Chapter 1

Time Independent S hrödinger

Equation
1.1 Revision of the S hrödinger equation
The S hrödinger equation is a time-dependent partial dierential equation.
In this introdu tory le ture, we will not onsider the time dependen e (that
will ome in le ture 3), and so we will instead onsider the simpler ase of
the time independent version. We an sometimes restri t things even more
by onsidering just one spatial dimension, in whi h ase, we only have a
reasonably straightforwards ordinary dierential equation to solve, subje t
to some set of boundary onditions. This is exa tly the type of problem that
you have solved analyti ally in the past.
We start by rewriting the time independent S hrödinger equation

~2 2
− ∇ ψ (r) + V (r) ψ (r) = Eψ (r) (1.1)
2me
in atomi ~ = me = e = 4πε0 = 1 [so from now on we shall
units, that is
always work in energy units of Hartrees (27.2eV ) and in length units of Bohr
radii (a0 = 0.529 Å)℄ and so :

1
− ∇2 ψ + V ψ = Eψ (1.2)
2
or in terms of the Hamiltonian operator b,
H

b = Eψ
Hψ (1.3)

We know that this is an example of an eigenvalue equation, and that for

a given potential V (r) and boundary onditions, there will in general be a
number of dierent possible solutions, orresponding to dierent eigenener-
gies Ei with orresponding eigenfun tions ψi .

7
8 CHAPTER 1. TIME INDEPENDENT SCHRÖDINGER EQUATION

The rst term in equation 1.1 orresponds to the kineti energy of the
state, and so we see that a smoothly varying wavefun tion (small ∇2 ψ ) will
generally be a lower energy state than a rapidly varying one. This an also
be related to the number of nodes (zero rossings) in the wavefun tion - a
smooth wavefun tion will have less nodes than a rapidly varying one.

Example 1 - parti le in innite square well

As an example, we an onsider the simple ase of a quantum parti le moving
in one dimension in an innitely deep square well:

V (x) = 0 −a ≤ x ≤ a
(1.4)
→ ∞ otherwise
whi h then gives the general solution of equation 4.1 as

ψ (x) = A cos (kx) + B sin (kx) (1.5)

with A and B being onstants given by the boundary onditions and

√
k= 2E (1.6)

If we now apply the boundary onditions that ψ (−a) = ψ (a) = 0 then

k = nπ/2a (1.7)

where n is an integer. For odd n we get cos (ka) = 0 and hen e A = 0 whilst
for even n we get sin (ka) = 0 and hen e B = 0. As the well is innitely deep,
then all solutions are bound states. As usual, bound states have quantized
energies, given in this ase by

π 2 n2
En = (1.8)
8a2
Note that as the potential has a denite parity then so do the eigenfun tions.
Note also that the ground state wavefun tion has no nodes, whilst the rst
ex ited state has one node, and su essive ex ited states have more nodes.
See gure 1.1 for a simple sket h of the three lowest energy solutions.
We an also solve analyti ally problems su h as the nite square well, or
the harmoni os illator, or the spheri al well, et . Remember that these an
be used to demonstrate the existen e of zero-point motion, tunnelling into
lassi ally forbidden regions and the Heisenberg Un ertainty Prin iple.

Example 2 - the hydrogen atom

In order to solve the S hrödinger equation for the hydrogen atom, we need
to deal with a 3D equation in spheri al polar oordinates. However, be ause
1.1. REVISION OF THE SCHRÖDINGER EQUATION 9

V
E3 ψ3

E2 ψ2

E1 ψ1
−a a x
Figure 1.1: Sket h of innite square well, showing lowest three eigenvalues
and orresponding eigenfun tions.
10 CHAPTER 1. TIME INDEPENDENT SCHRÖDINGER EQUATION

of the spheri al symmetry of the Coulomb potential, we an separate the

solution into a radial and an angular part:

ψ (r) = ψ (r, θ, φ) = Rnl (r) Ylm (θ, φ) (1.9)

where {n, l, m} are quantum numbers (integers) whi h hara terise the solu-
tion. If we substitute equation 1.9 into equation 4.1 and apply the method of
separation of variables, we get two equations - a 2D angular equation, whose
solution is given in terms of spheri al harmoni s and a 1D radial equation:

 
1 d2 χnl (r) l (l + 1)
− + V (r) + χnl (r) = Eχnl (r) (1.10)
2 dr 2 2r 2
where the substitution
χnl (r) = rRnl (r) (1.11)

is used both to simplify the equation and ease its interpretation. Remember
that the probability of nding the parti le in a small volume d3 r is

P r → r + d3 r = |ψ (r)|2 d3 r
2 2
= |ψ (r)| r dr · sin (θ) dθdφ (1.12)

= |χ (r)|2 dr · Ylm
2 (θ, φ) sin (θ) dθdφ

and so we see that the normalisation of χnl and Ylm an be hosen su h that

Z ∞
|χ (r)|2 dr = 1 (1.13)
0

i.e. |χ (r)|2 is the radial probability density.

Note that the se ond term in equation 1.10 looks very like a potential
energy term in the standard time independent S hrödinger equation, and so
is known as the ee tive potential:
l (l + 1)
Vef f (r) = V (r) + (1.14)
2r 2
and the
l(l+1)
2r 2 term is known as the entrifugal barrier.
What an we dedu e about the general form of χ (r)? Well, we know the
form of the Coulomb potential and so as long as V (r) → −∞ as r → 0 no
more qui kly than r −1 we get

d2 χ l (l + 1)
2
= χ (1.15)
dr r2
whi h gives a non-divergent solution as

χ (r → 0) ∼ r l+1 (1.16)
l
⇒ Rl (r → 0) ∼ r
1.1. REVISION OF THE SCHRÖDINGER EQUATION 11

R21
0.1

5 10
r/a0
R20
0.4 5
10
r/a0
R10

1.0

5 10
r/a0
Figure 1.2: Sket h of rst three radial eigenfun tions for the hydrogen atom.

Similarly, as r→∞ we get

d2 χ
= −2Eχ (r) (1.17)
dr 2
whi h gives two lasses of solution:

√
χ (r) ∼ e− √2|E|r E < 0 (bound states)
(1.18)
χ (r) ∼ e±i 2Er E > 0 (continuum)

For intermediate values of r we an either use more elaborate mathe-

mati s (whi h is possible for hydrogen but not for more omplex atoms) or
turn to numeri al solutions. We an therefore use this known asymptoti
behaviour to test our numeri al solutions - every bound state solution, or-
responding to an eigenenergy En , should tend to zero as r → ∞, and for
r → 0 should either tend to zero (for non-zero l) or tend to a usp (for l = 0).
See gure 1.2 for a sket h of the three lowest eigenenergy solutions whi h
demonstrates this behaviour. We now turn to the problem of how to solve
dierential equations numeri ally.
12 CHAPTER 1. TIME INDEPENDENT SCHRÖDINGER EQUATION

1.2 Revision of solving ordinary dierential equa-

tions numeri ally
Unfortunately, the number of ases where we an solve the S hrödinger equa-
tion exa tly are rather small. More usually, we are fa ed with the problem
of trying to nd numeri al solutions. As the S hrödinger equation is a linear
se ond-order dierential equation, this should not pose too mu h di ulty.
We therefore begin by revising what we already know about solving ordinary
dierential equations, before going on to explore parti ular te hniques that
are suitable for the S hrödinger equation.
Most methods are designed for rst-order dierential equations of the
general form

dy
= f (y (x) , x) (1.19)
dx
Se ond-order equations of the general form

d2 y
= f (y (x) , x) (1.20)
dx2
an be transformed into a pair of oupled rst-order equations using

dy
= z (x)
dx
dz
= f (y (x) , x) (1.21)
dx
and then solved using the same te hniques.
Most methods then pro eed to use an approximation to the derivative,
su h as the forward dieren e :
dy y (x + δx) − y (x)
≈ (1.22)
dx δx
the ba kwards dieren e :
dy y (x) − y (x − δx)
≈ (1.23)
dx δx
or the entred dieren e :
dy y (x + δx) − y (x − δx)
≈ (1.24)
dx 2δx
A straightforwards implementation of this approa h results in the Euler
method :

yn+1 ≈ yn + f (yn , xn ) h (1.25)

1.2. REVISION OF SOLVING ORDINARY DIFFERENTIAL EQUATIONS NUMERICALLY13

y(x)

x1 x2 x3 x4 x
Figure 1.3: Illustration of Euler's method - the derivative at ea h point xn
is extrapolated over the whole step length h to form the next point.

where the solution advan es from xn to xn+1 = xn + h = xn + δx and h is the

step size. See gure 1.3 for an illustration of the method. Comparison to the


Taylor series shows that this method has an error O h2 and hen e is alled


a rst-order method th
(an n -order method has an error term O hn+1 ).
Whilst the Euler method is the theoreti al basis of most s hemes, and as
su h is important, it should not be used as it stands as it suers from large
errors and is unstable. However, it an be improved upon to form the s hemes
we shall dis uss below.

