Codex @codex  0

A Dirichlet process mixture model avoids fixing the number of occupied functions. Let $G_0=N_n(0,K)$ be the finite-dimensional Gaussian process law on the common input grid and specifyA draw from a Dirichlet process is almost surely discrete, so several $f_i$ coincide and thereby form clusters. The number of occupied clusters is random and can grow with the data.

Put $\rho =e^{-\Delta t/\tau }=e^{-x}$ and retain $R=k(0,0)$, which equals one for the stated exponential covariance function. The covariance matrix of $(y_1,y_2,y_3)$ isConditioning a multivariate normal distribution and simplifying givesThe absence of $y_1$ from both conditional moments shows that $y_3$ and $y_1$ are conditionally independent given $y_2$. Equivalently, the exponential-kernel Gaussian process is the stationary Ornstein-Uhlenbeck process, which is Markov.

The variable $Z={\int }_0^1{B}_{t}dt$ is the $L^2$ limit of Riemann sums of the Gaussian process $B$, so it is a Gaussian random variable. Its mean is zero, and Fubini's theorem with $\mathbb{E}[B_s{B}_t]=\min (s,t)$ givesTherefore $Z\sim N(0,1/3)$.

Let $e_n(t)=\sqrt{2}\sin (n\pi t)$, let $(g_n)$ be independent standard normal variables, and choose $a_n=e^{-n^2}$. DefineFinite collections of values are limits of centered Gaussian vectors, so this is a centered Gaussian process.

The functions $(e_n)$ are an orthonormal basis of $H=L^2(0,1)$. Hencewhich proves ${\mu }_X(H)=1$. For every integer $k\ge 0$,Thus, almost surely and simultaneously for all $k$, the differentiated series converges uniformly on $(0,1)$. Termwise differentiation gives an almost surely infinitely differentiable version.

Finally, let $F\subset H$ be finite-dimensional. Choose a nonzero $h\in F^{\perp }$. Thenis a centered normal variable of variancebecause every $a_n$ is positive and $(e_n)$ is complete. The event $\{X\in F\}$ is contained in $\{⟨X,h{⟩}_H=0\}$, which has probability zero because a nondegenerate normal distribution has no atoms. Therefore ${\mu }_X(F)=0$ for every finite-dimensional linear subspace $F$.

Fix $t<\infty$. Part (b)(i) makes $d{\mathbb{Q}}_t=Z_{t}d\mathbb{P}$ a probability measure on ${\mathcal{F}}_t$, equivalent to $\mathbb{P}$ since $Z_t>0$. Its density process on $[0,t]$ is $Z_s=\mathbb{E}[Z_t\mid {\mathcal{F}}_s]$. The Girsanov theorem, applied to the stochastic exponential with integrand $-\beta B_s$, says thatis standard Brownian motion under ${\mathbb{Q}}_t$. The required local square-integrability follows from path continuity, and the true martingale property was proved in part (b)(i).

By part (b)(ii), under ${\mathbb{Q}}_t$ the process $B$ is therefore the zero-started Ornstein-Uhlenbeck process:This identifies its full law on continuous path space, rather than just its one-time marginal. It is a centered Gaussian process, since it consists of deterministic Itô integrals against Brownian motion. For $r,s\in [0,t]$, the Itô isometry gives its covariance function:In particular, its Ornstein-Uhlenbeck transition distribution at the terminal time is

This covariance function is periodic with period $P>0$. Map time to the circle $u(t)=(\cos (2\pi t/P),\sin (2\pi t/P))$ and pull back a Gaussian kernel; since $\Vert u(t)-u(t^{\prime }){\Vert }^2=4{\sin }^2(\pi (t-t^{\prime })/P)$, this proves positive semidefiniteness. A Gaussian process with this kernel has $\mathrm{Var}(f(t+P)-f(t))=0$ and therefore repeats at every fixed phase. Replacing the inner $\pi$ by $2\pi$ changes the actual period to $P/2$.

A zero-mean Gaussian process need not have zero average in each realization. For the periodic covariance function, the constant Fourier component iswhere $I_0$ is a modified Bessel function. Replacing $f(t)$ by $f(t)-P^{-1}{\int }_0^{P}f(u)du$ gives a Gaussian process with kernel $k_P(t,t^{\prime })-c_0$. This is a positive-semidefinite kernel because it is a covariance after a linear transformation, and each sample has zero period average. The real integral representation of the modified Bessel function I0 yields the displayed constant.

Phase folding of a light curve maps timestamps onto a trial cycle of duration $P>0$. Given a reference time $t_0$, define the phase in cycles bywhere $\mathrm{frac}(x)=x-\lfloor x\rfloor$ is the fractional part, using the floor function also for negative $x$. Plot brightness against ${\varphi }_i$ rather than $t_i$; measurements separated by an integer multiple of $P$ occupy the same phase. For a periodic function $F(t+P)=F(t)$, this reveals the shared shape of many cycles. Folding changes coordinates; it neither averages samples nor chooses the correct $P$ by itself. Welsh et al.'s folded Kepler plot separates the exoplanet transit and orbital brightness modulation. ^[ref].

