Observation of PT phase transition in a simple mechanical system

Carl M. Bender1∗ , Bjorn K. Berntson2† , David Parker1‡ , and E. Samuel1§

1
Department of Physics, King’s College London, Strand, London WC2R 1LS, UK
2
Blackett Laboratory, Imperial College, London SW7 2AZ, UK
(Dated: July 24, 2012)
Abstract
If a Hamiltonian is PT symmetric, there are two possibilities: Either the eigenvalues are entirely
real, in which case the Hamiltonian is said to be in an unbroken-PT -symmetric phase, or else the
eigenvalues are partly real and partly complex, in which case the Hamiltonian is said to be in
a broken-PT -symmetric phase. As one varies the parameters of the Hamiltonian, one can pass
through the phase transition that separates the unbroken and broken phases. This transition
has recently been observed in a variety of laboratory experiments. This paper explains the phase
transition in a simple and intuitive fashion and then describes an extremely elementary experiment
in which the phase transition is easily observed.

PACS numbers: 11.30.Er, 03.65.Ca, 03.65.Xp

∗
Permanent address: Department of Physics, Washington University, St. Louis, MO 63130, USA.
email: cmb@wustl.edu
†
email: bjorn.berntson11@imperial.ac.uk
‡
email: david.j.parker@kcl.ac.uk
§
email: ernest.samuel@kcl.ac.uk

1
I. INTRODUCTION

Quantum-mechanical PT -symmetric Hamiltonians often exhibit two parametric regions,

a region of unbroken PT symmetry in which the eigenvalues are all real, and a region of
broken PT symmetry in which some of the eigenvalues are real and the remaining eigenvalues
are complex. (Here, P represents space reflection, or parity, and T represents time reversal.)
A phase transition occurs at the boundary between these two regions. This paper describes
an elementary experiment that demonstrates the PT phase transition in a simple PT -
symmetric classical-mechanical system involving two coupled driven pendula.
If a quantum-mechanical Hamiltonian is Dirac Hermitian (that is, if H = H † , where
H † is the transpose and complex conjugate of H), then its energy eigenvalues are always
real. Thus, a Dirac Hermitian Hamiltonian can never exhibit a phase transition where its
eigenvalues go from being real to being complex. However, in 1998 it was shown that the
eigenvalues of a PT -symmetric Hamiltonian can be entirely real, even if the Hamiltonian is
not Dirac Hermitian.[1–3] The class of PT -symmetric Hamiltonians considered in Refs. [1–3]
has the form
H = p2 + x2 (ix)ε , (1)
where ε is a real parameter. These Hamiltonians are not Dirac Hermitian (except at ε = 0),
but they are PT symmetric because under space reflection x changes sign and under time
reversal i changes sign. The Hamiltonians (1) exhibit a region of unbroken PT symmetry
(ε ≥ 0) and a region of broken symmetry (−1 < ε < 0). Thus, at ε = 1 and at ε = 2 we
obtain the Hamiltonians
H = p2 + ix3 (2)
and
H = p 2 − x4 , (3)
and surprisingly, the eigenvalues of these Hamiltonians are all real, positive, and discrete.
However, at ε = −1/2 we obtain the Hamiltonian H = p2 + x2 (ix)−1/2 , which has an infinite
number of complex eigenvalues and only a finite number of real eigenvalues.
Recently, the PT phase transition has been observed in laboratory experiments on
many different PT -symmetric systems.[4–12] The coupled-electronic-oscillator experiment
by J. Schindler et al. [11] is particularly elegant because of its simplicity, and we were
motivated by this experiment to construct its mechanical analog by using a pair of coupled
driven pendula. We have been able to do this, and in this mechanical system the PT phase
transition is easy to observe.
Mechanical oscillators (coupled pendula) have been used in other contexts to model com-
plex physical phenomena. For example, in particle physics, it was proposed that transitions
in K0 K̄0 , B0 B̄0 , and D0 D̄0 systems could be visualized by using coupled pendula.[13] These
experiments have recently been performed.[14, 15] In these experiments, systems of cou-
pled, damped pendula were used. The experiment reported in this paper differs from these
experiments in that while one of our pendula is damped, the other is undamped (driven).
This paper is organized as follows. Section II presents an intuitive explanation of the
PT phase transition. The pendulum experiment is described in Sec. III. Section IV gives
concluding remarks and proposes some follow-up experiments that we believe can be readily
performed.

