The Proximity Effect: A Comparison of COMSOL® and Analytic

Solutions
C. Cooke1 and L. Shatz2,3
1. Massachusetts Institute of Technology, EECS, Cambridge, MA, USA
2. Suffolk University, Electrical Engineering, Boston MA, USA
3. Benjamin Franklin Institute of Technology, Boston, MA, USA

Introduction

Electromagnetic (EM) solver software can provide a In this paper we apply a new analytic solution
powerful tool to help develop complex electrical method for the Smith integral equation for the
apparatus with many useful applications including proximity effect and apply it to a greater range of
communications, transportation and electric power configurations with more conductors. Additionally
systems. Unfortunately numerical solver “accuracy” we use the COMSOL® with AC/DC Module solver
is typically expressed by the simple metric of to obtain numerical solutions for the same
iteration convergence, which means when very tiny configurations in order to compare the results
change per iteration is achieved the solution obtained between the analytic and numerical simulation
is essentially the final result. But this iteration solution methods.
convergence represents the numerical process and
does not quantify absolute accuracy. It would be
very convenient for the software user if the solver Parallel Wire Proximity Effect Analytic
software provided a metric for the absolute accuracy Models
it can achieve. But here the difficulty is finding a
problem, which both truly challenges the ability of Geometry
the software yet has a validated analytic solution. The calculation of proximity effects is based upon N
Hence a more severe type of problem, yet one with equally spaced parallel wires in the z-direction with
an analytic solution is needed to even begin to think wire radius, a, and separation between wire centers of
about “standard” or reference problems that would 2c. The arrangement is depicted in Figure 1. Note
help validate the capabilities of solver software. the gap spacing between adjacent wires, g = (2c-2a).

To help with this accuracy/validation topic we

introduce the problem of the “proximity effect”.
Broadly, the proximity effect concerns how currents CL

external to a conductor impact the location of wire-to-wire gap

= (2c - 2a)

currents within a given conductor. One practical edge

wire
center
wire
edge
wire
impact is that the proximity effect can increase the 2 1

effect resistance of a wire by forcing the currents to

flow mainly in a highly constricted region so only a a
2c
limited portion of the conductor carries all the
current, which acts to increase resistive losses. (1a) Side View (1b) Top View
Figure 1 Equally spaced parallel wire configuration, N= 3
As identified by Smith [1] the proximity effect in
parallel conductors causes an additive impact to the
For the analytic solution for the proximity effect it is
standard well-known skin-effect. The skin-effect
assumed: (1) the parallel round wires all carry equal
restricts high frequency AC currents to symmetric
currents, I, (2) these currents can vary around the
annular region adjacent the perimeter of a single
wire circumference, θ, and (3) are confined to be
isolated wire. Smith also provided an analytic
surface currents Kn(θ) different for each of the N
solution in the form of an integral equation and
wires. This condition of surface currents corresponds
furthermore made numerical solutions of the integral
to a high-frequencies where the skin-effect limits the
equations for a number of arrangements of parallel
currents to be very near the wire surface. The
wires with different separation spacing. And
problem is quasi-static without wave-propagation
additionally he also experimentally verified some of
effects.
the calculated configurations with excellent results.