1.2.1 The Runge-Kutta method

The Runge-Kutta method is a ommon lass of methods used for solving

many dierential equations. The simplest version improves upon the Euler
method by doing it twi e: we evaluate yn using a single Euler step of length
h and also using a midpoint step as follows:

k1 = f (xn , yn ) h
 
h k1
k2 = f xn + , yn + h (1.26)
2 2


yn+1 = yn + k2 + O h3

whi h is known as the se ond-order Runge-Kutta or midpoint method. It has

the advantage of being more stable than the Euler method, and so an be
used with a larger step size, and so requires less steps to span a given interval.
Unfortunately, the ost for this higher order method is that it requires two
fun tion evaluations of f (x) per step. Often, this is the most time onsuming
part of the al ulation. See gure 1.4 for an illustration of this method.
14 CHAPTER 1. TIME INDEPENDENT SCHRÖDINGER EQUATION

y(x)

x1 x2 x3 x4 x
Figure 1.4: Se ond-order Runga-Kutta method - the derivative at the start
of the step is used to nd a point half-way a ross the interval h and this
midpoint value is then used to improve the al ulation of the nal point at
the end of the step. The lled ir les represent nal values whi h are of
interest, whilst open ir les are intermediate values ki whi h are dis arded
at the end of ea h step.

This method an be extended to arbitrarily high orders. In pra ti e, it is

often found that the fourth-order method is the most e ient, in that higher
order methods require many more fun tion evaluations without bringing sig-
ni ant improvements in stability or step size. Note that a higher-order
method does not always have higher a ura y! The fourth-order Runge-
Kutta method is:

k1 = f (xn , yn ) h
 
h k1
k2 = f xn + , yn + h
2 2
 
h k2
k3 = f xn + , yn + h (1.27)
2 2
k4 = f (xn + h, yn + k3 ) h
k1 k2 k3 k4

yn+1 = yn + + + + + O h5
6 3 3 6
A signi ant advantage of the Runge-Kutta method is that it an be
used with a variable step size: at ea h step, an estimate of the error in
the solution an be made, and the step size redu ed if the error is larger
than some pres ribed toleran e, or the step size may be in reased if a larger
error an be a epted. In this way, more attention is paid to those regions
where the solution is rapidly varying, whilst not paying an unne essarily high
pri e of using an unne essarily small step size where the fun tion is slowly
hanging. This variable step size is a hieved by evaluating yn+1 twi e, on e
1.2. REVISION OF SOLVING ORDINARY DIFFERENTIAL EQUATIONS NUMERICALLY15

h
using a step size of h and then again using a step size of
2 . If the dieren e
in yn+1 for the two dierent step sizes is ∆, then it an be shown that for the
′
fourth order method, the required step size h to keep the pres ribed error
to within δ is given by

1
′ 15 δ 5
h = h (1.28)
16 ∆

′
as the error term is O h5 . That is, if h < h we must repeat the urrent
′ ′
step with the smaller step size, whilst if h ≥ h we may use h as the new
step size for the next step. There is, of ourse, some freedom in how to
hoose δ depending on the type of problem being solved - this may either be
an absolute error, a fra tional error, or an a umulated fra tional error sin e
the start of the al ulation.
The fourth-order Runge-Kutta method, with variable step size, is perhaps
the most widely used and trusted general purpose method for integrating
ordinary dierential equations. It has the advantage of being a self-starting
method, that is, does not require knowledge of previous values to omplete
ea h step. However, it does not exploit any parti ular properties of the
S hrödinger equation, and so in this instan e there may be more spe ialised
integration methods that an outperform it. We shall now onsider one su h
method.

1.2.2 The Numerov method

The Numerov method is a spe ialised method for solving a restri ted set of
dierential equations - those that an be written in the form:

d2 y
= f (x) y (x) (1.29)
dx2
whi h in ludes the time independent S hrödinger equation. The method
exploits the spe ial stru ture of this equation to produ e a method that has
fth-order a ura y but only requires two fun tion evaluations per step!
d2 y
We an derive the method by expanding both y (x ± h) and
dx2 x±h
in

Taylor series up to powers of h4 and ombining terms. There is a perfe t

an ellation of all odd-powers of h due to the symmetry of equation 1.29 and
6
so the resulting method is a urate to order h . The result is

z (x + h) = 2z (x) − z (x − h) + h2 f (x) y (x) + O h6 (1.30)

where  
h2
z (x) = 1 − f (x) y (x) (1.31)
12
16 CHAPTER 1. TIME INDEPENDENT SCHRÖDINGER EQUATION

The advantage of this method is that it has an error O h6 whi h is
an order of magnitude better than 4th-order Runge-Kutta and only requires
two (rather than four) fun tion evaluations per step. Hen e it is the pre-
ferred method for solving the S hrödinger equation. However, it does have a
drawba k in that it is not self-starting due to the z (x − h) term - that is, it
requires another method to generate the rst step, whereupon the Numerov
method an be used. This feature of the method will also give problems if
there are any dis ontinuities in the potential whereas the single-step Runga-
Kutta method will be OK.

