# Periodic Table science notes

Version 0.7.1 · 18 November 2026.

## Sample-state banner

`core/sample-state.js` reads the current structural diagnostics. It does not change forces, trajectories, temperatures, energy, atomic populations or the bulk phase reference. Contact/connectivity evidence supports gas-like, mixed or condensed descriptions. An explicitly prepared solid geometry with a dominant condensed cluster and retained neighbour distances supports a qualified solid-like label. The inherited 0.7 share/retention and 0.18 distance-RMS display criteria are heuristics, not calibrated phase boundaries.

Preparation-relative retention cannot distinguish a newly prepared liquid from a crystal: both can initially have zero relative distance change. Actual Hg/Br preparations expose this ambiguity. Only an exact initial declared liquid preparation with a liquid room reference is labelled liquid-like, with preparation-only confidence. Subsequent condensed snapshots remain solid/liquid unresolved unless a separate order and mobility model supplies evidence. A single snapshot of disorder does not establish liquid behaviour. See the primary [LAMMPS orientational-order](https://docs.lammps.org/compute_orientorder_atom.html) and [diffusion](https://docs.lammps.org/Howto_diffusion.html) methods; those calibrated analyses are not supplied by this banner.

The neutral material model has no ion or free-electron populations. The [IUPAC plasma definition](https://goldbook.iupac.org/terms/view/P04687) involves an at least partly ionised gas. Neither a temperature threshold, ionising source nor a drawn glow supplies those missing populations or charge dynamics. Plasma is explicitly unmodelled, including above the classical model's comfortable range.

The details dialog is a snapshot at its labelled time. It separates structural evidence from the bulk comparison near 1 bar. An initial Liquid-like to later Condensed label is an evidence limitation, not a detected freezing transition. Recorded readings retain the snapshot's classification and source settings; electron source frequency is omitted rather than relabelling a retained photon control as an electron frequency. Complete provenance is in `data/sample-state-sources.json`.

## Drawing corrections in 0.7.1

The Atom downward pulse now ends at the selected radiative lower energy, including retained lower excited helium states. Changing wavelength, current, irradiance or electron energy cancels/rebases pending representative paths at the current integrated counts. This prevents an old projectile from acquiring new source metadata while leaving its rate-network populations and event history intact. Crucible's sparse minus marks distinguish charged particles from photons; they do not add a charge/transport calculation. Full transition and lifetime-bound information remains in drawing notes and accessible labels.

## Electron-impact reference added in 0.7.0

The gun can instead specify monoenergetic electrons with energy E (1–3000 eV) and current density j (0–10⁶ A/m²). Number flux is j/e; kinetic power density is jE when E is in eV. The selected excitation coefficient is R=(j/e)σ(E). No electron collision rate is inferred solely from a level gap or optical selection rule.

`core/electron-impact.js` retains the first excitation column of Tables 1–2 in Stone, Kim and Desclaux, J. Res. NIST 107 (2002), [doi:10.6028/jres.107.026](https://nvlpubs.nist.gov/nistpubs/jres/107/4/j74sto.pdf). Published Å² values become m² by multiplication by 10⁻²⁰. Piecewise linear interpolation in incident energy is a display/model approximation between tabulated points, not a new fit or uncertainty estimate. No extrapolation fills the interval from the physical threshold to the first available data. Source Born scaling does not include near-threshold resonances.

Hydrogen's unresolved 1s–2p cross section drives an effective 2p term with g=6. Excitation energy and radiative A are weighted means of the NIST Handbook J=1/2 and J=3/2 levels (weights 2 and 4). One effective Lyman-alpha branch represents that mean; it is not a resolved-doublet prediction. The paper's rounded/scaled source excitation energy is retained as separate metadata. Helium uses its sourced selected singlet transition and existing radiative branches. The independent provenance is in `data/electron-sources.json`.

