Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University...

41
Approximate MR-CI methods eter G. Szalay otv¨ os Lor´ and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary [email protected]

Transcript of Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University...

Page 1: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

Approximate MR-CI methods

Peter G. SzalayEotvos Lorand University

Institute of ChemistryH-1518 Budapest, P.O.Box 32, Hungary

[email protected]

Page 2: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The Multireference Problem

Example: ozone (O3)

6?

6?

6?

6?

ψ(1) ψ(2)

Both determinants are important for a qualitative description!

1

Page 3: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The Multireference Problem

Example: ozone (O3)

6?

6?

6?

6?

ψ(1) ψ(2)

Both determinants are important for a qualitative description!

Both determinants should be used as reference in the truncated scheme!

1

Page 4: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The Multireference Problem

Example: ozone (O3)

6?

6?

6?

6?

ψ(1) ψ(2)

Both determinants are important for a qualitative description!

Both determinants should be used as reference in the truncated scheme!

ΦMR−CISD = c(1)ψ(1) + c(2)ψ(2)

1

Page 5: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The Multireference Problem

Example: ozone (O3)

6?

6?

6?

6?

ψ(1) ψ(2)

Both determinants are important for a qualitative description!

Both determinants should be used as reference in the truncated scheme!

ΦMR−CISD = c(1)ψ(1) + c(2)ψ(2)

+∑ia

c(1)aiψ(1)

ai +

∑ia

c(2)aiψ(2)

ai

1

Page 6: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The Multireference Problem

Example: ozone (O3)

6?

6?

6?

6?

ψ(1) ψ(2)

Both determinants are important for a qualitative description!

Both determinants should be used as reference in the truncated scheme!

ΦMR−CISD = c(1)ψ(1) + c(2)ψ(2)

+∑ia

c(1)aiψ(1)

ai +

∑ia

c(2)aiψ(2)

ai

+∑

i>j a>b

c(1)abij ψ(1)

abij +

∑i>j a>b

c(2)abij ψ(2)

abij

1

Page 7: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The Multireference Problem

6?

6?

6?

6?

ψ(1) ψ(2)

reference

2

Page 8: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The Multireference Problem

6?

6?

6?

6?

ψ(1) ψ(2)

reference

6?

6?

6?

6?

↓ ↓

double excited quadruply excited wrt ψ(1)

2

Page 9: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The Multireference Problem

6?

6?

6?

6?

ψ(1) ψ(2)

reference

6?

6?

6?

6?

↓ ↓

double excited quadruply excited wrt ψ(1)

MR-CISD includes higher than double excitation wrt to Hartree-Fock!!

2

Page 10: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The Multireference Problem

6?

6?

6?

6?

ψ(1) ψ(2)

reference

6?

6?

6?

6?

↓ ↓

double excited quadruply excited wrt ψ(1)

MR-CISD includes higher than double excitation wrt to Hartree-Fock!!

But not all, only a subset ⇒ substantial saving wrt CISDTQ!!

2

Page 11: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The Multireference Problem

What type of correlation will be included:

correlation type size number method descriptionof the individual contributions

static/non-dynamic large small MCSCF qualitativedynamic small large CI,CC,PT,... quantitative

Therefore:

• MR-CISD, MR-PT (CASPT2), MR-CC are the method of choice if bothdynamic and non-dynamic correlations are present in the system

• Reference space should be small and describe only non-dynamiccorrelation (i.e. what Hartree-Fock can not)

3

Page 12: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The Multireference Problem

orbital type reference CI space

frozen core double occupied double occupied

double occupied

or closed shell

double occupied reference −1,−2

active

or open shell

varying occupation reference ±1,±2

virtual empty empty, +1,+2

4

Page 13: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The (uncontracted) MR-CISD method

Reference space:

ψ(1), ψ(2), ..., ψ(nref)

Single excited space:

ψ(1)ai , ψ(2)

ai , ..., ψ(nref)

ai

Double excited space:

ψ(1)abij , ψ(2)

abij , ..., ψ(nref)

abij

Wave function:

