In 1921 Émile Borel published a short note on a problem of simultaneous allocation. It is remembered for two reasons. It was the first paper in the game theory literature to distinguish pure from mixed strategies, and it posed a puzzle that nearly thirty years later two RAND analysts, Oliver Gross and Robert Wagner, would name after a fictional officer: Colonel Blotto. Their 1950 memorandum, A Continuous Colonel Blotto Game (RAND RM-408), gave the problem its name and its first systematic solutions.
So: what is the Colonel Blotto game? Two players each distribute a fixed budget across several battlefields without seeing the other’s choice. Each battlefield is won by whoever commits strictly more resource to it, ties go to neither side, and the player who takes a majority of the battlefields wins the round. That is the whole specification. It contains no payoff asymmetry, no private information and no sequencing, and yet it has a property that makes it one of the most instructive objects in game theory: no pure strategy can be optimal. Randomisation here is not a hedge against ignorance about the opponent. It is a strictly necessary component of optimal play, and the proof fits in four lines.
The rest of this article gives that proof, states the continuous symmetric solution for three battlefields, documents the history, treats the asymmetric-budget case that makes the model practically interesting, and reports honestly on the applications.
Why no pure strategy can be optimal
Fix the standard case: three battlefields, a budget of units, integer or real allocations, majority of fields wins.
Suppose player 1 commits to a fixed allocation with . Label the fields so that is the largest. Then player 2 can respond:
- Concede field 1 entirely — allocate nothing there, surrendering it.
- Allocate to field 2 and to field 3, winning both strictly.
- The cost of that response is .
- Since is the largest of three parts summing to , we have , so the cost is at most . The counter is always affordable.
Player 2 wins two fields out of three, hence the round, against any fixed allocation. The argument never used a special feature of the allocation, so no allocation escapes it. The conclusion follows: in equilibrium both players must randomise, because any deterministic plan is exploitable with certainty.
Worked counters make the mechanism concrete:
| Player 1’s allocation | Player 2’s counter | Fields won by 2 |
|---|---|---|
| 2 and 3 | ||
| 2 and 3 | ||
| 1 and 3 | ||
| 2 and 3 | ||
| 2 and 3 |
Notice that the counter always gives up a field. That is the structural point: in a majority contest, resources spent on a field you lose anyway are wasted, so the profitable response to a spread is a concentration, and the profitable response to a concentration is a different concentration. There is no resting place. This is the same logic that forces mixing in a mixed-strategy Nash equilibrium, but Blotto states it with unusual starkness, because the exploiting response is explicit rather than merely existent.
Use the panel below to feel the force of it. Enter any allocation you like across the three fields — the remaining-budget readout will hold you to ten units — then set the opponent-strategy selector to best-counter mode and press the hundred-round button. Best-counter mode constructs precisely the response described above to your previous allocation, and the win-rate readout will show you losing essentially every round. That is the four-line argument made brutal. Any single plan, however clever, is a plan the opponent can answer.
The continuous symmetric solution
If pure strategies are all beatable, what does an optimal mixed strategy look like? For the continuous version of the game — real-valued allocations, three battlefields of equal value, equal budgets — the answer has a memorably simple description at the level of each field.
Each player draws a random allocation whose marginal distribution on each individual battlefield is uniform on . With that interval is . The uniform marginal has mean , so each field receives one third of the budget in expectation, as it must.
Why uniform? Consider what a marginal distribution does for you. Let be the opponent’s marginal cumulative distribution function on a given field. If you place there, you win that field with probability . For you to be indifferent between all the ways of shifting a marginal unit between fields — which is exactly what an interior equilibrium requires — the marginal return must be the same constant at every in the support. A constant density is a uniform distribution. The upper endpoint is then forced by the budget: a uniform marginal on has mean , and three fields must average each, so and .
Against such an opponent, placing on a field wins it with probability , which averages to one half over the opponent’s own distribution. By symmetry the round is won with probability one half. That is the equilibrium value, and it is what makes the strategy safe: nothing you do raises your win rate above one half.
Two pieces of honesty are required here. First, uniform marginals do not pin down the strategy. Roberson’s analysis establishes that uniform univariate marginals are necessary under this winner-takes-the-field rule, but a strategy is a joint distribution over triples summing to , and many different joint distributions share the same uniform marginals. The equilibrium is therefore not unique as a strategy, only as a set of per-field marginals and as a payoff. Constructing an explicit joint distribution with the right marginals and an exactly binding budget is the technically hard part of the problem; Roberson’s 2006 treatment does it with copula theory, and that is precisely a tool for building joint distributions with prescribed marginals. Second, sources differ on exactly how much of this was established when. Gross and Wagner’s 1950 memorandum is credited with the symmetric case for any finite number of fields, while the three-field symmetric case is sometimes attributed to Borel and Ville in 1938; summaries disagree on the scope of each contribution, and the safest statement is that the symmetric result is classical and the general asymmetric result is Roberson’s.
