Part 4b: Testing the Stellaris model against ARIES

3.1With the existing library

We first tried the existing library alone. Any change to the plant's assembly, connections and inputs was allowed, but no new or changed component definitions. We ran it once, on the repaired model. The run supplied the ARIES values that matched a model input directly: major radius, minor radius and peak ion temperature. The other 701 inputs stayed at the Stellaris design. That included the magnets' turns and current per turn, because the ARIES papers give total ampere-turns, which fix neither choice.

The run stopped at the conductor calculation. The model calculated the field of the Stellaris coils, with their turns and current, at ARIES's smaller major radius: 14.7 tesla on axis, where ARIES has 5.7, and 56.6 tesla at the coils. The conductor model covers 20 to 32 tesla, the limit Part 4a ends on, so it refused to calculate. We obtained no plant assessment and no LCOE, the lifecycle cost per unit of electricity.

Figure 3 puts both designs' fields on one axis, on the plasma axis above and at the coils below. Black squares are ARIES's published values and blue circles are our run. The shaded band is the range the conductor model covers. Our run's field at the coils lies far above that range. ARIES's own coil field lies below it, and the range belongs to a different conductor from the one ARIES uses.

That failure came from running ARIES with the Stellaris coils, so on its own it does not show the library is at fault. With ARIES's own coils and connections, the existing components might still have worked. So we compared each part of the ARIES design with the library directly, and found three gaps that no choice of inputs or connections could close:

• Magnetic field: the field calculation was calibrated at one coil geometry and leaves out a term that changes with geometry, so it cannot carry over to ARIES's coils. ARIES reports about 15.1 tesla at the coils.
• Conductor performance: ARIES uses Nb3Sn conductor at about 4 K. The library had only a REBCO model at 20 K.
• Heat removal and conversion: ARIES removes heat through separate helium and liquid-metal PbLi circuits and converts it in a helium Brayton cycle. Our plant had one helium circuit, a molten-salt intermediate loop and a steam cycle.

The existing library could not describe ARIES, because it lacked the definitions ARIES needs. Rewiring was allowed, but it could not supply the missing relationships.

Evidence: attempt and supplied inputs .project/active/aries-comparison-preparation/post-reveal-results/post-reveal-v1/report.md, field investigation .project/active/aries-comparison-preparation/post-reveal-investigation/findings.md, structural and numerical assessment .project/active/aries-comparison-preparation/final-assessment/report.md.

3.2Adding what ARIES needed

Next we allowed new definitions. New physics was expected: we had modeled a steam cycle, not a helium Brayton cycle. The question was whether the new components would connect to the existing ones and reuse their relationships, without changing the Stellaris model.

A hollow density profile has a lower density at the plasma's center than partway out. ARIES's published profile has that dip at the center and keeps a finite density at the edge. The Stellaris model's density peaks at the center and falls to zero at the edge (Part 3, section 6), so no choice of its inputs gives ARIES's shape.

The plasma-to-fuel connection is the clearest example of reuse. The new plasma calculation produced fusion power, and the existing fuel model used that power to calculate fuel demand. Raising the density by 50% increased fusion power from about 1.84 to 4.13 GW and exceeded the selected exhaust-processing capacity. The existing check caught that failure, with no ARIES-specific fuel equation and no automatic enlargement of the equipment.

Figure 4 follows that connection from left to right. The first two boxes are calculations added for ARIES, the third is the existing fuel model and the last is its existing capacity check. Each box gives its value at the design density and with 50% more.

The Stellaris model was unchanged: after the additions it still reproduced all 1,352 of its numeric outputs exactly.

Evidence: component additions, reuse and executed cases work/orchestration/aries-transfer-experiment/report.md, plant integration work/orchestration/goals/aries-integrated-heat-electricity/answer.md, equipment and cost integration work/orchestration/goals/aries-integrated-equipment-costs/answer.md.

3.3How close the extended model came

After the additions above, we evaluated the model against ARIES's published power balance and costs. We used explicit assumptions for unfinished physics, including magnet qualification and breeding. The purpose was to find what explained agreement or disagreement, not to keep adding detail until the totals matched.

Qualifying the magnets would mean calculating ARIES's field from its coils and checking its conductor at that field. The library still has neither calculation for ARIES (section 3.1), so the ARIES plant takes its magnets as selected inventories and costs from ARIES's own accounts.

