Landé g-Factor

Every fine-structure level of every atom carries one dimensionless number that tells you how it responds to a magnetic field: its Landé g-factor. Hand a spectroscopist a photographed Zeeman pattern and the g-factors of the two levels are what she reads off — often she knows the quantum numbers before the element, because $g_J$ is a fingerprint of $(L, S, J)$ arithmetic and nothing else. The number was extracted from data in 1921, four years before anyone knew what it counted; today it is two lines of algebra. This article owns the anatomy of that number, not its derivation. The rigorous proof — that inside a fixed- $J$ multiplet every vector operator acts as its projection along $\hat{\mathbf{J}}$ — is the property of Projection Theorem for Vector Operators, consumed by citation, never re-proved. What lives here is everything the formula says: the vector-model picture that makes it guessable in ten seconds, the special-case catalogue from pure orbit ( $g_J=1$ ) to pure spin ( $g_J=2$ ) and beyond both, the sensitivity to the QED value $g_s=2.0023$ , and one genuine surprise in the algebra — an exact family of levels with $g_J=0$ , atoms a magnetic field cannot touch at first order. The weak-field splittings this factor feeds belong to Weak-Field (Anomalous) Zeeman Effect; the strong-field regime where $g_J$ dies belongs to Strong-Field Paschen-Back Effect. Between those two owners sits this article: the g-factor itself, read until it is transparent.

Key takeaways

$g_J$ is the effective gyromagnetic ratio of a fine-structure level $|L,S,J\rangle$ : its weak-field rungs are spaced $g_J\mu_B B$ (the fan owned by Weak-Field (Anomalous) Zeeman Effect).
Vector-model reading: only the $\hat{\mathbf{J}}$ -projection of $\hat{\mathbf{L}}+2\hat{\mathbf{S}}=\hat{\mathbf{J}}+\hat{\mathbf{S}}$ survives, so $g_J=1+\langle\hat{\mathbf{S}}\cdot\hat{\mathbf{J}}\rangle/\langle\hat{\mathbf{J}}^2\rangle$ — rigor supplied by Projection Theorem for Vector Operators, consumed not re-proved.
The closed form $g_J=1+\dfrac{J(J+1)+S(S+1)-L(L+1)}{2J(J+1)}$ gives the catalogue: $S=0\Rightarrow1$ , $L=0\Rightarrow2$ , $^2P_{1/2}=\tfrac23$ , $^2P_{3/2}=\tfrac43$ , $^3P_2=\tfrac32$ , $^1P_1=1$ — all machine-verified.
$g_J$ is NOT boxed between $\tfrac23$ and 2: $^4P_{1/2}=\tfrac83$ , $^6F_{1/2}=-\tfrac23$ , and the family $(L,S,J)=(2J{+}1,\,J{+}1,\,J)$ — $^4D_{1/2}$ , $^5F_1$ , $^6G_{3/2}$ , … — has $g_J=0$ exactly: magnetically silent levels.
The QED wage $g_s=2.00232$ shifts $g_J$ at the $10^{-3}$ level and un-silences the zeros to $g_J=2-g_s=-0.00232$ .
Key idea: one number, fixed entirely by $(L,S,J)$ counting, converts a level's invisible spin-orbit composition into its visible magnetic response — because spin's double weight makes the moment mis-track $\hat{\mathbf{J}}$ by a computable factor.

The core insight

In a weak field only the projection of the total magnetic moment onto $\hat{\mathbf{J}}$ survives the fast spin-orbit precession; the projection theorem converts this geometric fact into the closed-form $g_J = 1 + [J(J{+}1)+S(S{+}1)-L(L{+}1)]/[2J(J{+}1)]$ — the one dial setting a level's first-order magnetic personality.

Where this lives today

The g-factor is now a workhorse label and a precision frontier. Spectral-line catalogues list $g_J$ beside wavelength — solar magnetometry picks large- $g_J$ lines for field maps and $g_J\approx0$ lines as wavelength references; electron paramagnetic resonance fingerprints a radical's orbital environment by its deviation from the free-spin 2.0023. As a frontier: Penning-trap experiments on single hydrogen-like ions measure bound-electron g-factors to parts in $10^{11}$ , testing bound-state QED and supplying the best value of the electron's atomic mass — descendants of Einstein's twisting rod, asking the same question with thirteen more digits. And the silent-level logic of Derivation C — engineer the g's so field noise cancels — is standard practice in optical clocks and trapped-ion qubits, where "magnetically insensitive transition" is a specification.

