Stability of Gyroscopic

Circulatory Systems
Firdaus E. Udwadia
Professor This paper presents results related to the stability of gyroscopic systems in the presence
Aerospace and Mechanical Engineering, of circulatory forces. It is shown that when the potential, gyroscopic, and circulatory
Civil Engineering, Information and Operations matrices commute, the system is unstable. This central result is shown to be a generaliza-
Management, and Mathematics, tion of that obtained by Lakhadanov, which was restricted to potential systems all of
430K Olin Hall, whose frequencies of vibration are identical. The generalization is useful in stability
University of Southern California, analysis of large scale multidegree-of-freedom real life systems, which rarely have all
Los Angeles, CA 90089-1453 their frequencies identical, thereby expanding the compass of applicability of stability
e-mail: [email protected] results for such systems. Comparisons with results in the literature on the stability of
such systems are made, and the weakness of results that deal with only general statements
about stability is exposed. It is shown that the commutation conditions given herein pro-
vide definitive stability results in situations where the well-known Bottema–Karapetyan–
Lakhadanov result is inapplicable. [DOI: 10.1115/1.4041825]

Introduction walking robots, self-oscillations (shimmy) in aircraft wheels, flut-

ter in aerospace systems, dynamics of brake squealing, and wear
The importance of the investigation of stability and instability
in paper calendars [10–16].
of linear dynamical systems is underlined by the fact that it stands
We consider a nonlinear system described by the n-vector of
at the cross-roads of our understanding of both natural phenom-
generalized coordinates, q 2 Rn , whose equilibrium point is
ena, like tornados and ocean waves, and engineered structural and
located, with no loss of generality, at q ¼ q_ ¼ 0. The dot repre-
mechanical systems such as rotorcraft, spacecraft, and aero-elastic
sents differentiation with respect to time, t. In the close vicinity of
systems. Such linear systems also provide an important inroad
this equilibrium point, we assume that the equation of motion of
into our understanding of the stability (or instability) of general
the system can be written as
nonlinear dynamical systems, which are ubiquitous in modeling
numerous and diverse systems such as those found in the sciences ~ q€ þ G~ q_ þ ðK~ þ NÞ
~ q ¼ Qðq; qÞ
and engineering, economics, sociology, and biology. For, the sta- M _ (1)
bility of such nonlinear dynamical systems at hyperbolic fixed
where M ~ is an n-by-n real (symmetric) positive definite constant
points and the capture of their local phase portraits in the vicinity
of their fixed points via the Hartman–Grobman result [1,2] rely on matrix and q is an n-component real column vector. The constant
our understanding of the stability of their linearized versions. matrices G~ and N~ are real and skew-symmetric, and the matrix K~
Therefore, in recent years, there has been a resurgence in the study is constant, real, and symmetric. The nonlinear generalized force
of the stability of linear dynamical systems through the develop- n-vector, Q, consists of terms that are quadratic (and/or higher) in
ment of new results and the availability of excellent expository _ Linearization about the equilibrium point leads to the
q; q.
texts (e.g., see Ref. [3]). equation
Instabilities in linear systems can be generated by damping
forces, gyroscopic forces, and/or circulatory forces. The first is ~ q€ þ G~ q_ þ ðK~ þ NÞ
M ~ q¼0 (2)
characterized by the notable Kelvin–Tait–Chetaev theory [4,5].
And the second is exemplified by the stabilization of unstable which would also have resulted were our system linear from the
potential systems through the use of gyroscopic forces [6]. The very outset.
third is dealt with by the celebrated Merkin’s theorem [7], which Using the transformation zðtÞ ¼ M ~ 1=2 qðtÞ and premultiplying
1=2
was generalized in Refs. [8] and [9]. Reference [8] does not deal by M~ , Eq. (2) can be rewritten as
with the quintessential problem of the severe restriction of
Merkin’s theorem, which is only applicable to systems whose
€z þ Gsk z_ þ ðKs þ Nsk Þz ¼ 0 (3)
vibrational frequencies are all identical, thereby making the result
very narrow in its applicability to many real-life systems. This
underlying restriction in Merkin’s theory is removed in Ref. [9], where Gsk ¼ M ~ 1=2 G~ M
~ 1=2 and Nsk ¼ M~ 1=2 N~ M~ 1=2 are skew-
and it is shown that for a potential circulatory system, even a single ~ 1=2 1=2
repeated vibrational frequency makes the system unstable. This symmetric matrices, and matrix Ks ¼ M K~ M~ is symmet-
generalization therefore expands the scope of its applicability to ric. Being skew-symmetric, the eigenvalues of the matrices Gsk
encompass large-scale, real-life, multidegree-of-freedom systems. and Nsk are either zero, or conjugate pairs of pure imaginary num-
While it is easy to visualize the presence of damping forces bers. Clearly, the dynamical system described by Eq. (3) is equiv-
(caused by dissipative processes in structural or mechanical sys- alent to the one described by Eq. (2). We will be using Eq. (3)
tems) and also gyroscopic forces (often caused by rotary motion from here on, and we will assume that the system is not the trivial
in rotating flexible machinery, spinning elastic systems, and system €z ¼ 0 in which Gsk ¼ Ks ¼ Nsk ¼ 0.
astrodynamics), the presence of circulatory forces is perhaps Forces expressed by the skew-symmetric matrix Nsk are called
some-what less intuitive. Yet, such forces arise in many areas of circulatory forces or nonconservative positional forces [15,16];
real-life applications. Some examples are control of two-legged forces caused by the symmetric matrix Ks are called potential
forces; and, forces expressed by the skew-symmetric matrix Gsk
are referred to as gyroscopic forces. The system described by
Contributed by the Applied Mechanics Division of ASME for publication in the
JOURNAL OF APPLIED MECHANICS. Manuscript received August 26, 2018; final
Eq. (3) may be thought of as a generalization of the one consid-
manuscript received October 24, 2018; published online November 16, 2018. Assoc. ered in Ref. [8] in which only circulatory forces (Nsk ) are included
Editor: George Haller. and no gyroscopic forces (Gsk ¼ 0).

