1: \begin{abstract}

2: 		This paper proposes an efficient general alternating-direction implicit

3: 		(GADI) framework for solving large sparse linear systems. The convergence

4: 		property of the GADI framework is discussed. Most of existing ADI methods can

5: 		be unified in the developed framework. Meanwhile the GADI

6: 		framework can derive new ADI methods. Moreover, as the algorithm efficiency

7: 		is sensitive to the splitting parameters, we offer a data-driven approach,

8: 		the Gaussian process regression (GPR) method based on the Bayesian inference, to

9: 		predict the GADI framework's relatively optimal parameters. The GPR method

10: 		only requires a small training data set to learn the regression prediction

11: 		mapping, can predict accurate splitting parameters, and has high

12: 		generalization capability. It allows us to efficiently solve linear systems

13: 		with a one-shot computation, and does not require any repeated computations.

14: 		Finally, we use the three-dimensional convection-diffusion equation,

15: 		two-dimensional parabolic equation, and

16: 		continuous Sylvester matrix equation to examine the performance of our proposed methods.

17: 		Numerical results demonstrate that the proposed framework

18: 		is faster tens to thousands of times than the existing ADI methods, such as

19: 		(inexact) Hermitian and skew-Hermitian splitting type methods in which the

20: 		consumption of obtaining relatively optimal splitting parameters is ignored.

21: 		As a result, our proposed methods can solve much larger linear systems which these existing ADI methods have not reached.

22: 		%	\keywords {}

23: 		% \PACS{PACS code1 \and PACS code2 \and more}

24: 		% \subclass{MSC code1 \and MSC code2 \and more}

25: 	\end{abstract}

26: