{"paper_id":"2510.02910","title":"Joint Stochastic Optimal Control and Stopping in Aquaculture: Finite-Difference and PINN-Based Approaches","authors":["Kevin Kamm"],"paper_date":"2025-10-03","paper_type":"methodological","paper_url":"https://arxiv.org/pdf/2510.02910v1","page_url":"https://thequant.space/flowcharts/joint-stochastic-optimal-control-and-stopping-in-aquaculture/","scores":{"empirical_rigor":6,"math_complexity":7.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a joint stochastic optimal control and stopping model for aquaculture operations and develops two numerical solution approaches: a finite difference benchmark and a Physics-Informed Neural Network (PINN) combined with a deep optimal stopping algorithm.","quote":"This paper studies a joint stochastic optimal control and stopping (JCtrlOS) problem motivated by aquaculture operations, where the objective is to maximize farm profit through an optimal feeding strategy and harvesting time under stochastic price dynamics. We develop two numerical solution approaches: First, a finite difference scheme that serves as a benchmark, and second, a Physics-Informed Neural Network","summary":"This paper studies a joint stochastic optimal control and stopping (JCtrlOS) problem motivated by aquaculture operations, where the objective is to maximize farm profit through an optimal feeding strategy and harvesting time under stochastic price dynamics. We introduce a simplified aquaculture…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"There exists an optimal deterministic biological feeding strategy f(t) that depends only on time.","type":"model","stated":"explicit","quote":"We will assume that there exists an optimal (in the sense of maximal growth) deterministic feeding strategy f : R+ →R+ only dependent on time.","support":"full"},{"assumption":"The price dynamics of feed and biomass follow geometric Brownian motions under a risk‑neutral measure.","type":"market","stated":"explicit","quote":"We will assume that the price dynamics are both by two geometric Brownian motions with dynamics under a risk-neutral measure Q given by dP F t = rP F t dt + σF P F t dW F t =: bpF (t, P F t )dt + σpF (t, P F t )dW F t , P F 0 = pF > 0, dP B t = rP B t dt + σBP B t dW B t =: bpB(t, P B t )dt + σpB(t, P B t )dW B t , P B 0 = pB > 0, 2We choose a quadratic penalty term for simplicity to penalize over- and underfeeding","support":"full"},{"assumption":"Fish mortality follows an exponential growth model with intrinsic mortality µ and additional mortality µF due to under‑ or overfeeding.","type":"model","stated":"explicit","quote":"We assume that the dynamics of hf,u t is given by the following exponential growth model assuming that there is no reproduction of the fish in the farm dhf,u t = \u0010 −µ −µF (ft −ut)2\u0011 hf,u t dt =: bh(t, hf,u t ; ft, ut)dt, hf,u 0 = h0 > 0, where µ > 0 is the intrinsic mortality rate of the fish and µF > 0 the mortality increase due to under- or overfeeding.","support":"full"},{"assumption":"A quadratic penalty term is used to penalize over‑ and underfeeding equally.","type":"model","stated":"explicit","quote":"We will assume that the price dynamics are both by two geometric Brownian motions with dynamics under a risk-neutral measure Q given by dP F t = rP F t dt + σF P F t dW F t =: bpF (t, P F t )dt + σpF (t, P F t )dW F t , P F 0 = pF > 0, dP B t = rP B t dt + σBP B t dW B t =: bpB(t, P B t )dt + σpB(t, P B t )dW B t , P B 0 = pB > 0, 2We choose a quadratic penalty term for simplicity to penalize over- and underfeeding","support":"full"},{"assumption":"The control space is discretized and the maximum over the discrete set of controls is taken.","type":"execution","stated":"explicit","quote":"In our numerical implementation, we will discretize the control space as well and take the maximum over the discrete set of controls.","support":"full"},{"assumption":"The neural network training uses a fuzzy boundary sampling with ε = 0.01 to improve stopping region accuracy.","type":"model","stated":"explicit","quote":"We chose ε = 0.01 for the fuzzy boundary sampling, which seems to perform best in our experiments. Results.","support":"full"}]},"datasets":{"status":"available","items":[{"name":"Python code and data sets for numerical experiments","provider":null,"kind":"synthetic","access":"public","period":null,"universe":null,"quote":"Keywords: Stochastic Optimal Control, Optimal Stopping, Deep Learning, PINN, Finite Differences, HJB, Numerics, Aquaculture Code availability: The Python code and data sets to produce the numerical experiments are available at https://github.com/kevinkamm/JointOptimalCtrlStopping_Aquaculture.","support":"full"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"synthetic"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R1","candidates":{"front_matter":"n/a","evidence_card":"no","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"none","findings":[{"finding":"Finite difference methods perform well in medium‑dimensional settings, serving as a benchmark.","kind":"performance","quote":"Numerical experiments demonstrate that while finite differences perform well in medium-dimensional settings, the PINN approach achieves comparable accuracy and is more scalable to higher dimensions where grid-based methods become infeasible.","support":"full"},{"finding":"The PINN approach achieves comparable accuracy to finite differences and scales to higher dimensions where grid‑based methods become infeasible.","kind":"performance","quote":"Numerical experiments demonstrate that while finite differences perform well in medium-dimensional settings, the PINN approach achieves comparable accuracy and is more scalable to higher dimensions where grid-based methods become infeasible.","support":"full"},{"finding":"Jointly optimizing feeding and harvesting decisions outperforms strategies that neglect either control or stopping.","kind":"theoretical","quote":"The results confirm that jointly optimizing feeding and harvesting decisions outperforms strategies that neglect either control or stopping.","support":"full"},{"finding":"The optimal stopping time from the joint problem is later than the optimal stopping time when using the optimal control alone.","kind":"theoretical","quote":"It is noteworthy that the optimal stopping time from the joint optimal control and stopping problem is later than the optimal stopping time when using the optimal control till time τ 1.","support":"full"},{"finding":"All three deep learning control approaches yield slightly lower mean farm values than the finite difference solution, suggesting possible issues with stopping time determination or control network refinement.","kind":"negative","quote":"However, all three approaches lead to slightly lower farm values than the finite difference approach, which suggests that the either the optimal stopping time is not determined correctly by the PINNs or the control network needs further refinement.","support":"full"}],"reported_performance":null,"out_of_sample":{"method":"none","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":24,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":true,"code_url":"https://github.com/kevinkamm/JointOptimalCtrlStopping_Aquaculture","note":null},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":2,"extracted_at":"2026-11-14T19:47:20Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}