Bosons That Talk to Each Other, and Two Masses to a Per Mille

Two masses measured to a fraction of a percent constrain a third particle nobody had found, which is why so much effort went into numbers that look like bookkeeping.

🎯 Why this matters

Precision here substitutes for energy. Loop corrections let a machine that cannot produce a particle measure its mass anyway, which is the only way this field ever knows what to expect before building the next collider.

§9.9 measured the $Z$ to twenty parts per million and found nothing wrong. One structural feature of §9.3 was still untested, though — the self-coupling of the vector bosons — that the $W$ carries weak charge, and therefore couples to the $Z$ and the photon directly. §9.10 is the test, and it comes with a beautiful argument: the theory is not merely consistent with the self-coupling, it is inconsistent without it.

§9.10 A divergence, fixed twice

The story starts with a problem §7.1 already flagged. Fermi’s point-like interaction gives $\nu_\mu e \to \mu\nu_e$ a cross-section growing without limit, which violates unitarity at high enough energy. Introducing the $W$ as a propagating mediator cuts it off.

But a mediator you can exchange is a particle you can produce, and the moment $W$ s exist so does $e^+e^-\to W^+W^-$ — which has a divergence of its own.

🪜 Fig. 9.33 — three curves, and only the third is finite

Step 1 of 3 — neutrino exchange alone

e^+e^- \xrightarrow{\;\nu\;} W^+W^-

Why you may do this: The obvious diagram, Fig. 9.32(a): the electron and positron annihilate by exchanging a neutrino in the t channel. Computing it alone gives a cross-section that grows without limit — the UPPER curve in Fig. 9.33.

A cross-section rising indefinitely with energy eventually exceeds the unitarity bound, which is not a numerical problem but a logical one: probabilities cannot exceed 1.

Bettini pp. 387–388. Each fix is forced by the previous one, which is what makes this an argument rather than a fit.

Fig. 9.32 — the three tree-level diagrams for e⁺e⁻ → W⁺W⁻

Click a vertex or an internal line.

Bettini Fig. 9.32. Click the third vertex: it is the one the whole section exists to test, and the only one that requires the mediators to carry the charge they mediate.

ν exchange only — diverges
ν + γ — still diverges
ν + γ + Z — finite, and what is measured
LEP2

Fig. 9.33 redrawn. The two rising lines are the incomplete calculations; the data follow the third. The threshold position gives M_W and the sharpness of the rise gives Γ_W — a wider W would soften the turn-on.

Erratum — Fig. 9.33’s curves are named twice with the same words

The text describes the neutrino-only calculation as giving “a diverging cross-section, the intermediate curve in Fig. 9.33”, and then, two sentences later, says that adding the photon “again gives a diverging cross-section, the intermediate curve in Fig. 9.33”.

They cannot both be the intermediate curve. The figure carries three, and the entire argument is that each fix helps and only the third works — so the first reference must be to the upper curve. A reader following the argument carefully is stopped by exactly the sentence that is meant to carry it.

Confirmed on the render of PDF p. 405.

💡 What this really says — the cancellation is the theory, not a coincidence

It is worth being precise about what the measured cross-section tests, because “the agreement confirms the self-coupling” undersells it.

Each of the three amplitudes, taken alone, grows with energy in a way that would violate unitarity. Added together with the couplings the gauge symmetry prescribes, the bad terms cancel exactly, order by order in the energy. That cancellation is not arranged; it is a consequence of the SU(2)⊗U(1) structure, and it breaks if you scale any one coupling even slightly.

So the experiment is not measuring a number and comparing it with a prediction. It is checking a cancellation between large terms — and those are the most sensitive tests there are, because a small fractional error in one term shows up as a large fractional error in the small residue.

An engineer knows the pattern from every bridge circuit and every differential measurement: null out two large quantities against each other, and the residue is exquisitely sensitive to their ratio. Here the “bridge” is enforced by a symmetry rather than by a trimpot, and the measurement asks whether nature’s trimpot sits exactly where the symmetry says.

This is also why the same reasoning appears in the Higgs argument of §9.12. Once the $W$ s have mass, longitudinal polarization states exist and $W_LW_L$ scattering diverges again — and the diagram that fixes it is a Higgs exchange. The pattern of §9.10 repeated one level up is, historically, one of the strongest reasons to have expected a Higgs below about a TeV.

