Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Unfold a neutron spectrum from foils

A set of activation foils irradiated in an unknown field records that field in its activities. actinv unfold turns the measured activities back into the group fluxes that made them, starting from a prior spectrum you supply, and tells you how well each energy range is now known.

actinv unfold problem.json measurements.json unfold-result.json

Unfolding codes usually expect activities already reduced to saturated reaction rates, leaving the irradiation history, decay during irradiation, burn-up and multi-step production to you. actinv unfold models each foil with a full ACTINV solve over the irradiation and cooling schedule you declare, so those effects are part of the model. Data you have already reduced can be given as reaction rates instead, and the two kinds can be mixed.

An example

examples/unfold_foils/ holds a nickel and an aluminium foil irradiated for one hour and counted after one hour of cooling. The prior is the JAEA-FNS iron spectrum of examples/fns_fe_5min.json. The measured values are illustrative, chosen to be consistent with a spectrum somewhat harder than the prior.

actinv data fetch
actinv unfold examples/unfold_foils/problem.json examples/unfold_foils/measurements.json unfold-result.json

It runs four solves (one per foil, two passes) in about 20 seconds once the data are installed, and prints chi-square 4.074 on 4. All four measurements respond mostly near the 14 MeV source peak, so the result shows what foils can and cannot tell you:

BandPosterior / prior fluxStandard deviation, prior → posterior
below 0.5 eV1.00018.5% → 18.5%
0.5 eV – 100 keV1.00316.3% → 16.3%
0.1 – 1 MeV1.04722.7% → 22.6%
1 – 10 MeV1.25722.0% → 17.3%
above 10 MeV1.07029.5% → 1.8%

Above 10 MeV the flux is now known to 2%. The Co58 and Ni58(n,p) responses start near 5 MeV, which moves and narrows the 1–10 MeV band a little. Below 1 MeV nothing responds, and the result is still the prior.

The problem: prior, foils and history

problem.json is an ordinary actinv-spec-1 neutron problem:

  • its spectrum is the prior φ⁰, on the library’s 709 groups (descending and total are honoured);
  • its schedule is the foils’ irradiation and cooling history, with flux multipliers as usual;
  • its material is the default foil material (each foil may declare its own);
  • its library and decay data are used for every foil.

Add an uncertainty block holding a covariance sidecar (and optionally confidence_level) to carry the library’s MF=33 cross-section covariance into the fit. Nothing else may appear there: unfold sets the responses and channels itself. Features whose flux is not the spectrum’s (power-driven or wall-loading steps, partitions, step spectra, feed or removal, rebinning) are refused.

The measurements

measurements.json uses the schema actinv-unfold-input-1. The example’s:

{
 "format": "actinv-unfold-input-1",
 "prior": {
  "relative_uncertainty": 0.3,
  "correlation": {"kind": "lethargy_exponential", "length": 1.0}
 },
 "foils": [
  {"id": "ni", "measurements": [
   {"nuclide": "Co58", "step": 2, "activity_Bq_per_g": 5450.0, "sigma_Bq_per_g": 164.0},
   {"nuclide": "Ni57", "step": 2, "activity_Bq_per_g": 52490.0, "sigma_Bq_per_g": 1575.0}
  ]},
  {"id": "al", "material": {"mass_g": 1.0, "basis": "wt_percent", "composition": {"AL": 100.0}},
   "measurements": [
   {"nuclide": "Na24", "step": 2, "activity_Bq_per_g": 1112200.0, "sigma_Bq_per_g": 33370.0}
  ]}
 ],
 "reaction_rates": [
  {"id": "ni58-np", "target": "Ni58", "mt": 103,
   "rate_per_atom_s": 3.39e-15, "sigma_per_atom_s": 1.0e-16}
 ]
}
  • prior. relative_uncertainty is the standard deviation of ln φ_g: one number, or one per group in the spectrum’s order. correlation is {"kind": "diagonal"} or {"kind": "lethargy_exponential", "length": L}, correlating two groups by exp(−|Δu|/L) with u the logarithm of the group energy. Groups with zero prior flux or zero uncertainty are held fixed.
  • foils. Each foil has an id, an optional material, and activities (Bq per gram of foil) at the end of a schedule step, each with its one-standard-deviation sigma_Bq_per_g and an optional model_relative_uncertainty for corrections you have not modelled (self-shielding, cover, geometry). A metastable state may be written In115m or In115m1; the result reports the canonical In115m1.
  • reaction_rates. Reactions per target atom per second at the declared spectrum: a target nuclide, an ENDF mt, and optionally the product state (without it the reaction’s total is used).
  • Optional: report_bands_eV (energy edges of the reported bands; default 1e-5, 0.5, 1e5, 1e6 and 1e7 eV and the top of the structure), emit_covariance (write the posterior covariance of ln φ), and iterations.

