This action might not be possible to undo. Are you sure you want to continue?

# Modeling of Purchase and Sales Contracts in Supply

Chain Optimization

Minhwan Park and Sunwon Park

Department of Chemical and Biomolecular Engineering, Korea Advanced Institute of Science and

Technology, 373-1 Guseong-dong, Yuseong-gu, Daejeon 305-701, Republic of Korea

Fernando D. Melle and Ignacio E. Grossmann*

Department of Chemical Engineering, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh,

Pennsylvania 15213-3890, USA

*CORRESPONDING AUTHOR: Ignacio Grossmann grossmann@cmu.edu, (412)268-3642

ABSTRACT

**This work presents a novel approach for modeling of different types of contracts that a company may
**

sign with its suppliers and customers. The main objective is to expand the scope of current planning and

supply chain optimization models by including the selection of the types of contracts as an additional

decision. The solution approach relies on representing the decision of choosing different contracts using

disjunctive programming for both short-term and long-term production planning models. The resulting

formulation is converted into a mixed-integer linear programming (MILP) problem. The advantages of

the proposed models are highlighted in two case studies of increasing complexity.

**Keywords: Supply Chain Management, Supply Chain Optimization, MILP, Purchase and Sales,
**

Contract

1

1 Introduction

If we examine the ever increasing amount of literature on problems belonging to the operation planning

of production plants and Supply Chain Management (SCM), it can be found that there are still many

issues related to the management of the relationships between a company’s Supply Chain (SC) and its

customers and suppliers. Particularly, the signing of contracts is a common practice in the business

world, through which the company aims mainly for two objectives: to reduce uncertainty by planning

capacity and ensuring a certain sales level, and to take advantage of the discounts for purchasing

materials in large amounts (economies of scale). For example, on reviewing the advances and

challenges in SCM, Shah (2005)1 identifies the negotiation of long-term contracts as an important SCM

problem at a strategic level.

A contract can be understood as a binding agreement in which the seller provides the specified product

and the buyer pays for it under specific terms and conditions. Current approaches for production

planning at different levels often neglect this important aspect focusing on other issues2. Therefore, in

this work, we propose the modeling of several types of contracts with the external entities of the

company, both suppliers and customers, which gives the opportunity of improving these approaches.

The paper is organized as follows. First, a literature review on recent work about contract modeling is

presented. Next, the problem is described as well as the proposed models for process network with

extensions to allow for contracts. These models are solved bye means of disjunctive programming

(Grossmann and Lee (2003)3). The performance of the models is then illustrated through two examples

and finally, the results obtained are discussed and some guidelines for future work are indicated.

2 Literature review

Literature on contracts, in the area of SCM, may be divided in three areas: negotiation process,

uncertainty management and contract modeling.

2

**The negotiation process is an area mainly belonging to the field of Computers Science in which the
**

emphasis is placed on the contract negotiation process itself. Many of these works are related to the ecommerce and software agents on the web. Within this area, a number of different negotiable attributes,

e.g. prices, quantities, due-dates, etc., are calculated by means of auction-like processes between the

interested parties. Such systems are simulators that mimic the negotiation process until reaching an

agreement in the contractual conditions (Sandholm (2002)4). Goodwin et al. (1999)5 present a

framework for providing decision support for an on-line exchange. They use a multi-agent system to

find matches of demand and supply on the exchange and provide the user with the best set of

transactions. The user then chooses the best match based on his/her discretion.

Most works devoted to uncertainty management apply an adapter strategy in which the company

controls the risk exposure of its assets by constantly adapting its operations to unfolding demand

realizations. In contrast, in the strategy known as shaper the SC aims to restructure the demand

distribution contracting agreements with the customer (Anupindi and Bassok (1999)6). Some of these

works are aimed at demonstrating how contracts are an effective tool against demand and supply

uncertainty. They are based on stochastic programming and use risk measures to compare the

approaches with and without signing up contracts. Nevertheless, these works do not put the emphasis on

modeling the type of contracts, in fact, contract models are rather simple because they consist of

including fixed quantities at given times in the horizon of the analysis.

A type of contract that is extensively used in these works is the option contract (Hull (1995)7). Options,

also known as derivative securities or contingent claims, are legally binding and negotiable contracts

that give the holder the right, but not the obligation, to purchase a certain quantity of an agricultural,

industrial, or financial product at a specified price and time for a one-time, upfront premium payment.

Applications in this direction can be found in the works by Gupta and Maranas (2003)8, Barbaro and

Bagajewicz (2004)9, Bonfill et al. (2004)10 and Aseeri and Bagajewicz (2004)11.

A different aspect within the area of contracts is the modeling of different types of contracts that a

company can set up during its planning process. The consequence of this is the modification of the

3

**objective function of the corresponding optimization program resulting in a more complex but also more
**

realistic cost function. There has been little work done in this area and the one that has been reported by

Kallrath (2002)12, Schulz et al. (2005)13 and Chandra et al. (2004)14 assumes that types of contracts are

given and not decision variables.

3 Problem statement

In this paper, we consider short and long-term multiperiod production planning of a chemical supply

chain network. The network involves NP processes, NC chemicals for which NI inventories are kept

over NT time periods. The operation of the network is constrained by existing capacities of all processes

in the network, limits in the supplies of raw materials and market saturation of some products.

Information is given for different types of contracts that can be made for purchasing raw materials and

selling the products. The objective in the short-term planning problem is to determine over a given time

horizon, typically weeks or months, the types of contracts for the purchase of raw materials and sales of

products in order to maximize the profit, which can be calculated by the data on sales revenues,

operating costs, marketing costs, inventory cost and shortfall penalties. For the case of the long-term

planning problem, we consider the possible capacity expansion of the processes. In this case, we

account for the capital costs in the NPV but exclude the effect of inventories since the length of the time

periods are assumed to be large (e.g. one year each). Our goal is to decide (a) which contract to make in

the purchase of the raw materials and selling the products in each time period, and (b) whether capacity

of each process should be expanded or not in each time period.

We consider various models for the supply/demand that involve different contract types which include

various types of discounts depending on the volumes and lengths of contracts. We will assume that all

the information is available in order to formulate the problem as a multiperiod MILP model.

3.1 Contract models for supply

4

5 . The types of contracts that will be considered when purchasing raw materials from suppliers include the following: (1) fixed price. t ∈ T (1) where ϕ jtf is the contract price of raw material j at time period t and Pjtf is the amount of raw materials j purchased at time period t . The difference between them is that the price in discount after σ djt is reduced to amounts that exceed σ djt . The cost of purchasing raw materials in the fixed price contract is given by: COST jtf = ϕ jtf Pjtf j ∈ JR. and (4) fixed duration. The treatment of demands is entirely analogous. we assume that contracts are made only for supplies. To model the various types of contracts. the discount after a certain amount and the bulk discount require the purchase at a minimum quantity of chemical j at time t ( σ djt and σ bjt . Figure 1. Fixed price contract: A fixed price contract simply means that we buy raw materials in any amount at the current market price. In both cases we assume that we renew the contracts at each time period. we use disjunctions in the following models.First. (3) bulk discount. In contrast bulk contract assumes that if the amount purchased exceeds σ bjt . then the price is reduced for the entire amount that is purchased. respectively). Discount contracts: On the other hand. (2) discount after certain amount. Discount after σ djt amount.