For an exciting collision, Eout=E−ΔE; recoil is neglected. The incident electron continues and is not added to the bound electron count. A later radiative photon has hν=Eupper−Elower for its branch. The shared analytic population solver uses R as its upward coefficient and zero optical stimulated-emission coefficient. Its historically named absorption counter counts exciting collisions here. It omits superelastic de-excitation and other excitation/ionisation channels; the excitation-to-radiative-rate ratio is exposed. Ionisation can compete above its threshold; the reference does not turn that threshold into a predicted plasma.

The electron display clock is 1 ms per real second, separate from the photon reference's 10 ns/s and the mechanical picosecond clock. A bounded drawing queue delays already counted representative radiation cues until their displayed collision finishes. This changes no populations, rates or energy integrals. Paths, scattering angle and drawn residence times are schematic. Moving pulses along arrows do not establish intermediate occupied energy levels. Intensity changes do not reinterpret historical event counts.

The classical electron receiver assumes complete deposition of admitted kinetic power into the wall (source absorptivity 1), while photon absorption remains 0.8. This is an ideal boundary, not a calculation of material stopping, backscatter, penetration, secondary emission or charge. Both retain the existing numerical guard and wall energy ledger. The electron source is omitted from the EM wavelength axis; thermal and neutral-LTE references remain independent. Atom-reference energy is not added again to the wall or particles.

Group/period filters use the existing reference fields and combine with exact/partial search. Detached f-block elements belong to periods 6/7 and retain an unassigned group field. Filtering never changes the material sample or ground-state configuration.

## Photon and material reference retained

The photon gun has two adjustable physical inputs: vacuum wavelength and irradiance. Standard preparations use 128 atoms with a 293.15 K kinetic start, fixed walls, exterior vacuum and an ideal mounted thermal wall. The initial velocities do not establish thermal equilibrium. Pressure, count and environment settings remain fixed. The beam drives the wall in Crucible and a separate isolated-atom ensemble in the atomic views. Kernel APIs retain separately tested library capabilities.

## Mechanical sample

Copper uses the complete published Mishin EAM1 table from the identified 2007 LAMMPS conversion. Other known forms use labelled shifted-force cohesive or restrained-molecule references; unknown forms use atomic placeholders. The [material notes](material-science-notes.md) retain the force definitions, provenance and limits. These references do not provide calibrated all-element chemistry or phase diagrams.

The density-table interpolant now preserves nonnegative density and its analytic derivative near the cutoff. An earlier unconstrained cubic could produce a tiny negative density from nonnegative table values, falsely leaving the embedding domain. The correction retains all upstream knots and does not extend the force table. Genuine table-domain failures still halt.

Material-wall drift now advances all atoms to the earliest hard-wall event, samples or reflects the outgoing velocity there and continues through the remaining drift interval. The former endpoint reflection used the incoming-speed overshoot after changing the thermal velocity. Wall impulses and signed kinetic-energy exchange are recorded at the actual drift collision time. Forces still use velocity Verlet with bounded substeps.

Earth gravity is 9.80665 m/s². At nanometre distances its energy is far smaller than bonding energy; visible floor contact comes from the preparation and declared substrate, not inflated gravity. A conservative Morse floor supplies assumed weak adhesion. Local stochastic floor contact and flux-weighted wall impacts exchange energy with the finite wall. Molecular bonds retain fixed topology; they cannot dissociate or react.

Preparation retains the relaxed floor contact. Earlier code relaxed with the floor potential, then moved the cluster down onto the hard boundary, partly undoing that relaxation. The seeded copper preparation began with an artificial substrate-energy offset of about 1.092 eV. Retaining the relaxed positions removes that mismatch without changing the published EAM table or adding a thermostat.

