PHYSICAL REVIEW E VOLUME 61, NUMBER 1 JANUARY 2000

Repulsion of resonance states and exceptional points

W. D. Heiss
Centre for Nonlinear Studies and Department of Physics, University of the Witwatersrand, P.O. Wits 2050, Johannesburg, South Africa
共Received 23 July 1999兲
Level repulsion is associated with exceptional points which are square root singularities of the energies as
functions of a 共complex兲 interaction parameter. This is also valid for resonance state energies. Using this
concept it is argued that level anticrossing 共crossing兲 must imply crossing 共anticrossing兲 of the corresponding
widths of the resonance states. Further, it is shown that an encircling of an exceptional point induces a phase
change of one wave function but not of the other. An experimental setup is discussed where this phase
behavior, which differs from the one encountered at a diabolic point, can be observed.

PACS number共s兲: 84.40.⫺x, 03.65.⫺w, 02.30.Dk

The dependence on parameters of the energies and widths erality for our specific purpose when the restriction to a two-
of resonance states has always been a central focus of inter- dimensional problem is made. For easy illustration we begin
est in virtually all domains of physics. One particular aspect with the discussion of

冉 冊 冉 冊
is the repulsion of levels in the complex energy plane. A
level repulsion in the complex energy plane can appear as a ⑀1 0 ␻1 0
H⫽ ⫹␭U U† 共1兲
crossing of, say, their real parts, since the corresponding 0 ⑀2 0 ␻2
imaginary parts still can avoid each other; likewise, a genu-
ine repulsion of the real parts can imply a crossing of the with

冉 冊
imaginary parts. These aspects have been discussed in a va-
riety of contexts: in nuclear and particle physics, for electro- cos ␾ ⫺sin ␾
magnetic resonators 关1,2兴, and in results found for absorptive U共 ␾ 兲⫽ . 共2兲
sin ␾ cos ␾
media in solid state physics 关3兴. An investigation on a more
theoretical footing is found in 关4兴. This is, up to a similarity transformation, the most general
The purpose of the present paper is twofold. First, we form of a real two-dimensional Hamilton matrix of the type
demonstrate that the various types of crossing and/or anti- H 0 ⫹␭H 1 . We emphasize again that our aim is not in par-
crossing can be understood from a common principle. They ticular directed at a physical model that is describable by a
are related to each other by the position of particular singular two-dimensional problem although there may exist interest-
points of the spectrum, which are called exceptional points ing problems in our special context. The example has been
共EP兲 关5兴. The second aspect deals with the fact that, if an EP chosen for illustration, while the physical application that we
is encircled, the phases of the associated wave functions have in mind is in general an infinite-dimensional situation.
change in a particular way which is different from the phase The eigenvalues of H are given by
behavior when a genuine degeneracy of levels 共a diabolic
point兲 is encircled 关6兴. At an EP two levels coalesce, but, as ⑀ 1 ⫹ ⑀ 2 ⫹␭ 共 ␻ 1 ⫹ ␻ 2 兲
E 1,2共 ␭ 兲 ⫽ ⫾R, 共3兲
is discussed below, an EP is not to be confused with a genu- 2
ine degeneracy of two resonant states. The fact that there are
different types of coalescence of resonance states was where
pointed out in 关4兴. However, in the quoted paper, the type of
singularity, in fact the concept of an EP was not explicitly
employed; rather the effect upon the Green’s function or the
R⫽ 再冉 2 冊 冉
⑀ 1⫺ ⑀ 2 2 ␭共 ␻ 1⫺ ␻ 2 兲
⫹
2 冊 2

冎
scattering amplitude was elaborated, which is of lesser inter- 1/2
est here. Genuine degeneracies of resonance states have been 1
⫹ ␭ 共 ⑀ 1 ⫺ ⑀ 2 兲共 ␻ 1 ⫺ ␻ 2 兲 cos 2 ␾ . 共4兲
discussed in the literature, including an associated phase be- 2
havior of the wave functions involved 关7,8兴. However, the
subject of the present paper which is a generalization and Clearly, when ␾ ⫽0 the spectrum is given by the two lines
further expansion of a previous publication 关9兴 addresses a
thoroughly different situation. E 0k 共 ␭ 兲 ⫽ ⑀ k ⫹␭ ␻ k , k⫽1,2
All essential aspects of exceptional points can be illus-
trated on an elementary level with a two-level model. In fact, which intersect at the point of degeneracy ␭⫽⫺( ⑀ 1 ⫺ ⑀ 2 )/
for finite- or infinite-dimensional problems an isolated ex- ( ␻ 1 ⫺ ␻ 2 ). When the coupling between the two levels is
ceptional point can be described locally by a two- turned on by switching on ␾ , the degeneracy is lifted and an
dimensional problem 关10兴. In other words, even though a avoided level crossing occurs. Now the two levels coalesce
high or infinite dimensional problem is globally more com- in the complex ␭ plane where R vanishes. This happens at
plex than the two-dimensional problem, we do not lose gen- the complex conjugate points

1063-651X/2000/61共1兲/929共4兲/$15.00 PRE 61 929 ©2000 The American Physical Society

930 BRIEF REPORTS PRE 61

⑀ 1⫺ ⑀ 2
␭ c ⫽⫺ exp共 ⫾2i ␾ 兲 . 共5兲
␻ 1⫺ ␻ 2

At these points, the two levels E k (␭) are connected by a

square-root branch point; in fact the two levels are the values
of one analytic function on two different Riemann sheets.
Obviously, this connection is not of the type encountered at a
genuine diabolic point. We stress again that the same nature
of singularity prevails also in an N-dimensional matrix prob-
lem of the type H 0 ⫹␭H 1 .
The question arises as to whether the existence of the EPs
is of physical interest in addition to their pertinent associa-
tion with level repulsion. Before we turn to actual proposals
of encircling an EP in an experiment we first discuss for-
FIG. 1. Level repulsion 共width crossing兲 and level crossing
mally the various effects of encircling an EP. Obviously, we
共width repulsion兲 for smaller 共top, ␮ ⫽0.35) and larger 共bottom,
obtain the same information by comparing the results of two
␮ ⫽0.5) absorption ␮ which is chosen such that the EP lies just
different sweeps over appropriate values of ␭, the one by below the real ␭ axis in the former and just above in the latter case.
passing an EP on its left-hand side and the other on its right- The other parameters are ⑀ 1 ⫽1,⑀ 2 ⫽2,␻ 1 ⫽1,␻ 2 ⫽⫺1,␾ 1 ⫽0.2.
hand side. In principle, this can be achieved by choosing
complex values of ␭. In order to get closer to an actual
it is demonstrated for a four-dimensional model that our
experimental situation we expand the Hamiltonian by con-
findings are not just the fluke of a two-dimensional
sidering the enlarged model
model 关11兴.

冉 冊 冉 冊
The top row shows a usual level repulsion among the four
⑀1 0 ␻1 0
levels together with the exceptional points without absorp-
H⫽ ⫹␭U 共 ␾ 1 兲 U †共 ␾ 1 兲
0 ⑀2 0 ␻2 tion, i.e., for a real symmetric Hamiltonian, while in the bot-

冉 冊
tom row the absorption has been made sufficiently strong to
␴1 0 enforce the crossing of all levels; the last EP which has just
⫺i ␮ U 共 ␾ 2 兲 U †共 ␾ 2 兲 . 共6兲
0 ␴2 slipped over the real axis with increasing absorption lies at
␭⬇⫺5.5. Note the symmetric positions of the EP with re-
The additional term ( ␮ real兲 can be used to describe an ab- spect to the real axis in the top row which prevails for a
sorption while adhering to real values of ␭. Also, the unper- self-adjoint Hamiltonian 关12兴. Also note that the crossing of
turbed energies ⑀ k may be chosen complex, that is including the real axis by an EP can happen in either direction; this is
a width. The EP of the enlarged model are situated at why, in the particular case considered, two EP are left in the
lower half plane for the absorptive Hamiltonian.

