An improved method to evaluate the “Joint Oxyde-Gaine” formation in (U,Pu)O 2 irradiated fuels using the GERMINAL V2 code coupled to Calphad thermodynamic computations

. In this work, two different thermodynamic softwares, A NGE using the TBASE database, and O PEN C ALPHAD using the TAF-ID (Thermodynamics of Advanced Fuels – International Database), have been integrated into the G ERMINAL V2 fuel performance code (of the P LEIADES platform) in order to evaluate the chemical state of (U,Pu)O 2 fuel and fission products in sodium cooled fast reactors. A model to calculate the composition and the thickness of the “Joint-Oxyde Gaine” (JOG) fission product layer in the fuel-clad gap has been developed. Five fuel pins with a final burnup ranging between 3.8 and 13.4% FIMA (Fissions per Initial Metal Atom) have been simulated, and the calculated width of the fission product layer have been compared with post irradiation examinations. The two different thermodynamic softwares have been compared in terms of computation time and predicted fuel-to-clad gap chemistry. The main elements and phases encountered in the fission productlayer have been identified, and the impact of the changing oxygen potential has been explored.


Introduction
When oxide fuel pins are irradiated in a fast breeder reactor (FBR), it has been observed that certain fission products (FP) migrate down the temperature gradient and form a layer between the fuel and the stainless steel cladding. This layer of fission product compounds is commonly called JOG (for "Joint Oxyde-Gaine" in French) [1], and the fact that its presence affects both heat transfer and corrosion rates [2,3] has warranted attempts to understand and predict its formation. Internal corrosion weakens the cladding and increases the probability of fuel failure, especially at high burnup [4]. As described in reference [1], JOG was first proposed as an explanation for an inconsistency found in these PIE: if the large fuel-toclad gap that appears at high burnup had only been filled with gas, it would certainly have caused fuel melting (due to the poor heat conductivity of the gas). However, if the gap was to be partly filled with fission product compounds with higher thermal conductivity compared with the gas plenum, the maximum fuel temperature would fall below the melting point of the fuel. These FP would need to be * e-mail: karlsam@kth.se * * e-mail: jean-christophe.dumas@cea.fr transported through the fuel towards the periphery due to the effect of the thermal gradient. This could later be confirmed by experimental observations and measurements. Inoue et al. [2] concludes, after studying irradiated MOX fuel pins in the fast neutron JOYO reactor, that JOG evolution is dependent on burnup, temperature, initial fuel microstructure, and fission gas release. These variables are of course not independent of one another. The exact composition of this JOG layer has never been determined, and the term itself can be seen as an umbrella term for any FP that has deposited in the fuel-to-clad gap. While it is believed to be rich in Mo and Cs oxides, the distribution of phases is likely heterogeneous [5].
The GERMINAL V2 [6] fuel performance code, developed by the CEA (French Alternative Energies and Atomic Energy Commission) within the PLEIADES simulation platform [7], is used to simulate the thermo-mechanical and the physico-chemical behavior of (U,Pu)O 2 fuel during irradiation in a fast neutron spectrum. In its current version, the prediction of JOG thickness is described by a model involving the amount of volatile FP (mainly caesium) based on a correlation to the kinetics of the release of the stable fission gases [6,8]. A threshold in burnup as well as a thermal activation term are respectively used to reproduce the post-irradiation observations showing no JOG formation at low burn-up and at low linear power.
In the past years, several groups have worked on implementing thermodynamic calculations inside fuel performance codes in order to improve predictive abilities. Baurens et al. [9] and later Konarski et al. [10] have coupled ANGE together with the ALCYONE (also in the PLEIADES simulation platform) in order to simulate, respectively, stress corrosion cracking and oxygen thermodiffusion. Simunovic et al. [11] have coupled THERMOCHIMICA [12] to the mass and heat transport models of the BISON [13] fuel performance code. Both these examples have been focusing on the simulation of light water reactor fuel. Uwaba et al. [14] at the Japan Atomic Energy Agency have recently coupled the MLCYONE [15,16] caesium behavior simulation code to the CEDAR [17] fast reactor fuel performance code. This has allowed for predictions on the JOG chemistry and geometry.
In this work, two different thermodynamic softwares, both based on the Calphad method [18,19], have been integrated into GERMINAL V2 in order to calculate the chemical state of the fuel. Full in-pile simulations have been performed on five fuel pins with different burnup ranging between 3.8 and 13.4 %FIMA burnup. JOG thickness has then been estimated on the basis of the predicted chemical composition of the gap and the known molar volumes of the involved phases. The two different thermodynamic solvers, ANGE [20] and OPENCALPHAD [21,22], and their respective databases have been compared in terms time and prediction of JOG thickness and its composition. When available, results have been compared with experimental results. In a separate set of stand-alone calculations, the thermodynamic codes have also been evaluated and compared in terms of computational cost.

