7.4 Supply Chain Logistics, Transportation Simplex & Distribution Network Design

Supply Chain Logistics, Transportation Simplex & Distribution Network Design

Logistics and physical distribution constitute the critical operational connective tissue linking manufacturing plants, distribution warehouses, retail outlets, and end customers. In modern industrial systems, logistics expenditures—comprising freight freight, facility leases, material handling, customs duties, and distribution inventory—typically represent 8% to 15% of a corporation's gross revenues.

Designing and operating an optimal distribution network requires industrial engineers to solve complex multi-facility location trade-offs, inventory centralization problems, linear network flow formulations (The Transportation Problem), and localized fleet dispatching (Vehicle Routing Problems).

---

1. Logistics Network Architectures & Inventory Centralization

Network Topology Paradigms

Industrial distribution networks are organized into four primary topological archetypes:

1. Direct Shipment Network: Goods are shipped directly from supplier factories to customer demand points without intermediate handling.
   • Advantages: Eliminates warehousing overhead, minimizes handling damage, and shortens transit duration.
   • Disadvantages: Demands high order volumes to achieve Full Truckload (FTL) rates; Less-Than-Truckload (LTL) shipments incur prohibitive freight premiums.
2. Hub-and-Spoke Distribution: Feeder shipments from multiple originating factories converge at a central terminal (hub), where loads are sorted, consolidated, and dispatched along major trunk lines to regional destinations (spokes).
3. Cross-Docking Facilities: Intermediate transfer hubs where inbound shipments from manufacturing plants are unloaded, deconsolidated, sorted according to destination, and loaded directly onto outbound delivery trucks with minimal or zero intermediate storage (dwell time typically < 24 hours).
   • Prerequisites: High-velocity product turnover, tight EDI/RFID shipment tracking, and synchronized dock scheduling.
4. Tiered Multi-Echelon Networks: Hierarchical architecture featuring a Central Distribution Center (CDC) supporting several Regional Distribution Centers (RDCs), which in turn replenish local forward fulfillment nodes.

The Square Root Law of Inventory Consolidation

A central strategic question in supply chain engineering is whether to distribute inventory across multiple regional warehouses (close to customers) or consolidate stock into a single centralized distribution hub.

Consolidating inventory captures the benefits of risk pooling: customer demand variations across geographic regions are rarely perfectly correlated. An unexpected demand surge in Region A is frequently offset by an unexpected slump in Region B, dampening the aggregate variance seen by a central warehouse.

Mathematical Derivation of the Square Root Law

Assume a corporation operates $k$ identical, independent regional warehouses. In each warehouse $j$, customer lead time demand has mean $\mu_j$ and standard deviation $\sigma_j = \sigma$. To achieve service level $z$, each decentralized warehouse holds safety stock:

$SS_j = z \cdot \sigma$

The total decentralized safety stock across all $k$ regional facilities is:

$SS_{decentralized} = \sum_{j=1}^k SS_j = k \cdot (z \cdot \sigma)$

If all $k$ regions are consolidated into a single centralized facility, total customer demand is the sum of $k$ independent random variables. The variance of aggregate demand is:

$\sigma_{centralized}^2 = \sum_{j=1}^k \sigma^2 = k \sigma^2 \implies \sigma_{centralized} = \sqrt{k} \cdot \sigma$

The safety stock required at the single centralized distribution center is:

$SS_{centralized} = z \cdot \sigma_{centralized} = z \cdot (\sqrt{k} \cdot \sigma) = \sqrt{k} \cdot (z \sigma)$

Dividing centralized safety stock by total decentralized safety stock establishes the Square Root Law:

$\frac{SS_{centralized}}{SS_{decentralized}} = \frac{\sqrt{k} (z \sigma)}{k (z \sigma)} = \frac{\sqrt{k}}{k} = \frac{1}{\sqrt{k}}$

$SS_{centralized} = \frac{SS_{decentralized}}{\sqrt{k}}$

Decentralized (4 Regional DCs):                     Centralized (1 Master Hub):
+---------+     +---------+
| DC 1    |     | DC 2    |
| SS = 500|     | SS = 500|                     +-------------------+
+---------+     +---------+      Consolidate    | Central Master DC |
+---------+     +---------+     ============>   |                   |
| DC 3    |     | DC 4    |                     | SS = 1,000 units  |
| SS = 500|     | SS = 500|                     +-------------------+
+---------+     +---------+                     Total SS Cut by 50%!
Total SS = 2,000 units                          (2,000 / sqrt(4))

For example, consolidating 4 regional warehouses ($k = 4$) cuts total required safety stock in half ($1/\sqrt{4} = 0.50$, a 50% working capital reduction); consolidating 9 warehouses reduces safety stock by 66.7% ($1/\sqrt{9} = 0.333$).