## Source and mounted wall

The intensity slider uses an Off detent followed by 10⁻⁶–10¹⁴ W/m², with 100 increments per decade. Wavelength spans 0.001 nm–1000 m on its own logarithmic strip, with exact entry in nm, µm, mm or m. Band boundaries are conventions. The default source remains 30 THz infrared. Both controls store physical SI values; inverse slider mapping does not quantise imported values.

```text
P_source_requested = 0.8 × irradiance × projected face area
C_wall = 3000 kB
P_mount = C_wall × (293.15 K − T_wall) / 25 ps
P_environment_radiation = 0.8 × sigma × total wall area × (293.15⁴ − T_wall⁴)
total wall area = 2(LxLy + LxLz + LyLz)
```

The receiver assumes absorption 0.8 at every selected wavelength. Changing wavelength changes photon energy and number flux, but not this assumed mean absorbed power at a fixed irradiance. This is an ideal calorimetric wall, not a prediction of gamma, X-ray, visible, infrared or radio absorption by the chosen element. Direct photon–atom cross-sections, penetration, transmission, photoelectrons and radiation damage are not added to the MD sample.

The mounting and its relaxation time are assumptions, not measured crucible properties. It supplies a conducting support while the exterior remains vacuum. Grey radiation is an ideal far-field wall approximation; the chamber is smaller than thermal wavelengths, and no nanostructure or near-field correction is inferred. Sigma is derived from SI defining constants. [Crucible sources](data/crucible-physics-sources.json) link the primary method descriptions and constants.

At 4000 K the default copper wall's grey radiative loss is about 3.1 × 10⁻¹⁰ W. This is much smaller than its former maximum absorbed source power of 3.55 × 10⁻⁶ W. Radiation alone cannot honestly create rapid cooling on picosecond clocks. The stated support conductance supplies the larger cooling channel. At room temperature both environmental exchanges have zero net power.

Linear source/reservoir segments are integrated analytically. Grey exchange uses a symmetric implicit-midpoint split with positivity, bounded-work and no-progress guards. Each accepted transfer is recorded with its own sign. `radiationJ` is accepted source input; `mountJ` and `environmentRadiationJ` are signed into the wall. Sample–wall contact is internal to the combined system and is not counted again as external heat. Legacy bath/controller transfers retain separate ledgers.

An 8000 K numerical source guard limits incoming source energy; it never deletes collision energy or stops motion. Delivery resumes when loss/contact makes room for further input. It is not a fitted-potential validity bound, an equilibrium phase boundary or an ionisation threshold. An extended-range notice appears during hot classical exploration. Ionisation depends on density as well as temperature, while molecular bond breaking is absent at every temperature.

Kernel-only environments default to disabled mount/radiative exchange so isolated conservation tests remain meaningful. The public adapter explicitly enables the standard environment. Loading a file rebuilds that boundary; it does not import a different physical environment.

## Temperature and light

For material/cohesive samples, the main value is the centre-of-mass-subtracted kinetic estimate `2K_internal / [(3N−3)kB]`. It is undefined for one atom and need not describe thermal equilibrium. Coherent rotation and expansion can contribute to it. Unknown-form atomic placeholders retain the container-frame estimate `2K_lab / (3NkB)` from their separate bounded gas kernel; their optical notes and exports identify this convention.

For a connected monatomic condensed cluster, the light reference removes the least-squares rigid angular velocity as well. An original dimensionless symmetric-inertia eigensolver resolves the rotational rank, including linear/degenerate geometries. The residual uses `3N−3−rank` degrees of freedom. It reads positions and velocities without changing them. A rigidly spinning cold cluster therefore does not become incandescent merely because of its mechanical rotation. Molecular-gas rotations remain in their kinetic estimate. Neither estimator supplies an electronic population or proves local equilibrium. [Method context](https://docs.lammps.org/compute_temp_rotate.html).

The separate optical reference retains Rakić bulk fits for Cu, Au, Ag and Al, held fixed with temperature, and an assumed grey comparison for other materials. Planck radiance uses `2hc²` in SI and is converted to per nanometre once. It describes unit condensed-reference area, with emissivity applied once. Unsupported optical-fit wavelengths stay null. [Optical ledger](data/optical-sources.json).

