Repository navigation
Electron table grid: the default 10 points per decade under-resolves the inelastic rate just above the Fermi level on the band-bottom axis #291
Description
Found while working on #241 (inelastic tables of materials with a band built on the band-bottom axis, rows at T = E - E_F; implemented in PR #289). Distinct from #287, which re-runs the committed δ tables: this is about the table grid those runs use.
What is the case
On the band-bottom axis the inelastic rate of a row rises from zero at the Fermi level over the first few eV above it (only losses below E - E_F count, so for Al most of the plasmon weight is excluded there). The shared table grid ([electron.tables] points_per_decade, default 10; min_energy_ev = 5 in validation/experiments/se_yield.py) has points about 3 to 6 eV apart at 15 to 30 eV, and the transport interpolates the rate linearly between them (electron::transport::rate). The secondary yield is sensitive to that interpolation.
Measured (diagnostic, not a gate)
lindhard-cli/examples/inelastic_low_energy.rs, mode clamp (the table lindhard run builds after #241), on the Al 400 eV default input that se_yield.py writes, 200 histories, seed 1, release build of a #241 implementation equivalent to #289 for Al (a metal, so E_F is the same under either choice of insulator offset):
| points_per_decade | δ |
|---|---|
| 10 (default) | 6.44 |
| 20 | 6.85 |
| 40 | 7.01 |
| 80 | 7.00 |
The same input with the pre-#241 table (model axis, Fermi energy 0) moves less: 6.89 at 10, 7.13 at 40. So on the new axis the default grid leaves a grid error of about 8 % in δ at this point, larger than the statistical error of the committed δ tables (200 histories here, about ±0.1 to 0.2).
Acceptance criteria
- Decide how the inelastic table grid resolves the region just above the Fermi level for a material with a band (for example a denser default, or extra inelastic grid points from
E_Fupward), and document the choice indocs/cli.md. - A convergence check (δ, or the interpolation error of the rate) showing the default within a stated tolerance.
- Coordinate with Re-run the secondary-yield tables and the electron oracle summaries with the band-bottom inelastic tables of #241 #287 so the re-run δ tables use the settled grid.
Activity
rjwalters commentedon Oct 10, 2026
Sweep (
sweep-20261010T000141Z-35105-6f5a1405): parkedloom:blockedon #241 (park record appended to the body). Concerns the table grid on the band-bottom axis introduced by #241, which is not on main yet. The prerequisite is in PR #289, which is approved but held by Champion for a human merge (loom:operator). Once it merges and #241 closes, this issue is ready to curate and build.rjwalters commentedon Oct 10, 2026
Cited blocker cleared: #241 just closed, and this issue's
loom:blockedcites it. The block now reads as stale:Re-check whether the block still applies and remove
loom:blockedif not. Posted at merge time byloom-daemon notify-cleared-blockers(#9102), rather than waiting for the nextcheck-stale-blockedsweep pass (#8927).loom-fleet-dispatch commentedon Oct 10, 2026
Unblocked by the daemon's deterministic release pass (#10556): every blocker declared in this issue's park record has resolved (#241). Removed
loom:blocked.loom:issuewas not restored: this issue never carried it before the block, so it re-enters triage/approval rather than the Builder queue.loom dashboard
rjwalters commentedon Oct 10, 2026
Sweep (
sweep-20261010T004216Z-6564-14875022): blocker #241 closed (PR #289 merged). Removedloom:blocked; back toloom:triagefor curation.rjwalters commentedon Oct 10, 2026
This issue is being worked by an in-session