The physical picture

Spin-orbit coupling locks $\hat{\mathbf{L}}$ and $\hat{\mathbf{S}}$ into rapid precession about their sum $\hat{\mathbf{J}}$ — the internal torque dwarfs anything a lab magnet exerts. The moment $\hat{\boldsymbol{\mu}}=-(\mu_B/\hbar)(\hat{\mathbf{L}}+2\hat{\mathbf{S}})$ rides along, but because spin is paid double ( $g_s=2$ against $g_\ell=1$ , the wage table of Magnetic Moment and the Gyromagnetic Ratio), $\hat{\boldsymbol{\mu}}$ does not point along $-\hat{\mathbf{J}}$ : it hangs off and sweeps a cone. A weak field is too slow to follow the wobble; it sees only the component of $\hat{\boldsymbol{\mu}}$ along the precession axis. The Landé factor is the length of that surviving average: above 1 when spin aligns with $\hat{\mathbf{J}}$ and its double wage adds, below 1 when spin hangs anti-aligned and the wage subtracts.

Figure

The tilted moment: spin's double weight pulls μ off the J axis, and precession averages it back

How long is the surviving projection? That depends only on the triangle $(L,S,J)$ , and the catalogue is worth staring at before any algebra: the values do not stay politely between the pure-orbit 1 and the pure-spin 2. High-multiplicity terms overshoot ( $^4P_{1/2}=\tfrac83$ ), anti-aligned terms dive below zero ( $^6F_{1/2}=-\tfrac23$ ), and one exact family sits at zero — levels a magnet cannot shift at first order.

Figure

The g-factor landscape: the catalogue is not confined between 1 and 2

The zero is the starred point on the landscape's floor. Put $^2P_{3/2}$ and $^4D_{1/2}$ in the same 1 T field: the first fans into four rungs spaced 77.2 μeV (the number Projection Theorem for Vector Operators computed); the second — two perfectly good $m_J$ states — does not move. Example 2 quantifies the contrast.

The moment, the bridge, the formula

A level's moment collects the orbital and spin contributions with their unequal g-factors — $g_\ell=1$ , $g_s=2$ , consumed from Magnetic Moment and the Gyromagnetic Ratio — and the identity $\hat{\mathbf{L}}+2\hat{\mathbf{S}}=\hat{\mathbf{J}}+\hat{\mathbf{S}}$ splits it into a part that tracks $\hat{\mathbf{J}}$ and a part that does not:

\hat{\boldsymbol{\mu}} \;=\; -\frac{\mu_B}{\hbar}\left(\hat{\mathbf{L}}+2\hat{\mathbf{S}}\right) \;=\; -\frac{\mu_B}{\hbar}\left(\hat{\mathbf{J}}+\hat{\mathbf{S}}\right) \qquad (1)

where: $\mu_B=e\hbar/2m_e=5.788\times10^{-5}$ eV/T is the Bohr magneton and the minus sign is the electron's charge. The $\hat{\mathbf{J}}$ piece is tame; every drop of g-factor anatomy lives in the $+\hat{\mathbf{S}}$ excess — spin counted once inside $\hat{\mathbf{J}}$ , then paid a second time. Inside a fixed fine-structure level, how much of that excess survives? The vector model answers "the projection of $\hat{\mathbf{S}}$ along $\hat{\mathbf{J}}$ ", and the projection theorem makes the answer exact. Defining $g_J$ as the moment's effective wage per unit of $\hat{\mathbf{J}}$ ,

g_J \;=\; 1+\frac{\langle\hat{\mathbf{S}}\cdot\hat{\mathbf{J}}\rangle}{\langle\hat{\mathbf{J}}^2\rangle} \qquad (2)

where: the 1 is the $\hat{\mathbf{J}}$ piece of (1) and the fraction is the surviving share of the spin excess. This bridge is the vector-model cartoon stated as an operator fact; its legality is the theorem of Projection Theorem for Vector Operators, consumed whole. Both expectation values are pure $(L,S,J)$ counting. With $\hat{\mathbf{S}}\cdot\hat{\mathbf{J}}=\tfrac12(\hat{\mathbf{J}}^2+\hat{\mathbf{S}}^2-\hat{\mathbf{L}}^2)$ — square $\hat{\mathbf{L}}=\hat{\mathbf{J}}-\hat{\mathbf{S}}$ , the same move as the dot identity of Spin-Orbit Coupling — equation (2) lands on the 1923 formula:

g_J \;=\; 1+\frac{J(J+1)+S(S+1)-L(L+1)}{2J(J+1)} \qquad (3)

where: everything is dimensionless counting — no radial integrals, no $Z$ , no field. The numerator is (twice) $\langle\hat{\mathbf{S}}\cdot\hat{\mathbf{J}}\rangle$ : positive when spin aligns with $\hat{\mathbf{J}}$ , negative when it anti-aligns — that sign is the whole story of $g_J\gtrless1$ . What the number then does: in a weak field the level's $2J+1$ sublevels shift by

E_{m_J} \;=\; g_J\,\mu_B B\,m_J \qquad (4)

where: this fan — equal rungs spaced $g_J\mu_B B$ — with its censuses and validity boundary is the property of Weak-Field (Anomalous) Zeeman Effect; this article supplies the $g_J$ that sets each spacing. When $\mu_B B$ rivals the fine-structure gap, $J$ dies as a label and $g_J$ with it — the crossover owned by Strong-Field Paschen-Back Effect. One refinement the catalogue will need — nothing in the projection argument cares that the wage is exactly 2. Letting it float,

g_J(g_s) \;=\; 1+(g_s-1)\,\frac{J(J+1)+S(S+1)-L(L+1)}{2J(J+1)} \qquad (5)

where: $g_s=2$ recovers (3); the QED value $g_s=2.00231930436$ shifts every $g_J$ at the $10^{-3}$ level. The base article's general form with both wages free is the same expression rearranged (verified symbolically).

Derivation AThe vector model made honest

We derive the closed formula (3) by making the precession-averaging cartoon into operator algebra, explicit about which step is consumed from Projection Theorem for Vector Operators and which steps are this article's own arithmetic. The strategy: split the moment into its $\hat{\mathbf{J}}$ -tracking and spin-excess parts, invoke the projection theorem once — only the $\hat{\mathbf{J}}$ -projection of the excess survives inside a multiplet — then evaluate the coefficient by pure Casimir counting. The delicate point: the theorem is a block statement, valid strictly within one $J$ .

Step 1. Split the operator. From (1), $\hat{\mathbf{L}}+2\hat{\mathbf{S}}=\hat{\mathbf{J}}+\hat{\mathbf{S}}$ : a part proportional to $\hat{\mathbf{J}}$ with coefficient exactly 1, plus the spin excess $\hat{\mathbf{S}}$ .

Step 2. License the average. Quantum "time-averaging" is illegitimate as stated — what is true is the projection theorem, consumed: within one fine-structure level, matrix elements of any vector operator $\hat{\mathbf{V}}$ equal those of $\hat{\mathbf{J}}$ times $\langle\hat{\mathbf{V}}\cdot\hat{\mathbf{J}}\rangle/\hbar^2J(J{+}1)$ . So the excess acts as that coefficient times $\hat{\mathbf{J}}$ — the cartoon's conclusion, now a theorem whose proof is not reproduced here. Why the cartoon works: a $J$ -multiplet has only one vector direction to offer, so every vector operator must line up with it on average — a representation-theory fact. Trust the fact, use the picture.

Step 3. Evaluate the coefficient. Square $\hat{\mathbf{L}}=\hat{\mathbf{J}}-\hat{\mathbf{S}}$ to get $\hat{\mathbf{S}}\cdot\hat{\mathbf{J}}=\tfrac12(\hat{\mathbf{J}}^2+\hat{\mathbf{S}}^2-\hat{\mathbf{L}}^2)$ , whence $\langle\hat{\mathbf{S}}\cdot\hat{\mathbf{J}}\rangle=\tfrac{\hbar^2}{2}[J(J{+}1)+S(S{+}1)-L(L{+}1)]$ .

Step 4. Assemble. $g_J=1+\langle\hat{\mathbf{S}}\cdot\hat{\mathbf{J}}\rangle/\hbar^2J(J{+}1)$ gives (3). For general wages, repeat with $\hat{\mathbf{V}}=g_\ell\hat{\mathbf{L}}+g_s\hat{\mathbf{S}}$ : the result is (5).

