Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

The Numerical Range of 3 X 3 Matrices: Dennis S. Keeler

Download as pdf or txt
Download as pdf or txt
You are on page 1of 25

View metadata, citation and similar papers at core.ac.

uk brought to you by CORE


provided by Elsevier - Publisher Connector

The Numerical Range of 3 x 3 Matrices

Dennis S. Keeler*
Department of Mathematics
Miami University
oxford, Ohio 45056

and

Leiba Rodman+ and Ilya M. Spitkovsl$


Department of Mathematics
College of William and Mary
Wi&amsburg, Virginia 23187-8795

Submitted by Hans Schneider

ABSTRACT

Let A be an n X n complex matrix. Then the numerical range of A, W(A), is


defined to be {r*Ax : x E C”, x*x = 1). In this article a series of tests is given,
allowing one to determine the shape of W(A) for 3 X 3 matrices. Reconstruction of
A, up to unitary similarity, from W(A) is also examined. 0 Elsevier Science Inc.,
1997

1. INTRODUCTION

Let A be an n X n matrix with complex entries. The numerical range of


A, W(A), is defined as {x*Ax : x E C”, x*x = 1). It is well known (see [l,
21) that W(A) is a convex compact subset of @, which contains all the

* Supported by a Research Experience for Undergraduates grant from tbe National Science
Foundation, Grant number 9300395.
’ Partially supported by the NSF Grant DMS 9123841.
’ Partially supported by the NSF Grant DMS 9401848.

LZNEAR ALGEBRA AND ITS APPLZCATZONS 252:115-139 (1997)

0 Elsevier Science Inc., 1997 0024-3795/97/$17.00


655 Avenue of the Americas, New York, NY 10010 SSDI 0024-3795(95)00674-5
116 DENNIS S. KEELER ET AL.

eigenvalues of A and therefore its convex hull conv(a( A)). For A normal
(that is, commuting with A*), W(A) = co&u ( A)); the converse statement
holds when n < 4.
For 2 X 2 matrices A a complete description of the numerical range
W( A) is well known. Namely, W(A) is an ellipse with foci at the eigenvalues
A,, A, of A and a minor axis of the length

s = (trace( A*A) - (h,(’ - (A,12)1’2. (1.1)

Of course, s = 0 for normal A, and the ellipse in this case degenerates into a
line segment connecting A, with h,. On the other hand, for 2 X 2 matrices
A with coinciding eigenvalues the ellipse W(A) degenerates into a disk.
For general n, the following procedure is useful: Write A = H + iK with
H, K Hermitian, and let

L,(u,u,w) = det(uH + UK + WI).

The equation L,(u, v, w> = 0, with u, v, w viewed as homogeneous line


coordinates, defines an algebraic curve of class n. The real part of this curve
we denote by C( A) and call the associated curve (the randerzeugende curve,
in German terminology of [3]) of A. Th e n real foci of C(A) correspond to
the eigenvalues of A [3, 41 and W(A) = conv C(A) [3]; see [5, Sect. 31 for a
detailed discussion of the connections between the polynomial L, and the
numerical range of A. Note that the usual point equation f(x, y) = 0 of the
curve C(A) also can be written down (see [6]), but for n > 2 it is much more
complicated than the line equation.
For n = 3, the following classification, based on factorability of L,, was
given by Kippenhahn in [3]:

Case 1. L, factors into three linear factors. Then C(A) consists of three
(not necessarily distinct) points, A is normal (and therefore reducible’), and
W(A) is the convex hull of its eigenvalues.

Case 2. L, factors into a linear factor and a quadratic factor. Then C( A)


consists of a point A, (the eigenvahre of A corresponding to the linear factor)

’ We say that a matrix A is reducible if there exists a unitary matrix U such that
U*AU = diag[A,, A,], where both diagonal blocks are of nonzero size. For reducible A,
W(A) = conv(W(A,), W(A,)).
NUMERICAL RANGE 117

and an ellipse E. The numerical range W(A) is either an ellipse (if A,, lies
inside E) or a “cone-like” figure otherwise; in the latter case A is reducible
(but not normal).

In the next two cases the polynomial L, (and therefore the matrix A) is
irreducible.

Case 3. The degree of C(A) (that is, the degree of its point equation)
equals 4. Then C(A) has a “double tangent,” and the boundary of W(A)
contains one flat portion but no angular points.

Case 4. The degree of C(A) equals 6. Then C(A) consists of two parts,
one inside another; an outer part (and therefore W(A)) has an “ovular”
shape.

This classification is complete: the same article [3] contains examples of


matrices A falling in each of the above-mentioned cases.
However, it does not provide a constructive procedure that would allow
one to determine, for an arbitrary given nonnormal 3 X 3 matrix A, the
shape of its numerical range. The main purpose of our article is to offer a
series of tests, in terms of a matrix A itself or its canonical unitarily
equivalent forms, to determine when W(A) is an ellipse, a set with a flat
portion on its boundary, or an ovular set. This is done in Sections 2 and 3.
The results obtained simplify dramatically for matrices with one-point spec-
trum; this is discussed in Section 4. Section 5 is devoted to the question of
when a 3 X 3 matrix A can be restored (of course, up to unitary equivalence)
from its numerical range.

2. W(A) IS AN ELLIPSE

We begin with the following general result concerning matrices of arbi-


trary size whose associated curve consists of ellipses and points.