Every measurement needs a positive value and a positive σ. A measurement the prior spectrum does not produce is refused, since it carries no information the fit can use.

The result

unfold-result.json (actinv-unfold-1) holds:

  • groups: prior and posterior flux per group, the standard deviation of ln φ before and after, and the resolution 1 − (posterior variance / prior variance): near 1 where the measurements determine the flux, near 0 where the result is still the prior;
  • bands: the flux in each report band with its standard deviation and an interval at the confidence level, and the correlation between bands;
  • measurements: each one’s prior and posterior prediction, its pull before and after the fit, its cross-section uncertainty when a sidecar is given, and its response window, the energies below which 5%, 50% and 95% of its response lies;
  • consistency: the χ² at the solution with one degree of freedom per measurement. A χ² far above the number of measurements says the measurements and the prior disagree more than their uncertainties allow; look at the pulls. Two foils that respond at the same energies but disagree by more than their σ (say Ni57 low and Na24 high, both from the 14 MeV peak) show up here, not as a confident spectrum.

What it does and does not tell you

Unfolding is under-determined. Outside the energies your foils respond to, the posterior is the prior; the resolution and response windows say where that is. The intervals are only as right as the prior uncertainty you declare. Self-shielding and flux perturbation by the foils, covers and counting corrections are inputs (through the foil material, the problem’s options or the model uncertainty), not something unfold estimates.

Method and evidence

The fit is generalized least squares on ln φ with the prior’s covariance, the measurement uncertainties and, with a sidecar, the cross-section covariance. It is solved by damped Newton with the local model’s exact second derivatives; every activity and its 709 derivatives come from one solve of the foil. The posterior covariance is the usual linearized one at the solution.

Protocol P195 and its Amendment A tested it (P195A-PASS):

actinv unfold estimates a neutron spectrum from foil activities, modelled with a full ACTINV solve of each foil’s irradiation history, and from reaction rates, carrying the library’s cross-section covariance into the fit.

On 40 synthetic spectra drawn from its prior, 94% of the true band fluxes fell inside its 95% intervals and 74% inside one standard deviation, and its band fluxes were closer to the truth than the prior’s (root-mean-square log error 0.124 against 0.208).

On IRDFF-II’s measured spectrum-averaged cross sections in the Cf-252 and U-235 fission fields, predicting each reaction left out of the fit, the median |ln(C/E)| was Cf-252-a prior (the other field): 0.321 → 0.087, UNFOLD CLOSER, p = 3e-08; Cf-252-b prior (a Maxwellian): 0.158 → 0.080, UNFOLD CLOSER, p = 0.00068; U-235-a prior (the other field): 0.246 → 0.044, UNFOLD CLOSER, p = 1.3e-05; U-235-b prior (a Maxwellian): 1.031 → 0.048, UNFOLD CLOSER, p = 2e-09; the Cf-252 reference spectrum used similar measurements in its evaluation.

Evidence: record, verdict, protocol and amendment.