You are on page 1of 40

CHAPTER 22

Additional Network and LP Algorithms

22.1

MINIMUM-COST CAPACITATED FLOW PROBLEM The minimum-cost capacitated flow problem is based on the following assumptions: 1. 2. 3. 4. All arcs are directional (one way). A (nonnegative) unit flow cost is associated with each arc. Arcs may have positive lower-capacity limits. Any node in the network may act as a source or as a sink.

The new model determines the flows that minimize the total cost while satisfying the flow restrictions on the arcs and the supply and demand amounts at the nodes.We first present the capacitated network flow model and its equivalent linear programming formulation. The linear programming formulation is then used as the basis for the development of a special capacitated simplex algorithm for solving the network flow model. 22.1.1 Network Representation Consider a capacitated network G = (N, A), where N is the set of nodes and A is the set of arcs and define xij = amount of flow from node i to node j uij(lij) = upper (lower) capacity of arc (i, j) cij = unit flow cost from node i to node j fi = net flow at node i Figure 22.1 depicts these definitions on arc (i, j,). The label [fi] assumes a positive (negative) value when a net supply (demand) is associated with node i.
22.1

22.2

Chapter 22

Additional Network and LP Algorithms


[ fi] FIGURE 22.1 Capacitated arc with external flow i [ fj] j

$ cij (lij, uij) xij

[100]

$1 (50, 80) $4 [50] $5 3 $6 $1

[150] 4 $2

$3

(70, 120) $2 5 [80] $4

6 (100, 120)

[120]

FIGURE 22.2 Capacitated network for Example 22.1-1

[200]

Example 22.1-1
GrainCo supplies corn from three silos to three poultry farms. The supply amounts at the three silos are 100, 200, and 50 thousand bushels and the demands at the three farms are 150, 80, and 120 thousand bushels. GrainCo mostly uses railroads to transport the corn to the farms, with the exception of three routes where trucks are used. Figure 22.2 shows the available route between the silos and the farms. The silos are represented by nodes 1, 2, and 3 whose supply amounts are [100], [200], and [50], respectively.The farms are represented by nodes 4, 5, and 6 whose demand amounts are [ - 150], [ - 80], and [ - 120], respectively. The routes allow transshipping between the silos. Arcs (1, 4), (3, 4), and (4, 6) are truck routes. These routes have minimum and maximum capacities. For example, the capacity of route (1, 4) is between 50 and 80 thousand bushels. All other routes use trainloads, whose maximum capacity is practically unlimited. The transportation costs per bushel are indicated on the respective arcs.

PROBLEM SET 22.1A


*1. A product is manufactured to satisfy demand over a 4-period planning horizon according to the following data:

Period 1 2 3 4

Units of demand 100 110 95 125

Unit production cost ($) 24 26 21 24

Unit holding cost ($) 1 2 1 2

Given that no back-ordering is allowed, represent the problem as a network model.

22.1

Minimum-Cost Capacitated Flow Problem

22.3

2. In Problem 1, suppose that back-ordering is allowed at a penalty of $1.50 per unit per period. Formulate the problem as a network model. 3. In Problem 1, suppose that the production capacities of periods 1 to 4 are 110, 95, 125, and 100 units, respectively, in which case the given demand cannot be satisfied without back-ordering. Assuming that the penalty cost for back-ordering is $1.5 per unit per period, formulate the problem as a network model. 4. Daw Chemical owns two plants that manufacture a basic chemical compound for two customers at the rate of 660 and 800 tons per month. The monthly production capacity of plant 1 is between 400 and 800 tons and that of plant 2 is between 450 and 900 tons. The production costs per ton in plants 1 and 2 are $25 and $28, respectively. Raw material for the plants is provided by two suppliers, who are contracted to ship at least 500 and 700 tons per month for plants 1 and 2 at the costs of $200 and $210 per ton, respectively. Daw Chemical also pays for the transportation cost of both the raw material and the final compound. The costs per ton of transporting the raw material from supplier 1 to plants 1 and 2 are $10 and $12, respectively. Similar costs from supplier 2 are $9 and $13, respectively. The transportation costs per ton from plant 1 to clients 1 and 2 are $3 and $4, and from plant 2 are $5 and $2, respectively. Assuming that 1 ton of raw material produces 1 ton of the final compound, formulate the problem as a network model. 5. Two nonintegrated public schools are required to change the racial balance of their enrollments by accepting minority students. Minority enrollment must be between 30% and 40% in both schools. Nonminority students live in two communities, and minority students live in three other communities. Traveled distances, in miles, from the five communities to the two schools are summarized in the following table:

Round-trip miles from school to Minority areas School Maximum enrollment 1 20 15 500 2 12 18 450 3 10 8 300 Nonminority areas 1 4 6 1000 2 5 5 1000

1500 1 2 2000 Student population

Formulate the problem as a network model to determine the number of minority and nonminority students enrolled in each school. 6. Charnes and Cooper (1967) During the next three months, farmers harvest their crops and sell it to a local silo. The silo in turn can sell the crop to retailers. The following table provides the purchasing cost and selling price per ton during the harvest season.

Month 1 2 3

Purchasing cost per ton ($) 200 190 195

Selling price per ton ($) 250 230 240

22.4

Chapter 22

Additional Network and LP Algorithms

The maximum capacity of the silo is 800 tons. The owner has $100, 000 cash on hand which can be used to purchase new crops. Also, initially, at the start of month 1, the silo is half full. It is estimated that the storage cost per ton per month is $10. Surplus cash earns 1% interest monthly. The objective is to determine the buy/sell policy the owner must follow to maximize the total cash accumulation at the end of the three-month season. Formulate the problem as a network model.

22.1.2 Linear Programming Formulation The formulation of the capacitated network model as a linear program provides the foundation for the development of the capacitated simplex algorithm, which we will present in the next section. Using the notation introduced in Section 22.1.1, the linear program for the capacitated network is given as Minimize z = a a cijxij
(i, j) H A

subject to
k (j, k) H A

a xjk - a xij = fj, j H N


i (i, j) H A

lij xij uij, (i, j) A The equation for node j measures the net flow fj in node j as (Outgoing flow from node j) - (Incoming flow into node j) = fj Node j acts as a source if fj 7 0 and as a sink if fj 6 0. We can always remove the lower bound lij from the constraints by using the substitution
xij = xij + lij The new flow variable, xij , has an upper limit of uij - lij. Additionally, the net flow at node i becomes fi - lij, and that at node j is fj + lij. Figure 22.3 shows the transformation of activity (i, j) after the lower bound is substituted out.

Example 22.1-2
Write the linear program for the network in Figure 22.2, before and after the lower bounds are substituted out.
FIGURE 22.3 Removal of lower bound in arcs [ fi] i [ fj] j [ fi lij] i [ fj lij] j

$ cij (lij, uij) xij

$ cij (uij lij) xij

22.1

Minimum-Cost Capacitated Flow Problem

22.5

The main constraints of the linear program relate the input and output flow for each node, which yields the following LP:
x12 Minimize Node 1 Node 2 Node 3 Node 4 Node 5 Node 6 Lower bounds Upper bounds 3 1 -1 x13 4 1 -1 -1 -1 0 q 0 q 50 80 0 q 0 q 70 120 x14 1 1 1 -1 1 1 -1 1 1 -1 -1 0 q 0 q 1 -1 100 120 x23 5 x25 6 x34 1 x35 2 x46 2 x56 4 = 100 = 200 = 50 = - 150 = - 80 = - 120

Note the arrangement of the constraints coefficients. The column associated with variable xij has exactly one + 1 in row i and one - 1 in row j, a typical property of network models. The rest of the coefficients are all 0. This structure is typical of network flow models. The optimum solution is z = $1, 870, 000 with x13 = 20 (thousand bushels), x14 = 80 , x23 = 20, x25 = 180, x34 = 90, x46 = 20, and x56 = 100. All the remaining variables are zero. The variables with lower bounds are substituted as
x14 = x14 + 50 x34 = x34 + 70 x56 = x56 + 100

The substitution leads to the following LP:


x12 Minimize Node 1 Node 2 Node 3 Node 4 Node 5 Node 6 Upper bounds 3 1 -1 x13 4 1 -1 -1 -1 q q 30 q q 50
x14

x23 5

x25 6

x34

x35 2

x46 2

x56

1 1

4 = 50 = 200 = - 20 = - 30 = - 180 = - 20

1 -1

1 1 -1 1 1 -1 q -1 q 1 -1 20

The corresponding network after substituting out the lower bounds is shown in Figure 22.4. Note that the lower-bound substitution can be done directly from Figure 22.2 using the substitution in Figure 22.3 and without the need to express the problem as a linear program first. The optimum solution is z = $1, 350 thousand (or z = 1350 + 50 * 1 + 70 * 1 + 100 * 4 = $1870 thousand) with x13 = 20 (thousand bushels), x14 = 30 (or x14 = 30 + 50 = 80 ), x25 = 180 , x34 = 20 (or x34 = 20 + 70 = 90 ), x46 = 20 , x56 = 0 (or x56 = 0 + 100 = 100 ), which, of course, is the same solution given by the before-substitution solution.

