Tritium Fuel Cycle in Fusion Reactors

• The Tritium Fuel Cycle (TFC) is an integrated set of functions that breed, extract, process, and manage tritium to sustain deuterium–tritium fusion.
• Current studies focus on dynamic, plant-wide models that balance key metrics such as breeding ratio, burn efficiency, and residence times across inner and outer cycles.
• Innovative approaches like multifunctional blankets and multiscale simulations aim to optimize tritium self-sufficiency while addressing capital and inventory challenges.

Searching arXiv for the cited TFC literature to ground the article in current research. to=arxiv_search.search 玩彩神争霸 ￣影音先锋={"query":"Tritium Fuel Cycle fusion tritium self-sufficiency surrogate TMAP8 tritium breeding ratio burn efficiency PathSim FESTIM 2026", "max_results": 10}let's see Searching arXiv for recent tritium fuel cycle papers. to=arxiv_search.search 六和彩 to=arxiv_search.search +天天中彩票 】!【json {"query":"Tritium Fuel Cycle fusion tritium self-sufficiency surrogate TMAP8 2026", "max_results": 10} to=arxiv_search.search code ＿天天json {"query":"(Yang et al., 21 Apr 2026)", "max_results": 5} The tritium fuel cycle (TFC) is the integrated set of breeding, extraction, processing, storage, fueling, exhaust, recovery, and accountancy functions required to sustain deuterium–tritium fusion when tritium must be produced onsite to offset burn, decay, and system losses. Its central constraints arise from tritium’s short half-life of 12.33 years, limited natural abundance, finite in-process residence times, and the coupling between plasma operation, breeder neutronics, material retention, and plant-wide losses. In contemporary fusion studies, the TFC is therefore treated not as an auxiliary subsystem but as a closed, dynamically evolving inventory network whose feasibility depends simultaneously on tritium breeding ratio, burn efficiency, extraction and processing performance, particle exhaust, component retention, and startup inventory (Yang et al., 21 Apr 2026).

1. System architecture and functional scope

In reactor studies, the TFC is commonly partitioned into an inner fuel cycle and an outer fuel cycle. In Tokamak Energy’s ST-E1 upgradable fusion pilot plant, the inner cycle comprises plasma fueling and exhaust through the Fusion Chamber Pump, Tokamak Exhaust Processing, isotope separation, isotope adjustment, and return to the plasma; the outer cycle comprises breeding and extraction from the blanket, the Tritium Extraction System, the helium loop, detritiation systems, the stack, and storage. The ST-E1 plant model explicitly represents 17 subsystems with residence times and loss coefficients, and treats tritium retained in plasma-facing components as a key inventory (Yang et al., 21 Apr 2026).

This partition is conceptually aligned with broader reactor analyses. Neutronics studies treat the breeding blanket as the source term that determines whether a reactor can be tritium self-sufficient and what extraction and processing throughput must be designed to support steady operation (Goel et al., 2023). Dynamic systems studies further emphasize that acceptable TFC behavior requires low inventories and low residence times across both inner and outer loops, because holdup directly enlarges startup requirements and decay losses (Delaporte-Mathurin et al., 20 Mar 2026). A plausible implication is that the TFC is best regarded as a plant-wide inventory control problem rather than merely a blanket-extraction problem.

The architectural importance of this distinction becomes sharper under detached plasma operation. Recent analysis argues that, in current TFC architectures, fuel puffing must contain tritium if Direct Internal Recycling is used, even though TFC models had assumed core fuelling to dominate. In that database, detached fuel puffing exceeds core fuelling by about an order of magnitude across present tokamaks and next-step stellarators, which materially changes pump sizing, isotopic balance, and inventory accounting (Meschini et al., 26 Jun 2026).

