The Helmholtz Decomposition and the Coulomb Gauge

Kirk T. McDonald
Joseph Henry Laboratories, Princeton University, Princeton, NJ 08544
(April 17, 2008; updated January 29, 2017)

1 Problem
Helmholtz showed in 1858 [1] (in a hydrodynamic context) that any vector eld, say E, that
vanishes suitably quickly at innity can be decomposed as1 ,2

E = Eirr + Erot, (1)

where the irrotational and rotational components Eirr and Erot obey3

Eirr = 0, and Erot = 0. (2)

For the case that E is the electric eld, discuss the relation of the Helmholtz decomposition
to use of the Coulomb gauge.4

2 Solution
The Helmholtz decomposition (1)-(2) is an articial split of the vector eld E into two parts,
which parts have interesting mathematical properties.
We recall that in electrodynamics the electric eld E and the magnetic eld B can be
related to a scalar potential V and a vector potential A according to
A
E = V , (3)
t
B = A. (4)

This results in another decomposition of the electric eld E which might be dierent from
that of Helmholtz. Here, we explore the relation between these two decompositions.
We also recall that the potentials V and A are not unique, but can be redened in
a systematic way such that the elds E and B are invariant under such redenition. A
1
The essence of this decomposition was anticipated by Stokes (1849) in secs. 5-6 of [2].
2
Radiation elds, which fall o as 1/r at large distance r from their (bounded) source, do fall o
suciently quickly for Helmholtz decomposition to apply, as reviewed in [3]. Doubts as to this were expressed
in [4], but see [5]. See Appendix A for the Helmholtz decomposition of the electromagnetic elds of a Hertzian
(point, oscillating) dipole, which illustrates that such a decomposition is readily made when radiation is
present.
3
The irrotational component is sometimes labeled longitudinal or parallel, and the rotational com-
ponent is sometimes labeled solenoidal or transverse.
4
Vector plane waves E ei(krt) do not vanish suitably quickly at innity, so care is required in applying
the Helmholtz decomposition Eirr = (E k) k, Erot = E Eirr of this mathematically useful, but physically
unrealistic class of elds. See, for example, sec. 2.4.2 of [6].

1
particular choice of the potentials is called a choice of gauge, and the relations (3)-(4) are
said to be gauge invariant.5
Returning to Helmholtz decomposition, we note that he also showed how
 
 E(r)  E(r)
Eirr (r) = dVol , and Erot(r) = dVol, (5)
4R 4R
where R = |r r |. Time does not appear in eq. (5), which indicates that the vector eld E at
some point r (and some time t) can be reconstructed from knowledge of its vector derivatives,
E and E, over all space (at the same time t).6 The main historical signicance of the
Helmholtz decomposition (1) and (5) was in showing that Maxwells equations, which give
prescriptions for the vector derivatives E and E, are mathematically sucient to
determine the eld E. Since E = total/0 and E = B/t, the elds Eirr and Erot
involve instantaneous action at a distance and should not be regarded as physically real.
This illustrates how gauge invariance in necessary, but not sucient, for electromagnetic
elds to correspond to reality.7,8,9
The Helmholtz decomposition (1) and (5) can be rewritten as

E = V + F, (11)

where  
 E(r)  E(r)
V (r) = dVol , and F(r) = dVol . (12)
4R 4R
5
The gauge transformation A A + , V V /(ct), leaves the elds E and B unchanged. A
consequence of this is that when the vector potential is decomposed as A = Airr + Arot , the rotational part
is actually gauge invariant. That is, Airr + Arot (Airr + ) + Arot , where the term in parenthesis is
the irrotational part of the transformed vector potential, so the rotational part, Arot , is unchanged by the
gauge transformation.
6
If the eld E is known only within a nite volume V, bounded by a closed surface S, then the Helmholtz
decomposition (1) and (5) becomes
  
