Rio Tinto solved decades-long open-pit mine plans with large neighborhood search

Open-pit mine planning over decades of extraction produces enormous, complex optimization models: pit sequencing, blending, equipment, and ore grades all interact. Off-the-shelf solvers either time out or exhaust memory, so planners fell back on few-scenario manual analysis.

The team's large neighborhood search solves the problem the other way: a sliding-window heuristic produces an initial feasible plan, then repeated restricted mixed-integer solves free only a subset of variables per iteration. Neighborhood formation strategies exploit the model's structure to choose which variables to free.

Running live at Rio Tinto, the approach solves large long-term planning models that commercial solvers could not handle, cutting solve times dramatically and enabling planners to test multiple scenarios — generating millions of dollars in value insights.

Fix most variables; re-optimize a neighborhood each time

Fixing most variables makes each subproblem small enough to solve exactly, while iterating over different neighborhoods escapes the local traps a single greedy pass would hit.

Structure-aware neighborhoods (blocks, benches, time periods) free the variables that actually interact, so the iteration converges fast.

Scenario exploration is the real product: planners can now ask 'what if' across many futures instead of rationing three runs.

When a model is too big to solve, don't shrink the question — fix most of the answer and re-solve a moving neighborhood of decisions until it converges.

Published in the INFORMS Journal on Applied Analytics in 2026. Rio Tinto uses the framework for long-term mine planning, with the paper reporting 'millions of dollars in value insights' generated by the faster scenario exploration.