Large scale physical effects of T violation in mesons

J. A. Vaccaro
Centre for Quantum Dynamics, Griffith University, 170 Kessels Road, Brisbane 4111, Australia.
(Dated: March 3, 2010)
An increasing number of experiments at the Belle, BNL, CERN and SLAC accelerators are con-
firming the violation of time reversal invariance (T). The violation signifies a fundamental asymmetry
between the past and future and calls for a major shift in the way we think about time. Here we
show that processes which violate T symmetry induce destructive interference between different
paths that the universe can take through time. The interference eliminates all paths except for two
that represent continuously forwards and continuously backwards time evolution. Evidence from
the accelerator experiments indicates which path the universe is effectively following. This work
resolves the long-standing problem of modeling the dynamics of T violation processes. It shows that
T violation has previously unknown, large-scale physical effects and that these effects underlie the
origin of the unidirectionality of time. It also provides a view of the quantum nature of time itself.

PACS numbers: 03.65.-w, 11.30.Er, 14.40.-n

Keywords: CP violation, T violation, kaons, arrow of time

I. INTRODUCTION towards our past, in one quantum equation of motion?

The physical nature of time has been an enigma for
centuries. The main tools for discussing its unidirection-
ality are the phenomenological arrows of time. A great
deal of progress has been made in recent decades in link-
ing various arrows together [1]. A notable exception is
the matter-antimatter arrow which arises from the vi-
olation of charge-parity conjugation invariance (CP) in
meson decay. CP violation, which was first discovered
by Cronin, Fitch and coworkers [2] in 1964 in the de-
cay of neutral kaons (K mesons), provides a clue to the
origin of the large-scale matter-antimatter imbalance of
the universe [3]. The combined CPT invariance has been
confirmed in kaon decay to a high degree of precision
[4] suggesting that neutral kaon decay also violates T
invariance. To date no violation of CPT has been ob-
served in B meson decay [5] and so the CP violation in B
meson decay also appears to be consistent with T viola-
tion. But more importantly, experimental evidence has
confirmed the direct T violation in K meson decay [6]
independently of CPT invariance. Nonetheless, T violat-
ing processes are relatively rare and the magnitudes of
the violations are relatively small, and so despite their
important asymmetric temporal nature, they are often
regarded as having little impact on the nature of time.
As a result, the physical significance of T violation has
remained obscure.
The CP violation due to the weak interaction is de-
scribed in the Standard Model of particle physics by the
Cabibbo-Kobayashi-Maskawa (CKM) matrix [7, 8]. Nev-
ertheless, a fundamental and intriguing question remains
open. In principle, the CKM model provides the basis for
the time evolution of mesons in terms of Schrödinger’s
equation and an associated Hamiltonian H. The evi-
dence of T violation implies that T HT −1 6= H where
T represents the time reversal operation [9]. But how
should one incorporate the two Hamiltonians, H for time
evolution towards our future and T HT −1 for evolution
At issue is the fact that Schrödinger’s equation describes
consistent unitary evolution in both directions of time:
the unitary evolution by negative time intervals merely
“back tracks” exactly the same time evolution described
by positive time intervals. Yet, paradoxically, the T vio-
lation appears to suggests that Nature does not work in
this way in general.
More specifically, we must choose the direction of time
evolution and a specific version, say H, of the Hamilto-
nian in order to write down the Schrödinger equation.
By applying the time reversal operation we then obtain
the Schrödinger equation involving the T HT −1 version of
the Hamiltonian for evolution in the opposite direction
of time. But at any point we are restricted to having
a dynamical equation of motion for a specific direction
of time evolution and a corresponding specific version of
the Hamiltonian. We do not have a dynamical equation
of motion for the situation where the direction of time
evolution cannot be specified and for which there is no ar-
gument for favoring one version of the Hamiltonian over
the other: if one version of the Hamiltonian is to appear
in the equation of motion then so must the other. This
problem becomes critical when we attempt to describe
the universe as a closed system as needed in cosmology.
In this case the direction of time evolution cannot be
specified because a closed system precludes any external
clock-like device for use as a reference for the direction
of time. There is, therefore, no satisfactory quantum for-
malism for describing a universe that exhibits T violation
processes. This failure of standard quantum theory is all
the more serious given that CP and, by implication, T
violation play a critical role in baryogenesis in the early
history of the universe [3, 10]. It is clearly an important
problem that lies at the very foundations of quantum the-
ory. Its resolution calls for a major shift in the way we
think about time and dynamical equations of motion.
We address this issue by deriving a single quantum
dynamical equation which incorporates the two versions

