Beruflich Dokumente
Kultur Dokumente
7, JULY 2014
1803
I. I NTRODUCTION
0018-9286 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
(1)
1804
where q(s), r(s), and t(s) are monic polynomials over the
reals; 1 R and 2 R are positive; and 0 deg q(s)
deg r(s) deg t(s). We use the following definition of the
positive QRL.
(2)
l = deg q(s),
d1 = m l,
m = deg r(s),
n = deg t(s)
d2 = n m,
d = d1 +d2 = nl
(3)
(4)
k = 1 k ej + 2 k 2 ej +
(5)
and
(n 1),
(2m l 1),
if
if
d2 ,
> d2 .
= deg t(s),
and t(s) are monic polynomials, m
= deg r(s), n
m.
Note that the ARL notation is consistent with
and d2 = n
the QRL notation, but we use hats to distinguish the ARL
from the QRL. The ARL asymptote rule implies that for i =
1, . . . , d2 and sufficiently large k > 0, d2 roots of pk (s) are ap 1/d2 ji
where i = (2i)/d2 ,
i,k =
k
e + ,
proximated by
= (
t
r )/d2 , and
t and
r denote the sum of the roots
i,k satisfies
of t(s) and r(s), respectively. For i = 1, . . . , d2 ,
Definition 2 with pk (s) replaced by pk (s), where 1 = 1, 2 =
n=n
, m = m,
d2 = d2 , and = (
n
0, = 1/d2 , = i , = ,
1,k , . . . ,
are asymptotic approximations of
1)/d2 . Thus,
d2 ,k
roots of pk (s), and Definition 2 is consistent with the asymptotic approximations given by ARL.
The following fact characterizes the asymptotic properties of
the QRL.
Fact II.4: As k , l roots of pk (s) converge to the roots
of q(s), and the d remaining roots of pk (s) satisfy the following
statements:
(a) Assume d1 < d2 2, and for i = 1, 2, . . . , d, define
i,k = 1d k d eji +
(6)
where
i =
2i
,
d
t q
.
d
(7)
i,k = 1 k d eji + i
(8)
1805
i,k =
k 1e +
2
(18)
where
2i
,
i =
d
(1)i1 2
t q
i =
+j
.
d
1 d
where
(9)
1
2
1
d
2
i,k = 1 k d eji + d1 k d ejd2 i +
(10)
d1d
where
2i
i =
d
(11)
2 (l2 + n2 l n 2md)
t q
+ 2
.
d
2d3 1
(12)
2 4
i,k =
+
2
1
2
2
2
d2 +i,k
k d eji +
1
d1
1
2
4
2
1
2
2
2
d
k e
ji
i =
d1
r q
1
+
d1
d1 22
(19)
(20)
where
i =
2i
,
d2
t r
1
.
d2
d2 22
(21)
i,k =
k 1e +
(22)
2
where
i =
2i
,
d1
r q
d1
(23)
(14)
(15)
+ 2 (t r ) 2 + 22 41
(16)
= 21 (q t )
+ 2 (t r ) 2 22 41
2d1 + 2 d2 2 22 41 .
(24)
where
= 21 (q t )
2d1 + 2 d2 2 + 22 41
2i
,
d1
(13)
where
2i + arg 2 22 41
i =
(17)
i =
2i
,
d2
t r
.
d2
(25)
1806
the roots of pk (s) are symmetric about the real axis. The
third rule of QRL provides conditions such that R is on
the QRL. Specifically, if (a) q()t() < 0; (b) q() = 0 and
r()t() < 0; or (c) w() 0 and q()r() < 0, then is on
the QRL. The fourth rule of QRL is that if d > 0, then d roots of
pk (s) are asymptotically approximated by 1,k , . . . , d,k given
by Fact II.4, where 1,k , . . . , d,k depend on 1 , 2 , d1 , d2 ,
q , r , and t . Finally, the fifth rule of QRL is that R,
where q( ) = 0, is a break-in or breakaway point if and only
(a) 1 ( ) > 0 and d1 ()/d|= = 0; or (b) 2 ( ) > 0 and
d2 ()/d|= = 0.
which maps the real numbers excluding the real roots of q(s)
to the complex numbers. Note that if 1 () > 0 or 2 () > 0,
then there exists k > 0 such that pk () = 0, and thus .