22.6

Chapter 22

Additional Network and LP Algorithms


[30] 4 $2 $1 [20] 3 (50) $2 5 [180] $4 (20) 6 [20]

[50]

$1 (30) $4

$3

FIGURE 22.4 Network of Example 22.1-2 after substituting out lower bounds [200] 2

$5 $6

Example 22.1-3

(Employment Scheduling)

This example illustrates a network model that initially does not satisfy the node flow requirement (i.e., node output flow less node input flow equals node net flow), but that can be converted to this form readily through special manipulation of the constraints of the linear program. Tempo Employment Agency has a contract to provide workers over the next 4 months (January to April) according to the following schedule:
Month No. of workers January 100 February 120 March 80 April 170

Because of change in demand, it may be economical to retain more workers than needed in a given month. The cost of recruiting and maintaining a worker depends on the length of the employment period, as the following table shows:
Employment period (months) Cost per worker ($) 1 100 2 130 3 180 4 220

Let xij = number of workers hired at the start of month i and terminated at the start of month j For example, x12 gives the number of workers hired in January for 1 month only. To formulate the problem as a linear program for the 4-month period, we add May as a dummy month (month 5), so that x45 defines hiring in April for April. The constraints recognize that the demand for period k can be satisfied by all xij such that i k 6 j. Letting si 0 be the surplus number of workers in month i, the linear program is given as
x12 Minimize January February March April 100 1 x13 130 1 1 x14 180 1 1 1 x15 220 1 1 1 1 x23 100 x24 130 x25 180 x34 100 x35 130 x45 100 -1 1 1 1 1 1 1 -1 1 1 1 -1 1 -1 = = = = 100 120 80 170 s1 s2 s3 s4

22.1

Minimum-Cost Capacitated Flow Problem

22.7

The preceding LP does not have the ( - 1, + 1) special structure of the network flow model (see Example 22.1-2). Nevertheless, the given linear program can be converted into an equivalent network flow model by using the following arithmetic manipulations: 1. In an n-equation linear program, create a new equation, n + 1, by multiplying equation n by - 1. 2. Leave Equation 1 unchanged. 3. For i = 2, 3, . . . , n, replace each equation i with (equation i) - (equation i - 1). The application of these manipulations to the employment scheduling example yields the following linear program, whose structure fits the network flow model:
x12 Minimize January February March April May 100 1 -1 x13 130 1 -1 -1 -1 x14 180 1 x15 220 1 1 -1 1 -1 -1 1 1 -1 1 -1 1 -1 x23 100 x24 130 x25 180 x34 100 x35 130 x45 100 -1 1 = 100 20 = = - 40 90 = = - 170 s1 s2 s3 s4

-1 1

-1 1

-1 1

Using the preceding formulation, the employment scheduling model can be represented equivalently by the minimum cost flow network shown in Figure 22.5. Actually, because the arcs have no upper bounds, the problem can be solved also as a transshipment model (see Appendix to this section). The optimum solution is x15 = 100, x25 = 20, x45 = 50, and all the remaining variables are zero. The following table summarizes the hiring and firing activities over the 4-month horizon. The total cost is $30, 600.
Month Number of workers hired 100 20 0 50 Number of workers fired 0 0 0 0 Net number of workers 100 120 120 170

January February March April FIGURE 22.5

Network representation of employment scheduling problem x15 x14 x13 x12 1 [100] s1 2 [20] s2 x23 x34 3 [40] s3 x24 x25 4 [90] x35 x45 s4 5 [170]

22.8

Chapter 22

Additional Network and LP Algorithms

PROBLEM SET 22.1B


*1. Write the linear program associated with the minimum-cost flow network in Figure 22.6, before and after the lower bounds are substituted out. 2. Use inspection to find a feasible solution to the minimum-cost network model of the employment scheduling problem in Example 22.1-3 (Figure 22.5). Interpret the solution by showing the pattern of hiring and firing that satisfies the demand for each month, and compute the associated total cost. 3. Reformulate the employment scheduling model of Example 22.1-3, assuming that a worker must be hired for at least 2 months. Write the linear program, and convert it to a minimum-cost flow network. 4. Develop the linear program and the associated minimum-cost flow network for the employment scheduling model of Example 22.1-3 using the following 5-month demand data. The per-worker costs of recruiting and maintaining a worker for periods of 1 to 5 months are $50, $70, $85, $100, and $130, respectively. (a)
Month Number of workers 1 300 2 180 3 90 4 170 5 200

(b)
Month Number of workers 1 200 2 220 3 300 4 50 5 240

5. Conversion of a capacitated network into an uncapacitated network. Show that an arc (i : j) with capacitated flow xij uij can be replaced with two uncapacitated arcs (i : k) and (j : k) with a net (output) flow of [ - uij] at node k and an additional (input) flow of [ + uij] at node j. The result is that the capacitated network can be converted to an uncapacitated transportation model (Section 5.1). Apply the resulting transformation to the network in Figure 22.7, and find the optimum solution to the original network by solving the uncapacitated transportation model.
FIGURE 22.6 Network for Problem 1, Set 22.1b
(3 $5 0, 40 )

[20] 3
$6 ) , (0

$4 (10, ) [50] 1 $1 (0, ) 2 [40] $3 (10, ) 4 [30]

22.1
2

Minimum-Cost Capacitated Flow Problem

22.9

[100]

$1 ) (60

$8 (50 )
4
$1 0) (9

[100]

$4 ( )

$2 (50)

FIGURE 22.7 3 Network for Problem 5, Set 22.1b

22.1.3 Capacitated Network Simplex Algorithm The algorithm is based on the exact steps of the regular simplex method, but it is designed to exploit the special network structure of the minimum-cost flow model. Given that fi is the net flow at node i as defined in the linear program of Section 22.1.2, the capacitated simplex algorithm stipulates that the network must satisfy
n i=1

a fi = 0

The condition says that the total supply in the network equals the total demand. We can always satisfy this requirement by adding a balancing dummy source or destination, which we connect to all other nodes in the network by arcs with zero unit cost and infinite capacity. However, the balancing of the network does not guarantee feasibility, as this depends on the capacities of the arcs. We now present the steps of the capacitated algorithm. Familiarity with the simplex method and duality theory (Chapters 3 and 4) is essential. Also, knowledge of the upper-bounded simplex method (Section 7.3) is helpful. Step 0. Step 1. Step 2. Determine a starting basic feasible solution (set of arcs) for the network. Go to step 1. Determine an entering arc (variable) using the simplex method optimality condition. If the solution is optimal, stop; otherwise, go to step 2. Determine the leaving arc (variable) using the simplex method feasibility condition. Determine the new solution, then go to step 1.

An n-node network with zero net flow (i.e., f1 + f2 + + fn = 0) consists of n - 1 independent constraint equations. Thus, a basic solution must include n - 1 arcs. It can be proved that a basic solution always corresponds to a spanning tree of the network (see Section 6.2). The entering arc (step 1) is determined by computing the objective coefficients for all the current nonbasic arcs ( i, j). If all the objective coefficients are 0, the current basis is optimum. Otherwise, we select the nonbasic arc with the most positive objective coefficients to enter the basis. The computation of objective coefficients is based on duality, exactly as we did with the transportation model (see Section 5.3.2). Using the linear program defined in

22.10

Chapter 22

Additional Network and LP Algorithms

Section 22.1.2, let wi be the dual variable associated with the constraint of node i, then the dual problem (excluding the upper bounds) is given as
n

