The peaks-over-threshold method fits a generalized Pareto distribution to exceedances of a high threshold and also estimates their frequency. Together these determine tail probabilities and return levels. For dependent observations, threshold exceedances may occur in clusters; declustering of extremes can separate extreme events before fitting a model for independent events.

A deviance residual is the signed square root of an observation's contribution $d_i$ to fitted-model deviance. For a Poisson regression, $d_i=2[y_i\log (y_i/{\hat{\mu }}_i)-(y_i-{\hat{\mu }}_i)]$, with $0\log 0=0$. Its squared sum is the Poisson deviance. Leverage and estimated dispersion can further standardize these residuals; a normal quantile-quantile plot is a diagnostic approximation, not a requirement that count errors have a normal distribution.

For a suitably regular generalized linear model, a residual deviance $D$ can be compared approximately with a chi-squared distribution on $n-p$ degrees of freedom to assess fit relative to the saturated model. Sparse or very small expected cell counts can invalidate this reference distribution. A large p-value means no detected lack of fit, not proof of correctness; dependence, wrong mean structure or overdispersion still require diagnostics.

The general linear group over a finite field $G{L}_2({\mathbb{F}}_2)$ acts faithfully on the four vectors of ${\mathbb{F}}_2^2$, since a matrix fixing all vectors fixes the standard basis. The zero vector is fixed by every element; the induced action on the three nonzero vectors identifies this six-element group with $S_3$. Swapping the two coordinates fixes $(0,0)$ and $(1,1)$ and exchanges the other two vectors, so its four-point permutation is odd. This action therefore embeds the group into $S_4$ with nontrivial composite sign homomorphism.

The pole contour in a Gaussian saddle coalescing with a simple pole changes the multiplier smoothly. For the lower contour, ${\int }_{{\mathcal{C}}_{-}}{e}^{-u^2-i\mu u}du/u=i\pi \mathrm{erfc}(\mu /2)$. This follows by integrating $-i\sqrt{\pi }{e}^{-{\mu }^2/4}$ and adding the indentation contribution $i\pi$. The transition width is inverse square root of the large parameter; on a Stokes line the multiplier takes half its switched-on value.

Gaussian process classification links latent variables from a Gaussian process to class probabilities. A binary probit regression likelihood is ${\prod }_i\Phi ((2y_i-1)f_i)$ under conditional independence. The posterior distribution of the latent values is not a normal distribution.

A Gaussian process has a mean-square derivative at $t$ when its difference quotients converge in L2 space. The mean-square convergence need not give differentiable sample paths. For a centered stationary process with twice continuously differentiable covariance function $K(t-s)$, the difference quotients are Cauchy in L2 space, since their covariance is a rectangular average of $-K^{\prime \prime }$. Their limits form a Gaussian process with covariance function $\mathbb{E}[Y(s)Y(t)]=-K^{\prime \prime }(s-t)$.

For centered multivariate normal distributions represented by vectors $U,V$ with equal corresponding variances and $\mathbb{E}[U_i{U}_j]\ge \mathbb{E}[V_i{V}_j]$ for all distinct indices, $\mathbb{P}(U_i\le u_i\text{ for all }i)\ge \mathbb{P}(V_i\le u_i\text{ for all }i)$ for every threshold vector. In particular the supremum of $U$ is stochastically smaller than the supremum of $V$. Equal corresponding variances are essential to this formulation. The related Sudakov-Fernique inequality compares expected values of maxima using variances of increments and does not require equal pointwise variances.

For centered multivariate normal distributions, if $\mathbb{E}(U_i-U_j{)}^2\le \mathbb{E}(V_i-V_j{)}^2$ for every pair, then the displayed comparison holds. For a quick proof, make $U,V$ independent and set $Z_r=\sqrt{1-r}U+\sqrt{r}V$. Apply Gaussian integration by parts to $F_{\tau }(z)={\tau }^{-1}\log {\sum }_i{e}^{\tau z_i}$. With softmax function weights $p_i$ and $D=\mathrm{Cov}(V)-\mathrm{Cov}(U)$, the derivative of $\mathbb{E}{F}_{\tau }(Z_r)$ is $\frac{\tau }{4}\mathbb{E}{\sum }_{i,j}{p}_i{p}_j(D_{ii}+D_{jj}-2D_{ij})\ge 0$. Since $\max z_i\le F_{\tau }(z)\le \max z_i+{\tau }^{-1}\log n$, let $\tau \to \infty$. Singular covariance matrices follow by adding identical small independent normal noise to both vectors and taking a limit.

