2.1 Linear Programming Formulation, Graphical Solutions & Sensitivity Analysis

2.1 Linear Programming Formulation, Graphical Solutions & Sensitivity Analysis

Linear Programming (LP) is the cornerstone mathematical optimization technique utilized by industrial and systems engineers to allocate scarce operational resources—including labor hours, machine capacities, raw materials, capital, and transportation pathways—to maximize profit or minimize cost. Developed mathematically by George Dantzig in 1947 through the introduction of the Simplex algorithm, LP models system interactions using linear relationships.

---

Foundations of Mathematical Programming in Industrial Systems

A mathematical optimization model seeks to determine the optimal values of decision variables that optimize an objective function subject to a system of structural constraints.

Core Model Components

1. Decision Variables ($x_1, x_2, \dots, x_n$): Controllable, quantifiable continuous entities representing operational levels (e.g., units of product $j$ produced per week, operating hours of facility $i$, or barrels of crude oil routed through pipeline $k$).
2. Objective Function ($Z$): A single mathematical expression to be maximized (e.g., total contribution margin, revenue, system throughput) or minimized (e.g., total production cost, scrap generation, delivery transit time):

$\max Z = \sum_{j=1}^n c_j x_j \quad \text{or} \quad \min Z = \sum_{j=1}^n c_j x_j$

where $c_j$ represents the objective function coefficient (unit profit or unit cost) associated with decision variable $x_j$.

1. Structural Constraints: Mathematical inequalities or equalities defining technological limitations, physical capacities, contractual commitments, or legal regulations:

$\sum_{j=1}^n a_{ij} x_j \le b_i \quad (\text{Capacity limitation, } i \in I_{\le})$ $\sum_{j=1}^n a_{ij} x_j = b_i \quad (\text{Exact requirement, } i \in I_{=})$ $\sum_{j=1}^n a_{ij} x_j \ge b_i \quad (\text{Minimum specification, } i \in I_{\ge})$

where $a_{ij}$ represents the technological consumption coefficient (the quantity of resource $i$ consumed per unit of activity $j$), and $b_i$ represents the right-hand side (RHS) parameter indicating the total availability of resource $i$ or the minimum requirement.

1. Non-Negativity Restrictions: Physical operational activities cannot take negative values:

$x_j \ge 0 \quad \forall j = 1, 2, \dots, n$

Fundamental Assumptions of Linear Programming

For an optimization model to be formulated strictly as a linear program, four axiomatic assumptions must hold:

• Proportionality: The contribution of each decision variable to both the objective function and the constraints is directly proportional to the value of the variable. There are no volume discounts, setup penalties, or diminishing marginal returns within the modeled domain.
• Additivity: The total contribution of all variables to the objective function and constraints is the exact algebraic sum of their individual contributions. No interaction effects or synergetic couplings exist between variables.
• Divisibility (Continuity): Decision variables are continuous and may assume any non-negative real value (fractional values are permitted). If decision variables must be integer-valued (e.g., whole airplanes or integer personnel counts), the model belongs to Integer Linear Programming (ILP).
• Certainty: All parameters ($c_j, a_{ij}, b_i$) are known constants with deterministic certainty throughout the operational planning horizon.

---

Graphical Solution Methodology for Two-Variable Problems

When a linear program involves exactly two decision variables ($x_1, x_2$), the entire optimization space can be evaluated geometrically on a two-dimensional Cartesian plane. This provides visual intuition for high-dimensional simplex geometry.