冉
␭ c ⫽ ⫺1⫹i ␮ e ⫾2i ␾ 2
⑀ 1⫺ ⑀ 2
e 冊
␴ 1 ⫺ ␴ 2 ⫾2i ␾ ⑀ 1 ⫺ ⑀ 2
1
␻ 1⫺ ␻ 2
. 共7兲
We present a topological argument why either the real
parts or the imaginary parts must cross when energy trajec-
tories of an absorptive Hamiltonian pass the vicinity of an
In an experimental situation a judicious choice of these EP. We denote by ␭ cross the real part of an EP and follow the
additional parameters can move one of the EP in the ␭ plane
close to the real axis. In fact, we now demonstrate that the
position of an EP can be arranged in various ways to lie just
above or below the real ␭ axis. The different effect of these
two situations for the energies, when sweeping over real val-
ues of ␭, turns out to be anticrossing for the real and cross-
ing for the imaginary parts in the one case, and crossing for
the real and anticrossing for the imaginary parts in the other.
In Fig. 1 we illustrate the real and imaginary parts of the
energies as a function of the real parameter ␭ for two differ-
ent values of ␮ . For simplicity, the choice ␾ 2 ⫽0,␴ 1 ⫽1, and
␴ 2 ⫽0 was made. Other choices lead qualitatively to the
same result if ␴ 1 ⫽ ␴ 2 . Also, if the unperturbed energies are
chosen complex, either in addition or instead of the choice
just made, the qualitative picture remains. The switching
from level avoidance to level crossing of the real parts—and
associated with it the switching from crossing to avoidance FIG. 2. Effect of strong absorption in an arbitrary four-level
of the related imaginary parts—is effected by the slipping of model. Complex conjugate pairs of EPs of a nonabsorptive Hamil-
the EP over the real ␭ axis. Within the model considered tonian 共top left兲 lead to genuine level repulsion 共top right兲. The
here it can be achieved by starting with different unperturbed absorption moves one member of each pair of EPs either into the
widths and/or an absorptive part of the Hamiltonian which upper or lower ␭ plane 共bottom left兲 thereby effecting level cross-
couples to the two channels with different strength. In Fig. 2 ing 共bottom right兲.
PRE 61 BRIEF REPORTS 931

for the wave functions behind the repulsion. In turn, if the

levels do cross, the wave functions pertaining to the same
levels will not undergo this rotation. As a consequence, be-
hind the crossing or anticrossing the bottom level is associ-
ated with ␺ 1 while the wave function of the top level is ⫹ ␺ 2
for the crossing and ⫺ ␺ 2 for the anticrossing. This intuitive
argument does not really specify which of the two wave
functions changes its sign; what matters is the relative
change of one of the wave functions. Which one it is in the
final comparison depends on the sense of direction by which
the EP is encircled.
To confirm more formally the statement about the phases
we consider the two situations displayed in Fig. 1. We pa-
rametrize the state vectors by the complex angle ␪ , viz.,

␺ 1共 ␭ 兲 ⫽ 冉 冊
cos ␪
sin ␪
, ␺ 2共 ␭ 兲 ⫽ 冉 ⫺sin ␪
cos ␪
冊 , 共9兲

with
FIG. 3. Schematic energy trajectories in the complex energy
plane. Explanations in main text. tan ␪ 共 ␭ 兲 ⫽ 关 ␭ 共 ␻ 1 ⫺ ␻ 2 兲 sin 2 ␾ 1 ⫺i ␮ 共 ␴ 1 ⫺ ␴ 2 兲 sin 2 ␾ 2 兴 /

trajectories E 1,2(␭) for real values of ␭ in the interval 关 E 1 共 ␭ 兲 ⫺E 2 共 ␭ 兲 ⫹ ⑀ 1 ⫺ ⑀ 2 ⫹␭ 共 ␻ 1 ⫺ ␻ 2 兲

关 ␭ cross⫺ ␦ ,␭ cross⫹ ␦ 兴 . In Fig. 3 the complex numbers
⫻cos 2 ␾ 1 ⫺i ␮ 共 ␴ 1 ⫺ ␴ 2 兲 cos 2 ␾ 2 兴 . 共10兲
E 1,2(␭ cross⫺ ␦ ) are indicated by A and A ⬘ . By definition they
have different real and imaginary parts. The end points of the
For the value of ␮ 共and ␾ 2 ⫽0) which yields the anticrossing
trajectories, which are at E 1,2(␭ cross⫹ ␦ ) and denoted by B
we read off from Eq. 共10兲 the expected result: ␪ (0)⫽0 and
and B ⬘ , must schematically be situated as indicated, since we
consider the vicinity of a square root singularity. Schemati- ␪ (␭)→ ␾ 1 for ␭Ⰷ 兩 ( ⑀ 1 ⫺ ⑀ 2 )/( ␻ 1 ⫺ ␻ 2 ) 兩 . In obtaining this re-
cally we may assume the singularity to be in the middle of sult use is made of E 1 ⫺E 2 ⫽2R→␭( ␻ 1 ⫺ ␻ 2 ) for ␭
the square like figure. When ␭ is sweeping over the interval Ⰷ 兩 ( ⑀ 1 ⫺ ⑀ 2 )/( ␻ 1 ⫺ ␻ 2 ) 兩 . For the other value of ␮ yielding
the crossing of the levels we now have to observe that we
关 ␭ cross⫺ ␦ ,␭ cross⫹ ␦ 兴 the energy trajectory starting at A can
crossed into the other sheet of the square root which means
move to B in which case A ⬘ must move to B ⬘ 共dotted lines兲.
This is the case of width crossing and level avoidance. The E 1 ⫺E 2 ⫽⫺2R→⫺␭( ␻ 1 ⫺ ␻ 2 ). As a consequence we find
other possibility is that the end points are interchanged which this time tan ␪ →tan( ␾ 1 ⫹ ␲ /2) which confirms the result.
is level crossing and width avoidance 共dashed lines兲. Only in This consideration also clarifies that it is the square root sin-
the special case where the parameter ␭ moves straight gularity that brings about this particular phase change. The
through an EP will both, the real and imaginary part, cross. values of ␭ which exceed ␭ cross get us into different Riemann
Since it is a square root singularity the angles between the in sheets depending on whether we pass the EP on its right-
and out trajectories at the EP must be at 90° in the energy hand or left-hand side.
plane 共solid lines兲. The different cases as illustrated in Fig. 1 have already
The different behavior of the energy trajectories, depend- been experimentally established in an electromagnetic reso-
ing on a left-hand or right-hand side passage of an EP, is also nator 关13兴. Two coupled resonators have been used as experi-
reflected in a different behavior of the corresponding wave mental setup. The levels of the one have been tuned by a
functions. It can be intuitively argued, and we confirm this parameter which plays the role of our parameter ␭. As a
formally below, that we should expect the phase of one but second parameter, the coupling strength between the resona-
not the other wave function to be different when comparing tors has been controllable; we denote this quantity by x. For
them behind the point of anticrossing and crossing of their fixed absorption which is achieved by suitable antennas the
real parts 共Fig. 1兲. In fact, for the situation of level avoid- necessary different widths have been adjusted. The situation
ance, it is well known that the wave functions after the point has also been modeled by two levels, however, without ref-
of repulsion are basically as if the levels would have crossed erence to the existence of EPs. With the parametrization used
but for a minus sign of one. In the vicinity of the point of in the present paper it corresponds to the choice ␮ ⫽0,␾ 1
repulsion the eigenvectors can be parametrized by an angle ␣ ⫽ ␲ /4, and ⑀ k ⫽E k ⫺i⌫ k /2 with ⌬⌫⫽⌫ 1 ⫺⌫ 2 ⫽0. The ex-
which ranges from 0 to ␲ /2 when ␭ is sweeping over the ceptional points are then situated at
repulsion point. Denoting by ␺ 1 and ␺ 2 the wave functions
of the top and bottom level before the repulsion, respec- ␭ c ⫽⫺⌬E⫺i⌬⌫⫾ix
tively, we have
with ⌬E⫽ ⑀ 1 ⫺ ⑀ 2 . In this way, the difference ⌬⌫ and/or the
␺ top⫽ ␺ 1 cos ␣ ⫺ ␺ 2 sin ␣ , coupling x can be adjusted such that one EP lies just above
共8兲 or below the real ␭ axis thus giving rise to the two cases
␺ bot⫽ ␺ 1 sin ␣ ⫹ ␺ 2 cos ␣ illustrated in Fig. 1. The equipment used in 关13兴 did not
932 BRIEF REPORTS PRE 61

allow a measurement of the phases of the wave functions, sible in a variety of other systems, systems where interacting
i.e., of the field strengths. It appears, however, that this is resonances prevail. The present paper focuses on level cross-
possible 关14兴. ing or anticrossing and related phase behavior for the asso-
We stress that exceptional points are a universal phenom- ciated wave functions. Interference effects between the two
enon in contrast to diabolic points. While diabolic points distinct cases—left-hand and right-hand passage of an EP—
and statistical aspects for a large number of resonance states
may arise when two real parameters are suitably chosen in a
will be the subject of forthcoming considerations.
Hamiltonian, exceptional points always occur whenever
there is level repulsion. The physically interesting aspect is The author acknowledges useful discussions with Peter
of course the access to one or more of these points in an von Brentano as well as the warm hospitality of the Theory
experiment. It was demonstrated that this is achievable in Group of the Max Planck Institute at Heidelberg where most
dissipative resonators. It is expected that it should be pos- of this paper was written.

关1兴 P. von Brentano and M. Philipp, Phys. Lett. B 454, 171 共1999兲, 3228 共1989兲.
and references therein. 关8兴 C. Mimiatura, C. Sire, J. Baudon, and J. Bellissard, Europhys.
关2兴 W.D. Heiss, M. Müller, and I. Rotter, Phys. Rev. E 58, 2894 Lett. 13, 199 共1990兲.
共1998兲 and references therein. 关9兴 W.D. Heiss, Eur. Phys. J. D 7, 1 共1999兲.
关3兴 O. Lefeuvre et al., Appl. Phys. Lett. 72, 856 共1998兲. 关10兴 W.D. Heiss and W.-H. Steeb, J. Math. Phys. 32, 3003 共1991兲.
关4兴 E. Hernandez and A. Mondragon, Phys. Lett. B 326, 1 共1994兲; 关11兴 Preliminary simulations in preparation of an experiment
A. Mondragon and E. Hernandez, J. Phys. A 26, 5595 共1993兲. clearly confirm the same pattern in an infinite-dimensional
关5兴 T. Kato, Perturbation Theory of Linear Operators 共Springer, problem: C. Dembowski 共private communication兲.
Berlin, 1966兲. 关12兴 W.D. Heiss and A.L. Sannino, J. Phys. A 23, 1167 共1990兲;
关6兴 M.V. Berry, in Quantum Chaos, edited by G. Casati 共Plenum, Phys. Rev. A 43, 4159 共1991兲.
London, 1985兲; Proc. R. Soc. London, Ser. A 239, 45 共1983兲.
关13兴 M. Philipp, Ph. D. thesis, University of Köln, 1997 共unpub-
关7兴 G. Nenciu and G. Rasche, J. Phys. A 25, 5741 共1992兲; T.B.
lished兲.
Kepler and M.L. Kagan, Phys. Rev. Lett. 66, 847 共1991兲; D.
关14兴 Achim Richter 共private communication兲.
Ellinas, S.M. Barnett, and M.A. Dupertuis, Phys. Rev. A 39,