This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
Recoverable Energy of Radiating Structures
Guy A. E. Vandenbosch, Fellow, IEEE
Abstract In this paper the maximum recoverable energy of a
radiator is determined. This energy is defined as the maximum
energy that can be recovered from the electromagnetic field
distribution in entire space generated by the radiator up to the
time point considered. The calculation procedure is based on
minimizing the future radiated energy. It is shown that the concept
of recoverable energy has no problems with so-called negative
energies, nor with any dependency on coordinate system, issues
that compromise many definitions found in literature of the
concept of stored energy for a radiator. The question can be
raised whether recoverable energy can be identified with stored
energy for a radiator.
Index TermsRadiative and reactive energy, stored energy, Q
factor, causality, time domain, frequency domain.
I. INTRODUCTION
The electric and magnetic energies stored around a device have
been studied mainly in the frequency domain. Although the first
papers were published about 50-60 years ago [1], [2], and many
authors have been investigating stored electromagnetic energies
and Q factors since then [3], [4], [5], [6], [7], [8], [9], [10], [11],
[12], [13], [14], even in frequency domain it was only recently
that expressions were formulated to calculate the reactive
energies explicitly in terms of the currents [15]. These
expressions have led to new, more effective studies of Q [16],
[17], [18], [19], for several types of radiators and are being used
in convex optimization problems involving electromagnetic
energies [20]. Introduced by Gustafsson and Nordebo, convex
optimization promises to become an extremely powerful tool in
antenna design.
However, this did not end the quest for the ultimate expressions
for stored energy. It was shown in [18] that in some cases the
expressions of [15] for reactive energies may lead to negative
values, which is of course unphysical for stored energy. This
has been picked up by several authors, who delivered additional
contributions to this field of research [21], [22]. The only
possible conclusion is that at this moment in time, there is still
no consensus in the community on the rigorous line of
reasoning to determine reactive and/or stored energy, and the
related expressions to calculate them. The rigorous solution of
the problem is still an open question.
Although there are literally dozens of papers on reactive energy
and Q factors in frequency domain, to the knowledge of the
author, there are only five papers that thoroughly study electric
and magnetic energies in time domain [23], [24], [25], [26],
[27]. In [24], Shlivinski uses a decomposition of the field into
The author is with the Department of Electrical Engineering, Katholieke
Universiteit Leuven, Belgium (e-mail: guy.vandenbosch@kuleuven.be). The
0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1
spherical modes. This automatically means that the resulting
analysis is an approximation, taking into account only the
energy outside a spherical surface with radius a. Typically, this
radius is the smallest radius that completely circumscribes the
radiator. The expressions derived in [25] and [26] do not rely
on this decomposition. They are general and rigorous.
However, in essence they only give the total energy
corresponding to the field distribution in entire space. They do
not directly give a physical concept for stored energy that
satisfies a necessary condition: being strictly positive for all
possible current distributions.
In essence, nowadays mainly two different techniques are used
to determine an approximation for the total stored energy: the
indirect technique as described in [5] based on the Q factor
calculated from the derivative of the impedance, and the
technique as introduced in [1] and rigorously worked out in
closed form in [15] based on the subtraction of the radiated
energy from the total energy. Other more recent techniques are
based on Brune synthesis [22], and the technique introduced by
Capek [27]. Note that all these techniques deal with total
energies, i.e. integrated over entire space, not with local energy
densities, as for example pioneered also by G. Kaiser [30], [31].
In this work, the following definitions are used. The total
energy is the energy deposited in the field throughout whole
space. It can be calculated by integrating the classical electric
and magnetic energy density over entire space. For signals with
a finite time span, no infinities occur and there is no problem to
perform the integration explicitly. Expressions for this energy
can for example be found in [25], [26], but also the classical
expression (1), see further, is applicable. In frequency domain,
this energy goes to infinity.
The reactive energy is the (finite) energy obtained when
applying the extraction technique as introduced by Collin and
Rothschild [1] in frequency domain. The expressions in [15]
involve a variant of this extraction technique, leading to closed
form expressions that overcome the dependency on coordinate
system. For some topologies, they deliver negative reactive
energies. However, this seems to be an inherent feature of many
subtraction techniques [18], [22].
As far as I can see, there is no unique definition of stored
energy widely accepted in literature and valid in all cases.
Stored energy is well-defined in the limit for static cases, and
also for pure circuits, but as soon as a radiator is involved,
problems occur. It is well-known however that the reactive
energy in the limit for frequencies going to zero (or for small
radiators) yields the stored energy. Further, in many
publications, it is implicitly assumed that stored and reactive
energy are the same, but this is not the best possible solution.
core of the work was performed within the framework of a sabbatical stay at
the Electronic Engineering Department, Tsinghua University, Beijing, China.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
2

Recovered energy is the energy that is going back to the radiator

from the field distribution in space. Physically, this can be

Wrad J 2 (t1 ) E21

(t1 ) dV2dt1 (4)
realized for example by switching off the excitation, after which V2

energy is received by the radiator and transferred to the load

resistance. In this work it will be shown that it is possible to The superscript indicates that time is taken up to infinity.
define a maximum recoverable energy. It is the maximum of With the simple time coordinate transformation
the energy that can be recovered at time t, i.e. in the future from t1 r21 / c t1 , it is easily shown that
time t up to infinity. Physically, one can imagine that this is
realized by modifying the load of the radiator in time in such a
way that the actually recovered energy is maximized. It will be J 2 (t1 ) E21

(t1 ) dV2dt1 J (t ) E
2 1

21 (t1 ) dV2dt1 (5)
proven that this energy shows several very important physical V2 V2
properties: 1) it is uniquely defined based on a physical line of
reasoning, 2) the problem of the coordinate dependency of the so that (4) can also be written as
reactive energy (as discussed thoroughly in [22]) does not
occur, 3) the problem of negative energies (as discussed
1
V J 2 (t1 ) (E21 (t1 ) E21 (t1 ))dV2 dt1

thoroughly in [18]) also does not occur since recoverable Wrad (6)
energy is always strictly positive. The calculation procedure is 2 2

based on choosing the future currents in such a way that

radiation is minimized. Examples and numerical results are An interesting question is: what is the energy that is recovered
given for harmonic time dependencies. They illustrate that for from the total energy deposited in the field distribution at a
small radiators, the new technique can be interpreted as a higher certain time t by the future current, i.e. in the interval t , .
order correction to the existing formulations of stored energy.
Only realistic signals, i.e. signals that are exponentially (or
faster) going to zero for t , are considered. The total
II. RECOVERABLE ENERGY IN TIME DOMAIN
radiated energy subtracted from the total energy deposited in
Consider an arbitrary current distribution J s flowing in a the field distribution at time t gives
volume Vs embedded in free space with permittivity 0 and
Wrec Wtot Wrad
permeability 0 . Looking further and further into the past, the t (7)
1
V J 2 (t1 ) (E21 (t1 ) E21 (t1 ))dV2dt1

current goes exponentially (or faster) to zero for t . The
2
classical expression for the total energy deposited in the field at 2

time t is 1
2 t V2

J 2 (t1 ) (E21 (t1 ) E21 (t1 ))dV2dt1
t
Wtot J 2 (t1 ) E21

(t1 ) dV2dt1 (1)
V2 When Wrec is positive, the energy is really recovered by the
current distribution. When it is negative, actually more energy
with
is radiated into space in the time interval t , .
r21
J1 (t1 )
1 1 c 1 ( (t r21 ) 1 ))dV (2) An even more interesting and (to the best knowledge of the
dE (t1 ) (
4 0 c 2 t1 author) unsolved question is: what is the maximum of this
21 2 1 1 1
r21 c r21
recovered energy in terms of the future current distribution, i.e.
E21 (t1 ) dE21