We now present a fact that characterizes the QRL break-in
and breakaway points.
Fact II.5: Let R be on the quadratic root locus, and
assume q( ) = 0. Then, is a break-in or breakaway point if
and only if either of the following statements holds:
(a) 1 ( ) > 0 and d1 ()/d|= = 0.
(b) 2 ( ) > 0 and d2 ()/d|= = 0.
We now summarize the QRL rules given by Facts II.1II.5.
The first rule of QRL is that for k = 0, the roots of pk (s)
start at the roots of t(s). The second rule of QRL is that
Fig. 1. The QRL for pk (s) = k2 q(s) + kr(s) + t(s), where q(s) = s +
10, r(s) = (s + 40)2 , and t(s) = (s + 20)7 , shows that [22.9, 10) is on
the QRL.
Fig. 2. The QRL for pk (s) = k2 q(s) + kr(s) + t(s), where q(s) = s +
10, r(s) = (s + 40)2 , and t(s) = (s + 20)7 , shows that six roots of pk (s)
diverge to infinity along the asymptotes centered at = 21.7 with angles
/6, /2, 5/6, 7/6, 3/2, and 11/6 radians.
1807
Fig. 3. The QRL for pk (s) = k2 q(s) + kr(s) + t(s), where q(s) = s +
10, r(s) = (s + 40)4 , and t(s) = (s + 20)7 , shows that [28.7, 10) is on
the QRL.
Fig. 4. The QRL for pk (s) = k2 q(s) + kr(s) + t(s), where q(s) = s +
10, r(s) = (s + 40)4 , and t(s) = (s + 20)7 , shows that three roots of pk (s)
diverge to infinity along asymptotes centered at 21.7 j16.4 with angles
4/9, 10/9 and 16/9 radians. Furthermore, three roots of pk (s) diverge
to infinity along asymptotes centered at 21.7 + j16.4 with angles 4/9,
10/9 and 16/9 radians.
1808
Fig. 7.
y(s) = G(s)u(s)
(27)
where G(s) = zp (s)/pp (s); u(s) is the input; y(s) is the output; R; and zp (s) and pp (s) are coprime monic polynomials, where deg zp (s) < deg pp (s). Next, consider the control
u(s) =
1
Gk (s) [v(s) y(s)]
(28)
k 2 zc (s)
(s + k)pc (s)
(29)
where
k (s) =
G
k (s) =
G
Fig. 6. The QRL for pk (s) = k2 q(s) + kr(s) + t(s), where q(s) = s +
10, r(s) = (s + 40)5 , and t(s) = (s + 20)7 , shows that four roots of pk (s)
diverge to infinity along asymptotes centered at = 47.5 with angles /4,
3/4, 5/4, and 7/4 radians. Furthermore, two roots of pk (s) diverge to
infinity along asymptotes centered at = 30 with angles /2 and /2
radians.
such that pk () = 0. Part (ii) of Fact II.3 implies that for all
(31.4, 20), there exists two distinct k > 0 such that
pk () = 0. A close-up view of the real axis of the QRL is
shown in Fig. 5.
Next, we examine the asymptotic properties of the QRL.
Since d1 > d2 + 1, part (f) of Fact II.4 implies that four
roots of pk (s) diverge to infinity along asymptotes centered at
= 47.5 with angles /4, 3/4, 5/4, and 7/4 radians.
Furthermore, two roots of pk (s) diverge to infinity along
asymptotes centered at = 30 with angles /2 and /2
radians. The QRL is shown in Fig. 6, which confirms the
asymptote angles.
=
=
k (s)
G(s)G
k (s)
1 + G(s)G
k 2 z(s)
k 2 z(s)
+ kp(s) + (s )p(s)
k 2 z(s)
pk (s)
(30)
r(s) = p(s)
t(s) = (s )p(s).
(31)
(32)
(33)
Fact IV.3 describes the segments of the real axis that are on
the QRL. The proof of Fact IV.3 relies on Fact II.3 and is given
in Appendix B.