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

Analytic Solution, Surface Current Density (after Using the formulas for g (θ ') and r 'ml in Eqns. (4) (5)
Smith [1]) and (6), for (7) and (8), we can see the solutions
involve integrals of the form below, Eqn (9):
Following the conventions used by Smith [1], the
long perfectly conducting wires carry a surface π
cos pθ '
1 + 2 c a (m − l ) cos θ − cos(θ − θ ')
,
current density, which varies with angular position θ, ∫
θ '=−π 4(m − l ) 2 c 2 + r 2 + a 2 − 2ar cos(θ − θ ') + 4(m − l )c(r cos θ − a cos θ ')
for the lth wire conductor given by:
The integration can be performed exactly by use of
K l (θ ', z ') = I (
g (θ ') f (z)
2π a l (1) ) the Cauchy’s Integral Theorem. [3] After integrating
where, with the problem symmetry the normalized exactly, the orthogonality property of cosines can be
angular variation function gm(θ’) is: used to determine the amp ' s coeffieicents for each m .
q
The calculation was implemented in Mathematica®
g m (θ ') = 1+ ∑ amp cos( pθ ') and typically 6 to 12 cosine terms were needed to
p=1 . (2)
achieve convergence in the sense that the coefficients
The z component of the magnetic vector potential
for higher order terms became several orders of
Amz (r ,θ , z ) at a point just off the surface of the mth magnitude smaller and thus insignificant.
conductor is related to the gm(θ’) function by:
∂( Amz (r ,θ , z ) µo I Example Results, Analytic Solution Current Density
− = g (θ ) f ( z )
(3)
∂r 2π a l
using the boundary condition at r = a . We represent The analytic calculated distribution of normalized
the magnitude of the displacement vector Rml as: surface current density gm as a function of
Rml = (z − z ') 2 + rml2 , rml = circumferential angle, θ, is presented in Figure 2 for
(4)
4(m − l) 2 c 2 + r 2 + a 2 − 2ar cos(θ − θ ') + 4(m − l)c(r cosθ − a cosθ ') 3, 11 and 25 wires all with normalized spacing of c/a
.
Applying Eqn. (4) to Eqn. (6) and rearranging, where = 2 (gap = 2a). Here by definition wire-1 is the outer
r 'ml = rml a . : “edge” wire and wire-2 is the center wire with 3
π m
wires and wire 6 is the center wire with 11 wires and
1 ⎛ ⎞
g m (θ ) = 1 + ∫π ⎜⎝ ∑ gl (θ ') K ml (θ ,θ ') dθ ' ⎟. wire 10 with 20 wires. All curves integrate to π, as
π (5)
− l =1,l ≠ m ⎠ gm(θ) = 1 for a uniform current density. (By
with symmetry the curves are symmetric for angles from 0
1 + 2 c a (m − l ) cos θ − cos(θ − θ ') to – π, and currents in wires beyond the center are
K ml (θ ,θ ') = ,
(r 'ml ) 2 . (6) mirror images of the first half.)
Symmetry about the center wire requires that gm( )
c/a= 2

g (θ ) = g (π − θ ) , which can be used to reduce the 2.5

number of equations to n 2 for n even, and 2.0

Wire 1
1.5 Wire 2
2 for n odd. Hence separating out the m=l
( n + 1) 1.0
Wire 3
Wire 4
term the set of integral equations become: 0.5 Wire 5
Wire 6
For n even: -0.5
0.5 1.0 1.5 2.0 2.5 3.0

1 π
g m (θ ) = 1+ (
∫ g (θ ')K m,n+1−m θ , π − θ ' dθ ' +
π θ '=− π m
) (7) -1.0

Figure 2a 11 wires
1 π % n/2 (
∫
π − π & l=1,l≠m
( ( ) (
'' ∑ g l (θ ') K m,l θ ,θ ' + K m,n+1−l θ , π − θ ' dθ '**
)
)) gm(θ)
c/a= 2.

3
for m = 2, 4,…, n / 2. Wire 1 Wire 9
Wire 2 Wire 10
2
Wire 3 Wire 11
For n odd, 1
Wire 4 Wire 12
Wire 5 Wire 13
h(m) π
g m (θ ) = 1+
π θ '=− π m
(
∫ g (θ ')K m,n+1−m θ , π − θ ' dθ ' ) (8) θ
Wire 6
Wire 7
0.5 1.0 1.5 2.0 2.5 3.0
Wire 8
1 π% (
(n+1)/2
+ ∫
π − π & l=1,l≠m
( ( ) (
'' ∑ g l (θ ') K m,l θ ,θ ' + K m,n+1−l θ , π − θ ' dθ '**
)
)) -1

h(m) π Figure 2b 25 wires

+ ∫ g (θ ')K m,(n+1)/2 θ ,θ ' dθ '
π θ '=− π m
( )
n +1 n +1 Figure 2 Normalized surface current density for parallel
where h(m) = 1 for m ≠ , h(m) = 0 for m = wires with spacing c/a = 2.00.
2 2 .