Step 5. Audit against the owner. The formula reproduces the case set audited in Projection Theorem for Vector Operators — pure spin 2, pure orbital 1, $^2P_{3/2}=\tfrac43$ , $^2P_{1/2}=\tfrac23$ , $^2S_{1/2}=2$ — exactly, quoted as its anchors; the catalogue extends the menu beyond them. Verified: equation (2) equals (3) symbolically; the general projection equals (5) and the base's rearranged form; the owner's five anchors reproduced exactly (article verify script, sympy).

The catalogue — what the formula actually says

Equation (3) is one line, but its value table is a landscape with cliffs — every entry an exact rational, machine-verified, each row a different lesson in how spin and orbit share a level.

term	$(L,S,J)$	$g_J$	reading
$^2S_{1/2}$	$(0,\tfrac12,\tfrac12)$	$2$	pure spin: the moment dictionary's $g_s$
$^1P_1$	$(1,0,1)$	$1$	pure orbit: singlets always give 1
$^2P_{3/2}$	$(1,\tfrac12,\tfrac32)$	$\tfrac43$	spin aligned: double wage adds a third
$^2P_{1/2}$	$(1,\tfrac12,\tfrac12)$	$\tfrac23$	spin anti-aligned: double wage subtracts a third
$^2D_{3/2}$ , $^2D_{5/2}$	$(2,\tfrac12,\cdot)$	$\tfrac45$ , $\tfrac65$	same pattern, diluted by a longer orbit
$^3S_1$	$(0,1,1)$	$2$	any $L=0$ term is pure spin
$^3P_1$ , $^3P_2$	$(1,1,\cdot)$	$\tfrac32$ , $\tfrac32$	different levels, same g — spacings match
$^4P_{1/2}$	$(1,\tfrac32,\tfrac12)$	$\tfrac83$	above 2: big spin against small $J$
$^4D_{1/2}$	$(2,\tfrac32,\tfrac12)$	$0$	magnetically silent (Derivation C)
$^6F_{1/2}$	$(3,\tfrac52,\tfrac12)$	$-\tfrac23$	negative: the fan inverts

Three structural readings. Within one doublet pair the $J=L+\tfrac12$ member always fans faster, and as $L$ grows both drift toward 1: a long orbital vector dilutes spin (symbolic limit in the verify script). The stretched states $J=L+S$ obey $g_J=(L+2S)/(L{+}S)$ — the wage-weighted average of 1 and 2, because fully aligned vectors have no geometry left to argue about (verified symbolically). And the folklore box " $\tfrac23\le g_J\le2$ " is a fact about one-electron doublets, not about the formula. Two cards complete the anatomy. The effective-moment card: what a susceptibility measurement calls the level's moment is

\mu_{\mathrm{eff}} \;=\; g_J\sqrt{J(J+1)}\;\mu_B \qquad (6)

where: $\sqrt{J(J+1)}$ is the length of the angular-momentum vector and $g_J$ its wage: $^3P_2$ carries $\tfrac32\sqrt6=3.67\,\mu_B$ , $^2S_{1/2}$ carries $\sqrt3=1.73\,\mu_B$ , and $^4D_{1/2}$ carries exactly nothing. This $\mu_{\mathrm{eff}}$ is what enters the Curie law of paramagnetism — the electromagnetism track owns that story; here it is one line. The wage-refinement card: QED shifts the spin wage to $g_s=2.00231930436$ , and by (5) every $g_J$ inherits $(g_s-2)\times$ its fraction. For $^2P_{3/2}$ the fraction is exactly $\tfrac13$ , so $\Delta g_J=+7.73\times10^{-4}$ : $g_J=1.3341$ instead of $1.3333$ — a part in $10^3$ , resolvable by precision measurement, invisible in a teaching lab. Textbooks use 2 and say so.

Go deeper · Derivation B · 1 minThe special-case catalogue, proved wholesale

Derivation BThe special-case catalogue, proved wholesale

We verify the catalogue as a single sweep rather than case-by-case arithmetic, because the formula's structure makes wholesale proof cheap: (3) is a ratio of counting polynomials, so entire families evaluate symbolically. The goal is threefold — pin the two textbook limits as identities rather than examples, land every rational in the table above, and extract the family rules (stretched states; the doublet drift toward 1) that organize the menu. The projection input was finished in Derivation A.