of a T-violating Hamiltonian. We begin with Feynman’s
path integral [11] method to construct the set of all pos-
sible paths that a large physical system can take through
time. The set represents every possible evolution zigzag-
ging forwards and backwards through time. We then
show how T violation induces destructive interference
that eliminates most of the possible paths that the sys-
tem can follow. The unidirectionality of time is shown
to emerge from physical evidence left in the state of the
system. We end with a discussion.
II. POSSIBLE PATHS THROUGH TIME
Consider a physical system which is composed of mat-
ter and fields in a manner which is consistent with the
visible portion of the universe. For clarity we assume
that the system is closed in the sense that it does not
interact with any other physical system, although this
assumption could be removed without affecting the main
conclusions drawn here. We also assume that there is no
clock external to our system and so our analysis needs to
be unbiased with respect to the direction of time evolu-
tion. Nevertheless it is convenient to differentiate two di-
rections of time as “forward” and “backward”, although
we stress that neither direction necessarily has any con-
nection to our actual everyday experience. Also, we be-
gin our analysis with the system in a state |ψ0 i which
we shall call the “seed state” without reference to time.
With this in mind we express the evolution of the sys-
tem in the forwards direction over the time interval τ as
|ψF (τ )i = UF (τ )|ψ0 i where
UF (τ ) = exp(−iτ HF ) (1)
is the forwards time evolution operator, and HF is the
Hamiltonian for forwards time evolution. Throughout
this article we use units in which h̄ = 1. If time rever-
sal symmetry were obeyed, the evolution in the opposite
direction of time would be given by exp(iτ HF ). To ac-
commodate the time asymmetry of T violation processes,
we need to replace HF in this expression with its time
reversed form. Thus we define the state of the system af-
ter backwards evolution over a time interval of the same
magnitude by |ψB (τ )i = UB (τ )|ψ0 i where
UB (τ ) = exp(iτ HB ) , (2)
and HB = T HF T −1 is the Hamiltonian for backwards
time evolution. This gives UB (τ ) = T UF (τ )T −1 , in ac-
cord with Wigner’s definition of time reversal [9].
The parameter τ in these expressions represents a time
interval that could be observed using clock devices which
are internal to the system. A simple model of a clock for
our purposes is a device which evolves unitarily through
a sequence of orthogonal states at sufficient regularity to
give a measure of a duration of time to any desired ac-
curacy. The orthogonal states would represent the eigen-
states of a pointer observable which acts a reference for
2
time. A harmonic oscillator is a device of this kind [12].
In order to avoid ambiguities in the definition of time in-
tervals for different time directions, we assume that the
Hamiltonian H (clk) describing our clock model is T in-
variant and so T H (clk) T −1 = H (clk) during its normal
operation. This gives an unambiguous operational mean-
ing of the parameter τ as a time interval.
The matrix elements hφ|UF (τ )|ψ0 i and hφ|UB (τ )|ψ0 i
represent the probability amplitudes for the system in
state |ψ0 i to evolve over the time interval τ to |φi via
two paths in time corresponding to the forwards and
backwards directions, respectively. Given that we have
no basis for favoring one path over the other, we follow
Feynman [11] and attribute an equal statistical weight-
ing to each. Thus, the total probability amplitude for the
system to evolve from one given state to another is pro-
portional to the sum of the probability amplitudes for all
possible paths through time between the two states. In the
current situation we have only two possible paths and so
the total amplitude is proportional to
hφ|UF (τ )|ψ0 i + hφ|UB (τ )|ψ0 i = hφ|UF (τ ) + UB (τ )|ψ0 i .
(3)
This result holds for all states |φi of the system and so
the unbiased time evolution of |ψ0 i over the time interval
τ can be written as
|Ψ(τ )i = [UF (τ ) + UB (τ )] |ψ0 i (4)
which we call the time symmetric evolution of the sys-
tem. To reduce unnecessary detail we use un-normalized
vectors to represent states of the system. We could have
included in Eq. (3) an additional phase factor eiθ to repre-
sent an arbitrary relative phase between the two paths,
such as hφ|UF (τ )|ψ0 i + eiθ hφ|UB (τ )|ψ0 i, but the factor
has no net physical effect for our analysis and so we omit
it.
It follows that the time symmetric evolution of the
system in state |Ψ(τ )i over an additional time interval
of τ is given by
|Ψ(2τ )i = [UF (τ ) + UB (τ )] |Ψ(τ )i
2
= [UF (τ ) + UB (τ )] |ψ0 i .
Repeating this for N such time intervals yields
N
|Ψ(N τ )i = [UF (τ ) + UB (τ )] |ψ0 i . (5)
Figure 1 gives a graphical interpretation of this result
in terms of a binary tree. It is useful to write the expan-
sion of the product on the right side as
N
X
|Ψ(N τ )i = SN −n,n |ψ0 i (6)
n=0
where Sm,n represents a sum containing (n+m n ) different
terms each comprising n factors of UF (τ ) and m factors
 3