THEOREM 2.1. Let A be an n X n matrix with eigenvalues A,, . . . , A,


and suppose that its associated curve C(A) consists of k ellipses, with minor
118 DENNIS S. KEELER ET AL.

axes of lengths sl, s2, . . . , sk, and n - 2k points. Then

i$ls: = trace( A*A) - 2 (hi12. (2-I)


i=l

For n = 2 the condition imposed in Theorem 2.1 is of no restriction since


C(A) always is an ellipse or a pair of points; formula (2.1) in this case takes
the form (1.1). In the original version of this article we considered the case
n = 3, in which, according to Kippenhahn’s classification, the conditions of
Theorem 2.1 are satisfied for C( A) being a union of three points (Case 1) or
an ellipse and a point (Case 2). Still there is not more than one ellipse, and
(2.1) in this case takes the form

s = (trace( A*A) - [A,)’ - l&l2 - l~sl~)“~. (2.2)

The generalization to the case n > 3 and its proof were suggested to us by
the referee.

Proof. Relabel the eigenvalues of A in such a way that A2i_ i, hsi


become the foci of the ith ellipse (i = 1,. . . , k), and hskfl, . . . , An--the
remaining points of C(A).
Along with A, consider the matrix

Since C( A) = C(B), the polynomials L, and L, have to be the same.


Compute now the coefficients of w”- 2 of these polynomials. When doing
that, due to unitarily invariance of L,, we may without loss of generality
suppose that A = [ajj]y j= 1 is in upper-triangular form. The coefficient of
wn- 2 in L, equals the sum of all 2 X 2 principal minors of uH + UK, that
is,2

=
l<i<j<n I u%aii + u3aii)(uiRajj + v3ajj) - f(U’ + U2)laij12
1
u%Ai + uZYA~)(U%A~ + uDAj) - f(u2 + u2)Iaij12 .
1
’ Here and in what follows, we denote by !Rz and 32 the real and imaging part,
respectively, of a complex number z.
NUMERICAL RANGE 119

Applying this formula to B (which already is in upper-triangular form) we


obtain

c [(t&hi + U3Ai)(U%Aj + U3Aj)] - i(u’+2) 5 ($12. i=l


l<i<jbn

Since L, = L,, it follows from here that

5 Isi12 = c laij12 = 5 laij12 - k la,,(’ = trace( A*A) - t lhi12.


i=l l<i<j<n i,j=l i=l i=l

Note that in the setting of Theorem 2.1 $1 the respective coefficients of


L, and L, are equal. In particular, equating the coefficients of u”, on yields

det H = fi %A2i_1%Ah,i - is:


i=l (
(2.3)

If n = 3 and A is in upper-triangular form

(2.4)

conditions (2.3) can be rewritten as ~x~~~c + Iy12iRb + lz12’iRa - %(xijz)


= s2!RA3, 1~1~3~ + ly123b + lz123a - 3(xijz) = s2DAhS, or simply

lx12c+ I y12b + lz12a - x& = s2A,. (2.5)

Due to (2.2),

s = 41x12+ I y12 + lz12. (2.6)


120 DENNIS S. KEELER ET AL.

Hence, conditions (2.5), (2.6) are necessary for matrices (2.4) and

1
A, s 0

B= 0
[ 0 0
h, 40

to have the same associated curves. Surprisingly, a direct computation shows


that they are also sufficient. Therefore, the following criterion holds.

THEOREM 2.2. LA A be in upper-triangularform (2.4). Then its associ-


ated curve C(A) consists of an ellipse ( possibly degenerating to a disk) and a
point if and only if

1. d = [xl2 + 1~1’ + Iz12 > 0 and


2. the number A = (~1x1~ + bl y12 + alz12 - xijz)/d coincides with at
least one of the eigenvalues a, b, c.

Zf these conditions are satisfied, then C(A) is the union of A with the ellipse
having its foci at two other eigenvalues of A and minor axis of length s = 6.

To obtain a unitary invariant form of Theorem 2.2, note that

cIx12+ bly12+ alz12 - xijz = d trace A - trace( A*A’) + i lh,(2h,.


j=l

Therefore, Theorem 2.2 admits the following reformulation.

THEOREM 2.3. The associated curve C(A) of a 3 X 3 matrix A consists


of an ellipse and a point if and only if

1. d = trace(A*A) - C~=rlAj12 > 0 and


2. the number A = trace A + (l/dXC~= ,lAj12Aj - trace(A*A2)) coin-
cides with at least one of the eigenvalues Aj of A.

Zf these conditions are satisfied, then C(A) is the union of A with the
ellipse having its foci at two other eigenvalues of A and minor axis of length
s = 0.

Now we are ready to formulate a necessary and sufficient condition for a


3 X 3 matrix A to have an ellipse as its numerical range.
NUMERICAL RANGE 121

THEOREM 2.4. LA A be a 3 x 3 matrix A with the eigenvalues Aj,


j = 1,2,3. Then W(A) is an ellipse if and only if conditions 1,2 of Theorem
2.2 (or 2.3) hold and, in addition,

3. (IA, - h,l + IA, - X,1)‘- - IA, - &I2 < d, where the eigenvalue coin-
ciding with A is labeled A,.

Proof. Conditions 1,2 are equivalent to C(A) being a union of the


ellipse E (with the foci A,, A, and minor axis of length 0) and the point A,.
Condition 3 means that A, lies inside E. According to Kippenhahn’s classifi-
cation, this is the only case when W(A) is an ellipse. ??

The results obtained allow us to describe all 3 x 3 matrices for which


W(A) is a disk.

