The Vector Potential

Mikael B. Steen

October 27, 2011

1 Introducing A
In electrostatics we found that the fact that ∇ × E = 0 allowed us to define a scalar potential with the
property that ∇V = −E. This electric potential gave us ways of visualising particles moving in electric
fields in terms of climbing the ’potential hill’, increasing or decreasing their electric energy. We also saw
it gave us a fully general way of calculating electric fields without referring to any symmetry or dealing
with any vectors in terms of Poisson’s equation

ρ
∇2 V = − (1)
ε0

which has the solution

ρ(r0 )
Z
1
V (r) = dv 0 . (2)
4πε0 |r − r0 |

So if we know the charge distribution1 ρ we have a straight forward, algorithmic way of finding the electric
field without appealing to any symmetry or dealing with vector addition. For simple distributions this
can be calculated by hand or for complicated distributions can be done by a computer (which you did
in Oblig 1).
It seems reasonable to ask weather we can do exactly the same thing for magnetic fields, but this can
not be since magnetic fields are not irrotational. Ampere’s law states that

∇ × B = µ0 J (3)

so we can not have scalar potential defining B such that ∇Vb = −B when the curl of any gradient field is
zero. This would be a contradiction to Ampere’s law! However as a consequence of there not being any
magnetic monopoles we know that magnetic fields are solenoid fields2 , meaning that they are divergence
less, which is encapsulated in ∇ · B = 0. This fact allows us to define a vector potential A with the
property that B = ∇ × A. We want a magnetic equivalent of equation 1 so let’s try to relate A to the
current density J. Using Ampere’s law (equation 3) we get

∇×B = ∇ × (∇ × A) (4)
2
= ∇(∇ · A) − ∇ A (5)
= µ0 J (6)
1 Often
this is the case since we in a given experiment can distribute the charge as we like.
2 Themagnetic field inside a solenoid is constant and therefore divergence less. This might be the reason for calling
divergence less fields for solenoid fields.

1
where we have used an identity for the curl of the curl of a vector field.3 Thus our relation reads

∇(∇ · A) − ∇2 A = µ0 J. (7)
Looking back at Poisson’s equation (equation 1) this looks a bit more nasty. If it we’re not for the pesky
∇(∇ · A) term at least our equation would look like vector form of Poisson’s equation. This term include
∇ · A, but what is the divergence of A? The answer is that it is up us to define it! Remember that the
electric potential V was only defined up to a constant. If ∇V = −E we could define a new potential
V 0 = V + c and this would give us exactly the same electric field since

∇V 0 = ∇(V + c) = ∇V + 0 = −E
meaning that we are not really changing the physics by adding a constant to V . Similarly we can add
any gradient field ∇Λ to A since the curl of any gradient field is zero and therefore this term would
vanish in taking the curl of A. Thus we can define

A0 = A + ∇Λ

and we’ll still have

∇ × A0 = ∇ × A + ∇ × (∇Λ) = ∇ × A = B
and this allows us to define ∇ · A since

∇ · A0 = ∇ · A + ∇2 Λ.

So if we want to clean up our relation in equation 7 we just choose Λ such that the divergence of A is
zero. This leads us to the result

∇2 A = −µ0 J. (8)

which indeed looks very much like Poisson’s equation. The only difference is really that we now have one
equation for each component of A, but we reached our goal in finding an equation which we can solve
algorithmically.

2 Calculating the vector potential

We set out to find the magnetic analog of Poisson’s equation and ended up with equation 8. We can
exploit the fact that this equation and equation 1 are similar. Looking just at the x-component of
equation 8 we have

∇2 Ax = −µ0 Jx .

which has the exact mathematical form of Poisson’s equation. We know that the solution to

ρ
∇2 V = −
ε0
is
3 This identity will also show up in the derivation of electromagnetic waves, so it might be worthwhile to get comfortable

with it now.

2
 ρ(r0 )
Z
1
V (r) = dv 0 .
4πε0 |r − r0 |

and identifying µ0 Jx with ρ/ε0 we can conclude that

Jx (r0 ) 0
Z
µ0
Ax = dv
4π |r − r0 |

or more generally

J(r0 )
Z
µ0
A= dv 0 . (9)
4π |r − r0 |

With this equation it is in principle straight forward to calculate the vector potential given a distribution
of currents. Then it’s just to take the curl to find the magnetic field.