The same cross-section gives two more numbers for free. Its threshold position is $2M_W$ , so the energy at which it turns on measures the $W$ mass; and the sharpness of the turn-on measures $\Gamma_W$ , since a wider $W$ softens the onset. Both were superseded by hadron-collider measurements, which is §9.11.

§9.11 Two masses, and why they are worth the effort

§9.8 established the stakes: $M_W$ carries a correction quadratic in $m_t$ and logarithmic in $M_H$ . Measure $M_W$ and $m_t$ precisely enough and the residue constrains the Higgs — which is exactly what was done, and it is why Fig. 9.44’s $M_W$ – $m_t$ plane became the most-watched plot in the field for fifteen years.

Bettini pp. 388–389. The energy barely moved; the luminosity moved by a factor of 63, and that is where the physics came from.

period↕	$\sqrt s$ ↕	peak $L$ (cm⁻²s⁻¹)↕	what changed↕
Run Ia	1.8 TeV	$5.4\times10^{30}$	the top is discovered by CDF in 1995
Run Ib	1.8 TeV	$1.6\times10^{31}$
1996–2001	—	—	major upgrade: a new main injector ring, built to raise the antiproton intensity
Run II	1.96 TeV	$\mathbf{3.4\times10^{32}}$	a 9 % gain in energy against a 63× gain in luminosity. CDF and D0 each collect ~10 fb⁻¹ before the 2011 shutdown

⚙️ Engineer’s bridge — you cannot raise the energy of a built machine, but you can always raise the luminosity

The table above contains a piece of accelerator strategy worth extracting, because it explains the shape of a whole era.

The Tevatron’s energy went from 1.8 to 1.96 TeV — 9 % — across a multi-year upgrade. Its luminosity went up by a factor of 63. That is not an accident of what was easy; it is what is possible. The energy of a synchrotron is set by the ring radius and the dipole field, both of which are civil engineering and superconductor metallurgy respectively. The luminosity is set by beam intensity, bunch count and focusing at the interaction point, all of which respond to better sources, better cooling and better optics.

So the natural upgrade path for an existing collider is always the same: leave the energy alone and buy statistics. The book says the top search needed “a collider of higher energy and higher luminosity than the Tevatron in its initial configuration”, and then notes that “while the energy of an existing machine cannot be increased much, the luminosity can.”

An engineer will recognise the trade. Bandwidth is fixed by the physical channel; SNR you can buy with integration time. When you cannot widen the channel, average for longer — and design the measurement so that what you need is statistics rather than reach.

The same logic sets the LHC’s future: the High-Luminosity upgrade (§9.20) leaves $\sqrt s$ at 14 TeV and raises the integrated luminosity by an order of magnitude.

Where it breaks: luminosity substitutes for energy only for processes that are already kinematically open. Below threshold no amount of running helps at all — the Spp̄S could not have found the top with any integrated luminosity, because 630 GeV cannot make a 173 GeV pair against the parton spectrum, and that is why the Tevatron had to be built rather than merely run longer. The trade also degrades what it buys: raising instantaneous luminosity raises pile-up, and past some point each extra collision costs you resolution on the one you wanted. You are not turning a knob that only goes one way — you are moving along a curve with a maximum in it.

The $W$ mass is measured by the same Jacobian peak as §9.7 — with one addition that makes all the difference:

🔢 Worked example — calibrating a hadron collider with someone else’s measurement

§9.7 established that the Jacobian method is limited by the calorimeter energy scale, not by statistics. The Tevatron’s solution is to import a calibration:

For leptons. Whenever a $Z$ decays to a lepton pair in the same data, its mass is reconstructed from the measured energies. The lepton energy scale is then scaled until the reconstructed peak sits at the $M_Z$ measured at LEP, which is known to 23 ppm (§9.9). A hadron collider borrows an $e^+e^-$ collider’s precision.

For jets. Harder, because jets have larger uncertainties. The trick is elegant: in the lepton+jets top topology, two of the jets are the decay products of a $W$ , whose mass is precisely known. So the jet energy scale is set by requiring the reconstructed $M_{jj}$ to equal $M_W$ — an in-situ calibration, taken from the same events being analysed.