ΨMR−CISD =

nref∑m

c(m)ψ(m) +∑ia

nref∑m

c(m)aiψ(m)

ai +

∑i>j a>b

nref∑m

c(m)abijψ(m)

abij

Expansion length: about nref times that of SR-CISD

5

Page 14: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The (uncontracted) MR-CISD method

Reference space:

ψ(1), ψ(2), ..., ψ(nref)

Single excited space:

ψ(1)ai , ψ(2)

ai , ..., ψ(nref)

ai

Double excited space:

ψ(1)abij , ψ(2)

abij , ..., ψ(nref)

abij

Wave function:

ΨMR−CISD =

nref∑m

c(m)ψ(m) +∑ia

nref∑m

c(m)aiψ(m)

ai +

∑i>j a>b

nref∑m

c(m)abijψ(m)

abij

Expansion length: about nref times that of SR-CISD

First MR-CI calculation of this type was performed by Liu [1] on H3 (1973).

5

Page 15: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

MR-CISD: Some important things to consider

• Some of the excited CSFs can be produced from several referencefunctions → redundancy

– in CI we can simply leave them out– for Coupled Cluster this is a big problem

• Some of the excited CSFs do not have non-vanishing matrix elementswith any of the reference functions → one can select just the first orderinteracting space [2]

– not a problem if excitation level defined wrt spinorbitals like in CC

• Single-double excitations out of non-symmetrical reference functionsmight have the right symmetry → these should be included if non-sym-metrical regions of the potential energy surface are also of interest.

• Orbitals: usually from MCSCF using the reference space, but this is notnecessary: it can be larger or even smaller (technical issue).

6

Page 16: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

The Multireference Problem

Shavitt graph for MR-CISD wavefunction:

• Thick lines: reference CSFs

– run together in the double occupiedand virtual space

– diverge in the “active” space

• both double occupied and virtual partof the graph has simple structure

7

Page 17: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Internally-contracted MR-CISDMeyer [3], Siegbahn [4], Werner and Reinisch [5]

Reference space (with a(m) fixed, e.g. from MCSCF calculation):

{ψ(1), ψ(2), ..., ψ(nref)} → ψ0 =

nref∑m

a(m)ψ(m)

Single excited functions:

ψai ≡ a

+a aiψ0 =

nref∑m

a(m)ψ(m)ai

Double excited functions:

ψabij ≡ a

+a a

+b aiajψ0 =

nref∑m

a(m)ψ(m)abij

Wave function – same expansion length as SR-CISD:

Ψic−MR−CISD = c0ψ0 +

∑ia

cai ψ

ai +

∑i>j a>b

cabij ψ

abij

8

Page 18: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

MR-CI ansatze

Expansion functions and their number in different CI ansatze:

Reference Singles Doubles

SR ψ0

ψai ψ

abij

1 nocc × nvrt(nocc2

)×(nvrt2

)

uc-MR a(1)ψ(1) + a(2)ψ(2) ψ(1)ai , ψ(2)

ai ψ(1)

abij , ψ(2)

abij

nref nref × nocc × nvrt nref ×(nocc2

)×(nvrt2

)

ic-MR ψ0 = a(1)ψ(1) + a(2)ψ(2) ψai ψ

abij

1 nocc × nvrt(nocc2

)×(nvrt2

)uc: uncontracted, ic: internally contracted

9

Page 19: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

ic-MR-CISD by Werner and Knowles [6, 7]

Reference space:

ψ(1), ψ(2), ..., ψ(nref)

Single excited space – not contracted:

ψ(1)ai , ψ(2)

ai , ..., ψ(nref)

ai

Double excited functions:

ψabij ≡ a

+a a

+b ajaiψ0 =

nref∑m

a(m)ψ(m)abij

Wave function:

Ψic−MR−CISD =

nref∑m

c(m)ψ(m) +∑ia

nref∑m

c(m)aiψ(m)

ai +

∑i>j a>b

cabij ψ

abij

Expansion length: not substantially larger than SR-CISD

10

Page 20: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Internally-contracted MR-CISD – newest developments:

• more complete contraction by Shamasundar, Knizia, and Werner[8]: alsothose single excitations are contracted which include active orbitals.

