PHYSICAL REVIEW B

VOLUME 58, NUMBER 7

15 AUGUST 1998-I

Relativistic separable dual-space Gaussian pseudopotentials from H to Rn

C. Hartwigsen, S. Goedecker, and J. Hutter
Max-Planck-Institut fur Festkorperforschung, D-70569 Stuttgart, Germany
~Received 20 March 1998!
We generalize the concept of separable dual-space Gaussian pseudopotentials to the relativistic case. This
allows us to construct this type of pseudopotential for the whole Periodic Table, and we present a complete
table of pseudopotential parameters for all the elements from H to Rn. The relativistic version of this pseudopotential retains all the advantages of its nonrelativistic version. It is separable by construction, it is optimal for
integration on a real-space grid, it is highly accurate, and, due to its analytic form, it can be specified by a very
small number of parameters. The accuracy of the pseudopotential is illustrated by an extensive series of
molecular calculations. @S0163-1829~98!05628-8#

I. INTRODUCTION

Pseudopotentials are a well-established tool in ab initio

structure calculations of molecules and solids. First, by replacing the atom by a pseudoatom, the number of orbitals
which have to be calculated is reduced, and, second, the size
of the basis set can substantially be reduced because the
pseudo-wave-functions are smoother than their all-electron
counterparts. In addition, relativistic effects which are relevant for heavier elements can be included in the pseudopotential construction, so that a nonrelativistic calculation can
reproduce these.
In 1982, Bachelet, Hamann, and Schluter1 published a list
of pseudopotentials for all elements up to Pu, that has found
widespread application. There have been many attempts
since to improve the pseudopotential transferability and their
numerical efficiency. One major advance was the introduction of a separable form by Kleinmann and Bylander,2 that
significantly reduces the computational effort for the calculation of the nonlocal part, especially when using a planewave basis set. Gonze, Stumpf, and Scheffler3 investigated
the Kleinmann-Bylander form carefully and computed a list4
of pseudopotentials for many elements up to Xe. Goedecker,
Teter, and Hutter proposed a dual-space Gaussian-type
pseudopotential which is separable and satisfies an optimality criterion for the real-space integration of the nonlocal
part. For large systems there is only a quadratic scaling with
respect to the system size if the integration of the nonlocal
part is performed on a real-space grid, compared to a cubic
scaling if a Fourier space integration is used.6 In contrast to
most other pseudopotential construction methods, the authors
of Ref. 5 also included unoccupied orbitals in their method,
thereby generating highly transferable pseudopotentials.
They gave the nonrelativistic pseudopotential parameters for
the first two rows of the periodic system, and showed that
their pseudopotentials give highly accurate results in molecular calculations. They obtained results which are much closer
to the quasi-exact all-electron LDA ~Refs. 7 and 8! ~localdensity approximation! value than what is obtained in allelectron calculations with a standard Gaussian 6-31G * basis
sets. In other words, the errors due to the pseudopotential
approximation were much smaller than the errors in an allelectron calculation introduced by incomplete basis sets.
0163-1829/98/58~7!/3641~22!/$15.00

PRB 58

In this paper we give the parameters of dual-space Gaussian pseudopotentials for all elements from H to Rn. In contrast to Ref. 5, all pseudopotentials are now generated on the
basis of a fully relativistic all-electron calculation, i.e., by
solving the two-component Dirac equation. The generalization of the norm-conservation property to the relativistic case
proposed by Bachelet and Schluter9 is used for the construction. We also introduced some slight modifications of the
analytic form of the pseudopotential. The parameters are
given in the context of the local-density approximation. Even
though the parameters change only slightly if the pseudopotential is constructed within the framework of a generalized
gradient approximation10,11 ~GGA! functional, we found that
molecular properties are less accurately described if LDA
pseudopotentials are inserted in a molecular calculation using GGAs. Since it is not possible to construct pseudopotential tables for all current GGA schemes, a program that
can construct pseudopotentials for the most common GGAs
can be obtained from the authors.
II. FORM OF THE PSEUDOPOTENTIAL

The local part of the pseudopotential is given by

V loc~ r ! 5

2Z ion
erf
r

S D F S DG

3 C 1 1C 2

A2r loc

1exp 2

1 r
2 r loc

S D S D S DG
2

r loc

1C 3

r loc

1C 4

r loc

~1!

where erf denotes the error function. Z ion is the ionic charge
of the atomic core, i.e., the total charge minus the charge of
the valence electrons. The nonlocal contribution V l (r,r8 ) to
the pseudopotential is a sum of separable terms
3

V l ~ r,r8 ! 5

1l

(( (

i51 j51 m52l

* ~ r8 ! ,
Y l,m ~ r! p li ~ r ! h li, j p lj ~ r 8 ! Y l,m
~2!

where Y l,m are the spherical harmonics, and l the angular

momentum quantum number. The projectors p li (r) are Gaussians of the form
3641

1998 The American Physical Society

3642

C. HARTWIGSEN, S. GOEDECKER, AND J. HUTTER

A2r l12 ~ i21 ! exp

AS

p li ~ r ! 5
rl

r2
2r l 2

4i21
G l1
2

l1 ~ 4i21 ! /2

32

~3!

p l51
3 ~ g !5

PRB 58

r 1 5 5/4
p g ~ 35214g 2 r 1 2 1g 4 r 1 4 !
1155
3 AV exp@ 21 ~ gr 1 ! 2 #

where G denotes the gamma function. The projectors satisfy

the normalization condition

8
p l52
1 ~ g !5

~4!

p li ~ r ! p li ~ r ! r 2 dr51.

16
It is a special property of our pseudopotential that it also
has an analytical form if expressed in reciprocal space. The
Fourier transform of the pseudopotential is given by
V loc~ g ! 52

4 p Z ion
Vg

e 2 ~ gr loc! /21 A8 p 3
2

1C 4 @ 1052105~ gr loc! 121~ gr loc! 2 ~ gr loc! # %

2

~5!
for the local part, and
3

V l ~ g,g8 ! 5 ~ 21 !

1l

(( (

i51 j51 m52l

Y l,m ~ g! p li ~ g ! h li, j p lj ~ g 8 !

* ~ g8 !
3Y l,m

~6!

for the nonlocal part. The Fourier transform of the projectors

p li (r) can be calculated analytically, and for the relevant
cases one obtains
p l50
1 ~ g !5

p l50
2 ~ g !5

16
p l50
3 ~ g !5

4 A2r 0 3 p 5/4

AV

AV

~7!

2r 0 3 5/4
8
p ~ 32g 2 r 0 2 !
15
exp@ 21 ~ gr 0 ! 2 #

~8!

2r 0 3 5/4
p ~ 15210g 2 r 0 2 1g 4 r 0 4 !
105
3 AV exp@ 21 ~ gr 0 ! 2 #

8
p l51
1 ~ g !5

16
p l51
2 ~ g !5

exp@ 21 ~ gr 0 ! 2 #

AV

r 1 5 5/4
p g
3

exp@ 21 ~ gr 1 ! 2 #

AV

exp@ ~ gr 1 ! #
1
2

~9!

~10!

r 1 5 5/4
p g ~ 52g 2 r 1 2 !
105

~11!

2r 2 7 5/4 2
p g
15

exp@ 21 ~ gr 2 ! 2 #

2r 2 7 5/4 2
p g ~ 72g 2 r 2 2 !
105

AV

r 3 9 5/4 3
p g
105

exp@ 21 ~ gr 3 ! 2 #

~12!

~13!

3 AV exp@ 21 ~ gr 2 ! 2 #

p l53
1 ~ g !5

~14!

~15!