The result, combining CDF, D0 and later the LHC experiments:

$M_W = 80.377 \pm 0.012\;\text{GeV}\quad(0.015\,\%), \qquad \Gamma_W = 2.085 \pm 0.042\;\text{GeV}$

Compare UA1’s $83\pm3$ : a factor of 250. And compare the tree-level prediction of §9.5, $\Gamma_W = 2.04$ GeV — 2 % low, which is the radiative correction again.

The top is rarer than anything else in this book:

how rare the top is, and how the fit closes

sig_tt, sig_tot = 7.50e-12, 81.9e-3          # barn
print("top production at the Tevatron, sqrt(s) = 1.96 TeV")
print(f"  sigma(t tbar)      = 7.50 +- 0.48 pb")
print(f"  sigma(pp-bar, tot) = 81.9 +- 2.3 mb")
print(f"  ratio              = {sig_tt/sig_tot:.1e}   -> one top pair in 10^10 interactions")
print("\n  (single-top production, by the weak interaction, is about half of this)")

L, H = 3/9, 6/9
print("\nthe three tt-bar topologies, from BR(W -> l nu) = 1/9 per family")
print(f"  leptonic fraction 3/9 = {L:.3f}, hadronic 6/9 = {H:.3f}")
print(f"  dilepton       (1/3)^2       = {L*L*100:4.1f}%    book: 10.6%")
print(f"  lepton + jets  2 x (1/3)(2/3)= {2*L*H*100:4.1f}%    book: 43.9%")
print(f"  all-hadronic   (2/3)^2       = {H*H*100:4.1f}%    book: 45.5%")

print("\nthe kinematic fit in lepton + jets -- why THAT topology:")
unk = [("neutrino px, py, pz", 3), ("two initial partons' longitudinal momenta", 2), ("the top mass itself", 1)]
con = [("energy-momentum conservation", 4), ("both reconstructed W masses = M_W", 2), ("the two top masses equal each other", 1)]
for i, (lab, n) in enumerate(unk):
    print(f"  {'unknowns   ' if i == 0 else ' '*11} {lab:41s} {n}")
print(f"  {' '*11} {'':41s} ---\n  {' '*11} {'':41s} {sum(n for _, n in unk)}")
for i, (lab, n) in enumerate(con):
    print(f"  {'constraints' if i == 0 else ' '*11} {lab:41s} {n}")
print(f"  {' '*11} {'':41s} ---\n  {' '*11} {'':41s} {sum(n for _, n in con)}")
print("  7 > 6, so the problem is OVERdetermined and is solved by minimising")
print("  a chi^2.  the leftover constraint is what makes the fit self-checking.")
print("\n  dilepton has a second neutrino: 3 more unknowns, same 7 constraints")
print("    9 unknowns vs 7 constraints -> UNDERdetermined by one, unsolvable")
print("    event by event, even though its background is the smallest")
print("  all-hadronic measures every momentum but drowns in QCD")
print("\nso the choice of topology is forced by COUNTING, not by rates.")

prints

top production at the Tevatron, sqrt(s) = 1.96 TeV
sigma(t tbar)      = 7.50 +- 0.48 pb
sigma(pp-bar, tot) = 81.9 +- 2.3 mb
ratio              = 9.2e-11   -> one top pair in 10^10 interactions

(single-top production, by the weak interaction, is about half of this)

the three tt-bar topologies, from BR(W -> l nu) = 1/9 per family
leptonic fraction 3/9 = 0.333, hadronic 6/9 = 0.667
dilepton       (1/3)^2       = 11.1%    book: 10.6%
lepton + jets  2 x (1/3)(2/3)= 44.4%    book: 43.9%
all-hadronic   (2/3)^2       = 44.4%    book: 45.5%

the kinematic fit in lepton + jets -- why THAT topology:
unknowns    neutrino px, py, pz                       3
            two initial partons' longitudinal momenta 2
            the top mass itself                       1
                                                      ---
                                                      6
constraints energy-momentum conservation              4
            both reconstructed W masses = M_W         2
            the two top masses equal each other       1
                                                      ---
                                                      7
7 > 6, so the problem is OVERdetermined and is solved by minimising
a chi^2.  the leftover constraint is what makes the fit self-checking.

dilepton has a second neutrino: 3 more unknowns, same 7 constraints
  9 unknowns vs 7 constraints -> UNDERdetermined by one, unsolvable
  event by event, even though its background is the smallest
all-hadronic measures every momentum but drowns in QCD

so the choice of topology is forced by COUNTING, not by rates.