Fact IV.3: R is on the quadratic root locus if and only if
either of the following statements hold:
(a) and p()z() < 0.
(b) < , p() = 0 and w() 0.
Furthermore, if R is on the quadratic root locus, then the
following statements hold:
(i) If , p()z() 0 or w() = 0, then there exists
exactly one k > 0 such that pk () = 0.
(ii) If < , p()z() < 0 and w() > 0, then there exists
two distinct k > 0 such that pk () = 0.
Part (a) of Fact IV.3 shows that the real axis to the right of
is consistent with ARL. Specifically, is on the QRL if
and only if lies to the left of an odd number of real roots of
p(s)z(s). In contrast, part (b) of Fact IV.3 shows that the real
axis to the left of differs from ARL. It follows from part (b) of
Fact IV.3 that real roots of w(s) that have odd multiplicity and
lie to the left of are boundary points that separate segments
of the real axis that are on the QRL from segments of the real
axis that are not on the QRL. Furthermore, part (b) of Fact IV.3
implies that real zeros (i.e., real roots of z(s)) can be closedloop poles (i.e., roots of pk (s)) for finite k. These features are
not present in ARL.
We now describe the asymptotic properties of this QRL.
d1 = m l, d2 = n m = 1, and d = d1 + d2 = n l,
which are consistent with (3) and (4) from Section II. Note that
d1 is a nonnegative integer and d2 = 1. Let z R and p R
denote the sum of the roots of z(s) and p(s), respectively.
Fact IV.4 characterizes the asymptotic properties of this QRL
and follows from Fact II.4.
Fact IV.4: As k , l roots of pk (s) converge to the roots
of z(s), and the d remaining roots of pk (s) satisfy the following
statements:
(a) Assume d = 1, and define
1,k = k 2 k +
where
2 (l2 n2 3l + n + 2ln)
+ p z
+
.
d
2d3
1809
where
= arg(
2 4),
2( + p z ) ( + 2 4)
,
=
2 4 + 2 4
2 4)
2( + p z ) (
=
.
2 4 2 4
+ , 2,k = j
+ ,
1,k = j
3,k = k + 2
where
p z + 2
.
2
i,k =
eji +
where
i =
2i
,
d1
p z
d1
and define
d,k = k + .
Then, 1,k , . . . , d,k are asymptotic approximations of
the d remaining roots of pk (s).
Parts (b)(d) of Fact IV.4 differ from ARL. In particular, part
(b) of Fact IV.4 implies that if d = 2, then as k tends to infinity,
two roots of pk (s) diverge through the open-left-half complex
plane with angles and . Parts (c) and (d) of Fact IV.4
imply that if d 3, then as k tends to infinity, one root of pk (s)
diverges to minus infinity and d 1 roots of pk (s) diverge to
infinity along the ARL asymptotes for relative degree d 1.
Specifically, part (d) of Fact IV.4 implies that d 1 roots of
pk (s) diverge to infinity along asymptotes centered at with
angles 1 , . . . , d1 , where and 1 , . . . , d1 are the ARL
center and angles associated with kz(s) + p(s). Part (c) of
Fact IV.4 is similar to part (d) of Fact IV.4, but the asymptote
center is shifted by + 2 /2.
We now present two examples to demonstrate controller
design using the QRL. In ARL, systems where the relative
degree exceeds two are not high-parameter stabilizable, that is,
stable for sufficiently large k > 0. In contrast, QRL can highparameter stabilize minimum-phase systems that are relative
degree one, two or three. Now, we consider the triple integrator,
which is relative degree three.
1810
Fig. 8. The QRL shows that the triple integrator G(s) = 1/s3 is high-gain
k (s) = k2 (s + 10)(s + 15)/((s + 5 + k)(s +
stabilized by the controller G
30)). In fact, the closed-loop system is asymptotically stable for all k > 1, 414.