2
II. INTUITIVE EXPLANATION OF THE PT PHASE TRANSITION

Originally, PT -symmetric quantum mechanics was studied at a highly mathematical

level. In Ref. [1] the techniques of complex variables, asymptotics, differential equations,
and perturbative theory were used to understand the analytic continuation of eigenvalue
problems and PT symmetry. The rigorous proof that the eigenvalues of H in (1) are real
and positive when ε ≥ 0 involves the use of sophisticated mathematical techniques (Baxter
TQ relation, Bethe ansatz, functional determinants), which are used to study integrable
systems and conformal field theory.[16, 17]
However, as experiments on PT -symmetric physical systems have been performed and
published, we have realized that the PT phase transition can be explained simply and
intuitively without resorting to sophisticated mathematics. We present this explanation
below.
Imagine two identical closed and isolated boxes, one on the negative-x axis at x = −a,
and the other on the positive-x axis at x = a (see Fig. 1). Inside the left box is a sink (an
antenna that absorbs energy) and inside the right box is a source (an antenna that radiates
energy at an equal rate). This system is PT symmetric because under parity P the left
box and the right box interchange positions and under time reversal T the sink becomes a
source and the source becomes a sink.

FIG. 1: A simple schematic PT -symmetric physical system: A box located at x = −a contains a

sink (an antenna that absorbs) and a box located at x = +a contains a source (an antenna that
radiates at an equal rate). Under space reflection P the boxes interchange, and under time reversal
T the sink becomes a source and the source becomes a sink. Thus, the system is PT symmetric.
If the boxes are isolated, the system cannot be in equilibrium because the energy decays to zero
in the left box and grows to infinity in the right box. Thus, in this case the PT symmetry of the
system is broken. The energies corresponding to these runaway modes are complex. However, if
the boxes are sufficiently strongly coupled, the system can equilibrate and the PT symmetry is
unbroken. When the system is in equilibrium, its energy is real.

The Hamiltonian that describes the time evolution of the one-dimensional system in the
left box is the 1 × 1 matrix H = [E1 ] = aeiθ , where a > 0 and 0 < θ < π so that Im E1 > 0.
The solution to the time-dependent Schrödinger equation for this system,
d
−i φ(t) = Hφ(t), (4)
dt
is ψ(t) = ψ(0)eiE1 t , which decays exponentially with time because there is a sink in the left
box.

3
 Similarly, the Hamiltonian that describes the time evolution of the one-dimensional sys-
tem in the right box is the 1 × 1 matrix H = [E2 ] = ae−iθ , so that ImE2 < 0. The solution


to the time-dependent Schrödinger equation for the right system is ψ(t) = ψ(0)eiE2 t , which
grows exponentially with time because there is a source in the right box.
The left and right systems taken together are described by the 2 × 2 diagonal matrix
Hamiltonian  iθ 
ae 0
H= . (5)
0 ae−iθ
This Hamiltonian is not Dirac Hermitian and its eigenvalues E1 and E2 = E1∗ are complex.
However, it is PT symmetric, where the parity operator P is the matrix
 
