Full text of DOI:10.1063/1.1288377

Full configuration interaction and multiconfigurational spin density in boron and
carbon atoms
Michael V. Pak and Mark S. Gordon
Citation: The Journal of Chemical Physics 113, 4238 (2000); doi: 10.1063/1.1288377
View online: http://dx.doi.org/10.1063/1.1288377
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/113/10?ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
Direct configuration interaction and multiconfigurational self-consistent-field method for multiple active spaces
with variable occupations. I. Method
J. Chem. Phys. 119, 9364 (2003); 10.1063/1.1615954
Density-functional study of intramolecular ferromagnetic interaction through m-phenylene coupling unit (II):
Examination of functional dependence
J. Chem. Phys. 113, 10486 (2000); 10.1063/1.1290008
A complete active space self-consistent field multireference configuration interaction study of the low-lying
excited states of BrO
J. Chem. Phys. 113, 8556 (2000); 10.1063/1.1318751
Quadratic configuration interaction versus coupled-cluster theory: Importance of orbital relaxation phenomena in
CuH and CuF
J. Chem. Phys. 106, 7185 (1997); 10.1063/1.473680
Multireference configuration interaction study of the mixed Valence–Rydberg character of the C 2 H 4 1 (π,π * )V
state
J. Chem. Phys. 106, 7208 (1997); 10.1063/1.473275
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.76.6 On: Tue, 25 Nov 2014 18:25:28

JOURNAL OF CHEMICAL PHYSICS
VOLUME 113, NUMBER 10
8 SEPTEMBER 2000
Full configuration interaction and multiconfigurational spin density
in boron and carbon atoms
Michael V. Pak and Mark S. Gordona)
Department of Chemistry, Iowa State University, Ames, Iowa 50011
͑Received 24 April 2000; accepted 14 June 2000͒
The reliability of spin polarization method results for atomic spin densities, obtained with several
widely used Gaussian basis sets, is examined by comparison with the results of full configuration
interaction ͑FCI͒ calculations. The spin densities obtained with these basis sets using the spin
polarization model and some other methods disagree with the FCI treatment. Since the FCI wave
function is exact for a given basis, it is not clear that the spin polarization model will be generally
reliable. A large active space multiconfigurational ͑CASSCF͒ calculation is shown to be inadequate
as an alternative to FCI treatment. The importance of accounting at least to some extent for
excitations to all orbitals in the complete space of basis functions is illustrated by very slow
convergence of CASSCF results with increasing size of active space. The FCI results reported here
can be used as benchmarks to test various approaches to spin density calculation. © 2000
American Institute of Physics. ͓S0021-9606͑00͒31034-0͔
I. INTRODUCTION
difficult and arguably the most interesting case is the only
type of hfcc calculations we will be concerned with here.
One way to describe the spin polarization effects is to
use a spin-unrestricted Hartree–Fock wave function ͑UHF͒.
The semiempirical INDO method, based on the UHF wave
function, in many instances gives quantitative agreement
with experiment.1 A variety of ab initio methods using UHF
wave functions has been used to calculate hfcc’s, sometimes
giving very accurate results. Some of the methods included
electron correlation, e.g., MBPT͑4͒,2 coupled-cluster single
double ͑CCSD͒,2 QCISD.3 The UHF-based methods also
provide a physical explanation for often observed negative
spin densities. However, the problem of spin contamination
makes the use of any spin-unrestricted methods suspect, even
if those methods include some procedures to remove the con-
tamination ͑PUHF,4 UHF-AA5͒.
Electronic spin density at the positions of the nuclei de-
termines the isotropic part of the interaction between the
magnetic moments of the electron and the nuclei, also re-
ferred to as the Fermi contact interaction. Since this interac-
tion is experimentally observable in electron paramagnetic
resonance spectroscopy ͑EPS͒ as an isotropic hyperfine cou-
pling constant ͑hfcc͒, calculation of the spin density at the
nuclei is an important problem for electronic structure meth-
ods. It is also a difficult problem, because unlike most other
electronic properties such as dipole moments, polarizabil-
ities, etc., the hfcc’s are determined by the amplitude of the
wave function at a single point in space. However, in most
quantum chemistry methods the wave function is found by
optimization based on some global energy criterion. As a
result, a variationally very good wave function may have
significant error at some particular point in space.