Fig. 9. The ARL shows that for sufficiently large k > 0, the two closed-loop
poles diverge along asymptotes centered at 5 with angles /2 and /2
radians.
k(s + 5)
[v(s) y(s)]
(s 5)(s + 20)
V. C ONCLUSION
We presented rules for constructing the QRL for a polynomial that is quadratic in k. These QRL rules describe: (i) the
behavior of the roots for small k; (ii) the symmetry of roots
about the real axis; (iii) the segments of the real axis that are on
Fig. 10. The QRL shows that for sufficiently large k > 0, two closed-loop
poles diverge along asymptotes centered at 5 j8.66 with angles 2.09
radians, respectively.
the root locus; (iv) the behavior of roots for large k; and (v) the
break-in and breakaway points.
We provided a controller synthesis using a control architecture that yields a closed-loop denominator polynomial that
is quadratic in k. The QRL rules for this control architecture
are derived from the general QRL rules in Section II. With
this architecture, we designed a controller that high-parameter
stabilizes the triple integrator.
The QRL rules can be implemented on a computer to generate QRL plots. Specifically, the QRL for pk (s) can be generated
using 1 , 2 , q(s), r(s), t(s), and w(s).
The techniques of this paper could be extended to address
cubic or quartic root loci. For example, the specialized cubic
root locus results in [22] could be extended using the techniques
in this paper. The general QRL rules are developed using the
quadratic discriminant but do not explictly use the solution to
the quadratic equation. Similarly, [22] suggests that a cubic root
locus could be developed using only the cubic discriminant
rather than the solution to the cubic equation. Moreover, [22]
shows that the cubic root locus can be high-parameter stabilizing for minimum-phase systems with relative degree less than 5.
Nevertheless, general rules for a cubic or quartic root locus is
an open problem.
A PPENDIX A
P ROOFS OF FACTS II.1II.5
Proof of Fact II.1: If k = 0, then pk (s) = t(s).
Proof of Fact II.2: Since for all k > 0, pk (s) has real coefficients, it follows that the roots of pk (s) are either on the real
axis or occur in complex conjugate pairs.
Proof of Fact II.3: First, we show that (a), (b) or (c) is
necessary for R to be on the QRL. Assume R is
on the QRL and consider four cases: (A) q()t() < 0; (B)
q() = 0; (C) t() = 0; and (D) q()t() > 0. First, assume
(A) q()t() < 0, which implies (a). Next, assume (B) q() =
0, which implies from (1) that pk () = 2 kr() + t(). Since
1811
1812
(34)
(35)
(36)
where a1 , a2 , . . . , al , b1 , b2 , . . . , bm , c1 , c2 , . . . , cn R. Note
lk = 1 k ej + 2 k 2 ej +
2 k e + 22 k 2 ej + 2 k (l2)
l
l lh j(lh)
+
e
h 1
h=3
h
2 k 2 ej + k (lh)
(37)
where
l
h
+
1mh ej(mh)
h
h=2
h1 (mh)
2 k 2 ej +
k
(40)
n
n jn n
k = 1 e k
2 k e + 22 k 2 ej + 2 k (n2)
n
n nh j(nh)
+
e
h 1
h=3
h
2 k 2 ej + k (nh)
(41)
n1
n1 (n1) j(n1)
e
k = 1 k
n1
n 1 nh j(nh)
+
1 e
h
h=3
h1 (nh)
2 k 2 ej +
k
.
(42)
Next, since n = l + d, note that
2 + 2l/d =
2 k 2 ej + k (l2)
l1
l 1 lh j(lh)
+
1 e
h
h=3
h1 (lh)
2 k 2 ej +
k
m
m jm m
k = 1 e k
m mh j(mh)
+
e
h 1
h=2
h
2 k 2 ej + k (mh)
1/d
1 k 2 q(i,k ) = 1d ejli k
2n
d
(n1)
+ (l + a1 )1 d ej(l1)i k
(2n3)
+O k d
(2n3)
2 kr(i,k ) = O k d
(n1)
d
ej(n1)i k
(45)
(46)
(2n2)
d
(47)
(39)
(2n2)
d
2n
d
+ (n + c1 )1
(2n3)
+O k d
(38)
(44)
t(i,k ) = 1 ejni k
(43)
n
d
2n
.
d
2n
d
(2n2)
d
+
ej(l1)i k d
1
l + a1 + ejdi (n + c1 )
(2n3)
.
