# Research Diary June2018

## Thursday-June 14-2018

1. A survey on KPZ and universality class by Ivan Corwin: arXiv 1106.1596. Another notes by Ivan Corwin: Exact solvability of some SPDEs. The stochastic heat equation (SHE) with multiplicative noise looks like (in differential form; formally, no exact meaning at the beginning)
$\left\{ \begin{array}{l} {\partial _t}z = \frac{1}{2}{\partial _{xx}}z + z\xi \\ z(0,x) = {z_0}(x) \end{array} \right.$ where $z:\mathbb R_{+}\times \mathbb R \to \mathbb R$ and $z_0$ is (possibly random) initial data which is independent of the white noise $\xi$ (will be described in detail later). The only thing to keep in mind right now is $\xi$ can be thought of as a random distribution. (SHE) makes sense by the following notion of solution.

Definition. A mild solution to the SHE for all $t>0$ and $x\in \mathbb R$ is $z(\cdot,\cdot)\in L^2(\Omega,\mathcal{F},\mathbb P)$ satisfies the Duhamel formulation
$z(t,x)=\int_{\mathbb R}p(t,x-y)z_0(y)dy+\int_0^t\int_{\mathbb R}p(t-s,x-y)z(s,y)\xi(s,y)dyds$ where $p(t,x)=\frac{1}{\sqrt{2\pi t}}e^{-x^2/2t}$ is the heat kernel or fundamental solution to $\partial_tp=\partial_{xx}p$ with $p(0,x)=\delta_{x=0}$. Also, we must have a priori that $\int_0^t\int_{\mathbb R}p^2(t-s,x-y){\mathbb E}[z^2(s,y)]dyds<\infty$ for the Ito integrals (stochastic integrals) to make sense and finite.

## Friday-June 15-2018

2. Martin Hairer's regularity structures (Link): Decomposition of heat kernel (Lemma 5.5, Theorem 5.12). "Canonical model" (Proposition 8.27). Renormalization is to "correct" Picard iteration at each step. To see how renormalization works, consider the term $({\mathcal D}{\mathcal I}(\Xi))^2$: $\Xi$ is the white noise, thought of as derivative of Brownian motion; hence, the action ${\mathcal D}{\mathcal I}$ on $\Xi$ basically recovers Brownian motion. To correct this, we have $M(({\mathcal D}{\mathcal I}(\Xi))^2)=({\mathcal D}{\mathcal I}(\Xi))^2-\frac{c}{\varepsilon}$ where $M=I-cL-\cdots$ where $L$ represents the action of "removing" Brownian motion.

3. Problem about feeble fish. In homogenization theory, the underlying vector field is set to be random. Roughly speaking, consider a (random) vector field $V:(\text{space},\text{time})\to \mathbb R$. Consider the following dynamics: $dX=Vdt,\quad \text{(or in discrete settings)}\quad X_{n+1}=X_n+V(X_n,n).$ We would expect in the long time, $X\sim \mathbb E(V) \cdot t$, or equivalently, $\frac{X}{t}\to \mathbb E(V)$ where the expectation is respect to the underlying probability measure. The main question is under what condition this would hold true.

Always to think about the purpose of defining an object.

We first work in the discrete settings. To even talk about expectation $\mathbb E$, we must have stationarity in both space and time. That's the first condition. We will consider its "ergodicity" (the relationship between time and space average). Let $W:\mathbb Z^2\to \mathbb Z^2$ is a deterministic velocity (with compact support to ensure convergence later), and $\xi_{ij}$ be i.i.d random variables (taking only $0$, $1$ or any finite set of values). One way to construct the random velocity $V$ is as follows $V(x,t)=\sum\limits_{ij}\xi_{ij}V(x+i,t+j).$ Consider the (canonical) space-time shift $T_{ij}:\mathbb Z^2\to \mathbb Z^2$ $T_{ij}(x,t)=(x+i,t+j).$ These shifts are measure-preserving $V(x,t,T_{ij}\omega)=V(x+i,t+j,\omega).$ The claim is that $V$ is strongly mixing with respect to these shifts. And we indeed get that $X_n/n\to \mathbb E(V)$ (link). The main question is can we reduce the strong mixing condition to just being ergodic, or is there a counterexample? "Ergodic" means that $\frac{1}{N}\sum\limits_{n=1}^{N} V(T_{ij}^{n}(x,t)) \to \int Vd\mathbb P$ where $\mathbb P$ is the underlying probability measure. Do we have $\frac{1}{N}\sum\limits_{n=1}^{N} V(X_n,n)\to\int Vd\mathbb P\;\;$ as $X_n$ follows the dynamics $X_{n+1}=X_n$$\;+ \;V(X_n,n)$?

4. How to solve stochastic heat equation: link

## Sunday-June 17-2018

1. The Birkhoff-von Neumann theorem states that the set of $n \times n$ doubly stochastic matrices (non-negative matrices whose every row and every column sums to $1$) is convex, and that the extremal points of the set are (all) the permutation matrices.

2. The Hoffman-Wielandt inequality states that for any two $n \times n$ normal matrices $A$, $B$, with sets of eigenvalues $\lambda_1(X),\; \lambda_2 (X),\;\ldots,\; \lambda_n(X)$ for $X \in \{A, B\}$, there exists a permutation $\sigma \in S_n$ such that $\sum_{i=1}^n |\lambda_i(A) - \lambda_{\sigma(i)}(B)|^2 \leq \text{Tr}\left((A-B)(A-B)^*\right).$ Proof. (kind of) Express $\text{Tr}\left((A-B)(A-B)^*\right)$ as a function on doubly stochastic matrices. Then use Birkhoff-von Neumann theorem to achieve the minimum and maximum.