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

When the wire spacing is made closer, especially less proximity effect for N parallel wires. Here RT is the
than c/a = 1.5, then the current distribution becomes total series-connected equivalent resistance per unit
more complex. This is depicted for the case of 7 length of the N wires, RT = Rp+Ro. And thus RT for
wires more closely spaced at c/a = 1.25 in Figure 3. the system of N series-connected parallel wires, each
Here inner wires 2 and 3 are seen to exhibit greater of radius a, becomes:
variation in current density for small angles. !R $ !R $
c/a= 1.25 RT = ## p +1&& Ro = ## p +1&& NR
skin
" Ro % " Ro %
gm(θ)

!R $ RS . (12)
2
= ## p +1&& N
1
Wire 1
" Ro % 2π a
Wire 2
Wire 3 Applying the method of Smith [1], the resistance for
0.5 1.0 1.5 2.0 2.5 3.0
θ
Wire 4 N series connected parallel wires (and q cosine terms)
-1
is determined from the above defined integral
equation coefficients, amp ' s , as:
-2
" q % N
Figure 3 Normalized surface current density for 7 parallel $1+ 1 2' RS
wires with smaller c/a = 1.25
∑ $ 2 ∑ mp ' ohm/m (13)
a RT =
2π a
m=1# p=1 &
Analytic Solution, Wire Resistance and hence the normalized proximity resistance, Rp/Ro,
becomes:
When the separation between parallel wires is very R p RT − NRskin N # q &
1 % 2
large so there is no proximity-effect, then only skin- = = ∑ ∑ amp ( (14)
Ro NRskin 2 % (
effect determines the total resistance per unit length m=1$ p=1 '.
of a set of N series-connected wires, Ro, and is: Figure 4 depicts the calculated values for the added
Ro = NRskin (ohm/m) (10) normalized resistance, (Rp/Ro) on a logarithmic scale,
where, a wire of radius, a, has a wire resistance per versus number of parallel wires, N, for several cases
unit length due to the skin effect Rskin of: of constant gap spacing, g, between wires normalized
RS 1 ωµo by wire diameter, dw = 2a., fixed c/a = ((g/dw)+1).
Rskin = = (ohm/m) . (11)
2π a 2π a 2σ
Added$Normalized$Wire$Resistance,$Proximity$Effect$
Here RS is the surface resistance of the wire at high 100$
gap$=$0.125*diam$
frequencies, where the skin-depth is much less than gap$=$0.25*diam$
Added$Wire$Resistance$$[Rp/Ro]$

the wire radius, a, so ds << a. The skin depth, ds, for gap$=$0.5*diam$
gap$=$diam$
a conducting wire material and frequency ω is: 10$
gap$=$1.50*diam$

2
ds =
ωµoσ (m).
1$

Proximity Effect Resistance Analytic Calculation

The impact of the adjacent wire geometry is to 0.1$

0$ 2$ 4$ 6$ 8$ 10$ 12$
increase the effective resistance, and hence increase Total$Number$of$Parallel$Wires$$$(N)$
the losses per unit length when the wires are closely
spaced. The base loss process is due to the skin Figure 4 Calculated normalized added resistance per unit
effect and yields an equivalent resistance per unit length, (Rp/Ro), due to the proximity effect for N series-
connected parallel wires, fixed gap/wire size, g/dw.
length of the N series connected wires, Ro. The
proximity effect introduces an added resistance loss,
The added resistance increases dramatically with
Rp, which is expressed here normalized to the skin-
number of parallel wires, more than 10-fold for 10
effect resistance, Ro. (Both resistance values scale
versus 2 wires. These losses also increase very
linearly with the square-root of frequency and
strongly as the gap between wires is reduced. And
linearly with the value of surface resistance RS.)
the effect is made more severe with increased number
of wires. For 10 wires the loss approaches 30 times
The added proximity-effect resistance, Rp, can be
increase for small gap spacing, but reduces to less
normalized by the no-proximity effect resistance, Ro,
than 2x loss for a larger wire gap spacing of 1.5 wire
to get a non-dimensional measure of the impact of the
diameters or more.
COMSOL® with AC/DC Module Solver
Solutions

The parallel wire proximity effect problem analytic

solution assumes quasi-static fields and hence the
AC/DC Module with 2D arrangements provides a
good platform for software numerical simulations.
The main challenge is setting the grid so as to obtain
accurate magnetic fields as well as resultant wire
resistance values by post-processing. Because the
wire resistance requires the solution for fields and
currents inside the wires the grid arrangement is (6a) Full 24 mm width
concerned with both inside and outside the wires.
Figure 5 depicts both an example of a full grid
arrangement for 3 wires (3.175mm diam.) at a
spacing of c/a = 1.50 and the corresponding detail at
1000x magnification near one outside of an outer
wires. To match the analytic solution a near-surface-
current must be enforced and this was implemented
by use of calculations with a 100 MHz frequency,
which yields a skin depth of 6.5 µm in comparison to
3.125 mm wire diameter. Figure 6 depicts the
corresponding magnetic flux density plots for these
grid views. All wires carry the same 1 ampere of
current. (6b) Expanded 24 µm width
Figure 6 Magnetic flux for simulation solutions for outer
wire of 3 wires

When there are more wires, the more severe field

redistributions occur at the outer wires and there is
less effect on the inner wires. This occurs because the
net disturbance external fields are greater at the outer
wires and smaller near the center. Figure 6 depicts
the resultant magnetic flux for 20 wires.

