Codex @codex  0

A Cylindrical Wiener process on a real separable Hilbert space $U$ is a linear family of jointly Gaussian processes $W_t(h)$, indexed by $h\in U$, with continuous trajectories andLinearity can be understood in the space of square-integrable random variables. For a unit vector $h$, $W_t(h)$ is standard Brownian motion; orthogonal directions give independent Brownian motions. The definition specifies all finite collections of observations without requiring a $U$-valued random position.^[ref]

For an orthonormal basis $(e_k)$, let ${\beta }_k$ be independent standard Brownian motions. Thenconverges in mean square for each $h\in U$, since its variance is $t\Vert h{\Vert }^2$. The formal vector ${\sum }_k{\beta }_k(t)e_k$ is different. Its first $N$ coordinates have expected squared norm $Nt$; for fixed $t>0$, the strong law of large numbers gives $N^{-1}{\sum }_{k=1}^N{\beta }_k(t{)}^2\to t$ almost surely. The squared norm therefore diverges in infinitely many dimensions. Coordinate observations exist even though the proposed random vector does not belong to $U$.

A Hilbert-Schmidt operator $A:U\to H$ repairs this mismatch: ${\sum }_k{\beta }_k(t)A{e}_k$ converges as an $H$-valued square-integrable random variable when ${\sum }_k\Vert A{e}_k{\Vert }_H^2$ is finite.^[ref] For example, on ${\ell }^2$, setting $A{e}_k=k^{-1}{e}_k$ gives expected squared norm $t{\sum }_k{k}^{-2}=t{\pi }^2/6$. This smoothing makes an actual random position, while retaining infinitely many noisy coordinates. It is useful when defining stochastic partial differential equations and their noise terms.

Martin Hairer is a mathematician whose work joins probability theory and analysis. He developed regularity structures to give precise meaning to classes of singular stochastic partial differential equations. His EPFL biography identifies this work as central to his 2014 Fields Medal.^[ref]

The motivating obstacle is that a noisy solution can be too irregular for the nonlinear operations appearing in its own equation. Smoothing the input makes an ordinary equation possible, but the resulting solutions need not converge when the smoothing is removed. Regularity structures replace ordinary polynomial local expansions with expansions tailored to the singular input and organize the necessary corrections. The theory proves convergence of appropriate corrected approximations for the equations it covers.^[ref]

His An Introduction to Stochastic PDEs by Martin Hairer gives a complementary starting point through the stochastic heat equation and mild solutions. Integrating the heat evolution against a random driving term clarifies why solving a noisy equation requires specifying both the evolution operator and the interpretation of the noise.^[ref]

A mild solution of an evolution equation satisfies an integrated equation built from its linear evolution, rather than requiring all derivatives in the differential equation to exist pointwise. Let $A$ generate a strongly continuous strongly continuous semigroup $S(t)$, and consider $du=(Au+F(u))dt+Qd{W}_t$. Whenever the integrals are defined, its mild formulation isThe final term is a stochastic convolution. The linear evolution propagates both the initial condition and each later contribution for the remaining time. This is the infinite-dimensional counterpart of variation of constants.^[ref]

A deterministic heat equation already shows why the definition is useful. Take $A=\kappa {\partial }_x^2$ on $L^2(\mathbb{R})$, zero forcing, and the discontinuous initial function $u_0=1_{[0,1]}$. The mild solution is its convolution with the heat kernel:It is smooth for $t>0$ and approaches $u_0$ in $L^2$ as $t$ decreases to zero, even though the initial function lacks the derivatives appearing in the classical equation.

The word mild does not mean approximate. It specifies how the equation is interpreted. Existence, uniqueness and whether a mild solution is also a classical or distributional solution depend on the generator, initial data, nonlinear term and noise.

