Вы находитесь на странице: 1из 15


3, MARCH 2012


Modeling High-Order Ferromagnetic Hysteretic Minor Loops and Spirals

Using a Generalized Positive-Feedback Theory
Robert G. Harrison, Life Senior Member, IEEE
Department of Electronics, Carleton University, Ottawa, ON K1S 5B6, Canada
A generalization of the previously-described positive-feedback theory of first- and second-order reversal curves permits modeling of
minor loops of arbitrarily high order . The validity of the theory is tested by using it to model various types of high-order phenomena
3), including symmetrical and asymmetrical spirals as well as closed minor loops, in a variety of different hard and soft ferromagnetic materials that exhibit the return-point memory property.
Index TermsAnhysteresis, demagnetizing spiral, energy summation, high-order loop, high-order return curve, hysteresis, magnetization, minor loop, recoil loop, return-point memory.

A. Theories and Models

HE successful modeling (and prediction) of high-order
minor loops in ferromagnetic materials has long been a
stringent test of the capability of a hysteresis model. Hitherto,
there seems to have been no single physical theory that can explain the detailed behavior of low- and high-order minor loop
phenomena in both hard and soft materials.
This paper introduces a generalized positive-feedback
(G-PFB) theory that can successfully model a variety of major
loops, symmetric and asymmetric minor loops, as well as
demagnetization spirals, in different types of soft and hard
ferromagnetic materials. The term generalized is used in the
sense that it can model minor loop segments of arbitrarily high
order, in many materials with parameters conforming to the
assumptions made in deriving the theory. The work described
here is an extended version of a previously-described physical
theory that was able to model first-order reversal curves [1] and
recoil loops [2].
In practical applications, high-order minor loops and their
associated hysteresis losses play an important part in characterizing the behavior of inductors, transformers, motors, generators, and numerous electromechanical devices. Examples include the calculation of iron-rotor hysteresis losses due to minor
- loops [3], and complex minor loops in motor cores due to
-fields with high-order time harmonics [4]. In addition, in the
area of material characterization, it has been suggested that by
measuring a set of minor loops with various field amplitudes,
it would be possible to obtain much information on lattice defects in ferromagnetic materials [5]. Conversely, by comparing
such loops with their theoretical counterparts, the G-PFB theory
could be related to the presence of lattice defects.
Models based on the G-PFB theory are expected to be ideally
suited for the formulation of finite-element codes because they
intrinsically provide as a function of , as required by FEM
procedures based on the magnetic vector potential .

Manuscript received June 22, 2011; revised September 14, 2011; accepted
September 27, 2011. Date of publication October 06, 2011; date of current version March 02, 2012. Corresponding author: R. G. Harrison (e-mail: rgh@doe.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TMAG.2011.2170846

Previous work on high-order ferromagnetic phenomena has

employed numerous different approaches, of which the two
dominant ones have been the Preisach model (PM) and its
descendents, and the Jiles-Atherton (JA) model and its variants.
Below, we survey these and some other models that have been
used for minor loops.
1) Preisach Models: The classical, generalized (nonlinear), average, and feedback1 versions of the PM have all
been used to simulate demagnetizing spirals of orders up to 6 in
various tape-recording materials [6][9]. A moving version of
the PM, together with a full set of measured symmetric closed
minor loops, was used to model minor loops in a very hard
SmFeCo material [10]. A PM incorporating a description of domain-wall motion could produce horizontally-congruent minor
loops and nested closed minor loops up to order 10 in very soft
materials [11]. Kdrs bilinear product model version of the
PM, which does not suffer from the non-physical congruency
property of the classical PM, has been used to model recoil
loops and demagnetization spirals up to order 14 in Alnico and
certain steels [12]. A measured major loop, together with a
simplified PM, was used to model low-order minor loop losses
in FeSi [13]. Preisach models have several problems:
a) They are not closely related to the physics of magnetic
materials. Instead, they rely on empirical techniques
involving the identification of the parameters of analytic functions such the Lorentzian, Log-Gauss, or
Gauss-Gauss distributions that are used to represent the
Preisach function. In the case of generalized versions
of the PM in its scalar or vector implementations, these
identification procedure can become computer intensive.
b) Because of its mathematical structure, the classical PM
predicts vertical congruency in its strict sense, that is, all
minor loops resulting from input field variations between
the same two extrema are congruent. This congruency is
not generally obeyed in reality [14], [15]. The nonlinear
[16] and average [17] versions of the PM both show the
property of equal vertical chords, a relaxed version of
congruency. In the average version this property is history-dependent.
1The feedback referred to here is a mathematical PM technique entirely
unrelated to the present G-PFB theory.

0018-9464/$26.00 2011 IEEE



c) In PM modeling, it is normally assumed that

be the sum of irreversible and reversible components
. The classical PM (CPM) provides only the
component, but in reality the magnetization has both irreversible and reversible properties. To include an
component, various different (empirical) modifications to
the CPM have been proposed. These include the moving
and product versions of the PM [18], [19], as well as an
approach that considers a distribution with an additional
Dirac delta-function singularity on the line
the Preisach plane,2 that produces a reversible term in
the PM [10], [17]. A problem with the assumption that
is that experiment shows that
is itself a function of
[18]. It has been concluded
that the concept of irreversible magnetization implicit
should be regarded as a convenient label and
should not be taken as a measure of thermodynamic irreversibility [20].
d) FEM electromagnetic-field calculations employing the
vector potential require the applied field as a function of the magnetization , whereas the PM provides
. Until recently, the computational cost of iteratively calculating
was relatively high [15],
[21], but Davino et al. have reported an algorithm that
overcomes this problem [22].
e) The classical PM does not include reversible
2) Jiles-Atherton Models: The original JA model [23] could
not model physically-reasonable minor loops. A subsequent
modification [24] provided a way to do so, but required the
magnetizations at both extremities of a minor loop to be known
a priori, and hence could not accommodate arbitrary applied
field variations. A proposed further modification requiring only
the initial point of the loop could deal with arbitrary excitations
and initial conditions; however no comparison with measured
data was provided [25]. An optimization procedure was used
to find the parameters of both the original JA model (and its
inverse), and thereby obtain agreement between modeled and
measured minor loops [26]; improved recoil loop representations were achieved through further empirical modifications
[27], [28]. An empirical variation of the classical JA model to
ensure minor-loop closure was also proposed [29].
Problems with JA models include the following:
a) They do not naturally produce closed minor loops; a fix is
needed to do that.
b) As in the Preisach models, see above, all JA models
involve a decomposition of the magnetization into reand irreversible
components, but the
physical reality of this is questionable [2], [20].
3) Other Models: Other prior approaches that have been investigated for minor loops include the following.
An empirical anisotropic vector hysteresis model,
involving a population of pseudoparticles and the
Prandtl-Ishlinskii friction-like representation of hysteresis [30].
An empirical-phenomenological algebraic limiting-loop
proximity model [31].
A quasi-physical energetic model [32].

the symbols and have their Preisach-model meanings.

An empirical model involving an equivalent elliptical loop that has the same area as the physical major loop (or
smaller ellipses in the case of minor loops), from which the
estimated irreversible hysteretic loss in soft core materials could be calculated using a transient FEM analysis. To
make this work, the wiping-out rule was imposed [33].
Stop and play models that employ input-dependent
shape functions. The main advantage of these models is
that they provide an inverse in closed analytic form. Although they still need function identification procedures,
they are sometimes claimed to be as efficient as, or more
efficient than, the PM [4], [34]. If a parallel architecture
(for example FPGA) is used, then the stop and play models
can indeed be more efficient than the PM. However for a
serial architecture, as in a PC, then the PM is faster, and
the JA model faster still (but less acccurate).
A copy-and-translate technique in a static hysteresis
model [35].
A procedure involving finite-element modelling plus
Madelungs rules to estimate minor hysteresis loops in
a representation defined by a measured initial-magnetization curve, a set of demagnetization curves, data for
straight-line recoil loops, and a measured symmetric
major loop [36].
An empirical piecewise-linear model defined as superposition of hysteresis operators in a resistor-inductor circuit
The difficulty with most of these empirical and quasi-physical
modeling approaches is that they need a fitting procedure or
special mathematical fixes to make them work.
Here we introduce an entirely different theory based on
quantum/classical physics that can be used to model high-order
minor-loop phenomena with good accuracy. No function identification procedures are needed. This paper generalizes the
positive-feedback hysteresis model described previously [1],
[2] in such a way that th-order minor loops and curve segments, closed minor loops of arbitrary order, and asymmetric
demagnetizing spirals can be calculated.
According to the present scalar theory, the basis of which
was explained in detail in [1], [2] and [38], ferromagnetic hysteresis always has its origin in a positive-feedback process of
QM origin occuring at the micromagnetic scale. This leads to the
logically necessary concept of quasi-macroscopic QM particles,
each consisting of a chain of magnetons, conveniently referred
to as a supermagneton, extending across the width of a domain
[1]. A parallel ensemble of many such spicular quasi-particles
constitutes a 3-D array occupying the volume of a domain. In
this theory the average length of a supermagneton is magnetons, where the distance between individual magnetons depends
on the crystal lattice structure. As shown in [38] this theory results in the following positive-feedback (PFB) equation for the
magnetization :

is the saturation magnetization, is the vacuum permeability,
is the Bohr magneton, is the Boltz-



mann constant, is the Kelvin temperature,

is the QM
component of the total applied field , and is the Weiss exchange-field coefficient. In normalized form (1) is simply

is the
. From this it is evident that
Curie temperature, and
the magnetization is a function of itself, thereby giving rise to
a PFB process. The result is an S-shaped
function, complete with unidirectional jumps that account for
hysteretic irreversibility. In this work, (2) is inverted to provide
a QM expression that is suited for field summation:
The hysteresis operator defined by (2), or equivalently (3),
shows a local memory hysteresis. That is, the output depends
only on the input and the current state of magnetization. As
shown in [1], [2], and [38], the overall magnetic state of the
system can be found by summing the relevant contributions to
(this is equivalent to an energy summation).
the total field
associated with the
Thus the anhysteretic field component
mesoscopic domain structure must be included in the sum. In an
isotropic medium in which magnetization changes are mediated
by rotation of domain magnetic moments
, the anhysteresis
is given by a Langevin function
. As demonstrated in [1],
[2], the corresponding anhysteretic field contribution is given

is the inverse Langevin function and

is the ratio between magneton and domain
moments. We remark that (4) could be written more generally
is a generalized anhysteresis function (or lookup
table) describing non-Langevin behavior. The field-sum should
also include a macroscopic global demagnetizing field term
is the demagnetization factor. Summing the above field terms we get the field sum (equivalently,
energy sum)
where O.T. stands for other terms (anisotropy-energy field, etc.)
as required. The minimum requirement for this summation is the
presence of the first two terms: if
is omitted then hysteretic
loops (major or minor) have perfectly vertical sides, seldom
seen in reality; without
there will be no PFB and consequently no hysteresis!
In the case of minor loops, a non-local memory hysteresis also
comes into play, where the output depends on the complete magnetization history. Explicit equations for the major loop, FORC,
and SORC according to the PFB theory, based on (1)(7), and
incorporating return-point memory (RPM) effects, have already

Fig. 1. Two theoretical demagnetization spirals calculated using the G-PFB

algorithm. Reversible segments are shown in thick line, irreversible segments in
thin line. Odd-order and even-order segments are shown in dark and light gray
respectively (blue and green online). Normalized parameters are given in the

been given in [2]. A discussion of the micromagnetic origin of

RPM was given in the same reference. In the next section we
generalize those equations in order to describe th-order reversal curves, under the condition that the physical requirements
for full RPM are satisfied. The th-order reversal curves will be
referred to as NORCs (symbolized as ). For each order of reversal curve, two different types are possible, one type having
both reversible and irreversible segments, the other having only
a reversible component.
The resulting generalized PFB theory is used here in a
fast-acting algorithm for computing arbitrary high-order minor
loops and spirals. To illustrate the versatility of this algorithm,
Fig. 1 depicts theoretical asymmetric demagnetization spirals
that commence from two different points on the descending
branch of a major loop.
The quasi-static scalar G-PFB model described above is limited to magnetic materials and situations that conform to the initial assumptions made in its derivation. Thus the material should
be homogeneous and isotropic (or else uniaxially anisotropic
with the -field parallel or perpendicular to an easy axis [1]),
and free from external or internal magnetomechanical stresses.
However, the model is often found to give useful results even
when some of these requirements are not fully met. According
to (1), inter-particle interactions are included as a mean-field approximation through the exchange-field parameter . In certain
circumstances, inclusion of higher-order interactions might improve the accuracy, but this has been found unnecessary in the
materials studied here.
A. Generalized Algorithm for Minor Loop Segments
It is important to notice that in this work the subscripts rev
and irr that refer to reversible and irreversible -fields have
an entirely different meaning from the usage in the Preisach



