PHYSICAL REVIEW A 73, 022503 共2006兲

Ground-state wave function and energy of the lithium atom

Mariusz Puchalski* and Krzysztof Pachucki†

Institute of Theoretical Physics, Warsaw University, Hoża 69, 00-681 Warsaw, Poland
共Received 28 November 2005; published 6 February 2006兲

Highly accurate nonrelativistic ground-state wave function and energy of the lithium atom are obtained in
the Hylleraas basis set. The leading relativistic corrections, as represented by the Breit-Pauli Hamiltonian, are
obtained in fair agreement with the former results. The calculational method is based on analytical evaluation
of Hylleraas integrals with the help of recursion relations.

DOI: 10.1103/PhysRevA.73.022503 PACS number共s兲: 31.25.Eb, 31.30.Jv, 31.15.Pf, 02.70.⫺c

I. INTRODUCTION accuracy is limited by the nonrelativistic energy. Moreover,

Theoretical predictions for the energy levels of light few
electron atoms are much less accurate than for the hydro-
genic systems. It is for two reasons. The nonrelativistic wave
function has to include electron correlations to a high degree
of accuracy. This can be achieved by using the Hylleraas
basis set, but it is quite difficult to evaluate integrals with
Hylleraas functions for three and more electrons. The second
reason is the difficulty in the accurate treatment of relativistic
and radiative corrections. The commonly used Dirac-
Coulomb Hamiltonian for few electron atoms does not in-
clude relativistic corrections properly as it cannot be derived
from quantum electrodynamic theory and its continuous
spectrum ranges from −⬁ to +⬁. One of the possible ap-
proaches is the derivation of an effective Hamiltonian 关1兴
within the so-called nonrelativistic quantum electrodynamic
共NRQED兲 theory. Matrix elements of this Hamiltonian give
exact correction to the energy at specified order in the fine-
structure constant ␣. However, this Hamiltonian becomes
quite complicated at higher orders and, for example, m␣6
corrections have been obtained for few low-lying states of
helium only 关2,3兴, not for lithium or beryllium atoms.
Theoretical predictions for light hydrogenlike atoms are at
present limited by uncertainty in higher-order two-loop elec-
tron self-energy corrections 关4兴, which is a few kHz for the
1S state. For helium-like atoms predictions are approxi-
mately 103 times less accurate. Since the nonrelativistic
wave function was computed very accurately using Hylleraas
关5兴 or exponential basis sets 关6兴, the uncertainty in its levels
comes mainly from the unknown m␣7 terms. These correc-
tions are currently under investigation in the context of he-
lium 23 PJ fine splitting. For lithium atoms, the Hylleraas
functions give very accurate nonrelativistic wave function
and energies 关7兴, but the precise calculation of three-electron
integrals with Hylleraas functions is very time consuming
关8,9兴, and so far no results for m␣6 corrections have been
obtained. For the beryllium atom the most accurate results
have been obtained with explicitly correlated Gaussian func-
tions 关10兴. Although it was possible to calculate accurately
the leading relativistic and QED corrections 关11兴, the final
this basis cannot be used for higher-order corrections since
the Gaussian wave functions do not fulfill the cusp condition.
So far the most accurate results for various states of the
lithium atom were obtained by Yan and Drake in Ref. 关7兴.
Here, we present even more accurate results for the lithium
ground state, as a demonstration of an analytic method to
compute the integrals with Hylleraas functions 关12兴. This
method is based on recursion relations between integrals
with different powers of electron-nucleus and interelectron
distances, which are fast and numerically stable for generat-
ing large basis sets. Our result for the ground-state energy
E = − 7.478 060 323 904 1共−50
+10
兲 共1兲
is significantly below the previous one, obtained in Ref. 关7兴,
which is E = −7.478 060 323 650 3共71兲. As a further applica-
tion of the analytic approach, we obtain the leading relativ-
istic corrections to the binding energy by the calculation of
the expectation value of the Breit-Pauli Hamiltonian in Eq.
共13兲. For this we used recursion relations for extended Hyl-
leraas integrals with 1 / r2ij and 1 / r2i terms. They have been
derived in Ref. 关13兴 and in this work respectively.
In the next section we construct the nonrelativistic wave
function, similarly to Ref. 关7兴, and obtain the ground-state
nonrelativistic energy and the wave function. In Sec. III we
compute the leading relativistic correction as given by the
Breit-Pauli Hamiltonian. In Sec. IV we derive recursion re-
lations for Hylleraas integrals containing 1 / r2i , which among
others, are necessary for relativistic matrix elements. In Sec.
V we summarize our result and present prospects for calcu-
lation of higher-order terms as well as the calculation of
Hylleraas integrals for four and more electrons.
II. NONRELATIVISTIC WAVE FUNCTION AND ENERGY
In the construction of the wave function we closely follow
the works of Yan and Drake in Ref. 关7兴. The ground-state
wave function ⌿ is expressed as a linear combination of ␺,
the antisymmetrized product of ␾ and the spin function ␹,
␺ = A关␾共rជ1,rជ2,rជ3兲␹兴, 共2兲