+O k d
(48)
For i = 1, . . . , d, it follows from (7) that 1+ejdi = 0. Furthermore, since a1 c1 = t q and d = nl, it follows from
(7) that l + a1 +ejdi (n+c1 ) = a1 c1 d = 0. Thus, (48)
implies that for i = 1, . . . , d, pk (i,k ) = O(k (2n3)/d ). Therefore, for i = 1, . . . , d, limk pk (i,k )/k (2n2)/d = 0. Thus,
Definition 2 implies that 1,k , . . . , d,k are asymptotic approximations of roots of pk (s), which confirms (a).
To show (b), assume d1 = d2 2, and let 1,k , . . . , d,k
be given by (8) and (9). Since d1 = d2 2, it follows that
d + 2m = n + m + d1 = n + m + (d2 2) = 2n 2, which
implies that
k
1+2m/d
=k
2(n1)
d
1813
1 k 2 q(i,k ) = 1d ejli k
(49)
(51)
ej(n1)i (ni + c1 )k
2(n2)
.
+O k d
(52)
2n
d
n
d
t(i,k ) = 1 ejni k
2(n2)
d
2n
d
(2n2)
d
(57)
2 (n1)
d
+ n
d1
1
d1d
2(n1)
d
ej(l1)i
(2n1)
d
2 +m1)i
ej(d
(2n3)
k
+O k d
+ 1
ej(l1)i k
(1d2 )
li + a1 + 1 d 2 ej(d1 +1)i
ej(d2 +n1)i k
(d1 +1)
d
li + a1 + 1
2 ej(d1 +1)i + ejdi (ni + c1 )
1
= a1 c1 + j(1)i1 2 /12 di
ej(n1)i
(n1)
d
(l2)
n(n 1)1 d 22 j(2d2 +n2)i
+
e
2d2
(2n3)
(2n2)
k d +O k d
.
(58)
(2n1)
d
+ (n + c1 )1
(53)
= 0.
(n1)
d
22 (m1)
d
+ m
d1
1
d1d
2(n1)
d
(n1)
d
2 kr(i,k ) = 1d 2 ejmi k
(n1)
d
(2n1)
d
(l2)
l(l 1)1 d 22 j(2d2 +l2)i
+
e
2d2
(2n3)
(2n2)
k d +O k d
,
(56)
(50)
2n
d
+ 1
n
d
ej(d2 +l1)i k
2(n1)
ej(l1)i (li + a1 )k d
2(n2)
+O k d
2(n2)
m
2(n1)
2 kr(i,k ) = 1d 2 ejmi k d +O k d
t(i,k ) = 1 ejni k
+ (l + a1 )1
(n1)
d
n
d
2n
d
2 (n1)
d
+l
d1
1
d
d1
2n
d
+ 1
(55)
1/d
1/d
1 k 2 q(i,k ) = 1 ejli k
(2n1)
d
k 1+2m/d = k
2n
d
1814
22
(l2 + n2 l n + 2mdejdi )
2d2 1
2
= d + a1 c1 + 22 (l2 + n2 l n 2md)
2d 1
= 0.
(60)
(61)
k 1+2m/d = k d .
1 k 2 q(i,k ) = 1 || d1 ejli k
2n
d
+ 1
(a1 + l)||
2(n2)
+O k d
2 kr(i,k ) = 2 ||
m
d1
ejmi k
(l1)
d1
t(i,k ) = ||
ejni k
2(n1)
d
(62)
2n
d
+ 2
(b1 + m)||
2(n2)
+O k d
n
d1
ej(l1)i k
(m1)
d1
ej(m1)i k
2(n1)
d
(63)
2n
d
(n1)
d1
+ (c1+ n)||
2(n2)
.
+O k d
ej(n1)i k
2(n1)
d
(64)
pk (i,k ) = d1 [1 + 2 + 2 ]k
(l1)
2n
d
2(n1)
+ d1 k d
1 (l + a1 ) + 2 (m + b1 ) + 2 (n + c1 )
2(n2)
.
