On the representation of the stability region in oscillation problems with the aid of the Hurwitz determinants


Material Information

On the representation of the stability region in oscillation problems with the aid of the Hurwitz determinants
Series Title:
Physical Description:
12 p. : ill ; 27 cm.
Sponder, E
United States -- National Advisory Committee for Aeronautics
Place of Publication:
Washington, D.C
Publication Date:


Subjects / Keywords:
Aerodynamics   ( lcsh )
Oscillations   ( lcsh )
federal government publication   ( marcgt )
bibliography   ( marcgt )
technical report   ( marcgt )
non-fiction   ( marcgt )


This report concerns the use of the Hurwitz determinants in defining boundaries of regions where oscillatory phenomena are to be stable or unstable. A simplification is suggested as an aid in reducing the computations usually required, although it is emphasized that point checks in the various regions defined are required using the complete set of Hurwitz determinants or some other complete stability determination.
Includes bibliographic references (p. 9).
Sponsored by National Advisory Committee for Aeronautics
Statement of Responsibility:
by E. Sponder.
General Note:
"Report date August 1952."
General Note:
"Translation of "Zur Darstellung des Stagilitätsgebietes bei Schwingungsaufgaben mit Hilfe der Hurwitz-Determinanten." Schweizer Archiv, March 1950."

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 003771595
oclc - 85872700
sobekcm - AA00006200_00001
System ID:

Full Text
'Jfj(M(VV 3Lfg

3y 5 6 / 4 C( 37L






By E. Sponder

For oscillation phenomena which may also have an unstable course,
it is customary to represent the regions where stability or instability
prevails in a plane as functions of two parameters x and y.

In order to determine whether stability exists at any point of the
plane represented (thus, whether a disturbance of the oscillation phe-
nomenon considered is damped in its course) it must be investigated
whether all roots X of the frequency (characteristic) equation

Xn + aln-1 + + a = 0

have a negative real part at this arbitrary point. The ,mathematical
condition for this is known to be that the n Hurwitz determinants D1
to Dn which are formed from the coefficients al to an and are
functions of the two parameters x and y mentioned before are all
positive for this point. Resulting from this criterion and completely
equivalent to it is the fact that the coefficients al to an and only
a few certain Hurwitz determinants must be positive.

If, conversely, all roots X of the frequency equation have a nega-
tive real part, all values D1 to Dn are positive. If one now visu-
alizes the point considered before (for which we assume stability to have
been established) as traveling in the representation plane of figure 1,
thete vary with its parameters -x and 'y also the coefficients al
to an, the real parts of the roots X, and finally the n Hurwitz-
determinants. If one arrives at a point where for instance a real root

*"Zur Darstellung des Stabilitatsgebietes bei Schwingungsaufgaben
mit Hilfe der Hurwitz- Determinanten." Schweizer Archiv, March 1950,
PP. 93-96.

NACA TM 1348

disappears, an also disappears since lani is the product of the values
of all roots; if one reaches, in contrast, a point where the real part
of a complex root becomes zero, it can be shown that then Dn-1 becomes
zero. For every case, however, the product anDn-1 disappears which is
nothing else but the Hurwitz determinant of the nth degree

anDn-1 = Dn

Thus the important theorem, the proof for which will be presented
later, is valid:

The limits of the (usually unique) stable region lie at
Dn = anDn1 = 0.

For graphical representation, it is therefore completely sufficient
to plot Dn = 0 or more simply an = 0 and Dn1 = 0 as separate
limiting curves of a region for which stability is known to prevail at
an arbitrary point, as illustrated by figure 1. If the limit Dn_1 = 0
is exceeded, the course of the oscillation process is "dynamically"
unstable because a damping becomes negative; beyond the limit an = 0,
one usually calls the oscillation process "statically" unstable.

In particular, the following is valid for oscillation phenomena
which lead to frequency equations of the 4th degree (reference i).

The Hurwitz determinant of the 4th degree formed from the coeffi-
cients A to E of the frequency equation

AX4 + BX3 + CX2 + DX + E = 0


D4 = ED3 = E(BCD AD2 B2E) = ER

with the expression in parentheses known as Routh's discriminant R; the
latter is therefore nothing else but the Hurwitz determinant of the

NACA TM 1348

3rd degree. Since the coefficient A is usually +1, it is generally
valid as dynamic stability condition that the expression (BC D)D B2E
turns out positive. The static stability is then guaranteed in addition
by E > 0 so that the graphic representation of the region of figure 2

Therewith, every requirement has been met; for it is impossible
that within the region denoted as stable a curve C = 0 or D = 0
could run and perhaps still further reduce this region.