A fractional Brownian motion with Hurst exponent $H\in (0,1)$ is a centered Gaussian process with the displayed standard covariance function and a continuous modification. Its increment variance is $|s-t{|}^{2H}$. Some sources omit the factor $1/2$, multiplying the process by $\sqrt{2}$; the corresponding increment variance is then $2|s-t{|}^{2H}$. Its canonical pseudometric of a Gaussian process is proportional to $|s-t{|}^H$, which gives finite expected value of its absolute supremum on compact intervals through the Dudley entropy integral.

The canonical distance is the L2 norm of an increment of a Gaussian process. It is a pseudometric, since distinct parameters may represent equal random variables almost surely. Its metric covering number appears in the Dudley entropy integral. For standard fractional Brownian motion, $d(s,t)=|s-t{|}^H$; a normalization with twice the covariance function multiplies this distance by $\sqrt{2}$.

For independent standard normal variables ${\xi }_i$ and ${\sum }_i{k}_i(t{)}^2<\infty$ at each fixed time, the series defines $X_t$ in $L^2$. Every finite linear combination is an $L^2$ limit of Gaussian variables and is Gaussian. Thus this constructs a Gaussian process with the Gram covariance kernel $K$. It supplies no sample-path continuity without additional assumptions.

For quadratic dispersion $p^2+t$ with no interacting anomalous dimension, the thermal and uniform-source scaling eigenvalues are $y_t=2$ and $y_h=(d+2)/2$. The generic homogeneous singular free-energy density has $2-\alpha =d/2$ and gap exponent $\Delta =(d+2)/4$. These Gaussian-model values below the upper critical dimension are not the exponents of an interacting scalar critical point.

A local field combination is a gauge-invariant operator when its gauge representations and charges combine to a singlet. In an effective field theory, such operators may be multiplied by Wilson coefficients and inverse powers of a heavy scale. Multiplication by ${\varphi }^{†}\varphi$ preserves the invariance of an existing Higgs Yukawa operator.

If a real nonzero field $\xi (x)$ appears in the Lorentzian density $-F_{ab}{F}^{ab}/4+({\partial }_a{A}^a{)}^2/(2\xi )$ and is varied, its algebraic Euler-Lagrange equation is $-({\partial }_a{A}^a{)}^2/(2{\xi }^2)=0$. Hence it imposes the Lorenz gauge condition on real classical fields. Varying $A_b$ gives ${\partial }_a{F}^{ab}-{\partial }^b[(\partial \cdot A)/\xi ]=0$, including derivatives of $\xi$. This constrained field theory differs from holding the gauge parameter fixed in a Gaussian functional integral.

A gauge-fixed action supplements a gauge-invariant action with a term imposing the gauge condition and with Faddeev-Popov ghost fields representing the Faddeev-Popov determinant. A local gauge condition gives local differential operators and a local action.

In gauge field theory its standard form is $A^{+}=n_{\mu }{A}^{\mu }=0$ for a fixed null direction $n$. A light-cone gauge uses light-cone coordinates to fix a longitudinal variable, often choosing $x^{+}$ as the evolution parameter for a relativistic particle phase-space action. Related field gauges set selected minus components to zero when a field has the required gauge invariance. A massive field without that gauge freedom must instead eliminate dependent components through its equations of motion.

The Jacobi identity for gauge covariant derivatives gives $D_{[\mu }{F}_{\nu \rho ]}=0$. In four dimensions this is $D_{\mu }{\tilde{F}}^{\mu \nu }=0$, with ${\tilde{F}}^{\mu \nu }=\frac{1}{2}{\epsilon }^{\mu \nu \rho \sigma }{F}_{\rho \sigma }$. It holds independently of the Yang-Mills equations.

In the standard Dirac representation of the gamma matrices, $({\gamma }^0{)}^{†}={\gamma }^0$ and $({\gamma }^i{)}^{†}=-{\gamma }^i$, so ${\gamma }^{a†}={\gamma }^0{\gamma }^a{\gamma }^0$. The transpose signs in that representation are positive for $a=0,2$ and negative for $a=1,3$. The charge-conjugation matrix expresses the basis-independent intertwining relation $C^{-1}{\gamma }^{a}C=-({\gamma }^a{)}^T$. Under an arbitrary nonunitary similarity transformation, the Hermitizing matrix changes by inverse Hermitian congruence; it need not equal the transformed temporal gamma matrix.

A Hermitizing matrix for a set of gamma matrices is an invertible Hermitian matrix $H$ satisfying ${\gamma }^{a†}=H{\gamma }^a{H}^{-1}$. Equivalently, each $H{\gamma }^a$ is Hermitian. It defines a covariant adjoint $\bar{\psi }={\psi }^{†}H$. If ${\gamma }^{\prime a}=M{\gamma }^a{M}^{-1}$, then $H^{\prime }=M^{-†}H{M}^{-1}$ has the same property. In a standard Lorentzian Dirac basis, $H={\gamma }^0$ is a conventional choice.