• GUGA based ic implementation by Wang, Han, Lei, Suo, Zhu, Song, andZhenyi Wen[9]

• DMRG-based internally contracted MRCI by Saitow, Kurashige, Yanai[10, 11]

11

Page 21: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Internally-contracted MR-CISD

What are the advantages?

• the size of the expansion is practically independent of the number ofreference functions

• the contracted configurations span exactly the first order interactingspace of the reference function

• much cheaper than uncontracted version

What are the complications?

• the contracted configurations are, in general, not orthogonal

• structure for the contracted configurations is extremely complex

12

Page 22: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Internally-contracted MR-CISD

What are the approximations?

• much less variational parameter

• coefficients are approximated by product: c(m)abij = a(m) cabij

• semi internal excitations: if {i, j} or {a, b} refer to active orbitals, notall reference can be excited.

What are the problems?

• problem for surface crossing: degeneracy point will be different for thereference space and for the full space.

• no way to systematically estimate the error

• very often not considered as approximation

13

Page 23: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Internal contraction (ic) vs. none-contraction (uc):ozone ground stateQUANTUM CHEMICAL CALCULATIONS ON THE POTENTIAL ENERGY SURFACE OF OZONE

!"###

#

"###

$###

%###

&###

'####

'"###

'$###

'%###

'&###

"(" "() "($ "(* "(% "(+ "(& "(, )

-./0

123.4.

/.51

-67

254.

7.-8

94:5

;!'

<

-' :0(=(<

>"34?9;;.1-254>=1@455!A!BCD!E0F2F4F.1

25!GH!IJ>>

GH!IJ>>

Figure 2: C2v Symmetric Cut and MEP-Cut were calculated with MR-AQCC and ic-MR-AQCC and a cc-pVTZ-basis set. The energy within the C2v Symmetric Cuts is relative to the ozone energy at r1 = 2.4 a.u.,r2 = 2.4 a.u. and ! = 117!, within the MEP Cut relative to r1 = 2.275 a.u., r2 = 3.4 a.u. and ! = 117!.

The dissociation energy predicted at uc-MR-AQCC level is larger by 86 meV to 95 meV than its inter-nally contracted counterpart (see Table 8). In fact, uc-MR-AQCC dissociation energy with a cc-pV5Zeven overestimates the experimental value by 0.030 eV.The results suggest, that the barrier is an artifact of the internal contraction which is also in line withthe experimental finding of mass-independent isotope enrichment effect in ozone. In addition, this mayindicate that the good agreement of the ic-MR-AQCC dissociation energy at CBS limit 1.126 eV andthe experimental value 1.143 eV is owed to error compensation. On the other hand, overestimating theexperimental dissociation energy at the CBS limit with the uc-MR-AQCC by 88 meV is larger than tobe expected at this level of theory.

3.3 (ic)-MR-CISD

The energies calculated with MR-CISD+QP and ic-MR-CISD+QP almost reproduce the respectiveMR-AQCC data (Fig. 4).The a posteriori Pople size consistency correction decreases the barrier height, lowers the van der Waalsminimum and shifts the barrier maximum below the dissociation limit.

Also the MR-CISD energies a posteriorily Pople corrected are in agreement with the observed trendsfor MR-AQCC: the dissociation energies derived from the internally contracted variants only slightlyunderestimate the experimental value and the size-extensivity correction amounts to 0.06 eV at theCBS limit. The uncontracted calculations yield at cc-pV5Z level a higher dissociation energy by0.062 eV (MR-CISD) compared to the ic calculation, a larger with the Pople correction (0.083 eV)also consistently leading to an overestimate of the experimental dissociation energies as compared touc-MR-AQCC.Within the framework of the ic-MR-CISD method it is possible to increasingly betterapproximate the uc-MR-CISD method by increasing the number of reference states. To investigatethe correctness of the uncontracted methods, the energies of r2 = 3.9 a.u. of the MEP (barrier top)

72

• uc curve is less steep

• barrier is disappearing

14

Page 24: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

ic vs. uc: reef in ozone dissociation channel disappearsQUANTUM CHEMICAL CALCULATIONS ON THE POTENTIAL ENERGY SURFACE OF OZONE

!"##

#

"##

$##

%##

&##

'###

'"##

'$##

" $ % & '# '" '$ '%

()*+

,-.)/)

*)0,

(12-0

/)2)(3

4/50

6!'