Step-by-Step Graphical Procedure

1. Establish the Coordinate Axes: Set $x_1$ on the horizontal axis and $x_2$ on the vertical axis. The non-negativity constraints ($x_1 \ge 0, x_2 \ge 0$) immediately restrict all feasible solutions to the first quadrant.
2. Plot Constraint Boundary Lines: Convert each inequality $\sum a_{ij} x_j \le b_i$ into an equality $a_{i1} x_1 + a_{i2} x_2 = b_i$. Determine the intercepts:
   • Set $x_1 = 0 \implies x_2 = \frac{b_i}{a_{i2}}$
   • Set $x_2 = 0 \implies x_1 = \frac{b_i}{a_{i1}}$ Connect these two points to establish the boundary line.
3. Determine the Half-Space: Identify which side of the boundary line satisfies the original inequality by testing the origin $(0, 0)$. If $0 \le b_i$ is true, the region containing $(0, 0)$ is feasible; otherwise, shade the opposite half-space.
4. Delineate the Feasible Region ($\mathcal{F}$): The intersection of all half-spaces forms the feasible region. In linear programming, the feasible region is always a convex polyhedron (or polygon in 2D), meaning that a straight line segment connecting any two points within $\mathcal{F}$ lies entirely within $\mathcal{F}$.
5. Evaluate Extreme Points (Corner Points): By the Fundamental Theorem of Linear Programming, if an optimal solution exists, it occurs at an extreme point (vertex) of the feasible region.
6. Objective Function Contours (Isoprofit / Isocost Lines): Set $Z$ to an arbitrary constant $Z_0$ to generate the contour line $c_1 x_1 + c_2 x_2 = Z_0$, with slope $m = -\frac{c_1}{c_2}$. The gradient vector $\nabla Z = [c_1, c_2]^T$ indicates the direction of steepest increase. Translating the contour line parallel to itself in the direction of $\nabla Z$ until it last touches an extreme point identifies the optimal solution $(x_1^*, x_2^*)$.

Binding vs. Non-Binding Constraints and Slack/Surplus

• Binding Constraint: A constraint that is satisfied as a strict equality at the optimal solution point ($a_{i1}x_1^* + a_{i2}x_2^* = b_i$). The optimal point lies directly on the boundary line. Modifying the RHS $b_i$ will change the optimal objective value $Z^*$.
• Non-Binding Constraint: A constraint satisfied as a strict inequality at the optimal point ($a_{i1}x_1^* + a_{i2}x_2^* < b_i$ for a $\le$ constraint). The boundary line does not pass through the optimal vertex.
• Slack Variable ($s_i$): Introduced into a $\le$ constraint to quantify unused capacity:

$s_i = b_i - \sum_{j=1}^n a_{ij} x_j \ge 0$

For a binding constraint, $s_i = 0$. For a non-binding constraint, $s_i > 0$.

• Surplus Variable ($e_i$): Subtracted from a $\ge$ constraint to represent excess over minimum requirements:

$e_i = \sum_{j=1}^n a_{ij} x_j - b_i \ge 0$

---

Simplex Algorithm Mechanics & Algebraic Geometry

While graphical analysis is limited to two variables, real industrial problems contain thousands of variables and constraints. The Simplex method solves these problems by moving systematically from one extreme point (basic feasible solution) to an adjacent extreme point with an improved objective value.

Conversion to Standard Form

To apply the Simplex method, the LP must be cast in standard algebraic form:

• Maximize the objective function (a minimization problem $\min Z$ is converted to $\max (-Z)$).
• All structural constraints are expressed as strict equalities.
• All right-hand side constants are non-negative ($b_i \ge 0$).
• All decision, slack, and surplus variables are non-negative.

For a system with $m$ constraints and $n$ total variables (after adding slacks and subtracting surpluses), where $n > m$:

• Basic Variables ($m$ variables): Selected variables solved in terms of the non-basic variables. These variables form the basis.
• Non-Basic Variables ($n - m$ variables): Set identically to zero ($x_j = 0$).
• Basic Feasible Solution (BFS): A basic solution where all basic variables satisfy $x_{B,i} \ge 0$. Geometrically, each BFS corresponds exactly to an extreme point of the feasible region.

Simplex Pivot Mechanics

1. Optimality Check: In a maximization problem, evaluate the reduced costs ($z_j - c_j$) in the objective row. If all $z_j - c_j \ge 0$, the current BFS is globally optimal.
2. Entering Variable Selection (Pivot Column): Choose the non-basic variable with the most negative reduced cost ($z_k - c_k < 0$). This variable offers the highest rate of objective improvement per unit increase.
3. Leaving Variable Selection (Pivot Row / Minimum Ratio Test): To maintain feasibility, the entering variable $x_k$ can only increase until the first basic variable drops to zero. Compute the ratio of current RHS values to positive technological coefficients in the pivot column:

$\theta = \min_{i : a_{ik} > 0} \left\{ \frac{b_i}{a_{ik}} \right\}$

The basic variable corresponding to the minimum ratio leaves the basis. 4. Gauss-Jordan Elimination: Perform elementary row operations to transform the pivot element to 1 and all other elements in the pivot column to 0, yielding a new adjacent BFS.

Special Solution Outcomes in Linear Programming

---

Duality Theory & Economic Interpretation

Every linear programming problem, termed the Primal, possesses a corresponding symmetric companion problem termed the Dual. The dual problem provides economic valuation of the primal system's resources.