0 1
P= , (6)
1 0

which interchanges the two systems, and the time reversal operator T is complex conjuga-
tion. The system described by H in (5) is not in equilibrium because the eigenfunction in
the left box decays exponentially and the eigenfunction in the right box grows exponentially.
Correspondingly, E1 and E2 , the eigenvalues of H, are complex. Thus, the system is in a
broken-PT -symmetric phase.
The combined two-box system can achieve equilibrium if we couple the boxes together
sufficiently strongly. We couple the boxes symmetrically in order to preserve the PT sym-
metry of the entire system:  iθ 
ae g
H= , (7)
g ae−iθ
where the coupling constant g is real. This coupling allows the excess energy in the right
box to leak into the left box.
The eigenvalues of H in (7) become real if g 2 > a2 sin2 θ. Thus, if the coupling is strong
enough, the excess energy in the left box can flow into the right box fast enough for the
system to remain in equilibrium. When this happens, the energy eigenvalues are real and
the system is in an unbroken-PT -symmetric phase. The phase transition occurs when the
coupling constant g exceeds the critical value given by
2
gcrit = a2 sin2 θ. (8)

Using this simple two-dimensional model, we can now understand heuristically why the
Hamiltonians in (1) have an unbroken PT symmetry when ε > 0 and a broken PT symmetry
when ε < 0. Consider, for example, the potential ix3 for H in (2); this potential has a
positive-imaginary part when x > 0 and a negative-imaginary part when x < 0. Thus, there
is a nonlocal source everywhere on the negative-x axis and a corresponding nonlocal sink
everywhere on the positive-x axis. Can such a system actually be in equilibrium? One might
think that it would be impossible for this system to reach equilibrium because the source
becomes infinitely strong as x → −∞ and the sink becomes infinitely strong as x → +∞.
However, by means of a simple classical argument, we can easily show that the system can
indeed achieve equilibrium: We determine how long it takes for particles at x = −∞ to flow
to x = +∞. The classical time of flight T is given by
Z Z Z ∞
dx dx
T = dt = = √ , (9)
p x=−∞ E − ix3

4
where we have used the condition that the Hamiltonian represents the classical energy:
E = p2 + ix3 . Since the integral in (9) converges, the time of flight from x = −∞ to
x = +∞ is finite, and thus the system can attain equilibrium. Clearly, if ε in (1) becomes
negative, the time of flight integral diverges and the system does not have enough time to
reach equilibrium. This is why the PT symmetry of H breaks and the eigenvalues become
complex when ε < 0.
This simple two-box source-and-sink model explains what was done in the microwave
cavity experiment in Ref. [12]. In this experiment two coupled microwave cavities, one
containing a source of microwaves and the other containing a sink of microwaves was used,
and as the coupling was varied, the PT phase transition was observed.
The source-and-sink model also explains the two-channel optics experiments in Refs. [6]
and [7]. In these experiments light travels down a coupled pair of wave guides. One wave
guide has loss and the other has gain. If the coupling of the wave guides is sufficiently strong,
then the system is in equilibrium, and one observes Rabi oscillations (power oscillations)
in which the optical energy oscillates between the two wave guides. When the coupling
between the wave guides becomes too weak, the Rabi oscillations cease and the system can
no longer remain in equilibrium; the power grows exponentially in one wave guide and decays
exponentially in the other.
The experiment in Ref. [11] is the electronic analog of the two-channel optics experiments.
The electronic experiment involves two inductively-coupled LRC circuits, one with gain and
the other with loss. Once again, as the coupling between the two oscillators becomes weaker
than a critical value, the system enters the PT -broken phase and is no longer in equilibrium.

III. DESCRIPTION OF THE EXPERIMENT

Before discussing the classical experiment described in this paper, we mention briefly
that the connection between PT -symmetric quantum systems and PT -symmetric classical
systems is simply that they are both described by a Hamiltonian H that exhibits PT
symmetry; that is, H is invariant under combined P and T reflection, that is, combined
space and time reflection. In the quantum case the variables x and p are operators, while
in the classical case they are numbers. A complete description of this connection may be
found in Ref. [18].
The experiment discussed here is a classical-mechanical analog of the electronic exper-
iment in Ref. [11]. Instead of two LRC circuits, there are two coupled pendula, one with
energy loss and one with energy gain. We first formulate some elementary equations to
model a coupled a two-pendulum system and then we describe the experiment itself.