(5a) Full 24 mm width

Figure 7 Magnetic flux plot for 20 wires, 1 amp each, c/a

= 2.00, Scale: Red = 350 µT to Blue = 0.

Comparison of Analytic and COMSOL®

Solutions

Both surface current distribution and added resistance

associated with the proximity effect were evaluated
(5b) Expanded 24 µm width by analytic and by COMSOL® methods. For
example with 3 wires the surface current distributions
Figure 5 Grid arrangement for simulation solutions for 3
yielded excellent agreement as seen in Figure 8.
wires with spacing c/a = 1.50
gm( )
c/a= 2
but the error became larger, as much as several
percent when 15 or more wires were present. The
1.5
cause for greater error with more wires may be due to
a more limited grid/calculation range external to the
1.0 Wire 1 wires when there were more wires. The external grid
Wire 2
range used for this figure was essentially constant
0.5 and did not increase in proportion to the number of
wires. This caused some distortion to the magnetic
fields, and hence could also disturb the surface
0.5 1.0 1.5 2.0 2.5 3.0
current densities and ultimately the resistance Rp.
(8a) Analytic, Mathematica®: blue = wire-1
Proximity Normalized Average Per Wire Loss, Rp/Ro
0.60

0.50

0.40

Rp/Ro
0.30
Analytic
0.20
COMSOL
0.10

0.00
0 5 10 15 20 25
(8b) Simulation, COMSOL®: green = wire-1
Number of Parallel Wires
Figure 8 Surface current distributions for 3 parallel wires,
c/a = 2.00. Normalized to single wire surface current Figure 9 Proximity factor Rp/Ro obtained with both
density, Js = 1.574 x 10-6 [A/m2] Analytic Mathematica® and COMSOL®

The ratio of corresponding normalized wire Conclusions

resistance values Rp/Ro for theory and by COMSOL®
are shown in the Table 1 for 3 wires and exhibit less The proximity effect provided an excellent vehicle to
than 0.4% deviations, where Rp is the calculated evaluate the abilities of electromagnetic simulation
resistance with proximity effect and Ro that software such as COMSOL® with the AC/DC
calculated for skin effect resistance of a single wire. Module since it demands both accurate field
calculations with very high dynamic range and post-
Table 1 Proximity Effect Resistance Comparison: processing to yield very localized currents and bulk
3-Parallel Wires Spaced c/a = 2.00 resistance properties. The solutions with COMSOL®
Method
Rp/Ro Rp/Ro Rp/Ro (ave were found to be in excellent agreement with the
outer center of 3-wires) analytic values calculated with Mathematica® for
Theory 0.4986 0.0390 0.3455 corresponding wire configurations, results differed by
less than ½%, for smaller numbers of wires. Both
COMSOL® 0.4968 0.0391 0.3443 surface current distributions and resultant net
resistance values were accurately quantified by the
Note that Table 1 also shows the losses for individual software in comparison to the analytic solutions for
wires were fully consistent between theory and the wide range of test conditions.
simulation. As always the 2 exterior wires exhibit
greater resistance than the interior/center wires. References
1. Smith, G., Proximity Effect in Systems of Parallel
The average proximity loss factor, Rp/Ro, was Conductors. J. Appl. Phys , 2196-22013 (1972)
determined for a range of different numbers of wires 2. Haus, H., & Melcher, J. Electromagnetic Field and
with a fixed spacing of c/a = 2.00. Figure 9 shows Energy (1998). Retrieved
the resultant steady increase in added average per- from:http://web.mit.edu/6.013_book/www/chapter12/
wire loss as quantified by the increased Rp/Ro values 12.3.html
according to the increased number of parallel wires. 3. Gradshteyn, I., & Ryzhik, I. (2007). Table of
Integrals, Series, and Products. San Diego: Elsevier
There is excellent agreement (less than 0.5% error) Inc. A. (2007)
between the COMSOL® solution and the analytic Acknowledgement This work partially
solution when the number of wires is less than ~10, supported by ProlecGE.