Variance Bounding of Delayed-Acceptance Kernels

A delayed-acceptance version of a Metropolis–Hastings algorithm can be useful for Bayesian inference when it is computationally expensive to calculate the true posterior, but a computationally cheap approximation is available; the delayed-acceptance kernel targets the same posterior as its associated “parent” Metropolis-Hastings kernel. Although the asymptotic variance of the ergodic average of any functional of the delayed-acceptance chain cannot be less than that obtained using its parent, the average computational time per iteration can be much smaller and so for a given computational budget the delayed-acceptance kernel can be more efficient. When the asymptotic variance of the ergodic averages of all $$L^2$$ L 2 functionals of the chain are finite, the kernel is said to be variance bounding. It has recently been noted that a delayed-acceptance kernel need not be variance bounding even when its parent is. We provide sufficient conditions for inheritance: for non-local algorithms, such as the independence sampler, the discrepancy between the log density of the approximation and that of the truth should be bounded; for local algorithms, two alternative sets of conditions are provided. As a by-product of our initial, general result we also supply sufficient conditions on any pair of proposals such that, for any shared target distribution, if a Metropolis-Hastings kernel using one of the proposals is variance bounding then so is the Metropolis-Hastings kernel using the other proposal.

Forward-reflected-backward method with variance reduction

We propose a variance reduced algorithm for solving monotone variational inequalities. Without assuming strong monotonicity, cocoercivity, or boundedness of the domain, we prove almost sure convergence of the iterates generated by the algorithm to a solution. In the monotone case, the ergodic average converges with the optimal O(1/k) rate of convergence. When strong monotonicity is assumed, the algorithm converges linearly, without requiring the knowledge of strong monotonicity constant. We finalize with extensions and applications of our results to monotone inclusions, a class of non-monotone variational inequalities and Bregman projections.

Weak ergodic averages over dilated measures

Let $m\in \mathbb{N}$ and $\mathbf{X}=(X,{\mathcal{X}},\unicode[STIX]{x1D707},(T_{\unicode[STIX]{x1D6FC}})_{\unicode[STIX]{x1D6FC}\in \mathbb{R}^{m}})$ be a measure-preserving system with an $\mathbb{R}^{m}$-action. We say that a Borel measure $\unicode[STIX]{x1D708}$ on $\mathbb{R}^{m}$ is weakly equidistributed for $\mathbf{X}$ if there exists $A\subseteq \mathbb{R}$ of density 1 such that, for all $f\in L^{\infty }(\unicode[STIX]{x1D707})$, we have $$\begin{eqnarray}\lim _{t\in A,t\rightarrow \infty }\int _{\mathbb{R}^{m}}f(T_{t\unicode[STIX]{x1D6FC}}x)\,d\unicode[STIX]{x1D708}(\unicode[STIX]{x1D6FC})=\int _{X}f\,d\unicode[STIX]{x1D707}\end{eqnarray}$$ for $\unicode[STIX]{x1D707}$-almost every $x\in X$. Let $W(\mathbf{X})$ denote the collection of all $\unicode[STIX]{x1D6FC}\in \mathbb{R}^{m}$ such that the $\mathbb{R}$-action $(T_{t\unicode[STIX]{x1D6FC}})_{t\in \mathbb{R}}$ is not ergodic. Under the assumption of the pointwise convergence of the double Birkhoff ergodic average, we show that a Borel measure $\unicode[STIX]{x1D708}$ on $\mathbb{R}^{m}$ is weakly equidistributed for an ergodic system $\mathbf{X}$ if and only if $\unicode[STIX]{x1D708}(W(\mathbf{X})+\unicode[STIX]{x1D6FD})=0$ for every $\unicode[STIX]{x1D6FD}\in \mathbb{R}^{m}$. Under the same assumption, we also show that $\unicode[STIX]{x1D708}$ is weakly equidistributed for all ergodic measure-preserving systems with $\mathbb{R}^{m}$-actions if and only if $\unicode[STIX]{x1D708}(\ell )=0$ for all hyperplanes $\ell$ of $\mathbb{R}^{m}$. Unlike many equidistribution results in literature whose proofs use methods from harmonic analysis, our results adopt a purely ergodic-theoretic approach.

Distributional chaos in multifractal analysis, recurrence and transitivity

There is much research on the dynamical complexity on irregular sets and level sets of ergodic average from the perspective of density in base space, the Hausdorff dimension, Lebesgue positive measure, positive or full topological entropy (and topological pressure), etc. However, this is not the case from the viewpoint of chaos. There are many results on the relationship of positive topological entropy and various chaos. However, positive topological entropy does not imply a strong version of chaos, called DC1. Therefore, it is non-trivial to study DC1 on irregular sets and level sets. In this paper, we will show that, for dynamical systems with specification properties, there exist uncountable DC1-scrambled subsets in irregular sets and level sets. Meanwhile, we prove that several recurrent level sets of points with different recurrent frequency have uncountable DC1-scrambled subsets. The major argument in proving the above results is that there exists uncountable DC1-scrambled subsets in saturated sets.

