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

Screen a material’s impurities

Which impurities in an alloy matter, at which cooling times, for which concern, and how much of each can you tolerate? actinv impurities answers with one command and one input. It solves each element of the alloy once as a pure material, forms every activity, heat, clearance-index and waste-class table sum of any composition as a weighted sum of those runs, and reports rankings, limits and worst cases over the impurity ranges you declare. Every number it reports is then checked by a full solve of the material at that composition.

actinv impurities impurities.json impurities-result.json
actinv impurities impurities.json impurities-result.json --no-verify

The example below is examples/impurities_ra_steel/: a reduced-activation ferritic-martensitic steel in the fusion first-wall spectrum of examples/optimize_ra_steel, after two full-power years at a total neutron flux of 5e14 n/cm2/s, at 1 d, 1 y, 10 y and 100 y of cooling. Its values are illustrative. They are not a qualified material specification, a design limit or a regulatory position.

Describe the screen

The document impurities.json uses the schema actinv-impurities-1:

{
 "schema": "actinv-impurities-1",
 "base_spec": "base_spec.json",
 "balance": "Fe",
 "matrix": {"Cr": 9.0, "W": 1.1, "Mn": 0.4, "V": 0.2, "Ta": 0.12, "C": 0.11},
 "impurities": {
  "Co": {"nominal": 0.005, "min": 0.0, "max": 0.01},
  "Ni": {"nominal": 0.005, "min": 0.0, "max": 0.05},
  "Nb": {"nominal": 0.001, "min": 0.0, "max": 0.005},
  "Mo": {"nominal": 0.005, "min": 0.0, "max": 0.02},
  "Cu": {"nominal": 0.005, "min": 0.0, "max": 0.02},
  "Al": {"nominal": 0.005, "min": 0.0, "max": 0.01},
  "N": {"nominal": 0.03, "min": 0.015, "max": 0.045},
  "Ag": {"nominal": 0.0001, "min": 0.0, "max": 0.0005}
 },
 "targets": [2, 3, 4, 5],
 "endpoints": [
  {"endpoint": "activity.total"},
  {"endpoint": "heat.total", "limit": 1.4e-2},
  {"endpoint": "dose.contact_proxy"},
  {"endpoint": "activity:Nb94", "limit": 100.0},
  {"endpoint": "clearance_index"},
  {"endpoint": "waste_class", "class": "C"}
 ],
 "waste": {
  "rules": "us-nrc-10cfr61.55-v1",
  "mass_g": 1000.0,
  "density_g_cm3": 7.8,
  "waste_type": "activated_metal",
  "external_tritium": {"status": "not_applicable"}
 }
}
  • base_spec is an actinv-spec-1 run (a path relative to this file, or an embedded object). Its flux, schedule and data are used; its material is replaced. The command forces mode: coupled and prune: reach, the configuration proven additive, and records that. The example’s schedule is two years of irradiation followed by cooling steps, so result steps 2 to 5 end at 1 d, 1 y, 10 y and 100 y of cooling.
  • balance takes the remainder to 100 wt%. matrix holds the fixed constituents. Each impurity gives wt% as nominal, min and max (min ≤ nominal ≤ max), or one number for all three.
  • targets are one-based result step numbers, as in actinv clearance.
  • Each endpoint is a name or an object with endpoint, an optional limit and, for waste_class, the class A, B or C you require. activity.total is in Bq/g, heat.total in W/g, activity:<nuclide> in Bq/g, dose.contact_proxy in Gy/h, clearance_index is the IAEA index (limit 1, or your own limit; the table can be replaced with limits, as for actinv budget). A limit on waste_class is refused: its limits are the Part 61 table sums.
  • waste is required by a waste_class endpoint: the rule pack, mass, one geometry denominator, waste type and external-tritium declaration, as for actinv waste budget. Nuclide properties are derived from the run’s decay data.
  • dose.contact_proxy needs the base spec’s photon section with a photon response that has attenuation curves for every element in the material.

The command refuses, and writes nothing, when the response would not be linear in composition: self_shielding, options.screen, a schedule feed, power-driven steps and an uncertainty section in the base spec (first-order bands do not compose across element runs; nuclear-data uncertainty is the sampling block, supported below). It also refuses a min above nominal or a nominal above max, an unknown endpoint, dose.contact_proxy without a photon section and a limit on waste_class.

