v 1.

05

Using FireFly in education and research @ home

A short introduction in Computational Chemistry & an overview of strength possibilities of PC-Gamess/FireFly and how to make calculations more efficient

Part II Benchmark Basis/Correlation Correction

This benchmark on basis set was done because calculation time strongly depends on basis set. There are a lot of different sets which describes chemical behaviour, some with good approximation to real functions and some with acceptable description. Smaller systems can be handled with large basis sets, but if we want to assay a molecule with more typical size we have to make some arrangements to handle such calculations. To see which smaller basis sets gave comparable descriptions to bigger ones in less computational effort there were done some benchmarks on typical molecules. We will also see when it is necessary to use bigger sets. This may help to get a feeling how calculation time rise with large basis sets. M. Checinski

Benchmarks of Basis/Correlation Correction

In this Chapter are the benchmarks collected, which i made to decide which basis sets and Correlation Corrections are useful (quality & time consumption) for typical questions in laboratory or which one should be used for educational demands.

For making a general statement about a good Basisset/Correlation Correction for smaller computer(-cluster) it seems to be useful to compare different chemical environments. We will study typical organic and inorganic molecules, to find out which basis sets are not advisable for some structures.

Who just want to see the result should jump to the end of this chapter. There is a kind of summary.

Benchmarks of Basis/Correlation Correction

At first we need an imagination of influence of basis sets on computation time. Because there are so many factors which have an influence on computational time it is impossible to say i.e. one set need two and a half times more computation time than an other. But to see how the tendency is i made a comparison of n-alkanes, to see how influence of an additional CH2-Group is.

Benchmarks of Basis/Correlation Correction

Dependence of basis set on computation time of alkanes
1000,0

800,0

600,0

MINI 3N21 6N31 6N31_dp 6N311 6N311_dp 6N311_2d2p TZV_2d2p cct

400,0

200,0

0,0 c2 c3 c4 c5 c6 c7 c8

Here we see that different sets have different slopes. There are hough differences in computational time of a C8-Alkane computed with a MINI and a cc-pVTZ set (here 1:100).

Benchmarks of Basis/Correlation Correction

Dependence of basis set on computation time of alkanes
350,0

250,0

150,0

MINI 3N21 6N31 6N31_dp 6N311 6N311_dp 6N311_2d2p TZV_2d2p cct

50,0

-50,0 c2 c3 c4 c5 c6 c7 c8

Here we see that a pd-polarized split-valence set needs more computation time than an unpolarized triple-valence set.

Benchmarks of Basis/Correlation Correction To compare the basis sets qualitatively we need some properties, which we can compare. As we have seen the absolute value of total Energy is not the only important information, difference of total Energy by stretching a bond could give a good hint about the quality.

Another property can be the dipole moment, which depends on bond partners and bond length. But we can only compare dipole moments with real ones if the molecule were measured in gas phase.

PC-Gamess gives us thermodynamic properties, too. But here we should, compare comparable (gas phase) molecules, too.

We will take a focus on 1D-Potential Energy Surfaces and dipole moment.

Benchmarks of Basis/Correlation Correction At first we will discuss the behaviour of simplest alkane.

In previous chapters we have discussed the differences of RHF/UHF and HF in general. We have discussed about the cheap correlation correction of Moller-Plesset and the popular hybrid calculation of DFT (especial Becke3-LeeYoungParr). Now we will try to compare them qualitatively.

For that we study energy changing by C-H Bond stretching. We compare how a HF, HF/MP2 and B3LYP influence the description of this system. As basis sets we use the small split valence set 3-21 and the hough triple valence set aug-cc-pTVZ with additional diffuse and polarized functions.

After that we will compare how the popular sets describe bond-stretching of many different molecules at UHF/B3LYP level.

Benchmarks of Basis/Correlation Correction

Ok, let's start with the 3-21 set.
Methan: stretch of C-H Bond
0,25

0,20

energydifference to lowest energy

3N21-RHF 3N21-UHF 3N21-RHF-MP2 3N21-RHF-B3L 3N21-UHF-B3Lx 3N21-UHF-B3L

0,15

0,10

0,05

0,00 -0,30 -0,10 0,10 0,30 0,50 0,70 0,90 1,10 1,30

distance to equibrillium length

What we can see is that at HF level the description is significant different. On the other hand all calculations say that the equilibrium bond distance (at 0.05 A stepping) have relative the lowest energy. We will later see that this have not be ususal, but this system is easy to describe. And we shall not forget that all basis sets were fitted on such general Molecules to give good functions.