*Electronic address: mpuchals@fuw.edu.pl

†
Electronic address: krp@fuw.edu.pl ␾共rជ1,rជ2,rជ3兲 = e−w1r1−w2r2−w3r3r23
n1 n2 n3 n4 n5 n6
r31r12r1 r2 r3 , 共3兲

1050-2947/2006/73共2兲/022503共7兲/$23.00 022503-1 ©2006 The American Physical Society

MARIUSZ PUCHALSKI AND KRZYSZTOF PACHUCKI PHYSICAL REVIEW A 73, 022503 共2006兲

TABLE I. Ground-state nonrelativistic energies and expectation values of Dirac ␦ functions obtained
using Drachman formulas 关16兴 for various basis lengths.

⍀ No. of terms E共⍀兲 兺 ␦ 共r 兲

a
3
a 兺 ␦ 共r
a⬎b
3
ab兲

3 50 ⫺7.477 981 524 089 7 13.843 446 803 98 0.544 164 351 92
4 120 ⫺7.478 052 334 642 2 13.842 288 641 67 0.544 331 564 16
5 256 ⫺7.478 059 463 915 8 13.842 509 174 63 0.544 327 870 45
6 512 ⫺7.478 060 208 663 7 13.842 637 966 67 0.544 325 260 63
7 918 ⫺7.478 060 310 362 9 13.842 606 662 38 0.544 324 788 85
8 1589 ⫺7.478 060 320 507 6 13.842 608 240 76 0.544 324 697 02
9 2625 ⫺7.478 060 323 450 1 13.842 610 098 57 0.544 324 629 45
10 4172 ⫺7.478 060 323 775 0 13.842 610 698 67 0.544 324 627 57
11 6412 ⫺7.478 060 323 861 0 13.842 610 779 19 0.544 324 631 50
12 9576 ⫺7.478 060 323 889 7 13.842 610 781 06 0.544 324 632 05
⬁ ⬁ +10
−7.478 060 323 904 1共−50 兲 13.842 610 783 46共100兲 0.544 324 633 96共50兲
Refs. 关7,15兴 ⬁ ⫺7.478 060 323 650 3共71兲 13.842 609 642 共55兲 0.544 329 79共31兲

␹ = ␣共1兲␤共2兲␣共3兲 − ␤共1兲␣共2兲␣共3兲, 共4兲 sector 1: all n3 , n1 = 0, n2 = 0;

with all ni non-negative integers and wi 苸 R+. The matrix sector 2: all n3 , n1 = 0, n2 ⫽ 0;
element of the Hamiltonian H sector 3: all n3 , n1 ⫽ 0, n2 = 0;

冉 冊
3 3 sector 4: n3 = 0, n1 ⫽ 0, n2 ⫽ 0;
pជ 2a Z␣ ␣
H=兺 − + 兺 , 共5兲 sector 5: n3 ⫽ 0, n1 ⫽ 0, n2 ⫽ 0;
a=1 2 ra a⬎b=1 rab
To avoid numerical instabilities, within each sector we drop
or of any spin independent operator can be expressed after the terms with n4 ⬎ n5 共or n4 ⬍ n5兲 and for n4 = n5 drop terms
eliminating spin variables, as with n1 ⬎ n2 共or n1 ⬍ n2兲. This division allows for a signifi-
cant improvements of nonrelativistic energies by optimiza-
具␺兩H兩␺⬘典 = 具2␾共1,2,3兲 + 2␾共2,1,3兲 − ␾共3,1,2兲 − ␾共2,3,1兲 tion of all five sets of wi’s. These nonlinear parameters are
− ␾共1,3,2兲 − ␾共3,2,1兲兩H兩␾⬘共1,2,3兲典. 共6兲 obtained by Newton’s method of searching zeros using ana-
lytic derivatives

冓冏冏 冔 冓冏 冔
In this way the calculation of this matrix elements is brought
to Hylleraas integrals, namely the integrals with respect to ri ⳵E ⳵⌿ ⳵⌿
=2 ⌿ H − 2E ⌿ . 共9兲
of the form ⳵w ⳵w ⳵w

