关于ORCA的DLPNO-CCSD(T)方法占用巨量磁盘和存储不足报错

DoorBell

老师们好！最近在算一个钴的团簇，已经用DFT完成了粗优化，之后又用ORCA 4.1.1的DLPNO-CCSD(T)计算单点能。

但是无论在PC上还是超算上都会在这一步产生几百G的.tmp文件：

1. ...
   
2. --------------------------------
   
3. INITIAL GUESS AND PNO GENERATION
   
4. --------------------------------
   
5. 
   
6. PNO truncation parameters                   ....
   
7.     PAOOverlapThresh   =        1.000e-08
   
8. 
   
9.     TCutPairs          =        1.000e-04
   
10.     TCutPNO            =        3.330e-07
    
11.     TCutPNOSingles     =        9.990e-09
    
12.     TCutMP2Pairs       =        1.000e-05
    
13.     TCutMKN            =        1.000e-03
    
14.     TCutDO             =        1.000e-02
    
15. 
    
16. Pair selection                          .... not used
    
17. Type of local MP2 treatment                 .... semi-local MP2
    
18. Strategy for PNO selection                  .... occupation number selection
    
19. Pair density normalization                  .... MP2 norm
    
20. Spin component scaling                      .... not used
    
21. 
    
22.    .... Finished loop over pairs
    
23. ...

复制代码

之后进行到此处就会报错OUT OF MEMORY!：

1. ...
   
2.    .... Finished loop over pairs
   
3. 
   
4. PNO Occupation Number Statistics:
   
5.   | Av. % of trace(Dij) retained        ...  95.447125430886
   
6.   | sigma^2 in % of trace(Dij) retained ...   2.27e+01
   
7.   | Av. % of trace(Di) retained         ...  99.997798967622
   
8.   | sigma^2 in % of trace(Di) retained  ...   2.96e-06
   
9. Distributions of % trace(Dij) recovered:
   
10.   |       >= 99.9  ...   293 (  9.3 % of all pairs)
    
11.   |   [60.0, 70.0) ...     1 (  0.0 % of all pairs)
    
12.   |   [70.0, 80.0) ...    42 (  1.3 % of all pairs)
    
13.   |   [80.0, 90.0) ...   304 (  9.7 % of all pairs)
    
14.   |   [90.0, 99.0) ...  1905 ( 60.6 % of all pairs)
    
15.   |   [99.0, 99.9) ...   600 ( 19.1 % of all pairs)
    
16. Distributions of % trace(Di) recovered :
    
17.   |       >= 99.9  ...    84 (100.0 % of all I-pairs )
    
18. 
    
19. Making pair pair interaction lists          ... done
    
20. 
    
21.                        ===========================
    
22.                        1740 OF 3145 PAIRS ARE KEPT
    
23.                        ===========================
    
24. 
    
25. Total time spent in the initial guess                 ...  8761.445 sec
    
26. SL-MP2 correlation energy (all non-screened pairs)    ...      -7.092094379993 Eh
    
27. 
    
28. Initial PNO correlation energy                        ...      -7.003322263097 Eh
    
29. sum of pair energies estimated for screened out pairs ...      -0.002482968226 Eh
    
30. sum of MP2 pair energies for pairs that were not kept ...      -0.054621149365 Eh
    
31. sum of PNO error estimates for the kept pairs         ...      -0.034236781988 Eh
    
32.                                                           --------------------
    
33. sum of all corrections                                         -0.091340899579
    
34. Initial total correlation energy                               -7.094663162676
    
35. Thresholds for map construction and integral transformation for fine MP2 and CCSD(T) calculation:
    
36.    TCutMKN   ...     1.0e-03
    
37.    TCutDO    ...     1.0e-02
    
38.    TCutPairs ...     1.0e-04
    
39.    TCutCMO   ...     1.0e-03
    
40.    TCutCPAO  ...     1.0e-03
    
41. 
    
42. Time for aux screen maps:            0.098
    
43. Time for maps after fine MP2:            0.360
    
44. -----------------------------
    
45. LOCAL RI TRANSFORMATION (IJV)
    
46. -----------------------------
    
47. 
    
48. Orbital window:      40 to 123
    
49. Basis functions:    920 (360 shells)
    
50. Aux. functions:    5940 (1660 shells)
    
51. 
    
52. Processing maps (0.0 sec)
    
53. Average map sizes:
    
54.    Aux shells -> MOs(i)           82.1
    
55.    Aux shells -> MOs(j)           84.0
    
56.    MOs        -> AO shells       349.5
    
57. 
    