Maximize z = a fiwi
i=1

subject to wi - wj cij, (i, j) H A wi unrestricted in sign, i = 1, 2, , n From the theory of linear programming, we have wi - wj = cij, for basic arc (i, j) Because the original linear program (Section 22.1.2) has one redundant constraint by definition, we can assign an arbitrary value to one of the dual variables (compare with the transportation algorithm, Section 5.3). For convenience, we will set w1 = 0. We then solve the (basic) equations wi - wj = cij to determine the remaining dual values. From Section 4.2.4, Formula 2, we know that the objective coefficient of nonbasic xij is the difference between the left-hand side and the right-hand side of the associated dual constraint, that is, wi - wj - cij. Letting zij = wi - wj the objective coefficient of the variable xij is computed as zij - cij = wi - wj - cij The only remaining detail is to show how the leaving variable is determined. The following numeric example provides this detail. Example 22.1-4
A network of pipelines connects two water desalination plants to two cities. The daily supply amounts at the two plants are 40 and 50 million gallons, and the daily demand amounts at cities 1 and 2 are 30 and 60 million gallons, respectively. Nodes 1 and 2 represent plants 1 and 2, respectively, and nodes 4 and 5 represent cities 1 and 2, respectively. Node 3 is a booster station between the plants and the cities. The model is already balanced because the sum of the supply at nodes 1 and 2 equals the sum of the demand at nodes 4 and 5. Figure 22.8 gives the associated network. Iteration 0 Step 0. Determination of a Starting Basic Feasible Solution: The starting basic solution must be a spanning tree. The feasible spanning tree in Figure 22.9 (shown with solid arcs) is obtained by inspection. Normally, we use an artificial-variable technique to find such a solution (for details, see Bazaraa et al., 1990, pp. 440446). In Figure 22.9, the basic feasible solution consists of (solid) arcs (1, 3), (1, 4), (2, 3), and (3, 5) with the feasible flows of 30, 10, 50, and 60 units, respectively. The remaining (dashed) arcs (1, 2), (2, 5), and (4, 5) represent the nonbasic variables. The notation x(c) shown on the arcs indicates that a flow of x units is assigned to an arc with capacity c. The default values for x and c are 0 and q, respectively.

22.1
Unit cost Plant 1 [40] 1
$7 ) 0 (1

Minimum-Cost Capacitated Flow Problem

22.11

Arc capacity $5 4 (35)

[30] City 1

$3 ()
$ (6 2 0)

Plant 2 [50]

( ) $1 (30)

$8

$4 () FIGURE 22.8 Network for Example 22.1-4

[60] City 2

w1 0 [40] 1
$7 0) (1 10

$5 30(35) w3 7 3
50 $2 (6 0)

w4 5 4 [30] z12 c12 0 (5) 3 2 z25 c25 5 (15) 1 9 z45 c45 5 (15) 4 6 Arc (2, 5) reaches upper bound at 30. Substitute x25 30 x52. Reduce x23 and x35 each by 30. 5 [60] w5 15

$3

$4 ()

$ 60 8 ( )

[50] 2 w2 5 FIGURE 22.9 Network for iteration 0

$1 (30)

Iteration 1 Step 1. Determination of the entering arc: We obtain the dual values by solving the current basic equations w1 = 0 wi - wj = cij, for basic arc (i, j) We thus get, Arc(1, 3): w1 - w3 = 7, hence w3 = - 7 Arc(1, 4): w1 - w4 = 5, hence w4 = - 5 Arc(2, 3): w2 - w3 = 2, hence w2 = - 5 Arc(3, 5): w3 - w5 = 8, hence w5 = - 15 Now, we compute zij - cij for the nonbasic arcs as Arc (1, 2): w1 - w2 - c12 = 0 - ( - 5) - 3 = 2 Arc (2, 5): w2 - w5 - c25 = ( - 5) - ( - 15) - 1 = 9 Arc (4, 5): w4 - w5 - c45 = ( - 5) - ( - 15) - 4 = 6 Thus, arc (2, 5) enters the basic solution.

22.12

Chapter 22 Step 2.

Additional Network and LP Algorithms

Determination of the leaving arc: From Figure 22.9, arc (2, 5) forms a loop with basic arcs (2, 3) and (3, 5). From the definition of the spanning tree, no other loop can be formed. Because the flow in the new arc (2, 5) must be increased, we adjust the flow in the arcs of the loop by an equal amount to maintain the feasibility of the new solution. To achieve this, we identify the positive ( + ) flow in the loop by the direction of flow of the entering arc (i.e., from 2 to 5). We then assign ( + ) or ( - ) to the remaining arcs of the loop, depending on whether the flow of each arc is with or against the direction of flow of the entering arc. These sign conventions are shown in Figure 22.9. Determination of the maximum level of flow in the entering arc (2, 5) is based on two conditions: 1. New flow in the current basic arcs of the loop cannot be negative. 2. New flow in the entering arc cannot exceed its capacity. The application of condition 1 shows that the flows in arcs (2, 3) and (3, 5) cannot be decreased by more than min{50, 60} = 50 units. By condition 2, the flow in arc (2, 5) can be increased to at most the arc capacity ( = 30 units). Thus, the maximum flow change in the loop is min{30, 50} = 30 units. The new flows in the loop are thus 30 units in arc (2, 5), 50 - 30 = 20 units in arc (2, 3), and 60 - 30 = 30 units in arc (3, 5). Because none of the current basic arcs leave the basis at zero level, the new arc (2, 5) must remain nonbasic at upper bound. However, to avoid dealing with nonbasic arcs that are at capacity (or upper bound) level, we implement the substitution x25 = 30 - x52, 0 x52 30 This substitution is effected in the flow equations associated with nodes 2 and 5 as follows. Consider Current flow equation at node 2: 50 + x12 = x23 + x25 Current flow equation at node 5: x25 + x35 + x45 = 60 Then, the substitution x25 = 30 - x52 gives New flow equation at node 2: 20 + x12 + x52 = x23 New flow equation at node 5: x35 + x45 = x52 + 30 The results of these changes are shown in Figure 22.10. The direction of flow in arc (2, 5) is now reversed to 5 : 2 with x52 = 0 , as desired. The substitution also requires changing the unit cost of arc (5, 2) to - $1. We will indicate this direction reversal on the network by tagging the arc capacity with an asterisk.

Iteration 2. Figure 22.10 summarizes the new values of zij - cij (verify!) and shows that arc (4, 5) enters the basic solution. It also defines the loop associated with the new entering arc and assigns the signs to its arcs. The flow in arc (4, 5) can be increased by the smallest of 1. Maximum allowable increase in entering arc (4, 5) = q 2. Maximum allowable increase in arc (1, 4) = 35 - 30 = 5 units

22.1
w1 0 [40] 1
$7 0) (1 10

Minimum-Cost Capacitated Flow Problem

22.13

$5 30(35) w 7 3 3
20 $2 (6 0)
$ 30 8 ( )

w4 5 4 [30] z12 c12 0 (5) 3 2 z52 c52 15 (5) (1) 9 z45 c45 5 (15) 4 6 Arc (4, 5) enters at level 5. Arc (1, 4) leaves at upper bound. 5 [30] w5 15 Substitute x14 35 x41. Reduce x13 and x35 each by 5.

$3

$4 ()

[20] 2 w2 5 FIGURE 22.10 Network for iteration 1

$1 (30)*

3. Maximum allowable decrease in arc (1, 3) = 10 units 4. Maximum allowable decrease in arc (3, 5) = 30 units Thus, the flow in arc (4, 5) can be increased to 5 units, which will make (4, 5) basic and will force basic arc (1, 4) to be nonbasic at its upper bound ( = 35). Using the substitution x14 = 35 - x41 , the network is changed as shown in Figure 22.11, with arcs (1, 3), (2, 3), (3, 5), and (4, 5) forming the basic (spanning tree) solution. The reversal of flow in arc (1, 4) changes its unit cost to - $5. Also, convince yourself that the substitution in the flow equations of nodes 1 and 4 will net 5 input units at each node. Iteration 3. The computations of the new zij - cij for the nonbasic arcs (1, 2), (4, 1), and (5, 2) are summarized in Figure 22.11, which shows that arc (1, 2) enters at level 5, and arc (1, 3) becomes nonbasic at level 0. Figure 22.12 depicts the new solution. Iteration 4. The new zij - cij in Figure 22.12 shows that the solution is optimum. Backsubstitution yields the values of the original variables, as shown in Figure 22.12.
FIGURE 22.11 Network for iteration 2 w1 0 [5] 1
$7 0) 1 5(

$5 (35)* w3 7 3
$ 25 8 ( )

w4 11 4 [5] z12 c12 0 (5) 3 2 z41 c41 11 0 (5) 6 z52 c52 15 (5) (1) 9 Arc (1, 2) enters at level 5. Arc (1, 3) leaves at level 0. Increase x23 by 5. 5 [30] w5 15

$3


20 $2 (6 0)

$4 5()

[20] 2 w2 5

$1 (30)*

22.14

Chapter 22
w1 0 [5] 1

Additional Network and LP Algorithms


w4 9 4 [5]

$5 (35)* w3 5 3
$ 25 8 ( )

z13 c13 0 (5) 7 2 z41 c41 9 0 (5) 4 z52 c52 13 (3) (1) 9 Optimum solution: x12 5, x13 0 x14 35 0 35 x23 25 x25 30 0 30 x35 25, x45 5 Total cost $490

$3 5()
25 $2 (6 0)

[20] 2 w2 3 FIGURE 22.12 Network for iteration 3

Solver Moment
The basic idea of the Excel Solver minimum cost capacitated model is similar to the one detailed in Example 6.3-6 for the shortest-route model. File solverEx22.1-4.xls provides the details.