Fig. 2. Theoretical asymmetric demagnetization spirals (continuous curves). (a) A low-order spiral in which NORC 0 has both lower-reversible (L ) and
ascender-irreversible (A ) segments and is therefore Type-LA. The major loop 0 is indicated in dashed line. (b) Part of a higher-order spiral in which NORC 0
has only a lower-reversible (L ) segment and is therefore Type-L. In (a) and (b), n is an odd integer. In this figure, P are reversal points, X are reversal fields,
Y are reversal magnetizations, and Y are bifurcation magnetizations [2]. Reversible segments are shown in bold line, irreversible segments as light arrows,
reversal points as open circles, and bifurcation points as diamonds. The connectors, due to the RPM effect, are shown as dotted lines with arrows.

and Jiles-Atherton models, which are formulated in terms of

reversible and irreversible magnetizations , see [2].
1) Reversible-Field Summation: We write the th-order field
for a reversible
(upper) or
(lower) segment
of a NORC, see Fig. 2, in the form of the normalized reversiblefield summation

3) The Four Key Arbitrary-Order Field Equations: The four

key field components occurring in (8) and (9) are:
The term
for the reversible part of the QM S-curve,
as given by the complete expression

is the reversible part of the hysteretic QM field
. The term
is a generalization of the the first- and
second-order center-point magnetizations of the symmetric but
-offset S-shaped curves that are caused by the QM-PFB
process, while
is the center-point magnetization of the
-offset th-order minor anhysteresis curve, as explained in [2].
The remaining terms in (8) are
the coercive field,
the domain-size parameter, and
the local demagnetizing
2) Irreversible-Field Summation: There is an analogous
th-order field sum
for an irreversible
(descending) segment of a NORC, written here as an
irreversible-field summation

The term
for the irreversible part of the QM
S-curve, which is

The term

for the anhysteresis:

The term

for the global demagnetizing field:


is the irreversible part of the hysteretic QM field
, and the other terms are as in (8). Generalizing the field
equations previously obtained for FORCs (1ORCs) and SORCs
(2ORCs) [2], we deduce the key th-order equations shown in
the following sub-section.

In (11) and (12) the integer is

when is odd and
is even.
In the following, the separation of minor branches into odd
and even types is only for descriptive convenience, and is due
to the choice of a FORC starting on the descender of the major
loop. If a FORC starts from the major-loop ascender then the
odd branches become even and the even ones become odd.


The algorithm will be explained by referring to Fig. 2(a),

which shows a low-order demagnetization spiral, and Fig. 2(b)
which depicts part of a higher-order demagnetization spiral.3 We
shall also refer to portions of the flow chart of Fig. 3. For simplicity, the sum field
will be written as
in what
Fig. 2(a) shows a fourth-order demagnetization spiral that
starts from point 1 on the descender branch of a major loop. In
. From
order to explain the algorithm this point is labeled
) there is a Type-LA NORC
point 1 to point 2 (labeled
(blue online) consisting of reversible lower segment
followed by irreversible ascender
. Then from point 2 to
) there is a Type-UD NORC
point 3 (labeled
online) consisting of reversible upper segment
(green) that aims at memory
by irreversible descender
point 1 (dotted line) because of the RPM effect. However, fulfillment of this aim is thwarted by the field reversal at point
3. The penultimate part of the spiral is a Type-LA NORC (labeled ) from point 3 to point 4 (blue), consisting of reversible
followed by irreversible ascender
lower segment
aims at memory point 2 (dotted line). Contrary to intuition,
the chains of ascending and descending connectors, shown in
dotted line, do not form smooth curves. Instead, each connector
aims along its own trajectory towards its own memory point.
(green) conThe final part of the spiral, a Type-UD NORC
followed by
, goes from point 4 directly
sisting of
to memory point 3, again because of RPM, uninterruped by any
further field reversal.
Fig. 2(b) depicts a seventh-order demagnetization spiral that
illustrates a Type-L fully-reversible NORC
going from point
). Here
7 (labeled ) back to memory point 6 (labeled
has no
segment because its bifurcation magnetization
greater than its reversal magnetization
. In spite of appearis actually Type-UD because,
ances, the preceding NORC
, it would have extended to
in the absence of reversal field
). The resulting closed loop between
point 5 (labeled
is a Rayleigh loop [39], [40].
4) Equations for an Odd-Order NORC: In the following discussion we refer to Fig. 2, and also to Fig. 3 which shows a simplified flow chart for the algorithm. When is odd, there are two
possible types of th-order return curve (NORC) , depending
on whether the bifurcation magnetization4
is greater or
less than the
th-order reversal magnetization
. If
, as it is in Fig. 2(a) and in frame 6 of Fig. 3, then
has both
(lower) and
(ascender) segments and is des, as in
ignated Type-LA (two components). But if
Fig. 1(b) and frame 5 of Fig. 3, then
has only an
segment and is called Type-L (one component). It is not known a
priori whether
is Type-L or Type-LA. This problem can be
resolved by initially assuming that
is Type-L, and then calculating the th-order bifurcation magnetization
. If it is
found that
, then the assumption was correct. Otherwise
must be Type-LA, in which case
has to be be re-calculated.

curves in Figs. 1(a) and (b) were all derived using the present theory.

bifurcation magnetization Y
is the value of y at which the QM
. It is the magnetization at which the NORC
S-shaped curve has @x =@y
segment changes from reversible to irreversible [2].



a) Finding the Reversal Magnetization

( Odd,
The investigation requires evaluation of the th-order reversal
at the th-order field-reversal point
, see
Fig. 2(a), (b). For a Type-L NORC, for which
occurs where th-order reversal field
, see for example Fig. 2(b)
the reversible upper segment
and frame 2 of Fig. 3, and is given by the zero of the bounded

denotes an error (depending on ), and the notameans that variable is to be restricted to
to . However, for a Type-LA NORC,
the search interval
, magnetization
occurs where
for which
cuts the irreversible descender
, see Fig. 2(a)
and frame 1 of Fig. 3, and is therefore given by the zero of the

