{"id":"4b1bc11f-a5ff-4048-8c07-d078c5fd42f5","arxiv_id":"2411.12080","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A comparison principle is proved for occupied PDEs, the dynamic programming equations of control problems whose state records the occupation measure of a diffusion, giving uniqueness of the viscosity solution.","lead":"This paper proves a mathematical result about control problems where the cost depends on the entire history of where the state has been, not just its current position. It shows the governing equations have a unique solution, which underpins pricing of exotic financial contracts such as timer options.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3.10 asserts uniqueness without a growth class: Theorem 3.9 compares only bounded functions, while Assumption 3.4 allows linear-growth payoffs that make v unbounded, so the claimed uniqueness is not a direct consequence and may be false as stated.","rationale":"The reader's verdict is CONDITIONAL and already notes in the rationale that Theorem 3.10 applies bounded comparison to potentially unbounded value functions without localization. My stress-test agrees with that criticism and sharpens it: the problem is not merely a missing proof detail, because the heat-equation example admits nonzero unbounded classical solutions with zero boundary data, so a literal reading of 'unique viscosity solution' without any growth class is false. The reader's weakest_assumption, however, focuses on the cylindrical-norm metrization and the vanishing-penalty step; I do not regard those as the primary load-bearing issue. The metrization assertion is indeed stronger than needed, and the proof only requires ρ to be weak-continuous along weak convergence and vanishing along the penalized maximizers, both of which follow from the uniform l² control on the separating family. Thus my concern partially disagrees with the stated weakest assumption but leaves the CONDITIONAL verdict unchanged: the paper needs an explicit growth class and a comparison theorem in that class.","tokens_in":26705,"tokens_out":44009,"duration_ms":505050,"concrete_test":"Set d=1, T=1, λ=1, b=0, σ=1, ℓ=0, A a singleton, so the occupied PDE is the heat equation -∂_o u - (1/2)u_xx = 0. Take g=0, so the value function is v=0. Construct a standard Tychonoff solution U of the forward heat equation ∂_t U = (1/2)U_xx with U(0,·)≡0 and U(1,0)≠0. Check that u(o,x)=U(1-|o|,x) is a classical, hence viscosity, solution of the occupied heat equation on ˚D_1 with u=0 on ∂D_1. If the example is valid, Theorem 3.10 has no uniqueness among all viscosity solutions; if the authors intend a restricted growth class, this test identifies exactly the missing hypothesis and the needed extension of Theorem 3.9.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central uniqueness claim (Theorem 3.10) is not supported by Theorem 3.9. Under Assumption 3.4 the running and terminal costs satisfy only |ℓ|,|g| ≤ c(1+ρ(o,x)) ≤ c(1+T+|x|), so the value function v is typically unbounded on DT. Theorem 3.9, however, requires a bounded-above subsolution and a bounded-below supersolution, and its proof uses boundedness essentially in Step 1 (to get sup(u_γ^β - w_β^γ)>0) and in Step 4 (to control level sets and the penalty (1/ε)ρ²). No localization, truncation, or stated growth class is supplied. This is not merely a proof gap: without a growth condition, uniqueness fails already for the occupied heat equation (7.2a)-(7.2b) with g=0. Classical Tychonoff solutions U of ∂_t U = (1/2)ΔU with U(0,·)≡0 and U(t,·) not identically zero for t>0 produce u(o,x)=U(T-|o|,x), which is a smooth nonzero viscosity solution of -∂_o u - (1/2)Δu = 0 with u=0 on ∂DT. Thus Theorem 3.10 is false as stated unless a growth class is added and a corresponding comparison result is proved.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a viscosity theory for parabolic PDEs associated with controlled occupied processes, whose state variable is a measure-valued occupation flow together with a finite-dimensional spatial component. The main technical result is a comparison principle (Theorem 3.9) for bounded viscosity sub- and supersolutions, obtained through an infinite-dimensional Crandall-Ishii-Lions lemma, finite-dimensional projections, and coercive approximations. The authors then claim in Theorem 3.10 that the control problem's value function is the unique viscosity solution of the dynamic programming equation, and they illustrate the framework with the occupied heat equation, timer options, and an uncertain-volatility model. The paper contains complete proofs of an Itô formula and of local 1/2-Hölder regularity of the value function. The central difficulty is that the uniqueness statement is made without any growth class, while the comparison theorem only applies to bounded functions; under the stated linear-growth assumptions the value function can be unbounded, and the uniqueness claim is false as stated.","tokens_in":26938,"tokens_out":35797,"duration_ms":371608,"significance":"If restricted to an appropriate class of solutions, the paper's comparison machinery is a meaningful extension of second-order viscosity theory to a measure-valued state variable. The finite-dimensional projection method, the coercive approximation via the gauge o(q)+q(x), and the complete appendices supplying the Itô formula and the local Hölder regularity are clear strengths. The examples are interesting and well chosen. However, the flagship uniqueness theorem substantially overclaims: no growth condition appears in Definition 3.7 or Theorem 3.10, and the proof of Theorem 3.10 invokes a comparison principle that is proved only for bounded functions. Since the stated assumptions allow unbounded value functions and in fact admit extra unbounded viscosity solutions even for the occupied heat equation, the uniqueness result cannot stand in its present form. The comparison proof also contains an unjustified infinity-limit step. These issues are load-bearing and must be resolved before the manuscript can be considered for publication.","major_comments":[{"comment":"Theorem 3.10 asserts that the value function v is the unique viscosity solution of (3.12a)-(3.12b), but no growth class is imposed. Theorem 3.9 compares only an upper-bounded subsolution and a lower-bounded supersolution, and the proof uses boundedness essentially in Step 1 and Step 4. Under Assumption 3.4 the payoff data have only linear growth, so v is typically unbounded; hence Theorem 3.9 cannot be applied to two arbitrary viscosity solutions. This is not a mere proof gap: for the occupied heat equation (7.2a)-(7.2b) with g=0, any nonzero Tychonoff solution U(t,x) of the classical heat equation with U(0,·)=0 produces u(o,x)=U(T-|o|,x), a smooth nonzero viscosity solution of -∂_o u - (1/2)Δu=0 with u=0 on ∂D_T. Thus Theorem 3.10 is false as stated. The theorem must be reformulated, either by adding boundedness of ℓ and g (and then proving v is bounded) or by proving and stating a comparison theorem in a specified growth class that excludes Tychonoff-type solutions.","section":"Section 3.3, Theorem 3.10; Section 7.1"},{"comment":"The final limiting argument is not justified as written. Equation (6.6) defines c(γ1)=sup_{ε,γ2∈(0,β0]} Q(x_ε^γ,x̄_ε^γ) and claims this is finite. But the coercivity bound derived from u^γ-w^γ yields only Q≤C/γ2, so the supremum over γ2∈(0,β0] can be infinite; for fixed γ1, Q can grow like 1/γ2 as γ2↓0. Consequently the asserted limits lim_{γ1↓0}(γ1 Q²)=0 and, more importantly, lim_{γ2↓0}lim_{γ1↓0}lim_{ε↓0} γ2 Q=0 are not established. The citation to [13, Lemma 3.1] covers the finite-dimensional vanishing-penalty statement, not these limits on the unbounded domain D_T with a measure-valued variable. Since the contradiction 0≤-c0β0 at the end of Step 4 depends on these terms vanishing, the comparison proof has a gap that must be closed, either by a complete limiting argument or by adding a hypothesis that controls the behavior of u and w at infinity.","section":"Theorem 3.9, Step 4, equations (6.5)-(6.6)"}],"minor_comments":[{"comment":"The assertion that the cylindrical norm ρ metrizes the weak topology on M_T is false. For example, with f_k∈C_0 chosen as a separating family, the sequence of Dirac measures δ_n satisfies ρ(δ_n,0)→0 but δ_n does not converge weakly to 0 because it is not tight. The paper does not appear to rely on this metrizability for the main proof, since the coercivity of ϑ and the upper semicontinuity of -ρ² suffice for the maximization arguments, but the statement should be corrected or removed.","section":"Section 2.2"},{"comment":"The displayed formulas for θ_K and θ̄_K write (o_*-ō_*)(f_k), but the definitions in Step 2 and the convergence argument in Step 3 give (o_K-ō_K)(f_k). The limit to Θ(x_*) is correct after a dominated-convergence justification, but the displayed formula should use the K-dependent measures.","section":"Lemma 5.1, Step 4"},{"comment":"The expression \"|θ| + θ|\" should read \"|θ| + |θ̄|\", and the intermediate bound for the term involving |θ| needs to be displayed consistently with the final ζ_ε definition.","section":"Lemma 6.2, estimate for I1+I2"},{"comment":"The claim that, for sufficiently small γ and ε, all maximizers of Φ_ε^γ lie in (˚D_T)² is stated without proof. A brief argument showing that a boundary maximizer would make Φ_ε^γ ≤ o(1) as ε↓0, contradicting the positive supremum from (6.2), would make the proof self-contained.","section":"Theorem 3.9, Step 3"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely to attract attention given the topic and the authors, but the uniqueness theorem is currently false as stated. The growth-class problem is the most serious issue; the Tychonoff counterexample in the occupied heat equation makes the overclaim explicit. The comparison proof's infinity-limit step also needs a real fix, not just a citation to the finite-dimensional lemma. I would encourage the editor to require the authors to either restrict the data so that the value function is bounded and Theorem 3.9 applies, or develop a weighted comparison theorem and state Theorem 3.10 in that class."