{"id":"f4a47a1d-8fde-4d10-9a30-6b4aa36a604d","arxiv_id":"2507.10604","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A finite-horizon mean field game of renewable capacity expansion yields a threshold time and a capacity threshold that separate investing from non-investing producers.","lead":"This paper models renewable electricity producers as players in a mean field game, where both electricity prices and installation costs respond to aggregate industry activity. It shows that a finite planning horizon creates a cutoff time after which all investment stops, and that smaller producers invest faster than larger ones.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified","rationale":"My reading supports the reader's verdict of CONDITIONAL. The paper delivers what it claims: a finite-horizon capacity-expansion MFG with both state and control interactions, a tractable homogeneous analysis with rigorous threshold results, and a heterogeneous HJB/FP framework with a concrete numerical approach. The central claim is conditional in nature, and the conditions are stated. The main weaknesses are (i) the continuum aggregate approximation (15) is not error-analyzed in the paper and refers to an 'In preparation' companion [22], (ii) well-posedness of the nonlinear coupled HJB/FP system (18)-(19) is not established, and (iii) convergence of the fixed-point iteration in Section 4.3.1 and of the finite-difference scheme in Section 4.3.2 is not proved. None of these is an internal inconsistency; each is an unproved but addressable hypothesis or deferred analysis. The weakest load-bearing assumption is indeed (i), exactly as the reader identified. Since the finite-player-to-mean-field approximation is what connects the motivating model to the studied MFG, the usefulness of the framework for real N depends on it, and the paper's reliance on a reference 'In preparation' is the single most important gap. However, the paper is explicit about this and all formal results are conditional on the approximation being reasonable; a nontrivial continuum model on its own is still well posed as a modeling claim. Therefore I do not oppose the reader's CONDITIONAL verdict, and no verdict adjustment is needed.","tokens_in":22494,"tokens_out":1687,"duration_ms":17267,"concrete_test":"A single verification worth running: contract the finite-player game with N+1 producers for small N (e.g., N=5,10,50) with linear price, compare each producer's capacity trajectory and payoff against the corresponding MFG prediction obtained from the algorithm in Section 4.3.1, and assess how the mean-field approximation (15) converges as N grows for the specific parameters used in Section 4.3. If the mismatch decays with N, the deferred approximation error in [22] is consistent with the paper's modeling claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"No significant objection identified. The paper's central claim is best understood as a modeling-and-methods contribution: equilibria of both the homogeneous and heterogeneous capacity-expansion MFG are characterized by the coupled HJB/FP structure (18)-(19), and the equilibrium exhibits the threshold/investment-stop structure formalized in Lemma 3.3 and the x*(t)/T* analysis of Section 4.2. Within the stated scope, the paper is careful and internally consistent: Proposition 3.1 proves existence/uniqueness for the homogeneous forward-backward system; Lemma 3.3 gives the threshold characterization; Lemma 4.1 and Lemma 4.2 give the concavity and stopping-region properties under assumptions that are stated explicitly; and Section 4.1 shows consistency between the two derivations (Pontryagin and HJB). The weakest point already identified by the reader is the continuum aggregate approximation (15), whose error analysis is deferred to reference [22], listed as 'In preparation.' This is a legitimate limitation of the paper, and it is acknowledged in the text. However, because the approximation is used as a modeling ansatz rather than as a theorem on which the other results depend, it does not undermine the internal validity of the results conditional on that ansatz. The paper is transparent about this, and the finite-player approximation error is a reasonable future-work item rather than a flaw in the argument as presented. Similarly, the lack of a full well-posedness proof for the coupled HJB/FP system and the lack of a convergence proof for the numerical algorithm are accurately described limitations, not hidden errors.