|y0 We now derive a more manageable expression for Sm,n .

The first step is to reorder the expression so that all the
UB UF UB factors are to the left of the UF factors using the
Zassenhaus formula [17]. This yields
UB UF UB UF
m
X s X̀
X £
Sm,n = UB (mτ )UF (nτ ) ··· exp (v + · · · + ` + k)τ 2 [
UB UF UB UF UB UF v=0 `=0 k=0

UB UF UB UF UB UF UB where there are n summations on the right side and [A, B]

UF
is the commutator of A and B. Appendix A gives the
A B C details of the calculation. Next, the analysis is made
simpler by expressing the summand in the eigenbasis
FIG. 1: Binary tree representation of the generation of the fi- of the Hermitian operator i[HF , HB ]. For this R we use
nal state |Ψ(N τ )i from the seed state |ψ0 i according to Eq. (5) the resolution of the identity given by 1 = Π(λ)dλ
with N = 4. States are represented as nodes (solid discs) and where the measure dλ accommodates both continuous
unitary evolution by links (arrows) between them. The root and discrete spectra and Π(λ) is the projection opera-
node (at the top) represents the seed state |ψ0 i and the leaf tor that projects onto the state space spanned by the
nodes (on the bottom row) represent components of the final eigenstates of i[HF , HB ] with eigenvalue λ. The degen-
state |Ψ(4τ )i, where for example, the node labeled A rep- eracy of the eigenvalue λ is given by the density function
4
resents the state S4,0 |ψ0 i = UB |ψ0 i, B represents the state
3 2 2 3 ρ(λ) = Tr[Π(λ)] where Tr is the trace. To emphasize the
S3,1 |ψ0 i = [UB UF + UB UF UB + UB UF UB + UF UB ]|ψ0 i and
4
C represents S0,4 |ψ0 i = UF |ψ0 i.
degeneracy of λ we write the identity operator as
Z
1 = ρ(λ)Π(λ)dλ (9)
of UB (τ ), and where (kj ) = k!/[(k − j)!j!] is the binomial
coefficient. Sm,n is defined by the recursive relation where Π(λ) = Π(λ)/ρ(λ) is an operator with unit trace.
R
Thus, for example, i[HF , HB ] = λρ(λ)Π(λ)dλ. Multi-
m
X plying Eq. (8) on the right by Eq. (9) and ignoring the
Sm,n = Sm−k,n−1 UF (τ )UB (kτ ) (7)
term of order τ 3 gives
k=0
Z
with Sm,n = UB (mτ )UF (nτ ) Im,n (λ)ρ(λ)Π(λ)dλ (10)
Pm Sm,0 = UB (mτ ) and gives, for example, Sm,1 =
k=0 UB [(m − k)τ ]UF (τ )UB (kτ ).
The expression hφ|SN −n,n |ψ0 i represents the evolution where
of the system from |ψ0 i to |φi over a set of (N n ) paths m
X s X̀
X
through time, where each path comprises a total of n £ ¤
steps in the forwards direction and N − n steps in the Im,n (λ) = ··· exp −i (v + · · · + ` + k)τ 2 λ .
v=0 `=0 k=0
backwards direction. The set of paths includes all possi-
(11)
ble orderings of the forwards and backwards steps. This
In the following analysis terms of order τ 3 are negligible
set of paths is the focus of the remaining analysis.
and are ignored. Performing the algebraic manipulations
First consider the size of the time steps τ . To explore described in Appendix B eventually yields
the consequences of the limit τ → 0 of infinitely-small
time steps for a fixed total time ttot , we set τ = ttot /N Qn−1 © ª
q=0 1 − exp[−i(n + m − q)τ 2 λ]
and consider Eq. (5) for increasing values of N . We can Im,n (λ) = Qn 2
. (12)
write 2−N [UF (τ ) + UB (τ )]N = {exp[−i 21 (HF − HB )τ ] + q=1 [1 − exp(−iqτ λ)]