Experiments
The operation of the Phénix reactor between 1973 and 2010 associated with numerous post irradiation examinations (PIE) by the CEA resulted in an extensive database of fuel pin behavior under irradiation in a fast neutron spectrum.
In this work, five fuel pins from the Phénix fast breeder reactor irradiated to different burnup (3.8, 7.0, 9.0, 11.2, and 13.4 %FIMA at the maximum flux plane) have been  [1], the burnup values refer to the local burnup at which the JOG was measured. For reference [8], the burnup refers to the maximum burnup reached in the fuel pin. simulated with GERMINAL V2. More information concerning the fuel pins can be found in Table 1. Previous PIE have given experimental values for measured JOG thickness of fuel pins irradiated in the Phénix reactor, see Figure 1. It should be noted that these experimental values are measurements of the fuel-to-clad gap, and are only assumed to be equal to JOG thickness for reasons mentioned above. The fuel pins were generating between 350 and 400 W/cm and the highest temperatures reached at the peak power nodes were, depending on the fuel pin between 2200 and 2400 K (based on the GERMINAL V2 simulations).

Thermodynamic software and databases
For the calculations, two different software-database combinations have been used and compared: -ANGE (Advanced Numeric Gibbs Energy minimizer) [20], co-developed by CEA and EDF (Electricité de France), based on the SOLGASMIX [23][24][25] software.
-OPENCALPHAD open source software [21,22] using the TAF-ID [26,27] database which is the result of the merging of several databases (including TBASE).
The main advantage of OPENCALPHAD is its ability to utilize better thermodynamic models in the newer (and still growing) TAF-ID database, but comes at the price of increased computational time as will be discussed in Section 3.2. The purpose of the TAF-ID project, coordinated by the Organization for Economic Co-operative Development Nuclear Energy Agency (OECD/NEA), is to provide a comprehensive thermodynamic database on nuclear fuel materials to perform a wide range of thermodynamic calculations for different applications of nuclear reactors. This database can be seen as a synthesis of different databases (including TBASE) developed independently in different countries and has been progressively extended for five years by introducing either models coming from research and/or databases of the participants of the project, or coming from the open literature. It has been decided to adopt a full Calphad modeling approach for this database in order to provide both phase diagram and thermodynamic data calculations. Here, the description of the (U,Pu,Ln)O 2±x phase is based on the Compound Energy Formalism (CEF) [28] model of Guéneau et al. [29]. This phase, made up by three sublattices, can be written as (Ba For the liquid phases, the two sublattice ionic model [30,31]  The TAF-ID describes the main metallic phase (also called "white phase") encountered in examinations of spent fuel [32] as an HCP structure with two sublattices: (Ba, Ce, Cs, Gd, Mo, Pd, Pu, Ru, U, Zr) 1 (O, Va) 0.5 .
One of the main oxide phases encountered is the perovskite structured BaZrO 3 [32]. This phase is sometimes referred to as the "gray phase", and in the TAF-ID it is expressed (within the CEF) as: 3 . Other fission product phases such as CsI, Cs 2 Te, Cs 2 MoO 4 , and BaMoO 4 are treated as stoichiometric compounds, which means that their compositions are fixed and their Gibbs energy functions depend only on temperature and pressure. Up to now, the TAF-ID can be directly used with THERMO-CALC [33] or OPENCALPHAD codes and a thermodynamic database converter has recently been developed in order to be able to use it with FACTSAGE (in CHEMSAGE format). Parts of the TAF-ID was converted to this format for use in the THERMOCHIMICA-BISON coupling mentioned in the introduction [34].
TBASE [35,36], on its side, is a thermodynamic database elaborated at ECN Petten (Netherlands) in the 1990's which contains mainly stoichiometric compounds from reference [37]. This is the case for most solid phases, and all liquid phases. The two notable exceptions concern the fluorite fuel phase and the metallic "white phase". The thermodynamic description of the fuel phase is represented by the variable stoichiometry species model of Lindemer & Besmann [38][39][40]. It can be written as a solution between the following constituents: The metallic phase is defined as an ideal solution between Mo, Ru, and Pd. It can be noted that in all definitions above, only the elements used in this work has been included in the expression of the phases. Moreover, the TAF-ID, unlike the TBASE description, includes heat capacity data for most phases. While heat capacity data is not required to perform the calculations presented in this work, a future improvement of the GERMINAL V2 code could be to couple the results of the thermodynamic model to the heat transfer model. If this were to be done, the heat capacity data for the involved phases would be necessary.