Journal of Applied Mechanics Copyright V

C 2019 by ASME FEBRUARY 2019, Vol. 86 / 021002-1

Downloaded From: https://appliedmechanics.asmedigitalcollection.asme.org on 11/18/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Main Result is unstable if the nonzero n by n matrices Ks , Gsk , and Nsk com-
mute pairwise, i.e.,
Our result will show that when the three matrices Ks , Gsk , and
Nsk pairwise commute, then the system described by Eq. (3) is
unstable. We begin with four lemmas, which deal with some of Nsk Ks ¼ Ks Nsk ; Ks Gsk ¼ Gsk Ks ; and Nsk Gsk ¼ Gsk Nsk (9)
the properties of these three matrices. They will be used later on
and will also establish the notation. Proof. Using the transformation zðtÞ ¼ T yðtÞ and premultiplying
LEMMA 1. If the n by n symmetric matrix Ks commutes with the by T T , Eq. (3) becomes
nonzero skew-symmetric matrix Nsk (Gsk ,) then the matrix Ks must
have multiple eigenvalues [9]. y€ þ G y_ þ ðK þ NÞ y ¼ 0 (10)
Remark 1. When Lemma 1 is satisfied, the n by n matrix Ks
therefore has k < n distinct eigenvalues k1 ; k2 ;    ; kk with cor- where the matrices K, G, and N are given in Eqs. (5) and (6).
responding multiplicities i1 ; i2 ;    ; ik , respectively. Since we Note that N ¼ diagðN1 ; N2 ; …; Nk Þ and G ¼ diagðG1 ; G2 ; …;
P
have a total of n eigenvalues, kj¼1 ij ¼ n. The symmetric matrix Gk Þ have the same block diagonal structure. The matrices N and G
Ks can be diagonalized using the orthogonal matrix T, and we therefore commute if and only if their corresponding blocks com-
have the block diagonal matrix K ¼ T T Ks T, which can be written mute, i.e., Nj Gj ¼ Gj Nj , j ¼ 1; 2; …; k. The matrices Nj and Gj
as are normal, and therefore unitarily diagonalizable; they also com-
mute, and therefore are simultaneously unitarily diagonalizable
2 3 [17]. We therefore have ij by ij unitary matrices Uj such that
k1 I1
6 k2 I2 7
K¼6
6
..
7
7 (4) Uj Nj Uj ¼ Nj ; and Uj Gj Uj ¼ Hj ; j ¼ 1; 2; …; k; (11)
4 . 5
k k Ik where Nj and Hj are diagonal matrices containing the eigenvalues
of the skew-symmetric blocks Nj and Gj , which are either zero or
Here, each block along the diagonal is the product of one of the conjugate pairs of pure imaginary numbers.
distinct eigenvalues, kj , of Ks multiplied by the ij by ij identity Hence, the block diagonal matrix G can be diagonalized by
matrix Ij , where ij is the (algebraic) multiplicity of kj . With no matrix U ¼ diagðU1 ; U2 ; …Uk Þ because
loss of generality, we list these blocks along the diagonal of the
matrix K so that i1  i2      ik : U GU ¼ diagðU1 ; U2 ; …; Uk Þ diagðG1 ; G2 ; …; Gk Þ
LEMMA 2. If the nonzero n by n symmetric matrix Ks commutes  diagðU1 ; U2 ; …; Uk Þ
with the nonzero skew-symmetric matrix Nsk (Gsk ), there exists an
orthogonal matrix T that diagonalizes the matrix Ks such that the ¼ diagðU1 G1 U1 ; U2 G2 U2 ; …; Uk Gk Uk Þ
diagonal matrix ¼ diagðH1 ; H2 ; …; Hk Þ :¼ H (12)