The 32 selected NIST neutral transitions across eight species retain rates, levels, angular momenta, original wavelengths and Ritz-vacuum positions. Optional line powers use a truncated neutral-LTE partition and `N_upper Aki hν`. Missing levels, ion stages, collisional excitation and optical depth limit the reference. Molecular samples receive no atomic LTE populations. Sample reference emission is not debited from dynamics; only the ideal wall's explicit radiative exchange affects its ledger. [Spectrum ledger](data/spectrum-sources.json).

Visible colour conversion uses the official tabulated CIE 1931 2° standard observer at 1 nm intervals from 360 to 830 nm. The earlier analytic approximation had a far-red tail error: normalising an extremely weak potassium-line colour could make it green. The table fixes that inversion. Conversion still clips negative linear-sRGB components and normalises hue for fixed-brightness markers. Display gamut, exposure and assumed optics remain limits on appearance; a glyph's brightness is not measured radiance. Nonvisible photon glyphs use labelled false colour. [Colour data, checksums and licence](data/cie-sources.json).

Crucible paths retain the declared receiver's 80% absorption, 20% diffuse reflection and zero transmission. They do not target atoms or calculate photon–atom cross-sections. Nonvisible colour, paths, wavelength spacing, transit speed and packet counts are schematic. Weak power can have zero drawn packet markers while numerical flux remains available. Mean source power is not Poisson sampling of discrete photon arrivals.

## Isolated-atom excitation reference

`core/photon-atom.js` reuses selected NIST transitions and adds sourced ground-state ultraviolet lines for He, Ne and Ar. It drives only transitions with a zero-energy lower state and available decay data. A gap between arbitrary plotted levels does not establish an allowed transition or its strength. The familiar visible noble-gas discharge lines have excited lower states and are not modelled as direct one-photon ground-state absorption. [Photon source ledger](data/photon-sources.json).

The rate reference assumes a stationary neutral atom, statistical mixing within fine-structure manifolds, and an incoherent Lorentzian beam of 1 GHz FWHM. The chosen wavelength is the beam centre; its near-resonant portion drives the transition. Photon flux uses the local narrow-line approximation `I/(h nu_centre)`. It is not a polarisation-resolved or coherent two-state Schrödinger calculation. Context temperature does not add Doppler broadening or set electronic populations. Hyperfine/isotope structure, velocity classes and competition between multiple simultaneously driven lines are omitted.

```text
nu0 = c / lambda0                 E_photon = h nu
A_total = sum(selected decay A)   gamma = 1 GHz / 2 + A_total / (4 pi)
phi(nu) = gamma / [pi ((nu − nu0)^2 + gamma^2)]
sigma_abs = (g_upper/g_lower) lambda0^2 A_upper_lower phi(nu) / (8 pi)
R_abs = irradiance sigma_abs / (h nu)
R_stim = R_abs g_lower/g_upper
p_upper_conditional = R_abs / (A_total + R_abs + R_stim)
```

The atomic and beam Lorentzian widths add under convolution. `g = 2J+1`; wavelengths and energy gaps use Ritz vacuum positions. The upward and stimulated rates are coefficients per atom in their starting manifold. The static `analyze()` result retains the conditional two-manifold projection above. For a closed pair, its high-intensity population tends to `g_upper/(g_lower+g_upper)`. The display instead uses the evolving populations below. The local line formula is evaluated only where relative line overlap is at least 10⁻³. Remote wings are not extrapolated; zero rate fields there mean no modelled event, not a measured zero absorption. [Cross-section method](https://superlidar.colorado.edu/Classes/Lidar2012/LidarLecture11_Classification.pdf), [lifetime and linewidth context](https://steck.us/alkalidata/sodiumnumbers.pdf).