Another approach to describing the spin polarization ef-
fects requires construction of a multideterminant wave func-
tion, to include all excitations accounting for orbital and spin
polarization of s, p, and sometimes d shells. This ‘‘spin po-
larization’’ wave function is then optimized using some kind
of multiconfiguration self-consistent field ͑MCSCF͒ or con-
figuration interaction ͑CI͒ procedures. An extensive review
of the spin polarization method can be found in Ref. 6. One
important feature of this approach is the incomplete character
of the wave function. Including into the wave function only
those configurations which appear to be important according
to a very simple physical picture of the phenomenon greatly
simplifies the calculations; it also seems to allow a better
understanding of the roles of different contributions to
hfcc’s. The latter problem has been a subject of thorough
analysis, since it is relevant not only in the context of the
spin polarization model, but also for various levels of CI
treatment. In fact, the spin-polarized MCSCF ͑SP-MCSCF͒
and various CI methods alike take advantage of the idea that
only some particular excitations are responsible for the spin
The local character of the hfcc’s makes the calculation
very sensitive to the quality and size of the atomic basis.
This sensitivity becomes extreme for systems in which un-
paired electrons do not contribute directly to the spin density
at the nuclei. For example, the unpaired electron in the
ground state of boron atom occupies a p orbital having a
node at the position of the nucleus, and thus cannot contrib-
ute directly to the spin density at the nucleus. The remaining
s electrons do have nonzero density at the nucleus, but they
all are paired. In this case the restricted open-shell Hartree–
Fock ͑ROHF͒ method predicts a value of zero for the isotro-
pic hfcc, while experimentally it is equal to 12 MHz. This
nonzero value is due to spin polarization contributions, re-
sulting from nominally paired electrons having different ex-
change interactions with the unpaired electrons. A great
number of open-shell radicals with an unpaired electron in a
␲-type orbital belong to this class of systems. This most
a͒
Electronic mail: mark@si.fi.ameslab.gov
0021-9606/2000/113(10)/4238/4/$17.00
4238
© 2000 American Institute of Physics
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.76.6 On: Tue, 25 Nov 2014 18:25:28

J. Chem. Phys., Vol. 113, No. 10, 8 September 2000
Spin density of B and C atoms
4239
density at the nucleus, while a great number of other con-
figurations are unimportant in this respect and can be omitted
from the wave function. A detailed discussion of the influ-
ence of various CI excitation classes on hfcc’s can be found
in Refs. 7 and 8, while Refs. 6 and 9 deal with the same
problem with respect to the SP-MCSCF approach. It should
be noted that unlike the UHF treatment of spin polarization,
all multideterminant-based methods account to some extent
for true electron correlation effects, which are found to be
significant for accurate description of hfcc’s.
rate hfcc’s, but adding double excitations ͑SD-CI͒ leads to
much poorer results.17 A number of studies observed dete-
rioration of results when larger and more flexible basis sets
were used with the same computational method.11,12,17 Sev-
eral other examples of such ‘‘paradoxical’’ behavior will be
given in the present paper.
One possible way to analyze this problem is to separate
the two main variables determining the accuracy of hfcc
calculations—the size and flexibility of the basis set and the
number and type of excitations to be included in the wave
function. Since a full CI ͑FCI͒ calculation gives an exact
wave function for a given basis set, the accuracy of hfcc’s
obtained by an FCI treatment reflects only the suitability of
the basis set used in calculations, and not the sufficiency of
the level of electron correlation. The suitability of various
Gaussian basis sets for hfcc calculations has been a topic of
several studies, with FCI ‘‘calibration’’ calculations used as
an ultimate test.6,7,13 Two of these works7,13 address the ni-
trogen atom. Several Gaussian basis sets demonstrated to be
inadequate for spin density calculations for N, however, are
shown to perform well for other systems.6 Among the first
row atoms, nitrogen is perhaps the easiest system for spin
density calculations ͑although it is much more difficult than
most molecules͒.
In this work we calculate the FCI limit for hfcc’s of the
boron and carbon atom ground states, using a Gaussian basis
set designed specifically for the purpose of spin density
calculations.18 The results in Ref. 18 are based entirely on
the spin polarization model. Having the exact ͑FCI͒ spin
densities allows us to place this method in perspective with
regard to the suitability of various basis sets for hfcc calcu-
lations, and also provides one with a good benchmark for
testing different approaches for these very difficult systems.
The experimental values for boron and carbon are not well
established.19 Therefore, despite the deficiencies of our FCI
results due to incompleteness of the basis sets used, they are
nevertheless the best results currently available for these ba-
sis sets and as such can be used to test less computationally
expensive methods.