Step 1. The limits as identities. Set $S=0$ (forcing $J=L$ ): the numerator becomes $J(J{+}1)-L(L{+}1)=0$ , so $g_J=1$ identically — every singlet. Set $L=0$ (forcing $J=S$ ): numerator $=2J(J{+}1)$ , so $g_J=2$ identically — every $L=0$ term.

Step 2. The doublet ladder $S=\tfrac12$ . For $J=L\pm\tfrac12$ the formula gives $g_J=1\pm\tfrac{1}{2L+1}$ : $\tfrac43/\tfrac23$ at $L=1$ , $\tfrac65/\tfrac45$ at $L=2$ — both members marching toward 1 as $L\to\infty$ (limit verified symbolically). One-electron atoms live entirely on this ladder — the source of the folklore box.

Step 3. The multiplet menu. Direct evaluation: $^3P_1=\tfrac32$ , $^3P_2=\tfrac32$ (equal by counting coincidence; $^3P_0$ has $J=0$ — a single state, no fan, no defined $g_J$ ), $^1P_1=1$ , $^4P_{1/2}=\tfrac83$ , $^4D_{1/2}=0$ , $^6F_{1/2}=-\tfrac23$ . Why g can leave the box: the fraction in (2) is the spin excess along J divided by J's own length. Pile up spin against small J and the excess can exceed the baseline (g above 2) or, anti-aligned, cancel it and keep going (g zero, then negative).

Step 4. The stretched-state rule. Substitute $J=L+S$ : the fraction simplifies to $S/(L{+}S)$ , so $g_J=(L+2S)/(L{+}S)$ — the wage-weighted average, exact for every stretched term (symbolic).

Step 5. Cross-checks against owners. The five anchors of Projection Theorem for Vector Operators sit inside the sweep untouched; the trio consumed by Weak-Field (Anomalous) Zeeman Effect for its censuses is the $L\le1$ corner of Step 2. Consistency both ways, no number recomputed as new physics. Verified: both limits symbolic; the doublet ladder and its L→∞ limit symbolic; all eleven catalogue rationals exact; the stretched rule symbolic; anchors matched (article verify script).

The silent levels

The catalogue's strangest entry is the exact zero, and it is not an accident of one term: the condition $g_J=0$ rearranges to a Diophantine statement about angular-momentum counting, and its solutions form a perfect one-parameter family.

Go deeper · Derivation C · 2 minThe hunt for magnetically silent levels

Derivation CThe hunt for magnetically silent levels

We solve $g_J=0$ exactly. The plan of the hunt: rearrange the vanishing condition into a constraint among the three Casimirs, scan the grid of allowed $(L,S,J)$ triples to see what nature's bookkeeping permits, then recognize the scan's output as a single algebraic family — proving the pattern rather than merely observing it. The subtlety: candidate triples must satisfy the triangle rule $|L-S|\le J\le L+S$ , or the "solution" describes no physical level.

Step 1. The condition. $g_J=0$ in (3) means $3J(J{+}1)=L(L{+}1)-S(S{+}1)$ . Immediately: silent levels need $L(L{+}1)>S(S{+}1)$ — orbit dominates the composition — yet the spin wage must fully buy the orbital moment back, forcing anti-alignment.

Step 2. The scan. Sweep all $L\le12$ , $S\le6$ , all triangle-legal $J>0$ : the solutions are $^4D_{1/2}$ , $^5F_1$ , $^6G_{3/2}$ , $^7H_2$ , $^8I_{5/2}$ , … — one per multiplicity, each with $J=L-S$ exactly: the anti-stretched member.

Step 3. The family, proved. Every hit satisfies $S=J+1$ , $L=2J+1$ . Substitute: $L(L{+}1)-S(S{+}1)=(2J{+}1)(2J{+}2)-(J{+}1)(J{+}2)=(J{+}1)(3J)=3J(J{+}1)$ — an identity for all $J$ . The family $(L,S,J)=(2J{+}1,\,J{+}1,\,J)$ is silent for every $J=\tfrac12,1,\tfrac32,\dots$ , and the scan finds nothing else in range. Why the physics allows it: in the anti-stretched state the spin projection along J is negative, and spin is paid double. Silence is the exact break-even where the doubled anti-aligned wage cancels the orbital wage — one more notch of spin (⁶F½) and the moment flips sign outright.