`core/photon-population.js` evolves one ground manifold `G`, one driven upper manifold `U` and the selected lower excited states `S_j`. The populations start with `G=1`. Let `a=R_abs`, `b=R_stim`, `A=sum(A_j)` and `A_ground` be the sum of branches returning to ground.

```text
dG/dt = -a G + (b + A_ground) U
dU/dt =  a G - (b + A) U
dS_j/dt = A_j U                   for branches ending above ground
G + U + sum(S_j) = 1
current spontaneous rate = A U
```

Copper's selected UV-excited 4p states can emit to the ground manifold or to lower excited D states. Helium's selected singlet 2p state also has a branch to an excited 2s state. Retained lower populations reduce later fluorescence instead of quietly replenishing a fresh ground-state atom after each cycle. Those lower states are unpumped absorbing sinks in this restricted network. Their recovery, cascades and repumping are absent; the long-time result is therefore a limit of this network, not a full-atom steady state. Selected branch probabilities are `A_channel/A`; `1/A` is an upper bound on radiative lifetime if unlisted channels contribute. [Helium source rows](https://physics.nist.gov/PhysRefData/Handbook/Tables/heliumtable3.htm).

Each constant-source interval uses the analytic two-mode solution for `G` and `U`; exponential-integral forms also integrate the branch populations and expected event counts. Stable `expm1` expressions retain short-time startup and stiff high-rate behaviour without a fine-step loop. Source-off sets `a=b=0` and preserves the ensemble: `U(t+dt)=U(t) exp(-A dt)`. Its remaining population decays into the selected ground/lower branches. [Einstein rates and source-off decay](https://sheffield-mps.github.io/PHY242/topic3.html).

Intensity edits and detuning within the same selected pair retain state. Changing element, selecting another upper-state pair or choosing Reset atoms starts an identified new ensemble. Unsupported ionising configurations do not advance a predictive ionisation model. The atomic clock represents 10 ns per real second, separately from the material clock, and pauses while the document is hidden.

The display uses independent cumulative expected counts `integral(a G dt)`, `integral(b U dt)` and `integral(A U dt)` for absorption, stimulated emission and spontaneous emission, with a separate `integral(A_j U dt)` for each spontaneous branch. These are per reference atom and can be fractional. It never reconstructs history as the current rate multiplied by elapsed time. Each channel is compressed independently to at most four weighted glyphs per display second. Drawn count ratios are therefore not physical flux ratios. The clock summary keeps the weighted cue and expected spontaneous count visible; its notes list all three rates, counts and current glyph weights. Source-off retains only the spontaneous upper-state decay. Travel and residence drawings are expanded for reading. They do not give physical flight speed or a sampled electron residence-time distribution. Wave marks identify radiation; their spacing is not a wavelength scale. A ring is a state cue, not an electron orbit.

The readouts distinguish the **Beam centre** wavelength and energy from the **Spont. branch** wavelength and photon energy `E_upper−E_lower`. An open transition can emit a lower-energy spontaneous photon and leave the atom above ground. Its upper→lower energies and landing level remain visible in motion and reduced motion. The separate stimulated cue follows the beam direction, with a representative beam-centre wavelength and photon energy. Its energy-diagram arrow returns to the driven lower state, independently of the selected spontaneous branch. With the finite-width beam, the near-resonant spectral component drives the selected transition; the existing energy ledger retains its selected-gap approximation. Stimulated emission enters the exciting field's mode, but this drawing does not propagate phase, polarisation or coherent field dynamics. [Primary rate and mode description](https://interactivetextbooks.tudelft.nl/interactive-optics/content/Chap7_Lasers/Lasers_2022_01Clean.html).

The reference remains a stationary, incoherent rate model. A 1 GHz linewidth alone does not establish a universal intensity cutoff for coherent behaviour; none is invented here. The existing strong-drive omissions still apply. Representative paths are not next-photon predictions. Visible hues have fixed screen brightness; nonvisible paths are labelled false colour. Photons, Waves, combined and Hidden drawing settings apply in both atomic views. Hidden and reduced motion suppress moving light glyphs while retaining populations, rates and energy references. These drawing changes expose a process already present in the rate equations; they add no light–matter coupling to the mechanical sample.