Unfortunately, many conclusions made regarding the
role of various contributions to the spin density appear to be
very much basis-set dependent. When Gaussian basis sets are
used for hfcc calculations, additional difficulty arises due to
the fact that Gaussian functions, having zero radial slope at
the nucleus, are unable to satisfy the correct cusp condition
associated with the singularity of the Coulomb potential at
the nucleus.10 Slater-type orbitals, on the other hand, do have
a cusp and can satisfy the cusp condition. However, using
Slater-type orbitals ͑STO͒ instead of Gaussians for hfcc cal-
culations, as suggested in Ref. 11 is generally not conve-
nient, since most quantum chemistry programs currently use
available Gaussian basis sets. Another approach, developed
in Ref. 12, suggests expanding an STO basis set in Gaussian
functions, calculating the wave functions with Gaussian type
orbitals ͑GTOs͒ and then replacing the Gaussians with the
corresponding STOs for the single purpose of hfcc evalua-
tion, assuming no change in the expansion coefficients. This
method, although quite effective, limits the choice of basis
sets to be used in calculations to only those sets for which
STO–GTO conversion is available. The deficiency of Gauss-
ian functions, on the other hand, does not necessarily make
them unsuitable for hfcc calculations. In many simple cases,
if the bulk of the spin density at the nucleus comes from an
unpaired electron and the total value is relatively large, even
small Gaussian basis sets do reproduce the experimental val-
ues for hfcc’s. It has been shown that with very large Gauss-
ian basis sets, high accuracy can also be achieved for the
most difficult systems, like a nitrogen atom.13
An entirely different approach, initially suggested by
14
Hiller, Sucher, and Feinberg ͑HSF͒ for the charge density,
and later developed by Harriman15 for spin density, substi-
tutes a global operator for the local delta function-type op-
erator, thus avoiding most of the problems discussed above.
With the recent development of HSF formalism to a more
general class of global operators,16 this approach allows one
to calculate hfcc’s with high accuracy. Another advantage of
this method is that unlike ␦ function-based approaches, it
always gives better values for hfcc’s as the wave function is
variationally improved.
In view of the previous discussion, it is understandable
that the choice of a method and a basis set for hfcc calcula-
tions can be a very confusing problem. There are numerous
examples for which a particular CI expansion used together
with some particular Gaussian basis set systematically give
quite accurate values for hfcc’s, whereas including more CI
configurations into the wave function or using a larger basis
set destroys the good agreement with experiment. For in-
stance, a single excitations CI ͑S-CI͒ often gives very accu-
II. COMPUTATIONAL METHODS
In this work all calculations were performed using the
GAMESS program.20 The Gaussian basis sets used in the
present work are those developed by Chipman.18 These basis
sets are various segmented contractions, suggested by
Dunning,21 of the commonly used (9s5p) primitive Gauss-
ian basis set of Huzinaga.22 The changes suggested by Chip-
man include uncontracting the outer member of the inner-
most contraction and adding diffuse s and p functions and
one or more d functions. For comparison, calculations were
also done with the original Dunning basis sets, with and
without diffuse sp and d polarization functions. A complete
list of all basis sets employed in our calculations is given in
Table I. The diffuse sp exponents used in all basis sets are
͑0.0330 s, 0.0226 p͒ for boron and ͑0.0479 s, 0.0358 p͒ for
carbon. When only one set of d polarization functions is
added to basis sets, its exponent is 0.32 for boron and 0.51
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.76.6 On: Tue, 25 Nov 2014 18:25:28

4240
J. Chem. Phys., Vol. 113, No. 10, 8 September 2000
M. V. Pak and M. S. Gordon
TABLE I. Gaussian basis sets used for spin density calculations in boron
and carbon atoms.
TABLE II. Full CI spin density of boron atom calculated with various basis
sets.