We next show how to calculate the NORC in each case.
): A Type-L
b) Calculating a Type-L NORC ( Odd,
segment , which is necessarily reversible and odd-order, see
Fig. 2(b) and frame 5 of Fig. 3, is given by
The unknown center-point magnetizations
can be determined by solving the following pair
of nonlinear simultaneous equations:
using a 2-D constrained optimization routine [2]. Suitable initial
conditions are
constraints are
. Having determined
, the bifurcation magnetization at the boundary between reversible and irreversible segments of the NORC is found to be
If the initial assumption that
turns out to be true,
is indeed Type-L (having only a reversible component
) and is given by a bounded form of (16):
However, if it is found that that
, then
must be
Type-LA, in which case revised magnetizations
need to be calculated, as discussed next.
c) Calculating a Type-LA NORC ( Odd,
): If it
is determined that
, then
has both
components and is Type-LA, see Fig. 2(a) and frame 6 of Fig. 3.
In this case revised magnetizations
have to be calculated by solving the following pair of simultaneous nonlinear



Fig. 3. Simplified flow chart for calculating demagnetization spirals and minor loops of arbitrary order n.


where (21) is the same as (17) but (22) is different from (18).
Again, a 2-D constrained optimization routine can be used, with
the same initial conditions and constraints as used for the Type-L
calculation using (17) and (18). Magnetization
is found by
recalculating (19), while the lower (reversible) segment



Fig. 4. Summary of the G-PFB algorithm depicted in the flow chart of Fig. 3.

is given by a recalculation of (20) using a different upper


However, for a Type-UD NORC, for which

occurs where reversal field
cuts the irre, see frame 3 of Fig. 3, and is therefore
versible ascender
given by the zero of the bounded function

The (irreversible) ascender component
NORC can be plotted (later, because

of the Type-LA
is not yet known)

5) Equations for an Even-Order NORC: When is even,
the calculations follow a similar pattern as when is odd, but
with certain differences. There are again two possible types of
NORC, depending on whether the bifurcation magnetization
is less or greater than the
th order reversal magnetization
. If
, as in frame 8 of Fig. 3, then
(upper) and
(descender) segments and is designated
Type-UD (two components). But if
, as in frame 7
of Fig. 3, then
has only an
component, and is Type-U.
segments appear in Fig. 2. Again, we
Examples of
is Type-UD consisting of both
do not know initially whether
segments, or Type-U having only an
We solve the problem by assuming that
is Type-U with only
segment, and then calculating
. If
, then
the assumption was correct. Otherwise
, and
has both
components, and is Type-UD, so that
must be recalculated.
a) Finding Reversal Magnetization
( Even,
, magnetizaFor a Type-U NORC, for which
occurs where field
cuts the reversible lower segment
, see frame 4 of Fig. 3, and is given by the zero of the
bounded function

We now show how to calculate the NORC in each case.
b) Calculating a Type-U NORC ( Even,
): A
Type-U segment, which is reversible and even-order, is given
by (16). The unknown magnetizations
are found
by re-solving nonlinear simultaneous (17) and (18), with initial
. The
constraints remain
. The NORC bifurcation field
and magnetization
are given by a recalculation of (19) and (20), respectively.
, see Frame 7 of
If it turns out to be true that
Fig. 3, then
has only an
component given by an expression similar to (23) but with different limits:
But if it is found that
, then
must have both
components, and is Type-UD. In this case revised
must be determined, as explained
in the next section.
c) Calculating a Type-UD NORC ( Even,
): If
, see Frame 8 of Fig. 3, then
it is found that
must have both
components, and is Type-UD. In this
case revised magnetizations
have to be determined
by re-solving simultaneous nonlinear equations (17) and (18),
using the same initial conditions and constraints as for Type-U.
Then the
segment is given by a recalculation of (27) with
different limits:


where recalculations of (20) and (21) apply.




Fig. 5. Measured third-order asymmetric demagnetizing spiral (dotted curve)

in Isomax [6] compared with theoretical results (continuous curves). Reversible
curve segments are shown in heavy line, irreversible segments in light line.

The irreversible descender segment

of the Type-UD
NORC can be plotted, once
is known, using

d) Including a Paramagnetic Magnetization Component:
A paramagnetic component of the magnetization can be included by adding to the calculated value of
a term
or in normalized form, by adding to the calculated value of a
, where
is the paramagnetic susceptibility.
6) Summary of the Algorithm: The above sequence of calculations is summarized in Fig. 4, which refers to the flow chart of
Fig. 3.
For an initial test of the generalized positive-feedback
(G-PFB) theory, Isomax was chosen. This is a recording
medium consisting of nonoriented Co-doped -Fe O particles of low acicularity [6], [41]. It has properties in close
conformity with the assumptions of this version of the theory,
that the material is isotropic and that magnetization changes
occur primarily through domain rotation, and consequently
it has a Langevin-type anhysteresis [1]. Then we investigate
how well the theory works for examples of ferromagnetic
materials ranging from very soft (
A/m) to very hard

Fig. 6. A measured eighth-order demagnetization spiral in -Fe O (dotted

line) compared with theory (gray, blue, green online). The 1ORC is almost coincident with the major loop. Reversible segments are in bold line, irreversible
segments in light line. Measured data are courtesy of I. Mayergoyz [7].

would be expected for an isotropic material with magnetization

via domain rotation. Relevant data are listed in Table I. The fundamental exchange-field and domain-width parameters and
occurring in (1) can be calculated from the measured data using
the following expressions [2]:

Quantities derived from the measured parameters in Table I

are the exchange field coefficient
and the domain
B. Eighth-Order Demagnetization Spiral: -Fe O

A. Third-Order Demagnetization Spiral

Fig. 5 compares data for a measured asymmetric third-order
demagnetizing spiral in Isomax with a modeled spiral according
to the present G-PFB theory. Reasonable agreement is observed
for the major loop
and for the
, and
segments, as

Fig. 6 compares theoretical and measured asymmetric eighthorder demagnetization spirals for a Co-coated -Fe O magnetic tape [7]. Even though the agreement for the outer loops is
imperfect, accuracy improves as increases, indicating the robustness of the algorithm. Relevant data are listed in Table II.



