{"id":"0d1ccaf8-9765-44ff-b279-42ce53a1f6cc","arxiv_id":"2504.15526","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"For a class of intensity-control mean field games, including cryptocurrency mining, the paper proves equilibrium existence and shows continuous-time equilibria arise as limits of discrete-time ones.","lead":"The paper proves that certain large-population games of cryptocurrency mining have well-defined equilibria, both in discrete and continuous time, and that solving the discrete game gives a valid approximation of the continuous one. For anyone building or regulating proof-of-work blockchains, this supplies a rigorous foundation for models that predict which miners stay and which drop out.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3.8's proof assumes an undefined equivalence between Scheme 2 and Scheme 1: the parameter sequence η_k := δ_{ζ_k} is type-inconsistent, so optimality of the discrete controls does not transfer to the interpolated dynamics.","rationale":"I read the paper in good faith: the architecture—discrete-time MFG existence via Kakutani, then tightness and a limit argument to continuous time—is standard, and the application to cryptocurrency mining is natural. The reader's weakest assumption identifies the same spot I find most load-bearing: the equivalence between Scheme 2 and Scheme 1 is asserted rather than defined, and the notation η_k := δ_{ζ_k} is type-inconsistent because λ takes real-valued second arguments while ζ_k is a measure. This is not a mere cosmetic typo: Step 4's optimality contradiction requires that the control from the Scheme 2 equilibrium be optimal for the Scheme 1 dynamics with the specified parameter sequence, and that implication is exactly what the undefined identification was supposed to provide. The gap is plausibly repairable by setting η_k := ζ_k and interpreting λ(a, η) as ∫ λ(a, h) η(dh), which is why I do not recommend rejection. However, as written, the proof of the main existence theorem is not fully rigorous. The reader's conditional verdict is therefore appropriate; I would keep it rather than accept or reject on the current text.","tokens_in":22131,"tokens_out":8723,"duration_ms":82201,"concrete_test":"Re-derive Step 1 with the minimal correction η_k := ζ_k (i.e., drop the Dirac notation) and check whether the Scheme 2 equilibrium optimality transfers to the Scheme 1 discretized chain. Concretely, compare the discrete-time Bellman equations for Scheme 1 and Scheme 2 with the same background measure ζ: if both have the same optimal policy for every ζ, the correction is safe; if they differ, then the convergence proof needs a new optimality argument for the interpolated controls, and the published proof of Theorem 3.8 is incomplete.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 3.8 turns on an undefined identification. In Step 1, a Scheme 2 equilibrium is obtained with population control law ζ_k ∈ P(U) satisfying L(a_k) = ζ_k, and the proof then defines the Scheme 1 parameter sequence as η_k := δ_{ζ_k}. But in Definition 3.6, Scheme 2 writes the jump probability as (1/2n) λ(a_k, η_k) with λ : U^2 → R_+, while Scheme 1 explicitly uses (1/2n) ∫ λ(a_k, h) η_k(dh). If Scheme 2's λ(a, η) is shorthand for ∫ λ(a, h) η(dh), then the natural identification is η_k := ζ_k, not η_k := δ_{ζ_k}; the latter is a probability on P(U) and cannot be inserted into the second argument of λ. If instead Scheme 2 is intended to use only the first moment η_k = ∫ h ζ_k(dh), then the consistency condition L(a_k) = ζ_k does not determine the Scheme 2 dynamics, and optimality of a_k for the original equilibrium does not imply optimality for the Scheme 1 chain used in the convergence argument. In either reading, Step 4's contradiction—that a^(n) is optimal for the fixed sequence η^(n) under Scheme 1—is not established. Since Theorem 3.8 is the paper's central existence result, this unstated equivalence is the load-bearing weak point.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops existence theory for mean field games of control with jump-process dynamics and mean-field interaction through the controls. Section 2 formulates a finite-horizon discrete-time MFG with Polish state/action/control spaces and transition kernels depending on the joint law of controls and states, and states a Kakutani-based existence theorem (Theorem 2.4). Section 3 introduces a continuous-time model in which agents control the intensity of a unit-jump process, with the jump intensity depending on the agent's control and on a flow of population control measures; the main result (Theorem 3.8) claims existence of a relaxed MFG equilibrium as the limit of discrete-time MFG equilibria, with sharpness under concavity. Section 4 applies these results to a cryptocurrency mining model: existence is claimed for the discrete-time and continuous-time versions, including a zero-epsilon limit, and a damped fixed-point algorithm is implemented, reproducing qualitative features of [23]. The paper's central contribution is Theorem 3.8 and its consequences for the cryptocurrency model.","tokens_in":22403,"tokens_out":13524,"duration_ms":118225,"significance":"If the gaps identified below are closed, the paper would establish a general existence result for a class of MFGs of controlled jump intensity with mean-field interaction via controls, a setting for which the authors state no general existence result is available. The discrete-to-continuous convergence result would also provide a rigorous justification for solving continuous-time MFGs by discrete-time algorithms, and the cryptocurrency application gives a concrete, economically motivated testbed with qualitative agreement with existing PDE-based numerics. The paper's use of relaxed controls and its explicit statements of assumptions are appropriate strengths. However, the significance is currently conditional: the central proof contains an undefined identification between the two discretization schemes, and several technical steps are asserted rather than proved.","major_comments":[{"comment":"The paper never defines λ(a,η) when the second argument is a probability measure, although Scheme 2 uses λ(a^{(n)}_k, η^{(n)}_k) with η^{(n)}_k ∈ P(U). The proof then sets the Scheme 1 parameter sequence to η^{(n)}_k := δ_{ζ^{(n)}_k}, where ζ^{(n)}_k ∈ P(U); δ_{ζ^{(n)}_k} belongs to P(P(U)), not P(U), so the interpolated flow η^{(n)}_t is not a measure flow on U and the consistency condition E[m_t]=η_t in Definition 3.5 is not well-typed. In addition, the equilibrium control a^{(n)}_k is optimal in Scheme 2 for the flow ζ^{(n)}, not in Scheme 1 for the flow δ_{ζ^{(n)}}; Step 4's contradiction relies on this transfer of optimality. The proof must either define λ(a,η)=∫λ(a,h)η(dh) and replace δ_{ζ^{(n)}_k} by ζ^{(n)}_k, or provide a separate argument establishing that the two schemes describe the same control problem.","section":"Section 3.2, Definition 3.6 and Theorem 3.8, Step 1"},{"comment":"The proof concludes with 'We omit the remaining details here' for the compactness of Ξ and the closedness of the graph of Γ, referring to [27, Propositions 3.9 and 3.10]. Since the present model allows mean-field interaction through the controls and uses weaker growth assumptions than [27], the reduction is not automatic; these are exactly the properties needed for the Kakutani–Fan–Glicksberg fixed-point theorem. Because Theorem 2.4 is the foundation for Step 1 of Theorem 3.8 and for the discrete-time cryptocurrency model, these details must be supplied rather than deferred.","section":"Section 2.4, proof of Theorem 2.4"},{"comment":"The verification that the limiting jump process N has the stated stochastic intensity is incomplete. The displayed computation conditions on the limit σ-field F_s, but the discrete-time martingale property is with respect to F^{(n)}_s, and the assertion that the limit 'can be seen to equal zero from the construction' is not a proof. The paper should provide a standard martingale-problem convergence argument, for example by establishing uniform integrability and the Aldous–Robin condition, before concluding that (X,m,N) satisfies the dynamics of Definition 3.2.","section":"Section 3.2, Step 2 of Theorem 3.8"},{"comment":"The proof asserts that the delayed block construction satisfies L(m^{ρ,∆})=L(m) 'by construction', but the randomization from the previous block is used in the current block, so the laws are not identical; no quantitative estimate on |W(t,x,m)-W(t,x,u^γ_m)| is given. Since Step 4 uses this lemma to produce a piecewise constant control that strictly outperforms the putative limit control m, the optimality assertion of Theorem 3.8 depends on this gap.","section":"Section 3.2, Lemma 3.10 (Chattering Lemma)"},{"comment":"The intensity map λ^{(ϵ)} is defined on U×U, but it is evaluated at ζ^{(n,ϵ)}_k ∈ P([0,L]) in (4.2), and Proposition 4.2 treats η_k as a scalar first moment (writing η_k = 0 and comparing intensities via Jensen's inequality). The paper never states the convention λ(a,η)=∫λ(a,h)η(dh) nor distinguishes between the measure flow and its first moment. This ambiguity affects the verification of Assumption 3.7 and the limiting arguments in Propositions 4.3–4.6, so the cryptocurrency existence claims are not rigorously grounded as written.","section":"Section 4, equations (4.1)–(4.2) and Propositions 4.2–4.6"}],"minor_comments":[{"comment":"The optimality clause compares the tuple with 'any other tuple' whose flow η may differ; the proofs compare controls only under a fixed η. The definition should explicitly restrict the comparison to tuples with the same flow.","section":"Definition 3.5"},{"comment":"The phrase 'for almost every ω∈F' should be 'for P-almost every ω'.","section":"Definition 3.5"},{"comment":"The displayed assumption contains '∀∈X' instead of 'for all x∈X'.","section":"Assumption 2.3(vi)"},{"comment":"Lemma 3.11 is not used in the final proof of Theorem 3.8, and its proof only shows convergence of the objective for a fixed control, not the claimed convergence of the value function; either add a uniformity argument or remove the lemma.","section":"Section 3.2, Lemma 3.11"},{"comment":"The statement that uniqueness of the discrete-time equilibrium 'allows one to conclude' sharpness of the continuous-time equilibrium is not justified; at most it is consistent with sharpness, not a proof.","section":"Section 4.4"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a timely and relevant topic, and the authors are honest about relying on earlier work. The central idea is plausible, and the main problems are technical incompleteness and an identification error in the proof of Theorem 3.8, both of which appear fixable within the scope of a revision. I would be willing to review a revised version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The short version: this paper proves a genuine new existence result for discrete-time mean field games with mean-field interaction through controls, and a discrete-to-continuous limit theorem for controlled-intensity jump MFGs. The cryptocurrency mining application is a useful worked example. But the proof of Theorem 3.8, the load-bearing continuous-time result, has an identification that is undefined as written. I think it is repairable, but a referee should require the fix.\n\nWhat is actually new: Theorem 2.4 extends the Saldi–Basar–Raginsky fixed-point framework from state-only interaction to control interaction, under a weaker growth condition. Theorem 3.8 is new for intensity-control MFGs, and the proof architecture—Kakutani for discrete time, tightness with Skorokhod for the limit—is standard but competently assembled. The chattering lemma and the value-function continuity lemma are useful standalone pieces. The authors also stay honest about scope: uniqueness is explicitly left open.\n\nSoft spots, in proportion.\n\nFirst and most important, Step 1 of Theorem 3.8. A Scheme 2 equilibrium gives a population law ζ_k ∈ P(U). The proof then sets the Scheme 1 parameter sequence to η_k := δ_{ζ_k}. That is a probability on P(U), not on U, and Scheme 1 needs η_k ∈ P(U) to plug into λ(a,h). If the intended identification is η_k := ζ_k, everything works because Scheme 2's λ(a,η) is presumably ∫ λ(a,h) η(dh); if the intended identification is the first moment, then optimality of the discrete controls does not transfer to the interpolated dynamics. The paper should just state which. I read this as a typo rather than a deep flaw, but as written it is a real gap in the proof of the main theorem.\n\nSecond, Theorem 2.4's proof omits the compactness and graph-closedness details, citing [27]. That is likely fine for the intended audience, but the paper should sketch how the control interaction affects those steps.\n\nThird, the numerical section is illustrative: no code, no stopping criterion, one damping factor, no sensitivity analysis. Acceptable for an existence paper, but it should be labeled as a demonstration.\n\nThe citation pattern looks honest. [23] is used for the model and comparison, and the new existence theorems do not depend on its conclusions. No circularity.\n\nBottom line: this deserves a serious referee. I would send it out and require the η_k identification and the Theorem 2.4 details to be addressed before acceptance. The core ideas are sound and the contribution is real.\n\nRecommendation: engage with it. Cite it for the discrete-time control-interaction existence, and push for a clean proof of Theorem 3.8.","headline":"Real contribution on discrete-time MFGs with control interaction, but Theorem 3.8 has an undefined discretization-scheme identification that should be fixed before the proof is accepted.","tokens_in":22944,"tokens_out":3847,"would_cite":true,"duration_ms":31906,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91A16","49N80"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that mean field games of controlled jump intensity admit equilibria, obtained as limits of discrete-time games, and that the result yields existence for a cryptocurrency mining competition.","keywords":["mean field games","controlled jump intensity","discrete-time mean field games","continuous-time limit","cryptocurrency mining","relaxed controls","fixed-point iteration","weak convergence"],"falsifier":"For a small case, such as a