PHYSICAL REVIE%'
8
VOLUME 37, NUMBER 9Comments
15MARCH 1988-II
Comments are short PaPers which comment on papers
of
other authors preuiously published in the physical Review. Each Comment should state clearly to which paper it refers and must be accompanied by a brief abstract. The same publication scheduleasfor regular articles isfollowed, and page proofs are sent to authors.
Comment on
"Psendoyotentials
that work: From
H to
Pn"
P.
J.
H.
Denteneer and%.
van HaeringenDepartment
of
Physics, Eindhouen Universityof
Technology, P.O.Box 513, NI.5600M-BEindhouen, TheNetherlandsF.
Brosens,J.
T.
Devreese,'
andO. H.
NielsenDepartment
of
Physics, Uniuersityof
Antwerp (Uniuersitaire Instelling Antwerpen) Uniuersiteitsplein 1, B-2610Antwerpen (Wilrijk), BelgiumP.
E.
Van Camp and V.E.
Van DorenUniuersity ofAntwerp (Rijksuniuersitair Centrum Antwerpen), Groenenborgerlaan 171,
B
2020-, Antwerpen, Belgium (Received 17December 1986}In this Comment we report on small differences found in the ion-core pseudopotentials ofSi
be-tween our results and the values tabulated earlier by Bachelet, Hamann, and Schluter. Itis shown
that the rounding ofthe parameters in their tables leads toinaccuracies. Forconvenience tofuture users ofthe Bachelet-Hamann-Schluter potentials a correct setofreference tables isgiven.
The introduction
of
norm-conserving ion-core pseudo-potentials' has been an important development in pseu-dopotential theory. These pseudopotentials are transfer-able by construction. Thereforeif
one accepts the under-lying exchange and correlation functional, they may be used in calculations involving the electronic propertiesof
atoms, molecules, surfaces, and solids. The set
of
ion-core pseudopotentials for the elementsof
hydrogento
plutonium have been tabulated by Bachelet, Hamann, and Schliiter (hereafter denoted by
BHS).
Using the tablesof
BHS,
calculationsof
electronic properties have been performed by several groups.The ion-core pseudopotentials
of BHS, hV/'"(r},
maybe considered to be very useful because
of
the fact that they are expanded in termsof
error functions and Gauss-ians, with expansion coefficients A; [given inEq.
(2.22)of
the BHS paper] making it possibleto
obtain analytical expressions for matrix elements not only for Gaussian basis functions but also for a basis setof
plane waves. However, as stated in theBHS
paper, the 6ttingcoeScients
A; can take on rather large values. SinceBHS
considered it not practicalto
tabulate numbers withtoo many digits, these
coeScients
A; were transformed tonew coe%cients C; by means
of
an orthogonality trans-formation. The triangular matrix Qof
this transforma-tion is given in closed form inEq. {2.
26)of
theBHS
pa-per. Accordingto
BHS
the advantageof
this transforma-tion is that an accuracyof
four digits in the C,.coe5cients
suIces
to calculate the ion-core pseudopoten-tialshVt'"(r)
The C; coeffici.ents are given in Table IVof
theBHS
paper. From these tabulated values the A s and subsequently the pseudopotentials6
V,'"{r}
areob-tained by applying the inverse orthogonalization pro-cedure.
Finally, in order to allow users to check the accuracy
of
their programs and their own inverse orthogonaliza-tion procedures,BHS
list in Table Vof
their article the I=0,
1,2 pseudopotentialof
Si. All the calculations in the BHS paper have been performed in single precision on aCray-1 computer,i.e.,
using 64-bit arithmetic.In comparing the results
of
calculations performed in-dependently by the present authors with Table Vof
Ref.
3, small difFerences were found in bVt'"(r) of
Si forr
close to zero. The deviations are
of
the order0.
5%
at smallr
and decrease rapidly with j.nereasingr.
Althoughthe discrepancies are small they nevertheless introduce some doubt about the accuracy
of
the programs.Upon investigation
of
this problem, we found thatseemingly trivial details in the computational procedures as well as the precision
of
various computers made the conversion from the coefficients C, to A, ambiguous. The small changes in the ion-core pseudopotential are dueto
numerical inaccuracies in the computationof
the overlap matrixS,
the transfer matrix Q and the inverse orthogonality transformation, given respectively by Eqs. (2.27), (2.26),and (2.28)of
Ref.
3.
The overlap matrix
S
can be calculated analytically in termsof
the tabulated parameters o.;,
and it is easily shown that the roundingof
these parameters allows the determinationof
the matrix elementsof
S
with a relative accuracyof
the orderof
0.
1%.
However, the transfer matrix Q is rather sensitive to the errors in the overlap matrix. The uncertainty in the matrix elements Q; isof
the order
of
the uncertainty inS,
-- fori=
l,
but it0.0 0.1 0.2 0.3 0.4 0.5
0.
6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4 2.5 aV(""(r) from Table V of Ref. 3 2.2360 2.1929 2.0610 1.8327 1,5002 1,0598 0.5170—
0.1078—
0.7729—
1.4175-1.
9743—
2.3894—
2.6395—
2.7354—
2.7133—
2.6181—
2.4893—
2.3534—
2.2245—
2.1071—
2.0016—
1.9062—
1.8196—
1.7403-1.
6676—
1.6007AVE""(r) calculated from Table IV of Ref. 3 2.2374 2.1942 2.0616 1.8327 1.5000 1.0598 0.5172
—
0.