Primal-Dual Relationships

Consider the symmetric primal-dual formulation:

Primal Problem: $\max Z = \mathbf{c}^T \mathbf{x} \quad \text{subject to} \quad \mathbf{A}\mathbf{x} \le \mathbf{b}, \quad \mathbf{x} \ge \mathbf{0}$

Dual Problem: $\min W = \mathbf{b}^T \mathbf{y} \quad \text{subject to} \quad \mathbf{A}^T\mathbf{y} \ge \mathbf{c}, \quad \mathbf{y} \ge \mathbf{0}$

Where:

• $y_i$ represents the dual decision variable associated with primal constraint $i$.
• Each primal constraint corresponds to a dual variable.
• Each primal variable corresponds to a dual constraint.
• The primal objective coefficients $\mathbf{c}$ become the dual RHS vector.
• The primal RHS vector $\mathbf{b}$ becomes the dual objective coefficients.
• The technological matrix $\mathbf{A}$ is transposed to $\mathbf{A}^T$.

Core Duality Theorems

1. Weak Duality Theorem: For any primal feasible solution $\mathbf{x}$ and any dual feasible solution $\mathbf{y}$:

$\mathbf{c}^T \mathbf{x} \le \mathbf{b}^T \mathbf{y}$

The objective value of any dual feasible solution provides an upper bound on the objective value of any primal feasible solution.

1. Strong Duality Theorem: If the primal problem has an optimal solution $\mathbf{x}^*$, then the dual problem also has an optimal solution $\mathbf{y}^*$, and their optimal objective values are identical:

$Z^* = \mathbf{c}^T \mathbf{x}^* = \mathbf{b}^T \mathbf{y}^* = W^*$

1. Complementary Slackness Theorem: Let $\mathbf{x}^*$ be primal feasible and $\mathbf{y}^*$ be dual feasible. They are mutually optimal if and only if:

$y_i^* \cdot \left( b_i - \sum_{j=1}^n a_{ij} x_j^* \right) = 0 \quad \forall i = 1, \dots, m$ $x_j^* \cdot \left( \sum_{i=1}^m a_{ij} y_i^* - c_j \right) = 0 \quad \forall j = 1, \dots, n$

Managerial Consequence: If primal constraint $i$ has positive slack ($s_i^* > 0$), its dual valuation is zero ($y_i^* = 0$). Conversely, if a resource has a positive shadow price ($y_i^* > 0$), its corresponding constraint must be strictly binding ($s_i^* = 0$).

---

Post-Optimality & Sensitivity Analysis

Sensitivity analysis examines how perturbations in input parameters ($c_j, b_i, a_{ij}$) affect the optimal solution without resolving the linear program from scratch.

1. Shadow Price (Dual Value $y_i^*$)

The shadow price of constraint $i$ is the marginal rate of change in the optimal objective value $Z^*$ resulting from a one-unit increase in the right-hand side $b_i$:

$y_i^* = \frac{\partial Z^*}{\partial b_i}$

• In a maximization problem, a $\le$ resource constraint has $y_i^* \ge 0$, indicating the maximum price an enterprise should be willing to pay for one additional unit of resource $i$ beyond its standard cost.
• For a non-binding constraint, surplus capacity exists; hence, $y_i^* = 0$.

2. Allowable Range of Feasibility (for RHS $b_i$)

The range of values for $b_i$ over which the current optimal basis (set of basic variables) remains unchanged. Within this range:

• The shadow price $y_i^*$ remains constant.
• The coordinates of the optimal extreme point change linearly as functions of $\Delta b_i$.
• The optimal objective value changes at the constant rate: $\Delta Z^* = y_i^* \cdot \Delta b_i$. If $b_i$ is changed beyond the allowable range, the shadow price changes because a different set of constraints becomes binding.

3. Allowable Range of Optimality (for Objective Coefficients $c_j$)

The range of values for an objective function coefficient $c_j$ over which the current optimal extreme point coordinates $(x_1^*, x_2^*, \dots, x_n^*)$ remain strictly identical.

• Within this range, the decision variables do not change, but the total objective value changes at rate $x_j^*$: $\Delta Z^* = x_j^* \cdot \Delta c_j$.
• In two dimensions, this corresponds to rotating the isoprofit line between the slopes of the two binding constraint lines intersecting at the optimal vertex.