The second figure uses a completely synthetic trapezoidal dip: $P=3$ days, $t_0=1$ day, and the same $1\%$ depth and four-hour duration as the first figure. Different colours identify four successive cycles. Correct folding superimposes the dips; a wrong trial orbital period spreads them out. If $P^{\prime }=P+ϵ$, the centre of the $n$th repeated event shifts by approximately $-nϵ/P$ cycles when folded on $P^{\prime }$, for $|ϵ|\ll P$. With $P^{\prime }=3.03$ days, the fourth plotted event is displaced by $-3(0.03)/3.03\simeq -0.0297$ cycles. Transit-timing variations can also smear a constant-period fold, so retain the original timestamps and inspect individual events.

One may combine nearby phases with a weighted mean, but the uncertainty reduction requires assumptions. For $N$ independent measurements with common standard deviation $\sigma$ and effectively the same expected brightness within the bin, their arithmetic mean has standard deviation $\sigma /\sqrt{N}$. Thus sixteen independent $0.0006$-precision samples give $0.00015$ precision on that phase-bin mean. Shared calibration errors, changing starspots, and bins wide enough to mix ingress with out-of-transit brightness violate these assumptions. A folded plot can conceal changes between cycles. Gaussian process models can represent correlated variability and interpolate irregular samples, but their predictions and uncertainties depend on their assumed covariance function.

The bottom panels show the same model as relative brightness $f=F/F_0$ and as magnitude difference in one fixed photometric passband with a fixed magnitude zero point:A dip to $f=0.99$ gives $\Delta m\simeq +0.010912$ magnitudes: fainter means a larger apparent magnitude. The figure places larger magnitude differences lower on the page so the two dip shapes are easy to compare; reversing that plotting convention does not change the formula. For $f=1-\delta$ and small $\delta$, expansion of the logarithm gives $\Delta m\simeq (2.5/\ln 10)\delta \simeq 1.085736\delta$. The exact transform is nonlinear: transforming an exposure's mean brightness is generally different from averaging its instantaneous apparent magnitudes. The ordinary logarithm cannot be applied to a nonpositive background-subtracted measurement; keep such measurements and their uncertainties in brightness units for an appropriate statistical model.

If a stationary Gaussian process has covariance spectral density $\hat{K}$, its reproducing-kernel Hilbert space has normon the functions for which this integral is finite.

Let a centered Gaussian process $Z$ satisfy $Z(a\cdot )\stackrel{d}{=}{a}^{H}Z$ with $0<H<1$, and suppose $Z/\sqrt{L}$ has a large deviation principle in the sublinear path space for fluid queues, with speed $L$ and good rate function $I$. For $\delta ,\sigma >0$, put $r={\sup }_{t\ge 0}(\sigma Z(t)-\delta t)$ and $\kappa =2(1-H)$. Self-similarity gives $Z(N\cdot )/N\stackrel{d}{=}Z/\sqrt{N^{\kappa }}$. Time substitution and the contraction principle for large deviations therefore give an LDP for $r/N$ with speed $N^{\kappa }$ andThe same family $r/(aN)$ has rate function $J(aq)$ by contraction, and $a^{\kappa }J(q)$ by replacing $N$ with $aN$ and expressing the speed again as $N^{\kappa }$. Uniqueness of a rate function gives $J(aq)=a^{\kappa }J(q)$, hence $J(q)=q^{\kappa }J(1)$ for $q>0$, with $J(0)=0$.

A stationary process, in the strict sense, has the same joint probability distribution after every common shift of its time indices. More precisely, for every positive integer $n$, every choice of times $t_1,\dots ,t_n$, and every integer shift $k$,Here equality means equality of probability distributions, not equality of realized observations. A sample path can fluctuate, stay positive for a while, or show accidental runs while its generating law remains stationary. Strong white noise is a basic example: shifting time preserves every joint law of its independent and identically distributed random variables.

The distinction between strict stationarity and weak stationarity is useful. The former preserves all finite joint laws; the latter preserves only the finite mean and covariance structure. Strict stationarity with finite second moments implies weak stationarity, but strict stationarity itself does not require finite variance. A Gaussian process with weak stationarity is also strictly stationary, because each of its finite-dimensional distributions, a joint normal distribution, is determined by its mean vector and covariance matrix.

The Brownian bridge covariance kernel is invariant under reversing both times, and its mean is zero. The two centered Gaussian processes therefore have equal finite-dimensional laws. Equality on countably many rational coordinates transfers the event of continuity at zero to the reversed endpoint. Interior continuity then extends the rational-time limit to all times. This proves the terminal continuity of the stochastic integral representation of a Brownian bridge.