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. The paper's genuine contribution is an infinite-dimensional Crandall-Ishii-Lions lemma with a cylindrical norm and a comparison principle for bounded viscosity sub/supersolutions of occupied PDEs. That is new and worth taking seriously. But Theorem 3.10 oversells it: uniqueness of the value function is claimed under linear-growth data, while Theorem 3.9 only compares bounded functions, and the gap is not cosmetic.\n\nWhat the paper does well: the finite-dimensional projection argument in the spirit of Lasry-Lions is a real adaptation to the measure-valued setting, Lemma 5.1 is a useful technical object, and the appendices give complete proofs of the Itô formula and the local 1/2-Hölder regularity of the value function. The financial examples are translated cleanly.\n\nThe main problem is Theorem 3.10. Assumption 3.4 gives only linear growth on l and g, so the value function need not be bounded. Theorem 3.9 requires a bounded-above subsolution and a bounded-below supersolution, and boundedness is used essentially in the proof. The claimed uniqueness therefore does not follow. Worse, as stated it is false: for the occupied heat equation with g=0, take any nontrivial Tychonoff solution U of the classical heat equation with U(0,·)=0 and set u(o,x)=U(T-|o|,x). This is a continuous viscosity solution of (7.2) with zero boundary datum and is not identically zero. So a growth class must be added and a corresponding comparison result proved; mere appeal to Theorem 3.9 cannot work.\n\nSecondary issues: the assertion that the cylindrical norm rho metrizes the weak topology on M_T is not automatic for an arbitrary separating family on a noncompact space and needs a convergence-determining argument. Likewise, the vanishing-penalty identity (6.5) is cited to the finite-dimensional [13, Lemma 3.1], but in this infinite-dimensional setting it needs proof. Both look fixable. There are also typos in core inequalities, for example missing absolute-value bars in Lemma 6.2.\n\nWho is this for? People working on path-dependent HJB equations, control in Wasserstein/measure spaces, and viscosity methods for nonlocal state variables. The CIL lemma and comparison architecture are valuable even after the growth issue is repaired. But with Theorem 3.10 false as written, the paper needs major revision before it can be relied upon. That said, it deserves a serious referee: the ideas are important enough and the technical core is mostly in place.\n\nRecommendation: send to peer review, but the referee should require a corrected uniqueness statement with a growth class, a proof of the metric/limit facts, and clean handling of the unbounded payoff case. I would not cite the uniqueness theorem in its current form.","headline":"The comparison machinery is a real step forward, but the stated uniqueness theorem is not supported by the bounded comparison result — without a growth class, the occupied heat equation has nontrivial Tychonoff solutions.","tokens_in":27612,"tokens_out":14780,"would_cite":false,"duration_ms":163509,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49L12","35K55","35R15","60J55","93E20"],"pacs":[],"model":"deepseek-v4-flash","headline":"The value function of an occupied-process control problem is the unique viscosity solution of its associated PDE, by a new comparison principle on the space of positive measures.","keywords":["Stochastic optimal control","occupation flow","occupied PDEs","viscosity solutions","comparison principle","Crandall-Ishii-Lions lemma","infinite-dimensional PDE","measure-valued state"],"falsifier":"Take $d=1$ and a separating family $(f_k)$ satisfying (2.4) whose span is not dense in $C_b(\\mathbb{R})$, and check whether $\\rho(o_n-o)\\to0$ with $o_n,o\\in M_T$ forces $o_n\\to o$ weakly. If a counterexample exists, then the claim that $\\rho$ metrizes the weak topology on $M_T$ fails, and the comparison proof's reliance on $\\rho$ as the metric of the state space needs a density assumption on $(f_k)$ to survive.","tokens_in":26371,"feed_emoji":"⚖️","tokens_out":9197,"duration_ms":85668,"temperature":0.7,"pith_summary":"This paper proves a comparison principle for viscosity solutions of the occupied PDE, the infinite-dimensional dynamic programming equation of stochastic control problems whose state includes the occupation measure $O_t=\\int_0^t\\delta_{X_s}\\,d\\Lambda_s$. The main theorem shows that any upper semicontinuous bounded-above subsolution and