COROLLARY 2.5. W(A) is a disk ifand only $

1. A has a multiple eigenvalue p (so that its eigenvalues equal p, CL, and
A)
2. 2ptrace(A*A) = trace(A*A2) + 2l/~‘l/~ + (2~ - A)lA12, and
3. 4) p - AI2 + 21~)~ + 1Al2 < trace( A*A).
For A in a triangular form (2.41, conditions 2 and 3 may be substituted by

2’. GJz = (p - AX6_(x12 + 6b,P(y(2 + 8,_1~1~>, where 6 is a usual


Kronecker symbol, and

3’. 41 p - Al2 < (xl2 + ( y12 + 1~1~.

If these conditions are satisfied, then W(A) is centered at p and has radius

i trace( A*A) - 21 k12 - IAl (= a~/lxl” + ly12 + 1~1 in the case (2.4)).

Proof. Indeed, W(A) is a disk if and only if it is an ellipse (that is,


conditions of Theorem 2.4 are satisfied) and, in addition, the foci of this
ellipse coincide. This means that A has a multiple eigenvalue (say, ~1, and its
third eigenvalue coincides with A defined by condition 2 of Theorem 2.2 or
Theorem 2.3. The rest is computation. ??

For triangular matrices, this corollary was first proved directly by Chien
and Tam, although in a very different manner [7]. Necessary and sufficient
conditions for W(A) to be the unit disk centered at 0 for a 3 X 3 matrix A
were obtained earlier by N. K. Tsing (unpublished) and stated in [7]; these
conditions appear as a particular case of Corollary 2.5, when we specialize
a = b = p = 0, (xl2 + Iy12 + Iz12 = 4.
122 DENNIS S. KEELER ET AL.

3. W(A) HAVING A FLAT PORTION ON ITS BOUNDARY

Throughout this section we assume that A is a 3 X 3 irreducible matrix


represented as A = H + iK with H and K Hermitian.
We begin by deriving a canonical form for an irreducible matrix with a flat
portion on the boundary of its numerical range.

THEOREM 3.1. Let A be a 3 X 3 irreducible matrix. Then afer unitary


similarity, translation, rotation, and scaling of A, A may be written in the
f O?Tl

where cl, cg, %( t> are positive, if and only if W( A) has a flat portion on its
boundary. In this form, W(A) has ajlat portion extending from 0 to i and is
contained in the closed right half-plane.

Proof. Let W(A) h ave a flat portion on its boundary. After rotation,
shifting, and scaling (by scaling we mean multiplication by a positive number),
we may assume that a flat portion stretches from 0 to i. Since W(A) is
convex, it must be contained entirely in the right or the left half-plane.
Applying yet another rotation and translation, if necessary, we may assume
that W(A) is in the right half-plane.
Since 0 and i are in W(A), there exist r,, x2 E C”, x:x1 = x2*x2 = 1
such that x:kr, = 0, x$kr, = i. Let 9= Spanix,, x2}. Since _!Z is a
2dimensional subspace, we may represent the linear transformation of A
restricted to 9, Al_Y = A’, by a 2 X 2 matrix. By choosing a proper basis for
A, A’ is the leading principal submatrix of A.
Now W( A') is an ellipse, as are the numerical ranges of all 2 X 2
matrices. Since W( A’) is convex, [0, i] c W( A’). Also, W( A’) E W(A). Since
W(A) does not extend into the left half-plane, the only possible ellipse

i
W( A’) can be is the degenerate ellipse [O, il. This implies that A’ is normal
with eigenvalues 0 and i. So with proper basis

1
0 v1

r
va.

1
o

A’= 1, 0 and A= [ cl0 c2


0 t
NUMERICAL RANGE 123

Since W( A) is in the closed right half-p1 ane, H is positive semidefinite. A


calculation shows that

1
-
Vl + Cl
-
v2 + c2 >

2%(l)

and therefore !R( 5) > 0, vi = - G, v2 = - G. Due to the irreducibility of


A, however, we know (ri([) > 0. By a diagonal unitary similarity, we’ may
assume that ci, c2 are nonnegative. If cr or c2 are 0, then A is reducible, so
ci and c2 are positive. We now have the form (3.1) of A we had hoped for.
Now suppose that after scaling, rotation, translation, and unitary similar-
ity, A is in the form expressed in the theorem. Consider the principal
submatrix A’ from the first two rows and columns of A. W( A’) is a line
segment from 0 to i. Clearly, W( A’) c W(A). But since H is positive
semidefinite, W(A) lies entirely in the right half-plane. So the line segment
from 0 to i must be on the boundary of W(A).
To see that the line segment does not go beyond 0 or i, note that any
point y on that line must be pure imaginary. So if r*Ax = x*Hx + ix*Kx =
y, then x*Hx = 0. Hence, x E Ker(H) = Span{[l, 0, OIT, [0, 1,OIT}. If ]]x]]
= 1, then x = v,[l,O,OIT + v,[O, 1,OlT with lv112+ lv212= 1, and 0 <
x*zzx = ]vJ < 1. ??

According to Kippenhahn’s classification, W(A) has a flat portion on


the boundary if and only if there exists a line, ux + uy + w = 0, tangent to
C(A) at two distinct points. This double tangent line corresponds to an eigen-
value -w of uH + UK, which has multiplicity 2; since u and v are real,
uH + UK is Hermitian and uH + UK + WI has rank 1. Conversely, if uH +
UK + WI has rank 1, then -w is an eigenvalue of uH + UK with multiplicity
2, and we get a double tangent. Observe also that, if A is irreducible,
uH + UK cannot have an eigenvalue of multiplicity 3 (for then the Hermitian
matrix uH + UK would be scalar, H and K would commute, and hence A
would be normal). We summarize:

PROPOSITION 3.2. Let A = H + iK be irreducible. Then the following


statements are equivalent:

1. W(A) has a flat portion on the bounda y;


2. rank(uH + UK + wZ) = 1 for some real u, v, w;
3. for some real u, v not both equal to zero, uH + UK has a multiple
eigenvalue.
124 DENNIS S. KEELER ET AL.