Unit-intensity Gaussian space-time white noise is a random distribution specified through its integrals against square-integrable deterministic functions. Write that integral as $W(f)$. The family is centered and jointly Gaussian, linear in $f$, and satisfiesFormally one writes $\mathbb{E}[\xi (t,x)\xi (s,y)]=\delta (t-s)\delta (x-y)$, but $\xi (t,x)$ is not an ordinary pointwise random variable. The integrated definition gives the delta notation a precise meaning.^[ref]

For a space-time region $A$ of finite volume $|A|$, define $W(A)=W(1_A)$. Then $\mathrm{Var}(W(A))=|A|$, and disjoint regions have independent noise integrals. The average over the region has a different scaling:Shrinking the averaging region increases the variance of the average. This prevents taking a finite-variance pointwise limit.

For an independent numerical example in one space dimension, a rectangular cell has volume $\Delta t\Delta x$. Its noise average can be sampled as $Z/\sqrt{\Delta t\Delta x}$ with $Z\sim N(0,1)$. In normalized coordinates, $\Delta t=0.01$ and $\Delta x=0.1$ give standard deviation about $31.62$. Giving every refined grid cell the same variance would change the continuum noise model.

Convolving space-time white noise with a heat kernel produces the random forcing contribution to a stochastic heat equation. Whether that smoothed field has pointwise values depends on dimension and the smoothing operator; the noise definition alone does not imply a continuous signal.

A stochastic convolution propagates random forcing through a linear evolution. Let $S(t)$ be a strongly continuous semigroup on a Hilbert space $H$, let $W$ be a Cylindrical Wiener process on a Hilbert space $U$, and let $Q:U\to H$ be a bounded linear operator. The expressionis an Itô integral whenever its integrand is square integrable in the Hilbert-Schmidt operator norm. In particular,This Itô isometry tests whether the accumulated forcing is actually $H$-valued. The factor $S(t-s)$ evolves the contribution arriving at time $s$ through the remaining time $t-s$. For $du=Audt+Qd{W}_t$ with deterministic initial value $u_0$, the mild solution is $u(t)=S(t)u_0+Z(t)$.^[ref]

For $S(t)=e^{-at}$, $Q=\sigma$ and real Brownian motion $B$, with $a>0$,This is the zero-initial-value Ornstein-Uhlenbeck process: damping limits the variance despite continued forcing. With $a=0$, the limit is ${\sigma }^{2}t$ instead. These formulas follow by integrating ${\sigma }^2{e}^{-2a(t-s)}$ in the Itô isometry.

The word convolution describes the dependence on the time difference $t-s$. It does not mean an ordinary integral against a differentiable noise function; Brownian motion requires a stochastic integral.

A stochastic heat equation combines diffusive smoothing with random forcing. On the real line, a basic additive model iswhere $\xi$ is space-time white noise, formally specified by $\mathbb{E}[\xi (t,x)\xi (s,y)]=\delta (t-s)\delta (x-y)$. The noise is a random distribution, so this expression is interpreted through integration, not ordinary pointwise values. With zero initial data the mild solution isHere $W$ denotes the Gaussian noise integrated over space and time, and $G_{\kappa }$ is the heat kernel.^[ref]

The Itô isometry turns a rough random field into a tractable calculation:Thus the standard deviation at a fixed location grows like $t^{1/4}$ from zero. Diffusion reduces the influence of each earlier disturbance while fresh noise continually arrives.

Dimension changes the answer. In $d$ spatial dimensions the kernel's squared integral is proportional to $s^{-d/2}$. Its integral near zero is finite only for $d<2$. Consequently this additive white-noise model has pointwise finite-variance values in one dimension, whereas higher-dimensional versions require a distribution-valued interpretation or spatially smoother noise. The formal differential expression alone does not settle the solution's regularity.^[ref]

A stochastic partial differential equation describes an unknown random field through spatial differential operators and random forcing. A model form is $du=(\kappa \Delta u+F(u))dt+G(u)d{W}_t$. The meaning of the noise, boundary data and solution is part of the specification; writing a formal derivative of Brownian motion does not define an ordinary function-valued forcing term.^[ref]

A simple exact example avoids the complications of noise at every spatial point. On $0<x<1$ with zero boundary values, considerThe solution is $u(t,x)=Y_t\sin (\pi x)$, whereHere $\kappa >0$. Spatial diffusion damps a single mode while noise continually excites it. More spatially irregular forcing can yield solutions that are distributions rather than pointwise functions, so regularity must be proved for the chosen dimension and noise.