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a finite-horizon mean field game for capacity expansion in renewable electricity markets, in which producers interact through both the electricity price (a decreasing function of aggregate installed capacity) and the installation cost (increasing in the aggregate installation rate). In the homogeneous case, the game is reduced to a forward-backward ODE system (Eq. (6)); the paper proves existence and uniqueness, identifies a threshold time T* after which investment ceases, and gives a semi-explicit solution for a linear price function. In the heterogeneous case, the authors introduce a continuum aggregate approximation, derive a coupled HJB-FP system (Eqs. (18)-(19)), establish concavity of the value function and a threshold capacity x*(t), and propose two numerical methods for the linear-price and inverse-price specifications. A stochastic extension with idiosyncratic capacity shocks is also presented.","tokens_in":22839,"tokens_out":28678,"duration_ms":308118,"significance":"If the results are correct, the paper provides a useful and relatively tractable framework for studying strategic capacity expansion with crowding effects, and it extends the infinite-horizon model of [5] to finite horizons with heterogeneous producers. The homogeneous analysis is careful: Proposition 3.1 and Lemmas 3.3-3.4 give a reasonably complete picture of the forward-backward system, and the threshold/investment-stop structure is an appealing economic insight. The heterogeneous part is also transparent about its structural assumptions, and the numerical methods are cross-checked against each other. However, the paper's central heterogeneous claims rest on an unverified continuum approximation (Eq. (15)), and one displayed semi-explicit formula (the constant C in Proposition 3.2) appears to be incorrect. These issues need to be addressed before the paper can be recommended for publication.","major_comments":[{"comment":"The constant C in Eq. (11) does not satisfy the stated boundary conditions. Solving X(0)=X0 and δX_T* + Xdot_T* = 0 for X = C e^{r1 t} + D e^{r2 t} + θ gives C = ((X0 - θ)(r2 + δ) + δθ e^{-r2 T*}) / ((r2 + δ) - e^{(r1-r2)T*}(r1 + δ)). The displayed formula instead has (X0 + θ)(r2 + δ) - δθ e^{-r2 T*} in the numerator. In the special case d2 = 0, one has r1 = r + δ and r2 = -δ, and the displayed formula yields C = +δθ e^{-(r+δ)T*}/(r + 2δ), whereas the boundary condition δX_T* + Xdot_T* = 0 requires C = -δθ e^{-(r+δ)T*}/(r + 2δ). This error propagates to Eq. (12) and to the claimed agreement between the shooting method and the semi-explicit solution in Figure 1b.","section":"Section 3.2, Proposition 3.2"},{"comment":"The continuum aggregate approximation X_t ≈ x_t + N xbar_t and K_t ≈ ν_t + N νbar_t is the only bridge from the finite-player game (1)-(5) to the HJB-FP system (18)-(19). No error estimate for this approximation is provided, and the cited analysis [22] is listed as 'In preparation'. Since all heterogeneous results and the numerical experiments are formulated in this approximated game, the claims about the original finite-player model are conditional on an unverified approximation. The authors should either supply a quantitative error analysis (even for a restricted class of parameters or price functions) or explicitly restrict the paper's claims to the continuum aggregate game rather than to the N-player game.","section":"Section 4, Eq. (15)"}],"minor_comments":[{"comment":"The line 'implying Xt = X0 e^{-δt} > X0' has the inequality direction reversed; since K_s = 0 on [0,t], one has X_t = X0 e^{-δt} < X0. The subsequent comparison 0 ≤ dot u_t ≤ (r+δ)α - (P(X0)-c) is valid only with this corrected direction. The final sentence of the same proof should identify the actual contradiction: from dot u_t1 = dot u_t2 = 0 one obtains P(X_t1) = P(X_t2), hence X_t1 = X_t2, contradicting the strict decay X_t1 > X_t2.","section":"Lemma 3.3, proof"},{"comment":"After the Gronwall estimate, the displayed exponent (r + δ + hLM/(2β))T contains an unexplained factor 2. Applying Gronwall to Δ(T) ≤ Δ(0)e^{(r+δ)T} + (hLM/β)∫_0^T Δ(t) dt gives the constant hLM/β, not hLM/(2β). This is a typographical issue, but it should be corrected.","section":"Proposition 3.1, proof"},{"comment":"The list of boundary conditions duplicates the 'at x = 0' bullet and writes 'at T = 0' where it should be 'at t = T'. The terminal condition for the value function should be imposed at the final time, not at T = 0.","section":"Section 4.3.2"},{"comment":"The running reward f is written with hP(x_s + N xbar_s) - c, which is ambiguous; it should be h(P(x_s + N xbar_s) - c) so that the factor h multiplies both the price and the cost term.","section":"Eq. (25)"},{"comment":"Proposition 3.1 is stated for a globally Lipschitz price function P. The discussion of P(x) = p/x correctly notes that X_t is bounded away from zero and hence P is locally Lipschitz on the reachable set, but this should be formalized as a separate argument rather than asserted as 'effectively Lipschitz', because the proof of the comparison principle in Lemma 3.1 uses the global Lipschitz bound.","section":"Section 3.3, inverse price"},{"comment":"The definition T* = inf{t : x*(s) = 0 for all s ∈ [t,T]} and the subsequent equalities x*(T*) = 0 and νbar_T* = 0 presume continuity of x* and νbar along the equilibrium. The paper does not state the regularity assumptions under which these properties hold; a brief regularity discussion or explicit assumptions would make the threshold characterization rigorous.","section":"Section 4.2, definition of T*"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid modeling contribution and the homogeneous ODE analysis is largely correct and clearly presented. The main obstacles are the incorrect constant in Proposition 3.2 and the unverified continuum approximation in Eq. (15). Both are fixable within the manuscript's scope, so I recommend major revision rather than rejection. If the authors can correct the semi-explicit formula and either prove an error bound for the approximation or carefully restate the results as applying to the continuum aggregate game, the paper would be a useful contribution to the applied MFG literature."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe headline: this is a finite-horizon extension of the renewable-capacity MFG in Alasseur et al. (2023), and it adds a genuinely heterogeneous state-and-control MFG with a threshold structure. The homogeneous part is solid and mostly careful; the heterogeneous part is plausible but leans on an approximation whose error is never quantified.\n\nWhat's new and worthwhile: Section 3 proves existence and uniqueness for the forward-backward ODE system (Prop 3.1), identifies a threshold time T* (Lemma 3.3), and gives a semi-explicit linear-price solution with a transcendental equation for T*. The shooting method and the turnpike behavior for long horizons are nice. Section 4 is the bigger contribution: a heterogeneous MFG with interaction through both capacity and installation rate, a free-boundary threshold x*(t), and a clean derivation of the non-installation value function. Recovering the homogeneous ODE system from the HJB formulation in Section 4.1 is a good consistency check.\n\nSoft spots, in proportion: the continuum aggregate approximation (15) is load-bearing for everything in the heterogeneous model, and its error analysis is deferred to an in-preparation paper. This is a real gap, though the authors flag it. The coupled HJB-FP system is not proved well-posed, and the numerical scheme's convergence is not established. The quadratic ansatz used in 4.3.1 is explicitly an approximation, matching only at T* rather than along the whole free boundary; they acknowledge this, but it means the reported threshold curves are approximate without a verified error bound. Minor slip-ups: a duplicated boundary-condition bullet in 4.3.2, a suspicious factor of 2 in the Gronwall exponent in Prop 3.1's proof, and a sign typo in Lemma 3.2's proof. None of these affect the main argument.\n\nOverall: within the stated scope, the central claims hold up, and the paper is transparent about what is not proved. It is a genuine modeling-and-methods contribution for MFG applications to energy investment, not a theorem machine.