Read the result

The result actinv-impurities-result-1 holds, for each target step, one entry per endpoint. From the repository root, with the data installed:

actinv impurities examples/impurities_ra_steel/impurities.json impurities-result.json

Output, the cobalt entry of activity.total at step 2 (1 d of cooling), in Bq/g:

            "Co": {
              "contribution_at_max": 310378679.26071066,
              "contribution_nominal": 155189339.63035533,
              "gradient_per_wt_pct": 31037867926.071064,
              "max_wt_pct": 0.01,
              "min_wt_pct": 0.0,
              "nominal_wt_pct": 0.005,
              "top_nuclides": [
                [
                  "Co58",
                  165792519.95582134
                ],
                [
                  "Co58m1",
                  17348137.37118505
                ],
                [
                  "Fe59",
                  10081400.746684032
                ]
              ]
            },

Cobalt at its nominal 0.005 wt% adds 1.6e8 Bq/g at one day, mostly Co-58 made by (n,2n) on Co-59.

For a scalar endpoint the entry gives:

  • matrix_only, nominal and worst_case: the value with the matrix alone, at the nominal composition and at the worst case over the declared ranges. worst_case.corner names, for each impurity, whether the worst case takes min or max.
  • impurities.<element>: the gradient_per_wt_pct at nominal, the contribution_nominal (the endpoint at nominal minus the endpoint with that impurity removed and the balance element taking its place), the contribution_at_max (the impurity at its range maximum against removed, the others at nominal) and the top_nuclides behind the contribution: the three largest per-nuclide shares of the contribution’s sign. For an impurity that lowers the endpoint, they are the balance element’s nuclides it displaces. ranking lists the impurities by contribution at nominal. top_nuclides is null for heat.total, which the run does not resolve by nuclide.
  • With a limit: limits.alone (the impurity is the only one present), limits.others_at_nominal and limits.others_at_worst (the others at their worst-case values), each with a wt_pct and a status, and joint_k, the factor by which every impurity’s nominal weight can be scaled to reach the limit.
actinv impurities examples/impurities_ra_steel/impurities.json impurities-result.json

Output, the cobalt entry of heat.total at step 2, in W/g against the 1.4e-2 W/g limit:

            "Co": {
              "contribution_at_max": 0.00005761499161947032,
              "contribution_nominal": 0.00002880749580973516,
              "gradient_per_wt_pct": 0.005761499161947032,
              "limits": {
                "alone": {
                  "status": "limit",
                  "wt_pct": 0.06871907545895065
                },
                "others_at_nominal": {
                  "status": "limit",
                  "wt_pct": 0.06435931557971897
                },
                "others_at_worst": {
                  "status": "limit",
                  "wt_pct": 0.04183787399387127
                }
              },
              "max_wt_pct": 0.01,
              "min_wt_pct": 0.0,
              "nominal_wt_pct": 0.005,
              "top_nuclides": null
            },

With the other impurities at nominal, cobalt may rise to 0.064 wt% before the steel’s heat one day after shutdown reaches the limit; with the others at their worst case, to 0.042 wt%.

A status is limit (the number exists), no response (the impurity does not raise the endpoint, or the limit lies beyond what the balance element can give up) or infeasible (the other constituents alone already reach the limit). For the contact proxy a status unavailable means the proxy is null there and the entry says why.

The summary gives, for each endpoint, the tightest_limit of each impurity across targets with the step that binds it, the impurities ranking by their largest contribution across targets, and the worst value or class across targets.

actinv impurities examples/impurities_ra_steel/impurities.json impurities-result.json

Output, the tightest heat.total limits with the others at nominal, from summary:

        "others_at_nominal": {
          "Ag": {
            "status": "limit",
            "step": 2,
            "wt_pct": 0.12560332896898269
          },
          "Al": {
            "status": "limit",
            "step": 2,
            "wt_pct": 0.1378665895018817
          },
          "Co": {
            "status": "limit",
            "step": 2,
            "wt_pct": 0.06435931557971897
          },
          "Cu": {
            "status": "limit",
            "step": 2,
            "wt_pct": 2.073400347459034
          },
          "Mo": {
            "status": "limit",
            "step": 2,
            "wt_pct": 0.8845944427497271
          },
          "N": {
            "status": "no response",
            "step": null,
            "wt_pct": null
          },
          "Nb": {
            "status": "limit",
            "step": 2,
            "wt_pct": 0.1891269323069253
          },
          "Ni": {
            "status": "limit",
            "step": 2,
            "wt_pct": 0.16134111838932721
          }
        },

Every heat limit binds at one day of cooling, where the iron, chromium and manganese products still dominate. Nitrogen gives no response. In place of iron it lowers the heat up to 10 y, and at 100 y the limit lies beyond what the balance element can give up.