7

(" 5+8987

:;</=>?@/:A!=BCD/E/F<G-0!:A!=BCD/E/F<

-0!:A!=BCD@/00!HI?J

-0!:A!=BCD@/00!HIFJ

:A!KF==@/00!HI?J

:A!KF==@/00!HIFJ

!"##

#

"##

$##

%##

&##

'###

'"##

'$##

" $ % & '# '" '$ '%

()*+

,-.)/)

*)0,

(12-0

/)2)(3

4/50

6!'

7

(" 5+8987

:;<!=>?@:A!=BCDGEF<

:A!=BCDEF<@/00!HI?J

:A!=BCDEF<@/00!HIFJ

:A!=BCD@/00!HI?J

:A!=BCD@/00!HIFJ

Figure 4: MEP-Cut was calculated with MR-CISD+QP and ic-MR-CISD+QP, and with MR-CISD and MR-CISD+QP. The energies of the MEP Cut are relative to ozone at r1 = 2.275 a.u., r2 = 3.4 a.u. and ! = 117!.

Table 9: Comparison of the dissociation energies calculated with ic-MR-CISD methods with and without sizeconsistency correction

basis set ic-MR-CISD ( eV) ic-MR-CISD+QD ( eV) ic-MR-CISD+QP ( eV) ic-MR-AQCC ( eV)

cc-pVTZ 0.857 0.895 0.893 0.887cc-pVQZ 0.982 1.032 1.030 1.022cc-pV5Z 1.032 1.089 1.087 1.078cc-pV6Z 1.051 1.111 1.109 1.099

(Q,5) 1.077 1.142 1.139 1.129(5,6) 1.075 1.138 1.136 1.126

Table 10: Comparison of the dissociation energies calculated with MR-CISD methods with and wihtout sizeconsistency correction

basis set MR-CISD (eV) MR-CISD+QD (eV) MR-CISD+QP (eV) MR-AQCC (eV)

cc-pVTZ 0.916 0.984 0.983 0.973cc-pVQZ 1.040 1.124 1.123 1.111cc-pV5Z 1.094 1.191 1.177 1.173

(Q,5) 1.140 1.235 1.234 1.231

Table 11: Relative energies of the barrier top and the van der Waals minimum calculated with ic-MR-CISD+QP,a cc-VTZ basis set and a number of reference wave functions and compared with the MR-CISD+QP-energy

Number of reference wave functions Relative energies ( cm!1)barrier top van der Waals minimum

r2 = 3.9 a.u. r2 = 5.0 a.u.

1 0 02 -75.11 -52.524 -133.11 -82.835 -143.14 -89.35

10 -191.73 -108.36

MR-CISD+QP -3175.07 -2813.97

74

15

Page 25: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Internally-contracted MR-CISD

Internal contraction is also used in:

• MC-CEPA

• CASPT2

• certain MR-CC ansatze

16

Page 26: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Externally contracted MR-CI

Introduced in 1983 by Siegbahn [12].

Expansion space consists of the following double excited functions:

ψ(m)ij =∑a>b

c(m)abij ψ(m)abij

i.e. the configuration with the same hole orbitals are contracted.

The contraction coefficients are determined from first-order PT:

cabij =〈ψ0|H|ψ(m)abij 〉

E0 − 〈ψ(m)abij |H|ψ(m)abij 〉

17

Page 27: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Externally contracted MR-CI

If the virtual orbitals are placed at the bottom of the Shavitt graph thanthe number of configurations equals the number of the internal walks.

18

Page 28: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Externally contracted MR-CI

Computational effort:

• comparable with a single iteration of an uncontracted MRCI calculation.