\n\nI'd send it to peer review. A good referee will ask for a finite-N numerical check or an error bound on the continuum approximation, and for a statement on well-posedness of the HJB-FP system. Those are addressable, not fatal.","headline":"Finite-horizon capacity expansion MFG with a solid homogeneous part and a plausible heterogeneous extension that leans on an unquantified continuum approximation.","tokens_in":23294,"tokens_out":2569,"would_cite":true,"duration_ms":26616,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91A16","49L12","91B74"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that in equilibrium, renewable producers invest only while their installed capacity sits below a time-dependent threshold, and that all investment ceases before the planning horizon ends.","keywords":["mean field games","capacity expansion","renewable energy investment","optimal control","Hamilton-Jacobi-Bellman equations","Fokker-Planck equations","threshold investment","crowding effects"],"falsifier":"Simulate the finite-player game with $N=10$ and $N=100$ producers using the paper's linear-price parameters and compare the resulting aggregate investment trajectories and the time at which investment stops with the mean field equilibrium; if the discrepancy does not shrink as $N$ grows, the continuum approximation and the coupled HJB-FP characterization fail to describe the finite game.","tokens_in":22322,"feed_emoji":"☀️","tokens_out":9625,"duration_ms":105577,"temperature":0.7,"pith_summary":"This paper sets out to model how many competing renewable electricity producers decide to build capacity over a finite planning horizon. In the model, the market price falls when total installed capacity rises, and the installation cost rises when everyone builds at once, so each producer's decision feeds back into everyone else's opportunity. The central claim is that the resulting mean field equilibrium has a threshold structure: at each date a producer installs new capacity only if its current capacity is below a cutoff, and after a critical time no producer invests at all. The authors prove this in the homogeneous case, derive semi-explicit solutions for linear and inverse price functions, and build numerical schemes for the heterogeneous version. If the claim holds, decentralized renewable investment should display a predictable aggregate pattern of early rapid build-out followed by a passive decay phase.","feed_headline":"Green investors stop building once capacity passes a threshold","feed_subtitle":"In a competition model, firms install only while their capacity is small; after a critical time, nobody builds.","key_machinery":"The load-bearing object is the threshold curve $x^\\star(t)$, defined by $\\partial V/\\partial x(t,x^\\star(t)) = \\alpha + \\beta N\\bar\\nu_t$ whenever the threshold is positive; here $V$ is the representative producer's value function, $\\alpha$ the marginal installation cost, $\\beta$ the crowding sensitivity, and $N\\bar\\nu_t$ the aggregate installation rate feeding back through the cost function. It separates a lower installation region, where producers add capacity, from an upper non-installation region, where capacity only decays and the value function has the closed form (26). The fixed-point condition (20) determines the mean installation rate from the density $m(t,x)$ and the value function, closing the mean field loop. In the homogeneous limit this same mechanism reduces to the adjoint variable $u_t$ crossing $\\alpha$ once, which is the content of Lemma 3.3.","core_discovery":"For a continuum of heterogeneous producers, the paper characterizes equilibrium by the coupled Hamilton-Jacobi-Bellman equation (18) and Fokker-Planck equation (19), joined through the mean capacity $\\bar x_t$ and the mean installation rate $\\bar\\nu_t$. The optimal installation rate is $\\nu^\\star(t,x)=\\frac{1}{2\\beta}(\\partial V/\\partial x-\\alpha-\\beta N\\bar\\nu_t)^+$, so the value function's slope against the full marginal cost of installing capacity decides behavior. The authors show that, when revenue $xP(x+N\\bar x)$ is concave in the producer's own capacity, the value function is concave and the optimal installation rate is non-increasing in capacity. Therefore the installation region is always an interval below a threshold $x^\\star(t)$; producers at or above the threshold never invest, and in the homogeneous limit the critical time $T^\\star$ is characterized by the adjoint variable $u_t$ crossing the marginal installation cost $\\alpha$ exactly once (Lemma 3.3).","pith_inferences":["The paper leaves open the magnitude of the error in replacing finite aggregates by continuum means; a direct finite-player simulation would show how large the number of producers must be before the threshold structure becomes a good description.","Because the threshold is pinned by the marginal cost of installation, policies that lower that cost or the crowding sensitivity would shift the investment window and the stopping time; this policy lever is present in the model but not developed.","The same state-and-control mean field machinery likely applies to storage or other