O(τ 2 )}N which becomes exp[−i 12 (HF − HB )ttot ] as N →
∞. In this limit, the system evolves according to the
Hamiltonian 12 (HF − HB ). This means that isolated sub-
systems that obey T invariance, such as our model of a
clock, would not exhibit any evolution. As clocks are
taken to measure time intervals, the system as a whole
would not exhibit dynamics in the conventional sense in
this τ → 0 limit. The lack of dynamics is an unphysi-
cal result if our system is to model the visible universe.
This suggests that the time interval τ should be a small
non-zero number for a universe-like system. Accordingly,
in the following we set the value of τ to be the smallest
physically-reasonable time interval, the Planck time, i.e.
τ ≈ 5 × 10−44 s.
There are three important points to be made about the
results in Eqs. (10) and (12). The first is that if the sys-
tem seed state |ψ0 i is a λ = 0 eigenstate of the operator
i [HF , HB ] then the integral in Eq. (10) can effectively
be replaced with Im,n (0)ρ(0)Π(0) and the resulting evo-
lution is equivalent to the T invariant case. In order to
focus on the implications of T violation we assume that
the system seed state |ψ0 i has a zero overlap with all
λ = 0 eigenstates, i.e.
Π(0)|ψ0 i = 0 . (13)
We call this the nonzero eigenvalue condition for con-
venience. We look at the implications of relaxing this