It is therefore completely superfluous to investigate further what
sign the other coefficients of the frequency equation have once the
boundaries for a region E = 0 and R = 0 have been fixed; however,
and this is important, it must be known that, at an arbitrary point of
this region, stability actually prevails. For this it is sufficient
to check for a point conveniently situated (for instance on an axis)
the signs of the remaining coefficients A to D of the frequency
equation which is equivalent with the fact that there all Hurwitz deter-
minants D1 to D4 turn out positive. That this is important can, for
instance, be recognized from the fact that, in the region indicated on
the left in the plane of representation of figure 2, instability may
prevail in spite of E > 0 and R > 0.

The advantage of these recognized facts given above does not result
in much simplification for frequency equations of the 4th, 5th; or even
6th degree; however, this advantage may save a great deal of unnecessary
calculations. The expressions for the Hurwitz determinants become more
and more cumbersome for higher degrees so that it will probably be a
very welcome facilitation for the outlining of the stability region if
the mere representation of the two curves an = 0 and Dn-1 = 0 will
be sufficient.

There now follows the proof of the theorem given before that the
limits of the (usually unique) stable region lie at Dn = anDn-1 = 0.

1. For the present consideration, the existence of at least one
stable region is presupposed. Not even one need necessarily always
exist; on the other hand, sVeral stability regions, separated from
one another, may occur. One can easily make sure of this by considering
a frequency equation of the 2nd degree which pertains to an ordinary

X2 + alk + a2 = 0

NACA TM 1348

The coefficients a, and a2 are dependent on two parameters x
and y, corresponding to the two coordinate axes of the plane of repre-
sentation. For simplicity's sake, we shall assume that a2 everywhere
has positive values only; then the sign of al alone is the criterion,
for the stability, since al is a measure of the damping of the oscil-
lation considered and this oscillation is stable only if al is posi-
tive. The limit of stability now depends entirely on the sign of the
analytical (or empirical) function of the coefficient al of the two
parameters x and y, or, in other words, what form the curves al = 0
have in the plane of representation which separate the positive from the
negative values. In this manner, one can easily understand that several
stability regions may exist but that just as well not even one need exist
anywhere in the entire plane of representation.

For frequency equations of higher degree, the probability decreases
that simultaneously in several regions all stability conditions found by
Hurwitz will be satisfied. Usually, one will deal with only one single
stability region which will then be considered more thoroughly.

2. Hurwitz' criterion (reference 2) signifies that a prescribed
equation of the nth degree with real coefficients

a0n + alkn-1 + + an = 0

(ao >O)

possesses roots exclusively
of the n-determinants

Dk =

with negative real parts only when the values

al a3 a9

a0 a2 a4

0 al a3

are all positive. Therein, one generally
subscript x is negative or larger than

has to put ax = 0 when the



S a2k-3

S ak

(k = 1,2,...,n)

NACA TM 1348 5

It shall now be proved that the Hurwitz determinant of the (n l)th
degree Dn_1 always disappears when the real part of a complex pair of
roots is zero.

Following the way Hurwitz pursued in deriving his criterion, one
finds the remarkable presupposition that no purely imaginary roots must
exist. This, however, is precisely the condition whose influence on a
certain Hurwitz-determinant is of special interest. However, for that
reason, one need not abandon this presupposition; it is sufficient to
interpret Dn-1 simply as a prescribed calculation rule which one applies,
without consideration of its derivation, to the coefficients of an equa-
tion of the nth degree.

It will now be expedient to consider (corresponding to the assump-
tion that the real part of a complex pair of roots is to be zero) the
equation of the nth degree as split into two factors: one is a quad-
ratic expression with the purely imaginary pair of roots and the other
is an expression of the (n 2)th degree, thus

(2 X aaon-2 + alk-3 + an-2) =

therein one may put the coefficient a0 = 1, without impairing the gen-
erality; the designation a0 will, however, be retained. For all
remaining coefficients al to an-2, no additional restriction will be
prescribed other than that they are to be real and not all zero (which
would be trivial for this statement of the problem). Multiplication of
the two factors yields

aoXn + an1Xn + (a5 + aao)n-2 + (a3 + aa + n- +

AO Al A2 A3

aan2 = 0


The combined coefficients A one now visualizes as substituted in
the Hurwitz determinant Dn-1, the general term of which has in the

sth column and zth line the subscript x = 2s z. Thus,
A2s-z or, expressed by the coefficients a, the term