In both real and Fourier space, the projectors have the

form of a Gaussian multiplied by a polynomial. Due to this
property the dual-space Gaussian pseudopotential is the optimal compromise between good convergence properties in
real and Fourier space. The multiplication of the wave function with the nonlocal pseudopotential arising from an atom
can be limited to a small region around the atom as the radial
projectors p li (r) asymptotically tend to zero outside the covalent radius of the atom. In addition, a very dense integration grid is not required, as the nonlocal pseudopotential is
reasonably smooth because of its good decay properties in
Fourier space.
The use of this form for the pseudopotential is also very
advantageous if atom-centered basis functions are used instead of plane waves. Because of the separability all threecenter integrals are products of two-center integrals, and so
only these two-center integrals have to be calculated. If
atom-centered Gaussian-type orbitals are used, these twocenter integrals can easily be evaluated analytically.
In the relativistic case the spin orbit coupling splits up all
orbitals with l.0 into spin-up and spin-down orbitals with
an overall angular momentum j5l61/2. So for each
angular-momentum l.0, one spin-up orbital and one spindown orbital with different wave functions and pseudopotentials exist. Following Bachelet and Schluter,9 we give a
weighted average and difference potential of these potentials.
The average pseudopotential is conveniently defined as
V l ~ r,r8 ! 5

1
@ lV l21/2~ r,r8 ! 1 ~ l11 ! V l11/2~ r,r8 !#
2l11
~16!

weighted by the different j degeneracies of the l6 21 orbitals.

The difference potential describes the spin-orbit coupling,
and is defined as
DV SO
l ~ r,r8 ! 5

AV

16

r loc3 2 ~ gr ! 2 /2
loc
e
3$C1
V

1C 2 ~ 32g 2 r loc2 ! 1C 3 @ 15210~ gr loc! 2 1 ~ gr loc! 4 #

p l52
2 ~ g !5

2
@V
~ r,r8 ! 2V l21/2~ r,r8 !# .
2l11 l11/2

The total pseudopotential is then given by

~17!

PRB 58

RELATIVISTIC SEPARABLE DUAL-SPACE GAUSSIAN . . .

V ~ r,r8 ! 5V loc~ r ! d ~ r2r8 ! 1

(l V l~ r,r8 ! 1DV SO
l ~ r,r8 ! LS,
~18!

where V loc(r) and V l (r,r8 ) are now scalar relativistic quantities but with the same form @Eqs. ~1! and ~2!# as the nonrelativistic case. To express DV SO
l (r,r8 ), we also use Eq. ~2!
just replacing the h li, j by different parameters k li, j , i.e.,
3

DV SO
l ~ r,r8 ! 5

1l

(( (

i51 j51 m52l

* ~ r8 ! .
Y l,m ~ r! p li ~ r ! k li, j p lj ~ r 8 ! Y l,m
~19!
DV SO
l (r,r8 )

Neglecting the contributions from

in Eq. ~18!
gives an average potential that contains all scalar parts of the
relativistic pseudopotential, whereas the total potential contains relativistic effects up to order a 2 .
III. DETERMINATION OF THE PSEUDOPOTENTIAL
PARAMETERS

The parameters of the pseudopotentials were found by

minimizing the differences between the eigenvalues and the
charges within an atomic sphere of the all-electron atom and
the pseudoatom. In most cases the radius of the atomic
sphere was taken to be the covalent radius of the atom. For
consistency we always performed a fully relativistic calculation for the all-electron atom, even when relativistic effects
are negligible. The exchange and correlation energy was calculated with the functional given in Ref. 5 This functional
reproduces very well the Perdew-Wang12 functional, but is
much easier to compute. To ensure transferability of the
pseudopotential, we also considered the next two or three
higher unoccupied orbitals for each angular momentum, and
the lowest orbital of the next two unoccupied angular momenta. However, in our calculations we never exceeded
l max53. The atom was put in an external parabolic confining
potential to have well-defined unoccupied orbitals. The
pseudopotential parameters given in Table I typically reproduce the eigenvalues of the occupied orbitals, with an error
of less than 10 25 a.u., and for unoccupied orbitals to within
10 23 a.u. Pseudopotentials containing semicore electrons
~Sec. IV! are an exception, as the errors for the semicore
orbitals are usually larger than for the valence orbitals. In
many cases we found it unnecessary to include all unoccupied orbitals in our fitting procedure. For most cases the
inclusion of only the first unoccupied orbital for an angular
momentum results in comparably good results for the following higher unoccupied orbitals. Nevertheless we always
checked all-electron eigenvalues, and pseudoeigenvalues and
charges of the unoccupied orbitals to verify this.
It was already discussed in Ref. 5 that our fitting procedure yields pseudopotentials that obey the normconservation13 condition, and meets several additional
conditions,1416 such as extended norm conservation and
hardness, thereby leading to pseudopotentials of a very high
quality. In Table II we give the transferability errors for several excited and ionized states for some elements.
The construction of our pseudopotential differs somehow
from the usual method because we fit the pseudopotential
parameters that give the best overall representation for the

3643

eigenvalues and charges of several orbitals, rather than producing pseudo-wave-functions that are identical to their allelectron counterparts beyond some cutoff radius. Therefore,
the wave functions of the pseudoatom and the all-electron
atom approach each other only exponentially. Nevertheless
the difference is very small beyond the core region, as can
been seen from Fig. 1. A second consequence of our fitting
procedure is that, contrary to most other pseudopotential
construction methods, the local part of our pseudopotential
does not correspond to a certain wave function.
It is a special feature of our method that we fit our parameters directly against the all-electron eigenvalues and
charges, rather than fitting analytical or numerical potentials
that reproduce pseudo-wave-functions, which themselves are
constructed from their all-electron counterparts. Therefore,
our pseudopotentials require significantly fewer parameters
than those tabulated by Bachelet, Hamann, and Schluter.1
During the generation of our pseudopotentials we found that
there is in general no single minimal parameter set that gives
the best overall pseudopotential for one atom. This finding is
different to the former study of Ref. 5, where only the first
two rows of the Periodic Table were considered. We always
tried to use a minimum parameter set, which is sufficient to
reproducing the desired accuracy of the fitted eigenvalues
and charges. Identical parameter sets were used for comparable elements, i.e., the same parameter set was used for all
3d or 4d elements, respectively. The fitting of the pseudopotential parameters is numerically demanding, as many local minima exist, so that sometimes up to some 105 pseudopotential evaluations are necessary until one finds good
parameter values. We used a slow Simplex-Downhill
algorithm17 for the optimization, that proved to be much
more robust than more sophisticated methods. The pseudopotential parameter r loc was set by hand, except for the first
row, because this parameter is not easily accessible by our
fitting procedure. For many elements we generated and
tested pseudopotentials with different values of r loc . After
selection of the optimum pseudopotentials, the r loc values for
the elements in between were interpolated so that no discontinuities occur.
IV. SEMICORE ELECTRONS

For many atoms there is no unambiguous separation of

the electronic system into a well-isolated core and valence
shell. For example, it is well known the n-1 p levels of the
heavy alkali atoms are relatively shallow in energy and extended in space. The 3d wave functions of the 3d elements
are strongly localized, so that there is a significant overlap
with the 3(s,p) wave functions, although the latter are much
lower in energy than the 3d and 4(s,p) valence wave functions. The same is true for the 4d and 5d elements. Analogously, the 4 f wave functions of the 4 f elements are so
localized that they overlap with the 5s and 5 p wave functions. In all these cases, where a non-negligible overlap between valence and core wave functions exists, the frozencore approximation underlying the construction of all
pseudopotentials is not well satisfied. One way to overcome
this problem is the inclusion of a nonlinear core correction18
that considers the contribution of the core charge to the
exchange-correlation potential. Another, more straightfor-

C. HARTWIGSEN, S. GOEDECKER, AND J. HUTTER

3644

PRB 58

TABLE I. LDA pseudopotential parameters in atomic units. The meaning of the entries is given in the
text.
H
He
Li