Nilsequences and multiple correlations along subsequences

The results of Bergelson, Host and Kra, and Leibman state that a multiple polynomial correlation sequence can be decomposed into a sum of a nilsequence (a sequence defined by evaluating a continuous function along an orbit in a nilsystem) and a null sequence (a sequence that goes to zero in density). We refine their results by proving that the null sequence goes to zero in density along polynomials evaluated at primes and along the Hardy sequence $(\lfloor n^{c}\rfloor )$. In contrast, given a rigid sequence, we construct an example of a correlation whose null sequence does not go to zero in density along that rigid sequence. As a corollary of a lemma in the proof, the formula for the pointwise ergodic average along polynomials of primes in a nilsystem is also obtained.

Pseudo-marginal Metropolis–Hastings sampling using averages of unbiased estimators

Summary We consider a pseudo-marginal Metropolis–Hastings kernel ${\mathbb{P}}_m$ that is constructed using an average of $m$ exchangeable random variables, and an analogous kernel ${\mathbb{P}}_s$ that averages $s<m$ of these same random variables. Using an embedding technique to facilitate comparisons, we provide a lower bound for the asymptotic variance of any ergodic average associated with ${\mathbb{P}}_m$ in terms of the asymptotic variance of the corresponding ergodic average associated with ${\mathbb{P}}_s$. We show that the bound is tight and disprove a conjecture that when the random variables to be averaged are independent, the asymptotic variance under ${\mathbb{P}}_m$ is never less than $s/m$ times the variance under ${\mathbb{P}}_s$. The conjecture does, however, hold for continuous-time Markov chains. These results imply that if the computational cost of the algorithm is proportional to $m$, it is often better to set $m=1$. We provide intuition as to why these findings differ so markedly from recent results for pseudo-marginal kernels employing particle filter approximations. Our results are exemplified through two simulation studies; in the first the computational cost is effectively proportional to $m$ and in the second there is a considerable start-up cost at each iteration.

Nonconventional ergodic theorems for quantum dynamical systems

We study the so-called Nonconventional Ergodic Theorem for noncommutative generic measures introduced by Furstenberg in classical ergodic theory, and the relative three-point multiple correlations of arbitrary length arising from several situations of interest in quantum case. We deal with the diagonal state canonically associated to the product state (i.e. quantum "diagonal measures") in the ergodic situation, and with the case concerning convex combinations (i.e. direct integral) of diagonal measures in nonergodic one. We also treat in the full generality the case of compact dynamical systems, that is when the unitary generating the dynamics in the Gelfand–Naimark–Segal representation is almost periodic. In all the above-mentioned situations, we provide the explicit formula for the involved ergodic average. Such an explicit knowledge of the limit of the three-point correlations is naturally relevant for the investigation of the long time behavior of a dynamical system.

Oscillation and the mean ergodic theorem for uniformly convex Banach spaces

Let $ \mathbb{B} $ be a $p$-uniformly convex Banach space, with $p\geq 2$. Let $T$ be a linear operator on $ \mathbb{B} $, and let ${A}_{n} x$ denote the ergodic average $(1/ n){\mathop{\sum }\nolimits}_{i\lt n} {T}^{n} x$. We prove the following variational inequality in the case where $T$ is power bounded from above and below: for any increasing sequence $\mathop{({t}_{k} )}\nolimits_{k\in \mathbb{N} } $ of natural numbers we have ${\mathop{\sum }\nolimits}_{k} \mathop{\Vert {A}_{{t}_{k+ 1} } x- {A}_{{t}_{k} } x\Vert }\nolimits ^{p} \leq C\mathop{\Vert x\Vert }\nolimits ^{p} $, where the constant $C$ depends only on $p$ and the modulus of uniform convexity. For $T$ a non-expansive operator, we obtain a weaker bound on the number of $\varepsilon $-fluctuations in the sequence. We clarify the relationship between bounds on the number of $\varepsilon $-fluctuations in a sequence and bounds on the rate of metastability, and provide lower bounds on the rate of metastability that show that our main result is sharp.

Ergodic Channel Capacity of MIMO Transmission in Spatial Correlated-Fading

In this article the MGF (moment generating function) obtained to determine the ergodic (average) channel capacity, and which is by passing the difficulty in calculating the pdf (probability density function) of SNR (signal-to-noise ratio) traditionally. Some numerical results are offered for validating the accuracy the theoretical deriver formulas. Furthermore, many plots work out from combination with different number of transmitter and receiver for comparison. It is valuable to note that the more the antenna numbers the larger channel capacity is not acceptable when the correlation coefficient is taken into account the system evaluation of a MIMO system.