(t1 ) (3)
look in the interval t , for the current that maximizes (7).
V1

The subscripts 1 and 2 both replace the subscript s. They are This problem can be solved with variational calculus. The
used to indicate a double integration over the volume. This solution is found by first assuming a current J s from time t up
means thus that V1 V2 Vs . r21 is the distance between to time infinity, and assuming on top an elementary change
points 1 and 2 and c1 0 0 is the speed of light in free dJ s of this current. After some straightforward calculations

space. The charge distribution generated by the current is the change dWrad due to dJ s is found to be
s J s dts . E21 is the normal causal electric field.

ts
dWrad dJ 2 (t1 ) ( E21

(t1 ) E21 (t1 ))dV2dt1
Already here, we define also E21 , which is the retrocausal t V2

electric field that will be used further in the line of reasoning. 1 r21 r 1 (8)
For currents radiating in a finite time interval only, for t

8 0c 2 dJ
V2
2 ( t ) ( J1 ( t
V1
c
) J1 (t 21 )) dV1dV2
c r21
the total energy deposited in the field becomes the total radiated
energy, so that dJ 2 (t1 ) (dE21

(t1 ) dE21 (t1 ))dV2dt1
t V2

0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
The second term in (8) is a consequence of the sudden jump at
time t if dJ s (t ) 0 . The third term gives the contribution of
the current dJ s and the fields generated by this same current
dJ s . It is well-known that any current that is finite in time
domain generates a total radiated energy over its time frame
that is always positive. Physically this means that without any
incident waves coming from outside, any radiator has to
radiate. Since actually dJ s is such a current, spanning the time
frame t , , and the third term can be identified with the total
energy radiated by dJ s , this third term is always positive. Note
that the first two terms are first order terms and the third term is
a second order term. The solution where the first order terms
yield zero contribution for any possible dJ s corresponds to the
situation where the total radiated energy is stationary.
Obviously, this is the case when
E21 (t1 ) E21
(t1 ) E21 0 t1 t and over Vs
r21 r 1
V (J1 (t c ) J1 (t c21 )) r21 dV1 0 at t and over Vs
1
This is actually a set of two equations in the future current
J s for the time interval t , . The current before t is of course
given since this is in the past. In the rest of this section we will
assume that this optimal current J s is known. The way to
determine it is discussed in the following section. The fact that
the third term in (8) is always positive automatically means that
this stationary point is a minimum. The maximum recoverable
energy is then
t t, since (9) and (10) have to be obeyed. However, essentially,
1
V J 2 (t1 ) (E21 (t1 ) E21 (t1 ))dV2dt1
max
Wrec
2 2
1
2 t V2
J 2 (t1 ) ( E21 ( t1 ) E21 ( t1 ))dV2dt1
J 2 ( t1 ) E ( t1 )dV2dt1 J 2 ( t1 ) E ( t1 )dV2dt1
21
21
t V2 t V2
The second and third line are obtained by using (5) and (9).
Since (5) is valid for any current, it is also applicable to the
isolated future current, i.e. put the past current identically equal
to zero. This also delivers
1
2 t V2
,mod
max
Wrec J 2 (t1 ) (E21 (t1 ) E21 (t1 ))dV2dt1 Wrad ( J 2f ) (12)
where the superscript mod means that this causal field has to be
calculated with the current in the past put to zero. The lower
boundary t- indicates that the integration has to start just to the
left of t, where the current is still zero. Since at times
t1 t max(r21 / c) with max( r21 ) the maximum distance
0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
3
between any two points of the radiator
,mod
E (t1 ) E (t1 ) E (t1 ) a last expression is
21 21 21
t max( r21 / c )
1
J (t ) ( E
,mod
max
Wrec 2 1 21 (t1 ) E21 (t1 ))dV2dt1 (13)
2 t V2
Expressions (11)-(13) give the maximum energy that can be
considered as recoverable at time t , in the sense that it is not
possible to extract more energy than this from the field
distribution. The newly defined quantity shows two very
important properties.
1) From the physical line of reasoning followed, it is clear
that the recoverable energy is strictly positive. Indeed,
according to the principle of conservation of energy, all energy
deposited in the field at time t at later times either has to be
absorbed back by the current distribution or has to be radiated
to infinity. Minimizing the total radiated energy naturally
corresponds to a situation where the current distribution indeed
absorbs a maximum positively valued energy again. Moreover,
expression (12) rigorously proves that recoverable energy is
(9)
positive, since this is also the expression for the energy radiated
(10) by the isolated future current J 2f , which is known to be
always positive.
2) The recovered energy is basically defined as a subtraction
of two terms, Wrad is subtracted from Wtot , see (7). The first
term is the total energy deposited in the field distribution at the
observation time t. Considering the fact that the expression used
to calculate this energy, i.e. (1), is obtained starting from the
basic definition of energy generated by the current distribution,
this term is gauge invariant, meaning that changing the gauge
does not change the value of the energy. The second term is the
total energy radiated by a current distribution with a very
special time profile for times greater than the observation time
this is nothing else than using (1) for time going to infinity. Also
this term is thus gauge invariant. This means that the
(11) recoverable energy is also gauge invariant: working in a
different gauge delivers the same energy values.
A. Modal Current based solution
In principle, the future current can be determined based on the
equations (9) and (10). A practical and very robust way is to
write the current in the interval t , as a superposition of a set
of decaying modes. The amplitudes of these modes can be
determined by minimizing the total energy radiated over the
whole time frame , . Using the decaying modes in the
future, the current can be written as
J s (t1, r) J e (t1, r)u(t t1 ) J mt ,ms e smt ( t1 t )u(t1 t ))J ms (r) (14)
mt ms
The subscript e refers to the known current in the past, the
subscript mt refers to an exponentially decaying time
dependency and the subscript ms refers to a spatial current
distribution mode ms. The function u (t ) is the Heaviside step
function. Note that this way of working automatically
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
4

incorporates a possible jump of the current at the observation of the integrals. On this axis, Plancherels theorem is valid, and
time t. In the bilateral Laplace domain, after applying the by inserting (15) the total radiated energy given in (6) can be
bilateral Laplace operator L, this becomes worked out as

j
L( J s (t1 , r )) J s ( s, r ) 1
4 j j V2

(15) Wrad J *2 ( s ) ( E21 ( s ) E21 ( s ))dV2ds
st
e
J e ( s, r) J mt ,ms J ms (r)
s smt 1
j

4 j j V2
mt ms
J *e ( s ) E21, e ( s )dV2 ds

with s j the Laplace parameter. Note that the same

1
j
(18)
symbols are used for quantities in the time domain and in the J ( s) Z
*
( s )dV2 Tmt ( s )ds J mt ,ms
Laplace domain. The difference is explicitly indicated by the 4 j j V2
e ms

mt ms
dependency on t or s. The region of convergence of J e ( s, r ) is
1
j
*
( t1 t ) J
*
E21, e ( s )dV2 Tmt ( s ) ds J mt ,ms
*
Re( s) where 0 and e

is the slowest decaying 4 j j V2
ms

mt ms
exponential in J e (t1 , r ) for t1 . The region of
1 *
j

convergence of the mode mt is Re( s) Re( smt ) where J mt ,ms Tmt ( s )Z ms ,ns ( s )Tnt ( s )ds J nt ,ns
*

mt ms nt ns 4 j j

Re( smt ) 0 . Note that the imaginary frequency axis with
,ns ( s ) J ns Z ms ( s )dV2 Z ms ,ns ( s ) Z ms ,ns ( s )

0 is always part of the region of convergence delivering Z ms *

V2
contributions to the current. In the rest of this section, all
expressions are in the Laplace domain.
In the Laplace domain, the difference between the causal and with Z ms ,ns
describing the (Laplace domain) causal impedance
the retrocausal fields generated by this current can be written as
relation between spatial modes ms and ns, and Z ms ,ns
its
retrocausal equivalent. Note that their difference in the case that
( 1) s 2
4 s 0 c 2 V1