condition in Section V. The second point is a useful
symmetry property of Im,n (λ). Multiplying both the nu-
merator and denominator of the right side of Eq. (12) by
Qn+m 2
k=1 [1 − exp(−ikτ λ)], canceling like terms and judi-
ciously relabeling indices yields the symmetry
Im,n (λ) = In,m (λ) . (14)
For the third point, it is straightforward to show for
n ≤ m that the zeroes of the denominator of Im,n (λ)
in Eq. (12) do not result in singularities, because if any
factor [1 − exp(−iqτ 2 λ)] in the denominator is zero for
λ = λ0 , there is a corresponding factor {1 − exp[−i(n +
m − `)τ 2 λ]} in the numerator, where ` = n + m − jq and
j is a positive integer, for which
2
1 − exp[−i(n + m − `)τ λ]
→1
1 − exp(−iqτ 2 λ)
as λ → λ0 . The symmetry of Im,n (λ) in Eq. (14) means
that the same result holds for the complementary case
n > m.
III. INTERFERENCE BETWEEN PATHS
We now analyze the behaviour of SN −n,n in Eq. (6) in
some detail. Let tF = nτ and tB = (N − n)τ be the ag-
gregate times that the system evolves forwards and back-
wards, respectively, where N is the total number of steps
in time and ttot = tF + tB = N τ is the corresponding
total time.
Certain features of IN −n,n (λ) are more apparent when
its modulus |IN −n,n (λ)| is written as the product of
diffraction grating interference functions, i.e.
¯ ¯
¯Yn sin(αq βq τ 2 λ) ¯
¯
|IN −n,n (λ)| = ¯ ¯ ,
q=1 sin(βq τ 2 λ) ¯
where αq = 1q (N + 1 − q) and βq = 2q . |IN −n,n (λ)| is a
periodic function with a period of 2π/τ 2 and has maxima
at λ = k2π/τ 2 of Πnq=1 αq = (N n ) for integer k. For the
times tF = nτ and tB = (N − n)τ to be physically mea-
surable intervals, both n and (N − n) need be very much
larger than unity. In these cases we can treat |IN −n,n (λ)|
as comprising narrow peaks at λ = k2π/τ 2 , for integer
k, and negligible elsewhere. Expanding |IN −n,n (λ)| as
a power series in λ gives the quadratic approximation
1 eters WN −n,n and λSD of IN −n,n (λ) and ρ(λ), respec-
|Im,n (λ)| ≈ (N 4 2
n )[1 − 24 n(N − n)(N + 1)τ λ ]. Fig. 2 il-
lustrates the accuracy of the approximation near λ = 0.
The zeroes of the quadratic approximation can be used as
a reasonable estimate of the width WN −n,n of the peaks.
This estimate is given by
s and so the integral in Eq. (10) is well approximated
24
WN −n,n = . plies that there is negligible destructive interference be-
τ 4 n(N − n)(N + 1)]
To estimate typical values for the eigenvalues λ we take
neutral kaon evolution as a prototypical T-violating pro-
cess. We first consider the problem for a single kaon for
4
1.0
0.8
N = 2000, n =100
|IN-n,n (l)|
0.6
N = 4000, n =400
0.4 N = 8000, n =400
0.2
0.0
0.0000 0.0006 2p-0.0006 2p
t2 l
FIG. 2: Comparison of the function |IN −n,n (λ)| (solid curves)
with the quadratic approximation (dashed curves) for various
values of N and n. All curves have been scaled to give a peak
value of 1.
which we label the eigenvalues and Hamiltonians with
(1)
a superscript (1). The eigenvalues λ(1) of i[HF , HB ]
(1)
in the 2 dimensional subspace spanned by the kaon and
anti-kaon states can be found using the phenomenolog-
ical model of Lee and Wolfenstein [13]. In Appendix C
we show that this gives λ(1) ≈ ±1017 s−2 (in units where
h̄ = 1) using empirical values of Yao et al. [18].
Next, we estimate the eigenvalues of the operator
i[HF , HB ] for the system containing M kaons. For this
we use the fact that the Hamiltonian of a non-interacting
set of subsystems is simply the sum of the Hamiltoni-
ans of the subsystems. Assuming that the collection of
kaons in our system do not interact we can approximate
each eigenvalue λ of the operator i [HF , HB ] as a sum
P (1) (1) (1) (1)
λ = j λj where λj is an eigenvalue of i[HF , HB ]
for the jth kaon. The values of λ range from −M |λ(1) |
to M |λ(1) |. The degeneracy of the eigenvalue λ = k|λ(1) |
for integer k is given by the density function ρ(λ) =
M !/( 12 M − k)!( 12 M + k)!. Note that 2−M ρ(λ) is approx-
imately a Gaussian distribution over λ with √ a mean of
zero and standard deviation of λSD = 21 M |λ(1) | and
so we shall take the typical values of λ to range over
−λSD , . . . , λSD . Setting M to be a fraction, f < 1, of the
average total number of particles√in the system, say 1080 ,
gives M ≈ f 1080 and so λSD ≈ f 1057 s−2 .
There are two important limiting cases to consider
that depend on the relative sizes of the width param-
tively, as illustrated in Fig. 3. The first is the regime
given by WN −n,n À λSD , i.e. relatively broad peaks
of IN −n,n (λ). In this limit IN −n,n (λ) can be treated
as approximately constant over the range |λ| < λSD
by (Nn )1 where 1 is the identity operator. This im-
tween the paths represented by hφ|SN −n,n |ψ0 i. The sec-
ond case is the regime WN −n,n ¿ λSD , i.e. relatively
narrow peaks of IN −n,n (λ). In the extreme version of
this, IN −n,n (λ) is zero except in the neighborhood of
 5