Under these conditions, the flat portion of the boundary lies on the line
ux + “fj + w = 0.

It also follows from Kippenhahn’s classification that an irreducible 3 X 3


matrix can have at most one flat portion on the boundary of its numerical
range.

COROLLARY 3.3. Let A be irreducible and unitarily similar to a real


matrix. Then W( A) has a flat portion on its boun& y if and only if H has a
multiple eigenvalue. lf W(A) does have a flat portion, it is parallel to the
imagina y axis.

Proof. If A is a real matrix, then W(A) is symmetric about the real axis.
So, the (unique) flat portion of the boundary of W(A) must be a vertical line.
According to Proposition 3.2, it happens if and only if the matrix 1 * H + 0 *
K = H has a multiple eigenvalue. ??

if A is not unitarily equivalent to a real matrix, Proposition 3.2 may be


difficult to use. We now present several statements, equivalent to Proposition
3.2 (and obtained from it>. As we see in Sections 4 and 5, these statements
are sometimes more suitable for application than Proposition 3.2 itself.

COROLLARY 3.4. Let an irreducible matrix A be written in the form

A= [$ z 5]+i[ $ g ::I. (3.2)

Then WC A) has a flat portion on its boundary if and only if there exist real
u,u not both zero such that

Iuh,, + r&,,l = Iu& + &,I = Iuh,, + uh,,l (3.3)

and

(uh,, + vuk,s) (uh,, + ukJ (uhss + r.k2s) is real. (3.4)


NUMERICAL RANGE 125

Proof. Since for any l E C W( A) h as a flat portion on its boundary if


and only if W(A - 51) h as a flat portion on its boundary, we may (and do)
assume that trace of A is zero: 5 = Q = 0. The characteristic polynomial of
uH + VK then equals

det(uH + UK - AZ)

= + IUhz3+ VkJ + Iuh,, + vk,,12)h


-A3 + (Iuh,, + vk1,12

+ 2%((uh,, + vk,,)(uh,, + vw(uh,, + vk,,)). (3.5)

According to Proposition 3.2, W(A) h as a flat portion on its boundary if


and only if for some real u, v not both zero uH + UK has a multiple
eigenvalue; that is, the discriminant of (3.5) equals zero. For an arbitrary
third-degree polynomial a, x3 + a, x2 + a, x + u3, the discriminant is
2 2
%% - 4u,u; - 4&z, - 27~:~: + 18u,u,u,u,; see, e.g., [8]. Due to our
assumption trace A = 0 the coefficient of A2 in (3.5) vanishes, and direct
computations show that the discriminant of (3.5) equals four times the
expression

(142 + vkl,12+ luh,, + t&l2 + It& + ~li,,l”)~

- 27(W((uh,, + vk,,)(uh,, + vk,,)(u~ + &))I”. (3.6)

Setting

x = uh,, + v/l,,, y = uh,, + uk,, and z = uh,, + vii,,, (3.7)

we see that (3.6) equals zero if and only if

M2 + ly12+ Id2
=“JM. (3.8)
3

Id2 + ly12 + lz12


3
&&iv >v-, (3.9)
126 DENNIS S. KEELER ET AL.

the equality (3.8) h o Id s exactly when both inequalities in (3.9) are actually
equalities, that is, when I xl = I y 1= Iz 1(for the first inequality in (3.9)) and
xZjz E R (for the second). The two conditions obtained are exactly the same
as (3.3) and (3.41, respectively. ??

Note that condition (3.2) is of no restriction, since it can always be


obtained by a unitary similarity. To rewrite Corollary 3.4 in a unitarily
equivalent form, put

= uH + uK - $trace(uH + uK))Z.

Then, in notation (3.7), 1x1’ + 1~1’ + lz12 = i trace B2 and !R(xIjx>


= $ trace B3, so that condition (3.8) is equivalent to

[trace( B3)12 = [trace( B2)13/6. (3.10)

COROLLARY 3.5. Let A = H + iK be an irreducible matrix. Then W(A)


has a flat portion on its bounda y if and only if there exist real u, v not both
zero so that for B = UN + UK - i(trace(uH + &))I the equality (3.10)
holds.

We see in Section 4 that Corollary 3.4 sometimes leads to explicit results


in spite of the fact that it refers to the existence of u, u without showing how
to construct them. The criterion not using u, 0 at all is given by the next
corollary. It is applicable to matrices A = H + iK with a diagonalized
summand K.

COROLLARY 3.6. Let A be an irreducible matrix written in the form

W(A) has a flat portion if and only if K has a multiple eigenvalue or

hdk2 - k3) + h,(k, - k,) + h,(k, - k2)


-
h,2 h23 Gh23
= (k, -k,)+ + (k, - k,)h + (k, - k2)F
23 13 12

(3.11)
NUMERICAL RANGE 127

and

h,,h,,h,, is real. (3.12)

Proof. According to Proposition 3.2, W(A) has a flat portion on its


boundary if and only if B = uH + vK + WI has rank 1 for some real u, v, w.
If K has a multiple eigenvalue A, then the latter condition is satisfied with
u = 0, v = 1, w = -A. Therefore, we need to consider only the case of K
having distinct eigenvalues.

Necessity. Let rank B = 1. Since the eigenvalues of K are all distinct, it


is possible only when u is nonzero. Without loss of generality we may (and
will) assume that u = 1.
To simplify further calculations, rewrite B in the form

1
h; + vk; + w’ h 12 h 13

B= G h’2 + vk; + w’ h,, ,

G j-G W’

where w’ = w + h, + vk,, hi = hi - h,, k: = ki - k, (i = 1,2).


Then all the off-diagonal elements h,,, h,,, h,, are nonzero (otherwise at
least two of them, located in the same row or column, equal zero, which
would imply reducibility of A), and

WI/h, = h,,/h,, = h,,/( h’, + vk; + w’) > (3.13)

-
w’/ h,, = &s/h12 = h,,/( h’, + vk; + w’). (3.14)

Solving (3.13) with respect to v, w’ we find that

h,,h,,
W’ = -
h 12 ’

For v, w’ defined by (3.15) the equalities (3.14) yield, respectively, (3.12) and

h13h,, h12 h23 hmh,, hl,h,z


h’2+h-- h’l+h-- . (3.16)
12
h 13 12 h 23
128 DENNIS S. KEELER ET AL.

It is easily checked that under the restriction (3.12) the latter is equivalent to
(3.11).

S@iZency. From (3.11) it follows, in particular, that h,alz,,hi, # 0.


Define o, w’ by (3.15). Th en, of course, (3.13) holds. Moreover, due to
(3.12) u, w’ are real, and

~7% = hnh23&hl;)
-- -
42 h3h,, . hu,/h,,
h, 4, h,, h,, = h,,/h,,.
= h,h,,h,,h,, = h,,h,,h,,
Therefore, the first of equalities (3.14) also holds.
Finally, (3.11) and (3.12) imply (3.16), which, in turn, leads to the second
of the equalities (3.14). D ue to (3.13) (3.14), I? is a (nonzero) matrix with
collinear columns and therefore has rank 1. ??

Note that condition (3.11) may be written as

where

k, - k,
___ k3- k,
~ k, - k2
~
p= lh,,12 + lh1J2 + lhJ2 ’

Since h,(k, - k,) + h,(k, - k,) + h,(k, - k,) and p are both real, it
means that condition (3.12) follows from (3.11) if p is nonzero.
The above corollary also works with H and not K diagonal. To see this,
multiply A by i. This makes H’ = - K and K’ = H. Clearly W(iA) has a flat
portion on its boundary if and only if W(A) has a flat portion.

4. W(A) FOR MATRICES WITH A TRIPLE EIGENVALUE

In this section we apply our results to the special case of matrices with a
triple eigenvalue. In their triangular form (2.4) of course, all diagonal
NUMERICAL RANGE 129

elements coincide:

(4-l)

with p, X, y, .z complex. Note that W( A) cannot be a noncircular ellipse


since such an ellipse requires two distinct foci (eigenvalues of A) of the
associated curve.

THEOREM 4.1. Let A be in the form(4.1). Then:

(1) W(A) is a disk if and only if ryz = 0; in this case the disk has radius

+dixI” + (~1’ + lz12 with centerp.


(2) W( A) has a flat portion on its boundary if and only if II ( = Iy I =
Iz) > 0; in this case C( A) is a cardioid.
(3) W(A) is of th e ocular shape if and only if xyz # 0 and Ixl, ( y 1,I ZJ are
not all equal.

A version of Part 1 of this theorem for the nilpotent case was first shown by
Marcus and Pesce, who also developed a unitarily invariant form of this
condition [9].

Proof. Part 1 follows easily from Corollary 2.5. In the rest of the proof
we may therefore suppose that xyz # 0, so that A is irreducible.

23(P)
Part 2: To simplify further calculations, consider the matrix 2 A instead of
A:

I[
Y -ix -iy
2%(p) 2 +i iX 2D( p) -iz .

2%(P) iij iZ 22(P) I

By Corollary 3.4, W(A) has a flat portion if and only if there exist real U, u
not both zero such that

lux + u( -ix)1 = by + u( -iy)l = luz + u( -iz)l,

arg(Ux - iux) + arg( uz - iuz) = arg(uy - ivy).


130 DENNIS S. KEELER ET AL.

From the first equation we see that we must have

I24- ivllrl = Iu - ivllyl = 124- ivllzl,

which implies that we must have )xl = ( y I = 1.~1since u - iv # 0. The


second equation becomes

arg( u - iv) = arg( y) - arg( x) - arg( 2).

We can easily choose u, v so that this is true. And so the only condition we
have is 1x1 = (y( = Iz(.
We now prove that under this condition C(A) is a cardioid. Using unitary
transformations A +B U*AU with U = diag[eiVl, eiv2, eiY3] (which do not
change C(A)) an d multiplying A by scalars (which rotate and dilate C(A))
we may reduce the general case to x = y = z = 1. Shifting then A by AI
(which shifts C(A) by A), we may suppose also that its eigenvalue is 5 (such a
choice of the eigenvalue ensures that the cusp of the cardioid would be at the
origin). In other words, without loss of generality

Using Fiedler’s formula (see [6]) for the point equation of C(A) and
transforming to polar coordinates we find

3P( -3r - 2 + 2cos 0)( -3r + 2 + 2cos 0) = 0.