1
2
1

0.200000
0.200000
0.787553
0.666375
1.079306

Lisc
Be

3
2

0.400000
0.739009
0.528797
0.658153

Besc
B

4
3

0.325000
0.433930
0.373843
0.360393

0.348830
0.304553
0.232677

0.289179
0.256605
0.270134

0.247621
0.221786
0.256829

0.218525
0.195567
0.174268

Ne

0.190000
0.179488
0.214913

Na

0.885509
0.661104
0.857119

Nasc

0.246318
0.141251
0.139668

Mg

0.651812
0.556478
0.677569

Mgsc

10

0.210950
0.141547
0.105469

Al

0.450000
0.460104
0.536744

24.180237
29.112023
21.892612
1.858811
20.005895
0.000019
214.034868
22.592951
3.061666
0.092462
0.000129
224.015041
25.578642
6.233928
0.000000
0.000878
28.513771
9.522842
0.000000
0.004104
212.234820
13.552243
0.000000
0.003131
216.580318
18.266917
0.000000
0.004476
221.307361
23.584942
0.000000
0.015106
227.692852
28.506098
20.000090
0.010336
21.238867
1.847271
0.471133
0.002623
27.545593
36.556987
210.392083
0.038386
22.864297
2.970957
1.049881
0.005152
219.419008
40.316626
210.891113
0.096277
28.491351
5.088340
2.193438
0.006154

0.725075
1.698368
0.286060

9.553476
0.354839

21.766488

0.084370

17.204014
0.804251

23.326390

0.165419

1.228432

1.766407

2.395701

3.072869

4.005906
21.076245

0.582004

1.125997

1.329941

2.871331

2.679700
0.000000
0.003947

PRB 58

RELATIVISTIC SEPARABLE DUAL-SPACE GAUSSIAN . . .

3645

TABLE I. ~Continued!.
Si

Cl

Ar

0.440000
0.422738
0.484278
0.430000
0.389803
0.440796
0.420000
0.361757
0.405285
0.410000
0.338208
0.376137
0.400000
0.317381
0.351619
0.950000
0.955364
1.086411
0.720606

Ksc

0.400000
0.294826
0.322359

Ca

0.800000
0.669737
0.946474
0.526550

Casc

10

0.390000
0.281909
0.310345
0.904330

Sc

0.750000
0.597079
0.847994
0.454653

Scsc

11

0.385000
0.359707
0.243234
0.252945

Ti

0.720000
0.528411
0.791146

27.336103
5.906928
2.727013
0.000373
26.654220
6.842136
3.282606
0.002544
26.554492
7.905303
3.866579
0.005372
26.864754
9.062240
4.465876
0.010020
27.100000
10.249487
4.978801
0.016395
0.914612
0.315462
20.004343
21.529514
0.000354
24.989348
11.238705
5.256702
0.015795
1.645014
0.585479
20.003362
23.032321
0.000814
24.928146
12.352340
5.722423
0.021701
0.016806
0.000371
1.835768
0.784127
20.008463
23.859241
0.001430
7.425036
6.119585
6.376618
0.083053
28.020892
0.004812
1.866613
0.967916

3.258196
0.000000
0.014437
3.856693
0.000000
0.017895
4.471698
0.000000
0.022062
5.065682
0.000000
0.025784
5.602516
0.000000
0.029171
0.287551
0.068194
0.010261

20.300224

20.756048
7.067779
0.938947
0.048923
1.523491
0.126329
0.012779

0.295996

21.232854
7.657455
0.923591
0.056661

1.734309
0.246733
0.020913

1.418483

20.489852
22.563453
26.016415
0.093017

1.440233
0.260687

3.658172

3646

PRB 58

C. HARTWIGSEN, S. GOEDECKER, AND J. HUTTER

TABLE I. ~Continued!.

0.408712
Tisc

12

0.380000
0.334235
0.242416
0.242947

0.690000
0.514704
0.743504
0.374890

Vsc

13

0.375000
0.326651
0.246407
0.240792

Cr

0.660000
0.498578
0.719768
0.354341

Crsc

14

0.370000
0.306011
0.241090
0.219577

Mn

0.640000
0.481246
0.669304
0.327763

Mnsc

15

0.365000
0.280753
0.254536
0.221422

Fe

0.610000
0.454482
0.638903
0.308732

Fesc

16

0.360000
0.269268
0.247686
0.223021

20.009333
24.826456
0.002010
7.548789
6.925740
5.079086
0.122395
29.125896
0.005822
2.208670
1.115751
20.010973
25.841633
0.002717
4.941291
7.659390
4.256230
0.156408
28.828518
0.006548
2.400756
1.145557
20.013176
26.615878
0.003514
5.113362
8.617835
3.161588
0.169781
211.157868
0.009007
2.799031
1.368776
20.013685
27.995418
0.004536
6.748683
9.379532
0.371176
0.164188
212.115385
0.009590
3.016640
1.499642
20.014909
29.145354
0.005722
5.392507
10.193723
0.145613
0.201450
212.026941
0.010322

0.025291

20.588377
23.142005
26.284281
0.057447

1.896763
0.286649
0.030816

3.076377

20.096443
23.892229
25.941212
0.008030

2.072337
0.278236
0.035625

2.952179

20.646819
24.137695
25.032906
0.000411

2.486101
0.316763
0.042938

2.565630

20.576569
25.575280
24.229057
20.039396

2.583038
0.326874
0.049793

20.030066
26.834982
25.234954
20.075829

3.257635

RELATIVISTIC SEPARABLE DUAL-SPACE GAUSSIAN . . .

PRB 58

TABLE I. ~Continued!.
Co

0.580000
0.440457
0.610048
0.291661

Cosc

17

0.355000
0.259140
0.251425
0.221665

Ni

10

0.560000
0.425399
0.584081
0.278113

Nisc

18

0.350000
0.245105
0.234741
0.214950

Cu

0.580000
0.843283
1.089543
1.291602

Cusc

11

0.530000
0.423734
0.572177
0.266143

Zn

0.570000
0.640712
0.967605
1.330352

Znsc

12

0.510000
0.400866
0.539618
0.252151

Ga

0.560000
0.610791
0.704596
0.982580

Gasc

13

0.490000
0.384713
0.586130

3.334978
1.634005
20.017521
210.358800
0.007137
3.418391
11.195226
20.551464
0.207915
212.075354
0.011475
3.619651
1.742220
20.020384
211.608428
0.008708
3.610311
12.161131
20.820624
0.269572
213.395062
0.013538

2.873150
0.356083
0.058766

3.091028

0.482078
27.845825
24.639237
20.108249

3.088965
0.386341
0.068770

3.058598

0.449638
29.078929
26.029071
20.143442

0.975787
0.024580
0.010792
20.065292
20.000730

20.822070
20.249001
20.006734

20.133237

3.888050
1.751272
20.024067
212.676957
0.010489

3.276584
0.374943
0.076481

2.290091

2.088557
0.163546
0.012139
0.010486
0.000225

20.218270
20.227086
20.004876

20.941317

4.278710
2.023884
20.025759
214.338368
0.012767

3.627342
0.431742
0.090915

2.849567

2.369325
0.746305
0.029607
0.075437
0.001486

20.249015
20.513132
20.000873

20.551796

4.831779
1.940527

4.238168
20.299738

2.833238

3647

3648

PRB 58

C. HARTWIGSEN, S. GOEDECKER, AND J. HUTTER

TABLE I. ~Continued!.