Basis set
4s,2p , 5s,3p
Reference
Total energy
͑a.u.͒
Spin density
͑a.u.͒
Spin density
͑a.u.͒
͑SP MCSCF͒
21
18
͓
͔
͓
͔
a
͑FCI͒
͑FCI͒
4s,2p , 5s,3p ϩdiff sp,
͓
͓
͔
͓
͔
4s,2p
4s,2p ϩsp
4s,2p ϩspϩd
4s,2p ϩspϩdd
Ϫ24.582 402
Ϫ24.582 889
Ϫ24.603 558
Ϫ24.606 311
Ϫ0.000 818 0
Ϫ0.000 552 4
0.016 391 9
0.019 047 6
Ϫ0.0019
͓
͓
͓
͓
͔
͔
͔
͔
4s,2p , 5s,3p ϩdiff spϩd
͔
͓
͔
5s ,2p , 6s,3p
͔
18
18
͓
Ј
͔
͓
5s ,2p , 6s,3p ϩdiff spϩd,
͓
͓
Ј
͔
͓
͔
2
5 s,2p
͓ Ј
5 s,2p ϩsp
͓ Ј
5 s,2p ϩspϩd
͓ Ј
5 s,2p ϩspϩdd
͓ Ј
Ϫ24.583 794
Ϫ24.584 257
Ϫ24.604 987
Ϫ24.607 736
Ϫ0.012 807 2
Ϫ0.009 803 1
0.009 149 5
0.011 206 5
͔
͔
͔
͔
5s ,2p , 6s,3p ϩdiff spϩdd
Ј
͔
͓
͔
Ϫ0.0079
for carbon. When two sets of d functions are added, the
exponents are ͑0.1600, 0.6400͒ for boron and ͑0.2800,
1.1200͒ for carbon.
It should be noted that Chipman’s basis sets use five-
component spherical d functions instead of the six Cartesian
components, and therefore all calculations were done with
spherical functions. Using Cartesian d functions changes re-
sults significantly, since they essentially add an extra s func-
tion to the basis set.
5s,3p
5s,3p ϩsp
5s,3p ϩspϩd
5s,3p ϩspϩdd
Ϫ24.587 073
Ϫ24.587 435
Ϫ24.608 220
Ϫ24.611 127
0.009 421 9
0.009 166 9
0.026 801 9
0.029 012 0
0.0133
0.0132
͓
͓
͓
͓
͔
͔
͔
͔
6s,3p
͓6s,3p͔ϩsp
6s,3p ϩspϩd
6s,3p ϩspϩdd
Ϫ24.589 024
Ϫ24.589 309
Ϫ24.610 144
Ϫ24.613 042
Ϫ0.021 444 3
Ϫ0.017 262 6
0.004 129 7
0.005 528 0
͓
͔
Ϫ0.0079
0.0140
0.0143
͓
͓
͔
͔
aSpin polarization MCSCF results Ref. 18.
bNotation introduces by Chipman and described in Ref. 18.
Full CI calculations with the basis sets discussed above
result in CI expansions containing millions of determinants
and were greatly facilitated by using a very fast determinant-
based CI code written by Ivanic and Ruedenberg.23
culations. For carbon, even with a single d function FCI
considerably overestimates the correct value for the spin den-
sity. It is reasonable to expect13,18 that adding the second and
then the third d function to the basis set would further in-
crease the spin density, thus making the results even less
accurate.
The results of spin polarization calculations apparently
do not agree well with the FCI limit, and there appears to be
no systematic dependence on the way in which the results of
these two methods differ. This would suggest that the spin
polarization model overlooks some excitations that are im-
portant for a correct description of the spin density at the
nuclei. On the other hand, the highly unsatisfactory results of
FCI calculations suggest that the basis sets considered here
are inadequate for the task; therefore, spin density obtained
2
Since electronic ground states of B and C atoms are P
and 3P, respectively, in all complete active space self-
consistent field ͑CASSCF͒ calculations reported in this work,
the electron density was averaged over all degenerate states,
2
ˆ
to ensure that the resulting wave function was a true L
͑angular momentum͒ eigenfunction.
III. RESULTS AND DISCUSSION
As noted in the Introduction, FCI provides the ultimate
test of the suitability of an atomic orbital ͑AO͒ basis set for
spin density calculations. FCI spin densities are essentially of
benchmarking value; if any other method gives better results
for spin densities using the same basis set, these results are
likely to be due to fortuitous cancellation of errors. In this
work we report FCI spin densities in boron and carbon at-
oms, obtained with several basis sets commonly used for
spin density calculations. Table II contains total energies and
spin densities for boron; Table III presents the same data for
carbon. The best available experimental spin densities are
0.0081 for boron24 and 0.0173 for carbon.25 Our choice of
Chipman’s Gaussian basis sets for this study was prompted
by the illustration18,26 that these basis sets are capable of
providing a reasonably good description of Fermi contact
spin densities. The last column of the tables gives the spin
density obtained by Chipman18 using the corresponding basis
sets and the spin polarization MCSCF model discussed
above.