Computation times
A complete fuel pin simulation with GERMINAL V2 can require millions of equilibrium calculations, implying a huge computational cost associated to the thermodynamic software.
A number of test equilibrium calculations were performed by OPENCALPHAD and ANGE over a temperature range of 500-2500 K, with a composition corresponding to a (U 0.78 ,Pu 0.22 )O 1.975 fuel pin irradiated to 13.4 %FIMA burnup. Here, in order to facilitate the performance evaluation, both solvers were used in their stand-alone mode, i.e., not coupled to GERMINAL V2. The composition was taken from previous calculations performed by the ERANSO code [41] using nuclear data from the JEFF-3.1 [42] project library. As can be seen in Table 2, 15 element groups representative of the FP, the actinides, and the oxygen were considered in the equilibria.

GERMINAL V2 fuel performance code
The GERMINAL V2 fuel performance code is being developed by the CEA, and works under the PLEIADES simulation platform [7]. The code implements a 1 1 /2-D approach for the discretization of the fuel pin geometry. This means that the pin is divided into axial cells, and each axial cell is then divided into radial cells by assuming cylindrical symmetry. Here, one radial cell may represent either the fuel itself, the gap, or the cladding. One simulation is then divided into different timesteps.
In reality, the relevant physical phenomena, e.g. swelling, temperature distribution, cracking, actinide and oxygen redistribution etc. are all coupled to one another. In order to describe all these phenomena, GERMINAL V2 uses a scheme of nested convergence loops. In practice this means that one timestep consists of one loop over the axial cells, and within the evaluation of each axial cell another convergence loop solves the necessary equations within each radial cell. The modeling of the thermal and mechanical behavior is treated by the finite element solver CASTEM2000 [43]. The description of clad mechanical behavior (irradiation and thermal-activated creep, irradiation-induced swelling, plasticity in transient conditions) allows to account for clad deformation when evaluating the fuel-to-clad gap width. The chemical composition at each radial node of the fuel is obtained from a simplified neutronic module implementing an isolated resolution of the Bateman equations. The coupling of GERMINAL V2 with a thermodynamic software (ANGE or OPENCALPHAD) elaborated in the frame of this work allows the thermodynamic equilibrium calculation at each node of the fuel pellet, and based on the amount of gas and liquid that is found, along with the fission gas release fraction, a corresponding amount is released into the fuel-to-clad gap. The volatile release fraction is taken to be equal to that of the inert fission gases, which is an assumption with some experimental justification [44]. The model used to calculate the fission gas release is described in reference [6].
This kind of thermodynamic calculation is used to find the equilibrium state of the chemical system defined by its composition, temperature, and pressure. It does not give information regarding the kinetics of the chemical reactions. For the calculations performed inside the fuel performance code in this work, the equilibrium state is assumed to occur instantaneously due to the high temperature.
Currently, the fuel equilibrium calculations involve 15 representative elements listed in Table 2, where the elements that have been regarded as identical to its representative element are shown in the parenthesis. For example, Ba (+Sr) means that the molar amount of Sr has been added to the amount of its representative element Ba.
-Ba and Sr were grouped together since they are both believed to be (mainly) found in the Ba(Zr,U)O 3 and Sr(Zr,U)O 3 [45,46]. Their binary phase diagram shows a large degree of mutual solubility [47]. -Ce and Pr are both expected to be found in solution with the fuel matrix [45]. -Cs and Rb are both alkali metals and are expected to behave similarly [45].  [50,51], and are represented by Te since it is the more abundant and well studied element of the two [32]. -Zr and Nb are commonly grouped together [45]. While this has been done in this work as well, it is of little consequence due to the low fission yield of Nb.
The decision to make groups of representative elements was made due to the demand to keep computational cost, complexity, and failure rates sufficiently low while still describing a chemical system as close as possible to the real one. In addition, elements needed to be grouped when they were not described in both databases, since the comparison required that the same input was used in all cases.
Among the main parameters for fuel chemistry simulation is the radial oxygen redistribution, and in GERMINAL V2 it is based on the work of Aitken [52]. At each axial cell, the average O/M ratio is calculated by a correlation based on the burnup, and then, depending on the radial temperature profile, the O/M radial redistribution is calculated, fixing for each radial cell its local O/M ratio. Here O/M refers the ratio between oxygen atoms and metallic atoms in the fluorite phase. An equilibrium calculation in each radial cell will tell the code how much of each element is found in a volatile phase.