4. Reduced Cost ($r_j$)

The reduced cost of a decision variable $x_j$ indicates the amount by which its objective function coefficient $c_j$ must improve before it becomes economical to produce a positive quantity ($x_j > 0$):

• If $x_j^* > 0$ in the optimal solution, its reduced cost is strictly zero ($r_j = 0$).
• If $x_j^* = 0$ (non-basic), $r_j$ represents the marginal penalty incurred per unit of $x_j$ introduced into the basis: $r_j = \sum_{i=1}^m a_{ij} y_i^* - c_j$ (for maximization).

---

Step-by-Step Worked Numerical Example

Problem Formulation: Precision Machining Facility

A precision manufacturing plant produces two advanced aerospace components: Component 1 ($x_1$) and Component 2 ($x_2$). Each unit generates unit contribution margins of:

• Component 1: $50/unit
• Component 2: $40/unit

Production requires processing across three dedicated departmental workcenters with finite available weekly hours:

1. Milling Workcenter: $2x_1 + 1x_2 \le 100$ hours available
2. Assembly Workcenter: $1x_1 + 1x_2 \le 80$ hours available
3. Finishing Workcenter: $1x_1 + 3x_2 \le 180$ hours available

Non-negativity requires $x_1 \ge 0, x_2 \ge 0$.

$\max Z = 50x_1 + 40x_2$ $\text{s.t.} \quad 2x_1 + x_2 \le 100 \quad (\text{Milling})$ $x_1 + x_2 \le 80 \quad (\text{Assembly})$ $x_1 + 3x_2 \le 180 \quad (\text{Finishing})$ $x_1, x_2 \ge 0$

Step 1: Constraint Boundary Analysis & Extreme Points

Plotting the boundary equations:

• Milling Boundary: $2x_1 + x_2 = 100 \implies (0, 100)$ and $(50, 0)$.
• Assembly Boundary: $x_1 + x_2 = 80 \implies (0, 80)$ and $(80, 0)$.
• Finishing Boundary: $x_1 + 3x_2 = 180 \implies (0, 60)$ and $(180, 0)$.

Evaluate all candidate intersections in the positive quadrant:

1. Origin $(0, 0)$: Feasible. $Z = 50(0) + 40(0) = \text{\textdollar}0$.
2. $x_1$-intercept: The tightest constraint along the $x_1$-axis is Milling at $(50, 0)$. Check other constraints:
   • Assembly: $50 + 0 = 50 \le 80$ (Feasible)
   • Finishing: $50 + 3(0) = 50 \le 180$ (Feasible)
   • $Z = 50(50) + 40(0) = \text{\textdollar}2{,}500$.
3. Intersection of Milling and Finishing: $2x_1 + x_2 = 100 \implies x_2 = 100 - 2x_1$ Substitute into Finishing: $x_1 + 3(100 - 2x_1) = 180 \implies x_1 + 300 - 6x_1 = 180 \implies -5x_1 = -120 \implies x_1 = 24$ $x_2 = 100 - 2(24) = 52$ Verify against Assembly constraint: $x_1 + x_2 = 24 + 52 = 76 \le 80 \quad (\text{Feasible! Slack } s_2 = 80 - 76 = 4 \text{ hours})$ Calculate objective value: $Z = 50(24) + 40(52) = 1,200 + 2,080 = \text{\textdollar}3{,}280$
4. Intersection of Assembly and Finishing: $x_1 + x_2 = 80 \implies x_1 = 80 - x_2$ Substitute into Finishing: $(80 - x_2) + 3x_2 = 180 \implies 2x_2 = 100 \implies x_2 = 50, x_1 = 30$. Check Milling constraint: $2(30) + 50 = 110 > 100$ (Infeasible! Violates Milling capacity).
5. Intersection of Milling and Assembly: $2x_1 + x_2 = 100$ and $x_1 + x_2 = 80$. Subtracting gives $x_1 = 20, x_2 = 60$. Check Finishing: $20 + 3(60) = 200 > 180$ (Infeasible! Violates Finishing capacity).
6. $x_2$-intercept: The tightest constraint along the $x_2$-axis is Finishing at $(0, 60)$. Check others:
   • Milling: $2(0) + 60 = 60 \le 100$ (Feasible)
   • Assembly: $0 + 60 = 60 \le 80$ (Feasible)
   • $Z = 50(0) + 40(60) = \text{\textdollar}2{,}400$.

