1costingfe

JAX-native fusion power plant costing framework. Replaces the PyFECONS adapter in the fusion-tea SysML pipeline with a differentiable, 5-layer customer-first model.

Install

pip install -e .
# or with dev dependencies:
pip install -e ".[dev]"

Requires Python 3.10+.

Quick Start

from costingfe import CostModel, ConfinementConcept, Fuel

# Create a model for a DT tokamak
model = CostModel(concept=ConfinementConcept.TOKAMAK, fuel=Fuel.DT)

# Forward costing: customer requirements -> LCOE
result = model.forward(
    net_electric_mw=1000.0,
    availability=0.85,
    lifetime_yr=30,
)

print(f"LCOE: {result.costs.lcoe:.1f} $/MWh")
print(f"Overnight cost: {result.costs.overnight_cost:.0f} $/kW")
print(f"Fusion power: {result.power_table.p_fus:.0f} MW")

Sensitivity Analysis (JAX autodiff)

sens = model.sensitivity(result.params)

# Engineering levers (sorted by |elasticity|)
for k, v in sorted(sens["engineering"].items(), key=lambda x: abs(x[1]), reverse=True):
    print(f"  {k:25s} {v:+.4f}")

# Financial parameters
for k, v in sens["financial"].items():
    print(f"  {k:25s} {v:+.4f}")

Elasticity = (dLCOE/dp) * (p/LCOE) -- dimensionless, comparable across parameters.

Batch Parameter Sweeps (JAX vmap)

# Sweep blanket thickness from 0.5m to 1.0m
lcoes = model.batch_lcoe(
    {"blanket_t": [0.5, 0.6, 0.7, 0.8, 0.9, 1.0]},
    result.params,
)

Cross-Concept Comparison

from costingfe import compare_all

results = compare_all(net_electric_mw=1000.0, availability=0.85, lifetime_yr=30)
for r in results[:5]:
    print(f"  {r.concept.value:15s} {r.fuel.value:5s} {r.lcoe:6.1f} $/MWh")

Backcasting

from costingfe.analysis.backcast import backcast_single

# What availability achieves 60 $/MWh?
avail = backcast_single(
    model, target_lcoe=60.0, param_name="availability",
    param_range=(0.70, 0.98), base_params=result.params,
)

Cost Overrides

Override any CAS account or CAS22 sub-account with a known value (M$). Downstream totals (CAS20, total capital, LCOE) recompute automatically.

# Override an entire CAS account
result = model.forward(
    net_electric_mw=1000.0,
    availability=0.85,
    lifetime_yr=30,
    cost_overrides={"CAS21": 50.0},  # "I know my building cost"
)
assert result.costs.cas21 == 50.0
print(result.overridden)  # ["CAS21"]

# Override a CAS22 sub-account (coils)
result = model.forward(
    net_electric_mw=1000.0,
    availability=0.85,
    lifetime_yr=30,
    cost_overrides={"C220103": 300.0},  # "Use this coil cost"
)
# CAS22 total is recomputed from patched sub-accounts
print(result.cas22_detail["C220103"])  # 300.0
print(result.cas22_detail["C220000"])  # Recomputed total

Available CAS-level keys: CAS10, CAS21-CAS28.

Available CAS22 sub-account keys: C220101 (blanket), C220102 (shield), C220103 (coils), C220104 (heating), C220105 (structure), C220106 (vacuum), C220107 (power supplies), C220108 (divertor), C220109 (DEC), C220111 (installation), C220112 (isotope sep), C220200 (coolant), C220300 (aux cooling), C220400 (rad waste), C220500 (fuel handling), C220600 (other equipment), C220700 (I&C).

CAS70 sub-accounts: CAS71 (annual O&M), CAS72 (annualized scheduled replacement — blanket/FW + divertor, PV-discounted at interest rate).