K ¼ T T Ks T (5) Similarly, the matrix N can be diagonalized by U because

and the skew-symmetric matrix U  NU ¼ diagðU1 ; U2 ; …; Uk Þ diagðN1 ; N2 ; …; Nk Þ

 diagðU1 ; U2 ; …; Uk Þ
N ¼ T T Nsk T ðG ¼ T T Gsk TÞ (6) ¼ diagðU1 N1 U1 ; U2 N2 U2 ; …; Uk Nk Uk Þ
both have the same block diagonal structure [9]. ¼ diagðN1 ; N2 ; …; Nk Þ :¼ N (13)
LEMMA 3. If the matrices Ks , Gsk , and Nsk commute pairwise, then
the matrices K, G, and N have the same block diagonal structure. Finally, the transformation U leaves the diagonal matrix K
Proof. By the last lemma, since Ks and Gsk commute, the mat- unchanged because
rices K and G have the same block diagonal structure. Also, since
Ks and Nsk commute, the matrices K and N too have the same U KU ¼ diagðU1 ; U2 ; …; Uk Þ diagðk1 I1 ; k2 I2 ; …; kk Ik Þ
block diagonal structure. Hence, the three matrices K, G, and N  diagðU1 ; U2 ; …; Uk Þ
have the same block diagonal structure, dictated by the block ¼ diagðk1 U1 I1 U1 ; k2 U2 I2 U2 ; …; kk Uk Ik Uk Þ
structure of the matrix K. ⵧ
Remark 2: When the matrices Ks , Gsk , and Nsk commute pair- ¼ diagðk1 I1 ; k2 I2 ; …; kk Ik Þ ¼ K (14)
wise, by the above lemma, K, G, and N have the same block diag-
onal structure, and we can write the matrices G and N as Using the transformation yðtÞ ¼ UxðtÞ and premultiplying by U  ,
2 3 2 3 Eq. (10) can now be decoupled as
G1 N1
6 G2 7 6 N2 7 x€ þ H x_ þ ðK þ NÞ x ¼ 0 (15)
6 7 6 7
G¼6 .. 7; and N ¼ 6 .. 7
4 . 5 4 . 5 where the matrices H, K, and N are diagonal. We see then that the
Gk Nk system described by Eq. (3) has been transformed into Eq. (15) by
(7) the transformation zðtÞ ¼ TU xðtÞ.
As N 6¼ 0, there must exist at least one conjugate pair of non-
whose jth (square) diagonal blocks Gj and Nj have the same size zero pure imaginary numbers 6in; along the diagonal of matrix
as the corresponding jth square diagonal block of K. In other N. Therefore, there exist at least two decoupled equations in the
words, Gj and Nj are each ij by ij skew-symmetric matrices. set of equations given by Eq. (15) that can be written in the form
LEMMA 4. The matrices Ks , Gsk , and Nsk commute pairwise, if
and only if the matrices K, G, and N also commute pairwise. ⵧ x€j þ ihx_ j þ ðk6inÞxj ¼ 0; n 6¼ 0; h 2 R (16)
We are now ready to prove our main result.
THEOREM. The system described by equation Here, xj is a component of vector x, and ih, k, in are eigenvalues
of matrices G, Ks , and N, respectively. Using the ansatz xðtÞ ¼
w expðltÞ in Eq. (16), it is easily seen that, for any nonzero n and
€z þ Gsk z_ þ ðKs þ Nsk Þz ¼ 0 (8) any real h, xj ðtÞ is unbounded since