E21
( s ) E21 ( s) ( J eG21 dV1 2 (1 J e )G21 dV1 )
ms equals ns, i.e. Z ms ,ms
describes the relation between the
V1
modal current distribution and the power that it is radiating. It
( 1) s 2

4 s 0 c 2 V1
( J msG21 dV1 2 (1 J ms )G21 dV1 )Tmt ( s ) J mt ,ms
is easily seen that Z ms
,ns ( s ) Z ms ,ns ( s )
, which means that
mt ms V1 (16)
Z ms ,ns
is even in s. It is also easily checked that on the
imaginary axis in the Laplace domain, i.e. the frequency axis,
e ( s ) E21,e ( s )) ( Z ms ( s ) Z ms ( s ))Tmt ( s ) J mt ,ms

(E21,
mt ms the self-impedance Z ms ,ms
is purely real, as it should. In other
e ( s ) Z ms ( s )Tmt ( s ) J mt ,ms

E21, words on the imaginary axis we can write
mt ms
s s
( ) r21
,ms ( j ) Rms ,ms ( ) jX ms ,ms ( )
( ) r21
(e c
e c
) e st Z ms
with G = and Tmt ( s )
s smt
21

,ms ( j ) Rms ,ms ( ) jX ms ,ms ( )

r21
Z ms (19)
Z
( j ) Z
( j ) Z
( j ) 2 Rms ,ms ( )
with Z ms the (Laplace domain) causal impedance vector, ms ,ms ms ,ms ms ,ms

describing the relation between the modal current distribution Z

ms ,ms ( j ) Z
ms ,ms ( j ) Z
ms ,ms ( j ) 2 jX ms ,ms ( )

ms and the three components of the causal field, and Z ms the
(Laplace domain) retrocausal impedance vector, describing the with Rms,ms and Xms,ms real, and with Rms,ms even and Xms,ms odd
relation between the modal current distribution ms and the three in . Again, the pure inductive effect and the pure capacitive
components of the retrocausal field. Note the minus sign, due effect may be subtracted from both the classical and the
to the fact that the radiator has to be considered as a generator. retrocausal impedances, without altering the result in (16).
It is easily seen that since J ms is independent of s, Using matrix notation the expression (18) can be written as

Zms ( s) Zms ( s) . It can be useful to write (16) with
Wrad Wee* MTen* J*m JTm Mme* JTm Mmn* J*m (20)

s s s s
( ) r21 ( ) r21 ( ) r21 ( ) r21
e c
e c (e c
1) (e c 1) (17) J m is a column matrix gathering all the current coefficients
J mt ,ms . The other quantities are defined as follows.
which means that both the pure inductive effect ( s and due to j
1
4 j j V2

current) and the pure capacitive effect ( s 1 and due to charge) Wee* J*e ( s) E21, e ( s )dV2 ds
(21)
may be subtracted from both the causal and retrocausal
impedance vectors, without altering the result. This quantity is the total energy radiated by the system when
Since the frequency axis is always part of the region of the current would suddenly drop to zero at the observation time
convergence, this axis can be chosen for the actual calculation

0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
5

t. M en* is an infinite column matrix with for each modal B. Wiener-Hopf method based solution
combination (nt,ns) the element An alternative way to solve the problem is to start from the
j
* e s*t equation
1
4 j j V2

M e,nt ,ns* J E ( s )dV 2 ds (22)
ns 21,e
s* snt*
J s ( s, r ) J ms ( s)J ms (r )
M mn* is an infinite square matrix with for each set of two modal ms (30)
combinations (mt,ms) and (nt,ns) the matrix element ( J msp ( s ) J msf ( s ))J ms (r )
j ms
e st e s t
*
1
4 j j s smt

M mt ,ms ,nt ,ns* Z ( s ) ds which means that the current is decomposed into spatial current
s* snt*
ms ,ns
(23) modes J ms with coefficients depending on s in the Laplace
1 1 domain, which yields time varying coefficients in time domain.
( Z ms ,ns ( smt ) Z ms ,ns ( snt ))
*

2 ( smt snt* ) These spatial current modes can be the same as in the previous
The second line is obtained from the first one after applying the section. The superscripts p and f indicate the past and future
technique as given in the appendix. The fact that Z ms
is even current, respectively. Testing (as in a method of moments
,ns
procedure) of expression (16) with a spatial mode J*ns (r )
in s is explicitly used here. M me* is an infinite column matrix
delivers
with for each modal combination (mt,ms) the element
e st *
j
(E

1 ( s ) E21 ( s )) J *ns dV2
J e ( s ) Zms ( s)dV2 ds

M mt ,ms ,e* (24) 21

4 j j s smt V2

V2
(31)
( 1) s 2
The total radiated energy in (20) can be minimized in terms of ( 2 J msG21 dV1 2 (1 J ms )G21 dV1 ) J *ns dV2 J ms ( s )
ms V2 4 s 0 c V1
the modal coefficients J mt ,ms by calculating the derivatives with V1

Z ms

,ns ( s )J ms ( s )
respect to each of these coefficients and equating the resulting ms
expressions to zero. This yields
Gathering the equations for all ns in matrix form yields

dWrad (MTen* JTm Mmn* ) dJ*m dJTm (Mme* Mmn* J*m ) 0 (25)
Em ( s) Zmn ( s)J ms ( s) (32)
It is easily proven that the second term is the complex conjugate
of the first term. The expression (25) is zero for all arbitrary where J ms is a column matrix gathering the coefficients J ms ,
dJ m only if
Zmn is a matrix gathering the elements Z ms ,ns
(note that these
are the same as in (18)), and Em is a column matrix gathering
J m (M*mn* )1 M*me* (MTmn* )1 Men* (26)
the projections of the field. Note that the minus superscript has
been dropped in the matrix notation, since it will be used a bit
The minimum total radiated energy is then given by the
further with a different meaning. Further, we will also drop the
expression
dependency on s. Note that on the imaginary axis Zmn is
1
Wee* MTen* Mmn Hermitian. Interpretation of (9) in the Laplace domain results in
* M me*
Wrad (27)
the fact that all elements in Em are analytic in the complex left
Based on (7), the maximum recoverable energy is written as half plane with 0 . This is a direct consequence of being
zero in the future. The core of the well-known Wiener-Hopf
1
Wtot Wee* MTen* Mmn method consists of writing the current column matrix explicitly
* Mme*
max
Wrec (28)
as a sum of the past and future current column matrix, analytic
in the left and right complex half plane, respectively, and
Since the total radiated energy by a current suddenly dropping writing the coupling matrix as a product of two matrices,
to zero at t is also given by (1) it is easily seen that Wtot Wee* analytic in the left and right complex half plane, respectively.
and the final expression for the maximum recoverable energy Being analytic in the left or right complex half plane is further
becomes indicated by a minus or plus superscript, respectively.

