Electroweak Precision Observables

Electroweak precision testing compares a correlated vector of pole, charged-current, low-energy, and radiative observables with predictions evaluated in one explicit input and renormalization scheme. Its power comes from shared parameter dependence and loop corrections—not from treating individually quoted residuals as independent measurements. Numerical inputs and conclusions must be attached to dated releases; this page develops the reusable construction without a current fit snapshot.

Required background. Standard Model pseudo-observables and unstable particles supplies the complex-pole and extraction conventions used for $Z$- and $W$-resonance quantities.

Helpful background. Validation and theory uncertainties supplies the correlation and approximation checks needed to construct a prediction covariance.

An electroweak observable vector

A precision analysis begins with an ordered vector, not a list of labels:

$$\mathbf y= \bigl( M_Z,\Gamma_Z,\sigma^0_{\rm had},R_\ell^0, \mathcal A_\ell,A_{\rm FB}^{0,f},M_W,\Gamma_W, O_{\rm low\ energy},\ldots \bigr)^T.$$

The entries shown are categories, not a prescribed contemporary dataset. Each actual component needs a pole or fiducial definition, units, bin or flavor label, release identity, and covariance ordering.

At the $Z$ pole, useful pseudo-observables include

$$\sigma^0_{\rm had} =\frac{12\pi}{M_Z^2} \frac{\Gamma_e\Gamma_{\rm had}}{\Gamma_Z^2}, \qquad R_\ell^0=\frac{\Gamma_{\rm had}}{\Gamma_\ell}.$$

For light fermions, one convenient effective-coupling convention is

$$\Gamma_f =N_c^f\frac{G_FM_Z^3}{6\sqrt2\pi} \left(|g_V^f|^2+|g_A^f|^2\right)\mathcal R_f,$$

where $\mathcal R_f$ collects the stated QED, QCD, mass, and nonfactorizable corrections. The associated polarization combination is

$$\mathcal A_f =\frac{2\operatorname{Re}(g_V^f g_A^{f*})} {|g_V^f|^2+|g_A^f|^2}, \qquad A_{\rm FB}^{0,f}=\frac34\mathcal A_e\mathcal A_f.$$

These are pole-level definitions. Measured cross sections and asymmetries require a specified treatment of photon radiation, $gamma$–$Z$ interference, acceptance, and nonresonant terms before they can be represented by these quantities LEP and SLD Electroweak Working Groups 2006, §§1.5 and 2.3–2.6, pp. 33–50.

Other sectors use different interfaces:

Sector	Representative object	Corrections that must be assigned	Boundary of the object
$Z$ line shape and asymmetries	Pole mass, width, residues, $sigma^0$, $R_f$, $mathcal A_f$	Electroweak form factors, QED/QCD radiators, nonresonant subtraction	Extraction convention and stable final states
Charged current	$W$ pole parameters or fiducial distributions	Self-energies, vertices, boxes, radiation, recoil and acceptance	Pole and reconstruction definitions
Low-energy neutral current	Effective charges, parity-violating asymmetries, neutrino or atomic observables	Matching, running, hadronic/nuclear matrix elements, process-specific boxes	Momentum scale and target convention
Radiative observables	Inclusive or fiducial rates and asymmetries	Infrared-safe photon definition, real–virtual cancellation, isolation	Measurement function and perturbative accuracy

It is usually wrong to place all of these in a single “effective weak mixing angle.” Different processes project different form factors and receive different vertex and box contributions.

Input schemes turn measurements into predictions

Choose a renormalized input vector $\mathbf I$—for example a set built from $\alpha$, $G_F$, and pole masses—and define every remaining quantity as a prediction in that scheme. In the on-shell example below, define $\Delta r$ by

$$s_W^2c_W^2 =\frac{\pi\alpha} {\sqrt2G_FM_Z^2\left(1-\Delta r\right)}, \qquad s_W^2=1-\frac{M_W^2}{M_Z^2}.$$

This equation fixes the sign convention for $\Delta r$ locally. At tree level $\Delta r=0$; loop self-energies, vertices, boxes, and counterterms shift the implicit prediction for $M_W$. A different input scheme moves finite pieces between “input” and “correction,” but a consistently truncated physical prediction agrees up to terms beyond the stated order. One-loop renormalization, gauge cancellation, and real-radiation organization are reviewed in Denner 1993, §§3–7, pp. 317–401.

More generally, write the prediction map as

$$\boldsymbol\mu =\boldsymbol\mu(\mathbf p,\mathbf I;S,\mu_R,Q),$$

where $\mathbf p$ are parameters being tested, $S$ names the input/renormalization scheme, and scales are included when relevant. The calculation must consistently include:

1. the Born term expressed in the chosen inputs;
2. renormalized self-energy, vertex, and box contributions appropriate to each process;
3. real and virtual QED/QCD radiation under the observable definition;
4. matching and running to the process scale when an effective description is used; and
5. a declared perturbative and power-accuracy remainder.

For pole pseudo-observables, part of the nonresonant and radiation dependence may reside in the extraction map rather than the pole coefficient. For low-energy scattering, process-dependent boxes remain part of the prediction. Moving a term between these layers without changing the definition double counts or omits it.

