Cogeneration: Principles, Technologies, and Applications

• Cogeneration is the coordinated production of two or more useful outputs from a shared resource or process, including electricity and heat in CHP, freshwater and power in OTEC, and software fixes with bug-reproduction tests.
• Effective cogeneration requires integrated optimization of coupled outputs, storage, network constraints, uncertainty, and opportunity costs rather than independent scheduling of each product.
• Applications range from district heating, nuclear and renewable energy systems to dark-matter physics, automated software repair, and module theory, where shared states create efficiencies, constraints, or predictive relationships.

Cogeneration is the coordinated production or generation of two useful outputs from a common process, resource, or decision trajectory. In energy systems, it conventionally denotes combined heat and power (CHP), in which a fuel, nuclear source, or thermal gradient produces electricity and useful heat; analogous coupled-output structures occur in electricity–freshwater production, hydrogen–power systems, and thermochemical conversion. The term is also used for asymmetric dark-matter and baryon production, joint generation of software fixes and bug-reproduction tests, and a module-theoretic invariant concerning embeddings into finite direct sums. Across these contexts, cogeneration denotes coupling: outputs are not independently selectable, and their joint production creates efficiencies, constraints, allocation problems, or predictive relationships.

1. Energy-system foundations

In CHP, a generating unit simultaneously produces electrical power and useful heat. The electrical and thermal outputs generally cannot be selected independently because they share fuel input, conversion equipment, operating constraints, and, in some systems, network states. Cogeneration therefore differs from separate generation, in which electricity and heat are produced by independent assets.

The principal system-level benefit is recovery of thermal energy that would otherwise be rejected. A single local-generation decision can displace grid electricity and external heat, while a CHP plant can continue producing electricity when recovered heat is stored or dissipated. The economic value of cogeneration depends on electricity prices, heat demand, fuel costs, heat-recovery efficiency, storage, transmission constraints, and the temporal coincidence of the two loads.

A CHP unit may be represented by a feasible operating region in the electricity–heat plane. Extraction-condensing units generally possess a two-dimensional operating region: extracting more steam for heat can reduce electrical production. Back-pressure units have an approximately fixed heat-to-power relationship. One-degree-of-freedom units are determined by a common fuel input, whereas two-degree-of-freedom units can vary an additional control, such as steam-valve opening, to alter the heat-to-power ratio (Gonzalez-Castellanos et al., 2018).

The combined-production constraint can be represented in simplified form by fixed efficiencies. For a gas-based CHP unit,

$$P_{\rm CHP}=\rho \frac{\dot M\cdot {\rm HHV}}{3600},$$

and

$$H_{\rm CHP}=\eta \frac{\dot M\cdot {\rm HHV}}{3600}.$$

Consequently,

$H_{\rm CHP}=\frac{\eta}{\rho}P_{\rm CHP}.$

This fixed-ratio formulation is useful for distributed scheduling but does not represent the richer feasible regions of extraction-condensing and back-pressure plants (Schrage et al., 19 Jun 2025).

Cogeneration is consequently a multi-energy problem rather than an electricity-only dispatch problem. Heat demand, electricity demand, renewable output, prices, storage states, ramping limits, startup costs, network congestion, and temperature requirements may all interact. Sequentially satisfying heat demand and then dispatching electricity can be economically inferior to integrated optimization.

2. Cogeneration technologies and applications

Combined heat and power

CHP systems include gas turbines with heat recovery, combined-cycle gas turbines (CCGTs), gas engines, fuel cells, coal and biomass plants, nuclear steam cycles, and Stirling engines. Their useful outputs may be electricity and district heat, process heat, cooling, or other thermal services.

CHP-SOFC systems for commercial buildings can be optimized jointly with batteries, boilers, water-tank storage, utility purchases, and thermal storage. Long-horizon planning distinguishes first-stage equipment investment from second-stage hourly operation. A two-level mixed-integer approach replaces millions of hourly binary decisions with daily on/off profiles and aggregated constraints while retaining a feasibility mapping from profile decisions to fine-scale schedules (Lin et al., 2015).

District-heating applications require explicit consideration of the heat network. Heat is transported through supply and return pipelines with thermal inertia and delayed temperature propagation, whereas electricity requires rapid balancing. An asynchronous dispatch formulation therefore uses shorter electricity intervals and longer heat intervals. Electricity may be priced through locational marginal prices, while heat is decomposed into a heat-energy marginal price and heat-grade marginal prices associated with supply- and return-side temperature requirements (Yi et al., 2022).