AMPL Moment
File amplEx22.1-4.txt is a general model that can be used to solve the minimum cost capacitated problem of any size. The idea of the model is similar to that of the shortest-route AMPL model for Example 6.3-6.

PROBLEM SET 22.1C


*1. Solve Problem 1, Set 22.1a, by the capacitated simplex algorithm. 2. Solve Problem 2, Set 22.1a, by the capacitated simplex algorithm, and also show that it can be solved by the transshipment model. 3. Solve Problem 3, Set 22.1a, by the capacitated simplex algorithm. 4. Solve Problem 4, Set 22.1a, by the capacitated simplex algorithm. *5. Solve Problem 5, Set 22.1a, by the capacitated simplex algorithm. 6. Solve the employment scheduling problem of Example 22.1-3 by the capacitated simplex algorithm. 7. Wyoming Electric uses existing slurry pipes to transport coal (carried by pumped water) from three mining areas (1, 2, and 3) to three power plants (4, 5, and 6). Each pipe can transport at most 10 tons per hour. The transportation costs per ton and the supply and demand per hour are given in the following table:
4 1 2 3 Demand $5 $6 $3 16 5 $8 $9 $1 6 6 $4 $12 $5 14 Supply 8 10 18

Determine the optimum shipping schedule.

$7 ) 0 (1

$4 5()

$1 (30)*

5 [30] w5 13

22.1
2 2 10

Minimum-Cost Capacitated Flow Problem


5 5 8 6

22.15

11

1 4

4 3 1 7 9

3 FIGURE 22.13 Network for Problem 8, Set 22.1c

8. The network in Figure 22.13 gives the distances among seven cities. Use the capacitated simplex algorithm to find the shortest distance between nodes 1 and 7. (Hint: Assume that nodes 1 and 7 have net flows of [ + 1] and [ - 1], respectively. All the other nodes have zero net flow.) 9. Show how the capacitated minimum-cost flow model can be specialized to represent the maximum flow model of Section 6.4. Apply the transformation to the network in Example 6.4-2. For convenience, assume that the flow capacity from 4 to 3 is zero. All the remaining data are unchanged. 10. Solve the following problems using Excel Solver and AMPL: (a) Problem 3, Set 22.1a. (b) Problem 4, Set 22.1a. (c) Problem 7, Set 22.1c. (d) Problem 8, Set 22.1c.

APPENDIX: Transshipment Model The transshipment model is an old technique that converts minimum cost transportation network to the regular transportation model presented in Chapter 5. This is a special case of the highly efficient minimum cost network presented above in that it does not allow capacity restrictions on the shipping routes. The transshipment model is rarely used in practice because it is not as efficient as the general network model. It is presented here, however, because it demonstrates the interesting use of buffers to convert transient shipping to direct shipping. Example 22.1.5
Two automobile plants, P1 and P2, are linked to three dealers, D1, D2, and D3, by way of two transit centers, T1 and T2, according to the network shown in Figure 22.14. The supply amounts at plants P1 and P2 are 1000 and 1200 cars, respectively, and the demand amounts at dealers D1, D2, and D3, are 800, 900, and 500 cars, respectively. The shipping costs per car (in hundreds of dollars) between pairs of nodes are shown on the connecting links (or arcs) of the network.

22.16

Chapter 22

Additional Network and LP Algorithms


D1 8 1000 P1 4 2 1200 P2 5 7 4 T2 9 3 D3 500 3 T1 6 5 D2 900 800

FIGURE 22.14 Transshipment network between plants and dealers

Transshipment occurs in the network in Figure 5.7 because the entire supply amount of 2200 (= 1000 + 1200) cars at nodes P1 and P2 could conceivably pass through any node of the network before ultimately reaching their destinations at nodes D1, D2, and D3. In this regard, each node of the network with both input and output arcs (T1, T2, D1, and D2) acts as both a source and a destination and is referred to as a transshipment node. The remaining nodes are either pure supply nodes (P1 and P2) or pure demand nodes (D3). The transshipment model can be converted into a regular transportation model with 6 sources (P1, P2, T1, T2, D1, and D2) and five destinations (T1, T2, D1, D2, and D3). The amounts of supply and demand at the different nodes are computed as Supply at a pure supply node = Original supply Demand at a pure demand node = Original demand Supply at a transshipment node = Original supply + Buffer amount Demand at a transshipment node = Original demand + Buffer amount The buffer amount should be sufficiently large to allow all of the original supply (or demand) units to pass through any of the transshipment nodes. Let B be the desired buffer amount; then B = Total supply (or demand) = 1000 + 1200 (or 800 + 900 + 500) = 2200 cars Using the buffer B and the unit shipping costs given in the network, we construct the equivalent regular transportation model as in the table below.
T1 P1 P2 T1 T2 D1 D2 B 3 2 0 M M M B T2 4 5 7 0 M M D1 M M 8 M 0 M D2 M M 6 4 5 0 500 D3 M M M 9 M 3 1000 1200 B B B B

800 + B 900 + B

22.1
D1

Minimum-Cost Capacitated Flow Problem


800

22.17

800
1000 P1
10

T1
00

400

D2
00

900

0 100
T2 500 FIGURE 22.15 D3 500 Solution of the transshipment model

1200

P2

The solution of the resulting transportation model (determined by TORA) is shown in Figure 22.15. Note the effect of transshipment: Dealer D2 receives 1400 cars, keeps 900 cars to satisfy its demand, and sends the remaining 500 cars to dealer D3.

PROBLEM SET 22.1D1


1. The network in Figure 22.16 gives the shipping routes from nodes 1 and 2 to nodes 5 and 6 by way of nodes 3 and 4. The unit shipping costs are shown on the respective arcs. (a) Develop the corresponding transshipment model. (b) Solve the problem, and show how the shipments are routed from the sources to the destinations. 2. In Problem 1, suppose that source node 1 can be linked to source node 2 with a unit shipping cost of $1. The unit shipping cost from node 1 to node 3 is increased to $5.
1 4 3 200 2 2 1 4 3 5 8 1 6 FIGURE 22.16 Network for Problem 1, Set 22.D

12

100

150

150 FIGURE 22.17 Network for Problem 3, Set 22.1D

900

1 1 .3 .8 4 .5 4.3 2 4.6 5 6 3 2.1 1.9 .2 4.5

1100

1400

1000

1000

1200

You are encouraged to use TORA, Excel Solver, or AMPL to solve the problems in this set.

22.18

Chapter 22

Additional Network and LP Algorithms

Formulate the problem as a transshipment model, and find the optimum shipping schedule. 3. The network in Figure 22.17 shows the routes for shipping cars from three plants (nodes 1, 2, and 3) to three dealers (nodes 6 to 8) by way of two distribution centers (nodes 4 and 5). The shipping costs per car (in $100) are shown on the arcs. (a) Solve the problem as a transshipment model. (b) Find the new optimum solution assuming that distribution center 4 can sell 240 cars directly to customers. 4. Consider the transportation problem in which two factories supply three stores with a commodity. The numbers of supply units available at sources 1 and 2 are 200 and 300, respectively; those demanded at stores 1, 2, and 3 are 100, 200, and 50, respectively. Units may be transshipped among the factories and the stores before reaching their final destination. Find the optimal shipping schedule based on the unit costs in the following table.
Factory 1 Factory 1 2 1 Store 2 3 $0 $6 $7 $1 $8 2 $6 $0 $2 $5 $9 1 $7 $5 $0 $1 $7 Store 2 $8 $4 $5 $0 $6 3 $9 $3 $1 $4 $0