021002-2 / Vol. 86, FEBRUARY 2019 Transactions of the ASME

Downloaded From: https://appliedmechanics.asmedigitalcollection.asme.org on 11/18/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 pffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi
Remark 4. Equation (23) is also a generalization of Merkin’s
2 2 2
ReðlÞ ¼ 6 ðh þ 4kÞ þ ð4nÞ  ðh þ 4kÞ 6¼ 0 (17)
2 2 theorem [8,9], which we discuss next.
4 Remark 5. The result given in Corollary 3 for the system
described by Eq. (22) was first obtained by Lakhadanov [18]. It is
The system described by Eq. (3) is therefore unstable, since a root often referred to in the literature as Lakhadanov’s theorem. In
of its characteristic equation has a positive real part. fact, to the dynamical system
Remark 3. It should be noted that the result obtained above
does not require the symmetric matrix Ks to be positive definite. It €z þ ðkI þ Nsk Þz ¼ 0 (24)
could be indefinite, semidefinite, or negative definite. This is
because, as seen from Eq. (17), ReðlÞ > 0 for at least one root of that was used by Merkin [7]—to establish his celebrated instabil-
the characteristic equation of the system described in Eq. (8) irre- ity theorem—Lakhadanov [18] added gyroscopic forces to obtain
spective of the value of k 2 R: the system described by Eq. (22).
COROLLARY 1. The system Merkin’s theorem states that for a stable potential system whose
frequencies of vibration are all identical, the addition of any (per-
€z þ Gsk z_ þ Nsk z ¼ 0; Gsk ; Nsk 6¼ 0 (18) turbatory) circulatory forces (Nsk 6¼ 0) renders the system unsta-
ble, i.e., the system described by Eq. (24) is unstable.
is unstable if the matrices Gsk and Nsk commute. Lakhadanov’s theorem, which we shall call L for short, therefore
Proof. We set the matrix Ks ¼ 0 in Eq. (8). Now, the first two expands upon Merkin’s instability theorem, and states that the
equalities in Eq. (9) are trivially satisfied. Since Ks ¼ 0, Ks has addition of gyroscopic forces—the term in Gsk in Eq. (22)—to
one eigenvalue, which is zero, with multiplicity n. Thus, the the unstable circulatory system (Eq. (24)) cannot render it stable if
eigenvalues of Ks are kj ¼ 0; j 2 f1; ng; and Eq. (17) becomes the matrices Gsk and Nsk commute.
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffi rq ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi As seen from Corollary 3, Lakhadanov’s theorem (L) [18] is a
2 2 special case of the result given in our theorem. The result of L
ReðlÞ ¼ 6 h4 þ ð4nÞ  h2 6¼ 0 (19) only deals with a potential system all of whose frequencies of
4
vibration are identical, which is then subjected to gyroscopic
guaranteeing that we have at least one eigenvalue of the system forces and circulatory perturbations (see Eq. (22)). Large scale,
described in Eq. (18) whose real part is positive, as long as multidegree-of-freedom physical systems that have all their fre-
Nsk 6¼ 0. quencies of vibration identical are indeed very rare, though they
COROLLARY 2. The system could occur in systems with a very small number of degrees-of-
freedom, especially those constrained by requirements of symme-
€z þ Nsk z ¼ 0; Nsk 6¼ 0 (20) try. In large-scale structural and mechanical systems that may
have hundreds, if not thousands, of degrees-of-freedom, though
is unstable. one often does find a few repeated frequencies—such as, say, the
Proof: We set Ks ¼ 0 and Gsk ¼ 0 in Eq. (8). Equation (17) now frequency of the fourth bending mode of vibration being coinci-
becomes dent with the second torsional mode—one almost never has a sit-
1 pffiffiffiffiffi uation in which all the frequencies of vibration are identical. This
ReðlÞ ¼ 6 pffiffiffi jnj (21) makes result L, though elegant, limited in its use to actual real-life
2 physical systems.
Remark 6: There is yet another way of interpreting the result
again guaranteeing that we have at least one eigenvalue of the sys-
given in the theorem obtained. It shows the effect of adding (per-
tem described in Eq. (20) whose real part is positive, as long as
turbatory) circulatory forces to gyroscopically stabilized systems.
Nsk 6¼ 0. This of course can be seen more directly by using a uni-
The result says then that the addition of a (perturbatory) circula-
tary transformation that diagonalizes the skew-symmetric matrix
tory force Nsk to gyroscopically stabilized potential systems,
Nsk . At least two eigenvaluespcorresponding
ffiffiffiffiffiffiffiffiffi to the system given
under the proviso that the commutators ½Gsk ; Nsk  ¼ ½Ks ; Gsk  ¼
in Eq. (20) will then become 6inj , and instability follows. ½Ks ; Nsk  ¼ 0, causes these systems to lose stability.
COROLLARY 3: For any real number k, the system Remark 7: To place the present result in a broader context, we
consider another elegant and useful theorem first apparently
€z þ Gsk z_ þ ðkI þ Nsk Þz ¼ 0 (22) obtained, again, by Lakhadanov [18], and independently by
is unstable when the matrices Gsk and Nsk commute. Karapetyan [19]. They generalized a result obtained previously by
Proof: The system described by Eq. (22) is a special case of Bottema for 2 and 3 degree-of-freedom systems [20].1 Their sig-
that described in the theorem obtained in this paper. It results nificant and outstanding contribution states that the system
from Eq. (8) when the n by n matrix Ks ¼ kI, where k is any arbi- described by the equation
trary real number. Using the theorem, we see that all the equalities
x€ þ Gsk x_ þ ðK þ Nsk Þx ¼ 0 (25)
in Eq. (9) are then satisfied by requirements laid out in the corol-
lary. The first two equalities are satisfied because the identity is unstable if traceðGsk Nsk Þ 6¼ 0. Here, K is a symmetric matrix,
matrix commutes with all matrices. When k ¼ 0, Corollary 1 while Gsk and Nsk are skew symmetric. The matrices K, Gsk , and
applies. Hence, the system is unstable. Nsk are general, noncommuting matrices. We shall call this the
COROLLARY 4. A potential system with a circulatory force, Bottema–Lakhadanov–Karapetyan result, or BLK for short.
described by the equation It should be noted that when traceðGsk Nsk Þ ¼ 0, BLK cannot be
applied, and it does not provide any information regarding the sta-
€z þ ðKs þ Nsk Þz ¼ 0 (23) bility or instability of the system described by Eq. (25). But when
the conditions (see Eq. (9)) of our main theorem are satisfied and
traceðGsk Nsk Þ ¼ 0, the system is assuredly unstable. Thus, our
is unstable when the matrices Ks and Nsk commute.
main theorem provided herein can yield information about the sta-
Proof: We set Gsk ¼ 0 in Eq. (8). Since the zero matrix com-
bility of the system (Eq. (25)) in situations where BLK cannot be
mutes with all matrices, the conditions of the theorem then simply
applied. However, the condition traceðGsk Nsk Þ 6¼ 0 is generic (see
require the matrices Ks and Nsk to commute. Equation (17) again
Remark 8) and therefore is much more widely applicable. This is
shows that the system is unstable. This, result, which was first
obtained in Ref. [8], arises as a natural consequence of the
Theorem. ⵧ 1
The author thanks an anonymous reviewer for pointing out this reference.

