Minimax Strategy

Prime #

995

Origin domain

Mathematics And Formal Systems

Subdomain

game theory and decision theory → Mathematics And Formal Systems

Aliases

Minimax, Maximin, Minimax Theorem

Core Idea

Minimax is a decision rule for adversarial or worst-case conditions: select the action whose worst possible outcome is the best worst possible outcome — minimize the maximum loss the environment can inflict. The structural commitment is the substitution of an adversary's selection step for a probability distribution: where expected-value reasoning integrates over a distribution, minimax takes a supremum over a feasible set.

How would you explain it like I'm…

Best Of The Worst

When you play against someone trying to beat you, look at the worst thing that could happen with each choice, and pick the choice whose worst thing is the least bad. You're not hoping for luck — you're getting ready for the meanest possible move against you. That way, even if things go badly, you've made them as not-bad as they can be.

Plan For The Meanest Move

Minimax is a way to choose when you're up against an opponent or a worst case. For each option you ask: what's the worst that could happen if I pick this? Then you choose the option whose worst outcome is the *least* bad of all the worst outcomes. It treats the world like a clever rival who will pick whatever hurts you most, so you guard against that. This is different from just hoping for the average or the best case — minimax cares most about not getting wrecked when things go wrong.

Minimize The Maximum Loss

Minimax is a decision rule for choosing under adversarial or worst-case conditions: pick the action whose worst possible outcome is the best of all the worst outcomes. Put another way, you minimize the maximum loss the environment can inflict — or, in payoff terms, maximize the minimum gain. It treats the environment as if it were a rational opponent deliberately selecting against you, and asks which option performs least badly against that worst response. This contrasts with expected-value reasoning, which averages over how likely each outcome is; minimax ignores the averages and stares only at the worst case. It's the right tool when a catastrophic outcome is so bad you can't afford to gamble on probabilities, and it shows up in games, in robust engineering, and in analyzing the slowest input an algorithm could face.

 

Minimax is a decision rule for choosing under adversarial or worst-case-relevant conditions: select the action whose worst possible outcome is the best worst possible outcome. Equivalently, minimize the maximum loss the environment can inflict given your choice — or, in payoff form, maximize the minimum gain. It treats the environment as if it were a rational adversary selecting against you and asks which option performs least badly against that worst response. In the two-player zero-sum game where it was first formalized, every player has a minimax value, and in finite matrix games with mixed strategies the value is unique and maximin and minimax coincide. Outside zero-sum games the structure generalizes in three directions sharing one skeleton: robust optimization chooses parameters performing well under the worst realization of an uncertainty set; worst-case algorithm analysis characterizes a procedure by its slowest input rather than its average; and risk-bounded decision accepts lower expected value for a smaller worst case when the catastrophic outcome is asymmetric. The structural commitment is the substitution of the adversary's selection step for a probability distribution: where expected-value reasoning integrates over an assumed distribution, minimax takes a supremum over a feasible set. That choice is itself the design lever — the two modes differ in how much they trust the distributional assumption, how heavily they weight catastrophe, and how far they presume the environment selects against the chooser. The dual fact is the minimax inequality — max-min is at most min-max, with equality at a saddle point under specific convexity and compactness conditions — and the size of any gap measures how much the player forced to commit first loses to the one who responds optimally.

Broad Use

• Game theory: the minimax value is the central solution concept for zero-sum games and a benchmark for non-zero-sum ones.
• Robust control: H-infinity control minimizes the worst-case effect of disturbances; structures are sized to worst-case load envelopes.
• Computer science: worst-case complexity is minimax over inputs; game-tree search uses the minimax algorithm.
• Statistics and ML: minimax estimators minimize worst-case risk; adversarial training and GANs are explicitly minimax.
• Policy: maximin rules are a principled response to catastrophic risk under deep uncertainty.
• Negotiation: a party's BATNA is a minimax value, the floor it can guarantee outside the deal.

Clarity

Naming minimax separates two reasoning modes that look alike: expected-value integrates over a distribution, minimax takes the supremum over a feasible adversary set — and makes the chooser's real question precise: what set are we taking the supremum over?

Manages Complexity

A sprawling, substrate-specific deliberation collapses to the specification of one adversary set and one supremum over it, plus an accounting of the insurance premium paid versus expected value.

Abstract Reasoning

Minimax is a quantifier-alternation move (min over my actions, max over their responses); the order of commitment is a design lever, and the minimax inequality's saddle-point gap prices what the first-mover loses.

Knowledge Transfer

• Engineering: enumerate the adversary set (load combinations, attack vectors) and minimax-optimize.
• Learning: add minimax structure (adversarial training, distributionally robust optimization) to produce models that degrade gracefully.
• Negotiation: a party's minimax value over the no-deal set is the floor below which any deal is rejected.

Example

A bridge engineer constructs an adversary set — the worst credible combination of dead, live, wind, seismic, and thermal loads — takes the supremum over it, and sizes members so stresses stay within limits with a safety factor. The bridge costs more than an expected-value design; that extra cost is the explicit insurance premium.

Relationships to Other Primes

Parents (1) — more general patterns this builds on

• Minimax Strategy is a kind of Optimization — Minimax is the SPECIFIC quantifier-alternation specialization of optimization — optimize over actions against a SUPREMUM over an adversary set (a sup-over-set rule), distinct from optimization in general. The file makes optimization/multiobjective_optimization the genus it is not identical to.

Path to root: Minimax Strategy → Optimization

Not to Be Confused With

• Minimax Strategy is not Optionality because optionality preserves upside by keeping favorable choices open, whereas minimax bounds downside by optimizing the worst case.
• Minimax Strategy is not Antifragility because antifragility benefits from volatility and stressors, whereas minimax merely limits worst-case loss without gaining from disorder.
• Minimax Strategy is not Expected Utility because expected utility integrates over a probability distribution, whereas minimax takes a supremum over a feasible set, refusing to trust the distribution.