0.240803
Ge

0.540000
0.493743
0.601064
0.788369

As

0.520000
0.456400
0.550562
0.685283

Se

0.510000
0.432531
0.472473
0.613420

Br

0.500000
0.428207
0.455323
0.557847

Kr

0.500000
0.410759
0.430256
0.517120

Rb

1.096207
0.955699
1.156681
0.664323

Rbsc

0.490000
0.282301
0.301886
0.514895

Sr

1.010000
0.837564
1.174178
0.743175

Srsc

10

0.480000
0.275441
0.302243
0.502045

0.026949
215.795675
0.015503

0.031400

3.826891
1.362518
0.043981
0.191205
0.002918

1.100231
20.627370
0.009802

21.344218

4.560761
1.812247
0.052466
0.312373
0.004273

1.692389
20.646727
0.020562

21.373804

5.145131
2.858806
0.062196
0.434829
0.005784

2.052009
20.590671
0.064907

21.369203

5.398837
3.108823
0.074007
0.555903
0.007144

1.820292
20.632202
0.068787

21.323974

5.911194
3.524357
0.087011
0.629228
0.009267
0.847333
0.887460
0.461734
20.043443
21.362938
0.003708
4.504151
9.536329
2.209592
20.867379
0.449376
0.008685
0.684749
1.200395
0.439983
0.004022
21.386990
0.002846
5.571455
9.995135
3.169126
20.576265
0.43728
0.008991

1.967372
20.691198
0.086008

21.458069

20.748120
0.903088
0.336113
0.057876

20.006750

20.741018
9.486634
5.475249
1.237532

20.062125
0.926675
0.018267
0.022207

21.079963
9.336679
4.049231
0.990062

20.315858

RELATIVISTIC SEPARABLE DUAL-SPACE GAUSSIAN . . .

PRB 58

3649

TABLE I. ~Continued!.
Y

0.900000
0.782457
0.949864
0.653851

Ysc

11

0.475000
0.414360
0.406442
0.513304

Zr

0.750000
0.649998
0.874408
0.630668

Zrsc

12

0.470000
0.396540
0.388558
0.520496

Nb

0.724000
0.699708
0.846672
0.516072

Nbsc

13

0.460000
0.393708
0.403626
0.513644

Mo

0.699000
0.678126
0.800771
0.453384

Mosc

14

0.430000
0.376255
0.361734
0.525828

Tc

0.673000
0.677612
0.784275
0.519890

20.343891
1.520655
0.780950
20.043336
21.256930
0.009198
13.217914
2.522621
20.569552
0.251526
21.571003
0.012450
20.782611
1.739877
1.018294
20.057486
21.173911
0.009380
15.782342
2.571767
20.794123
0.301247
21.548402
0.013187
4.021058
1.532651
0.609675
20.080816
22.696830
0.025653
13.505394
3.222025
20.822037
0.246821
21.489848
0.014064
7.995868
1.289607
0.301412
20.104124
22.809708
0.068972
16.237452
3.362426
20.379571
0.378681
21.543211
0.014460
13.315381
0.819218
0.028673
20.080760
25.984224
0.026025

1.484368
0.368739
0.079989
20.075368
20.011657
1.353178
24.363769
23.020044
20.077005
0.627616
20.007507

20.189013

2.388208
0.528223
0.104495
0.212179
20.011973
0.433648
24.714509
23.172114
20.098654
0.826127
20.010136

1.205349

1.428264
0.596788
0.125243
21.694967
20.031541
0.752434
24.599342
22.247366
20.086659
0.823817
20.012055
0.998113
0.741615
0.153906
26.820946
20.075591
1.496536
25.289276
24.067713
20.124561
1.074388
20.014769
0.348460
0.658363
0.140668
0.721822
20.041776

3650

C. HARTWIGSEN, S. GOEDECKER, AND J. HUTTER

TABLE I. ~Continued!.
Tcsc

15

0.430000
0.369721
0.357772
0.510487

Ru

0.647214
0.625656
0.746425
0.440358

Rusc

16

0.430000
0.364084
0.364053
0.495850

Rh

0.621429
0.598079
0.709586
0.369207

Rhsc

17

0.420000
0.350052
0.350253
0.496950

Pd

10

0.596000
0.582204
0.688787
0.442835

Pdsc

18

0.410000
0.342151
0.343111
0.494916

Ag

0.650000
1.012705
1.235842
1.016159

Agsc

11

0.570000
0.498900
0.630009
0.387660

Cd

0.625000
0.828465
0.972873

14.910011
3.917408
20.270000
0.340791
21.586709
0.015790
8.687723
1.637866
0.639012
20.095454
24.883365
0.046652
13.582571
4.480632
20.320372
0.337322
21.597870
0.017773
5.397962
2.242111
1.155278
20.114357
21.053058
0.135036
15.225012
4.715292
20.504694
0.389629
21.685594
0.018875
5.209665
2.411076
1.227253
20.136909
24.377131
0.033797
15.720259
5.177686
20.372561
0.406984
21.608273
0.020374
22.376061
0.897931
0.130081
0.019692
20.038842
0.009455
1.017053
2.990284
1.813968
20.201869
23.420076
0.045193
21.796838
1.485292
0.469208

1.046381
25.268399
23.737771
20.065948
1.132307
20.016485
1.329335
0.650376
0.164257
23.063746
20.061808
0.596227
25.268679
23.059714
20.103467
1.165495
20.019725
2.151597
0.704846
0.193923
210.923960
20.127750
0.415911
25.805525
23.373040
20.122856
1.387707
20.022837
2.318920
0.758021
0.220805
0.413271
20.047798
0.140765
25.852819
23.258728
20.110764
1.446609
20.026732
20.748323
20.277495
20.006821

0.029787

3.912395
1.304450
0.319708
21.019949
20.051565

2.205847

20.424753
20.448111

20.407986

PRB 58

PRB 58

RELATIVISTIC SEPARABLE DUAL-SPACE GAUSSIAN . . .

3651

TABLE I. ~Continued!.

1.240949
Cdsc

12

0.550000
0.491505
0.598565
0.377874

In

0.610000
0.770602
0.858132
1.088691

Insc

13

0.530000
0.474081
0.559819
0.360488

Sn

0.605000
0.663544
0.745865
0.944459

Sb

0.590000
0.597684
0.672122
0.856557

Te

0.575000
0.556456
0.615262
0.805101

0.560000
0.552830
0.562251
0.794325

Xe

0.560000
0.507371
0.541024
0.729821

Cs

1.200000
1.224737
1.280478
1.107522

0.037714
0.065412
0.003109
2.382713
3.207932
1.940150
20.241409
24.190072
0.049329
2.865777
1.256194
0.494459
0.066367
0.129208
0.004448
2.395404
3.554411
2.223664
20.306488
24.566414
0.058765
4.610912
1.648791
0.769355
0.103931
0.225115
0.007066
6.680228
1.951477
0.970313
0.139222
0.300103
0.009432
9.387085
2.046890
1.033478
0.172997
0.317411
0.010809
14.661825
1.338054
0.674496
0.213967
0.224345
0.010180
12.734280
2.236451
1.130043
0.236603
0.280131
0.013362
0.611527
0.244893
20.107627
20.542163
0.003259

20.002009

4.140963
1.515892
0.376830
20.770156
20.058835

1.584234

20.397255
20.380789
20.005563

20.278329

4.754135
2.035278
0.479435
20.773785
20.073414

1.565040

20.141974
20.445070
0.005057

20.576546

0.037537
20.466731
0.023513

20.786631

20.029333
20.481172
0.050641

20.881119

20.834851
20.577787
0.093313

20.467438

20.403551
20.752764
0.108473

21.132507

0.239830
0.227279
0.138132

20.294024

C. HARTWIGSEN, S. GOEDECKER, AND J. HUTTER

3652

PRB 58

TABLE I. ~Continued!.
Cssc

0.540000
0.456821
0.362467
0.761462
0.333507

Ba

1.200000
1.016187
1.249880
0.937158

Basc

10

0.540000
0.514776
0.375190
0.665403
0.304920

Lasc

11

0.535000
0.551775
0.476308
0.626672
0.299310

Cesc

12

0.535000
0.521790
0.470324
0.703593
0.306717

Prsc

13

0.532083
0.526850
0.458897
0.747610
0.300773

Ndsc

14

0.529167
0.503000
0.467013
0.325290
0.294743

Pmsc

15

0.526250
0.489879
0.472260

35.234438
20.282378
22.696733
22.748528
0.183754
0.010841
217.948259
0.004760

23.318070
22.780956
22.290940
1.815069

0.922593
0.447168
20.041017
20.718934
0.005321
24.478653
1.046729
20.202439
20.415083
0.378419
0.017660
218.795208
0.006151
19.909308
1.293272
1.172527
0.524623
0.328377
0.020900
218.269439
0.007193
18.847470
1.321616
0.972641
0.463710
0.074241
0.013265
217.214790
0.007568
18.424739
1.012621
1.117060
0.314280
0.017571
0.010905
217.897119
0.008547
17.815030
1.529110
0.721553
20.214396
20.543240
0.611413
218.520228
0.009598
18.251723
1.308978
0.160512

0.763495
0.151081
0.088800

0.000000
1.204825

20.333465

22.500850
20.977217
0.680331
0.599763

0.000000
0.927065

21.474830
21.121819
20.828810
20.030901

0.029857
0.142077

20.765636
21.700444
21.451337
0.090257

0.000000
0.012566

20.657669
21.717982
21.852109
0.350982

0.000000
20.165254

20.594798
22.153732
21.647499
1.100446

0.000000
20.832670

20.492107
22.507751
21.565305

0.000000

RELATIVISTIC SEPARABLE DUAL-SPACE GAUSSIAN . . .

