Opt

README.md

@@ -438,6 +438,10 @@ those diameters is a search, and `halbach-optimizer` runs it:
 ./build/optimizer/halbach-optimizer -o output/halbach_optimized.json
 ```
 
+Seven worked designs live in `designs/`, one folder each, holding the configuration, the resulting
+magnet, the field it produces and the convergence of the search. `designs/regenerate.sh` rebuilds
+them all.
+
 With no configuration file the defaults are the design of `HalbachOptimisation/`, the genetic
 algorithm this is a port of: 23 rings 22 mm apart, each position choosing one of 19
 (radius, magnet count) pairs, in two layers 21 mm apart, measured over a 200 mm sphere.
@@ -454,7 +458,7 @@ ignore. That object is also a configuration file, so the output of one run is th
 next:
 
 ```bash
-./build/optimizer/halbach-optimizer -c output/halbach_optimized.json --seed 7
+./build/optimizer/halbach-optimizer -c designs/nmr-5ring-10mm-tube/optimized.json --seed 7
 ```
 
 #### How it works
@@ -482,8 +486,8 @@ another position changes. So the run has three stages, and only the middle one i
 
 #### Configuration
 
-`halbach_optimization.json` in the project root is the full set of defaults, written out. Every
-key is optional:
+`designs/imaging-200mm-dsv/config.json` is the full set of defaults, written out. Every key is
+optional, and every design under `designs/` carries its own — see `designs/README.md`:
 
 | Key | Meaning |
 | --- | --- |
@@ -500,8 +504,9 @@ key is optional:
 
 The command line overrides the file: `--generations`, `--population`, `--seed`, `--symmetry`,
 `--objective`, `--field-model`. `--emit magnets` writes every magnet out one by one instead of the
-46 rings. `--history convergence.csv` writes the convergence curve. `--no-verify` skips stage
-three, for a smoke test. `halbach-optimizer --help` lists them all.
+46 rings. `--history convergence.csv` writes the convergence curve, and `--field field.csv` writes the field
+the verification measured, one row a sample, so the answer can be plotted or a lineshape fitted
+without simulating it again. `--no-verify` skips stage three, for a smoke test. `halbach-optimizer --help` lists them all.
 
 `--field-model dipole` replaces each cube with the point dipole of the same moment,
 `m = V·Br/µ0`, which is the model `HalbachOptimisation/halbachFields.py` uses. It is there to

designs/README.md

@@ -0,0 +1,109 @@
+# Magnet designs
+
+One folder per magnet. Each was produced by `halbach-optimizer` from the
+`config.json` beside it, and each folder holds both the answer and the evidence
+for it.
+
+| file | what it is |
+| --- | --- |
+| `config.json` | what was asked for. The `_`-prefixed keys are the record of *why* each parameter is what it is — the sweeps behind it, the constraints that bind, and what was traded against what. The optimizer ignores them. |
+| `optimized.json` | what came out. **An input file the simulator runs as it stands** — `arrangements` plus a `field_of_view` — with the homogeneity and field figures beside them in a `halbach_optimization` object the readers ignore. |
+| `field.csv` | the field the verification measured, one row a sample: `x,y,z,Bx,By,Bz,Bmag` over the whole sample sphere, at nine significant digits. |
+| `convergence.csv` | the search, one row a generation: `generation,best_ppm,best_ever_ppm,mean_ppm,seconds`. |
+
+## Where the homogeneity and field figures live
+
+In `optimized.json`, under `halbach_optimization`. Two blocks, and the
+difference between them matters:
+
+```json
+"optimized": { "homogeneity_ppm": 90.4,  "mean_T": 0.2127, "min_T": ..., "max_T": ...,
+               "peak_to_peak_T": ..., "points": 4662 },
+"verified":  { "homogeneity_ppm": 91.4,  "mean_T": 0.2127, ..., "points": 33401 }
+```
+
+`optimized` is the figure the search was steered by, over whatever symmetry
+reduction it used. `verified` is the same figure taken again over the **whole**
+sphere, from the magnets themselves, through the simulator the rest of the
+project uses. **Quote `verified`.** A number that was optimized cannot also be
+the evidence that the optimization worked, and the two do come apart: on
+`nmr-3ring-21magnets` the octant search said 531 ppm where the honest figure was
+719, because a 3-magnet ring carries a harmonic the octant reflections do not
+respect. Everything with ten or more magnets to a ring agrees to about 1%.
+
+Alongside them: `basis` (the size of the precomputation), `magnet_count`,
+`genome`, `chosen` (which candidate went to which ring position, with radii and
+counts), and `timing_s`.
+
+## The designs
+
+Field figures are `verified`, over the whole sample sphere.
+
+| design | magnets | cube | field | homogeneity | DSV | tube |
+| --- | ---: | ---: | ---: | ---: | ---: | --- |
+| `imaging-200mm-dsv` | 2971 | 12 mm | 49.1 mT (2.09 MHz) | 805 ppm | 200 mm | — imaging |
+| `nmr-21ring-10mm-tube` | 789 | 8 mm | 316.0 mT (13.46 MHz) | 215 ppm | 20 mm | 10 mm |
+| `nmr-5ring-10mm-tube` | 185 | 8 mm | 212.7 mT (9.05 MHz) | **91 ppm** | 12 mm | 10 mm |
+| `nmr-3ring-5mm-tube` | 99 | 8 mm | 178.6 mT (7.60 MHz) | 247 ppm | 8 mm | 5 mm |
+| `nmr-2ring-5mm-tube` | 86 | 8 mm | 82.8 mT (3.52 MHz) | 246 ppm | 6 mm | 5 mm |
+| `nmr-3ring-21magnets` | 21 | 24 mm | 171.2 mT (7.29 MHz) | 602 ppm | 4 mm | capillary |
+| `nmr-2ring-10magnets` | 10 | 24 mm | 83.9 mT (3.57 MHz) | 9396 ppm | 4 mm | — not usable |
+
+`imaging-200mm-dsv` is the port of `HalbachOptimisation/`, the Python project
+this optimizer replaces. The rest are NMR magnets for a sample tube along the
+bore, with B0 transverse to it.
+
+## Reading the ppm figure
+
+Peak-to-peak over a sphere is a worst-case number and it **overstates the line
+you would actually see**, often by a lot. What is left after optimization is
+mostly ring-discreteness ripple, which grows as (r/R)^(N-2) — for a 15-magnet
+ring that is r¹³, so almost all of it lives in a thin shell at the very edge of
+the sphere. On `nmr-5ring-10mm-tube`, half the volume sits within 4.8 ppm of the
+mean while the peak-to-peak is 91.
+
+The lineshape is just the histogram of `Bmag` over the detected volume, so
+`field.csv` gives it directly:
+
+| design | p-p ppm | line FWHM | as ppm | T2\* |
+| --- | ---: | ---: | ---: | ---: |
+| `nmr-21ring-10mm-tube` | 215 | 96 Hz | 7.2 | 3.3 ms |
+| `nmr-5ring-10mm-tube` | 91 | 48 Hz | 5.3 | 6.6 ms |
+| `nmr-3ring-5mm-tube` | 247 | 78 Hz | 10.3 | 4.1 ms |
+| `nmr-2ring-5mm-tube` | 246 | 22 Hz | 6.2 | 14.7 ms |
+| `nmr-3ring-21magnets` | 602 | 1024 Hz | 140 | 0.31 ms |
+| `nmr-2ring-10magnets` | 9396 | 5878 Hz | 1644 | 0.05 ms |
+
+Against a probe dead time of 10–50 µs, every T2\* above 1 ms loses well under a
+percent of the FID. None of these has the ~1 ppm linewidth chemical shift needs,
+which is what shimming is for; and none of it survives temperature drift, since
+NdFeB moves about −1100 ppm/K.
+
+## Regenerating
+
+```bash
+./regenerate.sh                       # every design
+./regenerate.sh nmr-5ring-10mm-tube   # one of them
+./regenerate.sh --check               # rebuild into a temp dir and compare, changing nothing
+```
+
+`--check` compares the genome and the `verified` figures, not the timings.
+Every design here reproduces exactly for its seed and thread count; across
+different thread counts the genetic operators draw from a thread-partitioned
+random pool and float sums reassociate, so the answers agree on the design but
+need not agree in the last digit.