at the lower price. we buy σ djt of raw materials at the price of ϕ djt1 and an excess amount. Bulk discount: Figure 2. Otherwise ( y djt1 = true ). t ∈ T 6 (5) . t ∈ T (4) where ϕ djt1 > ϕ djt 2 . if the amount of raw material required is Pjtd 2 which exceeds σ djt ( y djt2 = true ).Discount after a certain amount: The cost of purchasing raw materials in the discount after a certain amount (see Figure 1) is given by: COST jtd = ϕ djt1 Pjtd 1 + ϕ djt 2 Pjtd 2 ⎡ y djt1 ⎤ ⎡ y djt2 ⎤ ⎢ ⎥ ⎢ ⎥ d1 d d1 d ⎢0 ≤ Pjt ≤ σ jt ⎥ ∨ ⎢ Pjt = σ jt ⎥ ⎢ d2 ⎥ ⎢ d2 ⎥ ⎢⎣ Pjt = 0 ⎥⎦ ⎢⎣ Pjt ≥ 0 ⎥⎦ Pjtd = Pjtd 1 + Pjtd 2 j ∈ JR. t ∈ T (2) j ∈ JR. ϕ djt 2 . we buy raw materials at the price of ϕ djt1 which is higher than ϕ djt 2 . In this contract. Bulk discount. Pjtd 2 − σ djt . The cost of purchasing raw materials in the bulk discount (see Figure 2) is given by: ⎤ ⎡ y bjt1 ⎤ ⎡ y bjt2 ⎢ ⎥ ⎢ ⎥ b b1 b b b2 b ⎢COST jt = ϕ jt Pjt ⎥ ∨ ⎢COST jt = ϕ jt Pjt ⎥ ⎥ ⎢ ⎥ ⎢ b b b b ⎥⎦ ⎢⎣0 ≤ Pjt ≤ σ jt ⎥⎦ ⎢⎣ Pjt ≥ σ jt j ∈ JR. t ∈ T (3) j ∈ JR.

Otherwise. Fixed duration contracts. The cost of purchasing raw materials in the fixed length contract. the longer the contracts last. in this contract we purchase the raw material at the higher price ϕ bjt1 . assuming up to 3 time periods contract for simplicity. but the lower the purchase prices ( ϕ ljt1 > ϕ ljt2 > ϕ ljt3 ) are. the larger the minimum purchase quantities per month ( 0 < σ ljt2 < σ ljt3 ) are. In fixed duration contracts. The 3-month contract specifies the minimum quantity ( σ ljt3 ) and the purchase price ϕ ljt3 which is lower than ϕ ljt2 during the 3 months. The 2-month contract specifies a minimum purchase quantity ( σ ljt2 ) and the purchase price ( ϕ ljt2 ) which is lower than ϕ ljt1 for the 2 months. is given by: 7 . For example. Fixed duration contract: The fixed duration contract specifies the length of the time that contracts are valid and the minimum quantity that must be purchased. we buy the entire amount of raw materials at the lower price ϕ bjt2 if the amount of raw materials we buy is greater than σ bjt . the 1-month contract specifies no minimum amount at the current price ( ϕ ljt1 ) and lasts for 1 month. Figure 3 shows how each contract lasts during a given number of periods.In the bulk discount contract. Figure 3.

t ∈ T (6) 3. t ∈ T (9) j ∈ JP.2 Contract models for demand Concerning the contracts with the customers.t +1 ⎦ ⎢ j .⎡ y ljt3 ⎤ ⎢ ⎥ ⎡ y ljt2 ⎤ ⎢COST jtl = ϕ ljt3 Pjtl ⎥ ⎢ ⎥ ⎢ ⎥ l1 l l2 l l l3 l ⎡ y jt ⎤ ⎢COST jt = ϕ jt Pjt ⎥ ⎢COST j .t + 2 ≥ σ ljt3 ⎥ ⎣ ⎦ j ∈ JR. A fixed price contract means that we sell products in some amount with the current market price. the revenues of selling products in the bulk discount manner are given by: 8 . The income for selling products in the discount after certain amount is given by: REV jtd = ψ djt1 S djt1 + ψ djt 2 S djt 2 j ∈ JP.t +1 ⎥ ∨ ⎢COST j .t + 2 = ϕ jt Pj .t +1 ⎥ ⎢ Pjl.t +1 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ l l1 l l l2 l l l3 l ⎢COST jt = ϕ jt Pjt ⎥ ∨ ⎢COST j . t ∈ T (7) where ψ jtf is the contract price of product j at time period t and S jtf is the amount of j sold at time period t . fixed price contract in this case.t + 2 ⎥ ⎢ l ⎥ ⎢ l ⎥ ⎢ l l1 l2 l3 ⎥ ⎥ ⎢ Pjt ≥ σ jt ⎣⎢ Pjt ≥ σ jt ⎦⎥ ⎢ Pjt ≥ σ jt ⎥ ⎢Pl ≥ σ l 2 ⎥ ⎢Pl ≥ σ l3 ⎥ jt jt ⎣ j . The revenues of selling products in the fixed price contract are given by: REV jtf = ψ jtf S jtf j ∈ JP. The superscript stands for the type of contract. t ∈ T (10) where ψ djt1 > ψ djt 2 are the two different prices of product j and ρ djt is the minimum amount that is necessary to sell to give the discount.t +1 = ϕ jt Pj . Also similarly to the contract with the suppliers.t +1 = ϕ jt Pj . t ∈ T ⎡ z djt1 ⎤ ⎡ z djt2 ⎤ ⎢ ⎥ ⎢ ⎥ d1 d d1 d ⎢0 ≤ S jt ≤ ρ jt ⎥ ∨ ⎢ S jt = ρ jt ⎥ ⎢ d2 ⎥ ⎢ d2 ⎥ ⎢⎣ S jt = 0 ⎥⎦ ⎢⎣ S jt ≥ 0 ⎥⎦ S djt = S djt1 + S djt 2 (8) j ∈ JP. we also consider the four cases mentioned above.

t +1 ⎥ ⎢ S lj .t + 2 ⎥ ⎢ l ⎥ ⎢ l ⎥ ⎢ l l1 l2 l3 ⎥ ⎥ ⎢ S jt ≥ ρ jt ⎣⎢ S jt ≥ ρ jt ⎦⎥ ⎢ S jt ≥ ρ jt ⎥ ⎢S l ≥ ρ l 2 ⎥ ⎢S l ≥ ρ l 3 ⎥ jt jt ⎣ j .t + 2 ≥ ρ ljt3 ⎥ ⎣ ⎦ 9 j ∈ JP.t + 2 = ψ jt S j .⎡ z bjt1 ⎤ ⎡ z bjt2 ⎤ ⎢ ⎥ ⎢ ⎥ b b1 b b b2 b ⎢ REV jt = ψ jt S jt ⎥ ∨ ⎢ REV jt = ψ jt S jt ⎥ ⎢ ⎥ ⎢ b ⎥ b b b ⎢⎣0 ≤ S jt ≤ ρ jt ⎥⎦ ⎢⎣ S jt ≥ ρ jt ⎥⎦ j ∈ JP. t ∈ T (12) . The earnings for selling product in this type of contract.t +1 = ψ jt S j .t +1 ⎥ ∨ ⎢ REV j .t +1 ⎦ ⎢ j . assuming up to 3 periods contract. are given by: ⎡ z ljt3 ⎤ ⎢ ⎥ ⎡ z ljt2 ⎤ ⎢ REV jtl = ψ ljt3 S ljt ⎥ ⎢ ⎥ ⎢ ⎥ l l3 l ⎡ z ljt1 ⎤ ⎢ REV jtl = ψ ljt2 S ljt ⎥ ⎢ REV j .t +1 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ l l1 l l l2 l l l3 l ⎢ REV jt = ψ jt S jt ⎥ ∨ ⎢ REV j . t ∈ T (11) with ψ bjt1 > ψ bjt2 .t +1 = ψ jt S j .