1075—
0.7727—
1.4175—
1,9744—
2.3895—
2.6395—
2.7354—
2.7133—
2.6181—
2.4894—
2.3536—
2.2246—
2.1073—
2.0017—
1.9063—
1.8197—
1.7404-1.
6677—
1.6007TABLE
I.
Silicon ion-core pseudopotential forI=0
on areal space mesh, as derived from the coef6rients in Table IVofRef. 3,obtained on a Control Data Corporation Cyber 205 computer (128 bits), on a Digital Equipment Corporation VAX11j780
computer {64bits), and on a Burroughs 87900computer with
96-bit variables, compared toTable VofRef.3. AB data are in
hartree atomic units.
0.0
0.
1 0.20.
30.
4 0.50.
60.
7 0.80.
9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0 2.237 449 2.194 211 2.061 617 1.832669 1.499 964 1.0598060.
517183—
0.107 506—
0.772737—
1,417522—
1.974392—
2.389533—
2.639498—
2.735361—
2.713267—
2.618140—
2.489362—
2.353582—
2.224623—
2.107 282—
2.001678—
1.906330—
1.819 653—
1.740409—
1.667679—
1.600736—
1.538960—
1.481800—
1428766—
1.379425—
1.333399—
2.480 696—
2.485 788—
2.500993—
2.525832—
2.558 844—
2.597160—
2.636 615—
2.672411—
2.700025—
2.715882—
2.717591—
2.703 776—
2.673775—
2.627513—
2.565597—
2.489529—
2.401 817—
2.305 835—
2.205414—
2.104313—
2.005728—
1.911981—
1.824447—
1.743660—
1.669 533—
1.601 605—
1.539244—
1.481795—
1.428651—
1.379 294—
1.333290—
4.669 536—
4.720311—
4.856036—
5.037 463—
5.224481—
5.389462—
5.512 217—
5.565668—
5.510829—
5.310641—
4.954168—
4.472291—
3.931942—
3.411624—
2.973033—
2.643941—
2.418298—
2.269037—
2.164 132—
2.078153—
1.996563—
1.914294—
1.831876—
1.751781—
1.676210—
1.606305—
1.542 228—
1.483 555—
1,429638—
1.379829—
1.333574TABLE
II.
Silicon I=0,
1, and 2ion-core pseudopotentials5
VI""(r)on areal space mesh as derived from the coe%cients inTable IVofRef.3.All data are inhartree atomic units.
cally increases with increasing
i.
In practice, for angular momentum I=0
in Si,the first digit in Q66isdetermined by the eleventh digit in the matrix elementsS;
J.
The magnitudeaf
the diaganal elementsQ;;
turns autta
beaf
the order 10 ' (far
1=0
and l=2)
and the transfer ma-trix Q is thus rather ill conditioned.For
instance, with a 64-bit computation, the seventh significant digitaf
Q6 s is even inlluenced by the orderaf
the operations used in cal-culating terms like(a,
+a~ }"
~,
which occur inS,
.J..
Thisinaccuracy propagates further in the computation
of
the coefficients A;, as determined fromEq.
(2.28}af
Ref.
3.
It
seemsof
little useto
program the explicit expressions, worked out inRef.
5,since these are quite cumbersome. A mare elegant way ista
use the fact that the matrix Q is triangular and torewrite this equation as~6=
—
C6~Q6,6r
for i
=5,
4,..
.
,1.
Furthermore, even with sufFicient accuracy in the matrix Q, the accuracy
af
the parameters A; is clearly directly determined by the roundingof
the tabulated valuesof
C;.
In attempts
ta
reproduce Table Vaf
Ref.
3,we found that the resultsof
the ion-core pseudopotentials for Si de-pend upon the precisionof
our various computers. Be-causeaf
dependency an trivial detailsaf
the calculation we 6nd for the coefFicients A; an accuracyof
approxi-mately4
signi6cant digits with 64-bit arithmetic,9
significant digits with 96-bit arithmetic and 15 signi6cant digits with 128-bit arithmetic. However aur b,V/""(r} agree
ta
within 10 hartree (ifat least double precisian ana
32-bit computer isused). These last results however difFer slightly from Table Vof
Ref.
3.
This is demonstrat-ed in TableI
where theBHS
resultsaf
b,V&""(r}are cam-pared with our results obtained on various computers. This leads usto
speculate that the potential presented in Table Vof Ref.
3 has been calculated with slightly difFerent values for C,.than those presented in Table IVof
Ref. 3.
pseudo-potential b,V/'"(r) for Siaswe have derived it from the C,
coeScient
given in Table IVof Ref.
3. %e
thus hope that future usersof
theBHS
pseudopotentials mill not have to spend time debugging their procedures when in fact they may well be correct despite slight deviations from TableV
of
Ref.
3.
One
of
us(F.
B.
) was supported in part by the National Foundation forScientific Research, Belgium.'Also at University of A@twerp Rijksuniversitair Centrum Antwerpen (RUCA), 8-2020Antwerpen, Belgium and Eind-hoven University of Technology, NL-5600MB Eindhoven, The Netherlands.
Present address: Nordisk Institut fur Teoretisk Atomfysik (NORDITA), Blegdamsvej 17, DK-21 00Kgfbenhavn, Den-mark.
'D. R.
Hamann, M. Schluter, and C.Chiang, Phys. Rev. Lett.43,1494(1979).
2M.
T.
Yin and M.L.Cohen, Phys. Rev.8
25, 7403(1982). 3G.B.
Bachelet, D.R.
Hamann, and M.Schliiter, Phys. Rev.8
M, 4199 (1982).
4P.