Applicable:

• for low-lying valence states, but not highly excited states

Not popular nowadays (but see Wang, Ga, Su, Wen [13] or certain PTmethods).

19

Page 29: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Doubly Contracted MRCI

Introduced in 2004 by:

• Yubin Wang, Binbin Suo, Gaohong Zhai, Zhenyi Wen[14]

Simultaneous use of internal and external contraction.

Suggested for:

• potential energy surface calculation

Although the initial results were very promising, only a few applicationshave been reported.

20

Page 30: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Graphical Contraction Function method

The GCF method of Shepard and coworkers is also a special contracted CImethod which uses the Shavitt-graph.

For details see Ron’s talk.

21

Page 31: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Individual selection

Both internal and external contractions make the CI expansion space smaller,however, treat kind of average of the individual expansion functions.

One can also estimate the contribution of individual CSF’s by perturbationtheory, and leave out those with small contribution from the expansionspace.

Two steps of the selection procedure:

• estimation of the contribution

• correction for the missing contribution

22

Page 32: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Individual selection

Assume a multiconfigurational reference function:

Ψ0 =

k∑i

aiΦi

Two strategies are used to calculate the individual contribution of a CSF(Φj, j > k):

First, the energy of adding Φj to the space {Φi i = 1, k} can beestimated by perturbation theory (Ak procedure by Gershgorn and Shavitt[15]):

εj =|〈Φj|H|Ψ0〉|2

E0 −Hjj

with E0 = 〈Ψ0|H|Ψ0〉 being the reference energy.

23

Page 33: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Individual selection

The second is termed as the Bk procedure byGershgorn and Shavitt [15]:

– diagonalize the matrix which is complete inthe reference space, but only diagonal elementsare present for the external space

– the resulting coefficients cj are used to

calculate the contribution εj =(E−Hjj)c2j

1−c2j

24

Page 34: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Individual selection

The selection can be again performed in twoways:

• Threshold selection: if εj is smaller than a given number (usually10−4 − 10−5 hartree), the configuration is not included in the ansatz.

• Cumulative selection: CSF’s are excluded one by one in order ofincreasing contribution until the sum of the contributions of the excludedconfigurations exceeds a threshold (usually 10−1 − 10−2 hartree).

The later shows an advantage for potential surfaces calculation!

25

Page 35: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Individual selection

Correction for the neglected configurations: trivially, the εj estimates canbe added to the total energy of the CI procedure.

Disadvantage: perturbation theory overestimate the effect.

Extrapolation procedure by Buenker and Peyerimhoff [16]:

E(0) = E(T ) + λ∆E(T )

E(T ) is the energy with threshold T , ∆E(T ) is the corresponding correction(sum of the excluded energy contributions). λ will be fitted, its value issmaller than 1.

26

Page 36: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Individual selection

In praxis:

• MRD-CI by Peyerimhoff and Buenker [17]

• CIPSI by Malrieu et al. [18]

• MELDF by Davidson et al. [19]

• Diesel CI by Hanrath and Engels et al. [20]

27

Page 37: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Approximate MR-CI ansatze

• contracted CI (see above)

• individual selection (see above)

• local approaches

– first idea by Saebo and Pulay [21]– more elaborate implementation by Carter et al. [22]

• pseudospectral methods by Friesner [23] and Martinez and Carter [24]

• reduced virtual space methods

– PNO (Pair Natural Orbitals) [25] – also in MC-CEPA

28

Page 38: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

Inclusion of connected higher excitations

No routine technique yet, but there are several attempts:

• via dressed Hamiltonian by Malrieu et al. [26]

• using Coupled-Cluster equations by Meissner [27]

• including selected higher excitations by Sherrill and Schaefer [28]

• Sychrovsky and Carsky [29] used the so called Bk approximation

• Correlation Energy Extrapolation by Intrinsic Scaling by Bytautas andRuedenberg [30]

• MRCISD(TQ), a hybrid variational-perturbational approach by Khait,Hoffman et al. [31]

29

Page 39: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

References

[1] B. Liu, J. Chem. Phys. 58, 1925 (1973).

[2] A. Bunge, J. Chem. Phys. 53, 20 (1970).