Benchmarks of Basis/Correlation Correction

To see better how the different calculations differs we should look at a smaller part of this diagram.
Methan: stretch of C-H Bond

0,105

energydifference to lowest energy

3N21-RHF 3N21-UHF 3N21-RHF-MP2 3N21-RHF-B3L 3N21-UHF-B3Lx 3N21-UHF-B3L

0,095

0,085

0,075 0,550 0,600 0,650

distance to equibrillium length

Here we see that we have three groups ( HF, HF/MP2 and DFT ). For this system the MP2 correction is comparable to the computational heavier B3LYP calculation.

Benchmarks of Basis/Correlation Correction

If we use the hough aug-cc-pTZV set we see comparable descriptions of such bond-stretching.
Methan: stretch of C-H Bond

0,105

energydifference to lowest energy

aug-cc-pvtz-RHF aug-cc-pvtz-UHF aug-cc-pvtz-RHF-MP2 aug-cc-pvtz-RHF-B3L aug-cc-pvtz-UHF-B3L aug-cc-pvtz-UHF-B3Lx

0,095

0,085

0,075 0,550 0,600 0,650

distance to equibrillium length

But now the MP2 correction is not such good like in previous topic.

Benchmarks of Basis/Correlation Correction

Ok, lets put these descriptions together
Methan: stretch of C-H Bond

0,185

energydifference to lowest energy

3N21-RHF 3N21-RHF-MP2 3N21-UHF-B3L aug-cc-pvtz-RHF aug-cc-pvtz-RHF-MP2 aug-cc-pvtz-UHF-B3Lx

0,165

0,145 1,000 1,100 1,200

distance to equibrillium length

Here we see that the ACCT set is much better than 3-21, but this is nothing unexpected =) if we know that the ACCT set has for each Orbital 3 possible Orbitals which can be mixed and additional polarization and diffuse functions, to make the resulting Orbital more perfect for this chemical environment. Don't forget that this is a relative description, absolute values for ACCD are much lower than for 3-21.

Benchmarks of Basis/Correlation Correction

At this point we should look at the computation effort of these calculations. For better comparison we use a here a quite good information which we get from PC-Gamess on each calculation. The CPU utilization should be around 100% (or like here on a dual-core 200%), higher values means 100% :) But if we have lower values, that means HDD operations which are very very slow. In other words CPU have to wait for data to make next Operation. We will later see how dramatically this could be and how PC-Gamess settings can help us to avoid such problems. Ok, lets compare the WALL CLOCK times.
3N21-RHF 0,3 228,60% aug-cc-pvtz-RHF 349,9 188,05% 3N21-UHF 0,4 225,02% aug-cc-pvtz-UHF 412,3 186,20% 3N21-RHF-MP2 0,5 214,35% aug-cc-pvtz-RHF-MP2 823,5 181,34% 3N21-RHF-B3L 15,1 200,45% 3N21-UHF-B3Lx 10,2 200,24 3N21-UHF-B3L 29,0 200,25% aug-cc-pvtz-UHF-B3L 706,1 197,66%

Zeit [sec] Util 2 CPU

Zeit [sec] Util 2 CPU

aug-cc-pvtz-RHF-B3L aug-cc-pvtz-UHF-B3Lx 496,6 192,95%

Here we see a big difference between calculation with ACCT and 3-21, on UHF/B3LYPx level where the utilizations are most comparable we see a computational difference of 70:1 ! Ok, 12 minutes are not so long, but this is just a very small Molecule. For usual molecules it is a hough difference. At this point i have to say what B3Lx means. As mentioned PC-Gamess gives us a great flexibility in controlling calculations, some less accurate settings seems to make descriptions worser but other have low effects in qualitative descriptions but big in computational effort. I tested some settings with different molecules to check how to safe computational time with small loss of accuracy. In another chapter i will summarize these settings. The difference of B3Lx and B3L is just a smaller value of NRAD in $DFT part. This is by the way one cause why i made such benchmarks, to check which sets & settings gives best agreement in accuracy and time consumption.