PRB 58

3653

TABLE I. ~Continued!.

0.473709
0.291527
Smsc

16

0.523333
0.479677
0.490598
0.470840
0.284040

Eusc

17

0.520417
0.469043
0.445907
0.490038
0.278401

Gdsc

18

0.517500
0.462014
0.456953
0.482368
0.273390

Tbsc

19

0.514583
0.448694
0.424220
0.482809
0.268260

Dysc

20

0.511667
0.440590
0.434642
0.467229
0.261670

Hosc

21

0.508750
0.432212
0.420138
0.447131
0.254992

Ersc

22

0.505833
0.419948
0.414455

20.339955
20.429952
0.064044
219.305057
0.010619
17.206792
1.723635
20.082403
20.240300
20.410630
0.063352
219.984292
0.011924
17.373516
1.763638
0.518046
0.252258
20.426120
0.051028
220.946528
0.013267
17.512556
1.551856
20.058347
0.535540
20.562601
0.053128
221.923490
0.014666
17.603616
1.718481
0.562400
20.008725
20.625802
0.051754
222.911697
0.016197
16.994331
1.940320
0.014315
20.124371
20.668924
0.058286
223.922358
0.017938
16.781570
2.052797
0.253295
0.895947
20.742863
0.069483
225.197211
0.019830
17.105293
2.144503
0.054087
1.255135

1.359017

21.145883

20.532803
22.659367
21.111009
1.200867

0.000000
21.054041

20.648468
22.916932
22.135186
0.584324

0.000000
20.480586

20.719534
23.068703
21.697711
0.022024

0.000000
20.054301

20.828080
23.435239
22.781827
1.144007

0.000000
20.860425

20.955298
23.559798
22.046238
1.327404

0.000000
21.092142

21.173514
23.674534
22.355289
20.275149

0.000000
0.205729

21.430953
23.984203
22.294477
20.891960

0.000000
0.749717

C. HARTWIGSEN, S. GOEDECKER, AND J. HUTTER

3654

PRB 58

TABLE I. ~Continued!.
0.418385
0.249126
Tmsc

23

0.502917
0.413373
0.409923
0.392870
0.243917

Ybsc

24

0.500000
0.402309
0.414358
0.444025
0.238298

Lusc

25

0.497000
0.391206
0.393896
0.436518
0.232629

Hfsc

12

0.560000
0.422810
0.472681
0.426388

Ta

0.744000
0.581801
0.770646
0.534370

Tasc

13

0.550000
0.421853
0.461345
0.410994

0.719000
0.582463
0.742307
0.534959

Wsc

14

0.540000
0.418570
0.449555
0.399602

20.999006
0.091862
226.696809
0.021783
17.247293
1.947196
20.094493
2.045370
21.353308
0.119299
228.104159
0.023833
17.357144
2.120771
20.923212
20.349470
20.889967
0.070076
229.932854
0.025718
17.037053
2.184678
20.719819
0.152450
21.173245
0.072130
231.852262
0.028006
5.134801
2.564442
21.025275
0.607504
1.459363
0.222119
3.623116
2.005338
0.518567
20.485635
22.202200
0.086716
4.546236
2.708136
20.724853
0.649992
1.348495
0.205344
4.058450
2.161166
0.600973
20.509800
22.517063
0.075772
4.800251
2.692204
20.702084
0.715480
1.177436
0.205860

21.627697
24.121556
22.224087
22.347237

0.000000
1.902736

21.773916
24.802990
21.678540
2.074295

0.000000
21.814297

21.661610
25.432346
22.723799
1.395416

0.000000
21.238744

0.529191
26.013732
21.872548
20.331637
25.282764
20.121283
3.027036
1.185378
0.695148
21.666675
20.094635
0.779422
25.790959
22.215211
20.336371
25.386947
20.102353
2.741500
1.299943
0.751739
20.789137
20.086193
0.901544
26.022637
22.451680
20.385339
25.553621
20.100046

1.109760
0.000000
20.121021

0.947663
0.000000
20.101322

1.218316
0.000000
20.077093

PRB 58

RELATIVISTIC SEPARABLE DUAL-SPACE GAUSSIAN . . .

TABLE I. ~Continued!.
Re

0.693000
0.509816
0.745839
0.500954

Resc

15

0.530000
0.403252
0.440951
0.390395

Os

0.667000
0.510307
0.717553
0.487586

Ossc

16

0.520000
0.410578
0.422395
0.380252

Ir

0.641000
0.509960
0.684971
0.471745

Irsc

17

0.510000
0.404469
0.411426
0.376428

Pt

10

0.616000
0.520132
0.658976
0.451243

Ptsc

18

0.500000
0.409942
0.398652
0.367964

Au

Ausc

11

0.650000
0.919308
1.140351
0.590000
0.521180
0.630613
0.440706

8.180816
2.269379
0.496693
20.370589
23.689630
0.111557
5.592660
2.760720
20.900546
0.788715
0.875251
0.209737
9.440459
2.402367
0.499523
20.430746
24.142035
0.116941
5.613073
2.785758
20.590006
0.870817
0.880133
0.216440
10.720016
2.445999
0.461792
20.565347
24.545484
0.127199
4.904509
3.243278
20.380574
0.930121
0.754315
0.219517
11.027417
2.447430
0.408453
20.763296
24.552295
0.146912
5.445832
2.994366
20.225181
1.017060
0.632067
0.226472
21.963712
1.539599
0.471229
0.039349
11.604428
2.538614
0.394853
20.960055
24.719070

3.528529
0.925829
0.616765
21.894601
20.131595
0.943957
26.396415
22.511211
20.489984
25.672543
20.102862
3.046706
1.053284
0.701752
21.666100
20.139761
0.921955
26.692130
23.018323
20.412886
25.732892
20.103221
2.811037
1.304726
0.859620
21.635428
20.151950
1.313786
27.315509
23.504403
20.379865
25.875580
20.112145
2.640360
1.647716
1.065883
22.102396
20.169306
1.156382
27.448772
23.776974
20.348213
25.755431
20.114346
21.698123
20.468779
20.497538
0.132970
2.701113
2.057831
1.296571
21.650429

0.868732
0.000000
20.017012

2.247034
0.000000
20.139506

2.956978
0.000000
20.262716

4.243095
0.000000
20.331919

20.792039
20.209758
20.153427

3655

3656

C. HARTWIGSEN, S. GOEDECKER, AND J. HUTTER

TABLE 1. ~Continued!.