Step 4. What silence means — and its fine print. A $^4D_{1/2}$ level has two $m_J$ states and a field shifts neither: no fan, $\mu_{\mathrm{eff}}=0$ in (6). The fine print is the wage refinement (5): with the QED $g_s$ , the zero moves to $g_J=2-g_s=-2.32\times10^{-3}$ — silence at the $10^{-3}$ level, not absolute (at 1 T: 0.134 μeV residual, 575 times below a $^2P_{3/2}$ rung); second-order Zeeman survives too. First-order immunity is the claim — exactly the property clock builders hunt for in field-insensitive transitions (one line; the metrology story is not this article's). Verified: the scan yields exactly the family hits; the family identity symbolic for all J; every hit anti-stretched; the QED residual and 575:1 comparison computed (article verify script).

Limiting cases

case	result	meaning
$S=0$	$g_J=1$ identically	singlets are pure orbit — the normal Zeeman world
$L=0$	$g_J=2$ identically	any S-term is pure spin, whatever the multiplicity
$J=L+S$ (stretched)	$g_J=(L{+}2S)/(L{+}S)$	aligned vectors: the wage-weighted average
$S=J{+}1$ , $L=2J{+}1$	$g_J=0$	the silent family — anti-aligned spin buys back the orbit
$L\to\infty$ , fixed $S$	$g_J\to1$	a long orbital vector dilutes spin; classical limit
$J=0$	single state, no fan	$g_J$ undefined and unneeded: nothing splits
$\mu_B B\gtrsim\Delta E_{fs}$	$g_J$ loses meaning	$J$ dies as a label — owned by Strong-Field Paschen-Back Effect

Common trapEvery g-factor lies between two-thirds and two

False — that comfortable box belongs to the one-electron doublet ladder ( $g=1\pm\tfrac{1}{2L+1}$ ), not to the Landé formula. Higher multiplicities break out in every direction: $^4P_{1/2}$ pays $\tfrac83>2$ , the family $^4D_{1/2},{}^5F_1,{}^6G_{3/2},\dots$ pays exactly zero, and $^6F_{1/2}$ pays $-\tfrac23$ — its fan runs upside-down. Memorize the mechanism, not the box: aligned spin adds its double wage, anti-aligned spin subtracts it, and enough spin against little enough $J$ overshoots through zero. A second trap nearby: $^3P_1$ and $^3P_2$ share $g_J=\tfrac32$ — equal g-factors do NOT mean equal levels; the fans differ in rung count even when spacings match.

Worked examples

Example 1 — the working table: doublets plus a triplet multiplet. Build the g-table an alkali or helium-like spectrum needs, then attach numbers at $B=1$ T.

Step 1. Doublets from the ladder $g=1\pm\tfrac{1}{2L+1}$ : $^2S_{1/2}=2$ , $^2P_{1/2}=\tfrac23$ , $^2P_{3/2}=\tfrac43$ , $^2D_{3/2}=\tfrac45$ , $^2D_{5/2}=\tfrac65$ (exact, cross-checked against the anchors of Projection Theorem for Vector Operators).

Step 2. The $^3P$ multiplet from (3): $^3P_2=\tfrac32$ , $^3P_1=\tfrac32$ ; $^3P_0$ has $J=0$ — one state, no splitting, no $g_J$ : the formula's $0/0$ says the question is empty.

Step 3. Rung spacings $g_J\mu_B B$ at 1 T with $\mu_B=57.88$ μeV/T: $^2S_{1/2}$ : 115.8; $^2P_{1/2}$ : 38.6; $^2P_{3/2}$ : 77.2 (the sodium anchors of Projection Theorem for Vector Operators, quoted as its numbers — arithmetic re-checked); $^2D_{3/2}$ : 46.3; $^2D_{5/2}$ : 69.5; $^3P_{1,2}$ : 86.8 μeV each.

Step 4. Sanity reads. The $^2P_{3/2}$ quartet spans $3\times77.2=231.5$ μeV; $^3P_1$ (triplet) and $^3P_2$ (quintet) match in spacing but not rung count. Answer: the table above — the complete first-order magnetic personality of these terms.