Benchmarks of Basis/Correlation Correction

Methan: stretch of C-H Bond

-38,90

energydifference to lowest energy

STO2G MINI 6N31-2pd 6N311-3p2d cc-pVDZ aug-cc-pVTZ

-39,40

-39,90

-40,40

-0,30

-0,10

0,10

0,30

0,50

0,70

0,90

1,10

1,30

distance to equibrillium length

To see how the absolute differences are here are some impressions of methan. There are hough differences between STO2/MINI and a mulit valence set. Differences between bigger sets are in another scale.

Benchmarks of Basis/Correlation Correction

Methan: stretch of C-H Bond

-40,30

6N31-2pd 6N311-3p2d cc-pVDZ aug-cc-pVTZ

energydifference to lowest energy
-40,50

-0,30

-0,10

0,10

0,30

0,50

0,70

0,90

1,10

1,30

distance to equibrillium length

Here we see that split-valence- and triple-valence-sets are compareable.

Benchmarks of Basis/Correlation Correction

Methan: stretch of C-H Bond

-39,80

energydifference to lowest energy

3N21-RHF 3N21-UHF 3N21-RHF-MP2 3N21-RHF-B3L 3N21-UHF-B3Lx 3N21-UHF-B3L

-40,30

-0,30

-0,10

0,10

0,30

0,50

0,70

0,90

1,10

1,30

distance to equibrillium length

Here we see difference between HF / HF-MP2 / HF-B3LYP . Basis is 3-21. Here there are no differences between RHF & UHF, RHF/B3L & UHF/B3L , and there are low differences between RHF/B3L UHF/B3L & UHF/B3Lx

Benchmarks of Basis/Correlation Correction

Ok, with this experience we see that we should use UHF/B3Lx calculations. Now we can start to assay other chemical environments. Let's check how different basis sets describe C-C Bond breaking/creating on Ethan, which is a daily topic in Research.

C-C stretch of Ethan

0,250

0,200
energy relative to groundstate E

0,150

0,100

0,050

0,000 -0,10

0,00

0,10

0,20

0,30

0,40

0,50

0,60

0,70

0,80

0,90

1,00

1,10

1,20

1,30

distance to equilibrium lenght [A]

Benchmarks of Basis/Correlation Correction

To put focus only the Basis sets, we use a PM3 optimized geometry for this calculation
Ethan: variation of C-C Bondlength
0,26

0,16

STO-2 MINI MIDI 3N21 DZV 6N31-2pd TZV 6N311-3p2d ACCD

0,06

-0,04 -0,30 -0,10 0,10 0,30 0,50 0,70 0,90 1,10 1,30

At first we see, that the STO-2G & MINI set is significant different from the other. Ethan C-C-Bond is a very simple system, and if these sets have such quality problems, we should use them only for didactical usage or fast geometry preoptimization.

Benchmarks of Basis/Correlation Correction

To compare the other we should look at some smaller parts of the diagram.
Ethan: variation of C-C Bondlength
0,008

STO-2 MINI MIDI 3N21 DZV-pd 6N31-2pd TZV-2pd 6N311-3p2d ACCD ACCT
0,003

-0,002 0,05 0,10 0,15

For small stretch lengths we see that sets with no polarization functions seem not to be very accurate. Another informations is that from STO-2 to DZV the relative energy minimum is 0.05 A from equilibrium geometry of PM3 optimization.

Benchmarks of Basis/Correlation Correction

Ethan: variation of C-C Bondlength
0,067

0,062

STO-2 MINI MIDI 3N21 DZV-pd 6N31-2pd TZV-2pd 6N311-3p2d ACCD ACCT

0,057

0,052

0,047 0,45 0,50 0,55

On length around 0.5 A we have another picture. The order of sets is a little bit mixed. ACCT, ACCD, 6-311, TZV, DZV builds a close group. 6-31, 3-21 are not far away. The MIDI set is a little bit far away and seems to calculate this environment worser than 3-21

Benchmarks of Basis/Correlation Correction

Ethan: variation of C-C Bondlength

0,138

0,133

STO-2 MINI MIDI 3N21 DZV-pd 6N31-2pd TZV-2pd 6N311-3p2d ACCD ACCT

0,128

0,123 0,95 1,00 1,05

On a distance of ~ 1 A we can compare this situation with a interaction of 2 Methyl radicals. Here we see that the hough sets give lowest energy configuration, for such special environment it's not unusual that a set with so many additional Functions and Polarized & Diffuse-functions can describe this situation better. Not far away are the split valence sets 6-31 and the 3-21.