4 LP and MILP models for process network In this section. production of products of each process with fixed capacity. inventory cost. operating cost. and shortfall cost. In the long-term planning problem we consider the optimal selection and expansion of processes given time varying forecasts for the demands and prices of chemicals over a long time horizon. The indices. The short-term planning model is formulated as a multiperiod LP problem. parameters. j ∈ J i . 4. In the short-term planning problem we consider the schedule of purchase of raw materials from suppliers. inventories of each product.g. In the short-term planning model.1 Short-term planning model The objective function to be maximized is the operating profit of the network over a short-term horizon (e. We can additionally define COST jt = ϕ jt Pjt and REV jt = ψ jt S jt which respectively represents purchases and sales and will be defined later to account for different contracts. and variables defined in the model are given in the Nomenclature section. the operating profit is given by P1: PROFIT = ∑∑ψ jt S jt − ∑∑ ϕ jt Pjt j∈J t∈T −∑ j∈J t∈T ∑ ∑δ W i∈I j∈JM i t∈T it ijt − ∑∑ ξ jtV jt − ∑∑θ jt SF jt j∈J t∈T (13) j∈J t∈T where each term accounts for income from sales. The operating costs for each process are assumed to be proportional to the flow of the main product. t ∈ T (14) where J i is the set of chemicals involved by process i . purchase cost. for the short-term planning problem and for the long-term planning problem. and sales of products. several months) consisting of a set of time periods during which prices and demands of chemicals and costs of operating and inventory can vary. sets. 10 . respectively. The amount of chemical j being consumed or produced in process i during period t is represented by the variables: Wijt ≥ 0 i ∈ I . we consider two types of optimization models for process network.

The following equation relates the input to the output of processes: Wijt = µ ijWij 't i ∈ I . t ∈T (20) Finally. j ∈ JM i . and d Ljt . t ∈T (17) where a Ljt . Equation (18) corresponds to the mass balance of chemical j in the network which includes the inventory levels V jt : V j . t ∈T (19) 0 ≤ SF jt ≤ SF jtU j ∈ J . and µ ij are positive constants characteristic of each process i . equations (21) and (22) represent the upper or lower bounds for each variable. j ∈ J i . j '∈ JM i . The amount produced by process i cannot exceed the installed capacity: Wijt ≤ Qit i ∈ I . t ∈ T (16) As for the raw materials. intermediates and products. they are expressed by NC nodes of chemicals where purchases and sales are considered on single market. t ∈T (18) i∈I j where O j is defined as the set of processes that produce chemical j and I j as the set of processes that consume chemical j . aUjt are lower and upper bounds on the availabilities. Production shortfalls with respect to the demands (equation (19)) compensate loss of potential sales which is penalized in the objective function. d Ujt are lower and upper bounds on the demands. SF jt ≥ d Ujt − S jt j ∈ J .t −1 + ∑Wijt + Pjt = V jt + ∑Wijt + S jt i∈O j j ∈ J .All chemical flows in process i other than the main product are given by the mass balance coefficients. t ∈ T (15) where JM i is the set of main products of process i . They must satisfy the inequalities: a Ljt ≤ Pjt ≤ a Ujt ⎫⎪ ⎬ d Ljt ≤ S jt ≤ d Ujt ⎪⎭ j ∈ J . equation (13). 11 .

T . we consider the network which includes an existing system as well as potential new processes and chemicals. Pjt . linear models are assumed for the mass balances in the processes. limits on the investment cost at each time period can be specified. K . variable Qit represents the total capacity of process i that is available in period t . 1 year). the constraints that apply are: QEitL wit ≤ QEit ≤ QEitU wit ⎫ ⎬ i ∈ I . t ∈T wit ∈{0. A zero-value of the binary variables wit makes the capacity expansion at period t zero. a finite number of time periods is considered during which prices and demands of chemicals. Variable QEit represents the capacity expansion of process i which is installed in period t . while fixed-charge cost models are used for the investment cost. In the formulation of this problem. as well as constraints on the sales and purchases. t ∈T In equation (23). and investment and operating costs of the processes can vary. Parameter Qi 0 represents the existing capacity of a process at time t = 0 . If wit are the 0-1 binary variables which indicate the occurrence of the expansions for each process i at each time period t . t = 1. QEitL and QEitU are lower and upper bounds for the capacity expansions.V jt ≤ V jtU j ∈ J . No inventories will be considered since the length of each time period is assumed to be rather long (e. If the binary 12 . Wit . Also.2 Long-term planning model In the long-term planning model.t −1 + QEit (24) i ∈ I . This problem therefore becomes a multiperiod MILP model.g. QEit = 0 . Similarly as in the short term planning model. Also. The objective function to be maximized is the net present value of the project over the specified horizon in order to determine the capacity expansion for existing processes and sales and purchases of chemicals at each time period.1} ⎭ (23) Qit = Qi . V jt ≥ 0 (22) 4. t ∈T (21) S jt .

variable is equal to one. In order to determine the optimal planning of the network. subject to constraints (14)-(17) and (23)-(25). respectively. j ∈ JM i . t ∈ T i ∈ I . t ∈ T a Ljt ≤ Pjt ≤ a Ujt ⎫⎪ ⎬ d Ljt ≤ S jt ≤ d Ujt ⎪⎭ (16) j ∈ J . β it express the variable and fixed terms for the investment cost. j '∈ JM i . and parameters α it . Equation (24) simply defines the total capacity Qit that is available at each time period t . We consider additional constraints that include limits on the number of expansions of each process in equation (27) and limit on the capital for investment during each time in equation (28). ϕ jt are the prices of sales and purchases of the chemical j . j ∈ J i . the capacity expansion is implemented. All these parameters are discounted at the specific interest rate and include the effects of taxes and depreciations in the net present value. t ∈ T (14) Wijt = µ ijWij 't (15) Wijt ≤ Qit i ∈ I . the objective function is given by P2: NPV = ∑∑ψ jt S jt − ∑∑ ϕ jt Pjt j∈J t∈T −∑ j∈J t∈T ∑ ∑ δ W − ∑∑ (α i∈I j∈JM i t∈T it ijt i∈I t∈T it (26) QEit + β it wit ) where ψ jt . ∑W ijt i∈O j + Pjt = ∑Wijt + S jt j ∈ J . Equations (14)-(17) are also considered as constraints. δ it is the unit operating cost. ∑w it ≤ NEXP(i ) i ∈ I (27) t 13 . the multiperiod MILP model consists of maximizing the NPV in equation (26). Wijt ≥ 0 i ∈ I . t ∈T (17) Equation (18) is modified into equation (25) since there in no inventory in this problem. t ∈T (25) i∈I j Finally. j ∈ J i .