The separate atomic energy ledger uses the selected line gaps, rather than interpreting every beam-centre photon as exactly resonant:

```text
E_stored = U E_upper + sum(S_j E_lower_j)
E_absorbed = E_upper integral(a G dt)
E_stimulated = E_upper integral(b U dt)
E_spontaneous = sum[(E_upper - E_lower_j) integral(A_j U dt)]
delta(E_stored) = E_net_pump - E_spontaneous
```

Energies are eV per reference atom and the ground energy is zero. Net pump work is integrated with a stable population-balance form; directly subtracting enormous gross absorption and stimulated totals can lose precision at extreme intensity. The balance residual is accumulated by interval and reported relative to transferred energy. This checks the selected rate network. It does not add radiation transport, recoil or light energy to the MD ledger. The mechanical sample's ground-configuration record, velocities and energy remain unchanged.

Above an available first-ionisation threshold, no neutral bound-state cycle is drawn. A further energy check flags `h nu >= IE - E_upper`: resonant photons for H, He, Ne and Ar can also ionise an already excited atom. No cross-section, ion yield, product or competing probability is inferred from that condition. Multiphoton processes and reactive dynamics are absent. At high irradiance the omitted processes may dominate. No quantum validity is inferred from the wall's 8000 K numerical source guard.

## Time, size and references

Playback uses a bounded real-time accumulator with fixed simulation intervals. Its nominal physical pace is independent of screen refresh rate. Delayed work is dropped rather than taken as a larger force step. The displayed physical clock and exported readings report only time actually integrated.

Physical display-radius references retain the Cordero convention through Z=96 and an explicit 140 pm fallback thereafter. For legibility, marker pixel radii use a common 2.1 nm camera calibration across dense and dilute preparations, separately from the physical position scale. This fixes disappearing gas atoms without changing positions, chamber dimensions or interaction radii. The sphere diameter must not be read from the positional ruler as a measured atom size. Manual zoom still scales markers and positions consistently within a scene. Proximity controls surface fill and cluster diagnostics, not the force law or a thermodynamic phase probability.

The molecule dialog reads one actual model molecule without changing it. It enlarges its current geometry and gives current and reference bond lengths in pm. The nodes mark nuclei schematically; they are not measured atomic surfaces. This view supplies a readable molecular scale while the crucible retains its real chamber distances.

The bulk-state overlay compares the same temperature reference used for light with tabulated melting/boiling points near 1 bar. For coherent monatomic clusters this removes centre-of-mass translation and best-fit rigid rotation; molecular rotations remain in the kinetic estimate. The badge states the temperature used and its detail identifies the basis. It reports crossed reference points while retaining the MD structural label independently. Exact points identify coexistence boundaries, without assigning phase fractions. No force, latent heat, nucleation rate or position is changed by this overlay. A finite nonequilibrium vacuum sample can change structure elsewhere. Material forms remain explicit; unknown points are null, carbon and grey arsenic receive no invented normal-pressure liquid range, and helium has no normal-pressure freezing boundary. [Phase source ledger](data/phase-sources.json).

Atomic energy coverage remains 773 selected states for 99 spectra plus 19 ground-only entries, with 108 first-ionisation limits. These whole-atom references are separate from solid-state bands and evolving driven populations. Room-state symbol colours and photographs retain their stated conditions.

Version 6 setup files store beam irradiance and source frequency and retain signed environmental ledgers. Imports accept versions 1–6, normalise removed physical controls and start a new paused preparation and ground-state atomic ensemble. They do not restart a saved trajectory or population history. Versions 1–5 retain their legacy 30 THz source. Room reset creates a new 293.15 K kinetic start and wall, turns the beam off and retains wavelength; it is not a thermostat applied to the existing hot trajectory. The theme remains a local display choice. [README.md](README.md) gives the current test commands; the release audit records completed validation separately from the physical assumptions above.

## Wall-source power provenance

The intensity control commands power before the numerical wall guard. The guard admits a duty-averaged part of that beam. The admitted beam, rather than the withheld part, is split into 80% wall absorption and 20% diffuse reflection. Heat notes and the spectrum identify the command, admitted power, absorbed power, reflected power and guard withholding. The spectrum also retains the original requested grey absorption. After an integration interval, admitted and absorbed values use that interval's mean; before one, they describe the currently available power. These labels do not change the heat-transfer kernel, photon animation weighting or energy ledger.

