← Back to the searching page

Ternary Search & Unimodal Optimization

When a function rises then falls (or vice versa), two probe points per step shrink the search interval around the peak.

What to know

  • Requires strict unimodality — plateaus break the comparison logic.
  • Two midpoints m1, m2 discard the third of the range that cannot hold the optimum.
  • On integer domains, binary search on the slope is often simpler.

In the wild: Hyperparameter sweeps, physics trajectory optima, cost-curve minimization.

Algorithms to reach for

Ternary search

O(log n)

Maximize/minimize unimodal functions

Peak finding

O(log n)

Any local peak via slope binary search (LeetCode 162)

Practice problems — with full guides

Each problem has its own page: progressive hints, how to approach it, and the full solution.