f共n1,n2,n3,n4,n5,n6兲 = 冕 冕 冕d 3r 1
4␲
d 3r 2
4␲
d3r3 −w r −w r −w r
4␲
e 11 22 33
In the numerical calculations, we use sextuple precision
for recursion relations and quadruple precision for all other
arithmetics to obtain the wave function and the energy up to
⫻r23
n1−1 n2−1 n3−1 n4−1 n5−1 n6−1
r31 r12 r1 r2 r3 , 共7兲 ⍀ = 12. The results obtained for the ground-state energies are
presented in Table I. The next-to-last row is a result of ex-
with non-negative integers ni. These are performed analyti- trapolation to infinite length of the basis set, and the last row
cally for n1 , n2 , n3 = 0 , 1 关14兴 and by recursion relations for shows previous results of Yan and Drake 关7兴. The result for
larger ni using formulas derived in Ref. 关12兴. the nonrelativistic energy is significantly below the previous
The total wave function is generated from all ␾ in Eq. 共3兲 estimate 关7兴 and indicates that extrapolation to infinite basis
with ni satisfying the condition length does not always give the right result. In the same table
6 we present results for the Dirac ␦ functions, which also differ
兺
i=1
ni 艋 ⍀, 共8兲 from previous results in Ref. 关15兴. We observe that the num-
ber of significant digits for Dirac ␦ is increased by using
Drachman formulas 关16兴, namely

冓冏 冏冔
for ⍀ between 3 and 12. For each ⍀ we minimize energy
with respect to the free parameters wi in Eq. 共3兲. We noticed 1
4␲具⌿兩␦3共rab兲兩⌿典 = 2 ⌿ 共E⌿ − V兲 ⌿
that the use of only one set of wi’s does not lead to accurate rab

兺冓 冏 冏 冔
results, therefore, following Yan and Drake 关7兴, we divide the
whole basis set into five sectors, each one with its own set of − ជ ⌿ 1 ⵜ
ⵜ ជ ⌿ , 共10兲
c c
wi’s. This division goes as follows 关7兴: c rab

022503-2
GROUND-STATE WAVE FUNCTION AND ENERGY OF… PHYSICAL REVIEW A 73, 022503 共2006兲

TABLE II. Matrix elements of the Breit-Pauli Hamiltonian H共4兲 in atomic units.

⍀ 兺 − 8ⵜ
a
1 4
a 兺 2ⵜ
a⬎b
1 i
a 冉 ␦ij
rab
+
riabrab
r3ab
j
冊
ⵜbj H共4兲

3 ⫺78.587 286 690 90 ⫺0.438 632 545 84 ⫺12.080 670 336 80

4 ⫺78.557 331 859 61 ⫺0.436 096 586 40 ⫺12.053 111 944 61
5 ⫺78.556 355 905 97 ⫺0.435 697 344 91 ⫺12.050 709 116 55
6 ⫺78.556 714 503 43 ⫺0.435 616 426 50 ⫺12.050 388 076 38
7 ⫺78.556 195 780 85 ⫺0.435 602 362 02 ⫺12.050 004 294 51
8 ⫺78.556 162 642 13 ⫺0.435 599 523 90 ⫺12.049 961 162 16
9 ⫺78.556 137 477 61 ⫺0.435 598 217 44 ⫺12.049 926 149 76
10 ⫺78.556 135 734 01 ⫺0.435 598 047 58 ⫺12.049 921 414 27
11 ⫺78.556 131 596 34 ⫺0.435 597 963 57 ⫺12.049 916 800 81
12 ⫺78.556 128 632 10 ⫺0.435 597 910 50 ⫺12.049 913 772 96
⬁ ⫺78.556 112 88共200兲 ⫺0.435 597 765共50兲 ⫺12.049 897 86共200兲
Ref. 关15兴 ⫺78.556 135 55共148兲 ⫺0.435 598 001共137兲 ⫺12.049 909 94共180兲

4␲具⌿兩␦3共ra兲兩⌿典 = 4 ⌿ 冓冏 1
冏冔
ra
共E⌿ − V兲 ⌿ E共4兲 = 具⌿兩H共4兲兩⌿典 = 冓兺 再 −
pជ 4a
8m3 2m2
+
␲Z␣ 3
␦ 共ra兲 冎
兺冓 冏 冏 冔 再 冎冔
a

−2
c
ជ ⌿ 1 ⵜ
ⵜ c
ra
ជ ⌿ ,
c 共11兲 + 兺兺
a⬎b b
␲␣ 3
m 2 ␦ 共r ab 兲 −
2m
p冉
␣ i ␦ij rab
2 a
rab
+
i j
rab j
3
rab
pb 冊 .

共13兲
where V is a total potential energy in Eq. 共5兲. E共4兲 has already been obtained in Refs. 关15,17兴. Calculations
of these matrix elements involves the usual Hylleraas inte-
grals with all ni non-negative and extended integrals, namely
III. LEADING RELATIVISTIC CORRECTION TO with one parameter ni equal to −1. The direct numerical
BINDING ENERGY method to calculate these integrals was presented in Refs.
关8,9兴. Here we apply the analytic approach. Recursion rela-
The leading relativistic corrections to energy levels are tions for the case of n1 or n2 or n3 equal to −1 have been
given by the expectation values of the Breit-Pauli Hamil- obtained in Ref. 关13兴. Hylleraas integrals involving n4 or n5
tonian H共4兲, or n6 equal to −1 can in principle be obtained by the integra-
tion of the usual Hylleraas integral with respect to the corre-
sponding parameter wi 关13兴. However, some recursion rela-

H 共4兲
=兺 再 −
pជ 4a
8m3 2m2
+
␲Z␣ 3 Z␣ rជ
␦ 共ra兲 + 2 ␴ជ a · 3a ⫻ pជ a
4m ra
冎 tions may become unstable, for example in the case of n4 =
−1 the recursion in n1is numerically unstable for large w1. To

再
a avoid this problem we derive in the next section stable re-

+ 兺兺
a⬎b b
−
␲␣ 3
m 2 ␦ 共r ab 兲 −
2m
p
2 a 冉
␣ i ␦ij rab
rab
+
i j
rab j
3
rab
pb 冊 cursion relations for extended Hylleraas integrals with ni =
−1 for i = 4 , 5 , 6. Numerical results for matrix elements of the
Breit Hamiltonian using these recursion relations has been

−
2␲␣
3m 2
ជ
␴ a · ជ
␴ b ␦ 3
共r ab 兲 +
␣ ␴ia␴bj ij
2 3
4m rab
␦ 冉− 3
i j
rab
2
rab
rab
冊 presented in Tables I and II. One observes that the lowest
convergence is for the −p4 / 8 term, and in spite of the differ-
ences for separate matrix elements, the total relativistic
␣ correction is in good agreement with the former result in
+ ជ ជ ជ ជ ជ ជ
3 关2共␴a · rab ⫻ pb − ␴b · rab ⫻ pa兲 Ref. 关15兴.
4m2rab

ជ b · rជab ⫻ pជ b − ␴ជ a · rជab ⫻ pជ a兲兴 .

+ 共␴ 冎 共12兲 IV. RECURSION RELATIONS FOR THREE-ELECTRON
EXTENDED HYLLERAAS INTEGRAL WITH 1 / r21

In the former section we calculated relativistic correc-

For states with vanishing angular momentum L and spin S tions. For this we needed various extended Hylleraas inte-
= 1 / 2, the expectation value is simplified to the form grals, among them, integrals with 1 / r2i , which are being de-

022503-3
MARIUSZ PUCHALSKI AND KRZYSZTOF PACHUCKI PHYSICAL REVIEW A 73, 022503 共2006兲

rived here. To obtain recursion relations for the three- when n4 , n5, or n6 is equal to −1. Let us assume that n4 =
electron Hylleraas integral in Eq. 共7兲, one first considers the −1. The analytic expression for f共0 , 0 , 0 , −1 , n5 , n6兲 involves
integral G, powers of w2 − w3 in the denominator, which is not very con-

冕 冕 冕
venient in high precision numerical calculations. Instead, we
1 use recursions for f共0 , 0 , 0 , 0 , n5 , n6兲 and numerically inte-
G共m1,m2,m3 ;m4,m5,m6兲 = d 3k 1 d 3k 2 d 3k 3
8␲6 grate with respect to w1, namely
⫻共k21 + u21兲−m1共k22 + u22兲−m2
⫻共k23 + u23兲−m3共k32
2
+ w21兲−m4
f共0,0,0,− 1,n5,n6兲 = 冕 w1
⬁
dw1 f共0,0,0,0,n5,n6兲. 共16兲

⫻共k13
2
+ w22兲−m5共k21
2
+ w23兲−m6 , These recursions are derived as follows. We take id共i , i兲 with
共14兲 i = 1 , 2 , 3 and put ui = 0. Resulting three equations are
solved against three unknowns: G共1 , 1 , 1 , 2 , 1 , 1兲 ,
which is related to f by: f共0 , 0 , 0 , 0 , 0 , 0兲 G共1 , 1 , 1 , 1 , 2 , 1兲, and G共1 , 1 , 1 , 1 , 1 , 2兲. The solution for the
= G共1 , 1 , 1 , 1 , 1 , 1兲兩u1=u2=u3=0. The following nine integration last two G functions is the following:
by part identities are valid because the integral of the deriva-
tive of a function vanishing at infinity vanishes, 1
G共1,1,1,1,2,1兲 = 关G共0,1,1,1,1,2兲 − G共1,0,1,1,1,2兲
w22
0 ⬅ id共i, j兲 = 冕 冕 冕
d 3k 1 d 3k 2 d 3k 3
⳵
⳵ kជ i
− G共1,0,1,2,1,1兲 + G共1,1,0,2,1,1兲
+ G共1,1,1,1,1,1兲/2兴, 共17兲
⫻关kជ j共k21 + u21兲−1共k22 + u22兲−1共k23 + u23兲−1共k32
2
+ w21兲−1
⫻共k13
2
+ w22兲−1共k21
2
+ w23兲−1兴, 共15兲 1
G共1,1,1,1,1,2兲 = 关G共0,1,1,1,2,1兲 + G共1,0,1,2,1,1兲
w23
where i , j = 1 , 2 , 3. The reduction of the scalar products from
the numerator leads to the identities for the linear combina- − G共1,1,0,1,2,1兲 − G共1,1,0,2,1,1兲
tion of the G functions. If any of the arguments is equal to 0,
+ G共1,1,1,1,1,1兲/2兴. 共18兲
then G becomes a known two-electron Hylleraas-type inte-
gral. These identities are used to derive various recursion By differentiation with respect to w2 and w3 one obtains the
relations. Here, we derive a set of recursions for the case following recursion relations:

1
f共0,0,0,0,n5 + 1,n6兲 = 关共n5 + 1兲f共0,0,0,0,n5,n6兲w1w3 − 共n5 + 1兲n6 f共0,0,0,0,n5,n6 − 1兲w1 + n6 f共0,0,0,0,n5 + 1,n6 − 1兲
w 1w 2w 3
⫻w1w2 − n6⌫共n5,n6 − 1,− 1,w1 + w2,w3,0兲 + n6⌫共n6 − 1,n5,− 1,w1 + w3,w2,0兲 − ⌫共n6,n5,− 1,w1
+ w3,w2,0兲w1 + ⌫共n5 + n6,0,− 1,w2 + w3,w1,0兲w1 + ⌫共n5,n6,− 1,w1 + w2,w3,0兲
⫻w3 − ⌫共n6,n5,− 1,w1 + w3,w2,0兲w3兴, 共19兲

1
f共0,0,0,0,n5,n6 + 1兲 = 关共n6 + 1兲f共0,0,0,0,n5,n6兲w1w2 − n5共n6 + 1兲f共0,0,0,0,n5 − 1,n6兲
w 1w 2w 3
⫻w1 + n5 f共0,0,0,0,n5 − 1,n6 + 1兲w1w3 + n5⌫共n5 − 1,n6,− 1,w1 + w2,w3,0兲 − n5⌫共n6,n5 − 1,− 1,w1
+ w3,w2,0兲 − ⌫共n5,n6,− 1,w1 + w2,w3,0兲w1 + ⌫共n5 + n6,0,− 1,w2 + w3,w1,0兲w1 − ⌫共n5,n6,− 1,w1
+ w2,w3,0兲w2 + ⌫共n6,n5,− 1,w1 + w3,w2,0兲w2兴, 共20兲

where ⌫ is a known 关18–20兴 two-electron integral,

⌫共n1,n2,n3, ␣1, ␣2, ␣3兲 = 冕 冕 d 3r 1

4␲
d3r2 −␣ r −␣ r −␣ r n1−1 n2−1 n3−1
4␲
e 1 1 2 2 3 12r1 r2 r12 . 共21兲

The integration in Eq. 共16兲 is performed numerically using points and weights adapted to the function, which has logarithmic
end-point singularity, namely

022503-4
GROUND-STATE WAVE FUNCTION AND ENERGY OF… PHYSICAL REVIEW A 73, 022503 共2006兲

冕 0
1
dx关W1共x兲 + W2共x兲ln共x兲兴, 共22兲

where Wi are functions without any singularities. The method to obtain n adapted points and weights is presented in the
Appendix , and this integral is exact for Wi, being polynomials up to the order n − 1. In the actual calculations we achieved at
least 48-digit precision with only 100 points. Having obtained f共0 , 0 , 0 , −1 , n5 , n6兲 we construct recursion relations in n1 , n2,
and n3. This is achieved in two steps. In the first step we use integration by parts in momentum representation Eq. 共15兲, to form
the following linear combination:

id共2,2兲 + id共3,3兲 − id共1,1兲 = 2关G共0,1,1,1,1,2兲 + G共0,1,1,1,2,1兲 − G共1,0,1,1,1,2兲 − G共1,1,0,1,2,1兲 − G共1,1,1,1,1,1兲/2

− G共2,1,1,1,1,1兲u21 − G共1,1,1,1,1,2兲共u21 − u22兲 + G共1,2,1,1,1,1兲u22 − G共1,1,1,1,2,1兲共u21 − u23兲
+ G共1,1,2,1,1,1兲u23 + G共1,1,1,2,1,1兲w21兴 = 0. 共23兲

We integrate with respect to w1 and differentiate over u1 , u2 , u3 , w2, and w3 to obtain the main formula,