58. Calculating integrals (20.8 sec, 316.439 MB)
    
59. Sorting integrals (1.0 sec, 316.372 MB)
    
60. Total time for the integral transformation: 21.9 sec
    
61. --------------------------------
    
62. LOCAL RI TRANSFORMATION (VABPAO)
    
63. --------------------------------
    
64. 
    
65. Number of PAOs:     920
    
66. Basis functions:    920 (360 shells)
    
67. Aux. functions:    5940 (1660 shells)
    
68. 
    
69. Processing maps (0.0 sec)
    
70. Average map sizes:
    
71.    Aux shells -> PAOs            920.0
    
72.    PAOs       -> AO shells       360.0
    
73. 
    
74. Calculating integrals (1139.8 sec, 69876.943 MB)
    
75. Finished
    
76. 
    
77. 
    
78. -------------------------------------
    
79. Pair Pair Term precalculation with
    
80. RI-(ij|mn) and (im|jn) transformation
    
81. ON THE FLY
    
82. -------------------------------------
    
83. 
    
84.    IBatch   1 (of   1)              ...
    
85. [file orca_tools/qcmem.cpp, line 918, Process 3]:  OUT OF MEMORY ERROR!
    
86. 
    
87. [file orca_tools/qcmem.cpp, line 918, Process 3]:  OUT OF MEMORY ERROR!
    
88. 
    
89. [file orca_tools/qcmem.cpp, line 918, Process 1]:  OUT OF M
    
90. 
    
91. ...乱码...
    
92. 
    
93. [file orca_tools/qcmem.cpp, line 918, Process 1]:  OUT OF MEMORY ERROR!
    
94. 
    
95. cmem.cpp, line 918, Process 2]:  OUT OF MEMORY ERROR!
    
96. 
    
97. [file orca_tools/qcmem.cpp, line 918, Process 2]:  OUT OF MEMORY ERROR!
    
98. 
    
99. [file orca_tools/q
    
100. 
     
101. ...乱码...
     
102. 
     
103. [file orca_tools/qcmem.cpp, line 918, Process 5]:  OUT OF MEMORY ERROR!
     
104. 
     
105. ]:  OUT OF MEMORY ERROR!
     
106. 
     
107. [file orca_tools/qcmem.cpp, line 918, Process 4]:  OUT OF MEMORY ERROR!
     
108. 
     
109. [file orca_tools/qcmem.cpp, line 918, Process 0]:  OUT OF MEMORY ERROR!
     
110. 
     
111. [file orca_tools/qcmem.cpp, line 918, Process 0]:  OUT OF MEMORY ERROR!
     
112. 
     
113. 
     
114. job aborted:
     
115. [ranks] message
     
116. 
     
117. [0-5] process exited without calling finalize
     
118. 
     
119. ---- error analysis -----
     
120. 
     
121. [0-5] on DESKTOP-KEETL36
     
122. D:\ProgramFiles\ORCA\orca_mdci_mpi.exe ended prematurely and may have crashed. exit code 12
     
123. 
     
124. ---- error analysis -----
     
125. 
     
126. ORCA finished by error termination in MDCI
     
127. Calling Command: mpiexec -np 6  D:\ProgramFiles\ORCA\orca_mdci_mpi.exe F:\ORCA\test.mdciinp.tmp F:\ORCA\test
     
128. [file orca_tools/qcmsg.cpp, line 458]:
     
129.   .... aborting the run
     
130. 
     
131. [file orca_tools/qcmsg.cpp, line 458]:
     
132.   .... aborting the run
     

复制代码

输入文件：

1. ! RKS SlowConv NODIIS Grid6 DLPNO-CCSD(T) RI def2-TZVP AutoAux
   
2. 
   
3. %scf
   
4.     maxIter 1500
   
5.     MaxIntMem 40960
   
6.     MaxDisk 40960
   
7. end
   
8. 
   
9. %maxCore 1024
   
10. 
    
11. %pal
    
12.     nProcs 40
    
13. end
    
14. 
    
15. %output
    
16.     XYZFile true
    
17. end
    
18. 
    
19. * xyz 0 1
    
20. ...
    
21. *

复制代码

原子簇如图所示。

由于不知道哪些信息是有用的，所以就多复制了一些文字，还请各位老师原谅。

非常感谢！

DoorBell

产生的.tmp文件是这样的：

plus

%maxCore 10000
并行核数*10G<总内存

量化小菜鸡

这种计算特别费内存，10个G也未必够。你试着增大%MaxCore减小nprocs看看。

DoorBell

量化小菜鸡 发表于 2019-3-28 01:19
这种计算特别费内存，10个G也未必够。你试着增大%MaxCore减小nprocs看看。

已经降低了并行核数并提高了内存，但是仍然产生了几百G的文件...一共就只有500G的存储......