Multi-level atoms, hyperfine splitting, and the Hg-196 doping trick

The Holstein equation in Layer 1 assumed a two-level atom with a single resonance line. Real atoms are multi-level and most have hyperfine structure. Most experiments also have isotope structure that splits the resonance line into a comb of components.

What changes? The lineshape becomes a sum: $k(x) = \sum_i w_i\, k_{\text{single}}(x - \delta_i)$, with $w_i$ the relative amplitudes (M&O Eq 7.20). At high opacity the central components are optically thick — photons can't escape there — but the wings of the comb can be much more transparent than the wings of a single line, because the comb's outermost components carry only a fraction of the total absorption strength.

This insight is what the GTE Lighting research group exploited in the 1980s (Anderson et al. 1985; Grossman et al. 1986). They argued that adding 2.6% of the rare isotope ¹⁹⁶Hg (natural abundance 0.15%) to standard mercury fluorescent-lamp fillings provides an optically-thin "escape channel" at the blue wing of the 254 nm line — increasing lamp efficiency by over 1%. Across world fluorescent-lamp deployment this works out to $\sim 10^9$ kWh/year savings. A pure trapping-physics intervention with massive economic consequences.

Explore

Below: the natural Hg 254 nm absorption comb on the left, the composite trapping factor as a function of ¹⁹⁶Hg fraction on the right. Watch the non-monotonic optimum near 2.6%. Slider the opacity to see how the optimum shifts.

¹⁹⁶Hg fraction 2.6%

Total opacity $k_0 L$ 100

Cold-spot $T$ (°C) 40

Hg 254 nm absorption comb — composite $k(x)$

Bars: individual isotope contributions. Curve: composite. Blue: ¹⁹⁶Hg dopant channel (escape route).

Trapping factor vs ¹⁹⁶Hg fraction

Non-monotonic. Optimum visible. Dot: current ¹⁹⁶Hg setting.

Relative UV output (vs natural Hg)

$1/g$ normalized to natural-abundance reference. >1.01 = the GTE result.

FEYNMAN's worked example. At $k_0L = 100$ (typical fluorescent-lamp cold-spot opacity for the 254 nm line), the natural-abundance trapping factor is $g_0 \approx 90$. Increasing the ¹⁹⁶Hg fraction from 0.15% (natural) to 2.6% (Anderson 1985) reduces $g_0$ to $\approx 81$, a 10% improvement → about 1.5% more uv output. At very low opacity ($k_0L < 5$) the doping does nothing — every channel is already transparent. At very high opacity ($k_0L > 10^4$) the ¹⁹⁶Hg channel itself becomes optically thick and the benefit disappears. There is a window — and the lamp sits inside it.

Why this happens

Bezuglov's separated-component approximation (M&O Eq 7.23) gives the composite trapping factor as a weighted average:

\[ \frac{1}{g_{\text{composite}}} = \sum_i w_i \cdot \frac{1}{g_{\text{single}}(w_i \cdot k_0 L)} \]

Each component sees an effective opacity $w_i \cdot k_0L$ — its own share of the total. For natural Hg, $w_{196} = 0.0015$ → effective opacity $0.0015 \cdot k_0L$. At $k_0L = 100$ this is 0.15 — well below the "trapping starts" threshold of $\sim 3$, so the ¹⁹⁶Hg channel is essentially transparent. But the contribution to $1/g$ is also small — only 0.0015 of the total — so the benefit is hidden.

Increasing $w_{196}$ to 0.026 increases its contribution to $1/g$ by 17×, while its effective opacity goes only to $\sim 2.6$ — still in the transparent regime. That's the trick. Beyond about 5% the ¹⁹⁶Hg effective opacity crosses $\sim 5$ and the channel starts trapping its own photons; the benefit reverses.

Generalisation — does this work for other elements?

Sodium (Layer 8): only one stable isotope (²³Na). The trick can't work. Use isotope mixing → instead, use buffer-gas pressure-broadening (which we did in Layer 3) or hfs ground-state mixing.
Caesium: mono-isotopic (¹³³Cs). Sommerer 1993 demonstrated the same effect with Zeeman splitting — a magnetic field plays the role of isotopic separation.
Potassium: ³⁹K dominant (93.3%), ⁴⁰K rare (0.012%), ⁴¹K (6.7%). The ⁴⁰K abundance is even lower than ¹⁹⁶Hg's, so the equivalent doping experiment could in principle work for K-discharge lamps. Not done in the open literature. An untapped engineering opportunity for low-pressure sodium-lamp efficiency.
Rubidium: ⁸⁵Rb / ⁸⁷Rb (72%/28%), no rare isotope. Doping useless. But the natural 72/28 mix already provides hfs-splitting that helps; see DPAL literature.

What this layer does NOT do

Doesn't solve the full multi-state collisional-radiative system — that's Layer 6 (NLTE).
Uses Bezuglov's separated-component approximation (M&O Eq 7.23), valid when components are well-separated relative to Doppler width. For overlapping components, the full Eq 7.21/7.22 numerical solution is needed. Layer 6 will improve this.
Doesn't include collisional energy transfer between fine-structure components ($p_{1/2}$ ↔ $p_{3/2}$ intermixing). At lamp electron temperatures this matters; M&O Sec 7.3 has the corrections.
Doesn't address the Sommerer 1993 magnetic-field analogue for mono-isotopic Cs/Na.

FEYNMAN ✓ worked example · Victor ✓ live optimum visible · Kay ✓ reuses Layer 1 trapping factors · Scott ✓ reproduces GTE result qualitatively · Molisch ✓ Eqs 7.20, 7.21, 7.23 cited