Release Note of XMVB v3.1

Published Time:  2023-03-23 15:22:12

We are pleased to announce that XMVB@XACS version 3.1 is available now. Two major improvements are implemented in the update:

1. Hybrid MPI/OpenMP parallelization

2. Symmetry control during SCF

1. New Features

1.1. Hybrid MPI/OpenMP Parallelization

MPI and OpenMP are both supported in XMVB@XACS v3.1 to provide more powerful capability. The hybrid MPI/OpenMP parallelization strategy enables users to proceed ab initio valence bond computations with multiple nodes. Currently MPI/OpenMP parallelization is implemented for VBSCF, BOVB and VBCI.

Tables 1 and 2 show the efficiency of hybrid parallel calculations on cyclobutadiene and anthracene respectively. In some cases, the speedup is close to linear. In general, the speedup of each SCF iteration is half of the number of cores in parallel.

 

Table 1. VBSCF computation of C4H4 with cc-pVDZ basis set, where only 2 covalent structures are involved.

Cores

SCF Iteration

NMPI

NOMP

NPROC

Time (s)

Speedup

1

1

1

142.2

1.0

1

24

24

8.2

17.3

24

2

48

9.4

15.1

12

4

48

9.2

15.5

8

6

48

7.4

19.2

6

8

48

7.2

19.8

4

12

48

6.4

22.2

2

24

48

3.3

43.1

3

24

72

2.9

49.0

4

24

96

3.3

43.1

5

24

120

2.4

59.3

 

Table 2. VBSCF computation of C14H10 with cc-pVDZ basis set, where 2,760,615 structures are involved.

Cores

SCF Iteration

NMPI

NOMP

NPROC

Time(s)

Speedup

1

12

12

35196.3

1.0

1

24

24

19224.2

1.8

2

24

48

11658.2

3.0

4

12

48

11613.5

3.0

8

6

48

14038.8

2.5

3

24

72

10768.3

3.3

4

24

96

8955.3

3.9

8

12

96

9977.4

3.5

24

4

96

22746.0

1.5

48

2

96

38994.5

0.9

 

 

Figure 1. Geometry of C14H10.

1.2. Symmetry Control during SCF

Symmetry control is available in VBSCF and BOVB in XMVB@XACS v3.1. Currently only Abel groups are implemented. Tables 3 and 4 show the performance of symmetry control. Symmetry control with C2 decreases time consumptions of each SCF iteration half of those without symmetry, especially for the serial computations. For the number of iterations, computations with symmetry control decrease about 13% compared with those without symmetry. So, symmetry control decreases both time consumptions of each SCF iteration and number of iterations.

 

Table 3. Time consumption (in second) of each SCF iteration with different symmetries.

Cores

Symmetry

NMPI

NOMP

NPROC

C1

C2

D2

C4H4 with VBSCF/cc-pVDZ

1

1

1

15.4

8.4

5.7

1

24

24

1.6

1.3

1.2

2

24

48

1.6

1.2

1.0

C4H6 with BOVB/cc-pVDZ

1

1

1

38.0

20.7

--

1

24

24

2.0

2.0

--

2

24

48

2.0

1.7

--

 

Table 4. Number of SCF iterations with different symmetries.

 

Symmetry

 

C1

C2

D2

C4H4

61

53

53

2. How to Get

XMVB@XACS v3.1 is now available in the cloud computing platform of XACS. For those who are interested in our program, you may go to the XACS official site https://xacs.xmu.edu.cn to try the program on the cloud computing platform or download an copy and install the program on your own machine.