Once the amount of released FP has been calculated, the JOG thickness calculation can be summarized into the following steps: -Obtain the molar quantity of each phase in the gap by performing a thermodynamic calculation with the released element quantities as input data. -Estimate the molar volume of each phase found in the gap by thermodynamic calculation based on their density (see Tab. 3) and molar mass. -Calculate the total JOG volume by summing the volume contribution of each phase. Alternatively, in a simplified approach, the JOG volume can be approximated by omitting the thermodynamic evaluation of the gap, and assuming that all released FP will enter an imaginary phase with a molar volume equal to that of Cs 2 MoO 4 , as it is believed to be the main JOG component [1,53,54]. Since oxygen is not included in the transport model, one mole of volatile fission products produces one third of a mole of this Cs 2 MoO 4 like imaginary phase (two moles of volatile Cs and one mole of volatile Mo makes one mole of Cs 2 MoO 4 ). The amount of available oxygen is assumed to be sufficient to oxidize with all the released FP. From a more general point of view, choosing a Cs 2 MoO 4 like phase to represent all of the JOG is practical for the simulation of heat transfer in GERMINAL V2 since its thermal conductivity is relatively well studied [55,56]. -Regardless of how the JOG volume is obtained, by assuming that the JOG layer is uniform in thickness within each axial slice, JOG thickness, x JOG , can be calculated by the equation: where V JOG is the JOG volume, h is the height of the axial slice, r fuel-outer is the outer radius of the fuel.
The process can be summarized into the flowchart presented in Figure 2.
In the chemical simulation of the fuel, the 15 families of elements from Table 2 are considered for both softwares, and in the computational model, the following FP were considered volatile and thus to be potential components of the JOG: barium, caesium, iodine, molybdenum, palladium, and tellurium. In addition to the volatile fission products, the thermodynamic evaluation of the gap included uranium, plutonium, and oxygen. Here, uranium and plutonium were added to allow the gap components to react with the outer wall of fuel. Molybdenum and caesium were included since Cs 2 MoO 4 is commonly believed to be the main JOG component [1], while barium, tellurium, palladium, and iodine may vaporize at the relevant fuel temperatures and are considered as volatile fission products [3]. In the PIE of one of the fuel pins mentioned in Section 2, all the considered elements had elevated concentrations in the fuel-to-clad gap.
At room temperature, where the JOG width measurements were performed, there is no stable liquid phase. This is not the case for the in-pile conditions, where temperature can reach around 1000 K in the gap. Thus, when calculating the JOG thickness using the method above, there may be liquid phases present. Whether or not these liquids contribute to the JOG thickness is unclear, since it is not known to what extent they migrate axially. In any case, it is not expected to occur at the same rate as the radial migration since the temperature gradient is at least three orders of magnitude smaller. Available oxygen in the gap is another factor which complicates the JOG width calculations. While it is obvious that oxygen should be included in the thermodynamic evaluation of the gap, the true amount is not known. In this work, (U 0.8 ,Pu 0.2 )O 2±x was added to the equilibrium to allow the gap components to react with the fuel. Using this method, it was possible to adjust the oxygen content so that the impact of oxygen potential could be explored. This analysis was only done on the fuel pin with highest burnup, and was carried out by including slightly hypo-and hyper-stoichiometric fuel to the JOG composition. The oxygen redistribution that occurs due to the thermal gradient tends to keep the peripheral O/M ratio close to 2, both before and after the increasing burnup causes the global O/M ratio to reach or even surpass 2 [5,62]. The purpose of these calculations was to investigate how the JOG composition changes at Another factor that complicates the thermodynamic simulation of the JOG is the impact of the cladding elements. This has not been considered in the current work, but the treatment of the ROG ("Réaction Oxyde-Gaine" in French, also known as FCCI = Fuel Cladding Chemical Interaction) is planned to be included in future improvements of the GERMINAL V2 code.
Lastly, the method presented above assumes a smooth uniform layer of JOG in order to calculate the thickness while actual observations reveal a rough, porous structure. This point should be kept in mind when evaluating the results in this work since the porosity may increase the effective JOG volume considerably.