= 111

Fig. 7. (a) A theoretical major loop with a quasi-symmetric demagnetization spiral of order n
, computed for Isomax using the G-PFB theory (odd/evenorder NORCs are blue/green online). Open circles are reversal points, diamonds are bifurcation points. (b) The vertex curve obtained by plotting the calculated
demagnetization reversal points (circles).


Fundamental quantities derived from the measured parameters in Table II are the exchange field coefficient
the domain width
C. High-Order Demagnetization Spirals and Their
Relationship to the Initial-Magnetization Curve
As seen in Fig. 2, demagnetization curves of arbitrarily
high order can be calculated using the G-PFB theory. Fig. 7(a)
shows a quasi-symmetric example, again for Isomax. Here an
alternating -field of progressively decreasing amplitude is
applied to the sample, resulting in a demagnetization spiral
of order 111. The boundary between reversible
segments is demarcated by the bifurcation points (diamonds). The reversal points
circles) lie on the vertices of the minor
segments, and are
reproduced in Fig. 7(b) as a vertex curve. The parameters
used to generate these curves are as in Table I.
The central part of Fig. 7(a) is magnified in Fig. 8 to show the
nested Rayleigh loops that are typically seen at low
Since these loops consist of Type-U and Type-D segments only,
they identify a region of reversibility, as discussed in [39], [40].
Because the NORCs in Fig. 7 are so closely spaced, the demagnetization spiral strongly resembles a typical magnetization
spiral, as would be obtained by starting with an AC-demagnetized specimen and then applying an alternating sequence of

Fig. 8. Enlargement of central region of Fig. 7(a), showing in bold line the
lenticular Rayleigh quasi-loops that characterize the low-field reversible region.
The rest of the spiral is shown in light gray line.

-fields of progressively increasing amplitude. Such a set of

curves for an Sm-Fe-Co material has been obtained Cornejo et
al. [10]. Commenting on this set, it has been remarked that the
line connecting the loop tips is known as the normal or initial
magnetization curve [42]; a similar observation occurs in [43].
These comments prompt the idea that in an isotropic material
such as Isomax, wherein both magnetization and demagnetization processes are mediated primarily by rotation of the domain
, each half of the demagnetization vertex curve
shown in Fig. 7(b) should be equivalent to an initial magnetization (IM) curve. To test this notion, Fig. 9 compares the measured IM curve [6] with the theoretical vertex curve according
to the G-PFB theory. The rather close agreement seems to confirm this idea. Note that the vertex points associated with the
fully-reversible Type-U and Type-L segments, shown as black



E. Symmetric Minor Loops

Fig. 9. Enlargement of the theoretical first-quadrant vertex curve (open circles) of Fig. 7(b), compared with the experimental initial-magnetization (IM)
curve for Isomax (red online), from [6]. The portion of the theoretical IM curve
associated with low-field reversibility is shown as black dots.

dots in Fig. 9, occur in the reversible Rayleigh region, as would

be expected5.
This result suggests a possible new way of investigating certain aspects of the behavior of ferromagnetic materials. In Fig. 9,
the IM curves agree because a Langevin-type anhysteretic is
consistent with the properties of Isomax. As pointed out in [1],
the Langevin function is derived on the assumption that magnetization and demagnetization processes take place only via rotation of individual domain magnetization vectors
[44]. Thus,
if there is a significant difference between the measured and the
Langevin-theory IM curves, it means that other magnetization
processes, such as nucleation and domain-wall motion, are also
at work.

D. Closed Minor Loops

Fig. 10 illustrates how a set of closed minor loops can be generated by the G-FPB theory. In Fig. 10(a) the major loop (black)
contains three instances of a 1ORC (FORC, blue) starting at a
reversal field such as
, leading to a closed minor loop consisting of a 2ORC (SORC, green) followed by a 3ORC (red).
Each such trajectory is analogous to that in Fig. 5, but is constrained to being cycled between
. Loop closure occurs because (a) the G-PFB theory incorporates the return-point
memory (RPM) effect, and (b) the reversal fields
recur. In Fig. 10(b) the 1ORCs are suppressed, and the resulting
isolated minor loops are compared with measured data (dotted
line) [6]. The agreement is seen to be close. It is observed that
whereas the classical Preisach model imposes vertical congruency on closed minor loops [14], [15] the G-PFB theory imposes
no such non-physical behavior.
5The IM curve will of course revert to reversiblity when it reaches the upper
(reversible) segment of the major loop.

Experimentally, a set of symmetric closed minor loops

(S-CMLs) can be obtained by starting with a demagnetized
specimen, and then applying to it an alternating -field whose
amplitude is increased in a sequence of discrete steps [10],
[45]. In this theoretical discussion, normalized parameters will
be used for mathematical simplicity. In principle, an S-CML
can be generated by any pair of minor segments
such that their reversal fields and magnetizations satisfy
respectively. By definition an
S-CML has to be centered at the origin of the
Using the G-PFB theory, the simplest way an S-CML can be
generated is by a 2ORC followed by a 3ORC, similarly to Fig. 5,
but with the vertex (reversal) fields equal and opposite, so that
, see Fig. 11. It is also necessary that the vertex
. However, it is not sufficient
magnetizations satisfy
to just specify the reversal fields and then expect to get a perfect
S-CML: one also needs to know by what route the vertex points
of the S-CML should be approached. A simple possibility is to
start at point 1 on the major loop , and then to traverse
from point 1 to point 2, followed by
from point 2 to point 3,
and finally
from point 3 back to point 2. However, unless the
value of
is correctly chosen one will not even get a closed
minor loop. With luck, one may get a (maybe closed) minor loop
that has vertex fields at
, but uncontrolled vertex magnetizations. Neither is one free to choose any arbitrary point of the
plane as one of the two vertices of an S-CML. Both vertices have to be points on the initial-magnetization curve (IMC).
No other points are eligible.
In the next section, we describe a mathematical procedure for
generating perfectly symmetric CMLs.
1) Method: The situation is shown in Fig. 11, which was generated using typical parameters for Isomax, which closely conforms to the basic assumptions of the G-PFB theory as presently
formulated [1], [2]. To obtain a
type S-CML,
three conditions must be satisfied.
1) The fields at points 2 and 3 must be equal and opposite, so
2) The magnetizations at points 2 and 3 must also be equal
and opposite:
3) The two vertex points
must lie
upon the IMC.
For the case illustrated in Fig. 11, the sought S-CML can be
generated only by initiating a specific 1ORC at point 1 on the
is Type-LA and its irreversible ascender
major loop . Here
(thin line) has the general equation

see (24). Thus we can determine the centerpoint magnetizations
that characterize
by solving the simultaneous



Fig. 10. Three closed asymmetric minor loops in Isomax. (a) In this instance, each minor loop is reached via a 1ORC (blue online), and consists of a 2ORC (green
online) followed by a 3ORC (red online). In the case of the uppermost loop, the 3ORC overlaps the 1ORC. (b) Comparison between the theoretical curves (gray,
1ORCs suppressed) and measured data (dots) [6]. Reversible segments are shown in thick line, irreversible segments in thin line.