5. Consider the oil pipeline network shown in Figure 22.18. The different nodes represent pumping and receiving stations. Distances in miles between the stations are shown on the network. The transportation cost per gallon between two nodes is directly proportional to the length of the pipeline. Develop the associated transshipment model, and find the optimum solution. 6. Shortest-Route Problem. Find the shortest route between nodes 1 and 7 of the network in Figure 22.19 by formulating the problem as a transshipment model. The distances between the different nodes are shown on the network. (Hint: Assume that node 1 has a net supply of 1 unit, and node 7 has a net demand also of 1 unit.)
50,000 60,000 (gallons) FIGURE 22.18 Network for Problem 5, Set 22.1D 1 20 2 90,000 3 7 9 3 30 4 20,000

40 8

10 6

2 5

22.1

Minimum-Cost Capacitated Flow Problem


4 4

22.19

1 2 5 1 3 4 3 6 7 5 9

2 5 8 7

1 2 3 5

7 3 6

FIGURE 22.19 Network for Problem 6, Set 22.1D

7. In the transshipment model of Example 5.5-1, define xij as the amount shipped from node i to node j. The problem can be formulated as a linear program in which each node produces a constraint equation. Develop the linear program, and show that the resulting formulation has the characteristic that the constraint coefficients, aij , of the variable xij are aij =
- 1,

1, in constraint i in constraint j 0, otherwise

8. An employment agency must provide the following laborers over the next 5 months:
Month Number of laborers 1 100 2 120 3 80 4 170 5 50

Because the cost of labor depends on the length of employment, it may be more economical to keep more laborers than needed during some months of the 5-month planning horizon. The following table estimates the labor cost as a function of the length of employment:
Months of employment Cost per laborer ($) 1 100 2 130 3 180 4 220 5 250

Formulate the problem as a linear program. Then, using proper algebraic manipulations of the constraint equations, show that the model can be converted to a transshipment model, and find the optimum solution. (Hint: Use the transshipment characteristic in Problem 7 to convert the constraints of the scheduling problem into those of the transshipment model.)

22.20

Chapter 22

Additional Network and LP Algorithms

22.2

DECOMPOSITION ALGORITHM Consider the situation of developing a master corporate plan for several production facilities. Although each facility has its own independent capacity and production constraints, the different facilities are linked together at the corporate level by budgetary considerations. The resulting model includes two types of constraints: common, representing the corporate budgetary constraints, and independent, representing the internal capacity and production restrictions of each facility. Figure 22.20 depicts the layout of the resulting constraints for n activities (facilities). In the absence of the common constraints, all activities operate independently. The decomposition algorithm improves the computational efficiency of the problem depicted in Figure 22.20 by breaking it down into n subproblems that can be solved almost independently. We point out, however, that the need for the decomposition algorithm was more justifiable in the past when the speed and memory of the computer were modest. Todays impressive computer capabilities obviate the use of the decomposition algorithm. Nevertheless, we present the algorithm here because of its interesting theoretical contribution. The corresponding mathematical model is given as Maximize z = C1X1 + C2X2 + + CnXn subject to A1X1 + A2X2 + + AnXn b0 D1X1 D2X2 o b1 b2 o DnXn bn Xj 0, j = 1, 2, , n The slack and surplus variables are added as necessary to convert all the inequalities into equations.
FIGURE 22.20 Layout of a decomposable linear program Activity 1 Independent constraints Activity 2 Common constraints

Activity n

22.2

Decomposition Algorithm

22.21

The decomposition principle is based on representing the entire problem in terms of the extreme points of the sets {X|DjXj bj, Xj 0}, j = 1, 2, , n . To do so, the solution space described by each {X|DjXj bj, Xj 0} must be bounded. This requirement can always be satisfied for any set j by adding the artificial restriction 1Xj M , where M is sufficiently large. N , k = 1, 2, , K define the extreme points of {X|D X b , X 0}. We Let X jk j j j j j then have N , j = 1, 2, , n Xj = a b jkX jk
k=1 kj Kj

where b jk 0 for all k and a b jk = 1


k=1

We can reformulate the entire problem in terms of the extreme points to obtain the following master problem: N b + N N + Maximize = a C1X 1k 1k a C2X2k b 2k + a CnXnk b nk
k=1 k=1 k=1 K1 K2 Kn

subject to N N N + a A1X1k b 1k + a A2X2k b 2k + a AnXnk b nk b0


k=1 k=1 Kj k=1 K1 K2 Kn

k=1

a b 1k
Kj k=1

= 1 a b 2k o
Kj k=1

= 1 o a b nk = 1

b jk 0, for all j and k The new variables in the master problem are bjk . Once their optimal values, * bj , are determined, we can find the optimal solution to the original problem by k back-substitution: N Xj = a b jkXjk,
k=1 Kj

j = 1, 2, , n

It may appear that the solution of the master problem requires prior determination N , a difficult task indeed! Fortunately, it is not so. of all the extreme points X jk

22.22

Chapter 22

Additional Network and LP Algorithms

To solve the master problem by the revised simplex method (Section 7.2), we need to determine the entering and the leaving variables at each iteration. For the entering variable, given CB and B-1 of the current basis of the master problem, then for nonbasic b jk, we have zjk - cjk = CBB-1Pjk - cjk where N AjX jk 0 o = 1 o 0

N and P cjk = CjX jk jk

Now, to decide which, if any, of the nonbasic variables b jk should enter the solution, we need to determine zj*k* - cj*k* =
all j and k

min {zjk - cjk}

If zj*k* - cj*k* 7 0, then, according to the maximization optimality condition, b j*k* must enter the solution; otherwise, the optimum has been reached. We still have not shown how zj*k* - cj*k* is computed numerically. The idea lies in the following identity min {zjk - cjk} = min e min{zjk - cjk} f
j k

all j and k

The reason we are able to establish this identity is that each convex set {X|Dj Xj bj, Xj 0} has its independent set of extreme points. In effect, what the identity says is that we can determine zj*k* - cj*k* in two steps: Step 1. Step 2. For each convex set {X|DjXj bj, Xj 0}, determine the extreme point N Y jk* that yields the smallest zjk - cjkthat is, zjk* - cjk* = mink{zjk - cjk}. Determine zj*k* - cj*k* = min{zjk* - cjk*}
j

From LP theory, we know that the optimum solution, when finite, must be associated with an extreme point of the solution space. Because each of the sets {X|DjXj bj, Xj 0} is bounded by definition, step 1 is mathematically equivalent to solving n linear programs of the form Minimize wj = {zj - cj|DjXj bj, Xj 0} The objective function wj is a linear function in Xj (see Problem 8, Set 22.2a).

22.2

Decomposition Algorithm

22.23

The determination of the entering variable b j*k* in the master problem reduces to solving n (smaller) linear programs to determine the entering extreme point N Y j*k*. This approach precludes the need to determine all the extreme points of all n convex sets. Once the desired extreme point is located, all the elements of the column vector Pj*k* are defined. We can then determine the leaving variable, and subsequently compute the next B-1 using the revised simplex method computations. Example 22.2-1
Solve the following LP by the decomposition algorithm: Maximize z = 3x1 + 5x2 + x3 + x4 subject to x1 + x2 + x3 + x4 40 5x1 + x2 12 x3 + x4 5 x3 + 5x4 50 x1, x2, x3, x4 0 The problem has two subproblems that correspond to the following sets of variables: X1 = (x1, x2)T, X2 = (x3, x4)T The associated master problem is thus given as:

Subproblem 1 b 11 N C 1X 11 N A 1X 11 1 0 b 12 N C 1X 12 N A 1X 12 1 0 b 1K1 N C 1X 1K1 N A 1X 1K1 1 0 b 21 N C 2X 21 N A 2X 21 0 1

Subproblem 2 b 22 N C 2X 22 N A 2X 22 0 1 b 2K2 N C 2X 2k2 N A 2X 2K2 0 1 x5 0 1 0 0

Starting basic solution x6 -M 0 1 0 x7 -M 0 0 1 = 40 = 1 = 1

C 1 = (3, 5) A 1 = (1, 1) Solution space, D1X 1 b1: 5x1 + x2 12 x 1, x 2 0

C 2 = (1, 1) A 2 = (1, 1) Solution space, D2X 2 b2: x3 + x4 5 x3 + 5x4 50 x3, x4 0

Notice that x5 is the slack variable that converts the common constraint to the following equation: x1 + x2 + x3 + x4 + x5 = 40

