Towards a real-time fully-coherent all-sky search for gravitational waves from compact binary coalescences using particle swarm optimization
While a fully-coherent all-sky search is known to be optimal for detecting gravitational wave signals from compact binary coalescences, its high computational cost has limited current searches to less sensitive coincidence-based schemes. Following up on previous work that has demonstrated the effectiveness of Particle Swarm Optimization in reducing the computational cost of this search, we present an implementation that achieves near real-time computational speed. This is achieved by combining the search efficiency of PSO with a significantly revised and optimized numerical implementation of the underlying mathematical formalism along with additional multi-threaded parallelization layers in a distributed computing framework. For a network of four second-generation detectors with $60$~min data from each, the runtime of the implementation presented here ranges between $\approx 1.4$ to $\approx 0.5$ times the data duration for network signal-to-noise ratios (SNRs) of $\gtrsim 10$ and $\gtrsim 12$, respectively. The reduced runtimes are obtained with small to negligible losses in detection sensitivity: for a false alarm rate of $\simeq 1$~event per year in Gaussian stationary noise, the loss in detection probability is $\leq 5\%$ and $\leq 2\%$ for SNRs of $10$ and $12$, respectively. Using the fast implementation, we are able to quantify frequentist errors in parameter estimation for signals in the double neutron star mass range using a large number of simulated data realizations. A clear dependence of parameter estimation errors and detection sensitivity on the condition number of the network antenna pattern matrix is revealed. Combined with previous work, this paper securely establishes the effectiveness of PSO-based fully-coherent all-sky search across the entire binary inspiral mass range that is relevant to ground-based detectors.