capacity-constrained green technologies with supply-chain congestion, although the paper does not claim this extension.","The precautionary effect in the stochastic version suggests a testable prediction: measured uncertainty in renewable output should correlate with a later peak and a lower peak of aggregate investment, a comparison the paper does not run."],"forward_implications":["With a Lipschitz price function the homogeneous forward-backward system has a unique solution, and under Assumption 3.1 there is a unique critical time $T^\\star$ before which investment happens and after which capacity decays at rate $\\delta$.","For a linear price function, capacity follows a two-exponential formula during the investment phase, and the switch time solves a transcendental equation, so the threshold can be computed semi-explicitly.","For heterogeneous producers, smaller producers invest at a higher rate than larger ones, and a producer whose capacity enters the non-installation region never invests again; all producers stop by a common terminal time $T^\\star$.","With stochastic capacity dynamics driven by geometric Brownian motion, the installation threshold is smaller than in the deterministic case, giving a precautionary slowdown in investment.","For long planning horizons, the capacity path approaches the infinite-horizon steady state and stays near it until roughly 8.5 years before the terminal date in the reported parameter set."],"supporting_citations":[{"why":"Supplies the infinite-horizon mean field model of renewable capacity development that this paper reformulates with a finite horizon and heterogeneity.","marker":"[5]"},{"why":"Is cited as the source for the error analysis of the continuum aggregate approximation; the heterogeneous model's validity rests on that approximation.","marker":"[22]"},{"why":"Introduces the aggregate-replacement technique for state-and-control mean field games that the paper adapts for capacity and installation cost interaction.","marker":"[26]"}],"fun_headline_variants":["Capacity building halts at threshold in mean field game","Renewable investors stop installing above capacity threshold","Mean field model: when crowding stops green investment","Threshold behavior in renewable capacity expansion game","Optimal installation: build only below capacity threshold"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The heterogeneous equilibrium requires that replacing true aggregate capacity and aggregate installation rate by their continuum averages is accurate, and the paper does not quantify how far that approximation is from the finite-player game.","fun_headline_variants_meta":{"raw":{"variants":["Capacity building halts at threshold in mean field game","Renewable investors stop installing above capacity threshold","Mean field model: when crowding stops green investment","Threshold behavior in renewable capacity expansion game","Optimal installation: build only below capacity threshold"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00035,"raw_usage":{"total_tokens":1871,"prompt_tokens":865,"completion_tokens":1006,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":481,"completion_tokens_details":{"reasoning_tokens":936}},"tokens_in":481,"tokens_out":1006,"duration_ms":8886,"temperature":1.0,"reasoning_tokens":936,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T17:57:44.676099+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the finite-player game with $N=10$ and $N=100$ producers using the paper's linear-price parameters and compare the resulting aggregate investment trajectories and the time at which investment stops with the mean field equilibrium; if the discrepancy does not shrink as $N$ grows, the continuum approximation and the coupled HJB-FP characterization fail to describe the finite game.","supporting_citations":[{"cited_title":"Alasseur, M","cited_arxiv_id":null,"evidence_quote":"Supplies the infinite-horizon mean field model of renewable capacity development that this paper reformulates with a finite horizon and heterogeneity."},{"cited_title":"Garcia, M","cited_arxiv_id":null,"evidence_quote":"Is cited as the source for the error analysis of the continuum aggregate approximation; the heterogeneous model's validity rests on that approximation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the aggregate-replacement technique for state-and-control mean field games that the paper adapts for capacity and installation cost interaction."}],"review_version":1}