Fig. 12. Comparison between calculated (continuous line) and measured data
(dots) for four nested symmetric closed minor loops in a soft 6.5% Si-Fe material. As in other figures, thick lines indicate reversibility, thin lines irreversibility.
Measured data are courtesy of V. Basso [46].
Fig. 11. A (normalized) scenario for calculating a symmetrical closed minor
loop in Isomax. The 1ORC 0 , the 2ORC 0 , and the 3ORC 0 are in color
online. Thick lines indicate reversibility, thin lines irreversibility The theoretical
initial magnetization curve (IMC) is shown in dotted line (magenta online).

Once curve
is known, magnetization
can be found by
solving (33). The required starting field
for curve
is then
found from the intersection of curves and . The theoretical
IM curve is also plotted in Fig. 10: it does indeed pass exactly
through the vertices of the S-CML, at points 2 and 3.
2) Symmetric Loops: Comparison With Measurement:
a) Very Soft Si-Fe Material: Fig. 12 compares calculated
symmetric closed minor loops with measured data due to Basso

et al. for a very soft 6.5% Si-Fe specimen that was annealed for
1 hr at 1100 C, then ac-demagnetized [46]. This material has a
coercivity of only
A/m. Relevant data are listed in Table III.
The agreement is seen to be reasonable.
Fundamental quantities derived from the measured data in
Table III are the exchange field coefficient
and the domain width parameter
b) Very Hard Sm-Fe-Co Material: Fig. 13 gives a result
for symmetric closed minor loops in an exchange-coupled
nanocrystalline Sm Fe Co sample that has a high coercivity of
A/m. This value of
is 75 000 times
larger than for the material shown in Fig. 12, demonstrating the
very wide-range capability of the G-PFB modeling procedure.
Other relevant data are given in Table IV.



DATA FOR 6.5%Si-Fe

Fig. 13. Closed minor loops in Sm Fe Co . Measured data due to Cornejo

et al. [10], shown in dotted line, is compared with continuous curves calculated
using the G-PFB theory.

Quantities derived from the measured parameters in Table IV

are the exchange field coefficient
and the domain
This paper has introduced a generalized positive-feedback
(G-PFB) theory that can be used to model hysteretic major and
high-order minor loop phenomena in materials that display the
return-point memory property. Based on previous work concerning modeling FORCs and SORCs [1], [2], this energy-summation theory works for arbitrary-order NORCs. It is shown
that there are two different types of NORC, one being fully reversible, the other having both reversible and irreversible segments. The key features of this theory are:

The essence of the theory is contained in only four algebraic equations. Computation times are short because numerical iterations are required only when solving pairs of
simultaneous nonlinear equations.
No extra mathematical fixes or curve-fitting procedures are
neededonly the physical parameters of the material.
Unlike numerous other modeling approaches, the field
is treated as a function of magnetization
, rather
than the inverse. This is a major advantage when using
the G-PFB theory to develop codes for finite-element
modeling (FEM).
Hysteretic behavior and the associated irreversibility are
caused by discontinuous jump phenomena due to a theoretical S-shaped quantum-mechanical (QM) magnetization curve that is inherent in the fundamental physics.
The total field is the sum of (at least) QM, anhysteretic,
and demagnetizing field contributions. This is equivalent
to an energy summation.
The paper develops algebraic even and odd th-order equations for the two different NORC types, and provides detailed
explanations using both typical high-order curves and a flowchart of the algorithm.
Important results and conclusions:
The capabilities of the resulting algorithm have been
demonstrated for materials with coercivities ranging from
A/m to 18 A/m (a ratio of 75 000:1).
Unlike the well-known Preisach model (PM), the present
theory imposes no non-physical vertical or horizontal congruencies on minor loops [15].
Also unlike the PM, no computer-intensive function identification or curve-fitting procedures are required.
Minor loops and spirals are governed by the same laws as
the major loop.
The theory shows that there exists a close relationship
between a quasi-symmetric high-order demagnetizing
spiral and the initial magnetizing curve (IMC), in that
the calculated vertices of successive minor loop segments
all lie along the measured IMC. This is the inverse
of Bertottis experimental observation for magnetizing
spirals. The theory also demonstrates that there is a close
relationship between the IMC and the shape of the major
High-order demagnetizing spirals automatically produce
lenticular Rayleigh loops that tend to be reversible. This is
consistent with numerous experimental observations that
the IMC is reversible at low -fields [39], [47].
The accuracy of the theoretical curves does not deteriorate
as the order increases.
The G-PFB algorithm described in Section II.A constructs
the possible trajectory (not necessarily the whole branch)
from a reversal point to the next memory point (due to the
RPM effect), which may or may not be a reversal point.
This means that the set of reversal points does not need to
can be
be known a priori, because a new branch
initiated on the fly at any point on branch .
The accuracy of the derived algorithm has been tested by
using it to calculate results for comparison with measured data
for (a) demagnetizing spirals up to 8th order, (b) nested closed



Fig. 14. (a) CPU times to calculate up to 11 reversal branches. (b) The CPU times to calculate up to 111 branches.

minor loops in a variety of materials, and (c) a 111th-order demagnetization spiral, the vertices of which reproduced the measured initial-magnetization curve very closely. As would be expected, the most accurate results were obtained for a material
(Isomax) whose properties closely conform to the initial theoretical assumptions. However, other kinds of material can be
modeled quite well, all with the same theory.
Fig. 14(a) shows the measured CPU times (using a Dell PC
with Intel Core 2 Quad CPU at 2.39 GHz and 3.25 GB RAM,
with non-optimized Matlab code) to compute demagnetization
spirals with various numbers of reversal branches from 1 to
11, while Fig. 14(b) shows the corresponding result up to 111
branches, as shown in Fig. 7(a).