The factor of 3r2 is redundant since r = 0 is a solution to the other two


factors. The other two factors actually define the same curve. This is because
if one replaces r with -r and 0 with 8 + rr, the factors are identical within
a scalar multiple. In polar coordinates, this ‘means that the factors trace the
same curve. The equation therefore simplifies to r = $<l - cos 01, which is
the equation of a cardioid [lo].
Part 3: Since W(A) cannot be an ellipse without being a disk, the ovular
shape is the only case left. H
NUMERICAL RANGE 131

0
A computer image of W(A), where

1
1 1

A= i0
0 0
0 01,

was given in [9]. By Theorem 4.1, W(A) is the convex hull of a cardioid.
As in Sections 2 and 3, the unitarily invariant version exists:

THEOREM 4.2. Let A be a matrix with triple eigenvalue p, * =


trace(A*A) - 31p12, and R = trace(A*A’) - 2p(trace(A*A)) + 3plp1’.
Then

1. W(A) is a disk centered at p if and only if R = 0. In this case it has


radius i@.

2. W(A) has aflat portion on its boundary if and only if w = g3m >
0.

3. W(A) has an ovular shape if and only if 1I’# 33m > 0.

Pt-oaf. It suffices to consider A in the form (4.1). The direct computa-


tion then shows that * = ( xl2 + ( y12 + I zj2 and IR = xijz. Obviously, condi-
tion 1 is equivalent to xyz = 0. Due to the case of equality between
arithmetic and geometric means of Ix 12,Iy 12,(z 12, 2 holds if and only if
1x I = 1y I = 1z I > 0. Finally, 3 is the only logically possible case left. ??

5. RESTORATION OF A FROM W(A)

An inverse problem concerning numerical ranges may be formulated.


Given a numerical range W(A) for some A, can one reconstruct A? Since
W(A) = W(U*AU), we cannot restore A uniquely (with the exception of
W(A) being a single point), but we can sometimes find a unique unitary
equivalence class that generates W(A). The latter is always the case for 2 X 2
matrices (see, e.g., [11>.
In the remainder of this section, we deal with 3 X 3 matrices. For
reducible A, it can be easily seen that in this case A cannot always be
restored from W(A), but it can be restored from W(A) and the trace of A,
or equivalently C(A). We sh ow later that in the irreducible case, A cannot
always be restored, even if C(A) is known.
Unexpectedly, there is a case of a W(A) arising from an irreducible
matrix, which allows A to be restored up to unitary similarity:
132 DENNIS S. KEELER ET AL.

THEOREM 5.1. Let W(A) be a 2-dimensional shape with only one jlat
portion on its boundary. Then A is an irreducible matrix, which can be
restored up to unitary similarity.

Proof. Having only one flat portion on the boundary of its numerical
range, A belongs to Case 4 of Kippenhahn’s classification and is therefore
irreducible. After scaling, rotation, and shifting of W(A), we can have the flat
portion as the line segment [O, i] and W(A) lies entirely in the right
half-plane. We restore A in this case. After the restoration, one can obtain
the original A by reversing the scaling, rotation, and shifting.
According to Theorem 3.1, A must be unitarily similar to (3.1). Let us
assume A is in that form. The real part H of A is then diag(0, 0, 8( 5 )>, with
8( f > positive. Since W(%( A)) is the projection of W(A) onto the real axis,
which is a line segment from 0 to !R(c), we can determine 8(l).
Since there is only one flat portion, the real part of any point on that
portion is 0. So J is not on that portion. Because %( J’) is an endpoint of the
projection of W(A) onto the real axis and t is not on the flat portion, 5 is
uniquely determined as the point on the boundary of W(A) having a
maximum real part, namely YI( 5 ). So s( [ > is also determined.
The imaginary part K of A is

Since W( K > is a line segment, which is a projection of W(A) onto the


imaginary axis, we know two of the eigenvalues A,, A, of K, namely the
endpoints of the line segment. Calculating the characteristic polynomial of K
and substituting in A,, A, give us the system of linear equations in cf, ci

c,2(-A,) + c;(-A, + 1) = -A; + Af(l + W>) - WVC))


(5.1)
cf( -A,) + c2”( -A, + 1) = -A; + A;(1 + g(c)) - A@(l)).

The determinant of this system is Aa - A,, which is nonzero since the flat
portion is of nonzero length, causing the projection of W(A) onto the
imaginary axis to be of nonzero length. So the system (5.1) has a unique
NUMERICAL RANGE 133

solution:

cf = -(A, - l)(& - l)(A, + h, - S(S))

c; = A,h,(A, + A, - 1 - 5( 6)).

Since c,, c2 are positive, we thus have unique values for them. Therefore
we know all the elements of A in this canonical form, which determines A
up to unitary similarity. a

As it turns out, in cases of other shapes of W(A) for an irreducible A, the


matrix A cannot be uniquely (up to unitary similarity) restored by W(A). We
summarize all these cases, as well as the cases of a reducible A, in the
following theorem.

THEOREM 5.2. A 3 X 3 matrix A can be restored (up to unitary


similarity) from W(A) if and only if W(A) is one of the following: (1) a
point, (2) a triangle, (3) the convex hull of an ellipse and a point outside the
ellipse, (4) a %dimensional shape with only one flat portion on its boundary.
In the other cases, that of (5) a line segment, (6) an ellipse, and (7) an
ovular shape, the matrix cannot be restored. In the cases 5-7 there is a
continuum of nonunitarily equivalent matrices whose numerical range is
W( Al.