Thermal storage and sector coupling

Thermal storage decouples heat production from heat delivery. It permits CHP electricity production when electricity is valuable or when transmission conditions are favorable, while stored heat serves later thermal demand. In network-constrained CHP scheduling, thermal storage, two-degree-of-freedom CHP operation, and DC power-flow constraints must be optimized jointly. Removing thermal storage increased operating cost by approximately 1.3% in one reported test, while removing network constraints reduced cost by approximately 3.3% but produced an overloaded and physically infeasible schedule (Gonzalez-Castellanos et al., 2018).

Sector-coupled district-heating studies compare CHP with heat pumps, electric boilers, and storage. When fossil fuels are permitted, coal CHP, large heat pumps, and pit heat storage dominate the reported cost-optimal systems. When fossil fuels are excluded, heat pumps and storage become dominant, and the need for heat storage more than doubles because heat production becomes more closely coupled to electricity availability (Dahl et al., 2018).

Online scheduling is required when renewable output, electricity demand, heat demand, and grid prices are unknown. The CHASE algorithm—Competitive Heuristic Algorithm for Scheduling Energy-generation—uses a clipped cumulative gain process to decide when the evidence for operating a CHP unit has recovered its startup cost. For a fast-response single-generator model,

${\sf CR}(\textsf{CHASE}_s)\leq 3-2\alpha<3,$

where

$\alpha=\frac{c_o+c_m/L}{P_{\max}+\eta c_g}.$

An improved version combines CHASE with the external-only strategy and has guarantee

$${\sf CR}(\textsf{CHASE}_{s+}) \leq \min\left\{3-2\alpha,\frac{1}{\alpha}\right\}.$$

Under the stated assumptions, this is asymptotically optimal among deterministic online algorithms (Lu et al., 2012).

Nuclear cogeneration

Nuclear cogeneration varies the electricity–heat split while maintaining approximately constant reactor power. In the modeled nuclear CHP formulation,

$P^N_{e,t}+C_vP^N_{h,t}=P^N,$

where $C_v=0.3$. Steam extraction reduces electrical production while increasing useful heat. Heat storage then allows heat to be produced during periods of low electrical demand and delivered later, permitting greater electrical output during high-demand periods (Li et al., 2022).

The nuclear model is a scheduling abstraction rather than a complete reactor or turbine design. It excludes detailed steam-extraction pressure constraints, turbine off-design behavior, district-heating hydraulics, nuclear-renovation capital cost, and nuclear safety margins. In the reported North China case, nuclear cogeneration reduced modeled total operating cost from ¥2.212 million for electricity-only nuclear operation to ¥1.692 million and reduced emissions from 2,224 t to 1,207 t.

Cogeneration beyond heat

Open-cycle ocean thermal energy conversion (OC-OTEC) cogenerates electricity and freshwater. Warm seawater flashes into low-pressure steam, the steam drives a turbine, and turbine exhaust is condensed by cold deep seawater. The condensate becomes freshwater. Thus, steam production jointly determines electrical output and freshwater production, while pumping and condensation impose parasitic costs (Fu et al., 26 Jul 2026).

Fusion cogeneration can use waste heat for desalination, direct air capture, district heating, process heat, or hydrogen production. Its economic rationale is to spread capital and fixed costs over electricity and a second product. The modeled opportunity-cost relation for extracted heat is

$1~{\rm MWh_{th}}\rightarrow 0.15~{\rm MWh_e}$

of lost electricity production. At a fusion electricity cost of $50/MWh$_e$, modeled integration reduced effective fusion cost by 27% for multistage flash desalination and 35% for multi-effect distillation. These results depend on local offtake, temperature compatibility, infrastructure, and coproduct valuation (Handley et al., 2021).

3. Optimization, control, and market design

Cogeneration scheduling combines unit commitment, economic dispatch, storage management, network constraints, and multi-energy balances. The resulting models may be linear, mixed-integer linear, nonlinear, stochastic, robust, or dynamic-programming formulations.

A general network-constrained CHP unit-commitment model includes binary commitment and startup variables, minimum online times, ramp limits, thermal storage, district-heating zones, and lossless DC power flow. Nonlinear CHP performance curves are approximated by one- and two-dimensional piecewise-linear formulations. The model couples fuel consumption, electrical injection, heat delivery, storage operation, transmission flows, and commitment feasibility (Gonzalez-Castellanos et al., 2018).

