Equity Calculation in Pot-Limit Omaha

A reference treatment of the equity-calculation problem in Pot-Limit Omaha, including the combinatorial difference from No-Limit Hold'em and the methods used inside automated agents.

The equity problem

Hand equity is the probability, given known and unknown cards, that a hand will win or tie at showdown against a specified opponent range. In a complete-information setting — both hands face-up, board incomplete — equity is computed by enumerating all completions of the deal. Against a range, the equity is the expectation of the per-combination equities over that range.

The equity-calculation problem in PLO is computationally harder than in No-Limit Hold'em primarily because the number of relevant two-card combinations from four hole-cards is six per player, against one in NLH. A heads-up made-versus-made evaluation that requires one comparison in NLH requires up to thirty-six in PLO before any board cards are dealt.

Combinatorial structure

Each PLO hand contains C(4,2) = 6 two-card combinations. A heads-up matchup therefore involves up to 6 × 6 = 36 starting pairings, each of which must be evaluated against every legal completion of the five community cards. For a pre-flop equity computation the remaining-deck draw space is C(44,5) = 1,086,008 boards, giving an upper bound of roughly 3.9 × 10⁷ board-pairing combinations per matchup. Range-versus-range computations multiply this by the number of combos in each range.

Because each player must use exactly two hole-cards and three board cards, equity calculators cannot reuse a No-Limit Hold'em "best five of seven" evaluator. The constrained selection is part of the rules of the variant and must be enforced in the inner evaluation loop.

Exhaustive enumeration

Exhaustive enumeration produces an exact equity figure by visiting every board completion. For heads-up pre-flop calculations the work is tractable on modern hardware in the sub-second range when the evaluator is implemented with bitmask hand representations and a precomputed rank-lookup table. On the flop and turn the remaining-board space collapses to C(45,2) = 990 and 44 respectively, and enumeration becomes the standard method.

Monte Carlo sampling

For multi-way pots, deep ranges, or stacked range-versus-range queries, exhaustive enumeration is impractical. Monte Carlo sampling estimates equity by drawing N independent board completions and averaging the outcome. The standard error of the estimate scales as 1/√N, so an equity figure accurate to 0.5 percentage points typically requires on the order of 10⁴ samples. PLO bots that operate at the table use Monte Carlo with early termination when the running confidence interval clears a decision threshold.

Implementation notes

Bot implementations typically separate offline solver output (range-equity matrices, computed against canonical bucketed ranges) from online equity calls (per-decision Monte Carlo against the inferred opponent range). The OMPEval library and its successors are the reference open-source evaluators; commercial agents implement the same algorithms with SIMD-optimised inner loops. Equity is rarely the sole input to a decision: it is combined with positional information, blocker effects, and the bet-sizing constraint discussed in the companion reference.