2. Governing balances, metrics, and self-sufficiency criteria

At plant scale, TFC modeling is usually organized around component-wise and aggregate mass balances. A representative subsystem equation is

$$\frac{dI_i}{dt} = \sum_{j\ne i}\frac{I_j}{\tau_j} - \left(1+\varepsilon_i\right)\frac{I_i}{\tau_i} - \lambda I_i + S_i,$$

where $I_i$ is subsystem inventory, $\tau_i$ its residence time, $\varepsilon_i$ a nonradioactive loss fraction, $$\lambda = \ln 2 / 12.33\,\mathrm{yr} = 1.78\times 10^{-9}\,\mathrm{s^{-1}}$$, and $S_i$ a source term. The corresponding plant-wide scalar balance is

$$\frac{dN}{dt} = P_{\text{breed}} + P_{\text{sup}} - U_{\text{plasma}} - L_{\text{sys}} - \lambda N.$$

These forms make explicit that breeding, external supply, plasma consumption, nonradioactive losses, and decay all enter on comparable bookkeeping footing (Yang et al., 21 Apr 2026).

The tritium breeding ratio is the basic neutronic metric,

$$\mathrm{TBR}=\frac{T_{\mathrm{produced}}}{T_{\mathrm{consumed}}},$$

but system studies emphasize that neutronic TBR alone is not sufficient. One explicit self-sufficiency condition is

$$\mathrm{TBR}\times \eta_{\mathrm{capture}}\times \eta_{\mathrm{extraction}}\times \eta_{\mathrm{processing}}\times (1-L)\ge 1+M,$$

where the $\eta$ factors represent effective efficiencies for neutron capture in the breeder, tritium extraction, and processing, $I_i$0 is the fractional loss, and $I_i$1 is a reserve margin. This formulation makes clear that $I_i$2 is necessary but not sufficient (Goel et al., 2023).

System studies instantiate these balances with plant-specific operating points. In the ST-E1 model, the values used are availability factor $I_i$3, tritium breeding ratio $I_i$4, tritium burn efficiency $I_i$5, and tritium burn rate $I_i$6 (Yang et al., 21 Apr 2026). In ARC-class 0D reproductions, the storage inventory initially depletes while bred tritium transits through the outer fuel cycle; around $I_i$7 days the system reaches a pseudo steady state in which storage begins to refill while other subsystem inventories settle (Delaporte-Mathurin et al., 20 Mar 2026). These dynamic results underscore that residence time, not only total inventory, controls practical self-sufficiency.

3. Breeding physics, blanket performance, and experimental validation

TFC closure begins with breeding. Blanket studies based on OpenMC show that breeder material, neutron spectrum, lithium enrichment, and neutron multiplication strongly alter TBR. In a submersion tokamak survey under a monoenergetic DT spectrum, reported single-material breeder-zone TBR values include 1.082 for FLiBe, 1.094 for Li$I_i$8O, 1.109 for pure Li, and 1.145 for Li$I_i$9Pb$\tau_i$0 at 90% $\tau_i$1Li, with an additional “PbLi100” case reaching 1.249. Under a DD spectrum, the corresponding TBRs are roughly halved, for example 0.569 for FLiBe and 0.570 for Li$\tau_i$2Pb$\tau_i$3 (Goel et al., 2023).

The underlying reaction physics distinguishes the roles of the lithium isotopes. The $\tau_i$4 channel is exothermic and remains dominant as neutrons moderate, whereas $\tau_i$5 contributes through higher-energy channels. The 2026 lithium-enrichment analysis states that natural lithium consists of 7.5% $\tau_i$6Li and 92.5% $\tau_i$7Li, that the thermal-neutron cross section of $\tau_i$8Li is $\tau_i$9 barn at room temperature, and that designs commonly assume $\varepsilon_i$0Li enrichment of 60–90%, sometimes as low as 30% for solid breeders with neutron multipliers, to meet TBR targets in toroidal devices. The same study notes that every centimeter of steel first wall reduces TBR by 7–10%, and cites required TBR $\varepsilon_i$1 with design target $\varepsilon_i$2 in EU DEMO assessments (Ward et al., 6 May 2026).