Purchase of raw materials with several contracts. 4. Below we provide constraints to determine costs for each type of contract (see Figure 4).∑ (α i∈I it QEit + β it wit ) ≤ CI (t ) t ∈ T (28) where α it . The objective function in P1 changes as follows. For this purpose we assume that we can purchase the raw materials from suppliers with any of the few types of contracts at each time t (see Figure 4). Figure 4. t ∈ T (32) 14 . PROFIT = ∑∑ψ jt S jt − ∑∑∑ COST jtc j∈J t∈T −∑ c∈C j∈J t∈T ∑ ∑δ W it i∈I j∈JM i t∈T ijt − ∑∑ ξ jtV jt − ∑∑θ jt SF jt j∈J t∈T (29) j∈J t∈T whereas the corresponding one for P2 is: NPV = ∑∑ψ jt S jt − ∑∑ COST jtc j∈J t∈T −∑ j∈J t∈T ∑ ∑ δ W − ∑∑ (α i∈I j∈JM i t∈T it ijt i∈I t∈T it (30) QEit + β it wit ) where COST jtc represents the cost of purchasing raw materials j at time horizon t under contract c . β it are nondiscounted cost coefficients with regard to period t . t ∈ T (31) c∈C Pjt = Pjtf + Pjtd + Pjtb + Pjtl j ∈ JR.3 Extension for contract models We modify problem P1 in order to consider contract models mentioned in a previous section. COST jt = ∑ COST jtc j ∈ JR.

c ∈ C = { f . the purchasing cost is simply equation (1). Firstly.1} . we consider the disjunction in equation (3) under the discount after certain amount. respectively. l . d . We assume that for each chemical j . and Pjt . We use the 0-1 binary variables that decide the contract to make. t ∈ T (37) 0 ≤ Pjtd 11 ≤ y djt1σ djt j ∈ JR. the convex hull formulation by Balas (1985)15 is used to convert these disjunctions into an MILP. the bulk discount and the length contract (equations (2)-(6)). is also defined as the summation of Pjtc in equations (30) and (31). creating a variable for each disjunction in equation (36). t ∈ T (1) For the disjunctions under contracts of the discount after certain amount. COST jtd = ϕ djt1 Pjtd 1 + ϕ djt 2 Pjtd 2 Pjtd = Pjtd 1 + Pjtd 2 j ∈ JR. The variables Pjtc have the upper and lower bounds in equation (32). the cost of purchasing raw material j at time horizon t . COST jtf = ϕ jtf Pjtf j ∈ JR. the number of contracts that can be made at time t must be equal or less than one at each time period t . For the fixed price. t ∈ T (33) j ∈ JR. the amount of raw material j purchased at time horizon t . t ∈ T (38) Pjtd 12 = y djt2σ djt j ∈ JR.0 ≤ Pjtc ≤ Uy cjt ∑y c∈C c jt ≤1 c ∈ C . the continuous variables Pjtd 1 are disaggregated. To obtain the convex hull of equation (3). d . b. l} COST jt . c = f . t ∈ T (35) j ∈ JR. b. j ∈ JR. is defined as the summation of COST jtc . t ∈ T (36) Pjtd 1 = Pjtd 11 + Pjtd 12 j ∈ JR. t ∈ T (34) y cjt ∈{0. t ∈ T (39) 15 .

y bjt2 ∈{0. t ∈ T σ bjt y bjt2 ≤ Pjtb 2 ≤ U bjt y bjt2 y bjt1 + y bjt2 = y bjt (43) (44) j ∈ JR. we convert the disjunctions in equation (6) into the MILP constraints as follows. t ∈ T (47) j ∈ JR. we consider the disjunction in equation (5) under the contract of bulk discount. t ∈ T (42) j ∈ JR. t ∈ T (45) j ∈ JR.1} Variable bounds and modification equations are now rewritten in terms of the disaggregated and binary variables (equations (37). equation (40) enforces the requirement that only one binary variable be activated depending on the amount σ djt that is purchased. Secondly. t ∈ T y djt1 + y djt2 = y djt (40) j ∈ JR.1} In the fixed length contract. (38) and (39)). Applying the convex hull to this equation yields the following constraints: COST jtb = ϕ bjt1 Pjtb1 + ϕ bjt2 Pjtb 2 Pjtb = Pjtb1 + Pjtb 2 j ∈ JR. τ ∈ Tt p ⊂ T . COST jtl = Pjtl = ∑ ∑ϕ τ P τ p∈LC τ ∈Tt ∑ ∑Pτ p∈LC τ ∈Tt p lp j t lp j p∈LC lp j l j j ∈ JR. t ∈ T (46) y bjt1 . In case the discount contract is selected. y djt2 ∈{0. t ∈ T σ lpjτ y lpjτ ≤ Pjlpτt ≤ U lj y lpjτ ∑y τ ≤ y τ lp j t p (48) j ∈ JR. τ ∈ T (49) (50) 16 . t ∈ Tτ p ⊂ T . y djt is one. t ∈ T (41) y djt1 . t ∈ T 0 ≤ Pjtb1 ≤ σ bjt y bjt1 j ∈ JR.0 ≤ Pjtd 2 ≤ y djt2U djt j ∈ JR. p ∈ LC j ∈ JR.

τ is the time period at which the contract is made and t is the time period in which the raw material is purchased.1} .2.3} In the above equations LC is a set of contract lengths.y lpjτ ∈{0. while Tτ p is a set of time periods t in which the raw material is purchased with length contract p made at time period τ . Here Tt p is a set of time periods τ at which the contract is made in order to purchase the raw material in time period t for length contract p . customers can buy products from our company through different types of contracts at each time t (see Figure 5). The new objective function for P1 is as follows: PROFIT = ∑∑∑ REV c∈CP j∈J t∈T −∑ − c jt c∈CR j∈J t∈T ∑ ∑δ W it i∈I j∈JM i t∈T ∑∑∑ COST ijt c jt − ∑∑ ξ jtV jt − ∑∑θ jt SF jt j∈J t∈T (51) j∈J t∈T and for P2 is: NPV = ∑∑ REV jtc − ∑∑ COST jtc j∈J t∈T −∑ j∈J t∈T ∑ ∑ δ W − ∑∑ (α i∈I j∈JM i t∈T it ijt i∈I t∈T it (52) QEit + β it wit ) 17 . Pjlpτt represents the amount of raw materials j purchased in time period t under length contract p made at time τ ( τ ≤ t ). The 0-1 variables y lpjτ represent the decision of length contract p with raw material j at time τ . Sale of products by means of several types of contracts. LC = {1. Figure 5. In the same way as we can purchase raw materials from suppliers with any of the types of contract at each time t .

c ∈ C = { f . t ∈ T (60) S djt12 = z djt2 ρ djt j ∈ JP. REV jt = ∑ REV c∈CP j ∈ JP. t ∈ T c jt S jt = S jtf + S djt + S bjt + S ljt 0 ≤ S cjt ≤ Uz cjt ∑z c∈CP c jt ≤1 (53) j ∈ JP. z djt2 ∈ {0. t ∈ T (55) j ∈ JP. t ∈ T (57) j ∈ JP. the convex hull is also used to obtain the corresponding MILP. t ∈ T (59) 0 ≤ S djt11 ≤ z djt1 ρ djt j ∈ JP. t ∈ T (54) c ∈ CP.where REV jtc represents the revenues for selling product j at time t under contract type c . the bulk discount and the length contract. b. t ∈ T (56) z cjt ∈{0. Below we provide constraints to determine revenues for each type of contract. the continuous variables S djt1 are disaggregated to give: REV jtd = ψ djt1 S djt1 + ψ djt 2 S djt 2 S djt = S djt1 + S djt 2 j ∈ JP.1} . t ∈ T (58) S djt1 = S djt11 + S djt12 j ∈ JP. the revenues are simply given by: REV jtf = ψ jtf S jtf j ∈ JP. t ∈ T (63) z djt . t ∈ T (7) For the disjunctions under contracts of the discount after certain amount. l} For the fixed price. z djt1 . To obtain the convex hull of equation (9). t ∈ T 0 ≤ S djt 2 ≤ z djt2U djt z djt1 + z djt2 = z djt (61) j ∈ JP. j ∈ JP. t ∈ T (62) j ∈ JP.1} 18 . d .