Journal of Applied Mechanics FEBRUARY 2019, Vol. 86 / 021002-3

Downloaded From: https://appliedmechanics.asmedigitalcollection.asme.org on 11/18/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 because the set of matrices for which the commutation conditions represent periodic orbits—often a key to our understanding of
of the theorem are true is far smaller than the set of matrices for physical phenomena. Another way of saying this is that though
which traceðGsk Nsk Þ ¼ 0. Yet, as we will see later on, systems the chance that a (random) smooth curve drawn in 3D phase space
with nongeneric properties can contain uncountably infinite mem- would exactly close on itself is infinitesimally small, such closed
bers, and a particular system under consideration could have a curves represent closed periodic orbits that are quintessential to
nongeneric property (see Remarks 8 and 9). our understanding of physical phenomena. From a theoretical
As stated before in Remarks 4 and 5, Merkin’s famous theorem standpoint, one can think of the property of a square matrix being
has been generalized [8,9]. This Generalized Merkin result says singular. This is a mathematically nongeneric property relative to
that if to a stable potential system x€ þ Ks x ¼ 0 one adds a (pertur- the set of square matrices, and matrices with this property belong
batory) circulatory force Nsk that commutes with Ks , the new sys- to a nongeneric (nontypical) set. In common parlance, one might
tem obtained, namely, x€ þ ðKs þ Nsk Þx ¼ 0, will be rendered assert that: “in general,” square matrices are not singular. How-
unstable in a flutter instability. A critical aspect that connects the ever, this “nongeneral” or nontypical property has profound theo-
recent generalization given in Ref. [9] to Merkin’s theorem is that retical importance in our understanding and prediction of the
for such commutation to occur for a nonzero skew-symmetric behavior of almost every dynamical system. For example, the
matrix Nsk , the symmetric matrix Ks must have at least one eigen- eigenvalue problem, among others, rests on setting the determi-
value with multiplicity greater than 1. One can therefore place in nant of a matrix to zero, thereby utilizing and basing our under-
context the contribution of this paper through the following obser- standing of dynamical systems on a matrix that has a nongeneric
vation: just as the result of L given in Eq. (22) provides instability property, which we might describe as belonging to a very special
results that expand Merkin’s result through the addition of gyro- not-at-all-commonly-found set of matrices.
scopic forces to a circulatory system (see Remark 5), the theorem In like manner, the set of skew symmetric matrices Gsk and Nsk
in this paper similarly provides new instability results that expand that have the nongeneric (nontypical) property traceðGsk Nsk Þ ¼ 0
the Generalized Merkin result, also through the addition of gyro- could arise in the modeling of physical applications, and can con-
scopic forces to a circulatory system. In other words, L stands in stitute sets whose members are, in fact, uncountably infinite, as
relation to Merkin’s theorem in a manner analogous to the way shown below.
the present theorem stands in relation to the Generalized Merkin Remark 9. The theorem obtained herein opens up the possibility
result. But unlike Merkin’s theorem and L, both the Generalized of being useful in cases where traceðGsk Nsk Þ ¼ 0. It shows that if
Merkin result (as pointed out in Ref. [9]) and the present theorem traceðGsk Nsk Þ ¼ 0 and the conditions of the theorem are satisfied,
do away with the highly restrictive requirement that all the fre- the system is unstable. In this sense, the theorem complements the
quencies of vibration of the system be identical, making these BLK result. The BLK result and the present Theorem can together
newer results more applicable to real-life systems in various areas be thought of, in a sense, as the “circulatory counterpart” of the
of application in physics and engineering. In other words, our the- celebrated Kelvin–Tait–Chetaev theorem [4,5], which states that