Direct breeding experiments remain rare, which makes the BABY experiment notable. BABY used a $\varepsilon_i$3 mL molten-salt setup under 14 MeV D–T neutron irradiation and directly measured

$\varepsilon_i$4

That value is small because only $\varepsilon_i$5 of source neutrons were incident on the salt, but the experiment is important because it measured tritium chemistry and release rather than inferring them only from neutronics. In the reported ClLiF runs at $\varepsilon_i$6 with He sweep, collected tritium was entirely in the insoluble fraction, i.e. HT/T$\varepsilon_i$7, with no measurable soluble tritium in the pre-oxidizer vials. The measured TBR was a factor $\varepsilon_i$8 lower than the OpenMC prediction, and the authors hypothesize significant permeation losses through the Inconel crucible walls (Delaporte-Mathurin et al., 2024). This suggests that extraction chemistry and permeation control can be as consequential as nominal breeding performance.

A related GEANT4 study of a LiD sphere driven by a thermal point source found that the probability that a source thermal neutron produces tritium via $\varepsilon_i$9 exceeds 99% at 1% $$\lambda = \ln 2 / 12.33\,\mathrm{yr} = 1.78\times 10^{-9}\,\mathrm{s^{-1}}$$0Li and 99.9% for $$\lambda = \ln 2 / 12.33\,\mathrm{yr} = 1.78\times 10^{-9}\,\mathrm{s^{-1}}$$1 $$\lambda = \ln 2 / 12.33\,\mathrm{yr} = 1.78\times 10^{-9}\,\mathrm{s^{-1}}$$2Li in a 20 cm sphere. The same work observed strong self-shielding and substantial triton follow-on reactions, including $$\lambda = \ln 2 / 12.33\,\mathrm{yr} = 1.78\times 10^{-9}\,\mathrm{s^{-1}}$$3 at roughly 16–19% across enrichments (Fattori et al., 12 Mar 2025). Although that geometry is not reactor-realistic, it illustrates how moderation, enrichment, and secondary neutron production are tightly coupled in breeder media.

4. Plasma-side burn, fueling composition, and exhaust asymmetry

The TFC is constrained not only by how much tritium is bred, but also by how efficiently injected tritium is burned before re-entering the recycle loop. In Boozer’s treatment of DT plasmas, the tritium burn-time is

$$\lambda = \ln 2 / 12.33\,\mathrm{yr} = 1.78\times 10^{-9}\,\mathrm{s^{-1}}$$4

and the per-pass tritium burn fraction is approximated by

$$\lambda = \ln 2 / 12.33\,\mathrm{yr} = 1.78\times 10^{-9}\,\mathrm{s^{-1}}$$5

showing explicitly that burn fraction increases with $$\lambda = \ln 2 / 12.33\,\mathrm{yr} = 1.78\times 10^{-9}\,\mathrm{s^{-1}}$$6 and scales as $$\lambda = \ln 2 / 12.33\,\mathrm{yr} = 1.78\times 10^{-9}\,\mathrm{s^{-1}}$$7. Boozer further gives the minimum confinement scaling

$$\lambda = \ln 2 / 12.33\,\mathrm{yr} = 1.78\times 10^{-9}\,\mathrm{s^{-1}}$$8

so that halving $$\lambda = \ln 2 / 12.33\,\mathrm{yr} = 1.78\times 10^{-9}\,\mathrm{s^{-1}}$$9 from 0.5 to 0.25 doubles $S_i$0 while increasing required $S_i$1 by only $S_i$2 (Boozer, 2021). For the TFC, the practical consequence is that lower tritium fraction can reduce circulating throughput and startup inventory even if plasma confinement requirements rise modestly.

Recent reactor studies push this idea further. In an ARC-like tokamak producing 482 MW of fusion power with unpolarized 51:49 D:T fuel, the reported minimum startup tritium inventory is 0.677 kg, with $S_i$3. By spin-polarizing half of the fuel and using a 57:43 D:T mix, the same 482 MW operating point yields $S_i$4 and $S_i$5 kg; fully spin-polarizing the fuel with a 61:39 mix gives $S_i$6 and $S_i$7 kg (Parisi et al., 2024). The technical point is not merely that lower tritium fraction helps, but that higher effective reactivity can offset the fusion-power penalty normally associated with tritium-lean operation.