22.24

Chapter 22

Additional Network and LP Algorithms

Recall that subproblems 1 and 2 account for variables x1, x2, x3, and x4 only. This is the reason x5 must appear explicitly in the master problem. The remaining starting basic variables, x6 and x7, are artificial. Iteration 0 XB = (x5, x6, x7)T = (40, 1, 1)T CB = (0, - M, - M), B = B-1 = I Iteration 1 Subproblem 1 (j = 1). We have A1X1 1 - C1X1 0 11, 12a 1 0 x1 b x2

z1 - c1 = CBB-1

= 10, - M, - M2

- 13, 52a

x1 b x2

= - 3x1 - 5x2 - M Thus, the corresponding LP is Minimize w1 = - 3x1 - 5x2 - M subject to 5x1 + x2 12 x1, x2 0 The solution of this problem (by the simplex method) yields N = 10, 122T, z - c = w = - 60 - M X 11 1 1 1 Subproblem 2 (j = 2). The associated linear program is given as A2X2 0 - C2X2 1 11, 12a 0 1 x3 b x4

Minimize z2 - c2 = CBB-1

= 10, - M, - M2

- 11, 12a

x4 b x5

= - x4 - x5 - M

22.2 subject to x3 + x4 5 x3 + 5x4 50 x3, x4 0 The optimal solution of the problem yields

Decomposition Algorithm

22.25

N = 150, 02T, z - c = - 50 - M X 21 2 2
* * * * * Because the master problem is of the maximization type and z* 1 - c1 6 z2 - c2 and z1 - c1 6 0, N it follows that b 11 associated with extreme point X11 enters the solution. To determine the leaving variable,

P11

0 11, 12a b N A1X 12 12 11 = 1 = 1 = 1 0 0 0

Thus, B-1P11 = (12, 1, 0)T. Given XB = (x5, x6, x7)T = (40, 1, 1)T , it follows that x6 (an artificial variable) leaves the basic solution (permanently). The new basis is determined by replacing the vector associated with x6 with the vector P11, which gives (verify!) 1 B = 0 0 The new basic solution is XB = (x5, b 11, x7)T = B-1(40, 1, 1)T = (28, 1, 1)T N , - M) = (0, 60, - M) CB = (0, C1 X 11 Iteration 2 Subproblem 1 (j = 1). The associated objective function is Minimize w1 = - 3x1 - 5x2 + 60
* (verify!). The optimum solution yields z* 1 - c1 = w1 = 0, which means that none of the remaining extreme points in subproblem 1 can improve the solution to the master problem.

12 1 0

0 1 0 Q B-1 = 0 1 0

- 12 1 0

0 0 1

Subproblem 2 (j = 2). The associated objective function is (coincidentally) the same as for j = 2 in iteration 1 (verify!). The optimum solution yields N = 150, 02T, z - c = - 50 - M X 22 2 2

22.26

Chapter 22

Additional Network and LP Algorithms

N . The subscript 2 is used for notational N is actually the same extreme point as X Note that X 22 21 convenience to represent iteration 2. N From the results of the two subproblems, z 2 - c2 6 0 indicates that b 22 associated with X22 enters the basic solution. To determine the leaving variable, consider N A2X 22 = 0 = 1 11, 12a 0 1 50 b 0

P22

50 = 0 1

Thus, B-1P22 = (50, 0, 1)T. Because XB = (x5, b 11, x7)T = (28, 1, 1)T, x5 leaves. The new basis and its inverse are given as (verify!) 50 B = 0 1 12 1 0
1 0 50 -1 0 Q B = 0 1 1 - 50

- 12 50 1
12 50

0 0 1

XB = ( b 22, b 11, x7)T = B-1(40, 1, 1)T =

N ,C X N CB = 1C2 X 22 1 11, - M2 = 150, 60, - M2

11 T A 14 25 , 1, 25 B

Iteration 3 Subproblem 1 (j = 1). You should verify that the associated objective function is Minimize w1 = The associated optimum solution is N = 10, 02T, z - c = - 12M + 48 X 13 1 1 50 Subproblem 2 (j = 2). The objective function can be shown to equal (verify!) Minimize w2 = The associated optimum solution is N = 15, 02T, z - c = - 9M X 23 2 2 10 Nonbasic Variable x5 . From the definition of the master problem, zj - cj of x5 must be computed and compared separately. Thus, z5 - c5 = CBB-1P5 - c5 = A1 +
M 50 M 50 M AM 50 - 2 B x1 + A 50 - 4 B x2 12M 50

+ 48

AM 50 B (x3 + x4) - M

, 48 -

12M 50

, - M B (1, 0, 0)T - 0

= 1 +

22.2

Decomposition Algorithm

22.27

Thus, x5 cannot improve the solution. N enters the basic solution. From the preceding information, b 23 associated with X 23 To determine the leaving variable, consider N A2X 23 = 0 = 1 5 11, 12a b 0 5 0 = 0 1 1

P23

1 9 T 11 T Thus, B-1P23 = 1 10 , 0, 10 2 . Given that XB = ( b 22, b 11, x7)T = 1 14 25 , 1, 25 2 , the artificial variable x10 leaves the basic solution (permanently). The new basis and its inverse are thus given as (verify!)

XB = ( b 22, b 11, b 23)T = B-1(40, 1, 1)T = A N ,C X N N CB = (C2 X 22 1 11, C2 X23) = (50, 60, 5) Iteration 4

50 B = 0 1

12 1 0

1 5 45 -1 0 Q B = 0 1 1 - 45

- 12 45 1
12 45 23 45

5 - 45 0 50 45

, 1, 22 45

BT

* * Subproblem 1 (j = 1). w1 = - 2x1 - 4x2 + 48. It yields z* 1 - c1 = w1 = 0.

Subproblem 2 (j = 2). w2 = 0x4 + 0x5 + 48 + 48. Nonbasic Variable x5: z5 - c5 = 1. The preceding information shows that iteration 3 is optimal. We can compute the optimum solution of the original problem by back-substitution:
T T T N X 1 = 1x1, x22 = b 11 X 11 = 110, 122 = 10, 122 T N N X 2 = 1x4, x52 = b 22 X22 + b 23 X23 22 T T A 23 45 B A 50, 02 + A 45 B 15, 0 B

= 128, 02T

All the remaining variables are zero. The optimum value of the objective function can be obtained by direct substitution.

PROBLEM SET 22.2A


*1. In each of the following cases, determine the feasible extreme points graphically, and express the feasible solution space as a function of these extreme points. If the solution space is unbounded, add a proper artificial constraint. (a) x1 + 2x2 6 2x1 + x2 8 - x1 + x2 1 x2 2 x1, x2 0

22.28

Chapter 22 (b)

Additional Network and LP Algorithms

2x1 + x2 2 3x1 + 4x2 12 x1, x2 0 *(c) x1 - x2 10 2x1 40 x1, x2 0 *2. In Example 22.2-1, the feasible extreme points of subspaces D1X1 = b1, X1 0 and D2X2 = b2, X2 0 can be determined graphically. Use this information to express the associated master problem explicitly in terms of the feasible extreme points. Then show that the application of the simplex method to the master problem produces the same entering variable b jk as that generated by solving subproblems 1 and 2. Hence, convince yourself that the determination of the entering variable b jk is exactly equivalent to solving the two minimization subproblems. 3. Consider the following linear program: Maximize z = x1 + 3x2 + 5x3 + 2x4 subject to x1 + 4x2 2x1 + x2 5x1 + 3x2 + 4x3 8 9 10

x3 - 5x4 4 x3 + x4 10 x1, x2, x3, x4 0 Construct the master problem explicitly by using the extreme points of the subspaces, and then solve the resulting problem directly by the simplex method. 4. Solve Problem 3 using the decomposition algorithm, and compare the two procedures. 5. Apply the decomposition algorithm to the following problem: Maximize z = 6x1 + 7x2 + 3x3 + 5x4 + x5 + x6 subject to x1 + x2 + x1 + x2 x2 5x3 + x4 x3 + x4 + x5 + x6 50 10 8 12 x5 + x6 5 x5 + 5x6 50 x1, x2, x3, x4, x5, x6 0

22.3

Karmarkar Interior-Point Method

22.29

*6. Indicate the necessary changes for applying the decomposition algorithm to minimization LPs. Then solve the following problem: Minimize z = 5x1 + 3x2 + 8x3 - 5x4 subject to x1 + x3 + x3 + x4 25 5x1 + x2 5x1 - x2 20 5 x3 + x4 = 20 x1, x2, x3, x4 0 7. Solve the following problem by the decomposition algorithm: Minimize z = 10y1 + 2y2 + 4y3 + 8y4 + y5 subject to y1 + 4y2 - y3 2y1 + y2 + y3 3y1 y1 8 2 + y4 + y5 4 + 2y4 - y5 10 y1, y2, y3, y4, y5 0 (Hint: Solve the dual problem first by decomposition.) 8. In the decomposition algorithm, suppose that the number of common constraints in the original problem is r. Show that the objective function for subproblem j can be written as Minimize wj = zj - cj = (CBRAj - Cj)Xj + CBVr + j The vector R represents the first r columns of B-1, and Vr + j is its (r + j)th column.

