Nash equilibrium & the math of GTO
GTO poker rests on one central idea from game theory — the Nash equilibrium — and one family of algorithms that makes it computable for a game as large as poker — regret minimization. You do not need the math to use a solver, but understanding it tells you exactly what a solution is and why you can trust it.
Nash equilibrium
A Nash equilibrium is a set of strategies, one per player, where no player can do better by changing only their own strategy. Everyone is already playing a best response to everyone else, so nobody has an incentive to move. The concept was introduced by mathematician John Nash in 1950 and is one of the foundational results in game theory.
In a two-player zero-sum game — which heads-up poker is, since one player's win is the other's loss — a Nash equilibrium has a powerful property: playing it guarantees you cannot lose in expectation, regardless of what your opponent does. This is the precise mathematical meaning of "unexploitable."
The formulas: bluff ratio and calling frequency
The three numbers players quote most come from the simplest river spot. One player bets with a polarized range of value hands and bluffs; the other holds bluff-catchers that beat only the bluffs. Write the pot as 1 and the bet as b, a fraction of the pot (a half-pot bet is b = 0.5).
- Caller's required equity = b / (1 + 2b). The caller risks b to win 1 + b, so a call breaks even when it wins that often.
- Bluff share of the betting range = b / (1 + 2b). At equilibrium the bettor bluffs exactly often enough to make the caller's bluff-catchers indifferent.
- Minimum defense frequency (MDF) = 1 / (1 + b). The caller has to continue with at least this share of their range, or any two cards could bluff at a profit.
| Bet size | b | Caller needs | Bluffs in the betting range | Minimum defense |
|---|---|---|---|---|
| ⅓ pot | 0.33 | 19.9% | 19.9% | 75.2% |
| ½ pot | 0.5 | 25% | 25% | 66.7% |
| ¾ pot | 0.75 | 30% | 30% | 57.1% |
| Pot | 1 | 33.3% | 33.3% | 50% |
| 2× pot | 2 | 40% | 40% | 33.3% |
Bigger bets need more bluffs to stay balanced and force the caller to fold more often. Real spots are messier: ranges are not perfectly polarized, and earlier streets still have betting to come. A solver therefore works these frequencies out hand by hand instead of from a formula, but the formula explains why its numbers look the way they do. How you choose the bet sizes a solve uses is covered in Bet sizing & tree configuration.
To see these frequencies in a real solution, open a solved flop in the free browser viewer and compare the bet sizes and continuing ranges node by node.
When your hand is losing: bluff-catch, fold or bluff
Most river hands are neither the nuts nor nothing. A bluff-catcher beats only the opponent's bluffs, and at equilibrium it is indifferent: calling and folding have the same EV. That is why solvers often mix such a hand between the two.
Which bluff-catchers call first is decided by card removal. Hands that block the opponent's value hands and leave their bluffs intact call; hands that block the bluffs fold. How solvers build strategies walks through blockers in detail.
When you are the one betting and your hand cannot win at showdown, it loses nothing by bluffing. That is why the bluffs in a balanced range come from the weakest hands, often missed draws that also block the opponent's calls, while hands with some showdown value prefer to check.
Against a real opponent who calls or folds too often, the indifference breaks and one option becomes clearly better. That is where exploitative play starts: see GTO vs exploitative play.
Bet size against a narrow or nut-heavy range
When your range holds the strongest hands and the opponent's range is capped (it has few or no hands that beat your value bets), equilibrium strategies lean towards large bets, sometimes overbets. The opponent's medium hands are forced to call or give up a lot, and by the bluff ratio above, the big size also lets you bluff more often.
When it is the opponent whose range is narrow and full of strong hands, the picture flips: big bets only get called by hands that beat you, so your betting range shrinks and gets smaller sizes or checks. When the two ranges are close in strength, small bets with many hands are common. How solvers build strategies covers range and nut advantage and capped versus polarized ranges.
Exploitability: measuring distance from GTO
Because a true equilibrium is often only approximated, we need a way to measure how close a strategy is. That measure is exploitability: the most an optimal opponent could win against your strategy, usually quoted in milli-big-blinds per hand (mbb/h) or as a percentage of the pot.
- Exploitability of zero = a perfect, unbeatable equilibrium.
- Low exploitability = very close to GTO; a tiny edge remains for a perfect counter.
- High exploitability = a strategy with real, attackable weaknesses.
A solver works by driving exploitability down with each iteration. When you stop a solve at a target accuracy, that target is the residual exploitability of the solution. (ART/GTO's default target and how to read it are covered in Running a solve.)
Regret minimization
The practical engine behind modern solvers is regret minimization. The idea is intuitive:
- Play the game against yourself for many iterations.
- At each decision point, track regret — how much better you would have done had you taken a different action.
- Shift future play toward the actions you regret not having taken more.
- Average your strategy over all iterations.
A remarkable theorem guarantees that this average strategy converges to a Nash equilibrium in two-player zero-sum games. You never solve the equilibrium directly — you let it emerge from self-play.
Counterfactual regret
Poker is too big to track regret over whole strategies, so the key innovation was counterfactual regret: measuring regret locally, at each individual decision point (information set), weighted by the probability of reaching it. Minimizing counterfactual regret at every node provably minimizes overall regret — and that made equilibrium approximation tractable for games with billions of decision points.
This is the algorithm called Counterfactual Regret Minimization (CFR), introduced in 2007. Every major poker solver since — including ART/GTO — descends from it. ART/GTO specifically uses Discounted CFR (DCFR), which down-weights early, noisy iterations so the solve converges faster. The algorithms page covers ART/GTO's implementation; the research GTO is founded on traces the full academic history.
The pieces, together
| Term | What it means |
|---|---|
| Nash equilibrium | Strategy set where no one can gain by changing alone — the definition of GTO |
| Zero-sum | One player's gain equals the other's loss; true of heads-up poker |
| Exploitability | How much a perfect opponent could win against you — distance from GTO |
| Regret | How much better an alternative action would have done in hindsight |
| Counterfactual regret | Regret measured locally at one decision point, weighted by reach probability |
| CFR | The algorithm that minimizes counterfactual regret to approximate equilibrium |
| DCFR | A faster, discounted variant of CFR — the algorithm ART/GTO uses |
Key takeaways
- A Nash equilibrium is the formal definition of GTO: no profitable unilateral deviation exists.
- In heads-up (zero-sum) poker, the equilibrium is unexploitable — it cannot lose in expectation.
- Exploitability measures how far a strategy is from GTO; solvers drive it toward zero.
- CFR approximates the equilibrium through self-play and counterfactual regret, which is what made poker solvable.
- ART/GTO uses Discounted CFR, a modern, faster member of the CFR family.
Next
- How solvers build strategies — how board texture, range shape and position decide what the equilibrium looks like.
- Free browser viewer for solved spots — open a solved flop and step through the bet sizes and ranges.