 E(r )  n E(r ) 
Eirr (r) = dVol + dArea , (6)
V 4R S 4R
  
 E(r ) n E(r )
Erot (r) = dVol + dArea , (7)
V 4R S 4R

where n is the inward unit normal vector on the surface S. That is,

E = V only if E = 0 in V, and n E = 0 on S, (8)

E = A only if E = 0 in V, and n E = 0 on S. (9)

Neglect of the conditions on the surface S can lead to error, as remarked in [7, 8, 9, 10].
7
The forms (5) are not the only possible representations of Eirr and Erot . For example, we could write
   
 E(r )   E(r ) 
Eirr (r) = dVol + C , and Erot (r) = dVol + , (10)
4R 4R

for any constant C and any dierentiable scalar function , without changing the values of Eirr and Erot .
8
See Appendix A for an application of the Helmholtz decomposition to Hertzian dipole radiation
9
The Helmholtz decomposition leads to interesting interpretations of the momentum and angular mo-
mentum associated with electromagnetic elds [6].

2
 It is consistent with usual nomenclature to call V a scalar potential and F a vector
potential. That is, Helmholtz decomposition lends itself to an interpretation of elds as
related to derivatives of potentials.
When the vector eld E is the electric eld, it also obeys Maxwells equations, two of
which are (in SI units and for media where the permittivity is 0 )
B
E= , and E = , (13)
0 t
where is the electric charge density and B is the magnetic eld.
If we insert these physics relations into eq. (12), we nd

(r)
V (r) = dVol, (14)
40R

B(r)
F(r) = dVol . (15)
t 4R
The scalar potential (14) is calculated from the instantaneous charge density, which is
exactly the prescription (43) of the Coulomb gauge. That is, Helmholtz + Maxwell implies
use of the Coulomb-gauge prescription for the scalar potential.
However, eq. (15) for the vector potential F does not appear to be that of the usual
procedures associated with the Coulomb gauge. Comparing eqs. (11)-(12) and (15), we see
that we can introduce another vector potential A which obeys
A
F= , (16)
t
such that 
B(r)
A(r) = dVol , (17)
4R
and
A
E = V , (3)
t
which is the usual way the electric eld is related to a scalar potential V and a vector potential
A. Note also that eq. (17) obeys the Coulomb gauge condition (39) that A = 0.10
Thus, the Helmholtz decomposition (1) and (5) of the electric field E is equiva-
lent to the decomposition (3) in terms of a scalar and a vector potential, provided
those potentials are calculated in the Coulomb gauge.11
Using various vector calculus identities, we have
  
B(r )  1 B(r)   1 B(r)
A(r) = dVol = dVol = dVol
4R R 4 R 4
 
 B(r) B(r 
)
= dVol +  dVol
4R 4R
  
 B(r) B(r )  B(r )
= dVol dArea = dVol, (18)
4R 4R 4R
10
See, for example, sec. 3 of [11].
11
The elds Eirr = V Airr /(ct) and Erot = Arot /(ct) can be deduced from scalar potential
(C)
V and vector potential A = Airr + Arot in any gauge, but only in the Coulomb gauge is Airr = 0 such that
(C) (C)
the Helmholtz decomposition has the simple form Eirr = V and Erot = A /t.

3
provided B vanishes suciently quickly at innity. In view of the Maxwell equation B = 0,
we recognize eq. (18) as the Helmholtz decomposition B = A for the magnetic eld.12
We can go further by invoking the Maxwell equation
1 E
B = 0 J + , (22)
c2 t
where J is the current density vector, the medium is assumed to have permeability 0 , and
c is the speed of light, so that