A. An overly simple mathematical model of a two-oscillator system

The Hamiltonian
H = 21 p2 + 21 x2 + 12 q 2 + 12 y 2 + εxy (10)
describes the small-amplitude frictionless motion of two coupled pendula, where x(t) and
y(t) are the displacements of the pendula and p(t) and q(t) are the corresponding conjugate
momenta. The parameter ε represents the coupling of the two pendula. The classical

5
equations of motion for x(t) and y(t) are
x00 (t) = −x(t) − εy(t),
y 00 (t) = −y(t) − εx(t). (11)
If we solve these equations numerically, we observe Rabi oscillations (see Fig. 2).
A simple, but as we will see not a good, way to create a PT -symmetric system is to
introduce a damping (loss) term in the x equation and an undamping (gain) term in the y
equation:
x00 (t) + ax0 (t) + x(t) + εy(t) = 0,
y 00 (t) − ay 0 (t) + y(t) + εx(t) = 0, (12)
where the damping/undamping parameter a is positive. This system is PT symmetric
because the x and y oscillators are interchanged under P (just as the boxes in Fig. 1 are
interchanged under P) and the signs of the damping and undamping terms reverse under
time reversal T .
To solve (12) analytically we rewrite these equations as a system of four first-order
constant-coefficient differential equations:
x0 (t) = p(t),
p0 (t) = −x(t) − εy(t) − ap(t),
y 0 (t) = q(t),
q 0 (t) = −y(t) − εx(t) + aq(t). (13)
We then express these differential equations compactly in matrix form:
V0 (t) = MV(t), (14)
where the matrix M and the vector V are
   
0 1 0 0 x(t)
 −1 −a −ε 0 and V =  p(t)  .
 
M =  0 0 0 (15)
1  y(t) 
−ε 0 −1 a q(t)

To solve (15) we must find the eigenvalues E of the matrix M . To do so we calculate the
determinant D of the matrix M − EI:
 
−E 1 0 0
 −1 −a − E −ε 0 
D = det  . (16)
 0 0 −E 1 
−ε 0 −1 a − E

The determinant D is a simple quadratic form in E 2 ,

D = E 4 + (2 − a2 )E 2 + 1 − ε2 . (17)
The solution to the equation D = 0 is
 √ 
E 2 = 21 a2 − 2 ± a4 − 4a2 + 4ε2 . (18)

6
FIG. 2: Oscillatory motion of a pair of coupled pendula [see (11)] with no loss or gain. The
x displacement is shown in the left panel and the y displacement is shown in the right panel.
The coupling parameter is ε = 0.075. Rabi power oscillations occur in which the maximum and
minimum displacements x(t) and y(t) are 90◦ out of phase.

The system (13) exhibits oscillatory behavior if E 2 < 0. Two conditions must be met for
E 2 to be negative:

(1) a4 − 4a2 + 4ε2 > 0,

√
(2) a2 − 2 + a4 − 4a2 + 4ε2 < 0. (19)

Condition (2) immediately gives ε < 1. Then, Condition (1) implies that
 √ 
2
a < acrit = 2 1 − 1 − ε . (20)

Therefore, if the damping/undamping parameter a is below this critical value, the eigenval-
ues are imaginary and the system is in the unbroken-PT -symmetric region.
As the damping/undamping parameter a increases from 0, the frequency of the Rabi
power oscillations increases. Eventually, at the critical critical value of a the Rabi frequency

7
is the same as the oscillation frequency. Above this value of a we enter the region of broken
PT symmetry; the Rabi oscillations cease and we observe the onset of exponential behavior.
However, a numerical plot reveals the shortcomings of the overly simple model (12). In Fig. 3
we plot the x(t) and y(t) for the case a = ε = 0.075. Observe that y(t) grows exponentially,
but after an initial period of exponential decay x(t) also grows exponentially. This is because
a balanced damping and undamping violates energy conservation. For the model equations
(12), the rate of growth of y(t) due to undamping is proportional to the velocity of y(t)
and the rate of decay of x(t) is proportional to the velocity of x(t). Thus, as y(t) gets big
and x(t) gets small, the loss of energy in the x oscillator becomes insignificant while the
growth of energy in the y oscillator becomes immense. Then, because the two oscillators are
coupled, x(t) eventually becomes large because the energy in the y oscillator leaks into the
x oscillator.

