Stackelberg vs. Nash in the Lottery Colonel Blotto Game

The strongest honest case is a qualified yes: this paper admits an author-recognisable high-multiplicity mirror, although it is an extension mirror rather than a literal cloning of the paper’s two strategic players.

My lead anchor is Theorem 2, proved in this paper. It states that, for every possible follower-support set \(K\), the leader’s problem can be reduced in \(O(n)\) steps to maximizing a univariate continuous function over two half-open intervals, and that the resulting solution gives the leader’s optimal commitment. The paper presents this as yielding polynomial-time computation. Theorem 1, also proved here, is the structural result supporting that algorithm: after ordering battlefields by \(v_{aj}/v_{bj}\), only the \(n\) prefix supports \(K\in\{\{1\},\{1,2\},\ldots,[n]\}\) need be considered. By contrast, Lemma 1 is cited from Kovenock and Arjona (2019), so I would treat it as supporting prior machinery rather than as this paper’s computational anchor.

The natural continuous population is a large collection of local constituencies, advertising opportunities, or consumer segments. A type \(t\) is a complete local-contest description: the leader’s value \(v_a(t)\), the follower’s value \(v_b(t)\), and the common proportional contest rule. The society is a rational distribution \(\mu\in\Delta(T)\), where \(\mu_t\) is the fraction of local opportunities having type \(t\). The two strategic actors remain exactly the paper’s leader and follower; the population being continuized is the mass of local contests, not the number of strategic organizations.

Here is the precise problem I would propose.

HM-Lottery-Blotto-Stackelberg Commitment\(_\infty\) — given a finite type set \(T\), rational masses \(\mu_t\ge 0\) summing to \(1\), positive rational values \(v_a(t),v_b(t)\), and positive rational per-capita budgets \(B_a,B_b\), a strategy for player \(i\) is a nonnegative allocation density \(q_i:T\to\mathbb{R}_{\ge0}\) satisfying \(\sum_t\mu_tq_i(t)=B_i\). The utility is \(U_i(q_a,q_b)=\sum_t\mu_t v_i(t)\frac{q_i(t)}{q_a(t)+q_b(t)}\), with the paper’s convention that the follower receives the whole type-\(t\) value if both allocations are zero. The leader chooses \(q_a\) first; the follower chooses a best response \(q_b\). The output is an optimal commitment \(q_a^\star\), a follower best response \(q_b^\star\), and the leader utility \(U_a(q_a^\star,q_b^\star)\). An approximation version asks for a pair whose leader utility is within \(\varepsilon\) of the optimum.

This is not merely replacing a discrete budget by a fractional one: the paper already has continuous budgets. The new continuous object is \(\mu\), the population of repeated local contests. The correspondence with finite high multiplicity is concrete. If \(\mu_t=p_t/N\), make \(p_t\) identical battlefields of type \(t\), give each player total budget \(NB_i\), and allocate \(q_i(t)\) to every clone. The resulting utilities are exactly \(N U_i\). Conversely, at an optimum, Theorem 1’s equal-ratio clause forces identical clones to receive identical leader allocations, and the follower’s best response is then symmetric by the uniqueness used in Lemma 1. Thus the continuous problem preserves the optimal value, equilibrium response, and support structure after clearing denominators.

Writing \(x_{i,t}=\mu_tq_i(t)\) and \(w_{i,t}=\mu_tv_i(t)\) converts the problem into the paper’s own finite form: \(U_i=\sum_t w_{i,t}x_{i,t}/(x_{a,t}+x_{b,t})\), with \(\sum_tx_{i,t}=B_i\). The relative-value ordering is unchanged because \(w_{a,t}/w_{b,t}=v_a(t)/v_b(t)\). Consequently, Theorem 1 predicts only \(\tau=|T|\) possible active-support prefixes, and Theorem 2 supplies a one-dimensional optimization for each. I would therefore expect this mirror to be Class A: polynomial in \(\tau\) and the encoding length under the paper’s intended real-arithmetic model, and at least plausibly polynomial for an \(\varepsilon\)-version in the standard bit model.

The authors should recognise this as their problem. Their motivating applications explicitly include electoral competition, advertising, e-commerce, and cloud services. A large market with many repeated constituency or consumer types is a natural regime: the leader and follower still allocate budgets across valued battlefields, the outcome on each type is still proportional, and Stackelberg timing is unchanged. Nothing here replaces the game by fractional voting, a continuum of battlefields indexed by time, or a social-welfare objective.

The main further question suggested by the paper is a continuous Stackelberg–Nash Coincidence problem: given \((\mu,v_a,v_b,B_a,B_b)\), decide whether the optimal commitment and its follower response are also mutual best responses. Theorem 3 gives the relevant necessary-and-sufficient structure: the relative-value ratios must take at most two values, with the budget ratio satisfying the paper’s function \(f\). I would expect this type-compressed recognition problem to be tractable by grouping types by \(v_a(t)/v_b(t)\), although I would not use Theorem 3 as a separate computational anchor because the paper states it primarily as an equilibrium characterization. Corollary 2 likewise suggests studying the supremum of the leader’s Stackelberg advantage over continuous populations, but that is a quantitative follow-up rather than the lead result.

The weakest point is that the paper itself has only two strategic players. Treating its battlefields or constituencies as the continuous population is therefore an extension of the model’s ontology, not a direct population version of the original game. Also, Theorem 2 does not fully specify the bit complexity of maximizing its univariate continuous functions. If a referee insists that only the two strategic players count as agents, this mirror fails; cloning them would change individual deviations into coalition deviations and would not be faithful. The positive case survives only by keeping the two-player strategic game intact and interpreting the many repeated constituencies as the high-multiplicity society.