0  J(r)  E(r)
A(r) = dVol + dVol . (23)
4 R t 4c2 R
This is not a useful prescription for calculation of the vector potential, because the second
term of eq. (23) requires us to know E(r)/c2 to be able to calculate E(r).13 But, c2 is a big
number, so E/c2 is only a small correction, and perhaps can be ignored.14 If we do so,
then 
J(r)
A(r) = 0 dVol , (24)
4 R
which is the usual instantaneous prescription for the vector potential due to steady currents.
Thus, it appears that practical use of the Helmholtz decomposition + Maxwells equations
is largely limited to quasistatic situations, where eqs. (14) and (24) are suciently accurate.
Of course, we exclude wave propagation and radiation in this approximation. We can
include radiation and wave propagation if we now invoke the usual prescription, eqs. (44)-(45)
of Appendix B, for the vector potential in the Coulomb gauge. However, this prescription
12
We can verify the consistency of eqs. (17) and (18) by taking the curl of the latter. For this, we note
that
   
 B(r ) 1 1
= ( B(r ))   
= ( B(r )) . (19)
4R 4R 4R

The i-component of this is

i,j,k jlm (l Bm )k (1/4R) = ki      

lm (l Bm )k (1/4R) = (k Bi )k (1/4R) (i Bk )k (1/4R)
= k [Bi k (1/4R)] Bi 2 (1/4R) k [(1/4R) i Bk ] + (1/4R) i  B
= Bi (r ) 3 (r r ) + k [Bi k (1/4R) (1/4R) i Bk ]. (20)

The volume integral of this gives B(r) plus a surface integral that vanishes if the magnetic eld falls o
suciently quickly at large distances. That is, A = B for the vector potentials given by eqs. (17) and
(18).
We could also proceed by taking the curl of eq. (17), noting that
       
B(r ) B(r ) 1 B(r )
= B(r ) 2 = + B(r ) 3 (r r ). (21)
4R 4R 4R 4R

Then, integrating this over dVol gives B(r) plus a surface integral that vanishes for magnetic elds that fall
o suciently quickly at large distances. So, again we nd that A = B.
This footnote is due to Vladimir Hnizdo. See also [12].
13
Using the Helmholtz decomposition for E in eq. (23) permits us to proceed without knowing E, provided
we know the charge density and the time derivative B/t, which is no improvement conceptually.
14
See [13] for an argument that the second integral of eq. (23) vanishes in the quasistatic approximation.

4
does not follow very readily from the Helmholtz decomposition, which is an instantaneous
calculation.
Note that in the case of practical interest when the time dependence of the charges and
currents is purely sinusoidal at angular frequency , i.e., eit , the Lorenz gauge condition
(38) becomes
ic
V = A. (25)
k
In this case it suces to calculate only the vector potential A, and then deduce the scalar
potential V , as well as the elds E and B, from A.
However, neither the Coulomb gauge condition A = 0 nor the Lorenz gauge condition
(38) suces, in general, for a prescription in which only the scalar potential V is calculated,
and then A, E and B are deduced from this. Recall that the Helmholtz decomposition tells
us how the vector eld A can be reconstructed from knowledge of both A and A.
The gauge conditions tell us only A, and we lack a prescription for A in terms of
V.
[In 1 dimension, A = 0, so in 1-dimensional problems we can deduce everything
from the scalar potential V plus the gauge condition. But life in 3 dimensions is more
complicated!]

Appendix A: Helmholtz Decomposition of Hertzian Dipole

Radiation
The electric and magnetic elds of an ideal, point Hertzian electric dipole can be written (in
Gaussian units) as

2 cos(kr t) cos(kr t) k sin(kr t)
E = k p(r p) r + p[3(p r) r p] + , (26)
r r3 r2

2 cos(kr t) sin(kr t)
B = Brot = k p(r p) , (27)
r kr2
where r = r/r is the unit vector from the center of the dipole to the observer, p = p cos t p
is the electric dipole moment vector, = 2f is the angular frequency, k = /c = 2/ is
the wave number and c is the speed of light [14, 15].
The irrotational part of the electric eld is the instantaneous eld of the electric dipole,
cos t
Eirr = p[3(p r)r p] . (28)
r3
Thus, the rotational part of the electric eld is

cos(kr t)
Erot = E Eirr = k 2 p(r p) r
 r
cos(kr t) cos t k sin(kr t)
+p[3(p r) r p] + . (29)
r3 r2