Bettini Eqs. (9.96)–(9.98). The middle row wins on none of the individual criteria and wins overall, which is the usual shape of an experimental choice.

topology↕	fraction↕	final state↕	verdict↕
dilepton	10.6 %	2 opposite-sign leptons, 2 neutrinos, 2 $b$ jets	cleanest background — but underdetermined by one, so no event can be solved
lepton + jets	43.9 %	1 lepton, 1 neutrino, 4 jets of which 2 are $b$	the one used. Overdetermined by one constraint, and it supplies its own jet energy calibration through the hadronic $W$
all-hadronic	45.5 %	6 jets, 2 of them $b$	every momentum measured, no missing energy — and buried under QCD dijet production

📏 The top mass is not an observable, and the book says so

Two caveats attach to $m_t = 173.5 \pm 1.0$ GeV, and they matter more than the error bar.

First, a quark mass is not measurable in the way a hadron mass is. The top is coloured, and only colour singlets are observable. What is reconstructed from jets is sometimes called the “jet mass” or “Monte Carlo mass”, and its relation to the pole mass — the parameter in the Lagrangian — is known but carries its own theoretical uncertainty, which the book says is larger than the quoted experimental one.

Second, the world average is better and the caveat is bigger. ATLAS and CMS give $m_t = 172.69 \pm 0.30$ GeV — 0.17 % — and the book immediately adds that “an additional theoretical uncertainty of about 0.5 GeV must be added” once the mass-definition ambiguity and the colour structure of fragmentation are included.

That is a 0.30 GeV experimental error and a 0.5 GeV theoretical one. The measurement has been better than the theory that interprets it for over a decade — an unusual situation, and worth noticing, because the electroweak fit of §9.20 consumes $m_t$ as an input and inherits that limit.

D0, 4.3 fb⁻¹ — data and simulation are indistinguishable
background
M_W/2 = 40.2 GeV

Fig. 9.34 redrawn (D0, 4.3 fb⁻¹). Compare Fig. 9.17 from 1983 in §9.7: the same distribution with sixty thousand events per bin instead of sixty, and a systematic-limited 0.015 % instead of 4 %.

🔑 If you remember only three things

A divergence fixed twice is a structure, not a patch. The cross-section runs away without the self-coupling and is rescued by it, so the coupling is required rather than permitted.
Energy is fixed at construction and luminosity is not. That single asymmetry is why a machine’s best measurements often arrive at the end of its life.
The top’s mass was known before anyone made one. Virtual contributions shift what a precision experiment sees, so the answer came from a machine that could not have created it.

Where this goes next

Both inputs to the electroweak fit are now precise: $M_W$ to 0.015 % and $m_t$ to 0.17 %. What they constrain, through the logarithmic dependence of §9.8, is the one particle still missing.

§9.12 finally explains where the masses came from — the spontaneous breaking of the gauge symmetry, and why it is the only way to give a gauge boson mass without wrecking the theory. It also predicts a scalar, whose couplings are all fixed by its mass, and whose mass is fixed by nothing.

✅ Check yourself — self-coupling, and two precision masses

0/6 answered · 0 correct

1.What does the measured e⁺e⁻ → W⁺W⁻ cross-section actually test?
2.The book says the neutrino-only calculation gives 'the intermediate curve in Fig. 9.33', and two sentences later says the ν + γ sum gives 'the intermediate curve in Fig. 9.33'. What is wrong?
3.Across a multi-year upgrade the Tevatron's energy went from 1.8 to 1.96 TeV — 9 % — while its luminosity rose by a factor of 63. Why that asymmetry?
4.Of the three tt̄ topologies, lepton+jets has neither the smallest background nor the largest branching fraction. Why is it the one used for the mass?
5.How is the jet energy scale calibrated in the top-mass measurement?
6.m_t = 172.69 ± 0.30 GeV, and the book immediately adds that ~0.5 GeV of theoretical uncertainty must be added. What is that about?