(1).1} In the fixed length contract. we convert the disjunctions in equation (12) into the MILP constraints as follows. t ∈ T ρ lpjτ z lpjτ ≤ S lpjτt ≤ U lj z lpjτ ∑z τ ≤ z τ j ∈ JP.1 Description 19 . t ∈ T ρ bjt z bjt2 ≤ S bjt2 ≤ U bjt z bjt2 z bjt1 + z bjt2 = z bjt (66) j ∈ JP. 3} Then. the extension of model P1 for contracts with customers and suppliers are given by equations (51). For the long-term case. The convex hull is obtained with: REV jtb = ψ bjt1 S bjt1 +ψ bjt2 S bjt2 S bjt = S bjt1 + S bjt2 j ∈ JP. 2. t ∈ T (64) j ∈ JP. (7). we consider the disjunction in equation (11) under the contract of bulk discount. t ∈ T 0 ≤ S bjt1 ≤ ρ bjt z bjt1 (65) j ∈ JP. τ ∈ T (71) (72) z lpjτ ∈ {0. z bjt2 ∈{0. (7). τ ∈ Tt p ⊂ T . the extension of model P2 is given by equations (52). t ∈ T (68) z bjt1 . REV jtl = S ljt = ∑ ∑ψ τ S τ p∈LC τ ∈Tt lp j ∑ ∑S τ p∈LC τ ∈Tt p lp j t p∈LC lp j l j (69) j ∈ JP.1} . t ∈ T (67) j ∈ JP. (14)-(17). 5. (31)-(50) and (53)-(72). (31)-(50) and (53)-(72). t ∈ T lp j t p (70) j ∈ JP. (14)-(22). p ∈ LC j ∈ JP. 5 Example 1. t ∈ Tτ p ⊂ T .Secondly. (1). LC = {1.

All the cases (1. Existing capacities of each process are 27. 10 and 14. Process network for example. We assume that for process 3. 30. 2 and 3) are extensions of model P1. 83% of converted C makes E and the remaining 17% makes B. The system is a three-process network that manufactures products D and E from raw materials A. the inventory of each product and purchase of raw materials with no contract. The rest of data for this example are shown in Tables 2-6. the inventories and penalties are considered only for productions. B and C. (3) short-term planning with contracts with both the suppliers and the customers. and 25 tons. 20 .In this section. In this example. the example problem in Figure 6 is solved to illustrate the performance of the models in three cases: (1) short-term planning of the production of each process. For the purchase of the raw materials and the sale of the products. the market prices of raw materials varying with time period are considered as the price of the fixed price contract. (2) short-term planning with contracts only with the suppliers. respectively. we consider the four contracts described previously in the paper. 7-9 and 11-13. The upper bound for product demand varies with time period. We use a 6-month planning horizon. In Table 1. The prices of raw materials and products in each contract are given in Tables 1. and that reactants A and B are fed to process 2 in a 10:1 ratio16. Figure 6.

30 2.40 24. Operating cost coefficient ( δ it ) [100$/ton] at time period t Process P1 P2 P3 1 0.70 27.10 5.ψ jt ) [100$/ton] at time period t Chemical A B C D E 1 2.6 Time period 3 4 0.40 2. Prices of raw materials and products ( ϕ jt .50 5.7 0.70 23.5 2 2.6 0.9 Table 2.Table 1.8 6 2.90 5.40 2. Upper bound for raw material availability ( a Ljt .9 22. a Ujt ) [kton] at time period t Chemical A B C UB LB UB LB UB LB 1 100 0 30 0 100 0 2 100 0 30 0 100 0 Time period 3 4 100 100 0 0 30 30 0 0 100 100 0 0 21 5 100 0 30 0 100 0 6 100 0 30 0 100 0 .6 24.4 0.5 Table 4.6 6 0.4 22.7 24.4 0.6 0.20 1.6 2 0.5 0.6 5 0.5 0.6 0.20 5.70 23.5 Time period 3 4 2.10 5.40 2.9 21.20 2.1 20.40 5.5 21.20 22.7 0.5 0.5 0.7 0. Production capacity ( Qit ) [kton] of each process Process P1 P2 P3 Existing capacity 27 30 25 Table 3.2 5 2.20 2.5 0.

25 Time period 3 4 2.50 5.25 22 5 2.25 2.25 2.25 5.35 2.35 2.25 .10 2.25 2.10 5.30 2.00 5.50 5.30 2.50 5.30 2.50 5.35 2.30 5. UB Production D E 2 2 10 2 30 10 2 30 Table 7.30 6 2. Upper and lower bounds for product demand ( d Ljt .10 5.15 2. UB Invent.10 5.15 2.35 2.50 5.25 2.30 Time period 3 4 2.10 5.15 2.00 5.25 6 2.00 5.30 2.50 5. ϕ bjt2 ) [100$/ton] with the bulk contract at time t Chemical A B C Scheme 1 2 1 2 1 2 1 2.50 5.35 2.25 2.00 5.10 2.00 2.30 Table 8.35 2.15 2. d Ujt ) [kton] at time period t Chemical 1 20 5 51 5 UB LB UB LB D E 2 25 5 50 5 Time period 3 4 22 30 5 5 53 60 5 5 5 28 5 59 5 6 26 5 50 5 Table 6.30 2 2.15 2.10 2.15 2.35 2.50 5.35 2.30 2.35 2. penalty Shortf.35 2.Table 5. cost Invent.10 2. and inventory cost ( ξ jt ) [100$/ton] and upper bound ( V jtU ) [kton] for each product ( t = 1. ϕ djt 2 ) [100$/ton] with the discount contract at time t Chemical A B C Scheme 1 2 1 2 1 2 1 2.50 5.30 5 2.30 2.25 2. Prices of the raw materials ( ϕ djt1 . K . 6 ) Shortf. Shortfall penalty ( θ jt ) [100$/ton] and upper bound ( SF jtU ) [kton].10 2.00 5.10 5.25 2 2.10 2.10 2. Prices of the raw materials ( ϕ bjt1 .50 5.50 5.35 2.50 5.35 2.