5
Both elds Eirr and Erot have instantaneous terms.
The ow of energy in the electromagnetic eld is described by the Poynting vector S, so
the Helmholtz decomposition leads us to write
c c
S = S1 + S2 = Eirr Brot + Erot Brot. (30)
4 4
Using eqs. (27)-(28), we have that

ck 2p2 cos(kr t) sin(kr t)
S1 = [(3 cos2 1) r 2 cos p] cos t , (31)
4 r4 kr5
where is the angle between vectors r and p. Similarly,


c 4 2 2 cos2 (kr t) cos(kr t) sin(kr t)
S2 = k p sin r
4 r2 kr3

cos2(kr t) sin2 (kr t)
+k 2p2 [(3 cos2 1) r 2 cos p]
r4
 
k 1
+ cos(kr t) sin(kr t) 3 5
r kr
 
cos(kr t) sin(kr t)
cos t . (32)
r4 kr5
Neither S1 nor S2 describes the ow of energy at an identiable speed, so the Helmholtz
decomposition, which is based on present source terms, does not seem well suited to a general
characterization of the ow of energy in electromagnetic elds.
We can restrict our attention to the region very close to the source, where kr  1 and
we have
 
ck 2 p2 cos2 t cos t sin t
S1 (kr  1) = [(3 cos2 1) r 2 cos p] + , (33)
4 r4 kr5
and
  
c cos2 t cos t sin t
S2 (kr  1) = k 4 p2 sin2 r +
4 r2 kr3
 
2 2 2 k cos t sin t sin2 t
+k p [(3 cos 1) r 2 cos p] . (34)
r3 r4
Here, the separation of the total Poynting vector S into S1 and S2 is cleaner than for large
kr, but, to this author, this separation is still not associated with any crisp physical insight.
We can also consider only the time average of eqs. (31)-(32),
 
ck 2 p2 cos kr sin kr
S1  = [(3 cos2 1) r 2 cos p] , (35)
8 r4 kr5
and
c k 4p2 sin2
S2  = r S1  . (36)
8 r2
Again, there seems to be little physical insight associate with this decomposition.

6
Appendix B: Coulomb Gauge
The relations
A
E = V , and B=A (37)
t
between the electric and magnetic elds E and B and the potentials V and A permits various
conventions (gauges) for the potentials. One popular choice is the Lorenz gauge [16],

1 V
A= (Lorenz). (38)
c2 t
In situations with steady charge and current distributions (electrostatics and magneto-
statics), V/t = 0, so the condition (38) reduces to

A=0 (Coulomb). (39)

Even in time-dependent situations it is possible to dene the vector potential to obey eq. (39),
which has come to be called the Coulomb gauge condition.
We restrict our discussion to media for which the dielectric permittivity is 0 and the
magnetic permeability is 0. Then, using eq. (37) in the Maxwell equation E = /0
leads to

2 V + A = , (40)
t 0
and the Maxwell equation B = 0 J + E/c2t leads to
 
2 1 2A 1 V
A 2 2 = 0 J + A + 2 . (41)
c t c t

Thus, in the Coulomb gauge (39), eq. (40) becomes Poissons equation,

2 V = , (42)
0
which has the formal solution

1 (r, t)
V (r, t) = dVol (Coulomb), (43)
40 R
where R = |r r |, in which changes in the charge distribution instantaneously aect the
potential V at any distance.
It is possible to choose gauges for the electromagnetic potentials such that some of their
components appear to propagate at any specied velocity v [18, 19]. One can also choose that
the scalar potential has no time dependence, such that all time dependence of the electric
eld is associated with that of the vector potential [20].
For completeness, a formal solution for the vector potential in the Coulomb gauge is