1.0 side of Eq. (6); these paths are effectively “pruned” from
|IN-n,n (l)| the set possible paths that the system can take.
|IN-n,n (l)| N = 500, n =100
0.8 N = 8000, n =400 It is the condition tF ∼ tB À τ that is responsible
for this destructive interference. So to find paths that
|IN-n,n (l)|, r(l)

0.6 survive the interference we need to consider times tF ¿

r(l)
t2 lSD=0.0002
0.4
0.2
t2 WN-n,n t2 lSD
0.0
0 0.0002
FIG. 3: Comparison of the density function ρ(λ) (dotted
curve) with the integrand function |IN −n,n (λ)| (solid curve)
and its approximation (dashed curve). The curves represent-
ing |IN −n,n (λ)| and its approximation have been scaled to give
a peak value of 1, whereas for clarity, the curve representing
ρ(λ) has been scaled to give a peak value of 0.5. The standard
deviation of the density function is given by λSD = 0.0002/τ 2 .
The curve |IN −n,n (λ)| for N = 500 and n = 100 has a rel-
atively broad peak whereas for N = 8000 and n = 400 the
peak is relatively narrow compared to the peak in the density
function curve.
λ = 0. The integral in Eq. (10) therefore becomes
(N
n )Π(0) ∝ Π(0), where Π(0) is the projection opera-
tor which projects onto the zero eigenvalue subspace.
This means SN −n,n = UB [(N − n)τ ]UF (nτ )(N n )Π(0).
But according to the nonzero eigenvalue condition in
Eq. (13), the seed state of the system has no overlap
with any zero-eigenstate state of the commutator and so
SN −n,n |ψ0 i = 0. In other words, the corresponding paths
tend to undergo complete destructive interference in this
regime, and make negligible contribution to the sum on
the right side of Eq. (6).
The physical implications of these two limiting cases
become apparent on considering both forwards and back-
wards aggregated intervals tF = nτ and tB = (N − n)τ
to be very much larger than the Planck time τ , i.e.
n, (N − n) À 1. In this regime the width of the peaks of
IN −n,n (λ) can be approximated as
√ √
24 1 24
WN −n,n =p [1+O( )] ≈ 1/2 √
τ 4 n(N − n)N N τ
where ttot = tF + tB = N τ is the total time. Complete
destructive interference occurs for WN −n,n ¿ λSD which
implies that (tF tB ttot )1/3 À τ −1/3 (λSD )−2/3 . For tF ∼
tB this implies
tB or tB ¿ tF for total times ttot = N τ which are at least
measurable. For example, consider total times at the
current resolution limit in time metrology [14] in which
case ttot ≈ 10−17 s and so N ≈ 1027 . Clearly there is no
destructive interference for the cases n = 0 or n = N
(because there is only one path in each case). Also for
n ≈ 1 and n ≈ N − 1 the estimated width of the peaks
of IN −n,n (λ) are given by
0.0004 0.0006
t2 l
√
24 1
WN −n,n = [1 + O( )] . (15)
τ 2N N
In these cases WN −n,n ≈ 1061 s−2 À λSD which implies
negligible destructive interference. This means that the
sum in Eq. (6) includes a number of paths, however, the
paths are characterized by one of either tF or tB being
negligible compared to the total time of 10−17 s.
Finally consider the value of ttot for which there is
complete destructive interference for the sets of paths
represented by hφ|SN −n,n |ψ0 i for all values of n except
for n = 0 and n = N . In other words there is com-
plete destructive interference for n = 1, 2, . . . (N − 1) and
so Eq. (6) is simply |Ψ(N τ )i = SN,0 |ψ0 i + S0,N |ψ0 i. We
only need to find the value of ttot for the case n = 1 as this
also ensures the destructive interference in the remaining
cases. Thus we want to know the value of ttot that satis-
fies WN −1,1 ¿ λSD . The width
√ WN −1,1 is approximated
by Eq. (15) and so we have 24/(τ 2 N ) ¿ λSD which im-
plies ttot À f −1/2 10−13 s. In this regime, the two paths
that survive the destructive interference represent exclu-
sively forwards and exclusively backwards evolution. Fig-
ure 4 shows the effect of the destructive interference on
the binary tree representation of the evolution.
In summary, for total times t = P N τ > 10−17 s
P (6) can be replaced with |Ψ(t)i = [ n≈0 SN −n,n +
Eq.
n≈N SN −n,n ]|ψ0 i and the integral in Eq. (10) is effec-
tively the identity operator. We note that for n ≈ 0,
Eq. (10) then implies that SN −n,n = UB (N τ ) + O(τ ).
Similarly, it can be shown using the symmetry property
tF tB ttot Eq. (14) that SN −n,n = UF (N τ ) + O(τ ) for n ≈ N . On
ignoring the terms of order τ we arrive at a key result of
this paper, the bievolution equation of motion
|Ψ(t)i = [UB (t) + UF (t)] |ψ0 i . (16)