B. An improved energy-conserving oscillator model

Evidently, we need to construct a better model in which the source antenna radiates the
same amount of energy as is absorbed by the sink antenna. We do this by constructing a pair
of difference-differential equations: We begin with the coupled-oscillator equations in (11)
in which there is no loss or gain. We then impose the additional condition that whenever
x(t) reaches an amplitude maximum, we remove a given fixed fraction of the energy g from
the x oscillator (by decreasing the amplitude appropriately and reinitializing the motion).
Then, when y(t) is next at a maximum we transfer this exact amount of energy to the y
oscillator at its peak by increasing its amplitude accordingly.
Numerical solutions for this improved oscillator model are shown in Figs. 4 and 5. In
Fig. 4 we show what happens if we take ε = 0.05 and take g = 0.01. Here, we remove 1% of
the energy of the x oscillator each time it reaches its maximum amplitude and then transfer
this energy to the y oscillator when it reaches its next amplitude
√ maximum; this transfer
reduces the amplitude of the x oscillator at its maximum to 0.99 = 99.5% of its former
value. The coupling ε = 0.05 is large enough to keep the system in the region of unbroken
PT symmetry. The signal that the PT symmetry is not broken is that the Rabi oscillations
persist and the amplitudes of the oscillators remain constant.
In Fig. 5 we take the coupling smaller ε = 0.01 and g larger g = 0.3. [Now, at each swing
only 70% of the energy√in the x oscillator remains; when we make this transfer we reduce
the peak amplitude to 0.7 = 83.7% of its former value. These amplitude reductions are
clearly visible in the graph of x(t).] In this case the energy transfer overwhelms the small
coupling of the oscillators and we are in the broken-PT -symmetric region. The signal that
the PT symmetry is broken is that the Rabi oscillations cease; when this happens, the x
(and y) oscillators are no longer in equilibrium because the amplitudes of these oscillators
take infinitely long to decrease (and increase) to their limiting values.

C. Description of the experimental setup

The experiment approximates the energy-conserving mathematical oscillator model very

closely. Two identical pendula are suspended from a horizontal rope of total length 94 cm
(see Fig. 6). By adjusting the tension in the horizontal rope, we can vary the coupling of the
two pendula; increasing (decreasing) the tension in the horizontal rope decreases (increases)

8
FIG. 3: Numerical solution to the overly simple coupled-oscillator equations (12) for the parameter
choice a = ε = 0.075, which is deep in the broken-PT -symmetric region. The x displacement is
shown in the left panel and the y displacement is shown in the right panel. As expected, the y
oscillator exhibits gain, but after an initial period of decay, the x oscillator also exhibits gain. This
is because the system does not conserve energy, and the excess energy in the y oscillator leaks into
the x oscillator.

the coupling of the pendula. Each pendulum consists of a 50-gram cylindrical mass (diameter
2.3 cm, height 4.4 cm) hanging from a string of length 39 cm. The separation of the two
pendula is 34 cm.
In order to add or subtract energy from a pendulum, an electromagnet is situated about
5 cm from the top of each string. When the electromagnet is turned on, it applies a hor-
izontal force to a small iron nail that is attached to the string supporting the bob. The
electromagnets are only on for about 10 msec, so that they provide a brief impulse to the
string. The impulse is applied to the left pendulum when the pendulum is moving away
from the electromagnet, so the effect of the impulse is to subtract a small amount of kinetic
energy from the pendulum. However, the impulse is applied to the right pendulum when
the pendulum is moving towards the electromagnet, so the effect of the impulse is to add a
small amount of kinetic energy to the pendulum.