lower semicontinuous bounded-below supersolution, ordered on the boundary, are ordered on the whole domain. A direct corollary is that the value function of the control problem is the unique viscosity solution. This matters because it gives a nearly classical viscosity framework for a large class of path-dependent control problems, including pricing PDEs for timer options, Asian payoffs, and uncertain-volatility models, without treating full path-dependence.","feed_headline":"Value function is the unique solution of the occupied PDE","feed_subtitle":"A comparison principle on the space of measures gives unique prices for timer and Asian options.","key_machinery":"The argument is carried by three interlocking objects: the occupied process $(O_t,X_t)$ with $O_t$ a measure-valued occupation flow; the occupation derivative $\\partial_o\\varphi(o,x)=\\lim_{h\\downarrow0}(\\varphi(o+h\\delta_x,x)-\\varphi(o,x))/h$, which replaces the time derivative in parabolic PDEs; and a cylindrical norm $\\rho(o,x)=(\\sum_k|o(f_k)|^2+|x|^2)^{1/2}$ built from a separating family $(f_k)$ satisfying $\\sum_k\\|f_k\\|^2_{C^1}\\le1$. With this norm, the paper projects the infinite-dimensional problem onto finite-dimensional subspaces $\\pi_K(o)=(o(f_1),\\dots,o(f_K))$, proves a Crandall-Ishii-Lions lemma (Lemma 5.1) producing second-order jet inequalities in the limit, and uses the Hamiltonian's Crandall-Ishii-Lions property (Lemma 6.2) to run a doubling-of-variables comparison with coercive approximations $w_\\gamma=w+\\gamma\\vartheta$, $u_\\gamma=u-\\gamma\\vartheta$. The vanishing-penalty limit $\\varepsilon\\downarrow0$ then yields the contradiction proving comparison.","core_discovery":"Under Assumption 3.4 (growth, Lipschitz, and a nondegeneracy condition on the clock rate $\\lambda$), the paper establishes Theorem 3.9: if $u$ is a viscosity subsolution of $H(o,x,\\partial_o u,\\nabla u,\\nabla^2 u)=0$ that is upper semicontinuous and bounded above, $w$ is a viscosity supersolution that is lower semicontinuous and bounded below, and $u\\le w$ on the boundary $\\partial D_T$, then $u\\le w$ on $D_T$. The proof uses coercive approximations to overcome the non-compactness of the measure space, an infinite-dimensional Crandall-Ishii-Lions lemma obtained by finite-dimensional projection, and a Crandall-Ishii-Lions property of the Hamiltonian. Theorem 3.10 then identifies the value function $v(o,x)=\\inf_\\alpha J(o,x,\\alpha)$ as the unique viscosity solution, extending classical stochastic-control uniqueness to state spaces of the form $M_q\\times\\mathbb{R}^d$.","pith_inferences":["A testable extension: the same projection scheme should yield comparison for other HJB equations on measure spaces, such as controlled McKean-Vlasov dynamics, provided one verifies the cylindrical-norm metrization and the vanishing-penalty step in that setting; the paper does not prove those transfer.","An implicit numerical consequence: with uniqueness in hand, monotone finite-difference or semi-Lagrangian schemes for the occupied PDE, discretizing $o$ by finitely many test functions $f_k$, should converge to the value function, making path-dependent controls computable through measures rather than full paths.","The paper's Remark 7.1 conjecture, that the occupied heat equation is classically solvable for every weakly continuous terminal datum, suggests that occupation erasing chronology may yield regularity that path-dependent PDEs lack; if confirmed, it would strengthen the case for the occupied-state formulation.","A boundary probe: because the value function is only $1/2$-Hölder in $x$ (Remark 3.6), one can search for an example dropping the nondegeneracy $\\lambda\\ge1/c_*$ where the comparison theorem fails, delimiting how sharp Assumption 3.4 is."],"forward_implications":["The value function $v$ is the unique viscosity solution of the occupied PDE, so the dynamic programming equation is a complete characterization of the optimal-control problem, not just a necessary condition.","Comparison holds for fully nonlinear Hamiltonians on the space of positive measures, covering the occupied heat equation and the pricing PDEs of timer options, Asian-style payoffs, and uncertain-volatility exotic options.","Path dependence enters only through a first-order occupation derivative, so the proof uses classical second-order finite-dimensional tools; it does not require the heavy machinery developed for fully path-dependent PDEs.","The infinite-dimensional Crandall-Ishii-Lions lemma and the coercive-approximation technique provide reusable tools for viscosity theory on other non-compact measure spaces.","Because the state space is locally compact with coercive sub-levels, the comparison proof avoids Ekeland-type variational principles."],"supporting_citations":[{"why":"Supplies the classical viscosity framework, the Crandall-Ishii-Lions lemma, and the