Demagnetizing factor
Sample temperature [K]
Curie temperature [K]
Weiss exchange-field coefficient
Domain width, measured in magnetons
Vacuum permeability


Paramagnetic susceptibility
2) Normalized Physical Quantities:
Local demagnetizing parameter
Sample temperature
Ratio between magneton and domain

As in [2], the notation
is defined, for magnetizations in the range
that the field
to , by the nonlinear function
. Other specialized notations follow.
1) Physical Quantities:

Domain parameter
Global demagnetizing factor
Sum field6

Sum field


Hysteretic (QM) field component [A/m]

QM field component

Anhysteretic field component [A/m]

Anhysteretic field component

Global demagnetizing field component [A/m]

th order bifurcation field [A/m]

Global demagnetizing field component

Coercive field [A/m]

Boltzmanns constant

Sum field


Moment of a Bohr magneton [A-m ]

Coercive field

Moment of a domain [A-m ]

Magnetization [A/m]

th-order sum reversal field

th order reversal magnetization [A/m]

Saturation magnetization [A/m]


[2] the normalized fields x and x

were written as x .



th-order sum bifurcation field

th-order reversal magnetization
th-order sum bifurcation magnetization
Center-point magnetization of the th
order QM S-curve due to PFB
Center-point magnetization of the th
order anhysteretic curve due to domain
rotation and size-variation
3) Segments of an th-Order Reversal Curve (NORC)

Reversible upper segment

Reversible lower segment
Irreversible ascender
Irreversible descender
4) Specialized Functions:
Langevin anhysteresis function
Generalized anhysteresis function
Reversible normalized-field function
Irreversible normalized-field function.
The author thanks Dr. I. Mayergoyz of the University of
Maryland for permission to reproduce the measured data used
in Fig. 5, Dr. V. Basso of the Istituto Nazionale di Ricerca
Metrologica, Turin, Italy, and Dr. D. R. Cornejo of the Instituto
de Fisica, Universidade de So Paulo, Brazil, for permission
to re-use the measured data employed in Figs. 11 and 12,
respectively. He also thanks the anonymous reviewers for their
cogent comments. This work was supported by the Natural
Sciences and Engineering Research Council of Canada under
Discovery Grant 1340-2005 RGPIN.
[1] R. G. Harrison, Physical theory of ferromagnetic first-order return
curves, IEEE Trans. Magn., vol. 45, no. 4, pp. 19221939, Apr. 2009.
[2] R. G. Harrison, Positive-feedback theory of hysteretic recoil loops in
hard ferromagnetic materials, IEEE Trans. Magn., vol. 47, no. 1, pp.
175191, Jan. 2011.
[3] O. Bottauscio, G. Pellegrino, P. Guglielmi, M. Chiampi, and A. Vagati,
Rotor loss estimation in permanent magnet machines with concentrated windings, IEEE Trans. Magn., vol. 41, no. 10, pp. 39133915,
Oct. 2005.
[4] T. Matsuo, Y. Terada, and M. Shimasaki, Representation of minor
hysteresis loops of a silicon steel sheet using stop and play models,
Physica B, vol. 372, pp. 2529, 2006.
[5] S. Takahashi, S. Kobayashi, Y. Kamada, and T. Shishido, Similarity
rules of minor hysteresis loops in Fe and Ni materials, J. Magn. Magn.
Mater., vol. 320, pp. 20562061, 2008.
[6] K. Wiesen and S. Charap, A better scalar Preisach algorithm, IEEE
Trans. Magn., vol. 24, no. 6, pp. 24912493, Nov. 1988.
[7] I. D. Mayergoyz, G. Friedman, and C. Salling, Comparison of the classical and generalized Preisach hysteresis models with experiments,
IEEE Trans. Magn., vol. 25, no. 5, pp. 39253927, Sep. 1989.

[8] A. A. Adly and I. D. Mayergoyz, Experimental testing of the average

Preisach model of hysteresis, IEEE Trans. Magn., vol. 28, no. 5, pp.
22682270, Sep. 1992.
[9] I. D. Mayergoyz and A. A. Adly, Numerical implementation of the
feedback Preisach model, IEEE Trans. Magn., vol. 28, no. 5, pp.
26052607, Sep. 1992.
[10] D. R. Cornejo, M. LoBue, V. Basso, G. Bertotti, and F. P. Missell,
Moving Preisach model analysis of nanocrystalline SmFeCo, J. Appl.
Phys., vol. 81, no. 8, pp. 55885590, Apr. 1997.
[11] V. Basso, Hysteresis models for magnetization by domain wall motion, IEEE Trans. Magn., vol. 34, no. 4, pp. 22072212, Jul. 1998.
[12] O. Henze, W. M. Rucker, M. Jesenik, V. Gorican, M. Trlep, A. Hamler,