Set the opponent selector in the panel below to the equilibrium strategy — which draws real-valued allocations with uniform marginals on — and play a hundred rounds with whatever allocation you prefer. The win rate will hover near one half: perhaps , perhaps , drifting with the sampling noise but refusing to climb. Change your allocation and play another hundred; the number does not move in any systematic way. That is what an equilibrium opponent guarantees, and it is a more convincing demonstration than any algebra. Then try a degenerate allocation such as and watch the win rate collapse to roughly one in six: by conceding two fields outright you need to win both of them from zero, which happens only when the opponent draws zero there, an event of probability zero in the continuous model and vanishingly rare in the implementation. Concentration is only valuable when it is randomised concentration.
What the equilibrium is not
It is worth separating three claims that are easy to conflate.
The continuous game, with real-valued allocations, is the object described above. The discrete-budget game, where allocations must be whole units out of ten, is a different and harder object: it is a finite matrix game, its optimal strategies are solutions of a linear program rather than smooth densities, and its value and support depend on the arithmetic of the budget in ways that have no clean closed form. The widget below implements the continuous equilibrium, drawing real-valued allocations; it is not a solution of the integer game. Borel himself is sometimes credited with showing that small discrete versions behave differently from the continuous one, which is a reason to keep the two apart rather than a footnote.
The third claim to avoid is that the equilibrium is “the optimal Blotto strategy” in the singular. There is a set of them. What is unique is the value — one half under symmetry — and the per-field marginals.
History, verified
The lineage is well documented and worth getting right.
Borel (1921) posed the problem and introduced, in the same note, the distinction between pure and mixed strategies. Borel and Ville (1938) gave an optimal strategy for the symmetric three-battlefield case. Gross and Wagner (1950), in RAND research memorandum RM-408, A Continuous Colonel Blotto Game, named the game and extended the symmetric analysis; the memorandum frames it as Colonel Blotto committing units of force against an enemy’s units across several hills of differing value. Friedman (1958) supplied a partial characterisation for many battlefields with asymmetric resources. Roberson (2006), “The Colonel Blotto Game”, Economic Theory 29, pages 1–24, gave the modern general treatment: equilibrium characterisation and uniqueness of equilibrium payoffs for battlefields and arbitrary asymmetric budgets, using copulas.
On the experimental side, Ayala Arad and Ariel Rubinstein ran a large web-based tournament version of the game and reported the results in “Multi-dimensional iterative reasoning in action: the case of the Colonel Blotto game”, Journal of Economic Behavior & Organization 84(2), 2012, pages 571–585. Their finding is that people do not play the equilibrium. They choose the features of an allocation one at a time — how many fields to contest, how aggressively — and the strongest tournament entries reflected about two to three levels of this iterated reasoning. The gap between the equilibrium and observed play is not noise; it is structured, and the structure is about how the strategy space is mentally decomposed.
Asymmetric budgets
The symmetric case is the clean one; the asymmetric case is the interesting one. Suppose player 1 has budget and player 2 has . Does the larger budget win?
Not automatically, and the reason is instructive. Take three fields with and , and suppose the stronger player spreads evenly: . The weaker player responds and wins two fields out of three with certainty, despite being outspent by twenty per cent. Generalising, an even spread of per field is beaten for sure whenever , that is whenever . The stronger player must therefore randomise too; superiority of resources does not buy the luxury of a plan.
The weaker player’s equilibrium behaviour has a characteristic shape: concentrate, and concentrate randomly. Place nothing on some fields and a serious commitment on the others, with the identity of the contested fields drawn at random. Formally, the weaker player’s equilibrium marginal distribution puts an atom at zero — a positive probability of abandoning a given field entirely — and spreads the remainder over higher allocations than an even split would allow. The intuition is direct: with fewer resources you cannot be competitive everywhere, so buying a real chance on a subset beats buying a negligible chance on all of them. Roberson’s 2006 paper characterises the equilibria for general asymmetric budgets and confirms this qualitative picture, with the strong player’s advantage in equilibrium payoff growing with the budget ratio and the number of fields.
There is a threshold flavour to the result that matters for applications. When the budget ratio is large enough relative to the number of battlefields, the weak player’s chances approach zero, because even full concentration on a bare majority of fields cannot outbid the strong player’s per-field commitment. Between the symmetric case and that threshold lies the whole interesting region, in which a smaller force wins with substantial probability by choosing where to fight.
Applications
The model is used in four main places, and each use requires a judgement about how well the abstraction fits.
