# Equal torque and the material-mass budget

**Disposition:** thinning the radial magnet from 1.5 to 1.0 mm, while retaining the assumed SUS430-class ring, does not save material mass if added copper is used to preserve the original copper heat. That statement holds at all three directly solved reference loads. It does **not** rule out a short-duty application with spare thermal capacity. Two other arrangements retain conditional material budgets, which must pay for retention, loss, cooling and manufacturing changes before an installed-mass benefit can be claimed.

This is a public GL80-derived, native 10 mm stack calculation. The reference torque loads are bookkeeping choices from the earlier sweep, not an aircraft duty, a measured continuous rating or an identified customer's component. The original numerical and physical-validation limits remain in force.

## What changed in this extension

The existing 60-configuration field sweep was reused. At Br = 1.25 T, all five rotor arrangements and all four magnet thicknesses were compared by interpolating their existing torque curves: 20 configurations at three common torque targets, or 60 comparisons. The important 1.0 mm radial/SUS cases were then solved directly, using the existing nonlinear field solver. Two consequential alternatives were also directly solved at the middle reference torque: a 1.0 mm radial magnet with a generic-steel ring and a 2.0 mm ring-free Halbach rotor. No new material curve, physical cooling property or continuous-duty assumption was introduced.

The direct calculation corrects the earlier approximate current/heat comparison. At 3.0445 N m, the 1.0 mm radial/SUS option needs **11.48% more current and 24.27% more copper heat**, rather than the approximate 13% and 28%. Saturation makes this penalty load dependent.

## Declared comparison

Unchanged: 36 slots, 42 poles, native 10 mm axial stack, stator section, 0.7 mm radial air gap, Br = 1.25 T at the evaluated material state, recoil relative permeability 1.05, sinusoidal q-axis excitation and each comparison's required torque. The source section uses a 40/48 magnet arc fraction, an approximation to the public drawing. The model's generic steel and SUS430-class curves are assumptions, not certificates for supplied ring materials.

Winding bookkeeping: 25 turns per tooth coil, 12 series coils per phase, 29% initial slot fill, modeled mean turn length 33.26498 mm, copper density 8960 kg/m³. This yields **51.63089 g copper** and **0.893869 Ω phase resistance at 20°C**. These are calculations, not measured winding values. Three reference current amplitudes of 150, 300 and 500 A-turn peak per coil correspond to directly reproduced torques of 1.556221, 3.044496 and 4.673917 N m. The baseline middle case already dissipates 193.076 W copper at 20°C; no claim that this is continuously sustainable is made.

Magnet density is 7500 kg/m³ and ring density 7650 kg/m³. The coherent sweep grid gives baseline magnet/ring masses of 24.59274/20.66850 g. Using this same grid for candidates avoids silently mixing the earlier finer-grid native mass estimate into the sweep comparison. No statistical uncertainty bounds are available for these model-derived masses.

## Direct equal-torque results

| Arrangement | Magnet thickness (mm) | Target torque (N m) | Required peak AT/coil | Current change | Copper-heat change | Extra copper heat at 20°C (W) | Rotor material saved (g) | Added copper for equal heat (g) | Net material saved at equal copper heat (g) |
|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|
| Radial, SUS-class 1 mm ring | 1.0 | 1.5562 | 164.3065 | +9.54% | +19.98% | +9.647 | 8.536 | +10.318 | **−1.782** |
| Radial, SUS-class 1 mm ring | 1.0 | 3.0445 | 334.4267 | +11.48% | +24.27% | +46.856 | 8.536 | +12.530 | **−3.994** |
| Radial, SUS-class 1 mm ring | 1.0 | 4.6739 | 600.9823 | +20.20% | +44.47% | +238.512 | 8.536 | +22.961 | **−14.425** |
| Radial, generic-steel 1 mm ring | 1.0 | 3.0445 | 316.9715 | +5.66% | +11.63% | +22.463 | 8.536 | +6.007 | **+2.529** |
| Ring-free Halbach | 2.0 | 3.0445 | 268.5578 | −10.48% | −19.86% | −38.351 | 5.677 | −10.255 | **+15.933** |

Positive numbers in the final column are **remaining material budgets**, not installed-mass savings. In particular, the ring-free case contains 39.584 g of magnets and no modeled ring. Its mass accounting must include a feasible replacement retention structure. Reducing conductor area in that case is an accounting alternative, not a winding recommendation.