Summary of Extreme Points

The optimal production plan is $x_1^* = 24$ units of Component 1 and $x_2^* = 52$ units of Component 2, yielding a maximum contribution margin of $3,280.

---

Step 2: Exact Sensitivity Report Derivations

A. Shadow Prices ($y_1^*, y_2^*, y_3^*$)

Since Assembly is non-binding ($s_2 = 4$ hours), by Complementary Slackness: $y_2^* = \text{\textdollar}0.00/\text{hour}$

To find the shadow prices of the binding constraints (Milling $b_1$ and Finishing $b_3$), perturb $b_1$ by $\Delta b_1$: $2x_1 + x_2 = 100 + \Delta b_1$ $x_1 + 3x_2 = 180$ From the second equation, $x_1 = 180 - 3x_2$. Substitute into the first: $2(180 - 3x_2) + x_2 = 100 + \Delta b_1 \implies 360 - 5x_2 = 100 + \Delta b_1 \implies x_2 = 52 - 0.2\Delta b_1$ $x_1 = 180 - 3(52 - 0.2\Delta b_1) = 24 + 0.6\Delta b_1$

Calculate the rate of change in $Z$: $Z = 50(24 + 0.6\Delta b_1) + 40(52 - 0.2\Delta b_1) = 3,280 + 30\Delta b_1 - 8\Delta b_1 = 3,280 + 22\Delta b_1$ $\implies y_1^* = \frac{\partial Z^*}{\partial b_1} = \text{\textdollar}22.00/\text{hour}$

Similarly, perturbing Finishing capacity $b_3$ by $\Delta b_3$: $2x_1 + x_2 = 100$ $x_1 + 3x_2 = 180 + \Delta b_3$ Solving yields $x_1 = 24 - 0.2\Delta b_3$ and $x_2 = 52 + 0.4\Delta b_3$. $Z = 50(24 - 0.2\Delta b_3) + 40(52 + 0.4\Delta b_3) = 3,280 - 10\Delta b_3 + 16\Delta b_3 = 3,280 + 6\Delta b_3$ $\implies y_3^* = \frac{\partial Z^*}{\partial b_3} = \text{\textdollar}6.00/\text{hour}$

B. Allowable Range of Optimality for $c_1$

The binding constraints at $(24, 52)$ are:

• Milling: $2x_1 + x_2 = 100 \implies x_2 = 100 - 2x_1$ (Slope $m_1 = -2$)
• Finishing: $x_1 + 3x_2 = 180 \implies x_2 = 60 - \frac{1}{3}x_1$ (Slope $m_3 = -\frac{1}{3}$)

The slope of the objective function is $m_Z = -\frac{c_1}{c_2} = -\frac{c_1}{40}$. For the extreme point $(24, 52)$ to remain optimal, the slope of the objective contour must lie between the slopes of the two binding constraints: $-2 \le -\frac{c_1}{40} \le -\frac{1}{3}$ Multiply across by $-40$ (reversing the inequality signs): $\frac{40}{3} \le c_1 \le 80 \implies 13.33 \le c_1 \le 80.00$

• Allowable Increase for $c_1$: $80 - 50 = \text{\textdollar}30.00$
• Allowable Decrease for $c_1$: $50 - 13.33 = \text{\textdollar}36.67$

C. Allowable Ranges for the Binding Right-Hand Sides

The shadow prices stay valid only while the same two constraints remain binding and every variable stays non-negative.

• Milling ($b_1 = 100 + \Delta b_1$): the solution is $x_1 = 24 + 0.6\Delta b_1$ and $x_2 = 52 - 0.2\Delta b_1$, so Assembly uses $76 + 0.4\Delta b_1 \le 80$. Assembly becomes binding at $\Delta b_1 = +10$, and $x_1$ reaches zero at $\Delta b_1 = -40$. Allowable increase $= 10$ hours; allowable decrease $= 40$ hours.
• Finishing ($b_3 = 180 + \Delta b_3$): the solution is $x_1 = 24 - 0.2\Delta b_3$ and $x_2 = 52 + 0.4\Delta b_3$, so Assembly uses $76 + 0.2\Delta b_3 \le 80$. Assembly becomes binding at $\Delta b_3 = +20$, and $x_2$ reaches zero at $\Delta b_3 = -130$. Allowable increase $= 20$ hours; allowable decrease $= 130$ hours.

Comprehensive Sensitivity Report