J ms J ms J ms J ms
p
J ms
f (33)
1
max
Wrec MTen* Mmn * Mme*
(29)
Zmn K mn
K mn
with K mn
( s) K mn ( s) (34)
Expression (32) can then be written as
which is a positive semidefinite quadratic form. In the results Em K mn

K mn
(J ms
J ms ) (35)
section, this expression will be worked out in full details for a
which can also be written as
few representative (canonical) structures.
(K mn )1 Em K mn

J ms
K mn
J ms (36)

0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
6

Since also K mn J ms can be written as the sum of a part analytic
1 Z e,e ( se ) Z e,e ( se ) 1 2
*
in the left half plane and a part analytic in the right half plane,
Wee* Je (42)
i.e. 2 ( se se )
*
Ce,e se se*

K mn J ms P (K mn

J ms ) P (K mn

J ms ) (37)
1 Z e,ns ( se ) Z e,ns ( snt )
*
1

where P and P are the classical operators for this operation M e,nt ,ns* * e
J (43)
(see Appendix II and [33]), this also means 2 ( se snt )
*
Ce,ns se snt
1 Z ( s ) Z ms ,ns ( snt )

(K mn )1 Em P (K mn

J ms ) P (K mn

J ms
) K mn
J ms (38) *
1
M mt ,ms ,nt ,ns* ms ,ns mt *
(44)
Expression (38) equates a left hand side analytic in the left half 2 ( smt snt )
*
Cms ,ns smt snt
plane to a right hand side analytic in the right half plane. This
1 Z ( s ) Z ms ,e ( se ) *
*
is only possible when both sides actually consist of entire 1
M mt ,ms ,e* ms ,e mt * e
J (45)
functions. If both sides go to zero for s going to infinity, 2 ( smt se )
*
Cms ,e smt se
according to Liouvilles theorem both sides are identically zero.
The future current is then found as
Note that this means that the minimum total radiated energy is
J ms (K mn )1 P (K mn

J ms ) (39)
fully expressible in terms of the causal Laplace impedances
Using (12) in the Laplace domain the recoverable energy is then evaluated at the complex frequencies of the chosen modes and
j
the complex excitation frequency se . The physical
1
4 j j

max
Wrec ( J ms ( s ))T * Z mn ( s ) J ms ( s )ds interpretation is that these causal impedances contain all
necessary information on the spatial shape of the radiator. For
(40)
1
j
a purely sinusoidal regime e 0 , and (42) becomes
4 j j

((K mn ) 1 P (K mn

J ms
))T *K mn
K mn
(K mn ) 1 P (K mn

J ms )ds

1
j Re,e (e ) 1 X e,e (e ) 1
Wee* ( 2
2
( P (K mn (46)

p
J ms ))T * P (K mn
p
J ms )ds ) Je
4 j j 2e 2 2e Ce,e

Although formally, this method is very elegant, the actual Again, this contribution corresponds to the energy that is
calculation of (40) in many cases is formidable since the radiated by a current suddenly dropping to zero at the
Wiener-Hopf matrix factorization in (34) is absolutely not observation time t. It is obvious that the energy stored at that
trivial, but a well-known challenge in computational physics moment in the capacitor remains in the capacitor and does not
[34]. To date, a general technique for arbitrary matrices is not contribute to radiation. The expression for the maximum
available. A more detailed discussion is beyond the scope of recoverable energy becomes
this paper. In the examples section, two canonical topologies
will be discussed for which the Wiener-Hopf technique can be 1 1
max
Wrec MTen* Mmn * M me*
(47)
worked out in full detail: the spherical loop and the spherical 2 Ce,e
2
e
dipole.

III. RECOVERABLE ENERGY IN FREQUENCY DOMAIN Note that the matrix M mn* is Hermitian, so that M mn1 * is also

The term capacitive impedance is used for a causal Hermitian, and the matrices MTen* and M me* are each others
impedance with a pole at s 0 , i.e. in the limit to zero complex conjugate transpose. This ensures that Wrec max
is real, as
Z 1/ (Cs) with C the static capacitance. In this case, in the it should. Note that it is also easy to prove that in general
modal current based method it is mandatory to use the
1 X e,e (e ) 2 1 Z e,e (e )
technique of (17) and work with the modified impedances,

lim J e lim
2
where at least the capacitive effect has been subtracted.
max
lim Wrec Je (48)
e 0 e 0 2 e 0 2
Otherwise, convergence problems may occur when the
technique as outlined in the appendix is applied in order to
determine the matrix elements in (29). The procedure is linked The proof starts from the Taylor series expansion around 0
with the fact that a charge distribution constant in time domain of any arbitrary impedance, which is
does not contribute to radiation. For an exponential current in
the past given by J e (t1, r) J e ese ( t1 t ) J ue (r) the expressions (21) Z
1
j L R 2 O ( 3 ) (49)
- (24) can all be further worked out without any complications jC
based on the technique given in the appendix. The final result,
with the substitution For capacitive structures the question can be raised whether the
static zero radiative mode is allowed to contribute to the future
1 current. This mode leads to an infinite charge accumulation for
Z ( s) Z ( s) (41) times going to infinity, which is a matter that could seem quite
Cs
unphysical. Note that this mode is automatically taken into
account in the Wiener-Hopf technique, but can be omitted in
which eliminates the pole at s 0 , is

0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
the modal current technique. This matter will be further
investigated for the examples involving the spherical dipole and
the dipole with imposed current.
The term inductive impedance is used for a causal impedance
with a zero at s 0 , i.e. in the limit to zero Z Ls with L
the static inductance. The same expressions can be used as for
a capacitive impedance, but with C . consequence of the single port circuit paradigm. However, this
For currents that have an exponential behavior in the past it can
p
be written that J ms ( s) J ms
e
( 1/ ( s se )) , which means that
using the Wiener-Hopf solution, (40) becomes
1 j
( P ( K mn / ( s se )))T * P ( K mn
max
Wrec ( J ms )
e T*
/ ( s se ))ds J ms
4 j j
( J ems )T * Zmn
rec e
J ms
which is a quadratic form involving a positive semidefinite
matrix.
Although the procedures work for any set of spatial current
modes, i.e. a recoverable energy will always be found for that
specific set, it is advantageous to choose the modes in the most
optimal way in order to approach the theoretical maximum of
the recoverable energy as much as possible. Further in this
paper, we will first choose the spatial current distribution modes
ms in a dedicated way, and derive the complex modal
frequencies smt in such a way that
Z ms ,ms ( smt ) 0
We will call these frequency roots the zero radiative roots for
the chosen set of spatial current modes.
However, note that in principle even subsectional basis
functions can be chosen, so that (50) can be derived from the
classical subsectional method of moments matrix, which is very
advantageous in optimization problems.
IV. DISCUSSION
To the knowledge of the author, there are only two publications
investigating the concept of recoverable energy in
electromagnetics: Polevoi in [32] and Direen in his Ph.D. thesis
[33]. However, the approaches followed there are quite
different from the one here.
In [32] the author deals with fields in a medium, not with
radiating currents. With his approach it is not clear how to
calculate a recoverable energy for a real radiator by optimizing
the future current distributions within the finite volume of the
radiator. The statement in [32] if the medium under study
is not dissipative, all of the energy A (equal, in this case, to W)
can be extracted back. is clearly not true in case of a radiator,
where part of the energy A, (which is the energy distributed
over space as given in (1)) will be radiated to infinity. The
fundamental difference is that the author assumes that future
currents can be put everywhere, which is not the case for a real
radiator, where the currents have to be restricted to the volume
of the radiator. This work however could be very useful when
the theory as introduced here has to be extended to radiators
involving more complex dielectric materials.
In Chapter 4 of [33] Direen essentially discusses the case of a
single port radiator fed by a transmission line. The paradigm
followed in his work is based on circuit theory. It does not
0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
7
involve the actual current distributions nor the fields linked to
the radiator problem considered, but basically departs from a
circuit model for the radiator. Although the technique of [33]
can thus be used to calculate the recoverable energy for a given
radiator, it cannot be used to optimize the topology of the
radiator and the current flowing on that topology. This is a
type of optimization is essentially one of the challenges in
antenna design today, as explained for example in [20]. There,
optimal currents are found in a certain (possibly partial) volume
of a radiator for a given condition imposed on stored and
radiated electromagnetic energy. In our work, the paradigm is
e within the realm of vector field theory. The line of reasoning
(50)
works with currents and fields and basically results in quadratic
forms, see (29) and (40). This may open the way to design
optimized topologies with optimized currents under a given
constraint for recoverable energy related quantities, similar as
in [20].
Furthermore, for the modal technique the set of current modes
used in the optimization in principle can be freely chosen, for
example based on physical considerations. In this way, the
effect of certain modes can be investigated. One of the issues
that certainly needs further investigation is the fact whether the
static constant current mode has to be incorporated or not in the
optimization procedure. This mode leads to an infinite charge
accumulation for times going to infinity. Not including this
mode clearly leads to a system with a finite span in time
(51) domain, which could seem to be more physical. This issue can
be investigated with the modal current technique, but not with
the Wiener-Hopf technique.
It is important to emphasize that the concept of stored energy is
only well understood and widely accepted in the case of pure
circuits. In these cases, the electric energy is concentrated in the
capacitors, and the magnetic energy is concentrated in the
inductors. A remarkable circuit is the so-called Zobel network.
This network consists of two parallel branches, one with a
resistor R and a capacitor C, and one with a resistor also valued
R and an inductor L. It is easily proven that if L / C R2 the
overall input impedance is always R, independent of the
frequency. This means that any resistance in any circuit can
always be replaced by a corresponding Zobel network, without
changing the overall input impedance, but with an increased
stored energy. This clearly illustrates the well-known fact that
for a given circuit input impedance, an infinite number of circuit
realizations is possible. They all provide the same input
impedance, and thus the same recoverable energy, but they do
not have the same stored energy. This phenomenon can only be
fully understood by doing an analysis in time domain for a real
input signal with finite time span, making use of Laplace
transforms. For example for the Zobel network, this clearly
shows that although the energy at any time drawn from the
source is RI 2 , in a first phase this energy is (partially) used to
build up the stored energy in the capacitors and inductors. After
this a steady state regime is reached, where the inserted energy
equals the energy dissipated in both resistors. In the end phase,
when the source has stopped, and no further energy can be
drawn from it, the energy stored in the capacitors and inductors
is transferred to the resistors, where it is dissipated. As such, in
time domain there is no violation whatsoever of the energy
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
8