22.3

KARMARKAR INTERIOR-POINT METHOD The simplex method obtains the optimum solution by following a path of adjacent extreme points along the edges of the solution space. Although in practice the simplex method has served well in solving large problems, theoretically the number of iterations needed to reach the optimum solution can grow exponentially. In fact, researchers have constructed a class of LPs in which all feasible extreme points are visited before the optimum is reached. In 1984, N. Karmarkar developed a polynomial-time algorithm that cuts across the interior of the solution space. The algorithm is effective for extremely large LPs. We start by introducing the main idea of the Karmarkar method and then provide the computational details of the algorithm.

22.30

Chapter 22

Additional Network and LP Algorithms

22.3.1 Basic Idea of the Interior-Point Algorithm Consider the following (trivial) example: Maximize z = x1 subject to 0 x1 2 Using x2 as a slack variable, the problem can be rewritten as Maximize z = x1 subject to x1 + x2 = 2 x1, x2 0 Figure 22.21 depicts the problem. The solution space is given by the line segment AB. The direction of increase in z is in the positive direction of x1. Let us start with any arbitrary interior (nonextreme) point C in the feasible space (line AB). The gradient of the objective function (maximize z = x1) at C is the direction of fastest increase in z. If we locate an arbitrary point along the gradient and then project it perpendicularly on the feasible space (line AB), we obtain the new point D with a better objective value z. Such improvement is obtained by moving in the direction of the projected gradient CD. If we repeat the procedure at D, we will determine a new closer-to-optimum point E. Conceivably, if we move (cautiously) in the direction of the projected gradient, we will reach the optimum at point B. If we are minimizing z (instead of maximizing), the projected gradient will correctly move us away from point B toward the minimum at point A (x1 = 0). The given steps hardly define an algorithm in the normal sense, but the idea is intriguing! We need some modifications that will guarantee that (1) the steps generated
FIGURE 22.21 Illustration of the general idea of Karmarkars algorithm x2

2 A Gradient of z Maximize z = x1

C D E B

x1

22.3

Karmarkar Interior-Point Method

22.31

along the projected gradient will not overshoot the optimum point at B, and (2) in the general n-dimensional case, the direction created by the projected gradient will not cause an entrapment of the algorithm at a nonoptimum point. This, basically, is what Karmarkars interior-point algorithm accomplishes. 22.3.2 Interior-Point Algorithm Several variants of Karmarkars algorithm are available in the literature. Our presentation follows the original algorithm. Karmarkar assumes that the LP is given as Minimize z = CX subject to AX = 0 1X = 1 X 0 All the constraints are homogeneous equations except for the constraint n 1X = a j = 1xj = 1, which defines an n-dimensional simplex. The validity of Karmarkars algorithm rests on satisfying two conditions:
1 1 1 , n , , n B satisfies AX = 0. 1. X = A n 2. min z = 0.

Karmarkar provides modifications that allow solving the problem when the second condition is not satisfied. These modifications will not be presented here. The following example illustrates how a general LP can be made to satisfy the two stipulated conditions. Example 22.3-1
Consider the problem. Maximize z = y1 + y2 subject to y1 + 2y2 2 y1, y2 0 We start by defining the primal and dual problems of the LP:
Primal Maximize y0 = y1 + y2 subject to y1 + 2y2 2 y1, y2 0 Dual Minimize w0 = 2w1 subject to w1 1 f Q w1 1 2w1 1 w1 0

22.32

Chapter 22

Additional Network and LP Algorithms

The primal and dual constraints can be put in equation form as y1 + 2y2 + y3 = 2, y3 0 w1 - w2 = 1, w2 0 At the optimum y0 = w0, which yields y1 + y2 - 2w1 = 0 Selecting M sufficiently large, we have y1 + y2 + y3 + w1 + w2 M Now, converting (3) into an equation, we get y1 + y2 + y3 + w1 + w2 + s1 = M, s1 0 (4) (3) (2) (1)

Next, define a new variable s2. From (4) the following two equations hold if, and only if, the condition s2 = 1 holds: y1 + y2 + y3 + w1 + w2 + s1 - Ms2 = 0 y1 + y2 + y3 + w1 + w2 + s1 + s2 = M + 1 (5)

Now, given s2 = 1 as stipulated by (5), the primal and dual equations (1) can be written as y1 + 2y2 + y3 - 2s2 = 0 w1 - w2 - 1s2 = 0 Now, define yj = (M + 1)xj, j = 1, 2, 3 wj - 3 = (M + 1)xj, j = 4, 5 s1 = (M + 1)x6 s2 = (M + 1)x7 Substitution in equations (2), (5), and (6) will produce the following equations: x1 + x2 - 2x4 = 0 (6)

x1 + x2 + x3 + x4 + x5 + x6 - Mx7 = 0 x1 + x2 + x3 + x4 + x5 + x6 + x1 + 2x2 + x3 + x4 - x5 x7 = 1

- 2x7 = 1 x7 = 0

xj 0, j = 1, 2, , 7 The final step calls for augmenting the artificial variable y8 in the left-hand side of each equation. The new objective function will call for minimizing y8, whose obvious minimum value must be zero (assuming the primal is feasible). Note, however, that Karmarkars algorithm requires the solution X =
1 1 1 1 1 1 1 T A1 8, 8, 8, 8, 8, 8, 8, 8 B

22.3

Karmarkar Interior-Point Method

22.33

to be feasible for AX = 0. This will be true for the homogeneous equations (with zero righthand side) if the associated coefficient of the artificial x8 equals the (algebraic) sum of all the coefficients on the left-hand side. It thus follows that the transformed LP is given as Minimize z = x8 subject to x1 + x2 + x3 + x4 + x5 + x6 - Mx7 - 16 - M2x8 = 0 x1 + 2x2 + x3 x4 - x5 2x7 x7 + x7 + 2x8 = 0 x8 = 0 x8 = 1 x1 + x2 - 2x4 0x8 = 0

x1 + x2 + x3 + x4 + x5 + x6 +

xj 0, j = 1, 2, , 8 Note that the solution of this problem automatically yields the optimum solutions of the primal and dual problems through back-substitution. Figure 22.22 provides a typical illustration of the solution space in three dimensions with the homogeneous set AX = 0 consisting only of one equation. By definition, the solution space consisting of the line segment AB lies entirely in the two-dimensional simplex 1X = 1 and 1 1 passes through the feasible interior point A 1 3 , 3 , 3B.

Steps of the algorithm. Karmarkars algorithm starts from an interior point represented by the center of the simplex and then advances in the direction of the projected gradient to determine a new solution point. The new point must be strictly interior, meaning that all its coordinates must be positive. The validity of the algorithm rests on this condition. For the new solution point to be strictly interior, it must not lie on the boundaries of the simplex. (In terms of Figure 22.22, points A and B must be excluded.) To guarantee this result, a sphere with its center coinciding with that of the simplex is inscribed tightly inside the simplex. In the n-dimensional case, the radius r of this 1 sphere equals n(n . A smaller sphere with radius ar (0 6 a 6 1) will be a subset - 1) 3 of the sphere, and any point in the intersection of the smaller sphere with the homogeneous system AX = 0 will be an interior point, with strictly positive coordinates. Thus, we can move as far as possible in this restricted space (intersection of AX = 0 and the ar-sphere) along the projected gradient to determine the new (improved) solution point. The new solution point no longer will be at the center of the simplex. For the procedure to be iterative, we need to find a way to bring the new solution point to the center of a simplex. Karmarkar satisfies this requirement by proposing the following intriguing idea, called projective transformation. Let yi =
xi xki n x

, i = 1, 2, , n

j=1

i xkj

22.34

Chapter 22

Additional Network and LP Algorithms


x3

(0, 0, 1)

Simplex 1X 1

B Center of simplex

Intersection of AX 0 and 1X 1

1,1,1 3 3 3 (1, 0, 0) A

x1

(0, 1, 0)

x2 (a) Three dimensions (0, 0, 0, 1)

Simplex 1X 1 Intersection of AX 0 and 1X 1 Center of simplex 1,1,1,1 4 4 4 4 B (0, 0, 1, 0) (1, 0, 0, 0) C

(0, 1, 0, 0) (b) Four dimensions FIGURE 22.22 Illustrations of the simplex 1X = 1