(65)
+O k d
Since = (1/2)(2 + 22 41 ), it follows that 1 +
2 + 2 = 0. Furthermore, since a1 = q , b1 = r and
c1 = t , it follows from (16) that
1 (l + a1 ) + 2 (m + b1 ) + 2 (ni + c1 )
= (l1 + m2 + n2 ) + (1 a1 + 2 b1 + 2 c1 )
= 0.
(66)
Thus, (65) and (66) imply that for i = 1, . . . , d1 , pk (i,k ) =
O(k (n2)/d1 ). Therefore, for i = 1, . . . , d1 , limk pk (i,k )/
k 2(n1)/d = 0. Thus, Definition 2 implies that 1,k , . . . , d1 ,k
are asymptotic approximations of roots of pk (s).
k 2+l/d1 = k 1+m/d1 = k
(2ml)
d1
(67)
k d1 = k
(2ml1)
d1
(68)
2
(2ml2)
+ O k d1
(69)
dm
1 1 jmi (2ml)
2 kr(i,k ) = 2
e
k d1
2
(m1)
d1
1
+ 2 (m + b1 )
ej(m1)i
2
(2ml2)
(2ml1)
(70)
k d1
+ O k d1
dn
1 1 jni (2ml1)
t(i,k ) =
e
k d1
2
(2ml2)
.
(71)
+ O k d1
Therefore, it follows from (1) and (69)(71) that for i = 1,
. . . , d1
dl
(2ml)
1 1 jli
pk (i,k ) = 1
e [1 + ejd1 i ]k d1
2
(l1)
1 d1 j(l1)i (2ml1)
+ 1
e
k d1
2
l + a1 + (m + b1 )ejd1 i
(d+1)
d1
1
1
j(d+1)i
+1
e
2
(2ml2)
.
(72)
+ O k d1
For i = 1, . . . , d1 , it follows from (19) that 1 + ejd1 i = 0.
Since d + 1 = d1 + d2 + 1 = 2d1 , it follows that (d + 1)/
d1 = 2. Furthermore, since a1 b1 = r q and for i =
1, . . . , d1 , ej(d+1)i = ej(4i2) = 1, it follows from (19) that
(d+1)
d1
1
ej(d+1)i
2
= a1 b1 + 1 /22 d1 = 0.
l + a1 + (m + b1 )ejd1 i + 11
(73)
k 1+m/d2 = k d2 .
(n1)
d2
(74)
1815
(75)
2
1/d
2 2 ,
(n1)
d2
ej(m1)i (m + b1 )k
(n2)
+ O k d2
+ 2
t(d1 +i,k ) = 2 e
(n1)
d2
(77)
ej(n1)i (n + c1 )k
(n2)
.
+ O k d2
[1 + e
(78)
jd2 i
]k
n
d2
+ 2
ej(m1)i
(1d)
1 2 d2 ej(1d)i + m + b1
+ (n + c1 )ejd2 i k
+O k
(l1)
d
1
ejli
ej(l1)i
+ 1
(n1)
d2
(79)
1 2 d2 ej(1d)i + m + b1 + (n + c1 )ejd2 i
= b1 c1 + 1 /22 d2 = 0.
m
d1
1
2
(2ml)
d1
(m1)
d
1
(82)
ejmi
ej(m1)i
(80)
(83)
(84)
(n2)
d2
(2ml)
d1
(m + b1 )k d1
(2ml2)
+ O k d1
(2ml2)
.
t(i,k ) = O k d1
(n1)
d2
1
2
(2ml1)
(n1)
d2
1
2
+ 2
(n1)
d2
jmi
2 kr(i,k ) = 2
n
d2
+ 2
n
d2
(2ml1)
jni
dl
(l + a1 )k d1
(2ml2)
+ O k d1
(76)
n
d2
1
2
+ 1
1 k q (d1 +i,k ) = 1 2 e k
(n2)
+ O k d2
n
1 k q(i,k ) = 1
(n1)
d2
jli
(81)
1
2
dl
1
2
ejli (1 + ejd1 i )k
(l1)
d
1
(2ml)
d1
ej(l1)i
(2ml1)
l + a1 + (m + b1 )ejd1 i k d1
(2ml2)
.