/loom:sweeprunsweep-20261010T004216Z-6564-14875022on hosthost-d9142cf3, published at 2026-10-10T09:27:43Z. This comment is a lease record (Issue #6320, Epic #6165) in the identical formatloom-daemon's dispatch-time write uses (#6179) — its liveness signal is this comment's own forge-assignedupdated_at, never the timestamp in this text. It is renewed bysweep-lease-renew.sh(#6180) for as long as this sweep's own process is alive, and simply ages out afterwards. Readers: the daemon's reclamation gate (#6286) refuses to reclaim this claim while the lease is fresh, dispatch-time ordering (#6287) yields to it, andsweep-lease-fence.sh(#6309) fences this sweep's own push/PR-open against it. Seedefaults/docs/lease-record.md.rjwalters commentedon Oct 10, 2026
Curator Enhancement (2026-10-10)
Premise check on current
origin/main(read-only; #241/PR #289 is merged).EnergyAxis::BandBottom(lindhard/src/electron/inelastic/table.rs,model_energies_ev,loss_rows) stores the row atEas the model atT = E - E_F. Rows withT <= 0are empty, and losses stop atE - E_F. So the rate starts at 0 atE_Fand climbs steeply over the first few eV, and the transport interpolates it linearly in energy between grid rows (lindhard/src/electron/transport.rs, module docs line ~20). Chords of a convex, rising curve overestimate nothing but under-resolve the knee, so a coarse grid biases the rate low there. That fits the table in the body (δ 6.44 at 10 pts/decade, 7.00 at 80).DEFAULT_POINTS_PER_DECADE= 20 (docs/cli.md[electron.tables],min_energy_ev10). The 10 comes fromvalidation/experiments/se_yield.py(min_energy_ev = 5.0,points_per_decade = 10.0, ~line 328). By the body's own numbers, 20 gives 6.85 vs 7.01 at 40, about 2 %, inside the stated ±0.1 to 0.2 noise. The ~8 % error belongs to the validation input, not the library default.log_energy_gridinlindhard/src/electron/elastic/table.rs; built inlindhard/src/input/electron.rs~line 1300). Any inelastic-only refinement has to respect that sharing and the table-cache grid check, so a code change is the more invasive option.Bounded deliverable (no full Penn, no transport re-runs of δ(E), minutes of compute).
ratemode tolindhard-cli/examples/inelastic_low_energy.rs. For Al and Si (single-pole and Mermin of the input's ELF only), build the band-bottom inelastic table on a dense reference grid (for example 200 pts/decade overE_FtoE_F + 40 eV) and on the grids of 10, 20 and 40 pts/decade. Report the linear-interpolation error of1/λagainst the dense table atE - E_F= 0.5, 1, 2, 3, 5, 10, 20 eV. One table build per grid, no transport.E - E_F>= 2 eV; the region below that is dominated by one grid cell and is reported, not gated).se_yield.pyto the default grid (or 40) and document indocs/cli.mdthat a material with a band wants >= 20 pts/decade, naming the measured error. Bandless tables are untouched by construction.docs/cli.md. Do not change the library default unless the table build time at 40 is shown acceptable (one timing line).E_Fupward on the band-bottom axis. Not part of this issue, since it touches the shared-grid and cache contract.docs/validation.mdnext to the Re-run the secondary-yield tables and the electron oracle summaries with the band-bottom inelastic tables of #241 #287 note (~line 1947).Stop rule. No tuned constants and no per-material grid fits; the grid is chosen from the stated tolerance, not to match an experimental δ. No δ(E) or yield re-runs here (that is #287). No full Penn. Any engine change must leave a bandless material's tables byte-identical and keep results identical across thread counts.
Relation to #287. #287 re-runs the committed δ tables with these tables, so this issue must land first if it changes the validation grid. If it concludes "no change", #287 proceeds unblocked, and the conclusion should say so on #287.