Interval accepted source power uses energy and duration summed over the current wall-update segments. Subtracting two large cumulative radiation totals could erase the increment from a weak beam after strong heating, falsely giving zero delivery and a guard warning. The interval sums avoid both cumulative-energy and global-time subtraction. They do not reset earlier totals or increase weak-beam heating. Cumulative ledgers remain finite-precision totals.

A command-minus-admission difference within the source contract's relative 10⁻¹² tolerance is reported as zero withholding. This suppresses floating-point residue labelled as a guard action; it never floors the admitted or absorbed weak-beam power.

## Gas diagrams and screen exposure

The public crucible uses equal-size position symbols across elements and phases. Their 132 pm camera reference and 2.1 nm span are drawing conventions. The covalent-radius data remains available separately. Neutral symbols have a 0.10 translucent visibility layer; this does not estimate optical depth or gas transmission. Local bulk-surface weights fade continuously without changing symbol radius. Neither neutrality nor lack of a drawn glow establishes zero absorption.

Fixed-bond components are drawn around their actual homonuclear centre of mass. Current relative 3D vectors use a constant per-molecule diagram scale: `2.2 * symbolRadiusPx / shortestReferenceBondM`. This preserves orientation and relative vibration, while the centre retains the physical camera projection. The convention does not switch with phase or density. Drawn bond spacing is schematic; the molecule dialog gives current physical model distances. Molecular trails follow the centre. No diagram transform writes particle coordinates or changes collisions, wall contact or force evaluation.

The optional gas-light reference maps `P_visible / N` through `strength = -expm1(-powerPerAtomW / 1e-12 W)`. Exposure is applied before saturation, giving the same per-symbol brightness at fixed per-atom power when atom count changes. Physical LTE line powers still scale with count. A numerical strength cutoff of 1e-5 is a display choice, not an excitation or ionisation threshold. The hue uses the existing selected neutral lines and CIE table. It remains conditional on neutral LTE, incomplete line coverage and unresolved optical depth; molecular bands, plasma populations and direct beam excitation of the classical particles are not solved.

[NIST neutral sodium levels](https://physics.nist.gov/PhysRefData/Handbook/Tables/sodiumtable5.htm) place the familiar emitting upper levels below the neutral ionisation limit. Ionisation is therefore not a prerequisite for emission. Neutral molecular absorption also produces familiar bulk gas colours, including [chlorine](https://periodic-table.rsc.org/element/17/chlorine) and [iodine vapour](https://periodic-table.rsc.org/element/53/iodine). Those absorption paths are not calculated by the visibility symbols.

## Population roundoff and reference labels

Repeated analytic atomic intervals can accumulate tiny errors in the sum of ground, upper and shelved probabilities. Completed intervals now assign the residual `1 - sum(other populations)` to the largest population. Smaller physical fractions, including an arbitrarily small surviving afterglow, are retained. Strict input validation is unchanged. Rate coefficients, integrated photon counts and the stable ODE-derived net-pump increment are unchanged; any stored-energy correction from the probability projection appears in the energy residual. Curated weak Cu and He references are tested over 150,000 intervals, equivalent to 83 minutes 20 seconds at the declared display clock and representative 30 Hz population updates.

Gas-reference hue is calculated from the raw visible line powers and CIE functions. The parent spectrum's exposure floor cannot change a per-atom cue's hue or strength. Missing or inconsistent line inputs produce no invented cue. This changes a display mapping, not the LTE populations or physical line powers.

For a single material atom, the COM-subtracted temperature is unavailable. The existing numeric optical reference can use the container-frame kinetic estimate; its basis and note now say so explicitly. Translational motion of one atom is not a thermodynamic temperature or a measured electronic population. The standard interface still prepares 128 atoms, and its temperature method is unchanged.