orem states that given an unstable potential circulatory system in the addition of a (perturbatory) positive definite damping matrix
which the matrices Ks and Nsk ð6¼ 0Þ commute, the system cannot (force) D to gyroscopically stabilized systems will cause them to
be stabilized by the addition of a gyroscopic matrix Gsk that com- lose stability.
mutes with Ks and Nsk . It should be noted that while the statement of BLK (see Remark
Remark 8. As mentioned before, the BLK result is indeed more 7, Eq. (25)) is equivalent to the statement that if the system
widely applicable to the system described in Eq. (25) since the described by Eq. (25) is stable then traceðGsk Nsk Þ ¼ 0, it does not
property that traceðGsk Nsk Þ 6¼ 0 is, mathematically speaking, state that the converse is also true. And it is the acceptance of the
generic (or in common parlance, typical) among members of the converse, namely, that traceðGsk Nsk Þ ¼ 0 implies stability, which
set of skew symmetric matrices. The commutation conditions that is sometimes erroneously invoked in engineering practice.
are required to be satisfied in the Theorem herein require a much The cause of this misunderstanding is often the observation that
larger set of equations to be satisfied than the single condition that the property traceðGsk Nsk Þ 6¼ 0 is generic, and loosely speaking,
is required to be satisfied by BLK, and are therefore far less typical for the set of skew symmetric matrices Gsk and Nsk . So,
widely applicable. given any two skew symmetric matrices, this property is typically
More specifically, what is meant by a generic property is the satisfied, and by the BLK theorem (see Remark 7), the system
following [21]. Consider a given set S of objects (on which a described by Eq. (25) is unstable. Therefore, when this property is
topology has been defined so that we can speak of one element not satisfied and traceðGsk Nsk Þ ¼ 0—and this is the erroneous
being “close” to another). Let there be some property P that each step—the system must be stable.
of the elements, s, of S may or may not have. Then property P is To illustrate some of these ideas, we consider the following two
generic relative to S under the following circumstances: (a) if s 2 examples.
S possesses property P, then there exists a (open) neighborhood A Example 1. Consider the system described by Eq. (25) with the
of s such that every s 2 A also possesses property P, and (b) every following matrices:
neighborhood in S contains an element possessing property P. If 2 3
either one of these conditions is not satisfied, the property is called 0g 0 0
" # 6 7
nongeneric (mathematically, nontypical). Clearly then, the prop- k1 I 2 0 6g 0 0 07
erty traceðGsk Nsk Þ 6¼ 0 is mathematically generic relative to the Ks ¼ ; Gsk ¼ 6 7
6 0 0 0  g 7; and
set of skew-symmetric matrices. 0 k2 I 2 4 5
However, it should be pointed out that those properties P that 0 0 g 0
may not be mathematically generic relative to a given set S can 2 3 (26)
be, and very often are, important in the physical world, and espe- 0 n 0 0
6 7
cially so for our understanding of physical phenomena. For exam- 6 n 0 0 0 7
ple, an orbit that has the property P of being isolated and Nsk ¼ 6
6 0 0 0n 7
7
4 5
periodic—a limit cycle—of a nonlinear dynamical system is non-
generic (and nontypical) relative to the set S of all orbits of the 0 0 n 0
dynamical system in phase space. Yet, it is of great importance in
describing and understanding the behavior and physics of the non- where I2 is the 2  2 identity matrix, and k1 ; k2 ; g; and n are arbi-
linear system. More generally, smooth curves drawn in 3D phase trary real constants. We find that the conditions of our theorem are
space that have the property P of being closed, are also nontypical satisfied since the three matrices commute pairwise, and therefore
relative to the set of smooth curves that can be drawn; yet they are the gyroscopic system is unstable for arbitrary (infinitesimal) per-
of significant importance in physics and engineering since they turbations (given by the magnitude of n). In fact, the real parts of