9
 Each of the electromagnets is triggered by a pair of optical sensors located just above the
pendulum bobs. For the electromagnet to fire, the two optical sensors must be triggered in
a prescribed order, so the output of the optical sensors is fed into a simple logic circuit.
To record the motion of the pendula we use a small video camera mounted on a goose-neck
stand.[19] The camera looks up from underneath each pendulum and records the instanta-
neous position of the pendulum 15 times per second. We then draw a curve through these
data points to display the motion of the pendulum.
We begin our experiment by taking data with the magnets turned off (see Fig. 7). The
tension in the horizontal rope is 200 grams. The key characteristic feature of the swinging
pendula is that they exhibit Rabi power oscillations. The experimental data are in qualitative
agreement with the theoretical predictions in Fig. 2. [Note that the simple theoretical model
(12) with vanishing damping/undamping parameter a is the same as the improved energy-
conserving oscillator model with vanishing energy-transfer parameter g.]
Next, we turn on the magnets weakly and decrease the coupling of the pendula by in-
creasing the tension in the string from which the pendula hang from 200 to 400 grams. The
experimental data are shown in Fig. 8. Note that the Rabi power oscillations persist. Again,
the experimental data are in qualitative agreement with the theoretical predictions in Fig. 4.
We conclude that we are still in the region of unbroken PT symmetry.
Finally, we make the magnets stronger and increase the tension in the string from 400
grams to 600 grams in order to further weaken the coupling of the pendula. The experimental
data are shown in Fig. 9. If we compare these data to the theoretical predictions in Fig. 5, we
again observe good qualitative agreement: The Rabi power oscillations have ceased and we
can see that the amplitudes of the pendula level off and rapidly approach their asymptotic
values. We conclude that we have entered the region of broken PT symmetry.

IV. FINAL REMARKS

In summary, the phase transition between an unbroken and a broken PT -symmetric

phase is easy to explain at an intuitive level. It is simply a matter of whether two systems,
one with gain and the other with loss, are coupled strongly enough to be in equilibrium. The
phase transition, which takes place at a critical value of the coupling, can be observed in a
mechanical system consisting of two coupled pendula, one with damping and the other with
undamping. In the unbroken phase we observe periodic Rabi power oscillations. However,
when the damping/undamping becomes strong relative to the strength of the coupling, the
system enters a broken-PT -symmetric phase. At this transition the Rabi power oscillations
cease, and the amplitudes of the pendula approach limiting values.
A natural extension of the experimental work described in this paper to investigate PT -
symmetric chains of coupled pendula with alternating damping and undamping. It is pre-
dicted that an optical system consisting of coupled optical fibers with alternating loss and
gain will exhibit birefringence.[20] Our objective in studying this multiple pendulum system
is to observe the mechanical equivalent of birefringence.
We have also begun to study PT -symmetric systems of coupled acoustic wave guides.
We have already found that coupled acoustic wave guides with alternating loss and gain can
exhibit a rich and elaborate array of phase transitions.
We thank Profs. R. Pike and J. Schilling for many useful discussions. We also thank Prof.
Pike for the use of his laboratory facilities. CMB thanks the U.K. Leverhulme Foundation
and the U.S. Department of Energy for financial support.

10
FIG. 4: Mathematical simulation of the improved energy-conserving oscillator model in the
unbroken-PT region. In this graph ε = 0.05 and g = 0.01. (That is, 1% of the energy in the
x oscillator is removed each time x reaches a peak, and this exact amount of energy is then trans-
ferred to the y oscillator when it reaches a peak. This transfer reduces the peak-x amplitude to
99.5% of its former value, which is too small to be seen on this graph.) The motion of the x
oscillator is shown in the left panel and the motion of the y oscillator is shown in the right panel.
We can see that the PT symmetry is not broken: The Rabi oscillations persist and the amplitudes
of the oscillators remain constant.