Dynamic CCGT dispatch requires process memory because thermal output responds more slowly than electrical output. A fourth-order autoregressive model with delayed gas-input terms represents the heat process:

$$H_{\rm CHP}=\eta \frac{\dot M\cdot {\rm HHV}}{3600}.$$0

The state is augmented with historical thermal variables, allowing the problem to be written as a finite-horizon Markov decision process. Value-function-approximation approximate dynamic programming uses a piecewise-linear value function for the post-decision CCGT heat state. SPAR preserves monotonicity of the value-function slopes and thereby maintains convexity of the approximation. The reported method converged in fewer than 40 iterations and achieved an approximately 5% cost reduction relative to myopic and model-predictive-control policies (Lin et al., 2021).

Long-horizon investment problems can contain more than one million second-stage binary variables. Variable coarsening replaces hourly binary trajectories with a finite set of daily profiles, while constraint coarsening aggregates repeated constraints. Violated-constraint generation then restores feasibility of the semi-coarse model in finitely many steps. A feasible semi-coarse solution maps to a feasible fine-scale solution, and for minimization,

$$H_{\rm CHP}=\eta \frac{\dot M\cdot {\rm HHV}}{3600}.$$1

The method is a tightening rather than a relaxation of the original model; its accuracy depends on the representativeness of the profile library (Lin et al., 2015).

Market design must account for the nonhomogeneity and temporal dynamics of heat. In asynchronous CHP markets, the heat price is decomposed into heat-energy and temperature-grade components. The latter are nonzero when temperature-quality constraints bind. A CHP generator’s electricity price includes marginal electricity-production cost and an energy-coupling or opportunity-cost component; its heat price may depend on several electricity intervals contained within one heat interval (Yi et al., 2022).

Distributed scheduling addresses the case in which CHP units are independently owned and their technical and economic data are private. A gossiping and local-search algorithm allows agents to exchange schedules and aggregate consequences rather than complete plant models. Private objectives reward electricity and heat contributions while penalizing deviation from collective targets. In the reported gas-based CHP scenario, fixed-ratio CHP without storage achieved an aggregate fulfillment rate slightly below 70%, whereas high-penalty storage cases generally achieved fulfillment around or above 95% (Schrage et al., 19 Jun 2025).

4. Thermodynamic and materials contexts

Cogeneration is also relevant to materials whose decomposition produces multiple useful or undesirable products. In lithium boron nitride hydride, $$H_{\rm CHP}=\eta \frac{\dot M\cdot {\rm HHV}}{3600}.$$2, decomposition produces hydrogen and ammonia through coupled bond rearrangement, charged-defect generation, and bulk mass transport. A decomposition pathway is

$$H_{\rm CHP}=\eta \frac{\dot M\cdot {\rm HHV}}{3600}.$$3

The study identifies negatively charged hydrogen vacancies, $$H_{\rm CHP}=\eta \frac{\dot M\cdot {\rm HHV}}{3600}.$$4, as chemically decisive defects. Their migration breaks N–H bonds and produces $$H_{\rm CHP}=\eta \frac{\dot M\cdot {\rm HHV}}{3600}.$$5-like environments, while proton transport toward the surface enables formation of $$H_{\rm CHP}=\eta \frac{\dot M\cdot {\rm HHV}}{3600}.$$6. Lithium Frenkel disorder supplies highly mobile lithium vacancies and interstitials that help compensate slower hydrogen-, boron-, and nitrogen-containing defects (Hoang et al., 2014).

Stirling engines are externally heated heat engines that can be used for solar thermal power, waste heat recovery, space nuclear power, and distributed CHP. Conventional Stirling engines use a regenerator, but the proposed O-type loop engine eliminates it and uses a one-way loop with adiabatic compression and expansion. The proposed 8-type engine retains a regenerator but relocates it within the loop.

For an ideal Stirling cycle,

$$H_{\rm CHP}=\eta \frac{\dot M\cdot {\rm HHV}}{3600}.$$7

The O-type simulations reported efficiencies of 36.7%, 42.0%, and 44.4%, with simulated powers of 9.95, 7.77, and 6.25 kW under GPU-3-scale conditions. These are model-based results rather than experimental validation. Valve losses, leakage, mechanical friction, transient regenerator behavior, and dead volume are incompletely represented or omitted. The principal CHP relevance is that cooler heat could be recovered for domestic hot water, district heating, process preheating, drying, or absorption cooling (Deng, 1 Apr 2025).

These examples show that cogeneration is not necessarily synonymous with high efficiency. It may instead denote coupled formation of products whose quantities are constrained by shared transport, reaction pathways, defects, or heat flows. A coproduct can be economically valuable, environmentally problematic, or both.