The waste class

A waste_class entry reports nominal_class, and worst_case with the class over the ranges and its binding table, column and nuclides. Each Part 61 table sum is linear in composition, so each is maximised over the range box at its own corner and the class is read from the maximised sums. That is a conservative bound, since the class never decreases when a sum increases. The impurities’ class limits are those actinv waste budget gives, in summary.<endpoint>.impurity_class_limits; an open boundary is a supremum, not an attained maximum.

actinv impurities examples/impurities_ra_steel/impurities.json impurities-result.json

Output, the start of the waste_class:C entry’s worst_case at step 2:

          "worst_case": {
            "binding": {
              "column": null,
              "constraint": "table_1",
              "corner": {
                "Ag": "max",
                "Al": "max",
                "Co": "max",
                "Cu": "max",
                "Mo": "max",
                "N": "max",
                "Nb": "max",
                "Ni": "max"
              },
              "normalized_sum_max": 0.7039397258233535,

With every impurity at its maximum, the Table 1 sum reaches 0.70 of its Class C limit, so the steel stays Class C over the declared ranges. The entry’s ranking names the impurities behind it: nitrogen (C-14), then molybdenum and niobium (Nb-94).

Add nuclear-data uncertainty

Everything above uses the nominal nuclear data. Add a sampling block and every value and limit also gets a bound over the nuclear-data uncertainty. The worked example examples/impurities_ra_steel/impurities_uncertainty.json is impurities.json plus two keys:

 "confidence_level": 0.95,
 "sampling": {
  "channels": {"cross_section_mf33": true, "decay_constants": true},
  "covariance": {"path": "catalog:tendl-2025-neutron-709g-covariance"},
  "samples": 72,
  "seed": 158001
 }
actinv impurities examples/impurities_ra_steel/impurities_uncertainty.json impurities-uncertainty-result.json

sampling is the block of actinv waste budget uncertainty: samples from 20 to 4096, a seed, the channels and covariance of study.robustness, and an optional keep_samples folder, relative to this file. confidence_level runs from 0.5 to 0.999 and defaults to 0.95. The composition channel is refused, because composition is what the screen varies. So are a base spec that enables gas production, declares its own rate, decay or yield scales, or switches library, projectile or spectrum during the schedule, and a confidence_level without sampling. Every refusal is made before any solve, and nothing is written.

Method. At one fixed draw of the data the activation is again linear in composition, so the algorithm above applies to that draw unchanged. The command draws the factors of every sample once, from the sampling plan of the full alloy at the plan composition (the matrix, every impurity at its range maximum and the balance element taking the remainder, so every element the screen can contain is in the plan), and applies the same factors to every element solve of that sample. It then runs the whole screen on the sample’s element responses: the matrix-only, nominal and worst-case value of every endpoint at every target, each impurity’s contribution at nominal, the limits and joint_k, and for waste_class the nominal class, the worst class and the per-impurity class limits. The fields of the nominal screen are unchanged; the bands are added beside them.

Bounds. With n samples, k is the largest index for which the lower end of the two-sided Clopper-Pearson interval at the confidence level of (n − k + 1)/n is at least the confidence level. A value is reported by its median and its upper bound, the k-th largest per-sample value. A limit is reported by its lower bound, the k-th smallest per-sample limit. These are order-statistic (Wilks) tolerance bounds: exact for any distribution, one-sided, and they need no normality. At least 72 samples are needed at confidence 0.95; a document with fewer is refused, and the refusal names the minimum for your level (528 at 0.99).

Reading the result. Each per-target endpoint entry gains an uncertainty object:

  • matrix_only, nominal and worst_case, and each impurity’s contribution_nominal: the median, the upper_bound and its status (bounded, or unbounded when a failed sample or an unavailable proxy reaches the rank), with the value of every sample in by_sample.
  • With a limit, each impurity’s limits.alone, limits.others_at_nominal and limits.others_at_worst, and joint_k: the bound, the status and the per-sample limits (by_sample, sorted). A sample whose limit lies beyond the balance element’s reach enters at that cap (capped); a sample that fails or is infeasible enters at 0, so it can only lower the bound. The limit is not_established when any sample (or the nominal data) has a positive lower bound, and unbounded when the k-th smallest has no response.
  • For waste_class, the frequency of each class at nominal and over the ranges, and meets_target: the probability that the target class is met at nominal and at the worst case, with its Clopper-Pearson lower bound at the confidence level. Each impurity’s class limit, over all targets, is under uncertainty.impurity_class_limits.
  • ranking: for each impurity the fraction of samples in which it is the largest contributor at nominal.

The top-level uncertainty block gives the confidence_level, the order_statistic (k, n and minimum_samples), the sampling it ran, the plan composition, the coverage of the sampled channels, the failed_samples, not_claimed and verification_pass. With keep_samples, each sample writes sample_NNNN.json: its factor block, the factors each element solve actually carried, the element responses used and the sample’s screen values, enough to recompute every band.

Verification. Unless --no-verify is given, the nominal checks above still run. For every established limit, the samples at sorted ranks k − 2 to k + 2 are re-solved with the full alloy and that sample’s own factors at 0.99 and 1.01 times their own limit: the 0.99 point must meet the limit and the 1.01 point must not, unless the limit is capped. For every upper bound, the sample that holds it is re-solved at the composition the value belongs to (its own worst-case corner for worst_case) and must agree to 1e-6 relative. Differences smaller than 1e-12 times the endpoint’s largest value in that sample are ignored. That is round-off, such as a trace nuclide that one solve keeps at 1e-162 Bq/g and another sets to zero. A point decided this way records rel_dev_before_floor and floor. The results are in verification.uncertainty, and a miss exits 3 after writing.