Strategic Trade-off: Centralization dramatically slashes inventory holding costs, but increases outbound delivery distances and transit lead times, potentially increasing outbound freight charges.

2. Mathematical Formulation of the Transportation Problem

The classical Transportation Problem is a specialized class of linear programming models designed to determine the cost-minimizing distribution plan from $m$ supply origins (factories, ports) to $n$ demand destinations (warehouses, retail stores).

Linear Programming Formulation

Let:

• $m$ = Number of supply sources
• $n$ = Number of demand destinations
• $s_i$ = Supply capacity available at source $i$ ($i = 1, \dots, m$)
• $d_j$ = Demand requirement at destination $j$ ($j = 1, \dots, n$)
• $c_{ij}$ = Unit shipping cost from source $i$ to destination $j$
• $x_{ij}$ = Quantity transported from source $i$ to destination $j$ (Decision Variable)

$\min Z = \sum_{i=1}^m \sum_{j=1}^n c_{ij} x_{ij}$

Subject to:

$\sum_{j=1}^n x_{ij} \le s_i \quad \forall i = 1, \dots, m \quad (\text{Source Supply Constraints})$

$\sum_{i=1}^m x_{ij} \ge d_j \quad \forall j = 1, \dots, n \quad (\text{Destination Demand Constraints})$

$x_{ij} \ge 0 \quad \forall i, j \quad (\text{Non-negativity Constraints})$

Balanced vs. Unbalanced Transportation Networks

A transportation problem is balanced when total aggregate supply precisely matches total aggregate demand:

$\sum_{i=1}^m s_i = \sum_{j=1}^n d_j$

When networks are unbalanced, they must be converted into balanced formulations prior to applying transportation algorithms:

1. Excess Supply ($\sum s_i > \sum d_j$): Introduce a fictitious Dummy Destination ($n+1$) with demand equal to the surplus supply: $d_{n+1} = \sum s_i - \sum d_j$. Set all unit shipping costs to zero ($c_{i, n+1} = 0$), or to inventory holding costs if excess stock represents warehouse storage.
2. Excess Demand ($\sum d_j > \sum s_i$): Introduce a fictitious Dummy Source ($m+1$) with capacity equal to the unmet demand: $s_{m+1} = \sum d_j - \sum s_i$. Set unit costs to zero ($c_{m+1, j} = 0$), or to shortage penalty costs ($c_{m+1, j} = p_j$) if unmet demand represents contractual stockout costs.

Basic Feasible Solutions and Degeneracy

In a balanced transportation problem with $m$ sources and $n$ destinations:

• The linear program contains $m + n$ functional constraints.
• Because total supply equals total demand, one constraint is mathematically redundant.
• Consequently, any valid, non-degenerate Basic Feasible Solution (BFS) must contain exactly $m + n - 1$ basic variables (allocated cells).
• The allocated basic cells must be acyclic (free of closed rectangular loops).
• Degeneracy: Occurs when fewer than $m + n - 1$ cells receive positive allocations (happens whenever a subset of source capacities equals a subset of destination demands). Degeneracy is resolved by assigning an infinitesimal positive allocation $\epsilon > 0$ to an unallocated cell to maintain $m + n - 1$ basic cells, allowing dual multipliers to be evaluated.

3. Initial Basic Feasible Solution (IBFS) Heuristics

Before executing optimality testing, an initial basic feasible solution must be constructed. Three classic algorithms generate an IBFS:

1. Northwest Corner Rule

Begins at the upper-left (Northwest) cell $(1, 1)$ without regard to unit shipping costs:

1. Allocate $x_{11} = \min(s_1, d_1)$.
2. Subtract $x_{11}$ from $s_1$ and $d_1$.
3. If source 1 is exhausted, step down to cell $(2, 1)$. If destination 1 is satisfied, step right to cell $(1, 2)$.
4. Repeat until all supplies and demands are met. Assessment: Simple and fast, but completely ignores shipping costs, frequently yielding an expensive initial solution.