Benchmarks of Basis/Correlation Correction

For a better recapitulation we shall compare the computational time.
STO-2 13,8 186,41% MINI 13,5 200,41% MIDI 22,9 199,85% 3N21 22,4 199,75% DZV 62,9 195,20% 6N31-2pd 76,9 198,73% TZV 140,8 195,31% 6N311-3p2d 216,0 193,34% ACCD 219,0 191,59% ACCT

Zeit Util 2 CPU

Factor time in computational chemistry shall not be underestaminate. For this computation we can say that the MINI and STO-2 is qualitatively different. The MIDI and 3-21 set is qualitatively comparable to the bigger sets. For bigger molecules or a fast preview (or a slow cpu) they are a good agreement. Shown time relationship is not exact portable to other calculations, there are so many parameters which influence the calculations, but we can say the tendency is accepteable. The most CPU utilizations were ~ 100% per CPU, but the higher the molecule and the basis set is the larger is the number of stored Integrals. If they are larger than given RAM-Size they have to be stored to HDD with the consequence that the CPU utilization breaks significant down and the CPU time rise a lot, as seen for the ACCT calculation. To give a feeling how important this can be, we can see if we compare the same calculation with different RAM access. (MW for $SYSTEM MWORDS=xxx $END)
ACCT Zeit Util 2 CPU MW=180 MW=380 17524,6 38,77%

~ x times faster

Benchmarks of Basis/Correlation Correction As next we prove the basis sets on Acrolein.

C-H stretch of Acrolein
0,250

0,200
energy relative to groundstate E

0,150

0,100

0,050

0,000 -0,10

0,00

0,10

0,20

0,30

0,40

0,50

0,60

0,70

0,80

0,90

1,00

1,10

1,20

1,30

distance to equilibrium lenght [A]

Benchmarks of Basis/Correlation Correction

Acrolein : Vinyl(CO)-H bond stretch
0,26

0,16

STO-2 MINI MIDI 3N21 DZV-pd 6N31-2pd TZV-2pd 6N311-3p2d CCD CCT ACCD ACCT

0,06

-0,04 -0,30 -0,10 0,10 0,30 0,50 0,70 0,90 1,10 1,30

At first we see, that the STO-2G & MINI set is significant different from the other, again. With such quality problems, we should use them only for didactical usage or fast geometry pre-optimization (organic molecules only).

Benchmarks of Basis/Correlation Correction

Acrolein : Vinyl(CO)-H bond stretch
0,008

0,003

STO-2 MINI MIDI 3N21 DZV-pd 6N31-2pd TZV-2pd 6N311-3p2d CCD CCT ACCD ACCT

-0,002 0,00 0,05 0,10 0,15

On bond length near the equilibrium geometry of a PM3 optimized Acrolein we see that many sets haven't their minima.

Benchmarks of Basis/Correlation Correction

Acrolein : Vinyl(CO)-H bond stretch
0,067

0,062

0,057

STO-2 MINI MIDI 3N21 DZV-pd 6N31-2pd TZV-2pd 6N311-3p2d CCD CCT ACCD ACCT

0,052

0,047 0,45 0,50 0,55

Here we see the known groups. For better interpretation we should look two pages later on the total energies.

Benchmarks of Basis/Correlation Correction

Acrolein : Vinyl(CO)-H bond stretch
0,123

0,118

STO-2 MINI MIDI 3N21 DZV-pd 6N31-2pd TZV-2pd 6N311-3p2d CCD CCT ACCD ACCT

0,113

0,108 0,95 1,00 1,05

The same situation on long distances. The triple valence sets (TZV, 6-311, CCT) and some split valence sets (6-31, DZV, ACCD) runs parallel. STO-2G, MINI, MIDI are far away or like 3-21 run qualitatively in another way . For a better comparison we should look at the total energy at a distance of 1 A.
E at 1 A MIDI -190,67140 3N21 -190,73610 DZV-pd -191,83155 6N31-2pd -191,80680 TZV-2pd -191,87138 6N311-3p2d -191,86636 CCD -191,81550 CCT -191,87677 ACCD -191,83076 ACCT -191,88065

Benchmarks of Basis/Correlation Correction

Now let's compare computational effort.
STO-2G -183,80257 23,2 200,04% MINI -190,60694 25,4 200,19% MIDI -190,77909 49,9 200,09% 3N21 -190,84962 45,5 200,05%

