Schrödinger equation solved for the hydrogen molecule with unprecedented accuracy

Krzysztof Pachucki1, a) and Jacek Komasa2, b)

1)
Faculty of Physics, University of Warsaw, Pasteura 5, 02-093 Warsaw,
Poland
2)
Faculty of Chemistry, Adam Mickiewicz University, Umultowska 89b,
61-614 Poznań, Poland

(Dated: 14 April 2016)

The hydrogen molecule can be used for determination of physical constants, including
the proton charge radius, and for improved tests of the hypothetical long range force
between hadrons, which require a sufficiently accurate knowledge of the molecular
levels. In this work we perform the first step toward a significant improvement in
theoretical predictions of H2 and solve the nonrelativistic Schrödinger equation to
the unprecedented accuracy of 10−12 . We hope that it will inspire a parallel progress
in the spectroscopy of the molecular hydrogen.

a)
Electronic mail: krp@fuw.edu.pl
b)
Electronic mail: komasa@man.poznan.pl

1
I. INTRODUCTION

The spectroscopy of simple atomic systems like hydrogen1,2 , hydrogenic ions3 , muonic
hydrogen4 , muonium, and positronium has been used to determine fundamental physical
constants and to test the quantum electrodynamic theory. Although experiments for more
complicated atomic systems like helium or lithium can be as accurate as for hydrogen, the
precision of theoretical predictions, at the moment, is not sufficient to determine physical
constants, such as the fine structure constant α, the Rydberg constant (Ry), or the absolute
value of the nuclear charge radius. In contrast, the hydrogen molecule, thanks to its sim-
plicity, has already been used for the most accurate determination of the deuteron magnetic
moment from NMR measurements5 , and for the studies of the unknown hypothetical fifth
force at the atomic scale6 . The precision achieved recently for transition frequencies, of an
order of 10−4 cm−1 , has been verified by a series of measurements, which resulted in strong
bounds on the fifth force. We will argue in this work that it is possible to achieve 10−6 cm−1
accuracy for energy levels of the hydrogen molecule, which not only will improve tests of
quantum electrodynamic theory and will put stronger bounds on the fifth force, but also
will allow a resolution of the proton charge radius puzzle which stands as a violation of the
Standard Model of fundamental interactions7 .

The improvement in theoretical predictions for the hydrogen molecule can be achieved by
the calculation of the yet unknown higher order α4 Ry quantum electrodynamic correction
and by a more accurate solution of the nonrelativistic Schrödinger equation. This second
improvement is performed in this work, while the calculation of QED effects is in progress.
The solution of the Schrödinger equation for the hydrogen molecule has been pursued al-
most since the beginning of the quantum mechanic theory. Over time, there have been many
contributions to the development of methods that would ensure increased precision of theo-
retical predictions. Heitler and London8 , James and Coolidge9 , Kolos and Wolniewicz10 , and
many others have made their marks on the history of research on H2 . Every breakthrough
in the precision of theoretical predictions has been related to the progress in computational
techniques. Fig. 1 illustrates the progress made over many decades in the precision of the
theoretical determination of the nonrelativistic dissociation energy D0 for H2 .

Not always the results of calculations have been in agreement with the measured values,
which has questioned the validity of the theoretical approach. For example, in 1964 Kolos

2
 104
103
102
101 [9]
δD0/cm-1

100
10-1
10-2 [14]
10-3 [21]
[18]
10-4
[24]
10-5
1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
Year

FIG. 1. The accuracy of theoretical predictions of the nonrelativistic dissociation energy D0 of H2

versus time, with the linear fit on the logarithmic scale.

and Wolniewicz11 solved variationally the nonadiabatic Schrödinger equation for the hydro-
gen molecule. The calculated dissociation energy appeared to be higher than the measured
one12 , which was in contradiction with the variational principle. Six years later, Stwalley13
measured D0 again with the accuracy increased to a few tenths of reciprocal centimeter
and obtained a value higher than previously by 5 cm−1 , in agreement with the Kolos and
Wolniewicz result14 .
Further progress in theoretical predictions was related to the calculations of the lead-
ing relativistic corrections and to approximate treatment of QED effects15–18 . Later on,
due to rapid development of computer power, the methods based on exponentially cor-
related Gaussian (ECG) functions have been developed both in the Born-Oppenheimer
approximation19–22 and in the direct nonadiabatic approach23,24 . Very recently, we have in-
troduced a nonadiabatic perturbation theory (NAPT), which allowed the accuracy of about
10−3 cm−1 to be achieved for all the rovibrational levels of H2 and isotopomers25 . How-
ever, the complexity of NAPT in the higher order of electron-nucleus mass ratio26 makes
further improvements in accuracy quite complicated. For this reason, we propose another

3
approach to the direct solution of the Schrödinger equation resorting to the special integra-
tion technique for explicitly correlated exponential functions and we present its first results
in this work. Over fifty years after the pioneering work of Kolos and Wolniewicz27 employing
nonadiabatic exponential wave function, we approach the problem of solving the four-body
Schrödinger equation. Our new formulation aims at the precision level of 10−7 cm−1 for D0 .

II. THEORY

The main purpose of this work is to solve accurately the stationary Schrödinger equation
ĤΨ = EΨ for a diatomic molecule with the nuclei of charge ZA and ZB and finite masses
MA and MB
1 1 1 1
Ĥ = − ∇A2 − ∇B2 − ∇21 − ∇2
2 MA 2 MB 2 me 2 me 2
ZA ZB 1 ZA ZA ZB ZB
+ + − − − − , (1)
rAB r12 r1A r2A r1B r2B
using the variational approach. The trial wave function
K
X
~ A, R
Ψ(~r1 , ~r2 , R ~ B) = ~ A, R
ck Ŝ ψ{k} (~r1 , ~r2 , R ~ B) (2)
k=1

is expanded in properly symmetrized (Ŝ), four-particle basis of exponential functions

k0 k1 k2 k3 k4 k5
ψ{k} = exp [−α rAB − β (ζ1 + ζ2 )]rAB r12 η1 η2 ζ1 ζ2 , (3)

where ζi = riA + riB and ηi = riA − riB are the coordinates closely related to the prolate
spheroidal coordinates of i-th electron, rij are the interparticle distances, α and β are non-
linear variational parameters, and ki are non-negative integers collectively denoted as {k}.
For its resemblance to the electronic James-Coolidge function, we call this basis function the
nonadiabatic James-Coolidge (naJC) function.
Application of the naJC function for evaluation of the matrix elements leads to a cer-
tain class of integrals. Efficient evaluation of these integrals has become feasible since the
discovery of the analytic formulas28,29 and the differential equation30,31 which is satisfied by
the master integral
d3 ρ 1 d3 ρ 2 d3 ρ 3
Z Z Z
g=
4π 4π 4π
e−w1 ρ1 −w2 ρ2 −w3 ρ3 −u1 ρ23 −u2 ρ31 −u3 ρ12
. (4)
ρ23 ρ31 ρ12 ρ1 ρ2 ρ3

4
This differential equation has the form:
∂g 1 ∂σ
σ + g + Pa = 0 , (5)
∂a 2 ∂a
where a is ui or wi , and where

σ = u21 u22 w32 + u22 u23 w12 + u21 u23 w22 + w12 w22 w32

+u21 w12 (u21 + w12 − u22 − u23 − w22 − w32 )

+u22 w22 (u22 + w22 − u21 − u23 − w12 − w32 )
+u23 w32 (u23 + w32 − u22 − u21 − w12 − w22 ). (6)

The inhomogeneous term Pa is a combination of several logarithmic functions31 . Equa-

tion (5) enables the analytic evaluation of all necessary integrals. For example, when using
the naJC basis, the master integral g takes the following form
d rAB d3 r1B d3 r2B e−t rAB −u (r1A +r1B )−w (r2A +r2B )
Z 3

(4 π)3 rAB r12 r1A r1B r2A r2B

"
2u 2w



1 ln t+u+w ln t+u+w
= − −
4uw t−u+w t+u−w
   
2uw 2 (u+w) #
ln (u+w)(t+u+w) ln t+u+w
+ + . (7)
t+u+w t−u−w

The integrals with additional positive powers of interparticle distances are obtained by
differentiation of g with respect to the corresponding nonlinear parameter, which in turn
can be expressed in terms of straightforward algebraic recursion relations. These relations
are too long to be written explicitly here, but all of them can be derived using Eq. (5).

III. NUMERICAL RESULTS

Results of our calculations for H2 are presented in Tab. I in the form a sequence of energies
resulting from increasing length (K) of expansion (2). The selection of K was made on the
basis of the saturation of consecutive ’shells’ limited by 5k=1 ki ≤ Ω with k0 fixed at 30.
P

The observed regular convergence, obeying the inverse power low (∼K −4 ), permits a firm
extrapolation to the complete basis set as well as an estimation of uncertainty. The final
value agrees well with the previous estimation of −1.164 025 030 84(6) a.u. obtained by
Bubin et al.24 but has a significantly smaller uncertainty.

5
TABLE I. Convergence of the Schrödinger equation eigenvalue E (in a.u.) and of the corresponding
dissociation energy D0 (in cm−1 ) for H2 with the size of the basis set.

Ω K E D0

10 36642 −1.164 025 030 821 4 36 118.797 732 57

11 53599 −1.164 025 030 870 9 36 118.797 743 43
12 76601 −1.164 025 030 880 4 36 118.797 745 52
13 106764 −1.164 025 030 882 5 36 118.797 745 97
∞ ∞ −1.164 025 030 884(1) 36 118.797 746 3(2)

Further increase in the accuracy of eigenvalue of the four-body Schrödinger equation is

feasible, but the problem we face is the lack of the efficient parallel code in multiprecision
arithmetics for the LDLT matrix decomposition with pivoting, which results in a long com-
putation time. However, current uncertainties in the electron-proton (proton-deuteron) mass
ratio and in the Rydberg constant are much more significant than those due to numerical
uncertainties. For example, the CODATA 201432 electron-proton mass ratio has a relative
uncertainty of 9.5 · 10−11 , which affects the eigenvalue of H2 at the level of 4.3 · 10−12 a.u. and
the corresponding dissociation energy at 8.5 · 10−7 cm−1 . Similarly, the current uncertainty
in the Rydberg constant affects the conversion of D0 value from a.u. to reciprocal centime-
ters at the level of 2.1 · 10−7 cm−1 . This situation indicates that one cannot exclude the
possibility to determine the electron-proton mass ratio from future high precision studies of
H2 .

IV. CONCLUSIONS

The approach based on explicitly correlated exponential functions and the obtained re-
sults pave the way to a significant progress in the theory of the hydrogen molecule. Not only
the ground but also all the rovibrational levels can be obtained with a similar approach. From
the experience with James-Coolidge functions in the Born-Oppenheimer approximation33 ,
we know that they work very well up to the distances as large as 12 a.u., therefore they can
be used also for all excited states. Nevertheless, we see the possibility of using exponential

6
functions with arbitrary nonlinear parameters, for a more efficient representation of the total
wave function. In fact, the main problem we face for excited states, is the increased size of
the basis which can be solved by tuning values of all nonlinear parameters in exponential
functions and using a double or triple basis with different parameters.

Considering the leading relativistic corrections, they can be expressed in terms of an

expectation value with the nonrelativistic wave function, so their evaluation does not pose
a significant problem, and they have already been calculated in the BO approximation.
The leading quantum electrodynamic effects are more complicated due to Bethe logarithm
contribution which involves the logarithm of the nonrelativistic Hamiltonian. Its calcula-
tion beyond the BO approximation might be problematic. However, much more challeng-
ing is the calculation of the higher order α4 Ry contribution, which apart from hydrogen,
was calculated only for He atom. The dominant component of the α4 Ry—the radiative
correction—was calculated in Ref. 21 to be 16 · 10−4 cm−1 , whereas its remaining part was
estimated as 8 · 10−4 cm−1 . Another estimate of the non-radiative contribution, made on the
basis of an assumption that it is equal to 2 α2 multiplied by the known leading relativistic
correction to D0 (≈ 0.5 cm−1 ), leads to a much smaller value of 5 · 10−5 cm−1 . Because of
this discrepancy, a more accurate estimation of the α4 Ry correction and its uncertainty can
be obtained only by explicit calculations. In order to achieve the 10−6 accuracy, it would
probably be necessary to approximately evaluate also the α5 Ry correction as it is enhanced
by the presence of ln α−2 factors.

At the level of accuracy of 10−6 cm−1 , the proton charge radius, which contributes about
1.2 · 10−4 cm−1 to the dissociation energy of H2 , can be determined with 0.5% precision, pro-
vided that equally accurate measurement is performed. This certainly will resolve the proton
charge radius discrepancy, which is at the level of 4%, and will open a new era in precision
quantum chemistry. Regarding the tests of hypothetical forces, which are beyond those
in the Standard Model, the atomic scale is the natural region for the long range hadronic
interactions. Moreover, it has recently been shown that vibrational levels of the hydrogen
molecule34 are particularly sensitive to the interactions beyond the Coulomb repulsion be-
tween nuclei. So any deviation between hopefully improved theoretical predictions and the
experiment may signal a new physics.

7
ACKNOWLEDGMENT

We thank Prof. Bogumil Jeziorski for valuable comments to the manuscript. This research
was supported by the National Science Center (Poland) Grants No. 2012/04/A/ST2/00105
(K.P.) and 2014/13/B/ST4/04598 (J.K.), as well as by a computing grant from the Poznan
Supercomputing and Networking Center, and by PL-Grid Infrastructure.

REFERENCES

1
C. G. Parthey, A. Matveev, J. Alnis, B. Bernhardt, A. Beyer, R. Holzwarth, A. Maistrou,
R. Pohl, K. Predehl, T. Udem, T. Wilken, N. Kolachevsky, M. Abgrall, D. Rovera, C. Sa-
lomon, P. Laurent, and T. W. Hänsch, Phys. Rev. Lett. 107, 203001 (2011).
2
C. Schwob, L. Jozefowski, B. de Beauvoir, L. Hilico, F. Nez, L. Julien, F. Biraben, O. Acef,
J.-J. Zondy, and A. Clairon, Phys. Rev. Lett. 82, 4960 (1999).
3
S. Sturm, F. Kohler, J. Zatorski, A. Wagner, Z. Harman, G. Werth, W. Quint, C. H.
Keitel, and K. Blaum, Nature 506, 467 (2014).
4
R. Pohl, A. Antognini, F. Nez, F. D. Amaro, F. Biraben, J. M. R. Cardoso, D. S. Covita,
A. Dax, S. Dhawan, L. M. P. Fernandes, A. Giesen, T. Graf, T. W. Hänsch, P. Indelicato,
L. Julien, C.-Y. Kao, P. Knowles, E.-O. Le Bigot, Y.-W. Liu, J. A. M. Lopes, L. Ludhova,
C. M. B. Monteiro, F. Mulhauser, T. Nebel, P. Rabinowitz, J. M. F. dos Santos, L. A.
Schaller, K. Schuhmann, C. Schwob, D. Taqqu, J. a. F. C. A. Veloso, and F. Kottmann,
Nature 466, 213 (2010).
5
M. Puchalski, J. Komasa, and K. Pachucki, Phys. Rev. A 92, 020501 (2015).
6
W. Ubachs, W. Vassen, E. J. Salumbides, and K. S. E. Eikema, Annalen der Physik 525,
A113 (2013).
7
R. Pohl, R. Gilman, G. A. Miller, and K. Pachucki, Annu. Rev. Nucl. Part. Sci. 63, 175
(2013).
8
W. Heitler and F. London, Z. Phys. 44, 455 (1927).
9
H. M. James and A. S. Coolidge, J. Chem. Phys. 1, 825 (1933).
10
W. Kolos and L. Wolniewicz, J. Chem. Phys. 41, 3663 (1964).
11
W. Kolos and L. Wolniewicz, J. Chem. Phys. 41, 3674 (1964).
12
G. Herzberg and A. Monfils, J. Mol. Spectrosc. 5, 482 (1960).

8
13
W. C. Stwalley, Chem. Phys. Lett. 6, 241 (1970).
14
W. Kolos and L. Wolniewicz, J. Chem. Phys. 49, 404 (1968).
15
W. Kolos and L. Wolniewicz, J. Mol. Spectrosc. 54, 303 (1975).
16
L. Wolniewicz, J. Chem. Phys. 78, 6173 (1983).
17
L. Wolniewicz, J. Chem. Phys. 99, 1851 (1993).
18
L. Wolniewicz, J. Chem. Phys. 103, 1792 (1995).
19
B. Jeziorski and K. Szalewicz, Phys. Rev. A 19, 2360 (1979).
20
J. Rychlewski, W. Cencek, and J. Komasa, Chem. Phys. Lett. 229, 657 (1994).
21
K. Piszczatowski, G. Lach, M. Przybytek, J. Komasa, K. Pachucki, and B. Jeziorski, J.
Chem. Theory Comput. 5, 3039 (2009).
22
J. Mitroy, S. Bubin, W. Horiuchi, Y. Suzuki, L. Adamowicz, W. Cencek, K. Szalewicz,
J. Komasa, D. Blume, and K. Varga, Rev. Mod. Phys. 85, 693 (2013).
23
S. Bubin and L. Adamowicz, J. Chem. Phys. 118, 3079 (2003).
24
S. Bubin, F. Leonarski, M. Stanke, and L. Adamowicz, Chem. Phys. Lett. 477, 12 (2009).
25
K. Pachucki and J. Komasa, J. Chem. Phys. 130, 164113 (2009).
26
K. Pachucki and J. Komasa, J. Chem. Phys. 143, 034111 (2015).
27
W. Kolos and L. Wolniewicz, Rev. Mod. Phys. 35, 473 (1963).
28
D. M. Fromm and R. N. Hill, Phys. Rev. A 36, 1013 (1987).
29
F. E. Harris, Phys. Rev. A 55, 1820 (1997).
30
K. Pachucki, Phys. Rev. A 80, 032520 (2009).
31
K. Pachucki, Phys. Rev. A 86, 052514 (2012).
32
P. J. Mohr, D. B. Newell, and B. N. Taylor, ArXiv e-prints (2015), arXiv:1507.07956
[physics.atom-ph].
33
K. Pachucki, Phys. Rev. A 82, 032509 (2010).
34
E. J. Salumbides, J. C. J. Koelemeij, J. Komasa, K. Pachucki, K. S. E. Eikema, and
W. Ubachs, Phys. Rev. D 87, 112008 (2013).