(85)
+ O k d1
For i = 1, . . . , d1 , it follows from (23) that 1+ejd1 i = 0. Furthermore, since a1 b1 = r q , it follows from (23) that l+
a1 +(m+b1 )ejd1 i = a1 b1 d1 = 0. Thus, (85) implies
that for i = 1, . . . , d1 , pk (i,k ) = O(k (2ml2)/d1 ). Therefore,
for i = 1, . . . , d1 , limk pk (i,k )/k (2ml1)/d1 = 0. Thus,
Definition 2 implies that 1,k , . . . , d1 ,k are asymptotic approximations of roots of pk (s).
We now show that d1 +1,k , . . . , d,k are asymptotic approximations of roots of pk (s). Since d1 > d2 + 1, it follows that
2d2 + l d1 + d2 + l 2 = n 2.
(86)
1816
1/d
ej(m1)i (m + b1 )k
(n2)
+ O k d2
+ 2
n
d2
t (d1 +i,k ) = 2 k
n
d2
(n1)
d2
(88)
jni
(n1)
d2
ej(n1)i (n + c1 )k
(n2)
.
+ O k d2
+ 2
(n1)
d2
(89)
(n1)
+ 2 d2 k d2 ej(m1)i
m + b1 + (n + c1 )ejd2 i
(n2)
.
+ O k d2
(90)
A PPENDIX C
R ESULT ON Q UADRATIC P OLYNOMIALS
Lemma 1: Consider the polynomial a(k) = a2 k 2 + a1 k +
a0 , where a2 , a1 , a0 R, a2 = 0, and a0 = 0. Then, the following statements hold:
(a) a(k) has exactly one positive root if and only if
a2 a0 < 0.
(b) a(k) has repeated positive roots if and only if a21
4a2 a0 = 0 and a2 a1 < 0.
(c) a(k) has two distinct positive roots if and only if a21
4a2 a0 > 0, a2 a1 < 0 and a2 a0 > 0.
R EFERENCES
[1] W. R. Evans, Graphical analysis of control systems, AIEE Trans.,
vol. 67, pp. 547551, 1948.
[2] W. R. Evans, Control system synthesis by root locus method, AIEE
Trans., vol. 69, pp. 6669, 1950.
[3] W. R. Evans, Control System Dynamics. New York: McGraw-Hill, 1954.
[4] A. M. Krall, The root locus method: A survey, SIAM Rev., vol. 12, no. 1,
pp. 6472, 1970.
[5] W. Krajewski and U. Viaro, Root-locus invariance, exploiting alternative
arrival and departure angles, IEEE Control Syst. Mag., vol. 27, no. 1,
pp. 3643, 2007.
[6] A. M. Eydgahi and M. Ghavamzedeh, Complementary root locus revisited, IEEE Trans. Education, vol. 44, no. 2, pp. 137143, 2001.
[7] M. L. Nagurka and T. R. Kurfess, Gain and phase margins of SISO
systems from modified root locus plots, IEEE Control Syst. Mag., vol. 12,
no. 3, pp. 123127, 1992.
[8] R. W. Brockett and C. I. Byrnes, Multivariable nyquist criteria, root
loci, and pole placement: A geometric viewpoint, IEEE Trans. Autom.
Control, vol. 26, no. 1, pp. 271284, 1981.
[9] H. Kwakernaak, Asymptotic root loci of multivariable linear optimal
regulators, IEEE Trans. Autom. Control, vol. 21, no. 3, pp. 378382,
1976.
[10] R. W. Brockett and C. I. Byrnes, On the algebraic geometry of the
output feedback pole placement map, in Prod. IEEE Conf. Dec. Control,
Fort Lauderdale, FL, 1979, pp. 754757.
[11] C.-T. Chen, Linear System Theory and Design, 3rd ed. Oxford, U.K.:
Oxford Univ. Press, Inc., 1998.
[12] S. Sastry and C. Desoer, Asymptotic unbounded root loci-formulas and
computation, IEEE Trans. Autom. Control, vol. 28, no. 5, pp. 557568,
1983.
[13] S. S. Broussev and N. T. Tcharmov, Time-varying root-locus of largesignal LC oscillators, IEEE Trans. Computer-Aided Design Integr.
Circuits Syst., vol. 29, pp. 830834, 2010.
1817