The 8.536 g radial thinning credit comprises 8.29575 g magnet and 0.24033 g ring-radius change. The ring remains 1 mm thick. With fixed copper winding and a fixed 20°C resistance comparison, added cooling and all other installation burdens must stay below this entire credit. Expressed per extra watt of copper loss, the allowable added mass is at most **0.8849, 0.1822 and 0.03579 g/W** at the three thin-SUS loads. For the thin generic-steel case it is **0.3800 g/W**. These are break-even bounds, not achievable cooling coefficients. Other added losses make the bounds smaller; existing thermal headroom can change whether extra cooling is needed at all.

The equal-heat conductor fills would be 34.80%, 36.04% and 41.90% for the thin-SUS cases, 32.37% for thin generic steel, and 23.24% for ring-free Halbach. Manufacture at those fills has not been demonstrated. Slot insulation, conductor packing, bend radius, termination and AC loss remain unmodeled.

Rewinding changes fault damping as well as heat. At the middle reference torque, equal-heat conductor areas imply phase resistances at 20°C of 0.719308 Ω (thin-SUS), 0.800712 Ω (thin generic-steel) and 1.115427 Ω (ring-free Halbach), compared with 0.893869 Ω for the unchanged winding. A fault calculation using only the unchanged resistance does not establish the rewound candidate's fault performance. Both resistance choices are explicit in the fault-state handoff.

## Arithmetic and conditions

For peak ampere-turns `AT`, turns per coil `N = 25` and phase resistance `R`:

`I_phase_peak = AT/N; I_phase_rms = AT/(N sqrt(2)); Pcu = 3 I_phase_rms² R`.

Let `r = AT_candidate/AT_baseline`. Keeping turns, slot area, mean turn length and copper temperature fixed, conductor area must increase by `r²` to keep copper heat fixed. Consequently:

`Mcu_new = Mcu_baseline r²; fill_new = 0.29 r²`.

`net material saved = magnet-and-ring mass saved − Mcu_baseline (r²−1)`.

With 8.53609 g rotor credit and 51.63089 g baseline copper, the break-even current ratio is **1.079504**: only a 7.95% current increase can be paid for by copper without consuming more material than thinning saves. The thin-SUS direct solutions exceed that ratio at all three loads. The thin generic-steel case falls below it but leaves only 2.53 g for every other burden.

Changing turn count alone at fixed slot fill and mean turn length does not remove this copper-loss penalty. At fixed AT, current scales as 1/N and resistance as N², leaving copper heat unchanged. Turn count can still affect voltage, current rating, end turns, packing and manufacturability.

At a common winding temperature `Tw`, multiply the quoted 20°C copper powers by `1 + 0.00393 (Tw−20)`. This is not a rotor temperature model. If winding temperatures differ, the direct heat ratios must be recalculated. A mission energy penalty would require integrating the extra losses over the actual time history; battery and cooling masses are not inferred without that duty.

The controller must supply the solved phase current at the required speed. Retained phase-flux fundamentals imply no-load line-to-line RMS back-EMF coefficients of 0.022295, 0.020413, 0.021721 and 0.024714 V/rpm for baseline, thin-SUS, thin generic-steel and ring-free 2 mm cases respectively. These are 2D reference predictions. Loaded bus-voltage feasibility also needs inductive voltage, winding temperature, speed, inverter modulation and current limits. The lower-current Halbach case has higher back-EMF, so its current advantage can be defeated by the voltage constraint. No speed or voltage operating point is supplied here.

## Why the material comparison remains coupled

At the directly matched middle torque, the worst of the four sampled positions gives the following positive reverse-field magnitudes at the lowest 1% projected-induction quantile:

| Case | Required reverse field, q01 screen (kA/m) |
|---|---:|
| Baseline radial/SUS, 1.5 mm | 532.371 |
| Thin radial/SUS, 1.0 mm | 620.988 |
| Thin radial/generic steel, 1.0 mm | 608.058 |
| Ring-free Halbach, 2.0 mm | 923.990 |

These use `H = (Br−B_parallel_q01)/(mu0 × 1.05)` and assume uniform axial extrusion when interpreting area as volume. The 1% screen is an exploratory convention, not a damage acceptance limit, probability or physically validated demagnetization result. Four rotor positions do not establish a continuous-angle worst case. The calculations include no irreversible magnetic evolution. Full hot finished-part curves, direction-dependent behavior and circuit-coupled fault states may eliminate an apparent material budget. The exact equal-torque fields and currents are provided to the fault investigator in `fault_initial_conditions.json`.

## Numerical evidence and limits

The original portable FE source was copied byte-for-byte. Source hashes match the public package and the remote staged copies. The pre-run script specified relative equal-torque residual ≤1e−4. All five changed-design roots met it; the maximum observed residual is 5.471e−5. Baseline native-stack torque reproduces the retained per-metre result times 0.010 within 2.266e−9 relative. All 32 retained loaded field states converged; maximum summed phase-current imbalance is 3.13e−13 A-turn. The 12 no-load states per geometry also converged. These establish internal numerical consistency on this mesh and sampling scheme, not mesh convergence or physical prediction accuracy.