[1] C. M. Bender and S. Boettcher, “Real spectra in non-Hermitian Hamiltonians having PT

symmetry,” Phys. Rev. Lett. 80, 5243-5246 (1998).
[2] C. M. Bender, D. C. Brody, and H. F. Jones, “Complex extension of quantum mechanics,”
Phys. Rev. Lett. 89, 270401 (2002) (4 pages).
[3] C. M. Bender, D. C. Brody, and H. F. Jones, “Must a Hamiltonian be Hermitian?,” Am. J.
Phys. 71, 1095-1102 (2003).
[4] C. Zheng, L. Hao, and G. L. Long, “Observation of a fast evolution in a parity-time-symmetric

11
FIG. 5: Mathematical simulation of the improved energy-conserving oscillator model in the broken-
PT region. The coupling here is ε = 0.01, which is smaller than that in Fig. 4. Also, g = 0.3, which
is larger than that in Fig. 4. For this value of g, 30% of the energy in the x oscillator is removed each
time x reaches a peak, and this exact amount of energy is then transferred to the y oscillator when
it reaches a peak. This transfer reduces the peak amplitude to 0.837 of previous value, and this
change can be seen in the plot of x(t). The motion of the x oscillator is shown in the left panel and
the motion of the y oscillator is shown in the right panel. Observe that the Rabi oscillations cease
and that the x oscillations die down to a limiting amplitude and correspondingly the y oscillations
increase to a limiting amplitude. This is the characteristic behavior of an oscillator system having
a broken PT symmetry.

system,” arXiv:1105.6157 (4 pages).

[5] J. Rubinstein, P. Sternberg, and Q. Ma, “Bifurcation diagram and pattern formation of phase
slip centers in superconducting wires driven with electric currents,” Phys. Rev. Lett. 99,
167003 (2007) (4 pages).
[6] A. Guo, G. J. Salamo, D. Duchesne, R. Morandotti, M. Volatier-Ravat, V. Aimez,
G. A. Siviloglou, and D. N. Christodoulides, “Observation of PT -symmetry breaking in com-
plex optical potentials,” Phys. Rev. Lett. 103, 093902 (2009) (4 pages).
[7] C. E. Rüter, K. G. Makris, R. El-Ganainy, D. N. Christodoulides, M. Segev, and D. Kip,

12
FIG. 6: View of the two-pendulum experiment. Two pendula are suspended from a horizontal
rope and the tension in the rope is adjusted to increase or decrease the coupling of the pendula.
The horizontal rope runs around a wheel to the left (not shown) and is attached to a tray upon
which weights can be added or subtracted to change the tension. Electromagnets near the top of
the strings supporting the bobs apply brief impulses to small iron nails attached by white tape to
the strings. The electromagnets are triggered by pairs of optical sensors just above the pendulum
bobs. The electromagnets are timed so that on each swing a small amount of kinetic energy is
subtracted from the left pendulum and a roughly equal amount of kinetic energy is added to the
right pendulum.

“Observation of PT-symmetry in optics,” Nat. Phys. 6, 192-195 (2010).

[8] K. F. Zhao, M. Schaden, and Z. Wu, “Enhanced magnetic resonance signal of spin-polarized
Rb atoms near surfaces of coated cells,” Phys. Rev. A 81, 042903 (2010) (11 pages).
[9] Z. Lin, H. Ramezani, T. Eichelkraut, T. Kottos, H. Cao, and D. N. Christodoulides, “Uni-
directional invisibility induced by PT -symmetric periodic structures,” Phys. Rev. Lett. 106,
213901 (2011) (4 pages).
[10] L. Feng, M. Ayache, J. Huang, Y.-L. Xu, M.-H. Lu, Y.-F. Chen, Y. Fainman, and A. Scherer,
“Nonreciprocal light propagation in a silicon photonic circuit,” Science 333, 729-733 (2011).
[11] J. Schindler, A. Li, M. C. Zheng, F. M. Ellis, T. Kottos, “Experimental study of active LRC
circuits with PT symmetries,” Phys. Rev. A 84, 040101(R) (2011) (5 pages).
[12] S. Bittner, B. Dietz, U. Günther, H. L. Harney, M. Miski-Oglu, A. Richter, and
F. Schäfer, “PT symmetry and spontaneous symmetry breaking in a microwave billiard,”