NACA TM 1348

one has there

a2s-z + aa2s-z-2

The value of the determinant Dn-1 is now extended by multiplying
all terms of the first line by a8 and all terms of the second line
by al. Furthermore, one makes use of the theorem that a determinant
does not change its value if one adds to the elements of one line the
elements of another line multiplied by an arbitrary number. This is done
by adding to the'terms of the first line, which already have been multi-
plied by aO, the terms of the third, fifth, line multiplied by
a2, a, .; thus, one obtains as the general term of the sth column
in the first line


az-l(a2s-z + aa2s-z-2)

Likewise, one adds to the terms of the second line already multiplied
by a1 the terms of the fourth, sixth, line multiplied by a3,
85, .; one then finds as the general term of the sth column in the
second line


azl(a2s-z + aa2s-z-2)

In order to calculate these sums, it is sufficient to note that one
must generally put ax = 0 when the subscript x is negative; there
then results for a term of the first line

NACA TM 1348

z=l, 3,5,...

a,-l(a2s-z + aS2s-_z2) = aOa2s-1 + a232s-3 + +

a2s-433 + a2s-281 + (3a2s-3 + 8a22s-5 + + a2s-4a1)

and for a term of the second line


az-l(a2s- + 32s-z-2) = al2s-2 + 2s-4 + +

a2s-3a2 + a2s-la0 + (l2s-4 + s + ( + 2- 3ao)

One can see that both sums are of equal magnitude which, however,
does not signify anything else but that the corresponding elements of
the first two lines are equal and that therefore the Hurwitz-determinant
Dn-1 extended by a0al identically disappears:

aoalDn-1 0

From the derivation follows that one could have extended initially,
instead of the first and second line, two arbitrary other odd and even
lines by as_1 and then have proceeded further in the same manner; one
then would have obtained quite generally

even odd n-1

NACA TM 1348

Under the obvious assumption that there always exists such a pair
of values of the coefficients a0 to an-2, the product of which does
not disappear, thus

Dn_-l 0, Q.E.D.

is valid.

3. From the derivation of this proof, one may further conclude that
part of the curve Dn_- = 0 may run entirely in the unstable region;
thus it does not appear there as the required stability limit. Since
such a possibility had been pointed out before, the motivation of this
noteworthy indication should be mentioned.

In splitting the prescribed equation of the nth degree into two
factors, no more detailed data on the coefficients ao to an_2 of
the expression of the (n 2)th degree have been given and accordingly
none regarding the roots of the equation

( ,n-2 -n-i )
ao n-2 + alX n + + an_2 = 0


These roots may therefore have negative as well as positive real
parts in the neighborhood of points of the plane of representation for
which Dn-1 disappears; in other words, they may signify stability and
also instability. If there exists at least one root of the former
expression of the (n 2)th degree with a positive real part, this fact
signifies that the curve Dn-1 = 0 lies in an unstable region.

Figure 3 is to indicate how, for instance, three roots may vary
their position in the complex number plane if their point of reference
in the plane of representation shifts beyond the curve Dn-1 = 0. Thus
Dn_- = 0 is not necessarily always the boundary between two regions
here stability and instability prevail.

ileither does the sign of the coefficient a in the quadratic expres-
sion (X2 + a) play a role in the derivation of the proof. If a is

NACA TM 1348

positive, a purely imaginary pair of roots satisfies the equation
(X2 + a) = 0; this case has been considered in particular. However,
a could just as well be negative without altering the result of the
theorem. This signifies, however, that in the presence of two real
opposite-equal roots of the equation of the nth degree, the Hurwitz-
determinant of the (n l)th degree also disappears.

As figure 4 shows, here also an unstable region lies to both sides
of the curve Dn-l = 0 so that again it does not appear as the sta-
bility limit.

Thus, these examples show that it is indispensable to make sure
whether really all stability conditions on one side of the curve
Dn_1 = 0 have been fulfilled; only then that curve forms together with
the curve an = 0 the boundary against the unstable region.