Correlated uncertainty propagation

Let $\mathbf x$ be uncertain input quantities with covariance $C_x$. Linear propagation gives

$$J_{ia}=\frac{\partial\mu_i}{\partial x_a}, \qquad C_{\rm param}=J C_xJ^T.$$

This immediately creates correlations: one input can shift several observables coherently. Theory sources can be represented in the same way. If a vector of declared nuisance shifts $\boldsymbol\eta$ has covariance $C_\eta$ and response matrix $B$,

$$\boldsymbol\mu(\mathbf p,\boldsymbol\eta) =\boldsymbol\mu_0(\mathbf p)+B\boldsymbol\eta, \qquad C_{\rm th}=BC_\eta B^T.$$

When independent Gaussian components really are independent, a comparison may use

$$C_{\rm tot}=C_{\rm exp}+C_{\rm param}+C_{\rm th}, \qquad \chi^2=(\mathbf y-\boldsymbol\mu)^TC_{\rm tot}^{-1} (\mathbf y-\boldsymbol\mu).$$

But a covariance already obtained by profiling experimental nuisance parameters must not be combined with the same auxiliary constraints again. Likewise, a theory covariance and explicit theory nuisances are alternative encodings unless their components are demonstrably disjoint. Non-Gaussian or parameter-dependent uncertainties belong in a likelihood, not in a frozen symmetric matrix merely for convenience.

A release-ready observable record should therefore provide:

Item	Required specification
Observable	Formula or measurement function, pole/fiducial layer, units, flavor and bin ordering
Input scheme	Independent inputs, mass and width convention, renormalization scheme and perturbative order
Data object	Exact dataset/table identifier and version; covariance or statistical model version
Correlations	Source meaning, sign, scope across observables and experiments, and any profiled constraints
Theory	Calculation version, central prescription, scale/PDF/parametric components and correlation model
Applicability	Kinematic and EFT validity conditions, frozen exclusions, correction and supersession history

A representative sensitivity calculation

Suppose two observables depend on one uncertain input $x$ near $x_0$:

$$\boldsymbol\mu(x) \simeq \boldsymbol\mu(x_0) +\begin{pmatrix}a\\b\end{pmatrix}(x-x_0), \qquad \operatorname{Var}(x)=\sigma_x^2.$$

Then

$$C_{\rm param} =\sigma_x^2 \begin{pmatrix} a^2&ab\\ab&b^2 \end{pmatrix}.$$

Its determinant vanishes because one input produces a rank-one shift. The sign of $ab$ determines whether the induced correlation is positive or negative. Replacing this matrix by independent errors $|a|\sigma_x$ and $|b|\sigma_x$ discards the common direction and can manufacture apparent tension. The same Jacobian logic applies to electroweak inputs, hadronic corrections, and common missing-order components.

Independent checks and limits

Scheme translation. Reexpress a benchmark prediction in a second input scheme at the same perturbative order. The difference should have the size and parameter dependence of omitted higher orders, not a leading shift.

Gauge and infrared cancellation. Vary gauge-fixing parameters and infrared regulators before combining all required pieces. The observable must be gauge independent and regulator independent after the prescribed real–virtual combination.

Dimensions and normalization. Widths have mass dimension one, cross sections mass dimension minus two, and asymmetries are dimensionless. Setting all radiative form factors and radiators to their Born values must recover the tree-level normalization.

Covariance geometry. Verify the matrix is symmetric within numerical tolerance and positive semidefinite in the documented ordering. A Cholesky failure can reveal a transcription error, inconsistent rounding, or a covariance that needs its original nuisance representation.

Restricted interfaces. Oblique parameters summarize new physics only when its leading effects can be represented by the assumed gauge-boson two-point functions and expansion. They are not a universal replacement for vertex, box, flavor, or light-state effects.

Common pitfalls

Mixing pole and reconstructed masses. A resonance parameter extracted with a line-shape or template convention is not automatically the complex-pole mass. Translate the convention or keep the quantities distinct.

Holding derived inputs fixed twice. If $M_W$ is predicted from an input set, it cannot simultaneously be an independent fixed input in the same comparison. Define the independent coordinates before differentiating or fitting.

Adding errors component by component without correlations. Common electroweak inputs and theory variations generate coherent directions. Propagate their signed responses or use shared nuisances.

Inferring contemporary agreement from an evergreen formula. A formula defines a method, not the current numerical outcome. Any numerical comparison must identify the dated data, theory, and covariance releases used.

Informal self-check

Show why one shared uncertain input cannot produce a full-rank covariance for two observables at linear order.

Solution

With response vector $v=(a,b)^T$, the induced covariance is $C=\sigma_x^2vv^T$. Its image is the one-dimensional span of $v$, so $\operatorname{rank}C\le1$ and $\det C=0$. A full-rank two-observable covariance requires at least two independent response directions or an additional independent uncertainty component.

Handoffs

• Return to pseudo-observables and unstable particles when a pole or line-shape convention is unclear.
• Use collider measurements and likelihood provenance to bind a prediction to an exact fiducial release and statistical object.
• Use correlated Standard Model fits for shared nuisances, generalized least squares, goodness-of-fit, and stability tests.
• Use SMEFT, HEFT, and Standard Model observables only after declaring the EFT truncation and validity domain.

References

• Denner, Ansgar. “Techniques for the Calculation of Electroweak Radiative Corrections at the One-Loop Level and Results for $W$-Physics at LEP200.” Fortschritte der Physik 41 (1993) 307–420. DOI · Open PDF
• LEP Collaborations, ALEPH, DELPHI, L3, OPAL, SLD Collaborations, LEP Electroweak Working Group, SLD Electroweak and Heavy Flavour Groups. “Precision Electroweak Measurements on the $Z$ Resonance.” Physics Reports 427 (2006) 257–454. DOI · Open PDF