Hg

0.640000
0.812108
1.053714
1.100000

Hgsc

12

0.570000
0.521802
0.621648
0.401894

Tl

0.630000
0.754005
0.903742
1.063512

Tlsc

13

0.550000
0.502423
0.572016
0.393185

Pb

0.617500
0.705259
0.846641
0.971939

Bi

0.605000
0.678858
0.798673
0.934683

Po

0.592500
0.647950
0.748947
0.880468

At

0.580000
0.627827
0.709823
0.838365

Rn

0.570000
0.615182
0.676697
0.788337

0.148484
23.296329
1.765041
0.474056
0.092330
0.120638
0.020931
2.134572
3.293920
2.100960
0.084989
21.669886
0.155759
21.235846
1.875766
0.759668
0.168641
0.247614
0.022941
7.301886
3.326560
1.272807
0.012233
23.200652
0.186849
0.753143
1.979927
0.864420
0.207711
0.374967
0.029256
6.679437
1.377634
0.655578
0.305314
0.378476
0.029217
10.411731
1.144203
0.594562
0.396354
0.433232
0.033886
13.520411
0.945557
0.527078
0.480774
0.468948
0.037544
14.629185
0.981832
0.612279
0.549896
0.557746
0.045488

20.169493
20.466127
20.531816
20.001118

4.661001
1.689988
0.072771
22.473265
20.122282
20.303680
20.586721
0.004459

4.341390
2.992206
0.031664
23.008296
20.170651

20.799941

0.000000
0.653348

20.781337

0.000000
1.019164

20.164960
20.540969
0.012948

20.806060

20.513697
20.402932
20.023134

20.471028

20.735851
20.353595
20.031462

20.339386

20.965903
20.318821
20.034954

20.190429

21.038963
20.344122
20.023760

20.120456

PRB 58

RELATIVISTIC SEPARABLE DUAL-SPACE GAUSSIAN . . .

PRB 58

3657

TABLE II. Transferability errors.

Element
C

Si

Ge

Sn

Pb

Ti

Tisc

state

DE ~Ha!

Error ~Ha!

2s 1 2p 3 3d 0
2s 2 2 p 3 3d 0
2s 2 2 p 1 3d 0
2s 2 2 p 2 3d 1
3s 1 3 p 3 3d 0
3s 2 3 p 3 3d 0
3s 2 3 p 1 3d 0
3s 2 3 p 2 3d 1
4s 1 4 p 3 4d 0
4s 2 4 p 3 4d 0
4s 2 4 p 1 4d 0
4s 2 4 p 2 4d 1
5s 1 5 p 3 5d 0
5s 2 5 p 3 5d 0
5s 2 5 p 1 5d 0
5s 2 5 p 2 5d 1
6s 1 6 p 3 6d 0
6s 2 6 p 3 6d 0
6s 2 6 p 1 6d 0
6s 2 6 p 2 6d 1
4s 1 4 p 1 3d 2
4s 2 4 p 0 3d 0
4s 0 4 p 0 3d 0
4s 2 4 p 1 3d 2
4s 1 4 p 1 3d 2
4s 2 4 p 0 3d 0
4s 0 4 p 0 3d 0
4s 2 4 p 1 3d 2

0.3073
0.4294310 21
0.3612
0.3882
0.2529
20.5293310 22
0.2673
0.2462
0.2970
20.1682310 21
0.2681
0.1879
0.2605
20.2482310 21
0.2510
0.1084
0.3181
20.3587310 21
0.2505
0.7216310 21
0.1198
0.1166310 1
0.3258310 1
0.3593310 21
0.1198
0.1166310 1
0.3258310 1
0.3593310 21

20.40310 23
0.11310 23
20.44310 23
0.95310 24
20.20310 23
0.38310 24
20.28310 23
0.43310 24
20.48310 23
0.58310 24
20.23310 23
0.52310 24
20.40310 23
0.46310 24
20.35310 23
0.34310 24
20.35310 23
20.41310 25
20.46310 23
0.33310 24
20.17310 23
20.47310 21
20.10
0.65310 24
20.37310 24
20.31310 22
20.28310 22
0.15310 23

ward, solution is the explicit inclusion of the semicore electrons into the pseudopotential. In this work we decided on
the second method. This ensures that our semicore pseudopotentials still can be used with programs where nonlinear
corrections are not considered. In addition, the explicit inclusion of the semicore electrons ensures that our pseudopotentials still work well for systems where nonlinear core corrections fail. It is unnecessary that the eigenvalues and charge
distribution of the semicore wave functions have the same
accuracy as the valence wave functions of the pseudoatom.
We always tried to generate semicore pseudopotentials with
semicore pseudo-wave-functions that are as smooth as possible, but still yield accurate results for the valence wave
functions. Therefore the error for the eigenvalues of semicore wave functions for our pseudopotentials is within
10 23 10 22 a.u., which is about three orders of magnitude
worse than the typical error for the valence wave functions.
The choice of which electrons are treated as semicore
electrons also depends on the required accuracy. As we were
interested in generating pseudopotentials that can be used
together with plane-wave basis sets within a reasonable computational effort, we tried to include not too many semicore
electrons into our pseudopotentials. Our semicore pseudopotentials for the group Ia and IIa elements, the transition metals of groups IIIbVIIIb, and the lanthanides treat the (n21!
s and the (n21!p electrons as semicore electrons. For the
elements of groups Ib, IIb, and IIIa ~except B and Al!, all

FIG. 1. Relativistic all-electron ~solid line! wave function and

pseudo-wave-function ~dashed line! of the valence electrons of
gold. The difference between them is shown by the dotted line on a
logarithmic scale.

electrons of the completely filled nd shell are treated as

semicore electrons.
For all elements mentioned above semicore wave functions improve the description of highly positive charged ions.
In Table III the transferability error of two Ti pseudopotentials is listed for several states. For most states the calculated
excitation energies are much closer to the all-electron values
for the Ti semicore pseudopotential, including the 3s and 3p
semicore electrons. This is most significant for the
4s 0 4 p 0 3d 0 state, which corresponds to a Ti41 ion. For the
4e pseudopotential the error is 0.1 hartree, but only 0.28
310 22 hartree for the 12e semicore pseudopotential.
Pseudopotentials with semicore wave functions always require higher computational effort. They contain more electrons, and larger basis sets are necessary for a sufficient description of the localized semicore wave functions. In many
applications like molecular structure calculations semicore
pseudopotentials yield converged results with comparably
small basis sets even if the calculated total energy is still far
from its converged value. Therefore the inclusion of semicore electrons does not inevitably require the use of extremely large basis sets. In fact, in our molecular calculations
the highest plane-wave energy cutoff was needed for calculations with the fluorine pseudopotential which has no semicore electrons at all.
In many cases it is not quite clear if semicore electrons
play an important role or not. For most applications the need
to use semicore pseudopotentials depends on the required

C. HARTWIGSEN, S. GOEDECKER, AND J. HUTTER

3658

PRB 58

TABLE III. Comparison of bond lengths of molecules calculated with our pseudopotentials ~PSP! and
obtained with GAUSSIAN 94 or NUMOL ~Ref. 21! ~AE! and the experimental data. All GAUSSIAN 94 calculations
were done with a 6 31111G(3d f ,3pd) basis set if not otherwise mentioned. Bond lengths are given in
bohr.
Molecule
H2
Li2
LiH
Be2
BF3
BH
CH4
C2 H2
N2
NH3
HCN
H2 O
CO
CO2
F2
HF
CH3 F
Na2
NaH
Mg2
MgH
AlH
SiH4
SiO
SiF4
P2
PH3
PN
S3
H2 S
CS
CS2
Cl2
HCl
CH3 Cl
K2
KH
CaH
ScH
ScO
TiO
VO
CrO
MnO
FeO
CoO
NiO
NiH

r~CC!
r~HC!

r~CN!
r~HC!

r~CF!
r~HC!

r~CCl!
r~CH!

PSP

AE

Expt.

Ref.