Figure 5 has three panels. The top left compares net electricity from the same fusion power, with each plant's turbine inlet temperature on its bar. The top right compares cost of electricity with our model buying all its tritium, split into purchased tritium and everything else. The bottom panel applies ARIES's tritium assumption and accounting conventions to our model and plots the result at each discount rate from 0 to 10%, against ARIES's published figure.

• Power: an unresolved gap. Given ARIES's 2,436 MW of fusion power, our best tested steady alternative produced 891 MW net against the published 1,000 MW. Power-balance assessment work/orchestration/goals/aries-reference-heat-electricity-reconciliation/answer.md
  • Beyond using the published inputs, removing all the heat took two changes. We corrected an error in our ARIES model, which had connected the cycle's three heat exchangers in series where ARIES splits the flow between two of them. And we raised the cycle flow and compressor rating above what the sources state.
  • The higher flow carries the same heat in more helium, so the turbine inlet reached 628 °C against the published 708 °C. A large share of the heat, about 42%, arrives through the relatively cool blanket-helium circuit. An independent reviewer found that the published circuit heat loads and temperatures cannot all hold under the paper's stated exchanger temperature differences, and our sources do not say how ARIES reached its temperature. The gap remains unresolved.
• Cost: almost all of the difference is the tritium assumption. ARIES reported $77.6/MWh in 2004 dollars. For the 891 MW case our model gives $686/MWh, more than 90% of it tritium: the model cannot yet calculate ARIES's breeding, so it buys all its tritium at an assumed $30 million per kilogram. ARIES assumes the blanket breeds all the tritium the plant needs. Cost assessment work/orchestration/goals/aries-reconciled-alternative-economics/answer.md
  • With that assumption and ARIES's other accounting conventions, we calculate about $59/MWh at our assumed 5% real discount rate, and $32 to $105/MWh between 0 and 10%. Historical ARIES costing uses 4.35%, though other financial assumptions also differ, so the remaining difference is unresolved.
  • Much of our equipment cost came from ARIES's own accounts, so this is an accounting comparison, not an independent check of their estimate.

3.4The answer to question 1

Not with the library as built, and only partly once we added what ARIES needed. We made the big changes: a hollow plasma density profile, a blanket with separate helium and PbLi circuits, a helium Brayton cycle, ARIES's split exchanger network, and ARIES's equipment and costs. They assembled with the existing components into an ARIES plant the model can run, and the Stellaris model was unchanged. The remaining differences matter. Our best tested steady case produces 891 MW net against ARIES's 1,000 MW, and we could not reconcile the published heat and temperature data with ARIES's 708 °C turbine inlet. Our cost comes near ARIES's only once we adopt its assumption that the blanket breeds its own tritium.

We then stopped, for time, before closing every gap. The ARIES magnets and conductor, breeding for the ARIES geometry and the cycle-temperature reconciliation remain open, and the financial comparison differs in more than the discount rate. Each is recorded with what it would take to close. And because we made the comparison after reading ARIES, it describes the differences rather than predicting them blind.

4.1Parameters: how far can lowering compressor pressure improve output?

We connected the Stellaris helium cooling loop to the ARIES helium Brayton cycle and varied two operating parameters: cycle flow and the pressure ratio of each compressor stage. The reactor heat and the selected equipment stayed fixed. How much electricity could these components produce together, and what would limit further improvement?

At a cycle flow of 2,500 kg/s, lowering the stage pressure ratio from the plant's starting value of 1.518 to about 1.427 raised net electricity from 427 to 621 MW, 45% more from the same equipment. Lowering it further to 1.425 left 7.8 MW of reactor heat unremoved. That setting cannot hold the plant's heat balance, even though the model still reports about 620 MW.

The gain comes mostly from the compressors. At 2,500 kg/s, lowering the ratio from 1.518 to 1.427 cut their work from 2,452 to 2,062 MW, while the turbine's output fell only from 3,132 to 2,941 MW, so the shaft work left for the generator rose from 680 to 879 MW. The recuperator also recovered more of the turbine's exhaust heat, 1,440 MW instead of 677, and the turbine inlet rose from 404 to 453 °C.

The reactor's helium loop needs its helium back from the exchanger at 562 K (289 °C), so that after its circulators it re-enters the blanket at its design 300 °C. Holding that return temperature is the second check the red point in Figure 6 fails: it sends the helium back half a kelvin too warm.

In Figure 6, the top panel is net electricity and the bottom panel is the reactor heat the exchanger leaves unremoved, both against the compressor pressure ratio. The dashed line marks the exchanger limit.