conservation law. It only appears to be like that in frequency For capacitive cavity like structures the modal coupling matrix
domain, where the beginning and end phase are neglected and is not a diagonal matrix. This means that a matrix inversion is
a pure harmonic steady state regime is assumed. Note also that involved.
this means that in general the concept of stored energy (in
circuits) intrinsically cannot be linked for example to VI. EXAMPLES
bandwidth, since bandwidth is obviously only depending on In this section the newly developed theory will be illustrated by
input impedance, a point of view also taken by Direen in [33]. calculating the recoverable energy for five examples: 1. the
canonical case of a parallel opposite current double sheet
V. CAVITY LIKE STRUCTURES system, which is solved fully analytically, 2. the spherical
A very interesting type of spatial current distribution is the dipole and 3. the spherical loop, which are solved quasi-
cavity like type. This current distribution is the current that analytically, and finally 4. the loop with constant current over
flows at the surface of a cavity at resonance. For these modes the perimeter and 5. the dipole with imposed sinusoidal current,
the zero radiative roots lie on the imaginary axis. It is also easily which are solved numerically. The calculated (frequency
proven that in these cases these roots coincide with the natural domain) results involve the energies as listed in Table I.
roots, as defined in the Singularity Expansion Method [28].
The term cavity like is used because these topologies will not Table I. Energies calculated for the examples. Prad freq
is the
be considered solely as resonating cavities, i.e. at other complex frequency domain radiated power and tsto is the time it takes for
frequencies such a spatial current mode really behaves as a
radiator. the wave to travel over one radius a of the smallest
Assuming a single cavity like spatial current mode ms=1 at all circumscribing sphere of the radiator
times (both in the past and in the future) with a causal Symbol Name Reference
impedance Z ( s) , taking into account that for all its zero Wrea Reactive energy [15]
radiative roots Z ( jmt ) 0 , and only considering a purely Wrec Recoverable energy with all This paper
modes
sinusoidal regime with e 0 , the elements (43) (45)
Wdyn Recoverable energy without static This paper
become modes
Wt Wrea Prad
freq
tsto [26]
1 j 1
M e,nt ,1* ( ( R(e ) jX (e )) ) Je (52) WZ Impedance based energy [5], [15]
2 (e nt ) Cent
WWH Recoverable energy calculated This paper
1 1
M mt ,1,nt ,1* for mt nt (53) with Wiener-Hopf technique
2 Cmtnt
1 X (mt ) 1
M mt ,1,nt ,1* ( ) for mt nt (54) A. Two parallel opposite current sheets
2 Cmt2
Consider two parallel current sheets with opposite current. The
1 j 1 left sheet with current J J i z is located at x d / 2 and the
M mt ,1,e* ( ( R(e ) jX (e )) ) J e* (55)
2 (e mt ) Ce mt right sheet with current J J i z is located at x d / 2 . In the
Laplace domain the causal and retrocausal electric fields
It can be seen that for inductive cavity like structures (with generated at the location of the left plate by such a current
C ) the modal coupling matrix M mn* becomes a diagonal system are easily found as
matrix, so that the inversion in (26), (27), and (29) is easily
performed. For these structures the maximum recoverable 0 J
energy becomes E (1 e sd / c )i z (57)
0 2
1 ( R 2 (e ) X 2 (e )) 2 0 J
max
Wrec Je E (1 e sd / c )i z (58)
2 Le 2 0 2
1 (56)
X (mt )
The causal and retrocausal impedances relating the current

( R 2 (e ) X 2 (e )) J e
1 2 distribution to power are thus
mt (e mt )
2
2
mt 0
0
Z (1 e sd / c ) (59)
It involves an infinite summation with one term for each of the 0
modes. Note that the static term is extracted from the infinite 0
summation. It involves both the squared amplitude of the causal Z (1 e sd / c ) (60)
0
impedance and the static inductance of the structure.

Note that there is an extra factor two since there are two current
sheets. The radiated power per square meter is

0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
9

1 0 J s (t ) J m
j
(1 e je d / c )
1 (71)
Prad ( s) (2 (e sd / c esd / c )) J
2
(61) J
2 0 2 m d
m
(m e )
2 c
The zero radiative modes are found by solving the equation
With [29], formula 1.421.4, the infinite summation again can
0 be performed explicitly, and the result is simplified to
Z Z (2 (e sd / c e sd / c )) 0 (62)
0 1
J s (t ) J m (1 e je d / c ) J (72)
m 2
which delivers the zero radiative roots
This means that when e approaches one of the m , there is no
d
sm m( j 2 ) (63) current jump. This mode smoothly continues to flow. In that
c
case, other modes do not contribute to the future current. When
These modes thus lie on the imaginary axis and are perfectly e lies exactly in the middle between two consecutive m ,
equidistant. Since there is a zero at s 0 the topology is an there is a sudden current jump to 0. All modes contribute to the
inductive cavity like structure. The elements of the matrices in future current. It is easily checked that the current found
expressions (42)-(45) are easily calculated, even for e 0 , as satisfies equations (9), in a Fourier sense involving an infinite
summation, and (10).
Since all results are obtained analytically, and are thus rigorous,
1 0 this canonical example is very instructive.
Wee* (1 e d / c cos(e d / c)) J
2
e (64)
2 e 0 B. Spherical dipole and spherical loop
1 0 1 Consider the spherical electric dipole, i.e. a spherical shell with
M en* (1 ee d / c e je d / c ) J (65)
2 0 e j (e n ) radius a carrying a current given by J J sin i . Using
M mn* 0 for m n (66) classical spherical wave theory the causal surface impedance is
determined as
1 0 d
M mn* for m n (67)
2 0 c
0 sr sr
Z (rj ( )) (rh 2 ( )) (73)
1 0 1 0 r 1 jc r a r 1 jc r a
M me* (1 ee d / c e je d / c ) J * (68)
2 0 e j (e m )
The zero radiative roots are the zeros of the function
Note that (67) is calculated by taking the limit m 0 . For a
purely sinusoidal regime e 0 and the maximum sr sa sa ' sa
(rj1 ( )) j1 ( ) j1 ( ) (74)
recoverable energy per square meter becomes r jc r a jc jc jc