TABLE III. Full CI spin density of carbon atom calculated with various
basis sets.
Total energy
͑a.u.͒
Spin density
͑a.u.͒
Spin density
͑a.u.͒
͑SP MCSCF͒
͑FCI͒
͑FCI͒
5 s,2p
͓ Ј
5 s,2p ϩsp
͓ Ј
5 s,2p ϩspϩd
͓ Ј
5 s,2p ϩspϩdd
͓ Ј
Ϫ37.738 055
Ϫ37.739 280
Ϫ37.778 178
Ϫ37.784 688
0.001 432 1
0.016 308 2
0.035 938 2
0.041 574 9
͔
͔
͔
͔
0.0012
5s,3p
5s,3p ϩsp
5s,3p ϩspϩd
5s,3p ϩspϩdd
Ϫ37.744 352
Ϫ37.745 363
Ϫ37.784 815
0.069 208 7
0.074 916 9
0.093 504 7
¯
Ϫ0.023 558 8
Ϫ0.004 189 9
0.023 438 0
¯
0.0331
0.0340
͓
͓
͓
͓
͔
͔
͔
͔
a
¯
For each basis set studied, adding diffuse sp functions
and then one or two d functions results in a continuous in-
crease of the spin density at the nucleus. All sequences ap-
pear to converge to values considerably different from the
6s,3p
6s,3p ϩsp
6s,3p ϩspϩd
6s,3p ϩspϩdd
Ϫ37.746 956
Ϫ37.747 824
Ϫ37.787 278
¯
͓
͓
͓
͓
͔
͔
͔
͔
0.0018
0.0198
0.0212
experimental spin densities, even in the case of the 6s,3p
͓
͔
ϩspϩdd basis set specifically designed for spin density cal-
aToo large to permit FCI calculations.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.76.6 On: Tue, 25 Nov 2014 18:25:28

J. Chem. Phys., Vol. 113, No. 10, 8 September 2000
Spin density of B and C atoms
4241
TABLE IV. Convergence of CASSCF spin density with increasing size of
active space: from small CASSCF to full CI ͑all electrons included͒.
IV. CONCLUSIONS
It appears that the basis sets studied in this work are
largely inadequate for calculation of spin densities at the nu-
clei, at least in the case of boron and carbon atoms. The
results obtained with these basis sets using the spin polariza-
tion model and some other methods disagree with the FCI
treatment. Since the FCI wave function is exact for a given
basis, it is not clear that the spin polarization model will be
generally reliable. It is unlikely that a method that does not
account at least to some extent for excitations to all orbitals
in the complete space of basis functions would be successful
in correctly describing atomic spin densities at the nuclei.
Including excitations within only a subspace of the complete
space is not sufficient, even if all such excitations are in-
cluded. This is illustrated by very slow convergence of
CASSCF results with increasing size of active space. The
FCI results reported in this work can be used as benchmarks
to test various approaches to spin density calculation.
E͑CASSCF)
Ϫ24.570 213
Spin density
0.045 184 5
MCSCF͑9͒
1s2s2p3s3p
MCSCF͑13͒
1s2s2p3s3p4p4s
MCSCF͑14͒
1s2s2p3s3p4p4s5s
MCSCF͑15͒
Ϫ24.583 678
Ϫ24.584 180
Ϫ24.604 076
Ϫ24.604 223
Ϫ24.604 663
Ϫ24.604 715
Ϫ0.013 254 6
Ϫ0.009 920 8
Ϫ0.005 653 9
Ϫ0.005 732 2
Ϫ0.001 120 7
Ϫ0.000 938 1
1s2s2p 3s3p4s3d
͓
͔
MCSCF͑18͒
1s2s2p3s3p4p4s3d
MCSCF͑19͒
1s2s2p3s3p4p4s3d5s
FCIϭMCSCF͑22͒
1s2s2p3s3p4p4s3d5s6p
with these basis sets must be considered suspect. Since these
basis sets approach the accuracy of a complete basis set
within the spin polarization model,18 the current results illus-
trate the inadequacy of this level of theory.