2. Intuitive / Minimum Cost (Greedy) Method

Allocates flow based on lowest unit costs:

1. Search the entire transportation tableau for the cell $(i, j)$ with the smallest unit cost $c_{ij}$.
2. Allocate $x_{ij} = \min(s_i, d_j)$.
3. Cross out the saturated row or column and adjust remaining supply and demand.
4. Repeat among surviving cells until all constraints are satisfied. Assessment: Substantially better than Northwest Corner, but can inadvertently force massive allocations into expensive cells during final steps.

3. Vogel's Approximation Method (VAM)

Vogel's Approximation Method (VAM) is the gold-standard heuristic, capturing the opportunity cost (penalty) of not choosing the cheapest route.

VAM Algorithmic Steps:

1. Compute Penalties: For each row and column, calculate a penalty value defined as the absolute difference between the two lowest unit shipping costs in that row or column.
2. Select Maximum Penalty: Identify the row or column with the largest penalty (breaking ties arbitrarily). This row or column faces the worst economic consequence if its cheapest route is missed.
3. Allocate to Minimum Cost Cell: In the selected row or column, allocate the maximum possible flow $x_{ij} = \min(s_i, d_j)$ to the cell having the lowest unit cost.
4. Update Matrix: Reduce the corresponding supply and demand. Cross out the row or column that is fully satisfied. If both a row and column are satisfied simultaneously, cross out only one and assign supply/demand of zero to the other to avoid premature degeneracy.
5. Iterate: Recompute penalties for remaining uncrossed rows and columns, repeating steps 1–4 until all supplies and demands are allocated.

Assessment: Consistently delivers near-optimal (and frequently globally optimal) starting solutions.

4. Optimality Testing: MODI & Stepping-Stone Methods

Once an initial basic feasible solution with $m + n - 1$ basic cells is established, it must be evaluated for global optimality.

The MODI (Modified Distribution / $u-v$) Method

The MODI method applies linear programming duality directly to the transportation tableau.

Step 1: Compute Dual Multipliers ($u_i, v_j$)

Associate a dual variable $u_i$ with each supply row $i$ and $v_j$ with each demand column $j$. For every basic cell (where $x_{ij} > 0$), the dual equation must hold:

$u_i + v_j = c_{ij}$

This yields $m + n - 1$ independent linear equations in $m + n$ unknowns. Set one variable arbitrarily as a base (standard convention: set $u_1 = 0$) and solve algebraically for all remaining $u_i$ and $v_j$.

Step 2: Compute Opportunity Costs (Reduced Costs) for Non-Basic Cells

For every non-basic (empty) cell $(i, j)$ where $x_{ij} = 0$, compute its opportunity cost $\Delta_{ij}$:

$\Delta_{ij} = c_{ij} - (u_i + v_j)$

Step 3: Optimality Criterion

• If all $\Delta_{ij} \ge 0$: The current allocation is globally optimal. No reallocation can reduce total shipping costs.
• If any $\Delta_{ij} < 0$: The current allocation is non-optimal. Introducing shipments into a cell with a negative opportunity cost will reduce total system cost. The cell with the most negative $\Delta_{ij}$ is selected as the entering basic variable.

The Stepping-Stone Method

When a non-basic cell $(e_i, e_j)$ is selected to enter the basis, the flow shift is executed using the Stepping-Stone Loop:

1. Trace Closed Rectilinear Loop: Trace a unique closed loop of alternating horizontal and vertical line segments starting and ending at the entering cell. Every corner of the loop must turn on an existing basic cell.
2. Assign Alternating Signs: Label the entering cell with $+$. Proceed around the loop labeling consecutive corners with $-, +, -, +, \dots$
3. Determine Reallocation Flow ($\theta$): Identify all basic cells with a $-$ sign. The maximum flow that can be transferred without violating non-negativity is: $\theta = \min \{x_{ij} \mid (i, j) \text{ is a cell with a } - \text{ sign}\}$
4. Pivot Flow: Add $\theta$ to all cells with a $+$ sign, and subtract $\theta$ from all cells with a $-$ sign. The entering cell becomes basic with allocation $\theta$, while the donor cell that hit zero leaves the basis.
5. Re-evaluate: Compute new $u_i, v_j$ multipliers and repeat until all $\Delta_{ij} \ge 0$.

5. Vehicle Routing (VRP) & Traveling Salesperson (TSP) Fundamentals

While the transportation problem evaluates aggregate warehouse-to-store transfers, last-mile distribution involves dispatching vehicles along sequence-dependent road networks.

The Traveling Salesperson Problem (TSP)

The TSP seeks the shortest Hamiltonian cycle visiting $n$ distinct customer nodes and returning to the depot, visiting every customer exactly once.