1
ttot À ≈ f −1/3 10−23 s . The term “bievolution” here refers to the dual evolu-
τ 1/3 (λ SD )
2/3
tion generated by the two different Hamiltonians. The
The destructive interference means that the correspond- approximations made in deriving this equation become
ing set of paths do not contribute to the sum on the right exact in the limit t À f −1/2 10−13 s.
 6

(b) |y0 in the exponents of the definitions of UF and UB in

Eqs. (1) and (2) imply that they represent time evolu-
tion in opposing directions of time. We conclude that
the expressions UF (t)|ψi and UB (t)|ψi represent the sys-
tem evolving in opposite directions of time and in each
N
case the evolution leaves corroborating evidence of the
UBN (a) |y0 UF associated Hamiltonian in the state of the system.
UB UF We can compare this with our own experience of the
time evolution of the universe. All experiments to de-
UB UF UB UF
termine the nature of the Hamiltonian associated with
UB UF UB UF UB UF meson decay yield corroborating evidence of just one of
the Hamiltonians HF or HB . Our experience, therefore,
UB UF UB UF UB UF UB UF is consistent with just one of the terms on the right side
of the bievolution equation Eq. (16), and this underpins
the observed phenomenological unidirectional evolution
FIG. 4: The “pruning” of the binary tree in Fig. 1 by de- in the corresponding direction of time.
structive interference for the case HF 6= HB and N À 1. The To expand on this point a little, the bievolution equa-
relative contribution to the final state of each time evolution
tion Eq. (16) describes our observation of quantum sys-
step is depicted by the greyness of the corresponding arrow
with black and white representing maximal and minimal con-
tems as evolving according to either
tributions, respectively: (a) shows an expanded view of the
detail near the root node and (b) shows the whole tree on a
|Ψ(t)i = UF (t) |ψ0 i (17)
much coarser scale.
or