A different lever is asymmetric transport and pumping. In an ARC-class plant at $S_i$8 MW, increasing the particle diffusivity ratio to $S_i$9 and decreasing the divertor pumping-speed ratio to $$\frac{dN}{dt} = P_{\text{breed}} + P_{\text{sup}} - U_{\text{plasma}} - L_{\text{sys}} - \lambda N.$$0 raises tritium burn efficiency from 0.026 to 0.29 at fixed power. In that example, $$\frac{dN}{dt} = P_{\text{breed}} + P_{\text{sup}} - U_{\text{plasma}} - L_{\text{sys}} - \lambda N.$$1 falls from $$\frac{dN}{dt} = P_{\text{breed}} + P_{\text{sup}} - U_{\text{plasma}} - L_{\text{sys}} - \lambda N.$$2 to $$\frac{dN}{dt} = P_{\text{breed}} + P_{\text{sup}} - U_{\text{plasma}} - L_{\text{sys}} - \lambda N.$$3, and total exhaust throughput falls from about 51.1 to 12.4 Pa·m$$\frac{dN}{dt} = P_{\text{breed}} + P_{\text{sup}} - U_{\text{plasma}} - L_{\text{sys}} - \lambda N.$$4/s (Parisi et al., 2024). This is a direct TFC benefit, because lower tritium throughput relaxes pumping, isotope separation, and temporary storage burdens.

These optimizations are complicated by detached exhaust operation. A 2026 analysis reports that the fuel puffing rate $$\frac{dN}{dt} = P_{\text{breed}} + P_{\text{sup}} - U_{\text{plasma}} - L_{\text{sys}} - \lambda N.$$5 exceeds the core fuelling rate $$\frac{dN}{dt} = P_{\text{breed}} + P_{\text{sup}} - U_{\text{plasma}} - L_{\text{sys}} - \lambda N.$$6 by about an order of magnitude, with multi-machine scalings

$$\frac{dN}{dt} = P_{\text{breed}} + P_{\text{sup}} - U_{\text{plasma}} - L_{\text{sys}} - \lambda N.$$7

and

$$\frac{dN}{dt} = P_{\text{breed}} + P_{\text{sup}} - U_{\text{plasma}} - L_{\text{sys}} - \lambda N.$$8

For the database average, $$\frac{dN}{dt} = P_{\text{breed}} + P_{\text{sup}} - U_{\text{plasma}} - L_{\text{sys}} - \lambda N.$$9, $$\mathrm{TBR}=\frac{T_{\mathrm{produced}}}{T_{\mathrm{consumed}}},$$0, and $$\mathrm{TBR}=\frac{T_{\mathrm{produced}}}{T_{\mathrm{consumed}}},$$1 (Meschini et al., 26 Jun 2026). This directly challenges the simplifying assumption that core fuelling dominates TFC flows. The same study concludes that realistic TFC requirements can be met with D-rich, T-lean puffing at the cost of about 10% lower fusion power, or with reduced fuel puffing and stronger impurity seeding, albeit with higher core contamination (Meschini et al., 26 Jun 2026).

5. Retention, permeation, and multiscale component modeling

Even with acceptable breeding and plasma burn, the TFC can fail if tritium is retained too long in plasma-facing components, structural alloys, coolants, or extraction hardware. In the TMAP8-based ST-E1 study, tritium transport in solids is modeled by Fickian diffusion coupled to trapping and release,

$$\mathrm{TBR}=\frac{T_{\mathrm{produced}}}{T_{\mathrm{consumed}}},$$2

with a representative single-trap kinetic term

$$\mathrm{TBR}=\frac{T_{\mathrm{produced}}}{T_{\mathrm{consumed}}},$$3

The modeled components are the divertor, blanket first wall, center column first wall, and vacuum vessel, with tungsten, V-4Cr-4Ti, and SS 316LN transport parameters specified explicitly. The coolant-side boundary condition is taken as $$\mathrm{TBR}=\frac{T_{\mathrm{produced}}}{T_{\mathrm{consumed}}},$$4, which the study describes as conservatively overestimating tritium transport to coolant (Yang et al., 21 Apr 2026).

