hillier. (2003), Introduction to
Management Science, Vol. 24, No. 5, pp. 511-522. Management Science: A Modeling and Case Studies
Byrnes, J. P. and B. A. Wasik. (1991), "Role of Concep- Approach with Spreadsheets, McGraw-Hill/Irwin:
tual Knowledge in Mathematical Procedural Boston, MA.
Learning," Developmental Psychology, Vol. 27, Hinsley, D. A., J. R. Hayes, and H. A. Simon. (1977),
No. 5, pp. 777-786. "From Words to Equations Meaning and Rep-
resentation in Algebra Word Problems," in
Cognitive Processes in Comprehension, M. A. Just
Appendix
Linear Program Model Formulation Steps
1. Read and understand the problem. Ignore the numbers in the problem and concentrate on verbally identi-
fying the parts of the linear program:
• Objective (overall goal)
• Constraints (limitations and requirements)
• Decision variables (quantities under direct control)
Drawing a picture, making a table, or finding a trial solution may help you to better understand the problem
structure.
2. Write the measurable quantity formulation of the linear programming model. A measurable quantity is
the "# of units of something." Translate your English statement of the objective into a measurable quantity
and define each decision variable as a measurable quantity. Rewrite each constraint as a relation between
two measurable quantities, based on one of the following categories:
3.Translate the measurable quantity formulation into a mathematical model. Apply the Coefficient Rule de-
scribed below to each measurable quantity in the problem, using a "fill in the blanks" approach.
4. Add an equality constraint for each auxiliary variable. If you used any variables whose values are entirely
determined by the other variables in the problem, your final program must include one equality constraint (=)
for each such auxiliary variable, defining how it is calculated from those other variables.
5. Validate your model, by plugging in numbers to make sure your mathematical expressions are correct.