Proof. Cases l-3 are well known; case 4 was discussed in Theorem 5.1.
In case 5 A is normal, with at least two distinct eigenvalues and all three
eigenvalues collinear. The eigenvalues corresponding to the endpoints can be
determined, but the third eigenvalue cannot. There is a continuum choice for
this third eigenvalue.
If W(A) is an ellipse and A is reducible, A cannot be restored since the
point defined by its 1 X 1 block may be anywhere within the ellipse defined
by the 2 X 2 block. Again, there is a continuum of choices for the 1 X 1
block.
The proof in the remaining two situations (W(A) is an ellipse produced
by an irreducible A or an ovular shape) is based on a series of lemmas and is
therefore relegated to the end of this section. H

One might ask whether a matrix A can be restored (up to unitary


similarity) from W(A) and the trace of A. In this respect we note that for a
3 X 3 matrix A each of the following pieces of information completely
134 DENNIS S. KEELER ET AL.

determines two others: (1) W(A) and the trace of A; (2) W(A) and the
eigenvalues of A; (3) C(A).
Indeed, C(A) determines uniquely W(A) (because W(A) is the convex
hull of C(A) and th e ei g envalues of A (because there are the foci of C(A)).
On the other hand, if W(A) is known then the maximal and minimal
eigenvalues of every linear combination H cos tf + K sin 5 (here A = H +
iK with Hermitian H and K and 5 is a real number) are determined by
using the orthogonal projection of W(e?6AA) onto the real axis; note that H
cos 6 + K sin 5 is the real part of e -“(A. If, in addition, the trace of A is
known, then all eigenvalues of H cos 5 + K sin 6 are known, and therefore
the polynomial det(uH + vK + wZ) is known, which determines C(A).
It will be clear from the proof of Theorem 5.2 that, in addition to the
cases when a 3 X 3 matrix A can be restored from W(A), such a matrix can
be restored from C(A) (or equivalently from W(A) and the trace of A) if
W(A) is a line segment. On the contrary, if W(A) is ovular or W(A) is an
ellipse (without any information concerning the reducibility of A), then there
are uncountably many unitarily inequivalent matrices B such that C(B) =
C(A). However, if W(A) is an ellipse and it is known that A is reducible,
then A can be restored (up to unitary similarity) from C(A).
The rest of this section is devoted to completion of the proof of Theorem
5.2.
We use the two matrices

with czi > CQ > ~ys, cy; > (Y; > ai, Pi, &’ real, and off-diagonal elements
such that

e, f, e’, f’ > 0; if ef = 0, then g 2 0; if e’y = 0, then g’ 2 0.

(5.4)
NUMERICAL RANGE 135

LEMMA 5.3. Let A and B be written in the form (5.2) and (5.3). Then
L, = L, if and only if

1. All the diagonal elements are equal: czj = CX~,Pj = # for j = 1,2,X
2. e’ = e2 + ((al - CQ)/( c+ - a3))(lg’12 - lg12)
3. f’ = Jf2 - ((a1 ‘y3))(lg’12 - lg12)
- C$)/((Yz -
4. e’f)(g’ + z) = ef(g + 2) + (lg12 - lg’12X(aJP2 - PJ
+ cQ( & - PJ + q( PI - &))/(a2 - 4

Proof. For L, and L, to coincide it is necessary, in particular, that A


and B have the same sets of eigenvalues. Then their traces also are the same.
Without loss of generality we may assume that they equal zero. Assuming
that, redenote cri = a, cx2 = b, PI = c and p2 = d; then, of course, cxs =
-a - b, & = -c - d. Analogously, the diagonal elements of H’ and K’
are now a’, b’, -a’ - b’ and c’, d’, -c’ - d’, respectively.
From the equality L,(u, u, w) = L,(u, u, w) for u = 1, u = 0 it follows
that H and H’ have the same eigenvalues. Since in both matrices the
eigenvalues are ordered, it means that H’ = H and thus a’ = a, b’ = b.
Calculation shows that det(uH + vK + WI) =

(ab2 - a2b)zr3 + (-2abd - a2d - 2abc - cb2)u2u

+( -a2 - ab - b2)U2w

+( -2bcd + a(e2 - lg12) - bc2 - 2acd - ad2 + b(e2 - f”))zm”

+(c( -b - 2a) + d( -a - 2b))uvw

+(c(e2 - lg12) - c2d + ef( g + g) + d(e2 -f’) - cd2)v3

+ ( -d2 - cd -f 2 - c2 - lg12 - e2)v2w + w3.

Comparing
- - this to the equation
_ for det(uH’ + vK’ + wl) we see from the
coefficients of u2v and uvw that we must have

(-2ab - b2)c’ + (-2ab - a2)d’ = (-2ab - b2)c + (-2ab - a2)d

(-2a-b)c’+(-a-2b)d’=(-2a-b)c+ (-a - 2b)d.


136 DENNIS S. KEELER ET AL.

Considering this as a linear system of equations in c’, d’ of the form Cx = y,


det(C) = (-2ab - b2X-a - 2b) - (-2ab - a2X-2a - b) = (-a -
2bXa - bX2a + b) z 0 by our assumption that II > b > -(I - b. The
system therefore has a unique solution, which is obviously c’ = c, d’ = d.
We can now conclude that all the diagonal elements of A and B are the
same.
From the coeffkients of v2w and v2u we have, after elimination of
identical terms,

d2 +y2 = e2 +f2 + lg12- @I2

(u + b)e” - bf’l = (u + b)e2 - bf2 - a(lg12 - lg’12),

which can be viewed as a linear system of equations in e 12, f”. By solving the
system and using our assumption that e’, f’ are nonnegative, we obtain

f’=J.
Finally from the coefficient of v3 we have

e’fl(g’ + z) = ef(g + g) + c(lg’12 - lg12)

+(c + d)(e2 - er2) - d(f’ -fr2)

= ef(g + g) + (lg’? - lg12)

u-b 2a +b
c-(c+d)a-d-
I

3(bc - ad)
= ef( g + g) + (lg’i2 - lgt2)

Substituting our definitions of a, b, c, d into the above equations gives US


the equations stated in the lemma. ??
NUMERICAL RANGE 137

LEMMA5.4. Let A and B be in the forms (5.2) and (5.3). Then A is


unitarily similar to B if and only if A = B.