The field settings are unchanged: nonlinear 2D magnetostatic P1 elements; 120° periodic sector; 8 angular columns per base unit; radial subdivisions 4/8/20/4/6/10/6/4 for inner air/yoke/teeth/tip/gap/magnet/ring/outer air; Newton relative residual 1e−8, absolute 1e−12, maximum 60 iterations. No-load phase is fitted from 12 samples over an electrical period. Loaded torque is averaged over shifts 0, 4, 8, 12, following the retained sweep. Halbach uses four magnetization directions per pole pair. There was no additional mesh or angular refinement in this bounded branch.

The first two jobs ran serially, one CPU worker at a time, with BLAS/OMP/MKL thread counts 1 and nice 10 on the existing secondary Spark. The first job took 98.80 s; the two-case extension took 64.69 s. Outputs were recovered and all 36 result/log/field hashes matched across hosts. The later all-five initial-state extension below used the same controls. Existing workloads were left untouched. This branch has no running jobs.

The model omits axial ends, rotor eddy-current reaction, iron-loss changes, thermal transients, structural retention, manufacturing variation and measured material curves. The earlier prototype benchmark's failures are not repaired by matching one model against its own retained outputs. Neither solver convergence nor this root match validates hardware.

## Decision and smallest missing observations

1. **Thin radial/SUS:** close the equal-copper-heat mass claim as negative within the declared winding model. Retain only the conditional short-duty/spare-cooling possibility. A winding resistance and duty-temperature record would determine whether that possibility is relevant; a heavier hot finished magnet cannot repair its mass accounting alone.
2. **Thin radial/generic steel:** retain as a small conditional budget, not a recommended material. The actual ring B–H curve, magnetic/eddy losses and structural requirement could reverse its 2.53 g budget. A measured or certified ring constitutive curve and construction would distinguish an achievable substitution from the favorable generic assumption.
3. **Ring-free 2 mm Halbach:** retain a larger 15.93 g combined material budget only if feasible retention fits within it and finished magnets withstand its substantially larger loaded/fault reverse fields. Its 5.68 g rotor-only credit is the relevant bound if the winding stays unchanged. Hot parent-to-finished-arc magnetic comparison, plus credible retention geometry, can decide whether this route deserves development. Neither observation is supplied by these calculations.

The parent investigation owns fault evolution, independent reverse-field benchmarking and any axial-end check. Their outcomes can only narrow or revise these provisional budgets; they must not be read as already passed.

## Follow-up: all five equal-torque fault initial states

Four additional direct roots completed the five-construction comparison at 1.5 mm magnet thickness and the same 3.04449625618 N m reference torque. The baseline was already available. The native ncol8 geometry and predeclared 1e−4 root tolerance were retained. All new loaded and no-load states converged, and the maximum torque residual was 3.158e−5. The single-worker follow-up took 141.53 s; 18 additional output/log/field hashes matched across hosts. No new geometry search was performed.

| Construction, all with 1.5 mm magnets | Equal-torque peak AT/coil | Extra copper heat at 20°C (W) | Rotor material saved (g) | Net material budget with equal-heat rewind (g) | Loaded q01 reverse-field screen (kA/m) |
|---|---:|---:|---:|---:|---:|
| Radial, SUS-class 1 mm ring | 300.0000 | 0 | 0 | 0 | 532.371 |
| Radial, SUS-class 1.5 mm ring | 278.2531 | −26.977 | −10.515 | −3.300 | 502.468 |
| Radial, generic-steel 1 mm ring | 284.7844 | −19.088 | 0 | +5.104 | 508.202 |
| Ring-free Halbach | 312.6862 | +16.675 | +15.750 | +11.291 | 932.405 |
| Halbach, SUS-class 1 mm ring | 280.7793 | −23.948 | −4.919 | +1.485 | 999.933 |

This confirms that a thicker SUS-class ring improves the magnetic/copper trade but consumes its copper-mass advantage in added ring mass. The generic-steel substitution offers a conditional opportunity without thinning the magnet, contingent on real ring magnetic, structural and loss properties. Both Halbach options retain the earlier warning: their lower current or ring-removal credit does not remove their much larger internal reverse-field requirement. None is an installed-mass or finished-magnet acceptance result.

A later, explicitly bounded material bridge tested the thin generic-steel candidate at two public typical 100°C Br observations. Its nominal 2.529 g budget at Br1.25 T becomes only **0.556 g at Br1.227 T (45SH, supplier-labeled HRE-free)** and **4.280 g at Br1.272 T (52SH, supplier-labeled with Dy)**. The former lies close enough to the zero-budget region that source digitization alone is consequential. These results do not establish installation benefit, hot finished-part capability or fault passage. See `MATERIAL_BRIDGE.md` and the exact `material_bridge_summary.json`; exactly two further roots were run, then stopped.