Translated by Mary L. Mahler
National Advisory Committee
for Aeronautics


1. Price, H. L.: The Lateral Stability of Aeroplanes. Aircr. Engng.,
Vol. 15 (1943), No. 173, 174.

2. Hurwitz, A.: Ueber die Bedingungen, unter welchen eine Gleichung
nur Wurzeln mit negative reellen Teilen besitzt. Mathematische
Annalen XLVI.

NACA TM 1348

Dynamically unstable D

o Point at which
stability prevails

an =0

Statically unstable

Figure 1.- Travel of the point after establishment of stability.

NACA TM 1348



E>0,R> 0
possible J


S\ tatically

Figure 2.- Graphic representation of the region E > 0.


NACA TM 1348

S\\\\\\ s 1
\\\ \\% \ \\% '
,Unstable i L

Dn-l-O "
Unstable ~'
\\\^\\ ^\l

Figure 3.- Change of position of the roots in the complex number plane.

\\ \\\\\
Unstable\ ---
Dn-1, =0 i,
,Unstable \.--
S %

Figure 4.- Unstable region for the curve Dn-1 = 0.

So 5

*' 3 -g' .2g

.0 ed -n
CD Id Cs z
>- c e '
Q a- o oM'1^,
~ S E ^S^

a a iusS-zA S g
" c w a

I 4 U I U0)
u LzJ L. mN 11 .Jr j ^[ ilt*- C.CCI
ma o to W nl a I W A

s ^- 1 Cgje .14 -S -
WLbD f i -E < om-14bfl
: S : aS E V -0 >0

CQ -c L2 PE i2 i5s^ ^Ig ?S I tsi,
1 I- I 2 .1 I
c: 5 i C~ ow0 ,E.,r 4

e w0. Or- < ,;^ 5 <- a,0 9S M*f-s' "
o r a) L C ~, ulI- rE

8i i4i i P I!1 i

Sg~ra wS z c E 9c s bD L-~a a ig s 0a ~a N
z a)z--o a z 1
o4 0 '- 0-4 c
16c. c c cdX
a 9WA >M

c W 1 i z- t o

Sw^ osa "Wn- 4J- 41d-w fl s ~ n4 -d.C i i .a-~ s
gOz6ccM C1.4 | so-u
pwf z u F4 r a
E rN0 qE .c W oU 0 :ctd o
tiI3 w.C bf :; (rD L .C 0 ad I
~ U~ g,.E.0 't u w -3 '4
20 to = W w c:

~~~~ "!" C wE5 N0* C 1q4-p:d1
cc uzz w co flswj- -
V m (A aj o = rO ~m 7

U i LL. 4, CL CU 1 E -W wo..E
o o~
m 0'.0.0 19 m k 11 l
z T
06 Md R w
edrnc M w c o E c 0 w I c

0 0- >t

.|- tS

a zci c

4 4 r" H
I- a :

01W cT i-H

*o S



z -

41 C~

0 .1
S 3
a S
S) 4
I 2

.E .0

C4 9
Sa u

d C C
Cu -C i C -a"
_" S: *Cu S aoa oa

[-4 n4
u1 oog<< a o> *- c "
ad0| .4r.i' .#6

- -- C M ID
!> a a 3!

3n g CCES^
to to 8 0 o flc n
-4 &4 c~aS~

-, t" C .lM [ c (0'
0 to i

Or- a C;

I a 'l| a y
E i~ ri'


I a,
x h

a a

a) 0 0)
co c

o 0 0
o 0


4 -
z zF
ti a) ..

.8 .E


E- 0 E-0

a a a
oa g S
no m t o a;
a, S

a) a )~
, g

a~ 1: a 1

S 2 I S .sa |

z 4 tig h
1 u0


01 a) 0

o0 0! 0 0B,

s S.. s .. 4.

< fl.0 o

S ~ '8 g~ aS s .*,

sra s? p ^

i a isd -l

i 'ig lLl|

E i m i-i d B
Ow l 0 G -0
U2 c fl ; mm

. 0. i.
r .

. s os .c
ca IwP ijJ^|S
-; I l||^i^


4 i C4s
I s v,
^ e4 n E i-iH


*a '

4 S
~I- 0
' e o




O ,

* S
'a a

-0 0
o o

i t i

a 2 V

o 00


5. -
0 .6 C

.0 .0

C) ,! 0
.^ 8'

I11111I IIII 1111 11111 I 111111 III 1111111111111111 III
3 1262 08522 5794