The limit comes from the connection between the power cycle and the exchanger. Lowering the pressure ratio changes the turbine and recuperator temperatures. The recuperator, which recovers turbine exhaust heat, then sends warmer gas toward the reactor heat exchanger. Eventually that gas is too warm for the exchanger to take all the reactor heat. A compressor setting is therefore limited by heat transfer elsewhere in the plant. The best setting is where the exchanger exactly matches its duty. At the starting ratio it had capacity to spare, so about a third of the loop flow had to bypass it to hold the return temperature.

Changing cycle flow moves that limit. More flow allows a lower ratio, but it does not always give more electricity, because compressor work and turbine inlet temperature change too.

Figure 7 repeats that search at seven cycle flows. The left panel is the pressure ratio at which the exchanger reaches its limit, and the right panel is the net electricity at that ratio.

Every cost except fuel is fixed in this sweep, so more electricity lowers the cost excluding fuel from $133 to $91.5/MWh in 2004 dollars. The output is low for the 3,300 MW of heat the loop delivers because the cycle runs cool: its turbine inlet is about 450 °C, against 708 °C in the ARIES design. The components study below shows the same effect on a different plant. These results use fixed machine efficiencies rather than vendor performance maps, and the bypass hardware is not modeled. The exchanger was fixed equipment in this sweep, so sizing it is the next step.

Evidence: final parameter-study assessment work/orchestration/goals/design-study-parameters/answer.md, verified case results exploration/costed_loop_brayton/studies/20260926-design-study-parameters-b/results/readout.md, figure data archive/write-up/aries-study-assets/parameter-figure-data.json and reproducible renderer archive/write-up/aries-study-assets/render_parameters.py.

4.2Components: steam or helium Brayton conversion?

How do the tested steam and helium Brayton options compare on a reactor that supplies helium at a fixed 500 °C? At that temperature steam works well and a Brayton turbine needs a hotter source, so this comparison favors steam. We built steam and helium Brayton alternatives from the expanded library, with their connecting exchangers and cooling equipment, and attached each to the same Stellaris-derived reactor. For each, we calculated the whole plant: electricity exported after internal use, capital, fuel, operation and equipment replacement.

At this 500 °C source, steam gave the lower LCOE at both supported heat loads, because it exported more than twice as much electricity.

Costs are in 2025 dollars, at 80% availability over 30 years and a 5% real discount rate. Equipment was selected separately at each heat load from a fixed list of priced options, so the rows are different selections, not one plant at rising heat. At 3,000 MW the selected cooling and divertor hardware failed their checks, so neither option gave a supported plant.

The two options also had different freedoms within that list: the steam turbine was held at its supported temperatures, while the Brayton cycle's flow and pressure ratio could vary. So neither was optimized on equal terms, and the result does not rank the two technologies in general.

The difference starts with temperature. A gas cycle spends much of its turbine's work driving its compressors, and that share grows as the turbine inlet gets cooler. From 500 °C helium, the Brayton turbine inlet was about 413 °C, against about 708 °C in the ARIES design, and the compressors took about 1,900 MW of the turbine's 2,478 MW. At 2,500 MW, Brayton delivered 559 MW after its conversion equipment's own use, against 938 MW for steam. Both plants then spent another 274 MW on reactor circulation, heating, refrigeration and other services, leaving 664 MW and 286 MW to sell. That 274 MW includes 100 MW of electricity for plasma heating, which we supplied as an assumption. A burning plasma may not need it.

In Figure 8, each bar is one plant. Its full length is the electricity the plant generates, and the blue part is what remains for sale after the plant's own use.

Steam's conversion equipment cost $2.55 billion, against $1.54 billion for Brayton. Per kilowatt the two cost about the same: about $2,720 for steam and $2,750 for Brayton, dividing each by its conversion subsystem's net output of 938 and 559 MW. An earlier version of this study, which costed the conversion equipment alone, likewise found the two within $5/MWh of each other. Both plants also needed the same $10.23 billion of other purchases, and steam spread those common costs, and later reactor replacements, over more than twice as many MWh. So the LCOE gap comes from Brayton's lower output at this source temperature, not from its equipment price.

In Figure 9, each bar is one plant's LCOE, split by where the money goes; the first segment is the capital to build the plant.

