§ Overview

Two layers of decision-making, jointly.

The Capacitated Location Routing Problem (CLRP) is an optimization problem that combines two layers of decision-making: strategic decisions about where to open depots and operational decisions about how to route vehicles from those depots to serve customers.

Given a set of potential depot locations (each with an opening cost, a capacity limit, and a maximum number of vehicles) and a set of customers with known demands, the goal is to simultaneously decide which depots to open, assign customers to open depots, and design vehicle routes that minimize the total cost.

§ Three decisions

Open depots.
Assign customers.
Design routes.

These three decisions are tightly coupled. Solving them independently is feasible but far from optimal — the central challenge is exploiting their interaction.

DECISION 01

Depot opening

Which depots to open from a set of potential locations. Each depot has an opening cost fi, a capacity Wi (maximum total demand), and a limit on the number of vehicles it can accommodate.

Strategic
DECISION 02

Customer assignment

Every customer must be assigned to exactly one open depot and visited by exactly one route from that depot. No customer can be split across depots or routes.

Tactical
DECISION 03

Vehicle routing

For each open depot, design routes for a fleet of identical vehicles. Each route starts and ends at its depot; the total demand on a route cannot exceed the vehicle capacity Q.

Operational
§ Worked example

30 clients.
5 candidates.
Open 3.

A miniature CLRP instance — five depot candidates, three chosen. Every client is assigned to exactly one open depot; each open depot operates a small set of routes that respect vehicle capacity. Closed candidates are crossed out; each open depot and its routes share a color.

FIG 02 · n = 30 clients · |D| = 5 candidates · 3 opened · vehicle capacity Q = 65

§ Objective

Minimize total cost across three components.

A three-part objective captures the trade-off between opening fewer depots (cheaper strategically) and routing efficiency (which may require depots closer to customers).

# objective — minimize total cost
min   Σi∈D fi · yi    // depot opening costs
   + Σr∈R v · zr      // fixed cost per vehicle dispatched
   + Σ(i,j)∈E cij · xij   // total travel distance
yi
Binary — 1 if depot i is open, 0 otherwise
zr
Binary — 1 if route r is used, 0 otherwise
xij
Binary — 1 if edge (i, j) is traversed on some route, 0 otherwise
fi, v, cij
Depot opening cost, fixed vehicle cost, travel cost for edge (i, j)
§ Constraints

A feasible solution must satisfy…

Five constraint families define feasibility. Capacity tightness is a tuning dimension of the instance set — some instances will leave little slack.

ConstraintDescription
Customer serviceEvery customer is visited exactly once by exactly one route.
Depot assignmentEach route operates from exactly one open depot. All customers on a route are assigned to that depot.
Vehicle capacityThe total demand of customers on any single route does not exceed the vehicle capacity Q.
Depot capacityThe total demand assigned to a depot does not exceed its capacity Wi.
Vehicle limitThe number of routes operating from a depot does not exceed its maximum number of vehicles mi.
§ Why this problem

Real relevance.
Room to innovate.

The CLRP is directly relevant to Mexican industry. Companies such as OXXO, Bimbo, and FEMSA routinely face decisions about where to place distribution centers while simultaneously optimizing delivery routes. This problem captures the essence of those real-world decisions.

From an algorithmic perspective, the CLRP sits at a difficulty sweet spot: it combines strategic facility location with operational routing, so no single algorithmic paradigm dominates. The competition is open to metaheuristics, decomposition methods, branch-and-price, matheuristics, machine learning, and hybrid approaches.

Compared to the well-studied CVRP, there is significantly less benchmark saturation for the LRP. This challenge fills a genuine gap — any results carry higher novelty for the community.

§ Instance design

30 instances,
three scales, zero overlap.

Instances are generated by an open-source Python generator released in June 2026. Variation across dimensions prevents over-specialization.

customers
200 – 3,000
depots
10 – 50
distance
Euclidean symmetric (small / medium scales) and asymmetric (large scale)
distribution
Uniform, clustered, or mixed customer-placement patterns
capacity
Multiple tightness levels — from loose to near-infeasible
demand
Multiple distributions, controlling variance and skew
NOTE

Full technical specification — coming June 2026

The mathematical formulation, instance file format, solution file format, and the solution verifier will be published alongside the instance generator in June 2026. A set of mock instances with feasible solutions will also be provided at that time.