A key methodological contribution of that work is the replacement of constant component residence times with Gaussian-process surrogates trained on TMAP8 simulations. The surrogate approximates the steady coolant-side release flux $$\mathrm{TBR}=\frac{T_{\mathrm{produced}}}{T_{\mathrm{consumed}}},$$5 and a two-parameter back-flux model,

$$\mathrm{TBR}=\frac{T_{\mathrm{produced}}}{T_{\mathrm{consumed}}},$$6

For V-4Cr-4Ti pipe configurations, the minimum reported RMSPE is 9.92% for $$\mathrm{TBR}=\frac{T_{\mathrm{produced}}}{T_{\mathrm{consumed}}},$$7, with RMSPE$$\mathrm{TBR}=\frac{T_{\mathrm{produced}}}{T_{\mathrm{consumed}}},$$8 and RMSPE$$\mathrm{TBR}=\frac{T_{\mathrm{produced}}}{T_{\mathrm{consumed}}},$$9; for tungsten pipe configurations, the corresponding values are 1.60%, 8.35%, and 13.68%. The reported computational-cost reduction is by a factor of $$\mathrm{TBR}\times \eta_{\mathrm{capture}}\times \eta_{\mathrm{extraction}}\times \eta_{\mathrm{processing}}\times (1-L)\ge 1+M,$$0, and the two-parameter model reproduces the observed “elbow” behavior in component inventories better than one-parameter residence-time models (Yang et al., 21 Apr 2026).

The physical consequences are operationally significant. Under normal pulsed operation, divertor inventories saturate after $$\mathrm{TBR}\times \eta_{\mathrm{capture}}\times \eta_{\mathrm{extraction}}\times \eta_{\mathrm{processing}}\times (1-L)\ge 1+M,$$1 s with quasi-steady periodicity. In bake-out, for a divertor with 8 mm W armor and 1 mm W pipe, more than 95% of retained tritium is released after $$\mathrm{TBR}\times \eta_{\mathrm{capture}}\times \eta_{\mathrm{extraction}}\times \eta_{\mathrm{processing}}\times (1-L)\ge 1+M,$$2 hours at 873 K, while a vacuum chamber pressure model with continuous pumping shows a minimum pressure of approximately $$\mathrm{TBR}\times \eta_{\mathrm{capture}}\times \eta_{\mathrm{extraction}}\times \eta_{\mathrm{processing}}\times (1-L)\ge 1+M,$$3 Pa at about the same time (Yang et al., 21 Apr 2026). These numbers show that recovery schedules and availability assumptions can be anchored in explicit release kinetics rather than heuristic emptying times.

Other open-source multiscale frameworks pursue the same objective. PathSim/PathView combines 0D residence-time blocks, a 1D bubble-column tritium extractor, and high-fidelity FESTIM finite-element transport models within a single dynamic simulation environment. In the demonstrated ARC-class reproduction, the outer-fuel-cycle transit delay is about 10 days before pseudo steady state. In the embedded 1D gas–liquid contactor, extraction efficiency rises from just above 2.5% at $$\mathrm{TBR}\times \eta_{\mathrm{capture}}\times \eta_{\mathrm{extraction}}\times \eta_{\mathrm{processing}}\times (1-L)\ge 1+M,$$4 m to more than 13% at $$\mathrm{TBR}\times \eta_{\mathrm{capture}}\times \eta_{\mathrm{extraction}}\times \eta_{\mathrm{processing}}\times (1-L)\ge 1+M,$$5 m under the reported scan, while a single 3 m column gives about 10% steady extraction and three 1 m columns in series about 9% (Delaporte-Mathurin et al., 20 Mar 2026). The broader implication is that physically grounded component models can materially change predicted inventories and bottlenecks relative to single-$$\mathrm{TBR}\times \eta_{\mathrm{capture}}\times \eta_{\mathrm{extraction}}\times \eta_{\mathrm{processing}}\times (1-L)\ge 1+M,$$6 descriptions.