compact action space, intensity function equal to a divided by a plus ten times the background hash-rate, and linear terminal utility, compute the discrete-time equilibrium under Scheme 2. If running the resulting interpolated controls against the Scheme 1 parameter sequence given by point masses at the population control laws yields a different expected terminal reward than the Scheme 2 value function, the convergence proof's identification of the schemes is broken and the limit is not guaranteed to be a mean field game equilibrium.","tokens_in":21903,"feed_emoji":"⛏","tokens_out":7735,"duration_ms":69061,"temperature":0.7,"pith_summary":"This paper establishes existence of mean field game equilibria for models in which a large population of agents controls the intensity of a jump process and interacts through the population's distribution of controls. It first proves a general existence theorem for finite-horizon discrete-time mean field games, then shows that a continuous-time controlled-intensity game arises as a limit of these discrete-time games. Applied to cryptocurrency mining, the paper provides existence guarantees for both the discrete- and continuous-time games and demonstrates a numerical solver based on damped fixed-point iterations. The upshot is that difficult coupled PDE systems can be replaced by directly solving discrete-time games.","feed_headline":"Existence proven for mean field games of mining competition","feed_subtitle":"Discrete-time equilibria converge to continuous-time ones, so mining competition becomes solvable by fixed-point iteration.","key_machinery":"The argument runs on two machines. In discrete time, a set-valued operator is built from the Bellman optimality condition and the consistency, or Kolmogorov, condition, and a fixed-point theorem for set-valued maps supplies an equilibrium. In continuous time, the central object is the relaxed control, a random measure on the product of the action space and the time interval, together with the bilinear intensity formed by integrating the intensity kernel against the agent's relaxed control and the population's control measure. The proof interpolates discrete-time equilibria into cadlag processes, extracts weak limits via tightness, verifies that the limit has the correct stochastic intensity and satisfies the consistency condition, then proves optimality by a chattering lemma that approximates any relaxed control by piecewise-constant controls. Strict concavity of the intensity in the agent's control converts the relaxed equilibrium into one with sharp controls.","core_discovery":"The central claim is Theorem 3.8: under continuity, compact support, Lipschitz intensity, and boundedness assumptions, the continuous-time mean field game of controlled jump intensity has an equilibrium in relaxed controls, and if the terminal utility is strictly increasing and the intensity is strictly concave in the agent's control, the equilibrium uses sharp, non-randomized controls. The proof obtains this equilibrium as a weak limit of discrete-time mean field equilibria arising from Bernoulli-chain discretizations, using tightness of relaxed controls and a chattering lemma to rule out any gain from randomization. Along the way, a general existence theorem for finite-horizon discrete-time mean field games is proved through a set-valued fixed-point argument whose fixed points are exactly mean field equilibria. The same machinery yields existence for the cryptocurrency mining model motivating the paper, including the singular limit where the regularizing parameter is removed.","pith_inferences":["The proof's identification of Scheme 2 equilibria with Scheme 1 parameters is only as strong as an implicit rule for inserting a population control measure into the intensity function; making that rule explicit would let the continuity of equilibria in the population law be checked directly.","The same interpolation-compactness route should extend to state-dependent drift and intensity, since the chattering lemma and tightness criteria do not use state-independence except to simplify the dynamics; testing the method on a jump-diffusion with state-dependent coefficients would delimit the true boundary.","If the equilibrium hash-rate stays bounded away from zero uniformly, as the paper suggests, the uniqueness argument sketched near the end could be completed by verifying that the best-response map is a contraction in the sup norm; numerical continuation in the damping factor could expose any non-uniqueness."],"forward_implications":["Continuous-time mean field games of controlled jump intensity can be solved by computing discrete-time mean field equilibria directly, without solving coupled Hamilton-Jacobi-Bellman and Kolmogorov equations.","The cryptocurrency