1.2.3 Singularities

There is a problem as to what to do at singularities. For example, in the hy-

drogen atom both the Coulomb potential and the entrifugal barrier diverge
at the origin. However, for the hydrogen atom we have a known analyti
result for χ (r = 0) and so this an be substituted instead. But what about
the general ase of a singularity with non-zero l? Well, as long as the di-
vergen e is not too fast, then χ will still be well behaved, and we an do a
Taylor series expansion for χ (r) around the singularity:

∞
X
χ (r) = as r s+l+1 (1.32)
s=0
from whi h it an be shown that

   
2 d (rVef f )
as = rVef f as−1 + − E as−2 (1.33)
s (s + 2l + 1) dr

with a0 = 1 (unless l = 0) and a−1 = 0 . Therefore we an generate all the

terms for χ, with an arbitrary normalisation that an be xed later.

1.2.4 Other methods of solution

There are other methods that an be used to solve dierential equations.

For example, we might hoose to dis retize spa e and solve the equation on
a grid, repla ing dierentials by appropriate sums over neighbouring grid
points, and su essively iterating to self- onsisten y. This is known as the
nite-dieren e approa h. It an be in orporated in various ways, su h as
with a single xed grid, or with a hierar hy of grid sizes, or with a grid that
adapts to the solution (e.g. putting more points in the regions where the
fun tion is rapidly hanging). Su h methods an be very su essful and an
example will be onsidered in a later le ture.
There are also indire t methods that an be used to produ e information
about the solution without dire tly solving the equation. Su h methods
exploit spe ial properties of the equation, for example, that it has parti ular
1.3. APPLICATION TO THE SCHRÖDINGER EQUATION 17

symmetries, or is an eigenvalue equation, or an be expressed in terms of

a variational prin iple, et . This will form a very important theme to later
le tures in this ourse.

1.2.5 Errors

In general, when solving any dierential equation numeri ally, there will be
two dominant sour es of error:

• trun ation error

 this is aused by the trun ation of the Taylor series and represents
the fundamental limit to the a ura y of a given algorithm. In
general the error in a solution aused by trun ation error an be
redu ed by using a smaller step size h.

• rounding error

 this is aused by the nite representation of numbers within a digi-

tal omputer. For example, when using single pre ision arithmeti
in a typi al modern omputer, all oating point numbers will be
a urate to about 8 signi ant gures, and when using double
pre ision arithmeti they will be a urate to around 16 signi-
ant gures. In general, the error in the nal solution aused by
rounding error an be redu ed by using fewer integration steps so
that there is less a umulation of error, orresponding to a larger
step size h.

It an be seen then that there is a ompetition between these two sour es

of error. Usually trun ation error dominates, espe ially when using double
pre ision arithmeti (whi h is essential for any kind of s ienti omputing).
So when omputing any solution, it is always a good idea to repeat the
al ulation with a smaller step size and see if there is any signi ant hange
in the solution. If there is, then the al ulation must be repeated again with
an even smaller step size, et . until there is no signi ant hange. However,
if the step size is redu ed too mu h, then rounding error will start to be ome
signif ant and a ura y will be lost again.

1.3 Appli ation to the S hrödinger equation

How then shall we use either the Runge-Kutta or Numerov method to solve
the S hrödinger equation for any parti ular problem? If we are interested in
nding the bound states, i.e. the eigenenergies and eigenfun tions, then we
must:
18 CHAPTER 1. TIME INDEPENDENT SCHRÖDINGER EQUATION

• hoose a parti ular angular momentum (value of l )

• hoose a trial energy (value of Etrial )

• starting from r = 0 integrate the radial equation up to some appropri-
ate large value of r = rmax

• examine the behaviour of the solution - in parti ular χ (r → ∞)

• ifχ (r) → ±∞ then Etrial does not orrespond to a bound state so
adjust Etrial and repeat

This pro ess of tuning Etrial an of ourse be automated, resulting in a

variant of the shooting method.
Note that the ground state solution will have no nodes. A general solution
will have integer values for {n, l, m} - but we will get no information on m
from the radial equation and we have hosen l at the outset - so by ounting
the number of nodes we an dedu e the value of n:
n = (number of nodes) + l + 1

1.4 Final omments

A few nal points to highlight:

• The S hrödinger equation an often be written as an ordinary dier-

ential equation and solved using standard numeri al te hniques.

• Bound states should have a vanishing wavefun tion in the long-range

limit.

• The Numerov method is mu h more e ient than fourth-order Runga-

Kutta for solving the S hrödinger equation.

• Dierent algorithms have dierent hara teristi features, su h as the

ability to handle dis ontinuities or variable step sizes.

• A higher-order algorithm in general has a smaller trun ation error than

a lower-order one (but it depends on the unknown prefa tor!)

• Rounding error will dominate if the step size is made too small.

1.5 Further reading

• TISE in any undergraduate Quantum Me hani s textbook

• Numerov method dis ussed in Computational Physi s by J.M. Thi-

jssen, appendix A7