30 Table 10.50 22.Table 9. Prices of the raw materials ( ϕ lpjτ ) [100$/ton] with the length contract at time period t Chemical A B C Length 1 2 3 1 2 3 1 2 3 1 2.50 5.00 2 24.50 20.00 22.15 5. 6 ) Fixed Discount Bulk A B C 0 0 0 63 4 22 64 5 24 1 0 0 0 Length 2 63 3 22 3 66 4 24 Table 11.00 22.35 2.10 21.35 2.50 20.ψ djt 2 ) [100$/ton] with the discount contract at time t Chemical D E Scheme 1 2 1 2 1 24.25 2.25 2.50 20.40 5.40 5.10 21.40 6 24.50 20.25 2.25 2.35 2.25 2.50 20.35 2.50 20.25 2.15 2.15 2.10 21.40 20.50 22.50 20.30 Time period 3 4 2.10 24.50 5.00 22.50 20.40 5.50 22.40 Table 12.40 2 24.40 5.25 2.15 5.20 2.50 20.40 5.30 6 2.00 22.40 5.00 Time period 3 4 24.20 2.ψ bjt2 ) [100$/ton] with the bulk contract at time t Chemical D E Scheme 1 2 1 2 1 24.50 5.25 2.15 2.35 2.00 . Prices of the products (ψ bjt1 .15 2.20 2.50 21.10 22.50 20.10 21.50 22.50 5.50 22.50 22. Prices of the products (ψ djt1 .25 2.15 5.25 2.20 2.15 5.10 22.50 5.50 22.10 22.25 2.30 5 2.35 2.50 5.00 6 24. Minimum amounts ( σ cjt ) [kton] of raw materials in each contract Chemical t ( = 1.00 20.15 2.10 22.30 5.15 2.10 21.20 2.40 Time period 3 4 24.10 24.00 22. K .40 5 24.10 22.30 2 2.00 22.15 2.15 5.25 2.20 2.00 23 5 24.

20GHz Pentium PC.00 22. Cases 1 and 2 can be compared between them in order to see the effect of the contracts consideration in the model.10 23.50 21. Table 15. with 512Mbyte of RAM.50 22.73 214.40 20.40 5 24. the exploration of the branch and bound tree for this case is very large.05 22. The computational results are shown in Table 15.896. Minimum amounts ( ρ cjt ) [kton] of raw materials in each contract Chemical t ( = 1.50 21.91 7. 6 ) Fixed Discount Bulk D E 0 0 10 40 10 40 1 0 0 Length 2 10 40 3 20 50 5.848.45 20.40 Time period 3 4 24.05 22.45 21.00 22.45 20. K .05 23.10 23.00 22.Table 13.44 Solution [105 $] 7.10 23.40 2 24. Comparisons with case 3.05 22.16 The increase of CPU time in the last case is likely due to the existence of many choices of contract with the customers yielding very similar profit values.10 23.40 Table 14.00 22.40 6 24.00 22.2 Results The extended MILP problem P1 is modeled using the GAMS modeling language and solved using the CPLEX solver on a 3.05 22. Prices of the products (ψ lpjτ ) [100$/ton] with the length contract at time period t Chemical D E Length 1 2 3 1 2 3 1 24.83 8.50 21. however. Therefore.50 21. Computational statistics for the first example. 24 .05 22.10 23.00 22.45 20.50 21.45 20.45 20. are not possible since different data are used to express customer demand.10 24. Case 1 2 3 0-1 variables 0 198 336 Continuous variables 104 560 870 Constraints 159 626 938 CPU time [s] 0 8.193.

e. i. raw material B is not purchased. As it can be seen. results show.Case 2 2 in Example 1 70000 60000 fixed A Quantities [ton] 50000 fixed C discount A 40000 discount C bulk A 30000 bulk C length A 20000 length C 10000 0 1 2 3 4 5 Time periods Figure 7. that those choices with higher price (according to the prices used) are the preferred ones. as it can be expected. 2 and 3. Decision of contracts for suppliers in case 2 in Example 1. fixed contracts and length contracts with only one month duration. In both cases. Decision of Contracts with Suppliers . 25 6 . decisions for raw material contracts are exactly the same in both cases considered. and Figures 8 and 9 do the same for suppliers’ and customers’ contracts in case 3.Figure 7 shows which contracts are selected to purchase the raw materials in case 2. In the case of customers’ contracts.

Case 3 3 in Example 1 70000 60000 fixed A Quantities [ton] 50000 fixed C discount A 40000 discount C bulk A 30000 bulk C length A 20000 length C 10000 0 1 2 3 4 5 6 Time periods Figure 8.Case 3 60000 Quantities [ton] 50000 fixed D fixed E 40000 discount D discount E 30000 bulk D bulk E 20000 length D length E 10000 0 1 2 3 4 5 Time periods Figure 9.Purchase of raw materials . Decision of contracts for suppliers in case 3 in Example 1. 26 6 . 3 in Example 1 Sale of products . Decision of contracts for customers in case 3 in Example 1.

It can be seen that the slightly larger profit in the case of using contracts is due to a slightly smaller purchase cost in the case of using contracts whereas the other costs and revenues remain unchanged.Figure 10 shows a comparison between cases 1 and 2. Comparison in revenues and costs in cases 1 and 2. That behavior is typical of a case in which the system is working at full capacity. 10000 9000 8000 1e5 $ 7000 6000 Without contract 5000 With contract 4000 3000 2000 1000 Figure 10. 27 TF AL LS SH O R O PE IN VE N TO RY IO N RA T ES HA S PU RC EN RE V PR O FI UE S T 0 . This can be seen in Figure 11 that shows that demand satisfaction does not improve significantly with the choice of contracts.

This is an important difference from the case of suppliers’ contracts. The case of a fixed contract is the case where the demand in a given time interval is expressed as a fixed quantity between lower and upper bounds. Remarks on contracts with customers From the customer side in a supply chain. While bulk contracts are more realistic because they promote higher production levels. the fact of making contracts is tightly related to the way in which the customer demand is expressed in the model. length contracts are not profitable if one assumes deterministic demands. in which there are direct benefits for the company due to the economies of scale.96 Demand satisfaction [%] 95 94 93 92 91 Without contracts With contracts 90 89 0 1 2 3 4 5 6 7 Time periods Figure 11. In the case of customers’ length contracts. the equations considered for customer contracts are only useful in the context of stochastic programming approaches to address uncertainties in demand in real world applications 28 . Demand satisfaction per time period in cases 1 and 2. On the other hand. Therefore. benefits only derive if demands are uncertain as they have the effect of reducing the uncertainty. discount and bulk contracts are possible when the demand is expressed as a certain amount at a relatively high price or a larger quantity if a lower price is allowed.

29 .6 Example 2.1 Description This example includes four different case studies all based on a petrochemical network that involves 38 processes and 28 chemicals. 6. (1989)17). The process network considered in this study is shown in detail in Figure 12 and more schematically in Figure 13. (Sahinidis et al.