What would change the result at this 500 °C source? Steam stayed cheaper under every tested change in prices, reactor cost and operating assumptions taken one at a time. Only a combination reversed it, at 2,800 MW: Brayton machine efficiencies three points higher, steam turbine efficiencies three points lower, Brayton equipment at half price and steam equipment 50% dearer. Brayton then cost $397/MWh against steam's $427/MWh, an assumed scenario rather than a vendor offer. The next study is to design each cycle for its own source temperature and to settle the heating assumption. Sensitivity results exploration/whole_plant_conversion/studies/20260927-design-study-whole-plant-conversion-b/results/presentation/preference-sensitivity.svg

Evidence: study results and assumptions work/orchestration/goals/design-study-whole-plant-conversion/answer.md (2,496 cases, each recalculated by an independent check), earlier conversion-only comparison work/orchestration/goals/design-study-component-alternatives/answer.md.

4.3Architecture: can different exchanger connections produce more electricity?

We kept the reactor heat and selected equipment the same, then changed how the power-cycle helium passes through three heat exchangers. In series, the whole stream visits each exchanger in turn. In the split network, it passes through the blanket-helium exchanger first, divides between the PbLi and divertor exchangers, then mixes before the turbine. The split network is the arrangement ARIES uses; our first ARIES model had used series (section 3.3).

In Figure 11, follow the cycle helium from left to right, from the recuperator to the turbine. The top row is the series layout and the bottom row is the split network. Each box gives the helium's temperature entering and leaving it at the nominal operating pair compared below; watch the temperature at which the helium reaches the PbLi exchanger.

At 1,835 MW of fusion power, the best tested network operation produced 528 MW net against 498 MW for series, about 31 MW or 6.2% more. Both removed all the reactor heat, met the required return temperatures, and kept at least 30 K between the streams at both ends of every primary exchanger. ARIES cites 30 K for its exchanger bank as a whole; applying it to each exchanger is our assumption, and meeting it required a revised exchanger selection for both layouts.

Both layouts depend on a large bypass. To hold the reactor loops' return temperatures, roughly two-thirds of the blanket helium bypasses its exchanger: 63% in the network and 69% in series. With every bypass limited to 50%, no tested operation of either layout passed. The valves and their costs and pressure losses are not modeled, so the comparison holds only under that control assumption.

Figure 12 compares the two layouts at that operating point: cycle flow, then net electricity, then cost excluding fuel.

The split lets the cycle take the same heat with less circulating helium. In series, the divertor exchanger warms the entire stream before it reaches the PbLi exchanger. Splitting avoids that preheating, so the network needs about 1,246 kg/s of cycle flow where series needs 1,291 kg/s. Less flow means less compressor work and more electricity to export. At equal flow, when both remove all the heat, they produce exactly the same electricity: the split gives no separate efficiency bonus. With costs unchanged, the extra electricity lowers the cost excluding fuel from $104.29 to $98.24/MWh in 2004 dollars.

The study also shows what would reverse the choice. If the network loses 8% of the cycle pressure against 4.5% for series, its output falls to 491 MW, below series at 498 MW. The model does not calculate this loss from the piping. Each layout's loss is a supplied fraction, 4.5% in the nominal case, so calculating it is the next step. The network's real pressure losses and its extra piping and controls therefore decide whether it is worth building.

Changing connections can cut the circulation needed to carry the reactor heat, and so raise output, but added pressure loss can erase the gain. The study gives the gain and the supplied loss that would erase it: a reason to prefer the network, and a concrete thing to calculate before choosing it.

Evidence: final architecture assessment and replay work/orchestration/goals/design-study-exchanger-architecture/answer.md, verified sensitivities work/orchestration/goals/design-study-exchanger-architecture/evidence/r3-data/sensitivity-coverage.csv, figure data archive/write-up/aries-study-assets/architecture-figure-data.json and renderer archive/write-up/aries-study-assets/render_architecture.py.

4.4The answer to question 2

Yes, within the physics and prices we represent. Each study evaluated a design question across the whole plant and showed what limits the result:

• Parameters: an exchanger elsewhere in the plant limits the compressor setting, and the best setting raised output 45% with the same equipment.
• Components: the conversion system can be swapped with its own equipment and evaluated as a whole plant. On a fixed 500 °C source, which favors steam, steam exported more than twice as much electricity and so gave the lower LCOE. This does not rank the technologies in general.
• Architecture: splitting the exchanger flow gains 6%, but a few percent more pressure loss erases the gain.

Two of the studies combine components from both plants, which neither original model could do. The results are conditional on the represented physics and assumed prices, and each says where the design is limited and what to check before choosing it.