Sensitivity analysis works with overrides -- overridden accounts become constants with zero gradient:

sens = model.sensitivity(result.params, cost_overrides={"CAS21": 50.0})

Fusion-Tea Adapter

from costingfe.adapter import FusionTeaInput, run_costing

inp = FusionTeaInput(
    concept="tokamak",
    fuel="dt",
    net_electric_mw=1000.0,
    availability=0.85,
    lifetime_yr=30,
)
out = run_costing(inp)
# out.lcoe, out.costs (CAS-keyed dict), out.power_table, out.sensitivity

The adapter supports two additional override mechanisms for the fusion-tea pipeline:

inp = FusionTeaInput(
    concept="tokamak",
    fuel="dt",
    net_electric_mw=1000.0,
    availability=0.85,
    lifetime_yr=30,
    # Inject known CAS account values (M$)
    cost_overrides={"CAS21": 50.0, "C220103": 300.0},
    # Override costing constants (unit costs, fractions, etc.)
    costing_overrides={"blanket_unit_cost_dt": 1.0},
)
out = run_costing(inp)
print(out.overridden)           # ["CAS21", "C220103"]
print(out.costs["C220103"])     # 300.0 (CAS22 sub-accounts included in costs dict)

• cost_overrides -- replace computed CAS account values with known costs. Passed through to CostModel.forward().
• costing_overrides -- override fields on CostingConstants (unit costs, scaling coefficients). Applied via cc.replace() before model construction.
• out.overridden -- list of keys that were injected rather than computed.

Supported Concepts

Family	Concept	Key features
MFE	tokamak	Toroidal confinement, TF/CS/PF coils
MFE	stellarator	Steady-state, complex 3D coils
MFE	mirror	Cylindrical, end-loss DEC opportunity
IFE	laser_ife	Split laser drivers, target factory
IFE	zpinch	Pulsed power driver
IFE	heavy_ion	Heavy ion accelerator
MIF	mag_target	Magnetized target, liner factory
MIF	plasma_jet	Plasma jet driver

Fuels

• dt -- Deuterium-Tritium (breeding blanket, heavy shielding)
• dd -- Deuterium-Deuterium (no breeding, moderate shielding)
• dhe3 -- Deuterium-Helium-3 (mostly aneutronic)
• pb11 -- Proton-Boron-11 (fully aneutronic, minimal shielding)

Engineering Overrides

Pass any engineering parameter as a keyword argument:

result = model.forward(
    net_electric_mw=1000.0,
    availability=0.85,
    lifetime_yr=30,
    eta_th=0.50,        # Override thermal efficiency
    blanket_t=0.90,     # Thicker blanket
    axis_t=7.0,         # Larger major radius
)

See src/costingfe/data/defaults/ YAML files for all available parameters.

Power Cycle Selection

Select the thermal cycle used for power conversion. This sets eta_th and BOP cost coefficients (CAS23, CAS26) appropriate for the chosen cycle.

from costingfe import CostModel, ConfinementConcept, Fuel, PowerCycle

# sCO2 Brayton cycle
model = CostModel(
    concept=ConfinementConcept.TOKAMAK,
    fuel=Fuel.DT,
    power_cycle=PowerCycle.BRAYTON_SCO2,
)
result = model.forward(net_electric_mw=1000.0, availability=0.85, lifetime_yr=30)

Available cycles:

Cycle	eta_th	Description
RANKINE (default)	0.40	Steam Rankine (sub/supercritical)
BRAYTON_SCO2	0.47	Supercritical CO2 recompression Brayton
COMBINED	0.53	Gas topping + steam bottoming

You can override eta_th or BOP coefficients independently:

# sCO2 Brayton but with custom efficiency
result = model.forward(
    net_electric_mw=1000.0, availability=0.85, lifetime_yr=30,
    eta_th=0.50,  # Override preset's 0.47
)

Tests

pytest tests/ -v

Architecture

Customer Requirements (net_electric_mw, availability, lifetime_yr)
    |
    v
Layer 2: Physics (power balance, inverse for target p_net)
    |
    v
Layer 3: Engineering (radial build -> geometry -> volumes)
    |
    v
Layer 4: Costs (CAS 10-60 accounts, volume-based + power-scaled)
    |
    v
Layer 5: Economics (CAS 70-90, LCOE)