Example 2 — the silent level, quantitatively. A $^4D_{1/2}$ and a $^2P_{3/2}$ level sit in the same $B=1$ T field. Compare their responses, then apply the QED wage.

Step 1. $^2P_{3/2}$ : $g_J=\tfrac43$ , four rungs at $\pm38.6,\pm115.8$ μeV — the quartet that Weak-Field (Anomalous) Zeeman Effect fans into line patterns.

Step 2. $^4D_{1/2}$ : $g_J=1+\dfrac{\tfrac34+\tfrac{15}4-6}{2\cdot\tfrac34}=1+\dfrac{-\tfrac32}{\tfrac32}=0$ exactly. Both $m_J=\pm\tfrac12$ states shift by zero: the level ignores the magnet, and $\mu_{\mathrm{eff}}=0$ in (6).

Step 3. QED fine print via (5). The $^4D_{1/2}$ fraction is exactly $-1$ , so $g_J=2-g_s=-0.00232$ : residual splitting at 1 T is $0.134$ μeV — a factor 575 below the $^2P_{3/2}$ rung, and inverted.

Step 4. Why anyone cares: a transition between near-silent states barely feels field noise — the design logic of field-insensitive clock transitions, executed in real clocks at the hyperfine level (Hyperfine Structure's $g_F$ ). Answer: 77.2 μeV vs exactly 0 at $g_s=2$ ; with the QED wage, 0.134 μeV and sign-flipped — silence, with a whisper of QED underneath.

Origins · 1 minGyromagnetic Ratio and how this was discovered

Gyromagnetic Ratio

1915–1925 · Leiden, Zürich, Ann Arbor

Before quantum mechanics could compute a g-factor, rotating rods measured one. The classical prediction was crisp: orbiting charge gives $\gamma=q/2m$ , so $g=1$ . In 1915 Einstein and de Haas — Einstein's only true laboratory experiment, in Leiden — magnetized an iron rod and watched it twist, reporting $g\approx1.02$ : the expected answer, found partly because it was expected. Samuel Barnett in Ann Arbor ran the inverse experiment (magnetization by rotation) the same year and kept finding values near 2; a decade of magnetomechanical measurements converged on $g\approx2$ — an anomaly in plain sight with no theory to claim it. Meanwhile in Tübingen, Alfred Landé attacked the anomalous Zeeman patterns from the data side: in 1921 he assigned each level an empirical factor $g$ — half-integer quantum numbers and all — and by 1923 had the exact formula this article anatomizes, extracted from photographs four years before spin existed. The 1925-26 resolution (Goudsmit and Uhlenbeck's spinning electron with $g_s=2$ , then the projection arithmetic) turned both anomalies — twisting rods and fitted fractions — into one statement: spin is paid double.

History

1915

— Einstein and de Haas report g near 1 in iron; Barnett's rotation experiments the same year point toward 2

1919

— Repeat magnetomechanical measurements settle near g = 2: an anomaly without theory

1921

— Landé assigns empirical g-factors to levels, decoding anomalous Zeeman patterns with half-integer numbers

1923

— Landé publishes the exact g-formula, pure phenomenology fitted from spectra

1925

— Goudsmit and Uhlenbeck propose the spinning electron with g_s = 2; the double wage explains both anomalies

1926

— Heisenberg and Jordan derive the Landé formula from quantum mechanics

1947

— Kusch and Foley measure g_s = 2.00232: the g-factor becomes a precision probe of QED

Going further

Griffiths & Schroeter, Introduction to Quantum Mechanics, 3rd ed., §7.4 — the weak-field Zeeman context this factor feeds.
Sakurai & Napolitano, Modern Quantum Mechanics, §3.11 — the projection theorem route at full rigor.
Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. II, Complement D_X — the vector model with every step honest.
Condon & Shortley, The Theory of Atomic Spectra, Ch. 16 — the classic g-factor tables, exotic terms included.