22.3

Karmarkar Interior-Point Method

22.35

where xki is the i th element of the current solution point Xk . The transformation n is valid, because all xki 7 0 by design. You will also notice that a i = 1yi = 1, or 1Y = 1, by definition. This transformation is equivalent to Y =
-1 Dk X -1 1Dk X

where Dk is a diagonal matrix whose ith diagonal elements equal xki. The transformation maps the X-space onto the Y-space uniquely because we can show that the last equation yields X = DkY 1DkY

By definition, min CX = 0 . Because 1DkY is always positive, the original linear program is equivalent to Minimize z = CDkY subject to ADkY = 0 1Y = 1 Y 0 The transformed problem has the same format as the original problem. We can thus 1 1 1 ,n,,n start with the simplex center Y = A n B and repeat the iterative step. After each iteration, we can compute the values of the original X variables from the Y solution. We show now how the new solution point can be determined for the transformed problem. At any iteration k, the problem is given by Minimize z = CDkY subject to ADkY = 0 Y lies in the ar-sphere Because the ar-sphere is a subset of the space of the constraints 1X = 1 and X 0, these two constraints can be dispensed with. As a result, the optimum solution of the preceding problem lies along the projection of the gradient cp (minimization) and is given as cp Ynew = Y0 - ar ||cp|| where Y0 = a 1 1 1 t , , , b , and cp is the projected gradient, which can be shown to be n n n cp = [I - PT1PPT2-1P]1cDk2T where P = a ADk b 1

22.36

Chapter 22

Additional Network and LP Algorithms

The selection of a is crucial to enhancing the efficiency of the algorithm. Normally, we would select a as large as possible to acquire large jumps in the solution. However, if a is too large, we may come too close to the prohibited boundaries of the simplex. There is no general answer to this problem, but Karmarkar suggests the use 1 of a = n 3n . The steps of Karmarkars algorithm are Step 0. Start with the solution point X0 = 1) and a = (n 3n
1 1 1 ,n,,n An B , and compute r = 1 1n(n - 1)

General step k . Define Dk = diag{xk1, , xkn} ADk P = a b 1 and compute Ynew = a Xk + 1 where cp = [I - PT(PPT)-1P](cDk)T cp 1 1 T , , b - ar n n cp DkYnew = 1DkYnew

Example 22.3-2
Minimize z = 2x1 + 2x2 - 3x3 subject to - x1 - 2x2 + 3x3 = 0 x1 + x2 + x3 = 1 x1, x2, x3 0 The problem satisfies the two conditions imposed by the interior-point algorithmnamely, X = (x1, x2, x3)T = satisfies both constraints and the optimum solution
T T X = (x 1, x2, x3) = (0, .6, .4) , 1 1 T A1 3 , 3 , 3B ,

and yields z = 0.

22.3 Iteration 0

Karmarkar Interior-Point Method

22.37

c = (2, 2, - 3), A = ( - 1, - 2, 3) X0 =
1 3 1 1 T 1 A1 3 , 3 , 3 B , z0 = 3 , r =

1 26

,a =

2 9

0
1 3

D0 = 0 0

0 0
1 3

Using projective transformation, we get Y0 = Iteration 1 cD0 = 12, 2, - 32 0 0


1 3 1 1 T A1 3 ,3 ,3B

0
1 3

0 0 =
1 3 1 3

2 A2 3, 3, - 1 B

AD0 = 1 - 1, - 2, 32 0 0 1PPT2-1
T T -1

0
1 3

0 2 0 = A- 1 3, - 3, 1 B
1 3

-1 = a 3 1 0 1 0 25 =
1 42 - 20

-2 3 1

1 I - P 1PP 2 P = 0 0

-1 -1 1 9 3 1 0 2 b - 3 1 = a 14 1 b 1 0 3 1 1 0 -1 1 9 3 0 -1 -2 2 3 0 - - 3 1 a 14 1 b a 3 0 3 1 1 1 1 1

1 b 1

-5 Thus, cp = 1I - P 1PP 2 P21cD02 =


T T -1 T

- 20 16 4

-5 4 1

25
1 42 - 20

-5

- 20 16 4

2 -5 3 4 2 3 = 1 -1

25
1 126 - 20

-5

It then follows that 252 + ( - 20)2 + ( - 5)2

cp =

1262

= .257172

22.38

Chapter 22 Thus,

Additional Network and LP Algorithms

Ynew =

1 1 T A1 3, 3, 3 B -

2 9

1 116

1 .257171

1 126

(25, - 20, - 5)T

= (.263340, .389328, .347332)T Next, 1D0Ynew = Now, D0Ynew = 1D0Ynew


1 3 1 3

(1, 1, 1)(.263340, .389328, .347332)T =

1 3

X1 =

Ynew
1 3

= Ynew = (.263340, .389328, .347332)T

z1 = .26334 Iteration 2 cD1 = 12, 2, - 32 .263340 0 0 .263340 0 0 0 .389328 0 0 .389328 0 0 0 = 1.526680, .778656, - 1.0419962 .347332 0 0 .347332

AD1 = 1 - 1, - 2, 32

= 1 - .263340, - .778656, 1.0419962 1PPT2-1 - .26334 = a 1 = a I - PT1PPT2-1P 1 = 0 0 a 0 1 0 0 - .263340 0 - - .778656 1 1.041996 - .778656 1 - .449746 .322451 .127295 1 .567727 1 a 0 1 0 b .333333 .567727 0 - .778656 1 0 b .333333 - .263340 1.041996 b - .778656 1 1.041996
-1 1 1 1

- .263340 1

1.041996 b 1 - .177550 .127295 .050254

.627296 = - .449746 - .177550

22.3 Thus,

Karmarkar Interior-Point Method

22.39

.627296 cp = 1I - PT1PPT2-1P21cD12T = - .449746 - .177550 .526680 .778656 - 1.041996

- .449746 .322451 .127295

- .177550 .127295 .050254

.165193 = - .118435 - .046757 It then follows that ||cp|| = 3.1651932 + ( - .118435)2 + ( - .046757)2 = .208571 Thus,
1 1 T Ynew = ( 1 3,3,3) 2 9

1 116

1 .208571

(.165193, - .118435, - .046757)T

= (.261479, .384849, .353671)T Next, .263340 0 0 0 .389328 0 0 .261479 .068858 0 .384849 = .149832 .347332 .353671 .122841

D1Ynew =

1D1Ynew = .341531 Now, X2 = D1Ynew = 1.201616 1D1Ynew .438707 .3596772T

z2 = .201615 Repeated application of the algorithm will move the solution closer to the optimum point (0, .6, .4). Karmarkar does provide an additional step for rounding the optimal solution to the optimum extreme point.

PROBLEM SET 22.3A


1. Use TORA to show that the solution of the transformed LP given at the end of Example 22.3-1 does yield the optimal primal and dual solutions of the parent problem. (Hint: Use M = 10, and make sure that TORAs output gives at least 5 decimal points accuracy.)

22.40

Chapter 22

Additional Network and LP Algorithms

2. Transform the following LP to Karmarkars format. Maximize z = y1 + 2y2 subject to y1 - y2 2 2y1 + y2 4 y1, y2 0 3. Carry out one additional iteration in Example 22.3-3, and show that the solution is moving toward the optimum z = 0. 4. Carry out two iterations of Karmarkars algorithm for the following linear program: Maximize z = 2x1 + x2 subject to x1 + x2 4 x1, x2 0 (Hint: The problem must be converted to Karmarkar format first.) 5. Carry out two iterations of Karmarkars algorithm for the following problem: Maximize z = 4x1 + x3 + x4 subject to - 2x1 + 2x2 + x3 - x4 = 0 x1 + x2 + x3 + x4 = 1 x1, x2, x3, x4 0 (Hint: The problem must be converted to Karmarkar format first.)

BIBLIOGRAPHY
Ahuja, R., T. Magnati, and J. Orlin, Network Flows: Theory, Algorithms, and Applications, Prentice Hall, Upper Saddle River, NJ, 1993. Bazaraa, M., J. Jarvis, and H. Sherali, Linear Programming and Network Flow, 4th ed., Wiley, New York, 2009. Charnes, A. and W. Cooper, Some Network Characterization for Mathematical Programming and Accounting Applications to Planning and Control, The Accounting Review, Vol. 42, No. 3, pp. 2452, 1967. Hooker, J., Karmarkars Linear Programming Algorithm, Interfaces, Vol. 16, No. 4, pp. 7590, 1986. Lasdon, L., Optimization for Large Systems, Macmillan, New York, 1970.

You might also like