1: \begin{abstract}

2:  In the general framework of Bayesian inference, the target distribution can only be evaluated up-to a constant of proportionality.

3:  Classical consistent Bayesian methods such as sequential Monte Carlo (SMC) and Markov chain Monte Carlo (MCMC) 

4: have unbounded time complexity requirements.

5: We develop a fully parallel sequential Monte Carlo (pSMC) method

6: % , the parallel sequential Monte Carlo (pSMC) method, 

7: which provably delivers {\em parallel strong scaling}, i.e. the time complexity (and per-node memory) remains bounded

8: if the number of asynchronous processes is allowed to grow.

9: % in the general framework of the Bayesian inference problem where often the target distribution is only known up to a constant of proportionality. 

10: % When applied to large data sets, pSMC bounds the rapid growth of 

11: % The time complexity in single sequential Monte Carlo (SMC) and Markov chain Monte Carlo (MCMC) and maintains the efficiency of single SMC in high-dimensional sample space.  

12: % Comparably, 

13: More precisely, the pSMC has a theoretical convergence rate of MSE$ = O(1/NR)$\footnote{MSE is the mean square error, and $1/NR\equiv

14: % \frac1{NR}

15: (NR)^{-1}$.}, 

16: where $N$ denotes the number of communicating samples in each processor and $R$ denotes the number of processors.

17: In particular, for suitably-large problem-dependent $N$, as $R \rightarrow \infty$ the method converges to infinitesimal accuracy 

18: MSE$=O(\varepsilon^2)$ with a fixed finite time-complexity Cost$=O(1)$ and with no efficiency leakage, i.e. computational complexity Cost$=O(\varepsilon^{-2})$.

19: % , is guaranteed by the strong scaling property. 

20: A number of Bayesian inference problems are taken into consideration to compare the pSMC and MCMC methods. 

21: %\\

22: %\noindent \textbf{Keywords}: Sequential Monte Carlo, Parallel Computing, Bayesian Inference Problems, Markov Chain Monte Carlo 

23: \end{abstract}

24: