Lecture 8: Variational method:

the Helium atom (10/13/2005)

The main goal of the lecture will be to calculate a good enough estimate for
the binding energy of the Helium atom an example how the variational
method from the previous lecture may be applied in realistic cases.
Additional reading if you wish: Griffiths 7.2, Gasiorowicz 18

A few comments on chaotic systems

As you know, the classical Kepler problem - the Hamiltonian describing two
particles with the 1/r potential in between them can be solved exactly; the
trajectories are ellipses or hyperbolae or parabolae. The analogous quantummechanical problems is called the Hydrogen atom and the spectrum of the
Hamiltonian can also be found analytically. Such special solvable problems
are also called integrable systems.
You probably realize that this is a special feature of a few simple systems.
The three-body problem describing the dynamics of three planets is not
solvable analytically in classical physics; it is chaotic (see the picture). The
system never returns to the same state, although it can return arbitrarily
closely to it. According to ergodic theory, the time evolution will eventually
cover nearly all points in the phase space with the same energy and other
conserved quantities uniformly.
In quantum mechanics, there is no chaos (except for quantum chaos).
Even difficult Hamiltonians have a well-defined spectrum and the corresponding eigenstates are exactly stationary, in agreement with the general rule.
However, their wavefunctions may be pretty unpredictable (see a picture).
The simplest moral counterpart of the three-body problem in quantum mechanics is the Helium atom which is made of 1 nucleus and 2 electrons orbiting
it. The energy of its ground state is, according to experiments,
E0 = 78.975 eV
It means that you need to pump this energy to the Helium atom to kick both
electrons so that they leave the nucleus with a small velocity (ionization).
How close can we get to this number using the variational method?

Neglecting some small effects: our Hamiltonian

We must first define the Hamiltonian for the Helium atom. This includes the
first layer of approximation; another source of inaccuracy will be our variational solution, of course. The Hamiltonian, acting on functions (~r1 , ~r2 ),
is
!
Z
1
h
2
e2
Z
2
2

+
H=
( + 2 )
2m 1
40 r1 r2 |~r1 ~r2 |

The terms include the kinetic energy of the first electron, the kinetic energy of
the second electron, the attractive interaction of the first and second electron
with the nucleus that contains Z protons, and the repulsive self-interaction
of two electrons. For Z = 2 we solve the Helium atom but we may also use
the same Hamiltonian with Z = 1, 3, 4 . . . to solve various ions of hydrogen,
lithium, beryllium, and so forth. The Hamiltonian above is not quite exact
because it neglects various small effects (contributing about 0.1 percent) such
as

nuclear motion; we should be using a reduced mass of the electrons,

but let us ignore these issues because the nuclei are much heavier and
do not move much
relativistic effects implied by the Dirac equation; the speed of electrons
divided by the speed of light is comparable to the squared fine structure
constant and the fine structure constant itself is still a small enough
number (1/137.036) to trust non-relativistic equations
spin-orbit interactions; they are also implied by the Dirac equation,
and you may interpret them as the interaction between different components of the magnetic moments
orbit-orbit interactions; the magnetic interaction of the orbital magnetic moments of the electrons, roughly speaking
The ground state will be a spin singlet (S = 0):
1
= (| i| i | i| i)
2
The Fermi-Dirac statistics implies that the spatial wavefunction must be
symmetric:
(~r1 , ~r2 ) = (~r2 , ~r1 )
If the electron-electron interaction term were absent, the ground state could
be written in a factorized (but symmetrized) form
1
(~r1 , ~r2 ) = [n1 l1 m1 (~r1 )n2 l2 m2 (~r2 ) + n1 l1 m1 (~r2 )n2 l2 m2 (~r1 )] .
2
2

whose eigenvalue would simply be the sum

E = En1 + En2 .
Note that if (n1 , l1 , m1 ) = (n2 , l2 , m2 ), we should change the normalization of
the wavefunction.

Bounds on the ground state energy

What does it mean for the ground state? Recall that the Hydrogen had the
energy of 13.6 eV. Multiply it by Z 2 = 4 because the nucleus is stronger
and the energy scales like Z 2 and by 2 from two electrons to see that
E0 > 8 13.6 eV = 108.8 eV.
This is a lower bound on the energy because the repulsive interaction between
the electrons will definitely raise the minimal achievable energy. We can also
find an upper bound for the energy. Last time we showed that Eg0 + Eg1 , the
first two terms in the perturbative expansion, always overestimate the total
energy because E2 + . . . is negative. So let us treat the interaction between
the two electrons as a perturbation. Its expectation value is
E =



+


e2


 0 (~
r1 , ~r2 )
0 (~r1 , ~r2 ) 
 40 |~
r1 ~r2 | 

and can also be written as

=

e2
40

d 3 r1

d3 r2 |100 (r1 )|2

e|100 (r1 )|2 V (~r1 )d3 x

1
|100 (r2 )|2
|~r1 ~r2 |

i.e. as the integrated product of the charge distribution of the first particle
and the potential from the second particle
V (~r1 ) =

d3 r2 |100 (~r2 )|2

1
e
40 |~r1 ~r2 |

This way of phrasing the value of the interaction energy is useful once we
realize that the potential from a spherical shell of charge equals the potential
from the same charge at the origin at points outside the sphere, and it is constant inside where the constant is determined by continuity of the potential.
In fact, we can therefore
calculate V (r1 ) exactly for the Hydrogen ground
R
state as V (r1 ) = 0 dV (r1 , r2 ) where we integrate over r2 and
e 4r22 dr2
dV (r1 ) =
40 r1

Z3
e2Zr2 /a ,
a3
!

r1 r2
3

V (r1 ) =
=

E =
=

e 4r22 dr2 Z 3
e2Zr2 /a ,
r1 r2
40 r2
a3
 Z

Z
e 4Z 3 1 r1 2 2Zr1 /a
2Zr2 /a
r e
dr2 +
r2 e
dr2
40 a3 r1 0 2
r1
i
e2 h
1 (1 + Zr1 /a)e2Zr1 /a
40 r1
Z 2
i
r1 dr1 h
e2 Z 3 Z
2Zr1 /a
d
e2Zr1 /a
1

(1
+
Zr
/a)e
1
3
40 a
r1
!
2
5
e
Z
5
= Z (13.6 eV)
8 40 a
4
!

That has been a lot of mathematical symbols and integrals, but we calculated the first correction to the energy E from the self-interaction of two
electrons. If we add it to the zeroth order calculation from the doubled
Hydrogen atom, we obtain
E0 < 108.8 eV + E = 74.83 eV.
Note that the first two terms give us an upper bound on the energy because
the next (second order) correction is negative, as we discussed last time. Also
note that the result 74.83 eV is closer to the right value (5.2 percent error)
than the previous zeroth order estimate because more work often implies
more accurate results.

Improving the wavefunction

The result whose error is five percent is good but it is not good enough. To
get more accurate results, we will have to use more complicated wavefunctions. So far we only used the wavefunction of two independent hydrogen-like
electrons. What is the qualitative improvement we should do? It is based
on the observation that one electron, if it is far enough from the nucleus,
essentially sees a smaller charge of the nucleus because the charge of the
nucleus is partially shielded or neutralized by the other electron. We should
emphasize that we are not changing the exact Hamiltonian, of course; instead, we want to change the trial functions. Our new trial function will be
the exact eigenstate of a new approximate Hamiltonian in which we neglect
the self-interaction, but try to compensate it by changing the effective value
of Z to a smaller value Z :
H0

h
2
e2
=
(21 + 22 )
2m
40

Z Z
+
r1
r2

Thats not quite the same Hamiltonian as before, but we assume that they
could have rather similar eigenstates because the reduced value of Z may
4

have a similar effect on the wavefunction as the self-interaction of the electrons. But H0 can be solved exactly. The only change from the original
hydrogen problem are extra stars (which dont mean complex conjugation!):
(~r1 , ~r2 ) = 100 (r1 )100 (r2 ),

100 (r) =

Z 3 Z r/a
e
a3

Its total H0 eigenvalue is

E0 = 2Z 2 13.6 eV.