+
+The `field.csv` files are about 3 MB each, 22 MB in total. They are
+reproducible from `config.json` in seconds to a couple of minutes, so add
+`designs/*/field.csv` to `.gitignore` if you would rather not carry them.
+
+## Running one
+
+```bash
+./build/standalone/Greeter -i designs/nmr-5ring-10mm-tube/optimized.json \
+                           -s designs/nmr-5ring-10mm-tube/snapshot.json
+```
+
+The snapshot opens in the viewer. `optimized.json` needs no editing — the
+metadata block is ignored by the readers, and the `field_of_view` is already the
+sample sphere's bounding box.

designs/imaging-200mm-dsv/config.json

@@ -0,0 +1,59 @@
+{
+ "halbach_optimization": {
+
+  "_comment": "Every value here is the default, so an optimizer run with no configuration file at all is this file. It is the design of HalbachOptimisation/homogeneityOptimisation.py: a cylinder of 23 rings 22 mm apart, each position choosing one of 19 (radius, magnet count) pairs, in two layers 21 mm apart.",
+
+  "ring_count": 23,
+  "ring_separation": 0.022,
+
+  "candidates": [
+   {"radius": 0.148, "count": 50},
+   {"radius": 0.151, "count": 51},
+   {"radius": 0.154, "count": 52},
+   {"radius": 0.156, "count": 53},
+   {"radius": 0.159, "count": 54},
+   {"radius": 0.162, "count": 55},
+   {"radius": 0.165, "count": 56},
+   {"radius": 0.168, "count": 57},
+   {"radius": 0.171, "count": 58},
+   {"radius": 0.174, "count": 59},
+   {"radius": 0.177, "count": 60},
+   {"radius": 0.180, "count": 61},
+   {"radius": 0.183, "count": 62},
+   {"radius": 0.186, "count": 63},
+   {"radius": 0.189, "count": 64},
+   {"radius": 0.192, "count": 65},
+   {"radius": 0.195, "count": 66},
+   {"radius": 0.198, "count": 67},
+   {"radius": 0.201, "count": 68}
+  ],
+
+  "outer_radius_offset": 0.021,
+  "outer_count_offset": 7,
+
+  "order": 1,
+
+  "element": {
+   "size": 0.012,
+   "remanence": 1.3
+  },
+
+  "dsv": 0.2,
+  "resolution": 0.005,
+
+  "symmetry": "octant",
+  "objective": "ppm_bx",
+  "field_model": "cuboid",
+
+  "genetic": {
+   "population": 10000,
+   "generations": 100,
+   "cx_prob": 0.55,
+   "mut_prob": 0.4,
+   "gene_mut_prob": 0.05,
+   "tournament": 3,
+   "elitism": 1,
+   "seed": 42
+  }
+ }
+}

designs/imaging-200mm-dsv/convergence.csv

@@ -0,0 +1,102 @@
+generation,best_ppm,best_ever_ppm,mean_ppm,seconds
+0,8101.52,8101.52,136435,7.3e-08
+1,7849.12,7849.12,90612.5,0.202187
+2,7461.71,7461.71,69850.3,0.197998
+3,5674.92,5674.92,57813.9,0.204368
+4,5186.34,5186.34,49662.2,0.209631
+5,4779.94,4779.94,44547.3,0.209373
+6,3283.93,3283.93,40454.3,0.246036
+7,1974.54,1974.54,38118.9,0.19608
+8,1974.54,1974.54,34980.2,0.22197
+9,1974.54,1974.54,32737,0.210893
+10,1974.54,1974.54,30882.5,0.190113
+11,1974.54,1974.54,29640.9,0.221356
+12,1974.54,1974.54,29505.5,0.205366
+13,1974.54,1974.54,28744.8,0.203065
+14,1974.54,1974.54,28184.3,0.21039
+15,1974.54,1974.54,27868.5,0.214157
+16,1974.54,1974.54,27315.4,0.203219
+17,1974.54,1974.54,27648.9,0.199746
+18,1312.29,1312.29,26738.1,0.22743