Computation times
The test calculations, using the final (global) composition of the 13.4 %FIMA fuel pin, for obtaining the computational time for the different softwares are presented in Table 4. Results from these calculations can also be found in Figure 3, where the amount of gaseous and liquid phases are shown at different temperatures. Initially OC suffered from a considerable slow-down due to the complexity of the equilibria, but the implementation of an improved solving algorithm improved computation times. Further improvements were obtained by creating a database of solved equilibria, and using these as initial guesses for the calculations (see Sect. 5.1 for more details).

Release into gap
The maximum amount of released FP allowed by the model is the extreme case where all liquid and gaseous phases are transported into the fuel-to-clad gap. This case is illustrated in Figure 4 for the different fuel pins and softwares. Here, the distributions between the volatile fission products are presented, as well as the ratio between OC and ANGE regarding released amount of FP into the gap at the peak power node.   Table 1. The bottom row is the ratio between the total released amount predicted by OC and ANGE. All data is taken from the maximum power node.

JOG thickness vs burnup
The predicted JOG widths for the different fuel pins are presented in Figure 5. Here, the results are divided into three different subplots. The first is showing the results when the simplified mean molar volume method has been used. The second and third subplots correspond to gap  In (a), a uniform molar volume corresponding to that of Cs2MoO4 at room temperature has been assumed for all fission products. In (b) and (c) the gap composition has been evaluated thermodynamically at 800 K (b), and 1000 K (c). In Tourasse (red dots) [1], the burnup values refer to the local burnup at which the JOG was measured. In Melis (black dots) [8], the burnup refers to the maximum burnup reached in the fuel pin.
compositions that have been thermodynamically evaluated at 800 K and 1000 K, respectively. Here, in order to make the calculations of the fuel pins directly comparable to one another, the oxygen content was chosen to correspond to fluorite O/M ratio as close as possible to 2. The value was chosen based on results from the GERMINAL V2 calculations, and is representative of the increase of this ratio with increasing burnup. Included also in this figure is the predicted JOG thickness calculated by the previous, correlation-based model used by GERMINAL V2. Since both solvers predicted both solid and liquid phases in the gap, the results were divided into solid and liquid thickness. The gap width predicted by GERMINAL V2, defined by the distance between the outer fuel surface and the inner clad surface, has also been included.

Thermodynamic calculation of the JOG composition
In order to see how the oxygen potential affects the chemical state in the gap, calculations on the final gap composition of the 13.4 %FIMA irradiated fuel pin were performed together with (U 0.8 ,Pu 0.2 )O 2±x (where x is the deviation from stoichiometry and depends on the oxygen potential). In Figure 6, the results from calculations just above and just below O/M = 2 are presented. Figure 3 shows the predicted fraction of elements in a volatile state (here meaning both gaseous and liquid) in the test calculations using the input of Table 2. The TAF-ID calculations predict a higher tendency towards volatilization. Looking at the difference in the amount of liquid, the behavior becomes more pronounced. This is a consequence of the different models used for describing the liquid phases. As mentioned in Section 3.1, the TAF-ID uses the more advanced ionic liquid model based on the CEF, while the only liquid phases allowed in the TBASE calculations are the molten stoichiometric phases. This difference in models stabilizes the liquid phases (both against solid and gaseous phases) for the TAF-ID calculations. The TBASE calculations do not predict any melting of the precipitated metallic phase (Pd,Mo,Ru), while the TAF-ID calculations predict onset and completion of the melting of the metallic phases at around 1000 K and 2200 K, respectively. The increased complexity of the TAF-ID causes a noticeable increase in computation time compared to TBASE, which is an obvious drawback. In the case of a nuclear fuel simulation, however, most equilibria do not differ significantly from one another in terms of composition and temperature. This circumstance allows for considerable speedup when using OPENCALPHAD, because a large fraction of the computation time is normally spent finding a reasonable initial guess, using a grid minimizer described in references [21,22] based on the approach of Chen et al. [63,64]. If most calculations are similar in temperature and composition, a previous equilibrium can be used as an initial guess. In this work a method that saves and reads equilibria for initial guesses has been implemented, significantly improving computation times. Naturally, generating the initial library of necessary saved equilibria will still cost computational time. Another aspect of OPENCALPHAD which has been explored, but not yet implemented into GERMINAL V2, is the ability to solve several equilibria in parallel, which would obviously open the possibility for further improvements.