[3] W. Meyer, in Methods of Electronic Structure Theory, edited by H. F. Schaefer III,

page 413, New York, 1977, Plenum Press.

[4] P. E. M. Siegbahn, Int. J. Quantum Chem. 18, 1229 (1980).

[5] H.-J. Werner and E. A. Reinisch, J. Chem. Phys. 76, 3144 (1982).

[6] H.-J. Werner and P. J. Knowles, J. Chem. Phys. 89, 5803 (1988).

[7] P. J. Knowles and H.-J. Werner, Chem. Phys. Lett. 145, 514 (1988).

[8] K. R. Shamasundar, G. Knizia, and H.-J. Werner, J. Chem. Phys. 135, 054101

(2011).

[9] Y. Wang, H. Han, Y. Lei, B. Suo, H. Zhu, Q. Song, and Z. Wen, J. Chem. Phys.

141, 164114 (2014).

[10] M. Saitow, Y. Kurashige, and Y. T, J. Chem. Phys. 139, 044118 (2013).

[11] M. Saitow, Y. Kurashige, and Y. T, J. Chem. Theory Comput. 11, 5120 (2015).

[12] P. E. M. Siegbahn, Int. J. Quantum Chem. 23, 1869 (1983).

[13] Y. Wang, Z. Gan, K. Su, and Z. Wen, Chem. Phys. Lett. 312, 277 (1999).

[14] Y. Wang, B. Suo, G. Zhai, and Z. Wen, Chem. Phys. Lett. 389, 315 (2004).

30

Page 40: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

[15] Z. Gershgorn and I. Shavitt, Int. J. Quantum Chem. 2, 751 (1968).

[16] R. J. Buenker and S. D. Peyerimhoff, Theor. Chim. Acta 39, 217 (1975).

[17] R. J. Buenker and S. D. Peyerimhoff, in New Horizons In Quantum Chemistry, edited

by P.-O. Lovdin and B. Pullman, page Dodrecht, 1983, Reidel.

[18] S. Evangelisti, J.-P. Daudey, and J.-P. Malrieu, Chem. Phys. 75, 91 (1983).

[19] Meldf electronic structure codes was developed by l.e. mcmurchie, s.t. elbert, s.r.

langhof, e.r. davidson and d. feller and was extensively modified by d.c. rowlings and

r.j. cave.

[20] M. Hanrath and B. Engels, Chem. Phys. 225, 197 (1997).

[21] S. Saebo and P. Pulay, Annu. Rev. Phys. Chem. 44, 213 (1993).

[22] D. Walter, A. Venkatnathan, and E. A. Carter, J. Chem. Phys. 118, 8127 (2003).

[23] R. Friesner, Annu. Rev. Phys. Chem. 42, 341 (1991).

[24] T. J. Martinez and E. A. Carter, J. Chem. Phys. 102, 7564 (1995).

[25] R. Ahlrichs, F. Driessler, H. Lischka, V. Staemmler, and W. Kutzelnigg, J. Chem.

Phys. 62, 1235 (1975).

[26] J. Sanchez-Marin, I. Nebotgil, D. Maynau, and J.-P. Malrieu, Theor. Chim. Acta 92,

241 (1995).

[27] L. Meissner, Int. J. Quantum Chem. 108, 2199 (2008).

[28] C. Sherrill and H. F. Schaefer, J. Phys. Chem. US 100, 6069 (1996).

31

Page 41: Approximate MR-CI methods · Approximate MR-CI methods P eter G. Szalay E otv os Lor and University Institute of Chemistry H-1518 Budapest, P.O.Box 32, Hungary szalay@chem.elte.hu

P.G. Szalay: Approximate MR-CI methods Tianjin, October 10-14, 2016

[29] V. Sychrovsky and P. Carsky, Mol. Phys. 88, 1137 (1996).

[30] L. Bytautas and K. Ruedenberg, J. Chem. Phys. 121, 10905 (2004).

[31] Y. G. Khait, H. Song, and M. R. Hoffmann, Chem. Phys. Lett. 372, 674 (2003).

32