Proof. As in the proof of the preceding lemma, we assume that A and


B have zero trace. Suppose A = U* BU for some unitary U. Then obviously
C(A) = C(B), and therefore H = H’ by Lemma 5.3. Now H = U*HU,
which implies that U must be diagonal. Using the condition (5.4), the equality
K’ = U*KU implies K’ = K. ??

LEMMA5.5. Let A be an irreducible matrix in the form (5.2). Zf g is not


real, or condition

=
(a2 - 4ef + (a3 - al)eg + (a1 - c4fg (5.5)

is not satisfied, there exists a continuum of unitary equivalences classes of


matrices with the same associated curve C(A).

Proof. We construct matrices B of the form (5.3) with C(A) = C(B).


Let p = lg’? - lg?,

q = _ 4 P2 - P,) + cf2(P3- PA + 4 Pl - P2)


a2 - ff3

(5.6)
a1 - ff3
~ = a1 - a2 ,
7=
ff2 - a3 a2 - a3

Note that p, T > 0.


Then by Lemma 5.3, for C(B) = C(A) we must have

e’ = &F&T (5.7)

fl=dG (5.8)

e’fr(g’ + 2) = ef(g + g) + pq. (5.9)

To satisfy (5.71, (5.8), let us choose p E I = ( -e2/p, f 2/r). Note that the
length of I is positive, because otherwise f = e = 0, and A would be
138 DENNIS S. KEELER ET AL.

reducible. The last equation (5.9) is then equivalent to

2dfFg + Pq
%g’= * (5.10)
2 (e”+ pp)(f2 - TP)
For number g’ E C with 1g’12 = p + 1g I2 and % g’ given by (5.10) to exist, it
is necessary and sufficient that

(24% + Pd2 ~ p + ,g,2


4(e2 + w.p>(f” - 7~)

The latter inequality can be rewritten as fi p> > 0, if we denote

S(P) = (e” + pp)(f2 - TP>(~P + 41g12) - (2efsg + p412.

It g is not real, then P(O) > 0, so that there is an E > 0 such that
fi p) > 0 for ( pi < E. Every p E Z f~ (-E, E) generates a matrix (5.3) with
C(B) = C(A). Different values of p correspond to different matrices B, and
none of them are unitarily similar due to Lemma 5.4.
If g is real, then &O) = 0, and

CZY
= 4pf2g2 - 4re2g2 + 4e2ff2 - 4efgq.
dp p=o

Substituting in the values of Z.J, 7, and q from (5.6) we see that (5.5) is
equivalent to d4T/dpl,=o = 0. Hence, if (5.5) does not hold, there is a
one-sided neighborhood N of zero such that fi p) > 0 for p E N. Observe
that this neighborhood is positive if e = 0 and negative if f = 0, so that in
any case N n I is a continuum. All p E N n Z generate matrices with the
same associated curve as C(A), and, as above, all these matrices belong to
different unitarily equivalence classes. I

We now complete the proof of Theorem 5.2. Consider an irreducible


matrix A. Without loss of generality we may suppose that it is in the form
(5.2). Corollary 3.6 im pl ies that W(A) contains a flat portion on its boundary
if and only if g is real and (5.5) holds. From here and Lemma 5.5 it follows
that in all other cases (that is, when W(A) is an ellipse or has an ovular
shape) there is a continuum of unitary equivalence classes of matrices with
the same numerical range W(A) (and even the same associated curve C(A)).
??
NUMERICAL RANGE 139

We thank the referee for carefully reading the original version of this
article and suggesting many improvements in exposition, as well as Theorem
2.1.

REFERENCES

I. Glazman and Yu. Lyubich, Finite-Dimensional Linear Analysis: A Systematic


Presentation in Problem Form, MIT. Press, Cambridge, MA, 1974.
R. A. Horn and C. R. Johnson, Matrix Analysis, Cambridge Univ. Press,
Cambridge, 1985.
R. Kippenhahn, fiber den Wertevorrat einer Matrix, Math. Nachr. 6:193-228
(1951).
F. D. Mumaghan, On the field of values of a square matrix, Proc. Nat. Acad.
Sci. USA 18:246-248 (1932).
H. Shapiro, A conjecture of Kippenhahn about the characteristic polynomial of a
pencil generated by two Hermitian matrices, II, Linear Algebra AppZ. 45:97-108
(1982).
6 M. Fiedler, Geometry of the numerical range of matrices, Linear Algebra Appl.
37:81-96 (1981).
7 M. Chien and B. Tam, Circularity of the numerical range, Linear Algebra Appl.
201:113-133 (1994).
8 B. L. van der Waerden, Algebra, Vol. 1. Frederick Ungar, New York, 1970.
9 M. Marcus and C. Pesce, Computer generated numerical ranges and some
resulting theorems, Linear Multilinear Algebra 20:121-157 (1987).
10 J. D. Lawrence, A Catalog of Special Plane Curves, Dover, New York, 1972.

Received 22 September 1994; final manuscript accepted 20 August 1995

You might also like