Returning to the exact Hamiltonian

This Hamiltonian H0 may of course be converted to the exact original Hamiltonian if we add a perturbation
e2
V =
40

1
Z Z Z Z
+

.
r1
r2
|~r1 ~r2 |

We may therefore start with the better starting point H0 depending on

Z and use the first-order perturbation theory to estimate the energy. There
will be several terms,
e2
E = 2(Z Z )
40

! 

1
5
+ Z (13.6 eV)
r
4

But the expectation value of 1/r is something we know from the Hydrogen
atom;
 
1
Z
=
r
a
which allows us to calculate the answer, up to the first order terms, as
5
E = 2Z 2 4Z (Z Z ) + Z 13.6 eV
4


The variational recipe tells us that we should minimize this energy (although
it is just an estimate of it) with respect to the parameter Z :
0=

5
dE

4Z

4Z
+
8Z
+
dZ
4

Z = Z

5
16

Substituting it into E, we obtain

25
5
E = 2Z 2 + Z
13.6 eV
4
128


For Z = 2 this gives us 77.49 eV which is 1.9 percent higher than the
experimental value. Again, we were guaranteed to overshoot the right value
5

because we used two things variational method and the first-order perturbative calculation both of which overshoot the ground state energy.
You may think that it was too much work for such a poor improvement
of accuracy. And you may be right. Further improvements can only be done
once we admit some correlations between the two electrons wavefunctions.
The relevant trial wavefunctions are either more general ones that do not
factorize at all, or they factorize to a symmetrized product of two different functions of ~r1 and ~r2 . Perturbation theory never leads to this kind of
asymmetry.

A heroic 1958 calculation

Pekeris, in Phys Rev. 112, 1649 (1958), used a trial function with 1075 terms
and his or her obtained results were within experimental errors.

A wrong result for the hydrogen ion

For the case of the hydrogen ion, Z = 1, we would obtain the energy
5
25
7
E = 2Z 2 + Z
13.6 eV 1 +
13.6 eV
4
128
128


which would indicate that the ion is unbound because it can decay to a free
electron and the hydrogen atom. The reason is the 7/128 times the Rydberger energy we see above which equals +0.74 eV. Actually, this answer
is incorrect. The actual hydrogen ion exists, it is bound by 0.055 eV, and
the ion is easily seen in solar spectra. It is sometimes enough to make a
less-than-one-electronvolt error in order to spoil the qualitative results. Note
that the actual number of bound states that exist is likely to be larger, but
not smaller, than the number estimated by the variational method because
of its bias; we would never predict a bound state if it did not exist.

Helium: excited states

The term excited states is probably too ambitious. We will look at the
first excited state of the Helium atom only. Nevertheless, there are some
interesting phenomena already going on at this level.
The quantum numbers of the ground state we discussed previously may
be encoded in the following hieroglyph:
1

S0 = (2S+1) LJ

The superscript 1 describes the degeneracy (2S + 1): it is 1 for our singlet,
3 for a triplet, and so forth. The letter S informs us about
L

0(S),

1(P ),

2(D),

3(F ),

4(G),

5(H),
6

6(I), . . .

and the subscript 0 is the total J.