Energie Zeit Util 2 CPU

Energie Zeit Util 2 CPU

DZV-pd -191,94875 142,7 199,39%

6N31-2pd -191,92146 173,5 198,91%

TZV-2pd -191,98662 242,7 196,42%

6N311-3p2d -191,98066 455,9 197,38%

CCD -191,92643 137,3 199,18%

CCT -191,99154 1016,8 195,97%

ACCD -191,94354 447,6 196,03%

ACCT

Here we see again how bigger sets rise computational time and which additional effect cpu utilization can have on wall clock. If we compare the energies of triple and split valence we see a little difference, this is not unusual because we have a bigger set on inner and outer orbitals. If we find better inner orbitals with triple valence set we will always find lower energy, even if the outer chemical Orbitals are like from a split valence. With this assay on acrolein we can make a first conclusion. STO-2G, MINI & MIDI is good for fast calculations, we will see later that even this sets describes i.e. oxidation of ethan with peroxoaceticacid in a right way. But if we will make research more serious we should use at least a 6-31(pd) set.

Benchmarks of Basis/Correlation Correction Here we test trans-1-Cl-Propen

1-Cl-Propen : HCCl=CHMe bond stretch
0,250

0,200
energy relative to groundstate E

0,150

0,100

0,050

0,000 -0,10

0,00

0,10

0,20

0,30

0,40

0,50

0,60

0,70

0,80

0,90

1,00

1,10

1,20

1,30

distance to equilibrium lenght [A]

Benchmarks of Basis/Correlation Correction

1-Cl-Propen : HCCl=CHMe bond stretch

0,40

0,30

0,20

0,10

STO-2 MINI MIDI 3N21 6N31 6N31-pd DZV-pd TZV-2p2d 6N311-2p2d CCD CCT ACCD ACCT

0,00

-0,10 -0,30 -0,10 0,10 0,30 0,50 0,70 0,90 1,10 1,30

Benchmarks of Basis/Correlation Correction

1-Cl-Propen : HCCl=CHMe bond stretch

0,135

0,125

STO-2 MINI MIDI 3N21 6N31 6N31-pd DZV-pd TZV-2p2d 6N311-2p2d CCD CCT ACCD ACCT

0,115

0,105 0,45 0,50 0,55

Benchmarks of Basis/Correlation Correction

1-Cl-Propen : HCCl=CHMe bond stretch

0,27

0,26

STO-2 MINI MIDI 3N21 6N31 6N31-pd DZV-pd TZV-2p2d 6N311-2p2d CCD CCT ACCD ACCT

0,25

0,24 0,95 1,00 1,05

Benchmarks of Basis/Correlation Correction

Here is the corresponding data. The cpu utilization is comparable, so we can better compare computation time now. I.e. we get much better description with aug-cc-pVTZ set in comparison to 3-21 but it tooks 200 times more computational time.

Energie Zeit [min] Util 2 CPU

STO-2G -555,3482 0,5 200,16

MINI -574,5341 0,6 200,2

MIDI -574,6387 1,2 200

3N21 -574,6706 1,2 200,06

Energie Zeit Util 2 CPU

6N31 -577,4500 1,6 200,01%

6N31-pd -577,5057 3,6 199,95%

DZV-pd -577,5326 3,8 199,99%

TZV-2p2d -577,5850 11,6 199,98%

6N311-2p2d -577,5795 10,0 199,97%

CCD -577,5401 4,3 199,95%

CCT -577,5949 27,2 199,97%

ACCD -577,5507 18,9 199,94%

ACCT -577,5974 240 199,97%

E at 1 A

MIDI -574,3811

3N21 -574,4115

DZV-pd -577,2856

6N31-pd -577,2496

TZV-2pd -577,3339

6N311-3p2d -577,3258

CCD -577,2884

CCT -577,3425

ACCD -577,3028

ACCT -577,3464

Benchmarks of Basis/Correlation Correction Now lets take a look at an classical inorganic Molecule
3OCNi-CO stretch of Ni(CO)4
0,250

0,200
energy relative to groundstate E

0,150

0,100

0,050

0,000 -0,10

0,00

0,10

0,20

0,30

0,40

0,50

0,60

0,70

0,80

0,90

1,00

1,10

1,20

1,30

distance to equilibrium lenght [A]

Benchmarks of Basis/Correlation Correction

Ni(CO)4: Variation of 3(OC)Ni-CO
0,10

0,05

MINI MIDI 3N21 6N31 6N31-dp 6N31-2d2p TZV TZV-dp TZV-2d2p Mhs-tm Mhs-ptm Cct

0,00 -0,30 -0,10 0,10 0,30 0,50 0,70 0,90 1,10 1,30

The structure was taken by a TZV-dp UHF/B3Lx optimization. Here we see the problem of today basis sets in inorganic chemistry. There are so many orbitally changing in the metallic sphere that we need a hough choice of functions for every orbital to find a good approximation of the real one.