021002-4 / Vol. 86, FEBRUARY 2019 Transactions of the ASME

Downloaded From: https://appliedmechanics.asmedigitalcollection.asme.org on 11/18/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 four of the eigenvalues of this system, l1 ; l2 ; l3 ; and l4 ; can be Since G ¼ GT , the coefficient of the term l2n1 , namely,
explicitly obtained, and are a1 ¼ traceðGÞ  0. When a1 ¼ 0, if the system described by
Eq. (3) is stable, then a2i1 ¼ 0; i ¼ 2; 3; …; n (see Theorem 1, in
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi Ref. [19]). Hence, when even a single coefficient of an odd power
1 of l in the characteristic polynomial is not zero, the system is
Reðl1 and l2 Þ ¼ 2 ðg2 þ 4k1 Þ2 þ 16n2  2ðg2 þ 4k1 Þ (27)
4 unstable. It is a mathematically nongeneric property (see Remark
8) for all the coefficients of the odd powers of l in a 2n-degree
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi polynomial to be zero, relative to the set of all 2n-degree polyno-
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi mials (see Remark 8 for the definition of nongeneric). Alterna-
1
Reðl3 and l4 Þ ¼ 2 ðg2 þ 4k2 Þ2 þ 16n2  2ðg2 þ 4k2 Þ (28) tively put, it is a generic property that at least one coefficient of an
4
odd power of l in the polynomial in Eq. (30) is nonzero, relative
to G–K–N systems. And this is what is exactly meant, in a more
These roots are seen to be positive as long as n 6¼ 0, no matter precise manner, when Ref. [22] states that:
how small its value, confirming the result given by the theorem.
The real parts of the other four roots have the same absolute val- in general; G-K-N systems are unstable (31)
ues as those given in Eqs. (27) and (28), except that they are
negative. That is, such systems possess the mathematically generic property
Yet, we find that traceðGsk Nsk Þ ¼ 0. This makes the BLK result that not all the coefficients of odd powers in l of their characteris-
inapplicable. It cannot be used to answer whether the system is tic polynomials are zero, and therefore they are in general unsta-
stable or unstable, and the question of the stability of the system ble, relative to the set of G–K–N systems. Exceptions though,
would remain unresolved were the theorem developed herein not however infrequent, may arise when a particular system does not
available. Furthermore, this example points out that the condition satisfy the genericity property underlying such a statement.
traceðGsk Nsk Þ ¼ 0 does not, in general, imply stability. Use of a statement like (31) without explicit specification and/
Example 2. We next consider a system again described by or knowledge of the cause of the genericity has some important
Eq. (25) but now with the pairwise noncommuting matrices given drawbacks. It provides no direction to the stability analyst on
by whether a particular system under consideration falls in that gen-
eral category to which such a statement refers. One would need to
2 3 further explore whether the underlying property that defines the
0a 0 0 genericity is satisfied to make progress on assessing its stability,
6 7
6a 0 0 07 because a particular system under consideration may not satisfy
Ks ¼ diagðk1 ; k2 ; k3 ; k4 Þ; Gsk ¼ 6
60 0
7; and
4 0  b7
5
this genericity property. From a practical engineering standpoint,
this last consideration is paramount, and it rules out the sole use
0 0 b 0 of such statements in definitively ascertaining the stability or
2 3
0 a 0 0 instability of a given particular dynamical system unless consider-
6 7 ably more information about that particular system is obtained
6 a 0 0 07 and incorporated.
Nsk ¼ 6
6 0
7 (29)
4 0 0 b7 5 Remark 11. Though the gyroscopic circulatory system consid-
0 0b 0 ered in this paper is seemingly simple and has been worked on for
over 60 years now, its stability analysis still remains incomplete
since there can be, it appears, an uncountably infinite number of
where a and b are arbitrary constants at least one of which is dif- nongeneric G–K–N systems to which the BLK result and the pres-
ferent from zero, and the constants ki ; i ¼ 1; 2; 3; 4; are all ent theorem are both inapplicable. This is an important area for
unequal from one another. Since Ks now does not have the same future research and will not be discussed further here. It will be
block diagonal structure as Gsk ðNsk Þ, the matrices Ks and Gsk the topic of a later communication.
ðKs and Nsk Þ do not commute, and therefore the present theorem
is inapplicable, since the equalities given in Eq. (9) are not satis-
fied. However, traceðGsk Nsk Þ ¼ 2ða2 þ b2 Þ > 0, and hence by Conclusions
BLK [18–20], the system described by Eq. (25) with the set of This paper presents a set of conditions that assure the instability
matrices given in Eq. (29) is assuredly unstable. of gyroscopic circulatory systems. It is shown that when the
The two examples shown above can be extended to 4n by 4n potential, gyroscopic, and circulatory matrices pairwise commute,
matrices in a straightforward manner by adding 4  4 diagonal the system is unstable. Consequences of the commutation on the
blocks that have structures shown in Eqs. (26) and (29). We there- multiplicity of the eigenvalues of the potential system are devel-
fore notice that, in general, the present theorem as well as the oped. The well-known Lakhadanov instability theorem, which is
BLK result may be useful, depending on the application at hand, restricted to systems in which the potential part of the dynamical
though of course the genericity of the BLK result leads to its far system has all its frequencies of vibration identical, appears as a
wider applicability (see Remark 8). special case of the result obtained herein. The removal of this
Finally, going back to Example 1 above, we point out that the stringent restriction on the potential system renders the present
property traceðGsk Nsk Þ ¼ 0 is shared by all 4  4 matrices Gsk and result applicable to a wider class of real-life applications that deal
Nsk with structures shown in Eq. (26) (and more generally, by all with multidegree-of-freedom systems found in physics, engineer-
their 4n by 4n matrix counterparts mentioned above) irrespective ing, and other fields of application.
of the actual values of g and n. They therefore constitute an The contribution of the main result in this paper is shown in the
uncountably infinite set of dynamical systems that have this prop- context of the present literature. The result can be construed as an
erty. In what follows, we shall call systems described by Eq. (3) extension of the Generalized Merkin theorem when gyroscopic
as G–K–N systems, for short. forces are added to potential circulatory systems. The relationship
Remark 10. More generally, the characteristic polynomial of of the result to the Generalized Merkin theorem is analogous to
any n degree-of-freedom G–K–N system has degree 2n, and can the relationship of Lakhadanov’s result to Merkin’s theorem. It is
be written as shown that general statements like (31) about sets of dynamical
systems that simply state they are in general (un)stable need to be
used with some caution. The use of such statements without
l2n þ a1 l2n1 þ a2 l2n2 þ a3 l2n3    þ a2n1 l þ a2n ¼ 0 (30) proper knowledge and/or attribution of the reason for their