1.447
5.100 sc , 5.016
3.029 sc , 2.929
4.516 sc , 4.497
2.477
2.363
2.072
2.263
2.028
2.067
1.932
2.169
2.038
1.836
2.127
2.196
2.622
1.764
2.605
2.082
5.76
3.534 sc , 3.381
6.31
3.318 sc , 3.279
3.140
2.808
2.827
2.922
3.547
2.696
2.789
3.595
2.553
2.882
2.918
3.725
2.436
3.330
2.072
6.936
2.200 sc , 2.969
3.706 sc , 3.187
3.304 sc , 3.182
3.116 sc
3.018 sc , 2.840
2.943 sc , 2.845
2.955 sc , 2.903
2.996 sc , 2.991
2.958
2.932
3.027
2.726

1.447
5.120
3.029
4.510
2.469
2.371
2.072
2.270
2.030
2.070
1.932
2.174
2.039
1.834
2.130
2.195
2.613
1.760
2.592
2.082
5.64
3.529
6.41
3.313
3.174
2.820
2.861
2.952
3.542
2.704
2.808
3.609a
2.558
2.897
2.930
3.742
2.440
3.331
2.072
7.358a
2.303a
4.126a
3.300b
3.109b
3.020b
2.959b
2.997b
3.032b
2.994b
3.037b
3.035b

1.401
5.051
3.015
4.63
2.470
2.373
2.052
2.274
2.005
2.074
1.912
2.179
2.013
1.809
2.132
2.192
2.668
1.733
2.612
2.069
5.818
3.566
7.351
3.271
3.169
2.795
2.853
2.936
3.578
2.671
2.871

21
21
21
22
23
21
24
23
23
21
23
25
25
23
21
23
21
21
23
23
21
21
21
21
21
26
21
23
21
27
21

2.524
2.901
2.934
3.756
2.409
3.374
2.060
7.413
2.307
3.783

23
21
23
21
21
23
23
21
21
21

3.152
3.062
3.003
3.052
3.114
2.97
3.024

21
21
21
21
28
21
29

2.787

21

RELATIVISTIC SEPARABLE DUAL-SPACE GAUSSIAN . . .

PRB 58

3659

TABLEL III. ~Continued!.

CuO
CuH
Cu2
ZnO
ZnH
GaH
GaO
GaF
GeH4
GeO
As2
As4
AsH3
H2 Se
Se2
Se3 (C2 v )
SeO2
Br2
HBr
CH3 Br
r~CBr!
BrF
BrF3
Rb2
RbH
SrH
SrI2
SrO
ZrO
RhC
PdH
AgH
Ag2
CdH
InH
SnH
SnO
SnS
Sb2
SbCl3
Te2
H2 Te
HI
I2
Cs2
CsH
BaH
IrC
PtH
PtC
AuH
HgH
TlH
TlCl
PbH

r~CH!

r1 ~BrF!
r2 ~BrF!

3.153 sc , 2.749
2.731 sc , 2.463
4.086 sc , 3.472
3.157 sc , 2.886
3.010 sc , 2.858
3.176 sc , 3.095
3.167 sc , 3.022
3.376 sc , 3.183
2.861
3.010
3.912
4.551
2.875
2.783
4.089
4.142
3.041
4.296
2.701
2.069
3.636
3.326
3.437
3.289
7.671 sc , 7.333
4.368 sc , 3.072
4.009 sc , 3.522
5.560 sc , 5.327
3.591 sc
3.235 sc , 2.848
3.027
2.865
3.023 sc , 2.688
4.717 sc , 3.760
3.323 sc , 3.108
3.504 sc , 3.394
3.331
3.401
4.123
4.624
4.414
4.824
3.146
3.073
5.029
8.479 sc , 7.976
4.630 sc , 3.171
4.176 sc , 3.403
3.324
2.872
3.612
2.898 sc , 2.424
3.278 sc , 3.067
3.600 sc , 3.440
4.472
3.433

3.176b
2.755b
4.110b
3.161b
3.025b
3.173
3.173
3.367
2.896
3.074
3.951
4.573
2.887
2.788
4.097
4.146
3.045
4.299
2.705
2.063
3.635
3.322
3.434
3.285

3.512c

3.156c
3.073c

3.258
2.765
4.195

21
21
21

3.013
3.143

21
21

3.353
2.882
3.118
3.974
4.602
2.855
2.759
4.076

30
23
21
21
23
23
23
21

3.038
4.314
2.670
2.052
3.653
3.318
3.420
3.252
7.956
4.473
4.053
5.69
3.630
3.234
3.049
2.889
3.056
4.67
3.329
3.473
3.373
3.473
4.174
4.70
4.411
4.832
3.133
3.031
5.039
8.783
4.713
4.217
3.180
2.888
3.371
2.879
3.284
3.534
4.81
3.475

30
21
21
23
23
21
23
31
21
21
23
21
21
21
21
21
21
21
21
21
21
32
33
21
23
21
21
34
21
21
21
21
21
21
21
21
21
21

3660

PRB 58

C. HARTWIGSEN, S. GOEDECKER, AND J. HUTTER

TABLE III ~Continued!.
PbS
PbCl2
BiH

4.237
4.552
3.374

4.526
4.609
3.411

22
34
22

with 3 21G basis set.

with 6 31111G(3d,3pd) basis set.
c
Reference 20.
GAUSSIAN 94

GAUSSIAN 94

accuracy and necessary computational effort, and should be

tested carefully. Therefore we constructed both types of
pseudopotentials for most elements where semicore electrons
can play a significant role in electronic structure calculations.
V. MOLECULES

We tested our pseudopotentials by calculating the bond

lengths of a large number of molecules. In all calculations
we used our scalar relativistic pseudopotentials, neglecting
the terms for spin-orbit interaction. Whenever possible we
tried to determine values for the bond lengths, that are converged to '1023 bohr. To obtain this high level of accuracy,
extremely large boxes and high plane-wave energy cutoffs
were needed, so that for some molecules the accuracy of the
calculations was limited by our computational resources. The
calculated bond lengths, together with their experimental
values, are listed in Table III. As a reference for the quasiexact LDA value, we also list the bond lengths calculated
with GAUSSIAN 94 ~Ref. 19! using a 6 311G
11(3d f ,3pd) basis set ~for the 3d elements, no f polarization functions have been used!. With a few exceptions the
values calculated with GAUSSIAN 94 agree within a few thousandths of a bohr, with the LDA results published by Dickson and Becke,20 and therefore should be close to the LDA
limit. For some molecules where no high precision basis sets
are available, we took the all-electron results from Dickson
and Becke. To estimate the error arising from the pseudopotential approximation, our calculated values should be compared with these LDA results rather than with the experimental bond lengths. Unfortunately exact LDA values for
molecules containing heavier elements often are not available because of the lack of a sufficiently accurate basis set.
The bond lengths calculated with our pseudopotentials,
including semicore electrons where necessary, agree very
well with the all-electron values obtained with GAUSSIAN 94.
The error of the pseudopotential approximation for first row
atoms is nearly ten times smaller than the LDA error, and for
the heavier elements at least comparable to the LDA error. In
all cases except for the non-semicore pseudopotentials, the
accuracy relative to the exact LDA value is, however, better
than the results obtained with standard Gaussian 6 31G *
basis sets, and it is comparable to or better than the results
obtained with other all-electron methods. It must be mentioned that our results especially for molecules with heavier
elements are not exactly comparable to the values obtained
with GAUSSIAN 94 or the values of Dickson and Becke, as our
pseudopotentials also include relativistic effects.
For some non-semicore pseudopotentials, the error in the
calculated bond lengths is quite large. Nevertheless these
pseudopotentials may still be of interest for electronic structure calculations if no high precision is required, or if the

computational resources are limited. These pseudopotentials

require only small basis sets which is sometimes a necessity
for the study of large systems.
Our calculated bond lengths containing only first or second row atoms also agree to within one or two thousandths
of a bohr to those obtained with nonrelativistic versions of
these pseudopotentials that have already been published5
~differences in the case of HCN are due to the choice of a
different simulation box!. This clearly demonstrates that
relativistic effects do not influence the bond lengths for these
molecules on a relevant scale.
VI. PARAMETERS

In the following we list the parameters for all elements up

to Rn. The entries in Table I have the following meaning:
Element

Z ion

r loc C 1
h 01,1
h 11,1
k 11,1
h 21,1
k 21,1

r0
r1
r2

C2

C3

h 02,2
h 12,2
k 12,2
h 22,2
k 22,2

h 03,3
h 13,3
k 13,3
h 23,3
k 23,3

C4

Only the nonzero parameters are shown in Table I. Parameters for elements marked with sc correspond to semicore
pseudopotentials. In order to keep the table as small as possible, the coefficients h li, j and k li, j of the nonlocal projectors
for i5
j are not listed. To obtain the full parameter set, the
missing h li, j and k li, j have to be calculated from h li,i and k li,i .
The relevant equations for the h li, j are
1
2

h 01,252

h 01,35

1
2

h 02,352

1
2

h 11,252

3 0
h ,
5 2,2

~20!