mining mean field game, previously treated numerically, now carries an existence guarantee in both its regularized and singular forms.","When the intensity is strictly concave in the agent's control and terminal utility is strictly increasing, randomized controls are unnecessary: some deterministic, sharp equilibrium exists.","The discrete-time algorithm reproduces the qualitative equilibrium behavior, including wealth-dependent dropout and preferential attachment, seen in the PDE-based numerical solutions."],"supporting_citations":[{"why":"Supplies the fixed-point methodology and notation for discrete-time mean field game equilibrium existence.","marker":"[27]"},{"why":"Supplies the tightness criterion and weak-convergence machinery used to pass to the continuous-time limit.","marker":"[19]"},{"why":"Supplies the weak-convergence and chattering approximations used in the optimality step.","marker":"[18]"},{"why":"Supplies the cryptocurrency mining model and the numerical PDE-based solution that this paper's discrete-time solver reproduces.","marker":"[23]"},{"why":"Supplies the convergence-of-Nash-equilibria perspective for mean field games of controls that the continuous-time result complements.","marker":"[10]"},{"why":"Supplies the martingale characterization of stochastic-intensity jump processes used to identify the limit intensity.","marker":"[8]"},{"why":"Supplies the set-valued fixed-point theorem used in the discrete-time existence proof.","marker":"[2]"}],"fun_headline_variants":["Cryptocurrency mining competition solved as mean field game","Equilibria proven for mean field games with controlled jumps","Discrete-time limits yield equilibria for mining mean field games","Mean field games of intensity control have equilibria, via discretization","Mining competition equilibrium proven by discrete-time approximation"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes that a discrete-time equilibrium stated in terms of population control distributions can be repackaged as a sequence of point-mass parameters in the intensity function without losing optimality of the resampled controls, but the paper never defines the insertion rule.","fun_headline_variants_meta":{"raw":{"variants":["Cryptocurrency mining competition solved as mean field game","Equilibria proven for mean field games with controlled jumps","Discrete-time limits yield equilibria for mining mean field games","Mean field games of intensity control have equilibria, via discretization","Mining competition equilibrium proven by discrete-time approximation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001167,"raw_usage":{"total_tokens":4773,"prompt_tokens":834,"completion_tokens":3939,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":450,"completion_tokens_details":{"reasoning_tokens":3858}},"tokens_in":450,"tokens_out":3939,"duration_ms":24970,"temperature":1.0,"reasoning_tokens":3858,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T11:25:09.597822+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a small case, such as a compact action space, intensity function equal to a divided by a plus ten times the background hash-rate, and linear terminal utility, compute the discrete-time equilibrium under Scheme 2. If running the resulting interpolated controls against the Scheme 1 parameter sequence given by point masses at the population control laws yields a different expected terminal reward than the Scheme 2 value function, the convergence proof's identification of the schemes is broken and the limit is not guaranteed to be a mean field game equilibrium.","supporting_citations":[{"cited_title":"Saldi, T","cited_arxiv_id":null,"evidence_quote":"Supplies the fixed-point methodology and notation for discrete-time mean field game equilibrium existence."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the tightness criterion and weak-convergence machinery used to pass to the continuous-time limit."},{"cited_title":"Kushner.Weak convergence methods and singularly perturbed stochastic control and filtering problems","cited_arxiv_id":null,"evidence_quote":"Supplies the weak-convergence and chattering approximations used in the optimality step."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the cryptocurrency mining model and the numerical PDE-based solution that this paper's discrete-time solver reproduces."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the convergence-of-Nash-equilibria perspective for mean field games of controls that the continuous-time result complements."},{"cited_title":"Brémaud.Point processes and queues","cited_arxiv_id":null,"evidence_quote":"Supplies the martingale characterization of stochastic-intensity jump processes used to identify the limit intensity."}],"review_version":1}