1
f共n1,n2,n3,− 1,n5,n6兲 = 关共n1 − 1兲n1n5 f共n1 − 2,n2,n3,− 1,n5 − 1,n6 + 1兲 + 共n1 − 1兲n1n6 f共n1 − 2,n2,n3,
共n2 + n3 − n1兲w2w3
− 1,n5 + 1,n6 − 1兲 − 共n2 − 1兲n2n5 f共n1,n2 − 2,n3,− 1,n5 − 1,n6 + 1兲 − 共n3 − 1兲n3n6 f共n1,n2,n3 − 2,
− 1,n5 + 1,n6 − 1兲 + 共n1 − n2 − n3兲n5n6 f共n1,n2,n3,− 1,n5 − 1,n6 − 1兲 + n5n6 f共n1,n2,n3,0,n5 − 1,n6
− 1兲w1 − 共n1 − 1兲n1 f共n1 − 2,n2,n3,− 1,n5,n6 + 1兲w2 + 共n2 − 1兲n2 f共n1,n2 − 2,n3,− 1,n5,n6 + 1兲w2
− 共n1 − n2 − n3兲n6 f共n1,n2,n3,− 1,n5,n6 − 1兲w2 − n6 f共n1,n2,n3,0,n5,n6 − 1兲w1w2 − 共n1 − 1兲n1 f共n1
− 2,n2,n3,− 1,n5 + 1,n6兲w3 + 共n3 − 1兲n3 f共n1,n2,n3 − 2,− 1,n5 + 1,n6兲w3 − 共n1 − n2
− n3兲n5 f共n1,n2,n3,− 1,n5 − 1,n6兲w3 − n5 f共n1,n2,n3,0,n5 − 1,n6兲w1w3 + f共n1,n2,n3,0,n5,n6兲w1w2w3
+ n6␦共n3兲⌫共n5 − 1,n6 − 1,n1 + n2 − 1,w1 + w2,w3,0兲 + n5␦共n2兲⌫共n6 − 1,n5 − 1,n1 + n3 − 1,w1
+ w3,w2,0兲 − n5␦共n1兲⌫共n5 + n6 − 1,− 1,n2 + n3 − 1,w2 + w3,w1,0兲 − n6␦共n1兲⌫共n5 + n6 − 1,− 1,n2
+ n3 − 1,w2 + w3,w1,0兲 − ␦共n2兲⌫共n6,n5 − 1,n1 + n3 − 1,w1 + w3,w2,0兲w2 + ␦共n1兲⌫共n5 + n6,− 1,n2
+ n3 − 1,w2 + w3,w1,0兲w2 − ␦共n3兲⌫共n5,n6 − 1,n1 + n2 − 1,w1 + w2,w3,0兲w3 + ␦共n1兲⌫共n5 + n6,
− 1,n2 + n3 − 1,w2 + w3,w1,0兲w3兴. 共24兲

This general formula does not work in the case n1 = n2 + n3. In the second step we use integration by part identities in the
coordinate space to fill this hole. We limit ourselves only to a special case of these identities in the form

0 = id共i兲 ⬅ 冕 冕 冕d 3r 1 d 3r 2 d3r3共gⵜ2i h − hⵜ2i g兲, 共25兲

where

g = e−w1r1−w2r2−w3r3rn14−1rn25−1rn36−1,

n1−1 n2−1 n3−1

h = r23 r31 r12 . 共26兲

The identities id共2兲 and id共3兲,

f共n1,n2,n3,− 1,n5,n6兲 = 关共n1 − 1兲共n1 + n3 − 1兲f共n1 − 2,n2,n3,− 1,n5,n6兲 − 共n1 − 1兲共n3 − 1兲f共n1 − 2,n2 + 2,n3 − 2,− 1,n5,n6兲
+ 共n3 − 1兲共n1 + n3 − 1兲f共n1,n2,n3 − 2,− 1,n5,n6兲 − 共n5 − 1兲n5 f共n1,n2,n3,− 1,n5 − 2,n6兲 + 2n5 f共n1,n2,n3,
− 1,n5 − 1,n6兲w2 + ␦共n5兲⌫共n1 + n6 − 1,n3 − 2,n2,w3,w1,0兲兴/w22 , 共27兲

f共n1,n2,n3,− 1,n5,n6兲 = 关− 共n1 − 1兲共n2 − 1兲f共n1 − 2,n2 − 2,n3 + 2,− 1,n5,n6兲 + 共n1 − 1兲共n1 + n2 − 1兲f共n1 − 2,n2,n3,− 1,n5,n6兲
+ 共n2 − 1兲共n1 + n2 − 1兲f共n1,n2 − 2,n3,− 1,n5,n6兲 − 共n6 − 1兲n6 f共n1,n2,n3,− 1,n5,n6 − 2兲 + 2n6 f共n1,n2,n3,
− 1,n5,n6 − 1兲w3 + ␦共n6兲⌫共n2 − 2,n1 + n5 − 1,n3,w1,w2,0兲兴/w23 , 共28兲
022503-5
MARIUSZ PUCHALSKI AND KRZYSZTOF PACHUCKI PHYSICAL REVIEW A 73, 022503 共2006兲

replace the main recursion in Eq. 共24兲 for the case n1 = n2 nardelli for bringing the work 关29兴 to our attention. This
+ n3 and can be used also for all other ni under conditions work was supported by EU Grant No. HPRI-CT-2001-50034.
that n1 ⬎ 0, n3 ⬎ 0 or n1 ⬎ 0, n2 ⬎ 0, respectively.
APPENDIX: QUADRATURE WITH LOGARITHMIC
V. SUMMARY END-POINT SINGULARITY
We have demonstrated the advantages of the analytic ap- Consider the integral
proach to three-electron Hylleraas integrals by the calcula-
tion of nonrelativistic energy of the ground-state lithium
atom and the leading relativistic corrections. The achieved
accuracy is mainly due to the use of much larger basis sets.
I= 冕 0
1
dx关W1共x兲 + ln共x兲W2共x兲兴, 共A1兲