• Complexity: TSP is $NP$-hard; an $n$-city symmetric problem has $(n-1)! / 2$ tours.
• Nearest Neighbor Heuristic: Starting at depot, continuously visit the closest unvisited node. Fast ($O(n^2)$), but can force very long final closing arcs.
• Christofides Algorithm: Guarantees a tour length within $1.5 \times$ optimal for metric TSP problems by combining a Minimum Spanning Tree (MST) with minimum-weight perfect matching on odd-degree vertices and shortcutting Eulerian tours.

The Vehicle Routing Problem (VRP) & Clarke-Wright Savings

The Capacitated VRP (CVRP) extends TSP to multiple vehicles with capacity constraint $Q_{veh}$ servicing customer demands $q_i$ from a central depot $0$.

The Clarke-Wright Savings Algorithm (1964):

1. Begin with an initial trivial baseline where every customer $i$ is served by a dedicated direct out-and-back route from depot $0$ ($0 \to i \to 0$). Total cost is $\sum 2 c_{0i}$.
2. If customers $i$ and $j$ are consolidated into a single combined route ($0 \to i \to j \to 0$), travel between $i$ and depot, and $j$ and depot, is replaced by direct link $(i, j)$.
3. The Savings Metric $s(i, j)$ achieved by linking customer $i$ and customer $j$ is: $s(i, j) = c_{0i} + c_{0j} - c_{ij}$
4. Rank all customer pairs $(i, j)$ in descending order of savings $s(i, j)$.
5. Starting from the top of the savings list, merge routes provided that:
   • Neither customer is an interior node on an existing route (both must be directly linked to depot).
   • Merging does not violate vehicle capacity: $\sum_{k \in \text{Route}} q_k \le Q_{veh}$.

6. Worked Engineering Examples

Complete Transportation Problem: VAM, MODI & Optimality Testing

Problem Statement: A heavy equipment manufacturer ships components from three regional factories (S1, S2, S3) to three regional distribution hubs (D1, D2, D3). Supply capacities, destination demands, and unit transportation costs ($/unit) are:

Total supply = $30 + 40 + 30 = 100$. Total demand = $20 + 35 + 45 = 100$. The problem is balanced ($m = 3, n = 3$). A valid basic feasible solution requires $m + n - 1 = 3 + 3 - 1 = 5$ basic cells.

Step 1: Generate Initial BFS using Vogel's Approximation Method (VAM)

Calculate Initial Penalties (Difference between two lowest costs):

• Row S1: 4, 6, 8 $\implies$ Penalty = $6 - 4 = 2$
• Row S2: 3, 4, 9 $\implies$ Penalty = $4 - 3 = 1$
• Row S3: 5, 7, 8 $\implies$ Penalty = $7 - 5 = 2$
• Col D1: 4, 6, 7 $\implies$ Penalty = $6 - 4 = 2$
• Col D2: 5, 8, 9 $\implies$ Penalty = $8 - 5 = 3$ (Largest Penalty = 3)
• Col D3: 3, 4, 8 $\implies$ Penalty = $4 - 3 = 1$

Column D2 has the highest penalty (3). Lowest cost in Col D2 is cell (S3, D2) with cost $5.

• Allocate $x_{32} = \min(s_3 = 30, d_2 = 35) = 30$.
• S3 is exhausted (capacity = 0). D2 remaining demand = $35 - 30 = 5$.
• Cross out Row S3.

Recalculate Penalties for remaining uncrossed cells (S1, S2 vs D1, D2, D3):

• Row S1: 4, 6, 8 $\implies$ Penalty = $6 - 4 = 2$
• Row S2: 3, 4, 9 $\implies$ Penalty = $4 - 3 = 1$
• Col D1: 4, 6 $\implies$ Penalty = $6 - 4 = 2$
• Col D2: 8, 9 $\implies$ Penalty = $9 - 8 = 1$
• Col D3: 3, 4 $\implies$ Penalty = $4 - 3 = 1$

Ties exist between Row S1 and Col D1 (Penalty = 2). In Row S1, the cheapest cell is (S1, D3) with cost $4.

• Allocate $x_{13} = \min(s_1 = 30, d_3 = 45) = 30$.
• S1 is exhausted (capacity = 0). D3 remaining demand = $45 - 30 = 15$.
• Cross out Row S1.