Provenance note: A verifier should recompute the boxed Landé formula from the projection theorem and the identity $\mathbf{L}\cdot\mathbf{S}=\tfrac12(J^2-L^2-S^2)$ (Sakurai §3.11; Griffiths §7.4). The sodium g-factors $2,\,2/3,\,4/3$ and the $^3P_2$ value $3/2$ follow by direct substitution and match Cohen-Tannoudji Complement D_X. Numerical constants: $\mu_B=9.274\times10^{-24}\,\mathrm{J/T}$ and $g_S=2.00231930436$ are CODATA standard values. Limit checks ( $S=0\Rightarrow g_J=1$ , $L=0\Rightarrow g_J=2$ ) are reproduced in the derivation. — The projection-theorem derivation is executed where it is owned, in Projection Theorem for Vector Operators; this article consumes it by citation and quotes its audited anchors. Edition note (base error, fixed): the base's derivation displayed at its Step 5 the expression $g_J=[2J(J{+}1)+S(S{+}1)-L(L{+}1)]/[2J(J{+}1)]$ — a $J(J{+}1)$ dropped from the numerator, off by exactly $\tfrac12$ (it yields $\tfrac56$ instead of $\tfrac43$ for $^2P_{3/2}$ ), even though the base's prose combination and final Step 6 are correct. Discrepancy proven symbolically and numerically in the article's verify script; the rebuilt derivation carries the machine-verified line throughout.

Is the vector-model precession picture actually legitimate, or just a mnemonic?

It is a mnemonic for a theorem. "Fast precession averages out the perpendicular part" is not a quantum argument — nothing is precessing in an energy eigenstate. The honest statement is the projection theorem of Projection Theorem for Vector Operators: inside one $J$ -multiplet, every vector operator's matrix elements are proportional to $\hat{\mathbf{J}}$ 's; the cartoon happens to compute the same coefficient. Use the picture to remember, cite the theorem when asked why.

Why does spin count double in the first place?

That is the gyromagnetic anomaly, an input here, not an output: $g_s=2$ comes from the Dirac equation (plus the 0.2% QED correction), and the dictionary of moments is owned by Magnetic Moment and the Gyromagnetic Ratio. This article's job starts after the wages are set: given unequal g's, geometry decides how a coupled level pays out. If $g_s$ were 1, equation (3) would collapse to $g_J=1$ for everything and the anomalous Zeeman effect would not exist.

What happens to the Landé formula when LS coupling itself fails?

It keeps the form and loses the inputs. The projection logic needs only a good $J$ , which survives any coupling scheme. What fails in heavy atoms is the assumption of sharp $L$ and $S$ : in jj or intermediate coupling, $g_J$ becomes a weighted mix of the Landé values of the LS terms hiding in the state. Measured g's are therefore a diagnostic: match the formula and the term assignment stands; deviate systematically and you are watching coupling-scheme breakdown — the multi-electron story belongs to atomic-structure chapters.

Where does the magnetic moment go in a g equals zero level?

Nowhere — the operators are all still there; their projections cancel. In $^4D_{1/2}$ the orbital moment along $\hat{\mathbf{J}}$ is exactly $+1$ wage-unit and the anti-aligned spin moment exactly $-1$ . The level keeps its orbital and spin structure — anything coupling to $\hat{\mathbf{L}}$ or $\hat{\mathbf{S}}$ separately (spin-orbit energy, hyperfine coupling, selection rules) sees it normally. Only the combination $\hat{\mathbf{L}}+2\hat{\mathbf{S}}$ projected on $\hat{\mathbf{J}}$ — what a uniform field pays — sums to zero.

Is the g equals zero silence exact, or does something eventually break it?

Three things break it, in order of size. The QED wage: $g_s=2.0023$ moves the zero to $g_J=2-g_s\approx-0.0023$ — a 0.1%-of-normal residual, computed in Derivation C. Second-order Zeeman: the field mixes neighboring $J$ levels and shifts energies quadratically in $B$ . And in multi-electron atoms, coupling-scheme corrections shift g's at the percent level anyway. So "silent" means: the big, linear, first-order response — the one that dominates every ordinary Zeeman pattern — has coefficient zero.

Do g_J and the hyperfine g_F obey the same logic?

Identically — it is projection all the way down. Couple $\hat{\mathbf{J}}$ to the nuclear spin $\hat{\mathbf{I}}$ to form $\hat{\mathbf{F}}$ : in a weak enough field the moment projects onto $\hat{\mathbf{F}}$ , giving $g_F$ built from $g_J$ (plus a tiny nuclear term) by the same formula with relabeled letters. The construction and its own strong-field breakdown live in Hyperfine Structure. Every rung of the ladder — $g_s\to g_J\to g_F$ — is one more projection onto one more layer of coupling.