5 0
h ,
21 3,3

~21!

1
2

100 0
h ,
63 3,3

5 1
h ,
7 2,2

~22!

~23!

RELATIVISTIC SEPARABLE DUAL-SPACE GAUSSIAN . . .

PRB 58

h 11,35

1
6

h 12,352

h 21,252

h 21,35

1
2

h 22,352

35 1
h ,
11 3,3

1 14 1
h ,
6 A11 3,3
1
2

~24!

gives the equations for k li, j . We found that the treating all
h li, j for i j as independent pseudopotential parameters does
not improve the results.

~25!

VII. SUMMARY

~28!

We developed a complete set of relativistic LDA pseudopotentials for the whole periodic system up to Rn. The
pseudopotentials are easy to use as only a few parameters are
necessary. All terms for both Fourier and real space are
given analytically, and no tabulated functions are needed.
The pseudopotentials are highly accurate and transferable,
and have been tested in extensive atomic and molecular calculations.
Gaussian-type pseudopotentials for other exchange correlation functionals or gradient-corrected functionals can easily
be constructed using our LDA parameter sets as an initial
guess. The necessary programs are available from the authors upon request.

~29!

ACKNOWLEDGMENTS

7 2
h ,
9 2,2

~26!

63 2
h ,
143 3,3

~27!

1 18 2
h ,
2 A143 3,3

h li, j 5h lj,i .

By this procedure we minimize linear dependencies of the

h li, j during the fit. Replacing h 1i, j by k li, j in Eqs. ~20!~29!

3661

G. B. Bachelet, D. R. Hamann, and M. Schluter, Phys. Rev. B 26,

4199 ~1982!.
2
L. Kleinmann and D. M. Bylander, Phys. Rev. Lett. 48, 1425
~1982!; D. C. Allan and P. Teter, ibid. 59, 1136 ~1987!.
3
X. Gonze, R. Stumpf, and M. Scheffler, Phys. Rev. B 44, 8503
~1991!.
4
R. Stumpf, X. Gonze, and M. Scheffler, Fritz-Haber-Institute Research Report No. 1, April 1990 ~unpublished!.
5
S. Goedecker, M. Teter, and J. Hutter, Phys. Rev. B 54, 1703
~1996!.
6
X. Gonze, P. Kaekell, and M. Scheffler, Phys. Rev. B 41, 12 264
~1990!; D. J. Singh, H. Krakauer, C. Haas, and A. Y. Liu, ibid.
46, 13 065 ~1992!.
7
P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 ~1964!.
8
W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 ~1965!.
9
G. B. Bachelet and M. Schluter, Phys. Rev. B 25, 2103 ~1982!.
10
J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77,
3865 ~1996!.
11
A. D. Becke, Phys. Rev. A 38, 3098 ~1988!; C. Lee, W. Yang,
and R. G. Parr, Phys. Rev. B 37, 785 ~1988!; B. Miehlich, A.
Savin, H. Stoll, and H. Preuss, Chem. Phys. Lett. 157, 200
~1988!; J. P. Perdew and Y. Wang, Phys. Rev. B 33, 8800
~1986!; J. P. Perdew, ibid. 33, 8822 ~1986!; 34, 7406~E! ~1986!.
12
J. P. Perdew and Y. Wang, Phys. Rev. B 45, 13 244 ~1992!.
13
D. R. Hamann, M. Schluter, and C. Chiang, Phys. Rev. Lett. 43,
1494 ~1980!.
14
E. L. Shirley, D. C. Allan, R. M. Martin, and J. D. Joannopoulos,
Phys. Rev. B 40, 3652 ~1989!.
15
R. G. Parr and W. Wang, Density-Functional Theory of Atoms
and Molecules ~Oxford University Press, New York, 1989!; M.

Financial support by Hoechst AG is gratefully acknowledged. We thank Sverre Froyen and Mike Teter for making
available their relativistic atomic codes.

Teter, Phys. Rev. B 48, 5031 ~1993!.

S. Goedecker and K. Maschke, Phys. Rev. A 45, 88 ~1992!.
17
W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recepies ~Cambridge University Press, New
York, 1986!.
18
S. G. Louie, S. Froyen, and M. L. Cohen, Phys. Rev. B 26, 1738
~1982!.
19
M. J. Frisch, G. W. Trucks, H. B. Schlegel, P. M. W. Gill, B. G.
Johnson, M. A. Robb, J. R. Cheeseman, T. Keith, G. A. Petersson, J. A. Montgomery, K. Raghavachari, M. A. Al-Laham, V.
G. Zakrzewski, J. V. Ortiz, J. B. Foresman, C. Y. Peng, P. Y.
Ayala, W. Chen, M. W. Wong, J. L. Andres, E. S. Replogle, R.
Gomperts, R. L. Martin, D. J. Fox, J. S. Binkley, D. J. Defrees,
J. Baker, J. P. Stewart, M. Head-Gordon, C. Gonzalez, and J. A.
Pople, GAUSSIAN 94, Gaussian, Inc., Pittsburgh, PA, 1995.
20
R. M. Dickson and A. D. Becke, J. Chem. Phys. 99, 3898 ~1993!.
21
K. P. Huber and G. Herzberg, Constants of Diatomic Molecules
~Van Nostrand Reinhold, New York, 1979!.
22
V. E. Bondybey and J. H. English, J. Chem. Phys. 80, 568 ~1984!.
23
J. H. Callomon et al., in Atomic and Molecular Physics. Structure
and Data of Free Polyatomic Molecules, edited by K.-H. Hellwege and A. M. Hellwege, Landolt-Bornstein, New Series,
Group II, Vol. 7 ~Springer, Berlin, 1976!.
24
D. L. Gray and A. G. Robiette, Mol. Phys. 37, 1901 ~1979!.
25
S. Carter, I. M. Mills, and N. C. Handy, J. Chem. Phys. 97, 1606
~1992!.
26
M. D. Harmony et al., J. Phys. Chem. Ref. Data 8, 619 ~1979!.
27
G. A. McRae, M. C. L. Gerry, and E. A. Cohen, J. Mol. Spectrosc. 116, 58 ~1986!.
28
R. M. Gordon and A. J. Merer, Can. J. Phys. 59, 642 ~1980!.
16

3662

C. HARTWIGSEN, S. GOEDECKER, AND J. HUTTER

T. C. DeVore and T. N. Gallaher, J. Chem. Phys. 71, 474 ~1979!.

J. Demaison et al., in Atomic and Molecular Physics. Molecular
Constants, edited by K.-H. Hellwege and A. M. Hellwege,
Landolt-Bornstein, New Series, Group II, Vol. 6 ~Springer, Berlin, 1974!.
31
C. Amiot, P. Crozet, and J. Verges, Chem. Phys. Lett. 121, 390
~1985!.
29
30

32

PRB 58

A. A. Radzig and B. M. Smirnov, Reference Data on Atoms,

Molecules and Ions ~Springer, Berlin, 1985!.
33
E. Hirota et al., in Atomic and Molecular Physics. Structure Data
of Free Polyatomic Molecules, edited by K. Kuchitsu, LandoltBornstein, New Series, Group II, Vol. 21 ~Springer, Berlin,
1992!.
34
W. Weickenmeier et al., J. Chem. Phys. 82, 5354 ~1985!.