In fact it is possible to perform a calculation with ⍀ ⬎ 12 by where Wi are arbitrary polynomials of maximal degree n − 1.
using sextuple precision arithmetics. The typical evaluation We would like to find n nodes xi and n weights wi such that
time in sextuple precision for ⍀ = 12 is 24 h on a 2.4-GHz
Opteron, and most of the time is devoted to LU decomposi- n

tion. I = 兺 wi关W1共xi兲 + ln共xi兲W2共xi兲兴. 共A2兲

Having precise wave functions, we have calculated lead- i=1
ing relativistic corrections and the results only partially agree
In general it is a difficult numerical problem to find a solu-
with that of Yan and Drake 关15兴 and of King 关17兴. We are
tion of corresponding 2n nonlinear equations with j = 1 , n,
now in a position to calculate higher-order, namely m␣6,

冕
relativistic and QED corrections, for example to the lithium 1 n
1
ground-state hyperfine splitting 关21兴. However, this involves dxx j−1
= = 兺 wixij , 共A3兲
more complicated Hylleraas integrals containing two factors 0 j i=1
among 1 / r2i and 1 / r2ij, which have not yet been worked out

冕
by the recursion method of the authors. 1 n
1
Even more interesting is the possible extension of this dxx j−1ln x = − = 兺 wixijln x j . 共A4兲
analytic method to beryllium and berylliumlike ions, the 0 j2 i=1
four-electron systems. The use of a large Hylleraas basis set
will allow for a high precision calculation of the wave func- The work 关29兴 solves this problem and proves that wi are all
tion, energies, and transition rates. For example, knowing the positive. The solution is as follows. One defines 2n functions
isotope shifts, one can obtain charge radii as for the lithium ␾ i,
isotope 关22兴. General Hylleraas integrals for four-electron ␾k共x兲 = xk−1, for k = 1,n, 共A5兲
systems has not yet been worked out 关23,24兴. The so-called
double linked basis set, the functions with at most two odd
powers of rij, have been used by Büsse et al. in Ref. 关25兴 to ␾k共x兲 = xk−1ln x, for k = n + 1,2n. 共A6兲
obtain an accurate nonrelativistic energy, but still less accu-
Consider n points xi, which are not necessarily the solution
rate than the result of Komasa in Ref. 关10兴. It has not yet
of Eqs. 共A3兲 and 共A4兲 but are close to them, and construct
been attempted to calculate relativistic corrections with Hyl-
another set of functions ␴i, ␩i, for i = 1 , n,
leraas functions as they involve even more difficult integrals.
We think the integration by part technique should allow for 2n
the derivation of compact formulas for all four-electron Hyl- ␴i共x兲 = 兺 ␣ij␾ j共x兲, 共A7兲
leraas integrals. j=1
Our primary motivation for developing the Hylleraas ba-
sis set is the calculation of higher-order relativistic and QED 2n
effects, and to demonstrate that standard techniques used in ␩i共x兲 = 兺 ␤ij␾ j共x兲, 共A8兲
relativistic quantum chemistry, which are based on a multi- j=1
electron Dirac-Coulomb Hamiltonian, are not correct for
these principal reasons. This Hamiltonian does not include such that
properly negative energy states. The correct treatment has to
be based on quantum electrodynamics and several very ac- ␴i共xk兲 = 0,
curate results for few electron ions have already been ob-
tained within the so-called 1 / Z expansion 关26–28兴. Never- ␴i⬘共xk兲 = ␦ik ,
theless, there is no formalism yet that allows for systematic
inclusion of negative energy states and QED effects for ␩i共xk兲 = ␦ik ,
many-electron atoms.