A separately authorized single inverse-Br root subsequently corrected the coarse interpolation boundary: at fixed323.851224 AT peak/coil, the directly solved zero-material-budget contour is **Br1.220962428 T**, with torque residual1.004e−6. The nominal45SH reading therefore has6.038 mT headroom, compared with approximate7 mT reading resolution. Earlier interpolated boundary and endpoint budgets remain historical interpolation evidence; they must not be stated as direct field results. This check did not change the direct45SH0.556 g or52SH4.280 g nominal budgets.

A subsequent arithmetic-only audit found that solid generic-ring eddy diffusion cannot be assumed negligible on a millisecond fault timescale: ordinary-load local longitudinal slab estimates have median0.846–3.809 ms for the thin generic-steel case across an independent public resistivity bracket. This is an assumption diagnostic, not a conductive-field correction or conservative bound. The actual ring's conductivity, construction and three-dimensional current closure remain inputs. See `EDDY_TIMESCALE.md`, `.json` and `eddy_timescale.py`; no new field run was performed.

The exact phase, current vectors, geometry and separate fixed-winding/rewind resistances are in `fault_equal_torque_all_five.json`. Those currents solve the ncol8 model; passing them unchanged to a different grid does not itself establish equal torque on that grid. The original 60-row interpolation ledger remains preserved and labeled interpolation.

## Independent material-envelope implementation check

At the parent investigator's request, the corrected fault envelope was inspected without editing its implementation. The alternating mesh diagonals were checked against independently reconstructed diagonal intersections and barycentric point-in-triangle classification in the actual rotated mesh. Synthetic, deliberately unequal triangle fields were advanced through shifts 0, 1, 2, 7, nt−1, nt+1 and 2nt+5. Radial and both Halbach constructions each supplied 583,296 independently classified material-attached sample points.

Full running material-coordinate maxima, subtriangle areas and the weighted q99 screen agreed. Maximum field discrepancy was 3.98e−13 kA/m and maximum subtriangle-area discrepancy 1.21e−21 m². The audited `fault/coupled.py` SHA-256 was `b61f0eb9fa063f6b55f9160d4f2e4b07eb44a5e78e6c04932f1c50de9cf0a1f9`; later implementation changes need a fresh check. This verifies the integer-shift union bookkeeping for piecewise-constant triangle fields. It does not validate magnetic physics, physical irreversible damage, time convergence or mesh convergence. See `MATERIAL_ENVELOPE_INDEPENDENT_CHECK.json` and its executable `check_material_envelope.py`.

## Reproduction and artifacts

- `ledger.py`: coherent 60-row interpolation ledger from retained public sweeps.
- `direct_equal_torque.py` and `direct_inputs.json`: original three-target thin-SUS direct root solves and three baseline checks.
- `direct_promising.py` and `direct_promising_inputs.json`: the two authorized additional direct comparisons.
- `summarize_direct.py`: all mass, heat, voltage-coefficient arithmetic and fault-state handoff; run with the Python environment already supplying NumPy/SciPy.
- `direct_equal_torque_ledger.csv` / `.json`: eight baseline/candidate direct rows with units in column names.
- `interpolated_equal_torque_ledger.csv` / `.json`: all-five, all-thickness preliminary comparison. Unsolved rows remain interpolation evidence.
- `direct_equal_torque_results.json`, `direct_promising_results.json`, logs and 32 `field_*.npz`: retained field, convergence and root records.
- `fault_initial_conditions.json`: full geometry, rotor material/magnetization kwargs, exact phases, current conventions and fields for circuit coupling.
- `fault_equal_torque_all_five.json`, `equal_torque_all_five.csv`, `direct_all_five.py`, input/results JSON and log: all-five equal-torque initial-state follow-up and coherent mass accounting.
- `check_material_envelope.py`, `MATERIAL_ENVELOPE_INDEPENDENT_CHECK.json`: independent geometric audit of the corrected material-envelope bookkeeping.
- `DIRECT_NUMERICAL_CHECKS.json`, `REMOTE_STAGE_HASHES.json`, `REMOTE_PROMISING_HASHES.json`, `REMOTE_OUTPUT_HASHES.json`: numerical and transfer receipts.

Run the two FE scripts sequentially with `OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 MKL_NUM_THREADS=1`, preserving the original output directory if reproducing. The unmodified model sources are under `models/`. The source package's public geometry attribution and rights record remain authoritative; no supplier curve is needed or copied for this branch.