Allocate Remaining Demands to surviving Row S2 (capacity = 40):

• Allocate to D1: $x_{21} = 20$ (cost $4, fully satisfies D1).
• Allocate to D2: $x_{22} = 5$ (cost $9, fully satisfies D2).
• Allocate to D3: $x_{23} = 15$ (cost $3, fully satisfies D3).
• Total allocated from S2 = $20 + 5 + 15 = 40$. All constraints met!

VAM Initial Solution Allocation:

• $(S1, D3) = 30$ ($c_{13} = \text{\textdollar}4$)
• $(S2, D1) = 20$ ($c_{21} = \text{\textdollar}4$)
• $(S2, D2) = 5$ ($c_{22} = \text{\textdollar}9$)
• $(S2, D3) = 15$ ($c_{23} = \text{\textdollar}3$)
• $(S3, D2) = 30$ ($c_{32} = \text{\textdollar}5$)

Total VAM Shipping Cost: $Z_{VAM} = (30 \times 4) + (20 \times 4) + (5 \times 9) + (15 \times 3) + (30 \times 5)$ $Z_{VAM} = 120 + 80 + 45 + 45 + 150 = \text{\textdollar}440$

Step 2: Optimality Testing Using the MODI ($u-v$) Method

Number of basic cells = 5 ($m + n - 1 = 5$, non-degenerate). Using basic cells ($u_i + v_j = c_{ij}$), set $u_2 = 0$ (Row 2 has three basic cells):

• From cell $(S2, D1)$: $u_2 + v_1 = c_{21} \implies 0 + v_1 = 4 \implies v_1 = 4$
• From cell $(S2, D2)$: $u_2 + v_2 = c_{22} \implies 0 + v_2 = 9 \implies v_2 = 9$
• From cell $(S2, D3)$: $u_2 + v_3 = c_{23} \implies 0 + v_3 = 3 \implies v_3 = 3$
• From cell $(S1, D3)$: $u_1 + v_3 = c_{13} \implies u_1 + 3 = 4 \implies u_1 = 1$
• From cell $(S3, D2)$: $u_3 + v_2 = c_{32} \implies u_3 + 9 = 5 \implies u_3 = -4$

Compute Opportunity Costs $\Delta_{ij} = c_{ij} - (u_i + v_j)$ for all Non-Basic Cells:

• Cell $(S1, D1)$: $\Delta_{11} = 6 - (1 + 4) = 6 - 5 = +1 \ge 0$
• Cell $(S1, D2)$: $\Delta_{12} = 8 - (1 + 9) = 8 - 10 = -2$ (Negative! Not Optimal)
• Cell $(S3, D1)$: $\Delta_{31} = 7 - (-4 + 4) = 7 - 0 = +7 \ge 0$
• Cell $(S3, D3)$: $\Delta_{33} = 8 - (-4 + 3) = 8 - (-1) = +9 \ge 0$

Because $\Delta_{12} = -2 < 0$, shipping via $(S1, D2)$ reduces total system cost by $2 for every unit shifted.

Step 3: Stepping-Stone Pivot Loop

Trace the closed rectangular loop through basic cells starting at entering cell $(S1, D2)$: $\text{Loop: } (S1, D2)[+] \longrightarrow (S1, D3)[-] \longrightarrow (S2, D3)[+] \longrightarrow (S2, D2)[-] \longrightarrow (S1, D2)$

Evaluate donor cells with negative signs:

• Cell $(S1, D3)$ has allocation 30.
• Cell $(S2, D2)$ has allocation 5.
• Maximum feasible pivot flow: $\theta = \min(30, 5) = 5$ units.

Execute Shift by $\theta = 5$:

• $x_{12} = 0 + 5 = 5$ (Enters basis)
• $x_{13} = 30 - 5 = 25$
• $x_{23} = 15 + 5 = 20$
• $x_{22} = 5 - 5 = 0$ (Leaves basis)
• $x_{21} = 20$ (Unchanged)
• $x_{32} = 30$ (Unchanged)

New Total Optimal Shipping Cost: $Z_{optimal} = (5 \times 8) + (25 \times 4) + (20 \times 4) + (20 \times 3) + (30 \times 5)$ $Z_{optimal} = 40 + 100 + 80 + 60 + 150 = \text{\textdollar}430$

Recomputing MODI multipliers on the new basis yields all $\Delta_{ij} \ge 0$ (specifically: $\Delta_{11} = +1, \Delta_{22} = +2, \Delta_{31} = +5, \Delta_{33} = +7$), proving that $430 is the global minimum.