Acceptance criteria
1/λon the band-bottom axis for Al and Si at 10, 20 and 40 pts/decade against a dense reference (single-pole, and Mermin if cheap).se_yield.py) is shown within it, or the validation input is raised until it is.docs/cli.mdstates the grid guidance for materials with a band, and corrects "default 10" where it appears.docs/validation.mdrecords the numbers and the hand-off to Re-run the secondary-yield tables and the electron oracle summaries with the band-bottom inelastic tables of #241 #287.Affected Files
lindhard-cli/examples/inelastic_low_energy.rsvalidation/experiments/se_yield.pyvalidation/experiments/test_se_yield.pydocs/cli.mddocs/validation.mdTest plan. Run the new example mode locally once (release build, single table builds). If
se_yield.pychanges its grid, updatetest_se_yield.pyfor the written input. No engine change expected, so existing table and determinism tests (lindhard/tests/electron_bench_determinism.rs) are unaffected.9 remaining items
loom-fleet-dispatch commentedon Oct 10, 2026
Re-parked by the daemon's deterministic release pass (#10556): declared blocker(s) #241 resolved, so their park records were removed from the body. Still blocked by #339, #340;
loom:blockedstays.loom dashboard
rjwalters commentedon Oct 10, 2026
This issue is being worked by an in-session
/loom:sweeprunsweep-insession-s34613-FriOct91741382026on hosthost-d9142cf3, published at 2026-10-10T10:55:31Z. This comment is a lease record (Issue #6320, Epic #6165) in the identical formatloom-daemon's dispatch-time write uses (#6179) — its liveness signal is this comment's own forge-assignedupdated_at, never the timestamp in this text. It is renewed bysweep-lease-renew.sh(#6180) for as long as this sweep's own process is alive, and simply ages out afterwards. Readers: the daemon's reclamation gate (#6286) refuses to reclaim this claim while the lease is fresh, dispatch-time ordering (#6287) yields to it, andsweep-lease-fence.sh(#6309) fences this sweep's own push/PR-open against it. Seedefaults/docs/lease-record.md.rjwalters commentedon Oct 10, 2026
Cited blocker cleared: #340 just closed, and this issue's
loom:blockedcites it. The block now reads as stale:Re-check whether the block still applies and remove
loom:blockedif not. Posted at merge time byloom-daemon notify-cleared-blockers(#9102), rather than waiting for the nextcheck-stale-blockedsweep pass (#8927).loom-fleet-dispatch commentedon Oct 10, 2026
Re-parked by the daemon's deterministic release pass (#10556): declared blocker(s) #241, #340 resolved, so their park records were removed from the body. Still blocked by #339;
loom:blockedstays.loom dashboard
rjwalters commentedon Oct 10, 2026
Cited blocker cleared: #339 just closed, and this issue's
loom:blockedcites it. The block now reads as stale:Re-check whether the block still applies and remove
loom:blockedif not. Posted at merge time byloom-daemon notify-cleared-blockers(#9102), rather than waiting for the nextcheck-stale-blockedsweep pass (#8927).loom-fleet-dispatch commentedon Oct 10, 2026
Unblocked by the daemon's deterministic release pass (#10556): every blocker declared in this issue's park record has resolved (#241, #339, #340). Removed
loom:blocked. Restoredloom:issue.loom dashboard
rjwalters commentedon Oct 10, 2026
Sweep: blocker #339 closed (PR #355 merged — band-bottom rate refinement now on in
lindhard run). Removedloom:blocked; back to triage.loom-fleet-dispatch commentedon Oct 10, 2026
Champion Review: NEEDS REVISION
Curator should reconcile the acceptance criteria with current main and #339/#341, then record a terminal disposition if the work is complete or define the specific remaining hand-off to #287. Keep the curated marker; route to Curator rather than queue duplicate engine work.
Automated by Champion role
loom dashboard
rjwalters commentedon Oct 10, 2026
Sweep: resolved by PR #341 (diagnosis, Part of #291) + PR #355 (closes #339): band-bottom inelastic tables are now refined in
lindhard rununtil linear interpolation of the rate stays within the 1 % tolerance over E−E_F ∈ [2, 20] eV; every acceptance cell passes at the default grid (10 eV, 20 ppd) and these_yield.pygrid (5 eV, 10 ppd) for Al and Si, single-pole and Mermin (table indocs/validation.md, 'Table grid above the Fermi level'). The Si Mermin refusals (#340) are fixed. The δ(E) re-run on these tables is #287.