Skip to main content

Lot Sizing Optimization

Wagner-Whitin / Silver-Meal — Dynamic Programming

A production planner must decide in which periods to produce and how many units per batch, trading off fixed setup costs ($2,500 per run) against holding costs ($45/unit/period). Over a 52-week horizon, poor lot sizing wastes 8–15% of total production cost.

Where This Decision Fits

Manufacturing planning chain — the highlighted step is what this page optimizes

Plant Location Facility siting
Production Planning Lot sizing & scheduling
Shop Floor Machine scheduling
Assembly Line balancing
Supply Chain Distribution & logistics

The Problem

Minimizing total setup and holding cost over a finite horizon

Consider a single product over a 10-period planning horizon. Each period t has a known demand dt that must be satisfied (no backlogging). The manufacturer incurs a fixed setup cost K = $2,500 each time a production run is initiated, plus a holding cost h = $45 per unit per period for any inventory carried from one period to the next.

The decision is: in which periods should we produce, and how much? Producing every period avoids holding cost but incurs 10 setup costs ($25,000). Producing everything in period 1 requires only one setup ($2,500) but maximizes holding cost. The optimal policy balances these two opposing forces.

This is the classic uncapacitated lot-sizing problem, solvable exactly by the Wagner-Whitin algorithm (1958) using dynamic programming in O(n²) time, exploiting the zero-inventory ordering property: an optimal policy only produces when starting inventory is zero.

Wagner-Whitin DP Formulation minimize   Σt K · yt + h · It   // total cost = setup costs + holding costs
subject to
  It = It-1 + xt − dt   // inventory balance
  xt ≤ M · yt   // production only if setup (yt ∈ {0,1})
  It ≥ 0   // no backlogging
  I0 = 0   // initial inventory

// DP recurrence: f(t) = min over j≤t { f(j-1) + K + h · Σk=j..t-1 (t-k) · dk }

Where yt is the binary setup indicator, xt is production quantity, It is ending inventory, and dt is the demand in period t.

Try It Yourself

Optimize a 10-period lot sizing schedule

Lot Sizing Optimizer

K=$2,500 · h=$45/unit/period
Uniform demand of 100 units per period. A classic textbook scenario where the trade-off between setup frequency and holding cost is most transparent. EOQ-like patterns emerge.

Ready. Click “Solve & Compare All Algorithms” to run.

Algorithm Setups Setup Cost Holding Cost Total
Click Solve & Compare All Algorithms
Period schedule will appear here after solving.

The Algorithms

Three approaches to dynamic lot sizing

Exact

Wagner-Whitin (Dynamic Programming)

O(n²)  |  Guarantees global optimum

The Wagner-Whitin algorithm exploits the zero-inventory ordering property: in an optimal solution, production only occurs when beginning inventory is zero. This reduces the search space from exponential to O(n²). For each period t, it considers all possible last-production periods j ≤ t and picks the one minimizing cumulative cost. The DP recurrence builds the optimal solution forward, and backtracking recovers the production schedule. This is the gold standard for uncapacitated single-item lot sizing.

Heuristic

Silver-Meal

O(n)  |  Greedy cost-per-period minimizer

The Silver-Meal heuristic adds periods to the current production lot as long as the average cost per period (setup cost amortized plus accumulated holding cost) is decreasing. Once adding the next period would increase the average, the lot is closed and a new lot begins. This forward-looking greedy strategy typically achieves within 1–3% of optimal for smooth demand patterns but can deviate more significantly with lumpy or highly variable demand.

Baseline

Lot-for-Lot

O(n)  |  Zero inventory, maximum setups

The simplest policy: produce exactly the demand for each period, every period. This eliminates all holding cost but incurs a setup cost in every period with nonzero demand. Lot-for-Lot serves as an upper bound on setup cost and a lower bound on holding cost, making it a useful baseline for comparison. It is optimal only when holding cost is so high that carrying any inventory is more expensive than an additional setup.

Real-World Complexity

Why production lot sizing goes beyond a simple DP

Beyond the Basic Model

  • Capacity constraints — Production capacity limits per period make the problem NP-hard; requires Lagrangian relaxation or MIP solvers
  • Multiple products — Shared setups, joint replenishment, and resource contention across product families add combinatorial complexity
  • Stochastic demand — Uncertain demand requires safety stock calculations and robust or stochastic programming formulations
  • Setup carry-over — If the same product was produced last period, setup may be avoided (sequence-dependent setups)
  • Quantity discounts — Volume-based pricing from suppliers creates nonlinear cost structures that break DP optimality
  • Perishability — Shelf-life constraints limit how far ahead production can cover future demand
  • Multi-level BOM — Component dependencies in multi-echelon production (MRP) couple lot-sizing decisions across levels

Key References

Foundational works in lot sizing optimization

  • Wagner, H. M. & Whitin, T. M. (1958). “Dynamic version of the economic lot size model.” Management Science, 5(1), 89–96.
  • Silver, E. A. & Meal, H. C. (1973). “A heuristic for selecting lot size quantities for the case of a deterministic time-varying demand rate and discrete opportunities for replenishment.” Production and Inventory Management, 14(2), 64–74.
  • Pochet, Y. & Wolsey, L. A. (2006). “Production Planning by Mixed Integer Programming.” Springer Series in Operations Research and Financial Engineering.

Need to optimize production planning?

From single-item lot sizing to multi-product, multi-level production scheduling, mathematical modeling can transform manufacturing operations. Let’s discuss how Operations Research can work for you.

Disclaimer
Data shown is illustrative. Demand profiles, cost parameters, and production figures are representative scenarios for educational purposes and do not reflect any specific manufacturer or product line.
Back
ESC