5. Non-energy uses of the term

Asymmetric dark-matter cogeneration

In unified particle-physics models, cogeneration denotes the common origin and correlated redistribution of baryonic and dark matter asymmetries. Grand unified groups larger than $$H_{\rm CHP}=\eta \frac{\dot M\cdot {\rm HHV}}{3600}.$$8, such as $$H_{\rm CHP}=\eta \frac{\dot M\cdot {\rm HHV}}{3600}.$$9 or $H_{\rm CHP}=\frac{\eta}{\rho}P_{\rm CHP}.$0, naturally contain Standard-Model-singlet fermions that can serve as dark-sector states. An accidental global symmetry combined with a broken gauge generator can leave an unbroken stabilizing charge. In an $H_{\rm CHP}=\frac{\eta}{\rho}P_{\rm CHP}.$1 construction,

$H_{\rm CHP}=\frac{\eta}{\rho}P_{\rm CHP}.$2

A primordial asymmetry in a unified global charge is redistributed among quarks, leptons, and Standard-Model-singlet fermions. The singlet sector survives as asymmetric dark matter, while the ordinary sector produces the baryon asymmetry (Barr, 2011).

A related mechanism introduces a new non-Abelian gauge group $H_{\rm CHP}=\frac{\eta}{\rho}P_{\rm CHP}.$3. Its sphalerons violate $H_{\rm CHP}=\frac{\eta}{\rho}P_{\rm CHP}.$4, $H_{\rm CHP}=\frac{\eta}{\rho}P_{\rm CHP}.$5, and $H_{\rm CHP}=\frac{\eta}{\rho}P_{\rm CHP}.$6 while satisfying

$H_{\rm CHP}=\frac{\eta}{\rho}P_{\rm CHP}.$7

Electroweak sphalerons conserve $H_{\rm CHP}=\frac{\eta}{\rho}P_{\rm CHP}.$8 and $H_{\rm CHP}=\frac{\eta}{\rho}P_{\rm CHP}.$9. Chemical-equilibrium conditions then determine ratios such as ${\sf CR}(\textsf{CHASE}_s)\leq 3-2\alpha<3,$0 and ${\sf CR}(\textsf{CHASE}_s)\leq 3-2\alpha<3,$1. In the minimal model, the reported result is approximately

${\sf CR}(\textsf{CHASE}_s)\leq 3-2\alpha<3,$2

Combining this number-density ratio with

${\sf CR}(\textsf{CHASE}_s)\leq 3-2\alpha<3,$3

gives a dark-matter mass near ${\sf CR}(\textsf{CHASE}_s)\leq 3-2\alpha<3,$4 GeV. Here cogeneration does not mean literal conversion of baryons into dark-matter particles. It denotes common primordial origin, asymmetry sharing through sphalerons, and a resulting abundance relation (Barr et al., 2013).

Software-repair cogeneration

In agentic automated program repair, cogeneration means producing a source-code fix and a bug-reproduction test (BRT) in the same trajectory and returned patch. A BRT is a test that fails in the presence of the bug and passes once the bug is fixed.

The approach differs from separate pipelines that generate a test and a fix independently or use a temporary test only for validation. The shared trajectory preserves root-cause analysis, repository exploration, source edits, test edits, execution feedback, and the final patch. The evaluated strategies are BRT-only, fix-only, Freeform, test-driven development (TDD), and test-last development (TLD).

On 120 human-reported Google bugs, cogeneration produced plausible BRTs for at least as many bugs as a dedicated BRT-generation agent and plausible fixes for at least as many bugs as a fix-only agent. Freeform cogeneration achieved the highest joint success for patches containing both a plausible fix and a plausible BRT. TDD was strongest in converting candidate BRTs into plausible outcomes, while TLD performed best for candidate-BRT generation conditional on a plausible fix (Cheng et al., 27 Jan 2026).

The principal failure modes are omission of the BRT from the final patch, debugging loops and step exhaustion, fix–BRT interference, test-induced overfitting, and tool friction in locating executable test targets. A ranked test-aware patch selector increased the reported precision and recall for patches containing both a plausible fix and plausible BRT from 0.08/0.57 under the default selector to 0.16/0.71.

Module-theoretic cogeneration

In commutative algebra, the number of cogenerators of a Noetherian ${\sf CR}(\textsf{CHASE}_s)\leq 3-2\alpha<3,$5-module ${\sf CR}(\textsf{CHASE}_s)\leq 3-2\alpha<3,$6 with respect to an ${\sf CR}(\textsf{CHASE}_s)\leq 3-2\alpha<3,$7-module ${\sf CR}(\textsf{CHASE}_s)\leq 3-2\alpha<3,$8 is

${\sf CR}(\textsf{CHASE}_s)\leq 3-2\alpha<3,$9

It is the least number of copies of $\alpha=\frac{c_o+c_m/L}{P_{\max}+\eta c_g}.$0 into whose direct sum $\alpha=\frac{c_o+c_m/L}{P_{\max}+\eta c_g}.$1 embeds. The value is $\alpha=\frac{c_o+c_m/L}{P_{\max}+\eta c_g}.$2 if no finite embedding exists, and depends on $\alpha=\frac{c_o+c_m/L}{P_{\max}+\eta c_g}.$3, $\alpha=\frac{c_o+c_m/L}{P_{\max}+\eta c_g}.$4, and any restrictions imposed on the allowed homomorphisms.