and B. Stumberger,
Hysteresis model and examples of ferromagnetic
materials, IEEE Trans. Magn., vol. 39, no. 3, pp. 11511154, May
[13] E. Barbisio, F. Fiorillo, and C. Ragusa, Predicting loss in magnetic
steels under arbitrary induction waveform and with minor hysteresis
loops, IEEE Trans. Magn., vol. 40, no. 4, pp. 18101819, Jul. 2004.
[14] I. D. Mayergoyz, Mathematical models of hysteresis, IEEE Trans.
Magn., vol. 22, no. 5, pp. 603608, Sep. 1986.
[15] E. Dlala, J. Saitz, and A. Arkkio, Inverted and forward Preisach
models for numerical analysis of electromagnetic field problems,
IEEE Trans. Magn., vol. 42, no. 8, pp. 19631973, Aug. 2006.
[16] I. D. Mayergoyz and G. Friedman, Generalized Preisach model of
hysteresis, IEEE Trans. Magn., vol. 24, no. 1, pp. 212217, Jan. 1988.
[17] I. D. Mayergoyz, Mathematical Models of Hysteresis and Their Applications. Amsterdam, The Netherlands: Elsevier, 2003.
[18] E. Della Torre, Preisach modeling and reversible magnetization,
IEEE Trans. Magn., vol. 26, no. 6, pp. 30523058, Nov. 1990.
[19] M. Emura, D. R. Cornejo, and F. P. Missell, Reversible and irreversible magnetization in hybrid magnets, J. Appl. Phys., vol. 87, no.
3, pp. 13871394, 2000.
[20] D. C. Crew, S. H. Farrant, P. G. McCormick, and R. Street, Measurement of magnetic viscosity in a Stoner-Wohlfarth material, J. Magn.
Magn. Mater., vol. 163, pp. 299312, 1996.
[21] N. Takahashi, S. Miyabara, and K. Fujiwara, Problems in practical
finite element analysis using Preisach hysteresis model, IEEE Trans.
Magn., vol. 35, no. 3, pp. 12431246, May 1999.
[22] D. Davino, A. Giustiniani, and C. Visone, Fast inverse Preisach
models in algorithms for static and quasistatic magnetic-field computations, IEEE Trans. Magn., vol. 44, no. 6, pp. 862865, Jun. 2008.
[23] D. C. Jiles and D. L. Atherton, Ferromagnetic hysteresis, IEEE Trans.
Magn., vol. 19, no. 5, pp. 21832185, Sep. 1983.
[24] D. C. Jiles and D. L. Atherton, Theory of ferromagnetic hysteresis
(invited), J. Appl. Phys., vol. 55, pp. 21152120, Mar. 15, 1984.
[25] K. H. Carpenter, A differential equation approach to minor loops in
the Jiles-Atherton hysteresis model, IEEE Trans. Magn., vol. 27, no.
6, pp. 44044406, Nov. 1991.
[26] J. V. Leite, N. Sadowski, P. Kuo-Peng, N. J. Batistela, and J. P. A.
Bastos, The inverse Jiles-Atherton model parameters identification,
IEEE Trans. Magn., vol. 39, no. 3, pp. 13971400, May 2003.
[27] J. V. Leite, N. Sadowski, P. Kuo-Peng, and A. Benabou, Minor loops
calculation with a modified Jiles-Atherton model, J. Microw., Optoelectron. Electromagn. Appl., vol. 8, no. 1, pp. 4954, Jun. 2009.
[28] J. V. Leite, A. Benabou, and N. Sadowski, Accurate minor loops calculation with a modified Jiles-Atherton model, Int. J. Comput. Math.
Elect. Electron. Eng., vol. 28, no. 2, pp. 741749, 2009.
[29] A. Benabou, J. V. Leite, S. Clnet, C. Simo, and N. Sadowski, Minor
loops modelling with a modified Jiles-Atherton model and comparison with the Preisach model, J. Magn. Magn. Mater., vol. 320, pp.
10341038, 2008.
[30] A. Bergqvist, A. Lundgren, and G. Engdahl, Experimental testing of
an isotropic vector hysteresis model, IEEE Trans. Magn., vol. 33, no.
5, pp. 41524154, Sep. 1997.
[31] L. A. L. de Almeida, G. S. Deep, A. M. N. N. Lima, and H. Neff,
Limiting loop proximity hysteresis model, IEEE Trans. Magn., vol.
39, no. 1, pp. 523528, Jan. 2003.
[32] H. Hauser and R. Groessinger, Directional dependence and minor
loops of magnetization of Ba- and Sr-ferrites, IEEE Trans. Magn., vol.
39, no. 5, pp. 28872889, Sep. 2003.
[33] D. Lin, P. Zhou, W. N. Fou, Z. Baics, and Z. J. Cendes, A dynamic
core loss model for soft ferromagnetic and power ferrite materials in
transient finite element analysis, IEEE Trans. Magn., vol. 40, no. 2,
pp. 13181321, Mar. 2004.
[34] T. Matsuo, Y. Terada, and M. Shimasaki, Stop model with input-dependent shape function and its identification methods, IEEE Trans.
Magn., vol. 40, no. 4, pp. 17761790, Jul. 2004.


[35] S. E. Zirka, Yu. I. Moroz, P. Marketos, A. J. Moses, D. C. Jiles, and

T. Matsuo, Generalization of the classical method for calculating dynamic hysteresis loops in grain-oriented electrical steels, IEEE Trans.
Magn., vol. 44, no. 9, pp. 21132126, Sep. 2008.
[36] A. M. Michaelides, J. Simkin, P. Kirby, and C. P. Riley, Permanent
magnet (de-) magnetization and soft iron hysteresis effects: A comparison of FE analysis techniques, Int. J. Comput. Math. Elect. Electron.
Eng., vol. 29, no. 4, pp. 10901096, 2010.
[37] B. Cannas and S. Cincotti, Higher order reversal hysteresis curves
approximation by a piecewise linear circuit model of hysteresis, IEEE
Trans. Magn., vol. 39, no. 3, pp. 13491352, May 2003.
[38] R. G. Harrison, A physical model of spin ferromagnetism, IEEE
Trans. Magn., vol. 39, no. 2, pp. 950960, Mar. 2003.
[39] S. Chikazumi and C. D. Graham, Physics of Ferromagnetism. New
York: Oxford University Press, 1997, p. 498.
[40] S. Zapperi, A. Magni, and G. Durin, Microscopic foundations of the
Rayleigh law of hysteresis, J. Magn. Magn. Mater., vol. 242245, pp.
987992, 2002.
[41] J. U. Lemke, An isotropic particulate medium with additive Hilbert
and Fourier field components, J. Appl. Phys., vol. 53, no. 3, pp.
25612566, Mar. 1982.


[42] G. Bertotti, Hysteresis in Magnetism. San Diego, CA: Academic,

1998, p. 19.
[43] J. L. Hay and R. I. Chaplin, Dynamic simulation of ferromagnetic
hysteretic behaviour by digital computer, Simulation, pp. 185191,
Dec. 1975, see Fig. 1.
[44] S. Chikazumi and S. H. Charap, Physics of Magnetism. New York:
Wiley, 1964, pp. 6063.
[45] S. Koyabashi, N. Kikuchi, S. Takahashi, H. Kikuchi, and Y. Kamada,
Scaling analysis of minor loops in ferromagnets with large uniaxial
anisotropy, J. Magn. Magn., vol. 322, pp. 15161518, 2010.
[46] V. Basso, G. Bertotti, A. Fortuna, and M. Pasquale, Preisach model
study of the connection between magnetic and microstructural properties of soft magnetic materials, IEEE Trans. Magn., vol. 31, no. 6, pp.
40004005, Nov. 1995.
[47] R. M. Bozorth, Ferromagnetism. New York: Van Nostrand, 1951, p.