finite-dimensional vanishing-penalty estimate (the paper's Lemma 3.1) cited in Step 4 of the comparison proof.","marker":"[13]"},{"why":"Gives the dynamic programming principle and the classical argument that the value function is a viscosity solution, used in Theorem 3.10.","marker":"[21]"},{"why":"Introduces occupied processes and the occupation derivative, the foundational objects of the paper.","marker":"[35]"},{"why":"Provides the existence and uniqueness of strong solutions of the controlled occupied SDE (Theorem 4.2.11), used in Lemma 3.2.","marker":"[36]"},{"why":"Introduces the cylindrical-function regularization technique that the finite-dimensional projections adapt.","marker":"[24]"},{"why":"Establishes finite-dimensional approximations for uniqueness of viscosity solutions in Hilbert spaces, the template for Section 4.3.","marker":"[26]"},{"why":"Version of the Crandall-Ishii maximum principle on closed sets used in Step 2 of the CIL lemma proof.","marker":"[14]"},{"why":"Source for the existence of a separating family $(f_k)$ that defines the cylindrical norm in Section 2.2.","marker":"[28]"}],"fun_headline_variants":["Unique viscosity solution for occupied measure PDEs","Comparison principle yields unique prices for occupied options","Infinite-dimensional viscosity theory for occupied processes","Occupied PDEs: uniqueness from a new comparison principle","Measure-space viscosity solutions pin down value function"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes that the cylindrical distance between the two penalized maximizers collapses as the penalty parameter $\\varepsilon$ goes to zero—the vanishing-penalty identity $(1/\\varepsilon)\\rho^2(o_\\varepsilon-\\bar o_\\varepsilon,x_\\varepsilon-\\bar x_\\varepsilon)\\to0$—and that the cylindrical norm $\\rho$ metrizes the weak topology on the non-compact set $M_T$; both are asserted rather than fully established, and neither is automatic because $M_q$ is incomplete and $D_T$ is not compact.","fun_headline_variants_meta":{"raw":{"variants":["Unique viscosity solution for occupied measure PDEs","Comparison principle yields unique prices for occupied options","Infinite-dimensional viscosity theory for occupied processes","Occupied PDEs: uniqueness from a new comparison principle","Measure-space viscosity solutions pin down value function"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000455,"raw_usage":{"total_tokens":2237,"prompt_tokens":851,"completion_tokens":1386,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":467,"completion_tokens_details":{"reasoning_tokens":1317}},"tokens_in":467,"tokens_out":1386,"duration_ms":9994,"temperature":1.0,"reasoning_tokens":1317,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T18:00:38.455934+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $d=1$ and a separating family $(f_k)$ satisfying (2.4) whose span is not dense in $C_b(\\mathbb{R})$, and check whether $\\rho(o_n-o)\\to0$ with $o_n,o\\in M_T$ forces $o_n\\to o$ weakly. If a counterexample exists, then the claim that $\\rho$ metrizes the weak topology on $M_T$ fails, and the comparison proof's reliance on $\\rho$ as the metric of the state space needs a density assumption on $(f_k)$ to survive.","supporting_citations":[{"cited_title":"Crandall, H","cited_arxiv_id":null,"evidence_quote":"Supplies the classical viscosity framework, the Crandall-Ishii-Lions lemma, and the finite-dimensional vanishing-penalty estimate (the paper's Lemma 3.1) cited in Step 4 of the comparison proof."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the dynamic programming principle and the classical argument that the value function is a viscosity solution, used in Theorem 3.10."},{"cited_title":"Occupied Processes: Going with the Flow","cited_arxiv_id":"2311.07936","evidence_quote":"Introduces occupied processes and the occupation derivative, the foundational objects of the paper."},{"cited_title":"Tissot-Daguette","cited_arxiv_id":null,"evidence_quote":"Provides the existence and uniqueness of strong solutions of the controlled occupied SDE (Theorem 4.2.11), used in Lemma 3.2."},{"cited_title":"Lasry and P.-L","cited_arxiv_id":null,"evidence_quote":"Introduces the cylindrical-function regularization technique that the finite-dimensional projections adapt."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes finite-dimensional approximations for uniqueness of viscosity solutions in Hilbert spaces, the template for Section 4.3."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Version of the Crandall-Ishii maximum principle on closed sets used in Step 2 of the CIL lemma proof."},{"cited_title":"Controlled superprocesses and HJB equation in the space of finite measures","cited_arxiv_id":"2306.15962","evidence_quote":"Source for the existence of a separating family $(f_k)$ that defines the cylindrical norm in Section 2.2."}],"review_version":1}