The worked example’s 72 samples take about 45 minutes on one laptop core: 15 element solves per sample, 1,080 in all, plus 356 re-solves that verify the bounds. Its heat at step 2 (1 d of cooling) shows what the bounds add. With the nominal data, the worst case, 1.38e-2 W/g, is just under the 1.4e-2 W/g limit. Over the nuclear-data uncertainty, the steel exceeds that limit in 30 of the 72 samples even without its impurities. The 95% upper bound on the worst case is 2.5e-2 W/g, and the cobalt limit’s lower bound is 0: no cobalt share meets the limit at 95% confidence.

actinv impurities examples/impurities_ra_steel/impurities_uncertainty.json impurities-uncertainty-result.json

Output, the end of uncertainty.worst_case in the heat.total entry at step 2, after its 72 per-sample values:

              ],
              "median": 0.013621176480236313,
              "order_statistic_rank": 1,
              "status": "bounded",
              "upper_bound": 0.025141681784584535,
              "upper_bound_sample": 68,
              "valid_samples": 72
            }

What the bands do not claim. They are conditional on the sampled channels and assume, per sample, that the activation response is linear in composition, as the nominal screen does. Composition, geometry, flux-shape and model uncertainty are not sampled. Measured coverage of these channels on the FNS corpus is below nominal (docs/DATA_LIMITATIONS.md), so a bound can be optimistic. The upper bound on a value and the lower bound on a limit are one-sided order-statistic tolerance bounds. They are not two-sided intervals and not a design margin. The worked example’s composition and history are illustrative.

In the desktop app and the browser

The Impurities page, under STUDIES, screens the material of the problem that is open. Composition lists its elements: choose the Balance (the largest at first), keep an element Fixed at its share from Material, or make it an Impurity with a nominal, minimum and maximum in wt%. Add an impurity the material does not list by its symbol. When selects the steps to screen; every cooling step after the last irradiation is selected at first. What matters chooses the endpoints, each with an optional limit, the activity of single nuclides, and the waste class with the class to meet; the waste class uses the material’s mass and density as the component. Nuclear-data uncertainty is optional. It adds the sampling block of Add nuclear-data uncertainty: a confidence of 90, 95 or 99%, the number of samples (at least the minimum for that confidence), cross sections, decay constants or both, and a seed. The page shows how many calculations that takes before you start. The page needs a composition in wt%, and it explains on the page what it refuses, as the command does.

Screen impurities runs the screen in the background, with verification, and shows the result: whether it was verified, an overview with each endpoint’s worst case, largest contributor and tightest limit over all screened steps, and for a chosen step, each endpoint’s nominal, impurity-free and worst-case values with every impurity’s contribution, the nuclides it acts through and its limits. A screen with nuclear-data uncertainty shows each bound in grey beside its nominal figure: an upper bound on each value and contribution, a lower bound on each limit. A bound that only restates the nominal figure, such as no limit for an impurity that does not raise the endpoint, is left out. When the material exceeds a limit without impurities in some samples, the card says so once, instead of a 0 beside every limit. For the waste class, it shows how many samples meet the target class and the lower bound on that probability. Save screen… writes the actinv-impurities-1 document with the problem embedded, and Save result… and Open result… write and read results, including those of actinv impurities. The browser sets up screens and reads results but does not run them.

What to keep in mind

  • Linearity. At fixed flux the coupled activation is linear in the initial composition. That holds without self-shielding, screening or feed; it is a property of the tested neutron spectra and schedules, not a general one. Ranges that move the composition far from the tested mixtures are not covered by that evidence.
  • Contact dose is a slab proxy. dose.contact_proxy is the semi-infinite-slab screening proxy, not a dose for a geometry. It is not linear, because the material’s own attenuation enters it: the command sums the pure elements’ grouped photon sources and evaluates the proxy at the composition’s attenuation exactly as a run would. Its worst case is a corner maximum over the corners of the range box (at most 12 impurities with a range); an interior maximum is not excluded. Its limits are found by bisection to 1e-9 relative in the impurity’s weight fraction and assume the proxy rises with the impurity. Its gradient is a central difference. A proxy the response cannot evaluate is reported as null with the reason, never as a number.
  • Every number is verified. Unless --no-verify is given, the nominal composition, every worst-case corner and every reported limit is solved in full and compared with the prediction (verification.points, tolerance 1e-6 relative). The command exits 3, after writing the result, if any point misses.
  • The Part 61 class is the nominal U.S. classification, not disposal acceptance.