3 The vector potential in electrodynamics

A and V are usefull in electrostatics because they give us a straight forward way of calculating B and E
without appealing to any symmetry and we can find it by solving tree Poisson’s equations. However the
scalar and vector potential becomes even more important when we enter the domain of electrodynamics.
When we have moving charges and changing currents the laws of interaction between sources and fields
are expressed in in terms of Maxwell’s four equations

ρ
∇·E= (10)
ε0

∂B
∇×E=− (11)
∂t

∇·B=0 (12)

∂E
∇ × B = µ0 J + µ0 ε0 . (13)
∂t
Now because a changing B causes a curl in E, E will no longer be the gradient of any scalar V , but let’s
try to find a quantity that will. B is still divergence less so ∇ × A = B and substitution into equation
11 yields

∂ ∂A
∇×E=− (∇ × A) = −∇ × ( )
∂t ∂t
which implies that
 
∂A
∇× E+ =0
∂t
and thus the sum of the electric field and the time derivative of the vector potential will be an irrotational
field so it can be expressed as the gradient of a scalar as

∂A
∇V = −E − . (14)
∂t

3
When A does not change with time this just reduces to our old V from electrostatics and therefore is
a valid generalization of V for electrodynamics. We want only the potentials in terms of the sources J
and ρ and we can eliminate E by using Gauss law (equation 10) by taking the divergence of equation
14. This gives us
 
∂A ρ ∂
∇2 V = −∇ · E − ∇ · =− − (∇ · A)
∂t ε0 ∂t

such that

∂ ρ
∇2 V + (∇ · A) = − .
∂t ε0
This equation relates V and A to the charge distribution ρ, but we also need a relation in terms of the
current distribution J. This is obtained by Ampere’s law (equation 13) by inserting the vector potential
and our relation between V and E from equation 14 such that

∂2A
   
∂ ∂A ∂V
∇ × B = ∇ × (∇ × A) = µ0 J + µ0 ε0 −∇V − = µ0 J − µ0 ε0 ∇ − µ0 ε0 2
∂t ∂t ∂t ∂t

using the identity for the curl of the curl and rearranging terms we get

∂2A
 
2 ∂V
∇ A − µ0 ε0 2 − ∇ ∇ · A + µ0 ε0 = −µ0 J (15)
∂t ∂t

and together with

∂ ρ
∇2 V + (∇ · A) = − . (16)
∂t ε0
these two equations embody all the information in the four Maxwell equations!

4 Gauge transformations
Now you might argue that the two equations found in the previous section looked kind of hairy compared
to the neat Maxwell equations and I would agree. Even though we did reduce the six problems of finding
the components of E and B in to four (finding V and the components of A) these equations seems
very hard to solve due to their hairy nature. But remember what we did back in section 1 where we
exploited the fact that adding a gradient field ∇Λ to A isn’t changing the physics because ∇ × A is still
B and used this to clean up our equation by making A divergence less. The process changing V and A
in such a way that E and B does not change is called a gauge transformation and we’ll now use such
transformations to clean up our dynamical equations 15 and 16. Now in electrostatics E was uniquely
determined by V , but in electrodynamics E is also determined by A trough

∂A
E = −∇V − .
∂t
So if we’re adding ∇Λ to A we are not changing B, but we’re changing E to
   
∂A ∂Λ ∂Λ ∂A
E0 = −∇V − −∇ = −∇ V + − .
∂t ∂t ∂t ∂t

This is not allowed for our transformation to be a gauge transformation and we therefore have to com-
pensate by simultaneously subtracting ∂Λ/∂t from V . This ensures that neither E nor B changes. So
to conclude a gauge transformation must be of the form

4
 ∂Λ
V0 =V − A0 = A + ∇Λ. (17)
∂t
The transformation we did back in section 1 where we sat the divergence of A to zero by choosing

A0 = A + ∇Λ
such that

∇2 Λ = −∇ · A

is called the Coulomb gauge and is particularly usefull in electrostatics because of the simple relations
we get for A and V (equations 8 and 1). In electrodynamics however, we wish to make the equations 15
and 16 as simple as possible and a good choice here is to set

∂V
∇ · A = −µ0 ε0 (18)
∂t
which gets rid of one of the terms in equation 15 and we obtain

∂2A
∇2 A − µ0 ε0 = −µ0 J (19)
∂t2

and
∂2V ρ
∇2 V − µ0 ε0 2
=− . (20)
∂t ε0

In these two beautiful equations we have embodied everything in electrodynamics and we can in principle
find everything that we could ever wish for. Notice that they reduce to our familiar static equations
when A and V does not depend on time.