+19,1312.29,1312.29,26422.5,0.209309
+20,1312.29,1312.29,25514.4,0.262848
+21,907.155,907.155,26146.8,0.217538
+22,907.155,907.155,24771.9,0.230303
+23,907.155,907.155,24689.3,0.233633
+24,907.155,907.155,24079.5,0.276177
+25,907.155,907.155,22755.8,0.234103
+26,907.155,907.155,22347.6,0.209855
+27,907.155,907.155,21622.7,0.266339
+28,907.155,907.155,21059.3,0.272571
+29,842.94,842.94,19912.2,0.265702
+30,806.439,806.439,18900.1,0.221049
+31,806.439,806.439,17564.1,0.245743
+32,806.439,806.439,17029.7,0.24296
+33,806.439,806.439,14930.5,0.337188
+34,806.439,806.439,14745.6,0.325866
+35,806.439,806.439,12782.7,0.280995
+36,806.439,806.439,11514.9,0.224355
+37,806.439,806.439,10119.1,0.204933
+38,806.439,806.439,9513.87,0.283719
+39,806.439,806.439,9128.96,0.262237
+40,794.036,794.036,9267.33,0.277736
+41,794.036,794.036,9428.34,0.205629
+42,794.036,794.036,9704.23,0.226392
+43,794.036,794.036,9449.83,0.179688
+44,794.036,794.036,9333.96,0.193564
+45,794.036,794.036,8945.12,0.182924
+46,794.036,794.036,9246.31,0.190069
+47,794.036,794.036,9595.05,0.213226
+48,794.036,794.036,9406.76,0.179147
+49,794.036,794.036,9423.32,0.20157
+50,794.036,794.036,9456.54,0.208474
+51,794.036,794.036,9677.27,0.205819
+52,794.036,794.036,9208.88,0.232321
+53,794.036,794.036,9383.96,0.232358
+54,794.036,794.036,9616.24,0.235049
+55,794.036,794.036,9287.4,0.260572
+56,794.036,794.036,9158.29,0.18533
+57,794.036,794.036,9264.26,0.214221
+58,794.036,794.036,9354.95,0.181662
+59,794.036,794.036,9458.93,0.172563
+60,794.036,794.036,9440.98,0.184434
+61,794.036,794.036,8932.82,0.204836
+62,794.036,794.036,9592.97,0.211893
+63,794.036,794.036,9339.08,0.190144
+64,794.036,794.036,9303.53,0.182467
+65,794.036,794.036,9303.94,0.192508
+66,794.036,794.036,9445.74,0.211745
+67,794.036,794.036,9932.23,0.204642
+68,794.036,794.036,9408.34,0.184112
+69,794.036,794.036,9199.08,0.189697
+70,794.036,794.036,9245.36,0.18813
+71,794.036,794.036,9481.58,0.238724
+72,794.036,794.036,9326.61,0.275781
+73,794.036,794.036,9587.35,0.24268
+74,794.036,794.036,9399.89,0.198247
+75,794.036,794.036,9683.95,0.190395
+76,794.036,794.036,8892.99,0.212416
+77,794.036,794.036,9325.38,0.190637
+78,794.036,794.036,8853.66,0.190838
+79,794.036,794.036,9780.71,0.189715
+80,794.036,794.036,9323.76,0.19414
+81,794.036,794.036,9255.59,0.211951
+82,794.036,794.036,8741.57,0.207375
+83,794.036,794.036,9066.54,0.191849
+84,794.036,794.036,9082.9,0.197612
+85,794.036,794.036,9900.35,0.17581
+86,794.036,794.036,8826.4,0.2119
+87,794.036,794.036,9418.23,0.213331
+88,794.036,794.036,9819.57,0.194381
+89,794.036,794.036,9661.89,0.184161
+90,794.036,794.036,9152.66,0.193908
+91,794.036,794.036,9900.45,0.203714
+92,794.036,794.036,9691.71,0.219189
+93,794.036,794.036,9422.76,0.185709
+94,794.036,794.036,9715.7,0.180585
+95,794.036,794.036,9548.24,0.197988
+96,794.036,794.036,9482.2,0.198148
+97,794.036,794.036,9332.02,0.214273
+98,794.036,794.036,9206.61,0.19632
+99,794.036,794.036,9419.78,0.195931
+100,794.036,794.036,9062.5,0.192782