Jt (r, t = t R/c)
A(r, t) = 0 dVol (Coulomb), (44)
4 R

7
where the transverse current density is dened by

1 J(r , t)
Jt(r, t) = dVol . (45)
4 R
While the Coulomb-gauge vector potential (44) can be said to propagate (in vacuum)
with the speed of light, the part, (1/c)A/t, derived from it has pieces that propagate
instantaneously, as needed to cancel the instantaneous behavior of the part, V , derived
from the Coulomb-gauge scalar potential (43). For additional discussion, see, for example,
[21, 22, 23, 24].
Unless the geometry of the problem is such that the transverse current density Jt is easy
to calculate, use of the Coulomb gauge is technically messier than the use of the Lorenz
gauge, in which case the (retarded) potentials are given by are the retarded potentials

1 (r, t = t R/c)
V (r, t) = dVol (Lorenz), (46)
40 R
0  J(r , t = t R/c)
A(r, t) = dVol (Lorenz), (47)
4 R
where R = |r r|.
Analysis of circuits is often performed in the quasistatic approximation that eects of
wave propagation and radiation can be neglected. In this case, the speed of light is taken
to be innite, so that the Lorenz gauge condition (38) is equivalent to the Coulomb gauge
condition (39), and the potentials are calculated from the instantaneous values of the charge
and current distributions. As a consequence, gauge conditions are seldom mentioned in
ordinary circuit analysis.

References
[1] H. Helmholtz, Uber Integrale der hydrodynamischen Gleichungen, welche den Wirbel-
bewegungen entsprechen, Crelles J. 55, 25 (1858),
http://physics.princeton.edu/~mcdonald/examples/fluids/helmholtz_jram_55_25_58.pdf

[2] G.G. Stokes, On the Dynamical Theory of Diraction, Trans. Camb. Phil. Soc. 9, 1
(1849), http://physics.princeton.edu/~mcdonald/examples/optics/stokes_cpst_9_1_49.pdf

[3] A.M. Stewart, Does the Helmholtz theorem of vector decomposition apply to the wave
elds of electromagnetic radiation?, Phys. Scripta 89, 065502 (2014), http://physics.
princeton.edu/~mcdonald/examples/EM/stewart_ps_89_065502_14.pdf

[4] J.A. Heras, Comment on Causality, the Coulomb eld, and Newtons law of gravitation,
Am. J. Phys. 71, 729 (2003),
http://physics.princeton.edu/~mcdonald/examples/EM/heras_ajp_71_729_03.pdf

[5] F. Rohrlich, The validity of the Helmholtz theorem, Am. J. Phys. 72, 412 (2004),
http://physics.princeton.edu/~mcdonald/examples/EM/rohrlich_ajp_72_412_04.pdf

8
 [6] K.T. McDonald, Orbital and Spin Angular Momentum of Electromagnetic Fields
(March 12, 2009), http://physics.princeton.edu/~mcdonald/examples/spin.pdf

[7] T. Teichmann and E.P. Wigner, Electromagnetic Field Expansions in Loss-Free Cavities
Excited through Holes, J. Appl. Phys. 24, 262 (1953),
http://physics.princeton.edu/~mcdonald/examples/EM/teichmann_jap_24_262_53.pdf

[8] A.F. Stevenson, Note on the Existence and Determination of a Vector Potential, Quart.
Appl. Math. 12, 194 (1954),
http://physics.princeton.edu/~mcdonald/examples/EM/stevenson_qam_12_194_54.pdf

[9] K. Kurokawa, The Expansions of Electromagnetic Fields in Cavities, IRE Trans. Micro.
Meth. Tech. 6, 178 (1958),
http://physics.princeton.edu/~mcdonald/examples/EM/kurokawa_iretmtt_6_178_58.pdf

[10] J. van Bladel, On Helmholtzs Theorem in Multiply-Bounded and Multiply-Connected