Release into gap
The results in Figure 4 show that the OC+TAF-ID configuration predicts a larger amount of released FP in all cases except for the fuel pin with the lowest burnup. A similar trend can be seen in Figure 3, which shows the amount of predicted gas and liquid in the test calculations discussed in Section 3.2. This difference can be explained by the fact that ANGE+TBASE predicts almost no liquid Ba and Mo compared to OC+TAF-ID.

JOG thickness versus burnup evaluation
The results from the GERMINAL V2 calculations and subsequent thermodynamic evaluation of the chemical composition of the gap show that the method used allows the prediction of JOG widths at least comparable to the measured values. For the two fuel pins with lowest burnup, the calculations generally overpredict the widths. In the 7 %FIMA case, the predicted JOG widths even slightly (less than 5 µm) exceed the predicted fuel-clad gap width. This can probably be explained by the fact that the model (incorrectly) assumes instantaneous transport of the FP into the gap. This could be corrected by introducing a threshold burnup at which the JOG is expected to appear (as is the case for the GERMINAL V2 correlation based model).
The comparison between the results from the thermodynamic models presented in this work and the JOG widths predicted by the correlation based model in GERMINAL V2 shows a better agreement. The largest discrepancy between these models is seen in the fuel pin with highest burnup, where the thermodynamic calculation method underpredicts the JOG thickness compared to both the measurement and correlation model. In the 9.0 %FIMA fuel pin (Coucou-1), neither model was able to predict the jump in JOG width seen by the PIE. This could be explained by considering its comparatively low irradiation temperature. A lower temperature corresponds to a smaller fraction of volatile phases as well as a weaker propensity for radial migration due to the lower temperature gradient.
For the two fuel pins with highest burnup, the models slightly underpredict the JOG widths. One possible explanation here is that the measured gap was not entirely a consequence of the presence of FP compounds. As mentioned in Section 2, the experiments do not differentiate between the JOG and the gas gap when measuring the JOG width. The thickness of the JOG is assumed to be equal to the measured width since the gap is expected to be closed due to swelling at this point.
What can be noted regarding the results is the rather high amount of liquid phases in all calculations of the 1000 K case. When the temperature was decreased to 800 K, the amount of liquid decreased, and in some cases vanished completely. If the liquid phases are included in the JOG width calculations, there is no significant difference between the thermodynamic evaluation method and the mean molar volume method. While the simplified method appears adequate at predicting the JOG thickness, it can not be used when studying the consequences of the fission product layer, such as the change in thermal conductivity, and potential chemical interaction with the cladding. This is due to the fact that the simplified model does not give any information regarding the actual chemistry in the gap: everything is assumed to be an imaginary Cs 2 MoO 4 like phase.
In all cases, as can be seen, in Figure 4, Cs is the major contributor to the FP in the gap. In the ANGE calculations, only Cs, Te, and I were released (as well as low amounts of Ba). The OC calculations, using the more advanced database which allows ionic liquids, predicted the release of Mo and Pd in addition to those in the ANGE calculations. The TBASE database utilized by ANGE lacks a thermodynamic description of the liquid metallic Mo and Pd phases. For volatilization to occur, evaporation is required.
The varying fraction of released Mo in the OC calculations can be explained by the maximum temperatures during irradiation of the different fuel pins. Mo is only released when the temperature is above ∼2000 K, so in order to get a relatively high Mo content in the gap, there needs to be either high operating conditions all along the irradiation or at least one period of high temperature at the end of irradiation (when the total Mo inventory has accumulated). This is not the case in Coucou-1 (and Sphinx-1 to a lesser extent) irradiation experiments, and thus the Mo release is lower here.
When performing gap calculations, the most common solid phases were Cs 2 UO 3.5−4 . The metallic phase contained mostly Pd and Te, with low fractions of Mo. The most common liquid phases were Cs 2 Te, Cs 2 MoO 4 and CsI.