Benchmarks of Basis/Correlation Correction

As we see, there are much bigger differences in qualitative description of an OC Ni interaction, than in organic molecules. MINI & MIDI set have even a 0.10 A shorter equilibrium Bondlength and a significant different description. In principle it is useful, but in comparison to other there are bad. The same apply to 3-21 set.
0,020

Ni(CO)4: Variation of 3(OC)Ni-CO

0,010

MINI MIDI 3N21 6N31 6N31-dp 6N31-2d2p TZV TZV-dp TZV-2d2p Mhs-tm Mhs-ptm Cct

0,000 0,05 0,10 0,15

When we assay effect of polarization functions we see even at small bond-stretch differences.

Benchmarks of Basis/Correlation Correction

Ni(CO)4: Variation of 3(OC)Ni-CO

0,038

0,028

MINI MIDI 3N21 6N31 6N31-dp 6N31-2d2p TZV TZV-dp TZV-2d2p Mhs-tm Mhs-ptm Cct

0,018 0,45 0,50 0,55

We also see that a split set of polarization functions has lower effects than one set. But if we compare computational effort we see a kind of doubling cpu time.
6N31 225,7 198,48% 6N31-dp 582,6 197,18% 6N31-2d2p 1382,7 197,71% TZV 739,2 197,85% TZV-dp 1314,1 197,73% TZV-2d2p 2462,6 195,55%

Zeit [s] Util 2 CPU

Benchmarks of Basis/Correlation Correction

Here is an overview of total energy of each basis set and calculation time.
MINI -1951,9668 85,9 199,93% MIDI -1952,7950 170,1 199,29% 3N21 -1952,1341 176,5 198,41% 6N31 -1961,3597 225,7 198,48% 6N31-dp -1961,5631 582,6 197,18% 6N31-2d2p -1961,6069 1382,7 197,71%

Energie Zeit Util 2 CPU

Energie Zeit [s] Util 2 CPU

TZV -1961,7251 739,2 197,85%

TZV-dp -1961,9104 1314,1 197,73%

TZV-2d2p -1961,9254 2462,6 195,55%

Mhs-tm -1961,6495 372,7 194,56%

Mhs-ptm -1955,9510 6901,0 199,05%

Cct -1956,1494 ~1 day

On previous calculations we saw that except STO-2 and MINI most basis sets are acceptable good for H,C,O calculations. For research we should use there at least 6-31-(pd). On Ni(CO4) we see that even a good split valence set with polarization functions have problems. So one should use for calculations which contains transition metals at least a triple valence set. We know calculation time hardly depends on chosen basis set, so we should look for some agreements. Using of hybrid sets with hough sets for transition metals and smaller sets for first two row elements is one strategy. On the other hand we create a kind of artefacts because, different sets have different strategies in describing orbitals and we have no consequent description by using hybrids. But every one can decide which kind of accuracy he needs. For didactical of private usage 6-31 is acceptable, but for research one should use a least a triple valence set.

Benchmarks of Basis/Correlation Correction Here is another benchmark of a komplex. Cl-Mn(CO)5 streching Cl-Mn Bond
Cl-Mn(CO)5

0,10
energy relative to groundstate E

0,05

0,00 -0,10

0,00

0,10

0,20

0,30

0,40

0,50

0,60

distance to equilibrium lenght [A]