Journal of Applied Mechanics FEBRUARY 2019, Vol. 86 / 021002-5

Downloaded From: https://appliedmechanics.asmedigitalcollection.asme.org on 11/18/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 genericity (or nongenericity) do not guarantee either stability or [4] Kelvin, B. W. T., and Tait, P. G., 1879, Treatise on Natural Philosophy, Cam-
instability when applied to a given, particular dynamical system bridge University Press, New York.
[5] Zajac, E. E., 1964, “The Kelvin-Tait-Chetaev Theorem and Extensions,”
of interest, because they then lack the definitive knowledge J. Astronaut. Sci., 11(2), pp. 46–49.
required to unequivocally ascertain the system’s stability or insta- [6] Huseyin, K., Hagedorn, P., and Teschner, W., 1983, “On the Stability of Linear
bility. Equally relevant is the fact that the set of systems that may Conservative Gyroscopic Systems,” J. Appl. Math. Phys., 34(6), pp. 807–815.
have a mathematically nongeneric property could be: (1) uncount- [7] Merkin, D. R., 1974, Gyroscopic Systems, Nauka, Moscow, Russia.
[8] Bulatovic, R. M., 1999, “On the Stability of Linear Circulatory Systems,” Z.
ably infinite in number, and (2) very important in building theoret- Angew. Math. Phys., 50, pp. 669–674.
ical constructs. Besides being useful in modeling the physical [9] Udwadia, F. E., 2017, “Stability of Dynamical Systems With Circulatory
world, systems with nongeneric properties can often provide a Forces: Generalization of the Merkin Theorem,” Am. Inst. Aeronaut. Astro-
framework for understanding observed phenomena. It should be naut., 55(9), pp. 2853–2858.
[10] Beletsky, V. V., 1995, “Some Stability Problems in Applied Mechanics,” Appl.
noted that despite the simplicity of the system considered in this Math. Comput., 70(2–3), pp. 117–141.
paper and the fact that its stability has been a subject of interest [11] Zhuravlev, V. P., and Klimov, D. M., 2010, “Theory of the Shimmy Phenomen-
for more than 60 years, we do not yet have a general theory that on,” Mech. Solids, 45(3), pp. 324–330.
covers all the nongeneric systems that may arise. This leaves the [12] Hagedorn, P., Eckstein, M., Heffel, E., and Wagner, A., 2014, “Self-Excited
Vibrations and Damping in Circulatory Systems,” ASME J. Appl. Mech.,
general stability analysis of gyroscopic circulatory systems 81(10), p. 101008.
incomplete as of now. [13] Spelsberg-Korspeter, G., Hochlenert, D., and Hagedorn, P., 2011, “Self-
The new result obtained in this paper is also compared with the Excitation Mechanisms in Paper Calendars Formulated as a Stability Problem,”
elegant and long-established BLK instability theorem. Though far Tech. Mech., 31(1), pp. 15–24.
[14] Dowell, E. H., 2011, “Can Solar Sails Flutter?,” AIAA J., 49(6), pp.
less widely applicable than the BLK theorem, it is shown to be 1305–1307.
complementary to it in that it could be useful in providing infor- [15] Ziegler, H., 1953, “Linear Elastic Stability, a Critical Analysis of Methods—
mation on the nature of the stability of dynamical systems for Part 1,” Z. Angew. Math. Phys., 4(2), pp. 89–121.
which the elegant, older theorem might be inapplicable. [16] Ziegler, H., 1953, “Linear Elastic Stability, a Critical Analysis of Methods—
Part 2,” Z. Angew. Math. Phys., 4(3), pp. 167–185.
[17] Horn, R. A., and Johnson, C. R., 1991, Matrix Analysis, Cambridge University
Press, Cambridge, UK, p. 103.
[18] Lakhadanov, V. M., 1975, “On Stabilization of Potential Systems,” J. Appl.
References Math. Mech., 39(1), pp. 45–50.
[1] Arrowsmith, D. K., and Place, C. M., 1998, Dynamical Systems, Differential [19] Karapetyan, A. V., 1975, “About the Stability of Nonconservative Systems,”
Equations, Maps and Chaotic Behavior, Chapman and Hall, New York, Vestn. Mosk. Univ. Ser. 1: Mat. Mekh., 4, pp. 109–113.
pp. 77–81. [20] Bottema, O., 1955, “On the Stability of the Equilibrium of a Linear Mechanical
[2] Perko, L., 1991, Differential Equations and Dynamical Systems, Springer System,” Z. Angew. Math. Phys., 6(2), pp. 97–104.
Verlag, New York, pp. 120–127. [21] Casti, J., 1992, Reality Rules: I, John Wiley, New York, pp. 63–64.
[3] Kirillov, O. N., 2013, Nonconservative Stability Problems of Modern Physics, [22] Hagedorn, P., and Hochlenert, D., 2015, Technische Schwingungslehre, Verlag-
Walter de Gruyter, Berlin. Europa-Lehrmittel, Vollmer GmbH & Co. KG, Haan, Germany, pp. 155–157.

021002-6 / Vol. 86, FEBRUARY 2019 Transactions of the ASME

Downloaded From: https://appliedmechanics.asmedigitalcollection.asme.org on 11/18/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use