13
FIG. 7: Experimentally measured motion of the pendula with the magnets turned off. The tension
in the string is 200 grams. To produce these graphs we use a camera that records the instantaneous
position of each pendulum (tick marks on the graph) 15 times per second. We then fit a curve
through these data points. The motion of the pendula is qualitatively similar to that in the
theoretical curves in Fig. 2. In this configuration the PT symmetry is unbroken. The signal for
an unbroken PT symmetry is the presence of Rabi power oscillations, which are 90◦ out of phase.
One can observe a slight decay in the amplitudes of the pendula due to friction.

Phys. Rev. Lett. 108, 024101 (2012) (5 pages).

[13] J. L. Rosner and S. A. Slezak, “Classical illustrations of CP violation in Kaon decays,”
Am. J. Phys. 69, 44-49 (2001).
[14] K. R. Schubert and J. Stiewe, “Demonstration of K0 K0 , B0 B0 and D0 D0 transitions with a
pair of coupled pendula,” J. Phys. G: Nucl. Part. Phys. 39, 033101 (2012) (12 pages).
[15] A. Reiser, K. R. Schubert, and J. Stiewe, “Translation of time-reversal violation in the neutral
K-meson system into a table-top mechanical system,” arXiv: 1203.4703 (11 pages); see also V.
A. Kostelecký and Á. Roberts, “Analogue models for T and CPT violation in neutral-meson
oscillations,” Phys. Rev. D 63, 096002 (2001) (8 pages).

14
FIG. 8: Experimental data showing the motion of the pendula when the magnets are turned on
weakly, so that there is a weak loss and gain, and the pendula are strongly coupled (the tension in
the supporting rope is lowered to 400 grams). Observe that the Rabi power oscillations in Fig. 7
persist. This means that the system is in a region of unbroken PT symmetry.

[16] P. Dorey, C. Dunning, and R. Tateo, “Supersymmetry and the spontaneous breakdown of
PT symmetry,” J. Phys. A 34, L391-L400 (2001) and “Spectral equivalences, Bethe ansatz
equations, and reality properties in PT -symmetric quantum mechanics,” ibid. 34, 5679-5704
(2001).
[17] P. Dorey, C. Dunning, and R. Tateo, “The ODE/IM correspondence,” J. Phys. A:
Math. Gen. 40, R205-R227 (2007).
[18] C. M. Bender, S. Boettcher, and P. N. Meisinger, “PT -Symmetric Quantum Mechanics,”
J. Math. Phys. 40, 2201-2229 (1999); C. M. Bender, “Introduction to PT -Symmetric Quantum
Theory,” Contemp. Phys.46, 277-292 (2005); C. M. Bender, “Making Sense of Non-Hermitian
Hamiltonians,” Reps. Prog. Phys. 70, 947-1018 (2007).
[19] Details of the video camera system: Single user VisiLog, full version with advanced USB video
camera and goose-neck stand, supplied by 2012 ScienceScope.

15
FIG. 9: Experimental data showing the motion of the pendula when the magnets are turned on
strongly and the coupling of the pendula is weak (the tension in the supporting string is raised to
600 grams). Observe the the Rabi oscillations have ceased. This is the signal that the system is in
a region of broken PT symmetry.

[20] K. Makris, R. El-Ganainy, D. N. Christodoulides, and Z. H. Musslimani, “Beam dynamics in

PT -symmetric optical lattices,” Phys. Rev. Lett. 100, 103904 (2008) (4 pages).

16