6. Inventory economics, deployment constraints, and evolving design directions

The TFC is often described as a neutronics or process-engineering challenge, but recent work shows that it is equally a capital-inventory problem. The 2026 lithium-enrichment analysis argues that lithium is effectively one of the two elemental fuels for fusion because tritium must be bred in the blanket, and that the main economic difficulty is not annual consumption but the large stored $$\mathrm{TBR}\times \eta_{\mathrm{capture}}\times \eta_{\mathrm{extraction}}\times \eta_{\mathrm{processing}}\times (1-L)\ge 1+M,$$7Li inventory. Reported planning figures are roughly 50 tonnes of enriched $$\mathrm{TBR}\times \eta_{\mathrm{capture}}\times \eta_{\mathrm{extraction}}\times \eta_{\mathrm{processing}}\times (1-L)\ge 1+M,$$8Li in DEMO-scale blankets and about 100 tonnes of $$\mathrm{TBR}\times \eta_{\mathrm{capture}}\times \eta_{\mathrm{extraction}}\times \eta_{\mathrm{processing}}\times (1-L)\ge 1+M,$$9Li per GWth-scale reactor when blanket replacement and parallel preparation are included, while annual $\eta$0Li consumption is only about 100 kg per year per 1 GWth effective plant (Ward et al., 6 May 2026). The same study states that “possession costs more than consumption,” and identifies current mercury-based enrichment routes as too expensive, not scalable, environmentally risky, and entangled with controlled-substance regulation (Ward et al., 6 May 2026).

This changes how TFC design is framed. Lower inventories are no longer only a safety objective; they are also a capital-efficiency objective. That is one reason current TFC research places strong emphasis on lowering startup inventory, reducing in-process hold-up, shortening residence times, and minimizing conservative overestimates of plasma-facing-component retention (Yang et al., 21 Apr 2026). It also motivates interest in lower-enrichment or natural-lithium breeder concepts, although those concepts place more pressure on coverage, first-wall thickness, neutron multiplication, and processing performance (Ward et al., 6 May 2026).

Another emerging direction is the use of multifunctional blankets that preserve tritium self-sufficiency while performing an additional nuclear function. One example is an enriched $\eta$1Nd + $\eta$2Li channel in front of a FLiBe breeder. OpenMC scans of that concept report many design points with $\eta$3 and a substantial subset with $\eta$4 while producing more than 1000 kg/GWth-year of $\eta$5Pm; the nominal $\eta$6Pm yield reported in the blanket concept is about 1424 kg/GWth-year (Parisi, 18 May 2026). The relevance to TFC is not the radioisotope itself, but the demonstration that target materials can sometimes function as neutron multipliers rather than merely as parasitic absorbers.

Across these studies, the dominant trend is toward tighter coupling of plasma, edge, materials, and plant dynamics. Future improvements explicitly identified in current TFC modeling include parameter uncertainty quantification, multi-trap kinetics, irradiation-induced microstructural evolution, mixed-isotope surface chemistry, re-emission, more realistic convection and permeation boundary conditions, 2D and 3D component models, direct in-memory multiscale coupling, and extension of surrogate embedding to the full fuel cycle rather than only plasma-facing components (Yang et al., 21 Apr 2026). Joint optimization of core plasma, edge plasma, and TFC is increasingly treated as unavoidable rather than optional (Meschini et al., 26 Jun 2026).

A persistent misconception is that TFC feasibility can be certified by quoting a single favorable TBR number. The accumulated evidence suggests otherwise. Neutronic breeding, extraction efficiency, burn fraction, puffing and exhaust composition, materials retention, pump conductance, detritiation performance, enrichment logistics, and startup inventory must all close simultaneously. The modern TFC is therefore best understood as a multiscale, multiphysics, and multi-inventory closure problem whose limiting constraint may shift from blanket neutronics to divertor operation, from plasma composition to component trapping, or from isotope processing to lithium supply, depending on the reactor design point (Goel et al., 2023).