ACKNOWLEDGMENTS ␩i⬘共xk兲 = 0. 共A9兲

We are grateful to Vladimir Korobov for his source code The set of conditions 共A9兲 uniquely determines the matrices
of the fast multiprecision arithmetics and to Michał Ber- ␣ij and ␤ij. If xk are nodes, then

022503-6
GROUND-STATE WAVE FUNCTION AND ENERGY OF… PHYSICAL REVIEW A 73, 022503 共2006兲

冕 0
1
dx␴i共x兲 = 0,
␾k共x,t兲 = xk−1 for k = 1,n,

冕 0
1
dx␩i共x兲 = wi . 共A10兲 ␾k共x,t兲 = 共1 − t兲冑x + txk−1−nln共x兲 for k = n + 1,2n.
共A12兲
If xk are not exactly the nodes, but are sufficiently close, then
according to Ref. 关29兴, the iteration xi → x̃i,

冕0
1
dx␴i共x兲 At t = 0 , ␾共x , 0兲 are polynomials in 冑x. Therefore one ob-
tains xi = y 2i where y i are nodes for Gauss-Legendre quadra-
共A11兲
冕
x̃i = xi + 1 ,
ture. By slowly changing t from 0 one finds the solution at
dx␩i共x兲 t = 1. In the actual numerical calculations we found that the
0
steps ti = i / 100 were sufficiently small for the above iteration
converges to nodes, the solution of Eqs. 共A3兲 and 共A4兲. The to converge. This generalized Gaussian quadrature can also
only problem now is to find a sufficiently good initial value be constructed for other types of functions including various,
for xi. For this one constructs a homotopy ␾k共x , t兲 such that even nonintegrable, singularities.

关1兴 K. Pachucki, Phys. Rev. A 71, 012503 共2005兲.
关2兴 V. Korobov and A. Yelkhovsky, Phys. Rev. Lett. 87, 193003
共2001兲.
关3兴 K. Pachucki, Phys. Rev. Lett. 84, 4561 共2000兲; J. Phys. B 35,
3087 共2002兲.
关4兴 V. A. Yerokhin, P. Indelicato, and V. M. Shabaev, Phys. Rev. A
71, 040101共R兲 共2005兲.
关5兴 G. W. F. Drake and W. C. Martin, Can. J. Phys. 76, 679
共1998兲.
关6兴 V. I. Korobov, Phys. Rev. A 66, 024501 共2002兲.
关7兴 Z. -C. Yan and G. W. F. Drake, Phys. Rev. A 52, 3711 共1995兲;
Z. -C. Yan, M. Tambasco, and G. W. F. Drake, ibid. 57, 1652
共1998兲.
关8兴 P. J. Pelzl, G. J. Smethells, and F. W. King, Phys. Rev. E 65,
036707 共2002兲; D. M. Feldman, P. J. Pelzl, and F. W. King, J.
Math. Phys. 39, 6262 共1998兲.
关9兴 Z. -C. Yan and G. W. F. Drake, J. Phys. B 30, 4723 共1997兲.
关10兴 J. Komasa, Chem. Phys. Lett. 363, 307 共2002兲.
关11兴 K. Pachucki and J. Komasa, Phys. Rev. Lett. 92, 213001
共2004兲.
关12兴 K. Pachucki, M. Puchalski, and E. Remiddi, Phys. Rev. A 70,
032502 共2004兲.
关13兴 K. Pachucki and M. Puchalski, Phys. Rev. A 71, 032514
共2005兲.
关14兴 E. Remiddi, Phys. Rev. A 44, 5492 共1991兲.
关15兴 Z. -C. Yan and G. W. F. Drake, Phys. Rev. Lett. 81, 774
共1998兲.
关16兴 R. J. Drachman, J. Phys. B 14, 2733 共1981兲.
关17兴 F. W. King D. G. Ballageer, D. J. Larson, P. J. Pelzl, S. A.
Nelson, T. J. Prosa, and B. M. Hinaus,Phys. Rev. A 58, 3597
共1998兲.
关18兴 R. A. Sack, C. C. J. Roothaan, and W. Kołos, J. Math. Phys. 8,
1093 共1967兲.
关19兴 V. I. Korobov, J. Phys. B 35, 1959 共2002兲.
关20兴 F. E. Harris, A. M. Frolov, and V. H. Smith, Jr., J. Chem. Phys.
121, 6323 共2004兲.
关21兴 K. Pachucki, Phys. Rev. A 66, 062501 共2002兲.
关22兴 G. Ewald et al., Phys. Rev. Lett. 93, 113002 共2004兲.
关23兴 F. W. King, J. Chem. Phys. 99, 3622 共1993兲.
关24兴 F. E. Harris, A. M. Frolov, and V. H. Smith, J. Chem. Phys.
119, 8833 共2003兲.
关25兴 G. Büsse, H. Kleindienst, and A. Lüchow, Int. J. Quantum
Chem. 66, 241 共1998兲.
关26兴 S. A. Blundell, P. J. Mohr, W. R. Johnson, and J. Sapirstein,
Phys. Rev. A 48, 2615 共1993兲.
关27兴 V. M. Shabaev, Phys. Rep. 356, 119 共2002兲.
关28兴 A. N. Artemyev V. M. Shabaev, V. A. Yerokhin, G. Plunien,
and G. Soff, Phys. Rev. A 71, 062104 共2005兲.
关29兴 J. Ma, V. Rokhlin, and S. Wandzura, SIAM 共Soc. Ind. Appl.
Math.兲 J. Numer. Anal. 33, 971 共1996兲.

022503-7