HCN ACRYLONITRILE ACRYLONITRILE 1 2 ACETYLENE 3 14 PROPYLENE PHENOL CUMENE 4 5 ISOPROPANOL BYPRODUCTS 6 NAPHTA ACETONE 9 8 38 10 7 CHLOROBENZENE 11 BENZENE ETHYLOBENZENE 12 STYRENE 13 ACETALDEHYDE 15 ETHANOL 18 16 17 ETHYLENE 19 22 23 ACETIC ACID 33 20 VINYL ACETATE 27 21 ETHYLENE 24 29 25 CHLOROHYDRIN ACETIC ANHYDRIDE ETHYLENE DICHLORIDE 26 32 EST ERS 32" ETHYLENE GLYCOL 31 37 FORMALDEHYDE 30 36 METHANOL GLYCOLIC ACID 36" 34 28 CARBON MONOXIDE Local market International 35 Figure 12. 30 . Process network diagram showing the direction of material flows in Example 2.

in the short-term it is necessary to include the inventories. ethyl benzene. chloro benzene. benzene. The outputs are: acrylonitrile. ethanol. ethylene. vinyl acetate. byproducts and also methanol. styrene. phenol. acetone. ϕ jtk and Pjtk . the process networks use as inputs the following chemicals: hydrogen cyanide. therefore. ethylene glycol. The four case studies are the following: Case 1. naphtha and methanol. There are no investment considerations and the purchase of raw materials is performed directly. cumene. the short-term problem is posed along a horizon time divided into 10 months. there are two intermediate products: ketene and ethylene chlorohydrin. without contracts. This introduces a new subscript k in ψ jtk . 31 . Finally. propylene. acetic anhydride.Figure 13. isopropanol. ethylene oxide. All these products are assumed to be stored before selling. S jtk . acetic acid. acetaldehyde. The products manufactured can be sold to the local and to the international market and the raw materials can also be bought from the local or the international market as shown in figure 13. carbon monoxide. acetylene. As can be seen in Figure 13. ethylene dichloride. In this case. formaldehyde. k = 1 accounts for local market and k = 2 for international market. Inputs and outputs of the process network studied in Example 2.

06 5.7 6. This case is a modification of case (1) but with the possibility of signing contracts for the acquisition of three of the main raw materials: naphtha (Naph).161 12.616 Continuous variables 12.43 For the short-term planning problem. Data for these problems are not presented given their great length.160 152 2. The computational using the CPLEX solver on a 3. and Figure 15 shows comparatively the amounts in cases 1 and 2.269 14.2 Results The LP and MILP resulting problems are modeled using the GAMS modeling language and solved PC. Figure 14 shows which contracts are selected for purchasing raw materials in case 2. where decisions on capacity expansions are included.897. Case 1 2 3 4 0-1 variables 0 6. the problem is a long-term one. For the contracts.20 0.416 46.95 22. but they are available from the authors. 6.002 5.Case 2. The time horizon covers 4 time periods of one year each.085. Case 3.017 CPU time [s] 0. we consider those types described previously in the paper.073. ethylene (Ethy) and acetylene (Acet). in the local market.606 5. In this case. Table 16.95 1. ethylene and acetylene in the local market.329 Constraints 13.606 40. Inventory considerations are not taken into account and the purchase of raw materials is performed without contracts.20GHz Pentium results are shown in Table 16.89 Time periods 10 10 4 4 Solution [105 $] 18.18 0.935. 32 . Case 4: It is also a capacity expansion problem but with the additional feature of allowing the choice of contracts for naphtha. Computational statistics for the second example. with 512Mbyte of RAM. The biggest difference is the increase in the purchase of naphtha (Naph) in case 2.

1000 900 Ethy with contract 800 Acet with contract Naph with contract Quantity [kton] 700 Ethy without contract 600 Naph without contract 500 9 without contracts 400 300 200 100 0 0 2 4 6 8 10 Time periods Figure 15. Decision of contracts with suppliers in case 2 in Example 2. 33 12 .1000 900 fixed Ethy 800 fixed Acet fixed Naph Qauntities [kton] 700 discount Ethy 600 discount Acet discount Naph 500 bulk Ethy bulk Acet 400 bulk Naph 300 length Ethy length Acet 200 length Naph 100 0 1 2 3 4 5 6 7 8 9 10 Time periods Figure 14. Amount of raw material bought in cases 1 and 2 in Example 2.

34 . Comparison of revenues and costs in cases 1 and 2 in Example 2. revenues).Figure 16 shows a comparison of the distribution of revenues and costs in both cases. In the case of making contracts the demand satisfaction is clearly improved. In this example the capacity of the system allows for a larger volume of sales with the resulting increase in all costs.e. Difference from example 1 lies in a drop of purchase costs. 90000 80000 70000 1E5 $ 60000 50000 Without Contracts With Contracts 40000 30000 20000 10000 0 PROFIT REVENUES PURCHASES OPERATION STORAGE*100 SHORTFALLS Figure 16. That can also be seen in Figure 17 comparing the percentage of demand satisfaction. The higher profit in the second case is due to a larger level of sales (i.

Demand satisfaction per time period in cases 1 and 2 in Example 2. Figure 19 shows comparatively the total amounts purchased per period in cases 3 and 4.80 Demand satisfaction [%] 70 60 50 40 30 20 Without contracts 10 With contracts 0 0 2 4 6 8 10 12 Time periods Figure 17. 35 . Figure 18 shows the contracts that are selected to purchase raw materials in case 4. the long-term problem with possibility of capacity expansion. As in the first two cases. the biggest difference between cases 3 and 4 is also the increase in the purchase of naphtha.

5 1 1. 500 Quantities [kton] Naph with contracts Ethy without contracts 400 Acet * 10 without cont.5 2 2. Amount of raw material bought in cases 3 and 4 in Example 2.5 . 36 4 4.5 Time periods Figure 19. Decision of contracts with suppliers in case 4 in Example 2. Naph without contracts 300 200 100 0 0 0. 600 Ethy with contracts Acet * 10 with cont.600 500 fixed Ethy fixed Acet fixed Naph Quantities [kton] 400 discount Ethy discount Acet discount Naph 300 bulk Ethy bulk Ethy bulk Naph 200 length Ethy length Acet length Naph 100 0 1 2 3 4 Time periods Figure 18.5 3 3.

Figure 20 shows a comparison between the NPV. Demand satisfaction per time period in cases 3 and 4 in Example 2. the higher NPV in case 4 is due to a larger sales volume.5 . Comparison of investment and costs in cases 3 and 4 in Example 2. 30000 25000 Without contract With contract 1e5 $ 20000 15000 10000 5000 0 NPV INVESTMENT REVENUES OPERATION PURCHASES Figure 20. the investments associated to the expansions and the costs in both cases. by comparing the percentage of demand satisfaction.5 3 3. 37 4. 60 Demand satisfaction [%] 50 40 30 20 Without contracts 10 With contracts 0 0 0. This is evident from Figure 21.5 4 Time periods Figure 21.5 1 1. Again.5 2 2.

2 and 3 periods ( LC = {1. The values are as follows for Tt p : p = 1 .3} ). by considering the option of signing contracts. We assume that the problem has six time periods ( T = {1. The results have clearly shown the benefits for the cases when the models were applied for the contracts for sales to customers. T61 = {6} (A1) 38 .4. owing to the economies of scale.2. even in the deterministic case. Tt p is a set of time periods τ at which the contract is made in order to purchase the raw material in time period t for length contract p . Four basic cases have been considered: fixed price. direct benefits can be derived to the company. discount a certain amount. In the case of contracts signed with suppliers.5.6} ) and the contract lengths are of 1. … . This is an interesting avenue to pursue. and for both suppliers and customers. Modeling contracts with customers should become more useful when considering stochastic problems as then the contracts have a decisive effect for reducing uncertainty. Appendix: Time period sets for the fixed length contract To illustrate the definition of sets Tt p .2. Acknowledgements The authors acknowledge the financial support provided by ‘BK21’ and ‘Generalitat de Catalunya (FI programs)’. bulk discount and fixed duration contracts. T11 = {1} . The proposed models were applied to both short-term and long-term planning problem. Tτ p in equations (47) and (49).3.7 Conclusions New models have been presented in order to expand the scope of traditional models for the planning problem of a chemical processes. T21 = {2} .