0 1 d d 1 This topology is a capacitive cavity like structure. The

)) (69)
max 2
Wrec (1 cos( J maximum recoverable energy has to be evaluated with (47).
0 4 2 c e
c m e d 2
(m ) Note that the expression for the derivative of the surface
2 c reactance at the non-zero roots can be simplified to

With [29], formula 1.422.4, the infinite summation can be

X (m ) 0 a 2 (75)
performed explicitly, and the result is simplified to (1 )
0 c m a 2
( )
0 d c

max 2
Wrec J and the static capacitance and inductance are found as
0 2c (70)
1 X (e ) 1 d 1 Z (e ) 2
3 7
( R(e )) J Wt J WZ C a 0 and L a0
2
(76)
2 2c 2 2 15

Note that this means that for this simple case the maximum
recoverable energy 1) is independent of frequency, 2) can be
written as first suggested in [26] with tsto d / (2c) , and 3) can
also be written using the classical impedance derivative
expression. From (14) it is seen that the current at t+ is obtained
by summing all modal coefficients. Using (26), this yields
This canonical topology can be solved quasi-analytically. The
only numerical step really necessary is the localization of the
roots. For this the transcendental equation based on (74) has to
be solved and this cannot be done analytically. A matrix
inversion is necessary. For larger mode numbers, it is easily
shown that m (a / c) m .

0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
10

In the Wiener-Hopf technique the factorization in (34) can be

analytically performed as

0 sr sr
Z Z Z ( rj ( )) ( rj ( ))
0 r 1 jc r a r 1 jc r a
(77)
0 1 c c 2 c c 2 2 sac

0 2
(1 ( ) ) (1 ( ) )e
sa sa sa sa
c c c c 2
sa
(1 ( ) 2 ) (1 ( ) 2 ) e c
sa sa sa sa

This technique can then be used without any further special

issues.
The energies are given in Fig. 1, top. It is seen that Wrea is
oscillating and may become negative, as was already known.
All the other energies are strictly positive. The fact that Wrec and Fig. 1. Spherical dipole. Top: normalized energies
(normalization is performed by multiplying with
WWH are identical illustrates the fact that the modal current
0 / 0 J / c (8 a3 / 3) , which makes the energies
2

technique, including the static mode, and the Wiener-Hopf

technique have to give the same result. It is seen that for this dimensionless), note that Wrec WWH Wt , bottom: Q factors
topology, the formula of [26] also gives the same result. This is for small ka.
probably specific for the spherical topology. The modal current
result without the use of the static mode, i.e. Wdyn , is a bit For this spherical topology, the asymptotic solution for e
smaller. However, WZ is considerably different in the region can be obtained analytically. The causal surface impedance
goes to
ka [1, 2] . It even reaches zero at ka 2 . Notice the points
where all energies become equal. This is at the frequencies
0 a a a
where the spherical dipole is a resonating cavity, i.e. no power lim Z sin(e )(sin(e ) j cos(e )) (78)
is radiated and all energies correspond to the real energy e 0 c c c
accumulated at the inside. The future current consists of the
resonating mode which is just continued. The corresponding Q Since for e the surrounding zeros go to m (a / c) m
factors for small ka are given in Fig. 1, bottom. It is clearly seen and all contributions due to the static capacitance C can be
that all methods give quasi identical results. neglected, the maximum recoverable energy (47) is easily
worked out as

2
0 J 4 3
max
lim Wrec ( a )
e 0 c 3
1 X (e ) a 2 8 (79)
lim Wt ( R(e )) J ( a 2 )
e 2 c 3
1 Z (e ) 2 8 2

lim WZ J ( a )
e 2 3

The last factor on the second and third line is due to the
integration over the volume, as seen in the expression (18).
Note that this means that for this case the asymptotic maximum
recoverable energy is 1) independent of frequency, 2) equal to
the expression as first suggested in [26] with tsto a / c , and 3)
can also be written using the classical impedance derivative
expression.
Consider the spherical electric loop, i.e. a spherical shell with
radius a carrying a current given by J J sin i . The fields
generated by this current are easily determined using classical
spherical wave theory. The causal electric field shows the same
spatial dependency as the current, i.e. E E0 sin i Z J .
The causal surface impedance is

0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
11

current technique result Wrec and the Wiener-Hopf technique

0 ( s 2 a 2 ) sa 2 sa
result WWH would be identical, this was not the case. This issue
Z j1 ( ) h1 ( ) (80)
0 c2 jc jc was investigated in full detail. It turned out that loop type
topologies, i.e. topologies where there are no charges, need an
with j1 and h12 the spherical Bessel and Hankel functions of extra modal current that is proportional with (t1 t ) . This issue
order 1. Knowing this causal impedance, first the zero radiative is related to the fact that for loop type radiators lim Z s4 .
s0
roots can be calculated, which are the zeros of the function
Once this mode was introduced, Wrec and WWH became
sa sin x cos x identical. The fact that Wt is also identical is probably specific
j1 ( )( 2 ) (81)
jc x x x sa for the spherical topology. The original modal current technique
jc result (without taking the modal current proportional to (t1 t )
into account) is now called Wdyn in the figure. It is only slightly
It is seen that also in this case these roots are located on the
imaginary axis. Since there is a zero at s 0 the topology is an smaller, just as WZ . Notice the points where all energies
inductive cavity like structure. Using the modal current become equal. This is at the frequencies where the spherical
technique a recoverable energy is easily evaluated with (56). loop is actually a resonating cavity, i.e. no power is radiated and
Since the topology basically involves current loops, the static all energies correspond to the real energy accumulated at the
modal current is included. This does not pose any physical inside. The future current consists of the resonating mode which
interpretation problems. Note that the actual energies require an is just continued. The corresponding Q factors for small ka are
integration over the volume of the radiator, as in (18). This leads given in Fig. 2, bottom. All methods give quasi identical results.
8 2
to an extra factor a . The expression for the derivative of
3
the reactance at the roots can be simplified to

X (m ) 0 a
for m 0
0 c (82)
0 a 1
for m 0
0 c 3

This makes the evaluation of the infinite summation very easy.

The proof is based on (80) and (81) and straightforward, but
beyond the scope of this paper. The static inductance is found
as

0 a 1 1
L a (83)
0 c 3 3 0

It is important to note that also this canonical topology can be

solved quasi-analytically. For larger mode numbers, it is easily
shown that m (a / c) / 2 m .
In the Wiener-Hopf technique, the crucial step is the
factorization in (34). It is easily proven that

0 ( s 2a 2 ) sa sa
Z Z Z j1 ( ) j1 ( )
0 c2 jc jc (84)
0 1 c c 2
sa
c c 2
sa

(1 ) (1 )e (1 ) (1 )e
0 2
c c
sa sa sa sa

which means that this factorization is analytically known. The

Wiener-Hopf technique can then be used without any further
special issues. Fig. 2. Spherical loop. Top: normalized energies (normalization
The energies are given in Fig. 2, top. It is seen that Wrea is is performed by multiplying with 0 / 0 J 2 / c (8 a3 / 3) , which
oscillating and becomes negative. All the other energies are
strictly positive. Although it was expected that the modal

0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
12

makes the energies dimensionless), note that Wrec WWH Wt ,

bottom: Q factors for small ka.
C. Loop
The loop was the first topology for which it was proved that the
stored energy evaluated with the formulas of [15] yields
negative values, see [18]. This topology thus delivers a crucial
test for the new paradigm. The results for the loop with constant
current along its circumference are given in Fig. 3. The radius
Rl of the loop is 1 / (2 ) m and the wire radius 0.5 mm. Using
the thin wire approximation, the causal impedance is calculated
on the basis of the complex power, i.e.