Regions, J. Franklin Inst. 269, 445 (1960),
http://physics.princeton.edu/~mcdonald/examples/EM/vanbladel_jfi_269_445_60.pdf

[11] A.M. Stewart, Vector potential of the Coulomb gauge, Eur. J. Phys. 24, 519 (2003),
http://physics.princeton.edu/~mcdonald/examples/EM/stewart_ejp_24_519_03.pdf

[12] V. Hnizdo, Comment on Vector potential of the Coulomb gauge, Eur. J. Phys. 25, L21
(2004), http://physics.princeton.edu/~mcdonald/examples/EM/hnizdo_ejp_25_L21_04.pdf

[13] A. Shadowitz, The Electromagnetic Field (McGraw-Hill, 1977; Dover reprint, 1978),
sec. 11-2, http://www.hep.princeton.edu/~mcdonald/examples/EM/shadowitz_75_chap11.pdf

[14] H. Hertz, The Forces of Electrical Oscillations Treated According to Maxwells Theory,
Weidemanns Ann. 36, 1 (1889), Nature 39, 402 (1889); reprinted in chap. 9 of H. Hertz,
Electric Waves (Macmillan, 1900),
http://physics.princeton.edu/~mcdonald/examples/EM/hertz_electric_oscillations.pdf

[15] Sec. 9.2 of J.D. Jackson, Classical Electrodynamics, 3rd ed. (Wiley, New York, 1999).

[16] The gauge condition (38) was rst stated by L. Lorenz, On the Identity of the Vibrations
of Light with Electrical Currents, Phil. Mag. 34, 287 (1867),
http://physics.princeton.edu/~mcdonald/examples/EM/lorenz_pm_34_287_67.pdf
Lorenz had already used retarded potentials of the form (46) in discussions of elastic
waves in 1861, and Riemann had discussed them as early as 1858 [17].

[17] J.D. Jackson and L.B. Okun, Historical roots of gauge invariance, Rev. Mod. Phys. 73,
663 (2001), http://physics.princeton.edu/~mcdonald/examples/EM/jackson_rmp_73_663_01.pdf

[18] J.D. Jackson, From Lorenz to Coulomb and other explicit gauge transformations, Am.
J. Phys. 70, 917 (2002),
http://physics.princeton.edu/~mcdonald/examples/EM/jackson_ajp_70_917_02.pdf

9
[19] K.-H. Yang, The physics of gauge transformations, Am. J. Phys. 73, 742 (2005),
http://physics.princeton.edu/~mcdonald/examples/EM/yang_ajp_73_742_05.pdf

[20] K.T. McDonald, Static-Voltage Gauge (March 25, 2008),

http://physics.princeton.edu/~mcdonald/examples/static_gauge.pdf

[21] O.L. Brill and B. Goodman, Causality in the Coulomb Gauge, Am. J. Phys. 35, 632
(1967), http://physics.princeton.edu/~mcdonald/examples/EM/brill_ajp_35_832_67.pdf

[22] C.W. Gardiner and P.D. Drummond, Causality in the Coulomb gauge: A direct proof,
Phys. Rev. A 38, 4897 (1988),
http://physics.princeton.edu/~mcdonald/examples/EM/gardiner_pra_38_4897_88.pdf

[23] F. Rohrlich, Causality, the Coulomb eld, and Newtons law of gravitation, Am. J. Phys.
70, 411 (2002), http://physics.princeton.edu/~mcdonald/examples/EM/rohrlich_ajp_70_411_02.pdf

[24] B.J. Wundt and U.D. Jentschura, Sources, Potentials and Fields in Lorenz and Coulomb
Gauge: Cancellation of Instantaneous Interactions for Moving Point Charges, Ann.
Phys. (N.Y.) 327, 1217 (2012),
http://physics.princeton.edu/~mcdonald/examples/EM/wundt_ap_327_1217_12.pdf
http://arxiv.org/abs/1110.6210

10