Thermodynamic calculation of the JOG composition
In the case of the highest burnup fuel pin, an examination of the impact of oxygen potential, which is known to increase with burnup, was performed. Calculations on the gap compositions together with (U 0.8 ,Pu 0.2 )O 2±x were performed in order to see what happens to the predicted phases as the O/M ratio shifts from below to above the critical value of 2. In the ANGE case, as can be seen in Figure 6, the added oxygen causes the amount of Cs 2 UO 4 to increase, while absorbing some of the caesium of the Cs 2 Te phase. In the OC case where a significant amount of molybdenum is present (unlike the ANGE case), a partial solidification of the Cs 2 MoO 4 phase is seen. The molybdenum which is in the metallic form at lower oxygen potential oxidizes, as expected. All of the caesium from the Cs 2 Te is absorbed into the Cs 2 MoO 4 and Cs 2 UO 4 phases. The tellurium, together with the metallic palladium, form the intermetallics Pd 8 Te 3 and PdTe 2 . In both the OC and ANGE cases, the liquid CsI remains unchanged by the rise in oxygen potential.
The presence of Cs 2 MoO 4 in the calculations discussed here and in Section 5.3 justifies, at least to some extent, the assumptions of the mean molar volume method presented in Section 3.3.
Different options regarding the treatment of the gap chemistry have been explored in this work. There are still several variables that affect the outcome of the thermodynamic calculations in the gap, mainly which elements to include and their respective quantities. Including or discarding fuel and cladding elements in the equilibrium calculations directly or indirectly determines what chemical phases are stable in the gap. In reality, it may also be the case that the inner part of the JOG layer contains fuel elements, while the outer part contains cladding elements. Indeed, small quantities of uranium as well as cladding material was found inside the JOG in the PIE of one of the studied fuel pins. If this is the case, the fuel-to-JOG and JOG-to-clad borders become blurred, and the JOG thickness becomes more complicated to properly define. This can be seen in thermodynamic evaluation of the fuel composition performed in this work. The caesium, together with uranium, forms the oxide phases Cs 2 UO 3.5−4 , when there is not enough molybdenum to form Cs 2 MoO 4 .

Outlook
The volatile fission product release fraction has been approximated to correspond to that of stable fission gases, based on the correlation used by GERMINAL V2. A future improvement could be to couple the JOG prediction model with the MARGARET [65] fission gas transport code in order to describe more accurately the volatile fission product release rate. Indeed, new functionalities have been added to the current version of the code, designated MARGARET PAF, in order to describe the creation, destruction (by decay or by neutronic reaction), and transport of isotopes in the grain and along the grain boundaries.
One factor which has not been accounted for in this work is the JOG porosity. The theoretical density has been used in both of the methods for calculating the JOG thickness, which is expected to underpredict the thickness. Future experiments could help improve the understanding of the JOG microstructure, and thus provide a value for the porosity that could be used in the model. It may, however, be speculated that any measurements would yield a range of values for the porosity, due to the known heterogeneity in composition in the JOG. More detailed PIE focusing on the JOG could illuminate which phases are present in the gap, and the ratios between the present volatile FP. This could also help to clarify whether or not the high fraction of liquid phases in the gap, which is encountered in some of the calculations, is possible. This information could then be implemented to improve the approach presented in this work.

Conclusions
By coupling thermodynamic calculations to the GERMINAL V2 fuel performance code, a method for calculating the JOG width has been developed. Additionally, the approach is able to illuminate the question of which elements and phases are likely to be encountered in the JOG.
In the ANGE+TBASE setup, neither molybdenum nor palladium was found released into the fuel-to-clad gap. Based on PIEs, both of these elements are expected to migrate towards the periphery and into the gap. Thus, the added computational cost of using OC+TAFID over ANGE+TBASE becomes justified by the improvement in predicted fuel chemistries.
In this work, only Ba, Cs, I, Mo, Pd, O Te, U, and Pu have been included in the gap calculations. While it is not obvious how to choose the oxygen content in the gap, different approaches have been tested in this work.
The implementation of thermodynamic calculations into a fuel performance code allows for the possibility of coupling several additional models. For example, the modeling of heat transfer, oxygen and actinide redistribution (by solid or gaseous diffusion), axial redistribution of JOG components, and internal cladding corrosion could all benefit from the calculations performed by the thermodynamic software. This last aspect is planned to be implemented in the future.