… . a). Equation (48) represents the amount of raw materials purchased in time period t .6} (A6) If we make a 2-period contract at time period 2. 4 (see Fig. T13 = {1.5} .2} .2. On the other hand. the cost of raw materials purchased at time period 5 with 3-period contract is the multiplication of the amounts purchased at time period 5 and the prices of contract made at time periods 3. T42 = {4.3} . T12 = {1. T22 = {2. and Pjl44 (see Fig. T43 = {4. the contract must be made at time periods 1. T62 = {5} (A2) p = 3 . T33 = {1. A1. we must buy some material in time period 2 and 3 (see Fig.5} . A1. T53 = {3. T11 = {1} . T23 = {2. A1. and the amount Pjl24 . … .6} (A5) p = 3 . T61 = {6} (A4) p = 2 . … . T21 = {2} . d). A1.4} . b).2.p = 2 . T13 = {1} . Tτ p is a set of time periods t in which the raw material is purchased with length contract p made at time period τ : p = 1 . Equation (49) expresses the terms of length contract described previously in the paper.3} .5. T63 = {4} (A3) If we buy some raw materials in time period 3 with the 3-period contract ( p = 3 ). Pjl343 . For example.4} . the amount Pjl342 and Pjl442 in 2- 3 3 period contract.2} . c). T52 = {5. … .3} . … . T22 = {1. The purchase amount Pjl4 of raw material j in time 1 period 4 is the summation of the amount Pjl44 with 1-period contract. 39 .2} . T23 = {1. 2 and 3 (see Fig.3. T52 = {4. T12 = {1} .

Nomenclature Indices subscripts i process j chemical k t type of market time period time period of the contract signature τ superscripts c L p U type of contract lower bound duration of the length contract upper bound Sets C I Ij Ji JM i JP JR LC types of contracts processes process that consume chemical j chemicals involved by process i main products of process i products raw materials contract lengths 40 . Length contract example with 6 time periods.Figure A1.

ρ jt ρ cjt ρ cp . jt greek letters U big-M value (large enough value) α it α it variable investment cost for process i at time t β it β it . lower and upper bounds of the availabilities of raw materials d Lj d Uj . .Oj T Tt p Tτp processes that produce chemical j time periods time periods τ at which the contract is made to purchase material for length contract p time periods t in which materials are purchased with length contract time period τ Parameters a Lj a Uj . jt SF jt selling price of material j at time period t under type of contract c maximum capital investment in time period t cost of purchasing material j at period t under type of contract c number of expansions allowed for process i total net present value amount of material j purchased at period t under type of contract c total profit over the planning time horizon installed capacity of process i at time period t capacity expansion of process i in period t revenues of selling product j at period t under type of contract c amount of j sold at period t under type of contract c V jt shortfall respect to the demand for chemical j at time period t inventory level of chemical j at time period t Wijt amount of chemical j being consumed and produced at t in process i integer (0-1) 41 p made at . . fixed investment cost for process i at time t δ it operating cost associated to process i at time t θ jt shortfall penalty of product j at time t µ ij mass factor of the product j in process i ξ jt inventory cost associated to product j at time t . NEXPi NPV P jt P jtc P jtcp . jt purchase price of material j at time period t under type of contract c ψ jt ψ cjt ψ cp . S jt S cjt S cp . . lower and upper bounds of the demands Uc . jt Variables real-positive CI t COST jt COST jtc . . . . jt minimum sellable quantity of j at time t under type of contract c σ jt σ cjt σ cp . PROFIT Qit QE it REV jt REV c jt . jt minimum purchasable quantity of j at time t under type of contract c ϕ jt ϕ cjt ϕ cp .

In Artificial Intelligence for Electronic Commerce. Wu. M. Special issue on Agent Technology for Electronic Commerce. process i is expanded in period t if true. (4) Sandholm. (9) Barbaro. Industrial and Engineering Chemistry Research. S. 1846-1857. MA: Kluwer Academic Publishers. AAAI Workshop 99.. (8) Gupta.. 29. C. Use of inventory and option contracts to hedge financial risk in planning under uncertainty managing financial risk in planning under uncertainty. 18. R. 802-810. (3) Grossmann. Process industry supply chains: Advances and challenges. Keskinocak. Computational Optimization and Applications. Supply Chain Contracts with Quantity Commitments and Stochastic Demand: Quantitative Models for Supply Chain Management. NJ: Prentice Hall. 2005. Akkiraju. R. “Enterprise-wide Optimization: A New Frontier in Process Systems Engineering. S. 42 . I. J. Intelligent decision support for the esupply chain. 83-100. 51. F. (5) Goodwin.. Computational Intelligence. (6) Anupindi... Murthy. F. (2) Grossmann. T. E..” AIChE Journal. 26. 990–998. (7) Hull. 2003.w it cp y cjt y jt . Y. 1225-1235. AIChE Journal. eMediator: A Next Generation Electronic Commerce Server. 656-676. 42. 1999. 2005... N. Englewood Cliffs. 50. p. Market-based pollution abatement strategies: Risk management using emission option contracts. Bagajewicz. Introduction to Futures and Options Markets. z cp jt if true. Generalized disjunctive programming: nonlinear convex hull relaxation and algorithms. P. R. 2002. Norwell.1995. A. product j is sold in period t using type of contract c References (1) Shah. z cjt . 1999. Bassok. Lee. A. Computers and Chemical Engineering.E. 2004. 770. I. Maranas. 2003. raw material j is purchased in period t using type of contract c if true.

29. S. E. Grossmann. I. Diaz. 24. 2004. Chathrathi.. Combined strategic and operational planning: An MILP success story in chemical industry. I. A. (15) Balas. Karimi. Computers and Chemical Engineering. P. J.. 2002. 41. Computers and Chemical Engineering. Optimization model for long range planning in the chemical industry. Grossmann. Bandoni. (17) Sahinidis. E. 1305-1316.. 1049-1063. A. J.. A. 13. 2791-2821. R. OR Spectrum. 43... Risk Management in the Scheduling of Batch Plants under Uncertain Market Demand. Puigjaner. (14) Chandra Prakash Reddy. Ind Industrial and Engineering Chemistry Research. M. New measures and procedures to manage financial risk with applications to the planning of gas commercialization in Asia. J.. Srinivasan. M. E. R. 3761-3770. M. Disjunctive programming and a hierarchy of relaxations for discrete optimization problems. 2004. 50. Computers and Chemical Engineering. P. 6. R. V.. N. J. 28. 1989. Bagajewicz. 466-486. 2004. (11) Aseeri. Bagajewicz. SIAM Journal of Algebraic Discrete Methods. 1985. 43 . Novel solution approach for optimizing crude oil operations. Industrial and Engineering Chemistry Research. Supply chain optimization of large-scale continuous processes. L. A.. E. Fornari. J.. (13) Schulz. 2005. High-level optimization model for the retrofit planning of process networks.. 1177-1197. 741-750. 2002. Espuña. AIChE Journal. 315-341. I.(10) Bonfill. (12) Kallrath. (16) Jackson. M. E.