If $\alpha=\frac{c_o+c_m/L}{P_{\max}+\eta c_g}.$5 is commutative, $\alpha=\frac{c_o+c_m/L}{P_{\max}+\eta c_g}.$6 is arbitrary, and $\alpha=\frac{c_o+c_m/L}{P_{\max}+\eta c_g}.$7 is Noetherian, then

$\alpha=\frac{c_o+c_m/L}{P_{\max}+\eta c_g}.$8

The supremum can be restricted to maximal members of $\alpha=\frac{c_o+c_m/L}{P_{\max}+\eta c_g}.$9. No finite-generation assumption on $${\sf CR}(\textsf{CHASE}_{s+}) \leq \min\left\{3-2\alpha,\frac{1}{\alpha}\right\}.$$0 or dimension bound on $${\sf CR}(\textsf{CHASE}_{s+}) \leq \min\left\{3-2\alpha,\frac{1}{\alpha}\right\}.$$1 is required. The Noetherian hypothesis on $${\sf CR}(\textsf{CHASE}_{s+}) \leq \min\left\{3-2\alpha,\frac{1}{\alpha}\right\}.$$2 is essential because injectivity can be tested through kernels, associated primes, and localized socles (Baidya et al., 2021).

The same framework has restricted versions for a submodule $${\sf CR}(\textsf{CHASE}_{s+}) \leq \min\left\{3-2\alpha,\frac{1}{\alpha}\right\}.$$3, enhanced simultaneous localization statements, quotient and trivial-intersection formulations, and a graded analogue. The graded theorem requires a common homogeneous degree, homogeneous localization, and sufficiently large degree-zero residue fields; it does not state a full graded cogenerator-number localization formula.

6. Common properties, limitations, and design principles

Cogeneration problems share several structural properties despite their different domains. First, outputs are coupled through a common state or resource: fuel and heat in CHP, steam and freshwater in OTEC, a primordial charge in dark-matter models, source edits and tests in APR, or a common module embedding in algebra. Second, an apparently favorable marginal decision can impose an opportunity cost on the other output. Extracting steam reduces electricity, generating fixed-ratio CHP electricity may overproduce heat, and adding a BRT may alter patch selection. Third, storage, buffering, or state augmentation can relax instantaneous coupling. Thermal storage, electrical storage, freshwater sinks, augmented CCGT states, and persistent test artifacts all serve this role in their respective settings.

The principal limitations are domain-specific. Energy models may omit detailed thermodynamics, network hydraulics, unit commitment, outages, price feedback, uncertainty, or capital costs. Cogeneration economics can reverse when coproduct markets are absent, infrastructure is expensive, temperatures are incompatible, or electricity opportunity costs exceed coproduct value. In fusion applications, desalination is presented as a stronger early opportunity than district heating, process heat, or hydrogen because it has a direct market and can use low-grade heat (Handley et al., 2021). In district heating, fossil-free systems are more exposed to electricity-price variation and require substantially more storage (Dahl et al., 2018).

Algorithmic guarantees also depend on assumptions. CHASE’s competitive optimality applies to the fast-response model, while slow-response generators receive only a multiplicative upper bound. Profile-based MILP coarsening guarantees feasibility relative to the restricted profile model, not automatic exactness relative to the original model. Distributed gossiping and local search produce globally near-optimal solutions empirically but do not provide a formal approximation guarantee. Nuclear, OTEC, and Stirling results are model-based and require engineering validation before being treated as plant-level performance claims.

A general design principle follows: cogeneration should be modeled at the level at which coupling actually occurs. Static heat-to-power ratios are insufficient when transient thermal states matter; synchronous dispatch is insufficient when electricity and heat evolve at different time scales; independent market prices are insufficient when heat grade and network delay affect value; and separate generation pipelines are insufficient when two software artifacts must remain semantically coherent. The most informative analyses therefore combine physical coupling, intertemporal state variables, coproduct valuation, storage or buffering, uncertainty treatment, and explicit accounting of opportunity costs.