One alternative to FCI is a smaller CI expansion. For a
nitrogen atom, which is a much simpler system with respect
to spin density calculations than boron or carbon,19,27 it was
demonstrated that at least quadruple excitations are required
if a single reference wave function is used. A multireference
CI wave function based on a relatively small CASSCF ref-
erence function predicts a N spin density that is very close to
FCI.13,28 There is a number of studies of the relative roles of
single, double, triple, and higher excitations available in the
literature.7,8 In this work we take a different approach and
investigate the convergence of spin density with increasing
size of the CASSCF active space. In other words, instead of
including a limited number of excitations within the space
spanned by all basis functions, we include all excitations
within a smaller space, and then gradually increase the size
of that active space, up to FCI. It should be emphasized that
we use a complete active space wave function, which in-
cludes all excitations within the active space. This is quite
different from the spin polarization MCSCF wave function,
which includes only a limited number of spin polarization
configurations. The results of these CASSCF-based calcula-
1 J. A. Pople, D. L. Beveridge, and P. A. Dobosh, J. Am. Chem. Soc. 90,
4201 ͑1968͒.
2 H. Sekino and R. J. Bartlett, J. Chem. Phys. 82, 4225 ͑1985͒.
3 I. Carmichael, J. Phys. Chem. 95, 108 ͑1991͒.
4 J. Harriman, J. Chem. Phys. 40, 2827 ͑1964͒.
5 A. T. Amos and L. C. Snyder, J. Chem. Phys. 41, 1773 ͑1964͒.
6 D. M. Chipman, Theor. Chim. Acta 82, 93 ͑1992͒.
7 B. Engels, S. D. Peyerimhoff, and E. R. Davidson, Mol. Phys. 62, 109
͑1987͒.
8 B. Engels, L. Eriksson, and S. Lunell, Adv. Quantum Chem. 27, 297
͑1996͒.
9 D. M. Chipman, Phys. Rev. A 39, 475 ͑1989͒.
10 T. Kato, Commun. Pure Appl. Math. 10, 151 ͑1957͒.
11 D. Feller and E. R. Davidson, J. Chem. Phys. 80, 1006 ͑1984͒.
12 H. Nakatsuji and M. Izawa, J. Chem. Phys. 91, 6205 ͑1989͒.
13 C. W. Bauschlicher, S. R. Langhoff, H. Partridge, and D. P. Chong, J.
Chem. Phys. 89, 2985 ͑1988͒.
14 J. Hiller, J. Sucher, and G. Feinberg, Phys. Rev. A 18, 2399 ͑1978͒.
15 J. E. Harriman, Int. J. Quantum Chem. 17, 689 ͑1980͒.
16 V. A. Rassolov and D. M. Chipman, J. Chem. Phys. 105, 1470 ͑1996͒.
17 D. Feller and E. R. Davidson, Theor. Chim. Acta 68, 57 ͑1985͒.
18 D. M. Chipman, Theor. Chim. Acta 76, 73 ͑1989͒.
19 I. Carmichael, J. Phys. Chem. A 101, 4633 ͑1997͒.
20 M. W. Schmidt, K. K. Baldridge, J. A. Boatz, S. T. Elbert, M. S. Gordon,
J. H. Jensen, S. Koseki, N. Matsunaga, K. A. Nguyen, S. J. Su, T. L.
Windus, M. Dupuis, and J. A. Montgomery, J. Comput. Chem. 14, 1347
͑1993͒.
tions for the 5 s,2p ϩspϩd ͑d exponent 0.40͒ basis set
͓
͔
Ј
are given in Table IV.
As can be seen from the table, the convergence of the
spin density is very slow; the CASSCF value for an active
space which is only three orbitals short of the full space is
still 20% off the FCI value. It appears that at least some
excitations to orbitals of very high orbital momentum should
be included in the wave function,29 and therefore we are
forced to conclude that CASSCF calculations probably can-
not be used as a substitute for FCI.
21 T. H. Dunning, J. Chem. Phys. 53, 2823 ͑1970͒.
22 S. Huzinaga, J. Chem. Phys. 42, 1293 ͑1965͒.
23 J. Ivanic and K. Ruedenberg ͑unpublished͒.
24 J. S. M. Harvey, L. Evans, and H. Lew, Can. J. Phys. 50, 1719 ͑1972͒.
25 I. Carmichael, J. Phys. Chem. 95, 108 ͑1991͒.
26 D. M. Chipman, J. Chem. Phys. 91, 5455 ͑1989͒.
27 D. Feller and E. R. Davidson, J. Chem. Phys. 88, 7580 ͑1988͒.
28 C. W. Bauschlicher, J. Chem. Phys. 92, 518 ͑1990͒.
29 I. Carmichael ͑private communication͒.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.76.6 On: Tue, 25 Nov 2014 18:25:28