(2 Rl ) E (e ) I *
Z (e ) (85)
I I*
Fig. 3. Loop with constant current along its circumference. Top:
with E the component of the causal electric field tangential to normalized energies of Table I and normalized electric +
the loop. The same normalization is used as in [26] so that the magnetic energy (normalization is performed by multiplying
results can be easily compared. In the numerical calculations with (16 2 0 ) / (kI02 ) , which makes the energies dimensionless),
the number of segments on the loop was 720 and the number of bottom: Q factors depicted as in [26].
zero-radiative modes was 50. Including the modal current
proportional with (t1 t ) delivers Wrec , excluding this modal D. Dipole with imposed current
current delivers Wdyn . It is seen that a recoverable energy is The results for a dipole with an imposed sinusoidal current
distribution are given in Fig. 4. The length L of the dipole is 1
obtained which is almost identical to the magnetic energy m and the wire radius 0.5 mm. Using the thin wire
linked to the currents, with much smaller values compared to approximation, the causal impedance is calculated on the basis
the values obtained in [26]. However, the negative energies of the complex power, i.e.
have disappeared totally. A small oscillation remains in the
recoverable energy. L /2
Ez (e , z ) I * ( z )dz
Z (e )
L /2 (86)
I (0) I * (0)

with E z the component of the causal electric field in the

direction of the dipole. The same normalization is used as in
[26]. In the numerical calculations the number of segments on
the dipole was chosen very high, i.e. 5000, and the singular
1/ r21 coupling was rigorously taken into account by using an
extraction technique for r21 0 . This leads to an extremely
accurate calculation of the capacitor C for this capacitive
structure. The number of zero-radiative modes was chosen as 5.
This is sufficient since the exponential decay of these modes in
time domain increases rapidly with increasing mode number.
No negative energies occur and there is no oscillation
whatsoever in any of the energies. Moreover, the recoverable
energy is almost identical to the reactive energy. The physical
reason for this is that all non-static zero-radiative modes show
a very large negative real part for sm , so that the current profile
very much resembles the profile with an almost instantaneous
current drop to zero at t for Wdyn or with a constant future
current for Wrec . This line of reasoning would explain why the
energy expressions introduced in [15] work quite well for non-
cavity dipole-like structures, but not always for loop-like
structures.

0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
13

APPENDIX I
Consider the integral

j
e st e s t
*
1
I
4 j j s s1
F ( s) * * ds
s s2
(87)

with F ( s) an even function in the Laplace parameter s, and s1

and s2 with a negative real part. Choosing the integration path
on the imaginary axis, the two exponentials cancel each other,
so that the integral simplifies to

j
1 1 1
I
4 j j s s1
F ( s) * * ds
s s2
(88)

Applying Plancherels theorem and the convolution property,

this integral can be written as

1 s2*t

I F (t t ')e u (t ')dt ' e u (t )dt
s1 t '

2
1 s2*t

F (t t ')e dt ' e dt
s1 t '

2 0 0 (89)

1

F (t t ')e
s1 ( t ' t )
dt ' e( s1 s2 )t dt
*

2 0 0
1 ( s1 s2* )t

F (t '')e dt '' e
s1 t ''
dt
2 0 t

with F (t ) the inverse Laplace transform of F ( s) . Since F ( s)

Fig. 4. Dipole with a sinusoidal current. Top: normalized is even in s, F (t ) is even in t. Integration by parts yields
energies of Table I and normalized electric + magnetic energy
(normalization is performed by multiplying with
(16 2 0 ) / (kI 02 ) , which makes the energies dimensionless), 1 1
*
I F (t '')e s1 t ''dt '' e( s1 s2 )t
*

bottom: Q factors. 2 ( s1 s2 ) t 0 (90)

1 1
*
F (t ) e s1 t e( s1 s2 )t dt
*
CONCLUSIONS
2 ( s1 s2 ) 0
In this paper, it was clearly shown that it is possible to define a
concept for recoverable energy in frequency domain that
satisfies all physical requirements: based on a physical line of Since s1 and s2 have a negative real part the value at infinity is
reasoning, solely depending on currents, independent of zero, yielding
coordinate system, and yielding purely positive values. On top,
the new formulation opens possibilities for topology and 1 1

*
I
*
s1 t
F ( t ) e dt F (t ) e s2t dt
current distribution optimization. The core paradigm shift with 2 ( s1 s2 ) 0 0 (91)
respect to the traditional definitions already published for stored
1 1
energy in open literature is the necessity to build a line of [ F ( s1 ) F ( s2* )]
reasoning fully in time domain. Two techniques have been 2 ( s1 s2* )
described to calculate the recoverable energy: one based on
modes, and a Wiener-Hopf based technique. Whereas both The integrals in (91) involve the Laplace transform of the causal
techniques give the same result in case all quasi-static modes part of F (t ) , evaluated in s1 and s2* .
are taken into account, the modal technique gives the possibility
It is important to note that a similar line of reasoning can be
to study the effect of certain modes. The new definition and the
properties following therefrom were clearly demonstrated for followed in cases that s1 and/or s2 have a positive real part.
five simple basic radiator topologies. The result is then slightly different.

0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
APPENDIX II
A complex function f can be decomposed into a sum of two
functions f and f , analytic in the left and the right half
complex plane, respectively. The related operators are
1 f ( s ')
2 j
f ( s) ds '
C
s ' s
1 f ( s ')
2 j C
f ( s) ds '
s ' s
with the contours C and C parallel to the imaginary axis and
passing to the right and left of the point s ' s , respectively.
Similarly, the function f can be factorized into a product of two
functions analytic in the left and the right half complex plane,
by first taking the logarithm, and then performing the sum
decomposition.
ACKNOWLEDGEMENT
The author would like to thank dr. Xuezhi Zheng and dr. Martin
Stumpf for the inspiring discussions on operator theory and
Laplace transforms, Prof. Fan Yang from Tsinghua University,
Beijing, China, for receiving me within the framework of a
sabbatical, and the Fund for Scientific Research Flanders for the
funding under grant K8.020.14N.
REFERENCES
[1] R. E. Collin and S. Rothschild, Evaluation of antenna Q, IEEE
Trans. Antennas & Propagat., Vol. AP-12, pp. 23-27, Jan. 1964.
[2] L. J. Chu, Physical limitations on omni-directional antennas, J.
Appl. Phys., Vol. 19, pp. 1163-1175, Dec. 1948.
[3] J. S. McLean, A re-examination of the fundamental limits on the
radiation Q of electrically small antennas, IEEE Trans. Antennas &
Propagat., Vol. 44, pp. 672-676, May 1996.
[4] S. R. Best and D. Yaghjian, The lower bounds on Q for lossy
electric and magnetic dipole antennas, IEEE Antennas Wireless
Propag. Letters, Vol. 3, pp. 314-316, Dec. 2004.
[5] D. Yaghjian and S. R. Best, Impedance, bandwidth and Q of
antennas, IEEE Trans. Antennas & Propagat., Vol. 53, pp. 1298-1324,
Apr. 2005.
[6] H. L. Thal, New radiation Q limits for spherical wire antennas,
IEEE Trans. Antennas & Propagat., Vol. 54, No. 10, pp. 2757-2763,
Oct. 2006.
[7] M. Gustafsson, C. Sohl, and G. Kristensson, Illustrations of new
physical bounds on linearly polarized antennas, IEEE Trans. Antennas
& Propagat., Vol. 57, No. 5, pp. 1319-1327, May 2009.
[8] R. C. Hansen and R. E. Collin, A new Chu formula for Q, IEEE
Antennas and Propagat. Magazine, Vol. 51, No. 5, pp. 38-41, Oct.
2009.
[9] W. Geyi, A method for the evaluation of small antenna Q, IEEE
Trans. Antennas & Propagat., Vol. 51, pp. 2124-2129, Aug. 2003.
[10] A. D. Yaghjian and H. R. Stuart, Lower Bounds on the Q of
Electrically Small Dipole Antennas, IEEE Trans. Antennas &
Propagat., Vol. 58, No. 10, pp. 3114-3121, Oct. 2010.
[11] M. Capek, P. Hazdra, and J. Eichler, A Method for the
Evaluation of Radiation Q Based on Modal Approach, IEEE Trans.
Antennas Propagat., Vol. 60, No.10, pp. 4556-4567, Oct. 2012.
[12] H. L. Thal, Antenna Q calculation using radiated modes and
reflection coefficient, IET Microwaves Antennas & Propagat., Vol. 9,
No. 4, pp. 328-335, March 2015.
0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
14
[13] M. Gustafsson and L. Jonsson, Stored Electromagnetic Energy
and Q, Progress in Electromagnetics Research PIER, Vol. 150, pp.
13-27, 2015.
[14] L. Jelinek, M. Capek, P. Hazdra, J. Eichler, An analytical
evaluation of the quality factor Q(Z) for dominant spherical modes,
IET Microwaves, Antennas & Propagation, Vol. 9, No. 10, pp. 1096-
1103, July 2015
[15] G. A. E. Vandenbosch, Reactive energies, impedance, and Q
(92) factor of radiating structures, IEEE Trans. Antennas Propagat., Vol.
58, no. 4, pp. 1112-1127, April 2010.
[16] G. A. E. Vandenbosch, Simple procedure to derive lower
bounds for radiation Q of electrically small devices of arbitrary
topology, IEEE Trans. Antennas Propagat., Vol. 59, No. 6, pp. 2217-
2225 , June 2011.
[17] G. A. E. Vandenbosch, Explicit relation between volume and
lower bound for Q for small dipole topologies, IEEE Trans. Antennas
Propagat., Vol. 60, No. 2, pp. 1147-1152, Feb. 2012.
[18] M. Gustafsson, M. Cismasu, and B. L. G. Jonsson, Physical
bounds and optimal currents on antennas, IEEE Trans. Antennas
Propagat., Vol. 60, No. 6, pp. 2672-2681, June 2012.
[19] P. Hazdra, M. Capek, and J. Eichler, Radiation Q-factors of
thin-wire dipole arrangements, IEEE Antennas Wireless Propag.
Lett., vol. 10, pp. 556560, 2011.
[20] M. Gustafsson, and S. Nordebo, Optimal antenna currents for
Q, superdirectivity, and radiation patterns using convex optimization,
IEEE Trans. Antennas Propagat., Vol. 61, No. 3, pp. 1109-1118,
March 2013.
[21] M. Capek, L. Jelinek, P. Hazdra, and J. Eichler, The Measurable
Q Factor and Observable Energies of Radiating Structures, IEEE
Trans. Antennas Propagat., Vol. 62, No.1, pp. 311-318, Jan. 2014.
[22] M. Gustafsson and B. L. G. Jonsson, Antenna Q and Stored
Energy Expressed in the Fields, Currents, and Input Impedance, IEEE
Trans. Antennas Propagat., Vol. 63, No.1, pp. 240-249, Jan. 2015.
[23] Shlivinski and E. Heyman, Time-domain near-field analysis of
short-pulse antennas Part I: Spherical wave (multipole) expansion,
IEEE Trans. Antennas & Propagat., Vol. 47, pp. 271-279, Feb. 1999.
[24] Shlivinski and E. Heyman, Time-domain near-field analysis of
short-pulse antennas Part II: Reactive energy and the antenna Q,
IEEE Trans. Antennas & Propagat., Vol. 47, pp. 280-286, Feb. 1999.
[25] G. A. E. Vandenbosch, Radiators in time domain, part I:
electric, magnetic, and radiated energies, IEEE Trans. Antennas
Propagat., Vol. 61, No. 8, pp. 3995-4003, Aug. 2013.
[26] G. A. E. Vandenbosch, Radiators in time domain, part II: finite
pulses, sinusoidal regime, and Q factor, IEEE Trans. Antennas
Propagat., Vol. 61, No. 8, pp. 4004-4012, Aug. 2013.
[27] M. Capek, L. Jelinek., and G. A. E. Vandenbosch, Stored
electromagnetic Energy and Quality Factor or Radiating Structures,
Proceedings of the Royal Society A, Vol. 472, No. 2188, April 2016.
[28] C. E. Baum, E. J. Rothwell, K.-M. Chen, and D. Nyquist, The
Singularity Expansion Method and Its Application to Target
Identification, Proc. IEEE, Vol. 79, pp. 1481-1492, 1991.
[29] L. S. Gradshteyn and L. M. Ryzhik, Table of integrals, series,
and products, Sixth edition, Academic Press, San Diego, USA, 2000.
[30] G. Kaiser, The Reactive Energy of Transient EM Fields,
arXiv:.1201.6575v2.
[31] G. Kaiser, Electromagnetic Inertia, Reactive Energy and Energy
Flow Velocity, J. Phys. A: Math. Theor., Vol. 44, pp. 1-15, 2011.
[32] V. G. Polevoi, Maximum Energy Extractable from an
Electromagnetic Field, Radio Physics and Quantum Electronics, Vol.
33, No. 7, pp. 603-609, 1990.
[33] R. H. Direen, Fundamental Limitations on the Terminal
Behavior of Antennas and Nonuniform Transmission Lines, Ph.D.
thesis, University of Colorado, 2010.
[34] V. G. Daniele and R. S. Zich, The Wiener-Hopf Method in
Electromagnetics, Mario Boella Series on Electromagnetism in
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TAP.2017.2702659, IEEE
Transactions on Antennas and Propagation
15

Information and Communication, Raleigh, NC: SciTech Publishing;

2014.

Guy A. E. VANDENBOSCH received the M.S. and

Ph.D. degrees in Electrical Engineering from the
Katholieke Universiteit Leuven, Leuven, Belgium, in
1985 and 1991, respectively. Since 1993, he has been a
Lecturer, and since 2005, a Full Professor at the same
university. Guy Vandenbosch has taught or teaches courses on
Electromagnetic Waves, Antennas, Electromagnetic
Compatibility, Fundamentals of Communication and Information
Theory, Electrical Engineering, Electronics, and Electrical Energy,
and Digital Steer- and Measuring Techniques in Physics. His
research interests are in the area of electromagnetic theory,
computational electromagnetics, planar antennas and circuits, nano-
electromagnetics, EM radiation, EMC, and bio-electromagnetics. His
work has been published in ca. 260 papers in international journals and
has led to ca. 360 presentations at international conferences. From
2001 to 2007, he was the President of SITEL, the Belgian Society of
Engineers in Telecommunication and Electronics. From 2008 - 2014,
he was a member of the board of FITCE Belgium, the Belgian branch
of the Federation of Telecommunications Engineers of the European
Union. In the period 1999-2004, he was vice-chairman, in the period
2005-2009 secretary, and in the period 2010-2016 chairman of the
IEEE Benelux Chapter on Antennas and Propagation. In the period
2002-2004 he was secretary of the IEEE Benelux Chapter on EMC. In
the period 2012-2014 he was secretary of the Belgian National
Committee for Radio-electricity (URSI), where he is also in charge of
commission E. From September to December 2014, he was a visiting
professor at Tsinghua University, Beijing, China.

0018-926X (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.