Benchmarks of Basis/Correlation Correction

Cl-Mn(CO)5
0,06

MINI 6N31-pd CCD CCT SVP TZVP ptm2 ptm3

0,00

-0,05 -0,15 -0,05 0,05 0,15 0,25 0,35 0,45 0,55 0,65 0,75 0,85

Text

Benchmarks of Basis/Correlation Correction

Cl-Mn(CO)5
0,006

MINI 6N31-pd CCD CCT SVP TZVP ptm2 ptm3

0,001

-0,004 0,05 0,10 0,15

Text

Benchmarks of Basis/Correlation Correction

Cl-Mn(CO)5
0,033

0,028

MINI 6N31-pd CCD CCT SVP TZVP ptm2 ptm3

0,023

0,018 0,45 0,50 0,55

Text

Benchmarks of Basis/Correlation Correction Here is a benchmark of Ni(CO)4 again. With additional basis sets and new hardware.
3OCNi-CO stretch of Ni(CO)4
0,250

0,200
energy relative to groundstate E

0,150

0,100

0,050

0,000 -0,10

0,00

0,10

0,20

0,30

0,40

0,50

0,60

0,70

0,80

0,90

1,00

1,10

1,20

1,30

distance to equilibrium lenght [A]

Benchmarks of Basis/Correlation Correction

Ni(CO)4
0,06

MINI 6N31-pd CCD CCT SVP TZVP ptm2 ptm3

0,00

-0,05 -0,15 -0,05 0,05 0,15 0,25 0,35 0,45 0,55 0,65 0,75 0,85

Text

Benchmarks of Basis/Correlation Correction

Ni(CO)4
0,006

MINI 6N31-pd CCD CCT SVP TZVP ptm2 ptm3

0,001

-0,004 0,05 0,10 0,15

Text

Benchmarks of Basis/Correlation Correction

Ni(CO)4
0,033

0,028

MINI 6N31-pd CCD CCT SVP TZVP ptm2 ptm3

0,023

0,018 0,45 0,50 0,55

Text

Benchmarks of Basis/Correlation Correction

Another important question in chemistry is the difference between some conformations. To prove how energy difference depends on basis set we compare cis/trans conformer of but-2-ene. Additionally we can compare calculated dipol of each conformer.

Energy Cis -150,7977 -156,2157 -156,3047 -156,3717 -157,1904 -157,2370 -157,2521 -157,2812 -157,2892 -157,2928 Trans -150,7994 -156,2168 -156,3067 -156,3736 -157,1925 -157,2392 -157,2539 -157,2833 -157,2912 -157,2949 Diff 0,001680 0,001119 0,002003 0,001948 0,002071 0,002135 0,001797 0,002122 0,002040 0,002022

Dipol Cis 0,164597 0,229661 0,179939 0,202529 0,195935 0,197658 0,228780 0,212560 0,249976 0,262794 Trans 0,000196 0,000370 0,000298 0,000283 0,000281 0,000254 0,000283 0,000245 0,000247 0,000252

Time Cis 1,8 2,3 4,1 4,0 5,6 13,5 16,3 40,5 49,9 1024,6 Trans 1,8 2,2 3,5 3,8 5,0 11,8 14,5 38,0 43,5 1038,8

# Iter Cis 9 9 9 9 10 10 11 9 10 12 Trans 9 9 8 9 9 9 10 9 9 12

STO-2 MINI MIDI 3N21 6N31 6N31-dp DZV-dp 6N311-2p2d TZV-2p2d ACCT

Here we see the same tendency that STO-2 & MINI set is significant different to split or triple zeta sets.

Benchmarks of Basis/Correlation Correction

Now let's assay the dipole moment of DMF. Geometry was optimized by 6-311-dp/B3LYP Table contain energy, dipole moment, mulliken population diff. and computational time.

Dipol moment for liquid DMF is 3,86 Debye.

Benchmarks of Basis/Correlation Correction

Because in previous test we get some worse mulliken populations with good sets, here is another molecule. On 4-aminobenzonitril (6-311(pd) Geometry) we test differences of basis sets.

Here we see again that there is some kind of inconsistency in describing electron density. In some cases we have partial negatively in other partial positively charged nitrogen of nitril-group. We would expect a partial negative charge, so here we have a good example that bigger sets don't meen automatically better/realer descriptions.

.. to be continued
- more benchmarks of metal-organic molecules - benchmarks of excited states - benchmarks of different hybrid basis sets

Leibniz-Institut fr Katalyse e.V.

This document is free available. It can be used for private or educational requirements. It must not be used for commercial aim without agreement of the author. It is literary property of Marek Pawel Checinski.
http://www.catalysis.de/ http://www.chemie.hu-berlin.de/ mail: marek.checinski catalysis.de