Political campaigns. The clearest fit is a system where units of competition are winner-take-all, as with electoral college states or single-member constituencies: resources committed to a state that is lost anyway are wasted exactly as in Blotto. The modelling literature on campaign allocation under the electoral college is substantial, and recent work applying a Blotto-style framework to the 2008 and 2020 United States presidential elections concludes that the best response to the parties’ actual allocations was to concentrate on a subset of states and write off others — a recommendation that departs both from standard theory and from what the campaigns did. That is an interesting finding and also a warning: the result is a statement about a model fitted to those elections, not a demonstrated fact about how to win one. Mapping the clean model onto a real election requires assuming that spending translates into votes field by field, that the two sides’ budgets are fixed and known, that commitments are simultaneous and irreversible, and that fields are of known and separable value. All four are strong. Related questions about how votes aggregate into outcomes are taken up in our article on the mathematics of voting.
Advertising budgets across markets. Where market share in each region is effectively awarded to whoever outspends, the same structure applies. The assumption doing the work is the sharpness of the contest rule; where spending buys a smooth share rather than the whole market, Blotto is the wrong model and a smooth contest function is the right one.
Research and development portfolios. Several firms race for several patents or standards, each prize going to whoever invests most. Here the simultaneity is realistic and the winner-take-all rule is close to literal for patents, which makes it one of the better-fitting applications.
Military deployment. The original motivation, and the one that gave the game its colonel. It is also the application where the model’s silence about information, terrain and sequencing is most obviously costly.
A fifth use is methodological. Blotto is a standard testbed for experimental and computational work on how people and algorithms handle large, unordered strategy spaces — the Arad–Rubinstein tournament is the best-known instance — and for problems of network attack and defence, where the fields are nodes.
Why the game matters mathematically
Most introductions to randomised strategies motivate them defensively: you mix because you do not want to be predictable, or because you are uncertain what the opponent will do. Colonel Blotto removes the hedging interpretation entirely. The opponent in the proof above is not uncertain and not exploiting your ignorance; the argument shows that every deterministic plan loses with certainty to a specific, cheap, explicitly constructed response. Randomisation is therefore not insurance. It is a component of optimal play, in the same sense that a particular number is a component of the answer to an arithmetic problem.
That is the content of von Neumann’s minimax theorem made vivid: a zero-sum game has a value, and attaining it may require the space of strategies to be enlarged from the pure allocations to probability distributions over them. Blotto is the clearest case in the literature where the enlargement is not a technical convenience but the substance of the solution. It also shows how quickly a transparently specified game can become mathematically hard — the three-field symmetric case is a classroom exercise, the general asymmetric case needed copula theory and waited until 2006, and the discrete-budget version remains awkward.
For the broader setting, see our overview of game theory and strategic decisions. For contrast, the prisoner’s dilemma has a pure equilibrium and the problem is that it is bad for everyone; Blotto has no pure equilibrium at all and the problem is that you cannot have a plan. The finite version is solved by the methods of linear programming, and the stability of the hundred-round win rates in the panel below is simply the law of large numbers doing its work on a sequence of fair coin flips.
Related topics: Minimax Theorem, Mixed-Strategy Nash Equilibrium, Prisoner’s Dilemma, Linear Programming, Law of Large Numbers, Mathematics of Voting, Second-Price Auction, Combinatorial Game Theory, Game Theory and Strategic Decisions
Frequently asked
What is the Colonel Blotto game?
A two-player zero-sum game in which each player simultaneously divides a fixed budget across several battlefields without seeing the opponent's choice. Each battlefield goes to whoever commits strictly more there, and the player who takes a majority of battlefields wins the round.
Why is there no pure strategy equilibrium in Blotto?
Because every fixed allocation can be beaten. An opponent who sees it can concede your strongest field entirely, freeing up resources to outspend you by one unit on each of the other two. With three fields and a budget of ten, that counter is always affordable, so no single allocation is safe.
What is the optimal Blotto strategy?
A randomised one. In the continuous three-field game with equal budgets, each player draws a real-valued allocation whose marginal distribution on each field is uniform on the interval from zero to two thirds of the budget. Against such an opponent every allocation wins with probability one half.
How is Blotto used in politics?
As a model of campaign resource allocation across constituencies or electoral college states, where each unit is winner-take-all. It explains why campaigns randomise and concentrate rather than spreading evenly, though mapping the clean model onto a real election requires strong assumptions.
What is a resource allocation game?
A game in which players divide a fixed endowment across several simultaneous contests. Blotto is the canonical example. The family also covers advertising budgets across markets, research portfolios, attack and defence of networks, and anything else where commitments are made once and compared contest by contest.