While the ground state was a singlet, the first excited state can either be
a singlet or a triplet. The Fermi statistics implies that the spatial wavefunction will be symmetric or antisymmetric, respectively. Using the hydrogen
notation, the two heliums electrons can either occupy
(1s)(2s)

or

(1s)(2p)

which means that the total orbital angular momentum L is either 0(S) or
1(P ). Adding S and L using the rules for the addition of angular momenta,
the total J can take several values and the candidates for the first excited
state are therefore
1

S0 ,

S1 ,

P1 ,

P0 or 1 or 2

Again, the zeroth approximation is to take the simple (anti)symmetrized

pairs of Hydrogen wavefunctions with (nlm) = (100) or (2lm) as our trial
functions. The energy correction from the e e interaction (calculated as
the expectation value) thus depends on l:
i
h
e2 1
El =
d3 r1 d3 r2 100 (~r1 )2lm (~r2 ) 100 (~r2 )2lm (~r1 )
40 2
1

[100 (~r1 )2lm (~r2 ) 100 (~r2 )2lm (~r1 )]

(1)
|~r1 ~r2 |

The factor 1/2 came from the normalization factors 1/ 2 of the wavefunctions. The signs in both factors are identical. If we now expand 2 2 = 4
terms above, there will be 2 terms of one type and 2 terms of the other type:
Z

El = Dl El
where
e2
1
d 3 r1 d 3 r2
|100 (~r1 )|2 |2lm (~r2 )|2 ,
40
|~r1 ~r2 |
Z
e2
1
=
d 3 r1 d 3 r2
100 (~r1 )100 (~r2 )2lm (~r2 )2lm (~r1 )
40
|~r1 ~r2 |

Dl =
El

The first integral Dl is manifestly positive while the second integral El is

not manifestly positive but it is actually still positive. The positive sign of
El implies that the triplet state whose spatial wavefunction is antisymmetric
the sign minus used for will be moved below the singlet state whose
spatial wavefunction is symmetric. This is also the reason for Hunds rules:
when all other things are equal, then the highest spin states are preferred
(because they have a lower energy).
7

The original degenerate double-hydrogen states, neglecting the electrons interaction, were (1s)(2s) and (1s)(2p), both of them having energy
of 68.0 eV. Once we incorporate the electrons interaction, the energy is
shifted up but the degeneracy is lifted, and the levels starting with those
with the lowest energy are (in electronvolts) draw also a simple picture:
3

S1

(59.2),

S0

(58.4),

P0,1,2

(58.0),

P1

(57.8).

Spin-spin interactions
At the beginning, we mentioned that we neglected various spin-spin interactions, including the fine structure (and hyperfine structure, which comes from
the interactions of the nuclear magnetic moments with the electrons magnetic moments). These terms are very small, of order 1 meV, because they
are essentially proportional to v/c where c is the speed of light. And (v/c)
goes like O(Z)2. The interactions El we computed above are much more
important, by two orders of magnitude, and they can still be interpreted indirectly as a spin-spin interaction whose origin is not the magnetic interaction
but rather the Fermi-Dirac statistics. The original formula El = Dl El
may be written as
1
El = Dl (1 + ~1 ~2 )El .
2
Why is it the same formula? We need to prove that (1 + ~1 ~2 )/2 is
equal to +1 for the singlet (symmetric spatial wavefunction) and 1 for the
triplet (antisymmetric spatial wavefunction). To prove it, you should first
understand that the total squared spin of S = S1 + S2 satisfies
~2 = S
~ 2 + 2S
~1 S
~2 + S
~2
S
1

which allows us to write the inner product (a sum of three terms, as always,
each of them being a tensor product of two operators or matrices acting on
two different sets of degrees of freedom)


~1 S
~2 = 1 S 2 S 2 S 2
S
1
2
2
2
2
2
Noting that S1 = S2 = 3
h /4 and S 2 = 0 for the S = 0 singlet and S 2 = 2
h2
for the S = 1 triplet, we see that the inner product has eigenvalue
~1 S
~2 =
S

3
h2 /4 for the singlet
+
h2 /4 for the triplet

Then, ~1 2 has the same eigenvalues without the h

2 /4 factor, i.e. 3 for
the singlet and +1 for the triplet, which implies that (1 + ~1 ~2 )/2 is equal
to +1 for the singlet and 1 for the triplet.
Werner Heisenberg was the first guy who realized that the magnetic moments of the elementary particles were far too small to explain ferromagnetism. Instead, the Fermi-Dirac statistics is essential for ferromagnets.
8