Abstract:We consider the precise upper large deviations estimates for the maximal displacement of a branching random walk. In addition, we obtain a description of the extremal process of the branching random walk conditioned on this large deviations event. This introduces a family of point measure playing a role similar to the decoration measures introduced in [9] for branching Brownian motion.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is about the precise large - deviation upper bound estimates of the maximum displacement in branching random walks. Specifically, the author focuses on the asymptotic behavior of the probability \(P(M_n\geq nx)\) that the maximum displacement \(M_n\) in the branching random walk exceeds its expected value \(nx\), where \(x > x^*\) and \(x^*\) is the asymptotic speed of the maximum displacement in the branching random walk. In addition, the paper also studies the description of the extreme process under the condition of such large - deviation events.
### Main contributions of the paper
1. **Large - deviation upper bound estimates**:
- Under nearly optimal conditions, the author obtains the precise asymptotic behavior of \(P(M_n\geq nx)\). Specifically, when \(\psi(0)>0\), we have
\[
\lim_{n\rightarrow\infty}\frac{1}{n}\log P(M_n\geq nx)=-\psi^*(x),\quad\text{for any }x > x^*.
\]
- When \(\psi(0)\leq0\), we have
\[
\lim_{n\rightarrow\infty}\frac{1}{n}\log P(M_n\geq nx)=-\psi^*(x),\quad\text{for any }x > \frac{E\left(\sum_{|u| = 1}V(u)\right)}{E(Z_1(\mathbb{R}))}\in\mathbb{R}\cup\{-\infty\}.
\]
2. **Precise asymptotic behavior**:
- The author further studies the precise asymptotic behavior of \(P(M_n\geq n\psi'(\theta)+y)\). When \(y = O(\sqrt{n})\), we have
\[
\lim_{n\rightarrow\infty}\sup_{|y|\leq a_n}\left|\sqrt{\frac{n e^{y^2/(2\sigma^2n)}}{2\pi\sigma^2n}}e^{\theta y}e^{n(\theta\psi'(\theta)-\psi(\theta))}P(M_n\geq n\psi'(\theta)+y)-C(\theta)\sqrt{\frac{2\pi\sigma^2}{\theta}}\right| = 0,
\]
where \(C(\theta)\in(0, 1)\) is a well - defined constant and \((a_n)_{n\geq0}\) satisfies \(a_n = O(\sqrt{n})\).
3. **Joint limit distribution of extreme processes**:
- The author also studies the joint limit distribution of the extreme processes \(E_n\) and \(M_n - n\psi'(\theta)\) under the condition \(M_n\geq n\psi'(\theta)\). Specifically, there exists a point process \(D(\theta)\) such that
\[
(E_n, M_n - n\psi'(\theta))\text{ converges under the condition }\{M_n\geq n\psi'(\theta)\}\text{ to }(D(\theta), e),
\]
where \(e\) is an exponential random variable with parameter \(\theta\), independent of \(D(\theta)\).
### Methods and techniques
- **Ridge decomposition theorem**: The author uses the ridge decomposition theorem to deal with the extreme values of branching random walks.