IV. UNIDIRECTIONALITY OF TIME AND |Ψ(t)i = UB (t) |ψ0 i (18)

STANDARD QUANTUM THEORY
Consider a situation where the Hamiltonians HF and
HB leave distinguishable evidence in the state of the sys-
tem. For example, imagine that the system state at (to-
tal) time t1 is
|Ψ(t1 )i = |φi + |θi
where |φi = UB (t1 )|ψ0 i and |θi = UF (t1 )|ψ0 i. Let
Uµ (t1 + a) = Uµ (a)Uµ (t1 ) and Uµ (a)|xi = |xµ i where
µ is either F or B, x is either φ or θ, and a is a suitable
fixed time interval. Then, according to Eq. (16)
|Ψ(t1 + a)i = UB (t1 + a)|ψ0 i + UF (t1 + a)|ψ0 i
= UB (a)UB (t1 )|ψ0 i + UF (a)UF (t1 )|ψ0 i
= UB (a)|φi + UF (a)|θi
= |φB i + |θF i.
The Hamiltonians leave distinguishable evidence in the
state of the system if they generate orthogonal states
in the sense that hθB |θF i = hφB |φF i = 0. We assume
this to be the case. Next, let this process occur twice
such that Uν (a)Uµ (a)|xi = |xµ,ν i and Uµ (t1 + 2a) =
Uµ (a)Uµ (t1 + a) where µ and ν are each either F or B.
We find, again by Eq. (16), that
|Ψ(t1 + 2a)i = UB (t1 + 2a)|ψ0 i + UF (t1 + 2a)|ψ0 i
= UB (a)UB (t1 + a)|ψ0 i + UF (a)UF (t1 + a)|ψ0 i
= UB (a)|φB i + UF (a)|θF i
= |φB,B i + |θF,F i
which shows that corroborating evidence is left in each
term of the superposition. Recall that the opposite signs
depending on whether we observe the version HF or HB ,
respectively, of the corresponding Hamiltonian in meson
decay experiments. Each of these equations is the solu-
tion of the conventional Schrödinger equation for the cor-
responding version of the Hamiltonian and its associated
direction of time. This demonstrates how the unidirec-
tional nature of time emerges phenomenologically from
our physical observations. It also demonstrates the con-
sistency between the bievolution equation and the conven-
tional quantum theory inasmuch as those observations al-
low.
V. EIGENSTATES OF i[HF , HB ] WITH ZERO
EIGENVALUES
We now temporarily relax the nonzero eigenvalue con-
dition in Eq. (13) and set the seed state to be an eigen-
state of the commutator i[HF , HB ] with eigenvalue zero,
i.e. we set
Π(0)|ψ0 i = |ψ0 i .
As previously mentioned, the integral in Eq. (10) in this
case can effectively be replaced with Im,n (0)ρ(0)Π(0) =
(n+m
n )Π(0) and so Eq. (6) becomes
XN
¡ ¢
|Ψ(N τ )i = UB [(N − n)τ ]UF (nτ ) N
n |ψ0 i
n=0
which represents all 2N possible paths through time. We
now reapply the analysis of the previous section for the
processes UF (a) and UB (a) with t1 = N τ and a = kτ .

There is no destructive interference in this case, and so
the state of the system at the total time t1 + 2a is, ac-
cording to Eq. (5),
|Ψ(t1 + 2a)i = [UB (τ ) + UF (τ )]2k
"N #
X ¡N ¢
× UB [(N − n)τ ]UF (nτ ) n |ψ0 i
n=0
which represents 2N +2k paths through time. These
paths include the operator combinations UF (a)UF (a),
UF (a)UB (a), UB (a)UF (a), UB (a)UB (a) (among many
others) and so any distinguishable evidence of the Hamil-
tonians left in the state of the system is not necessarily
corroborated over repetitions of the process. This is not
in accord with present experimental evidence. Moreover,
consider the seed state of the most general form
k lation processes are responsible for the phenomenological
|ψ0 i = |ψ0⊥ i + |ψ0 i
k k
where Π(0)|ψ0⊥ i = 0 and Π(0)|ψ0 i = |ψ0 i. For this case
the bievolution equation (16) is replaced with
|Ψ(t)i = [UB (t) + UF (t)] |ψ0⊥ i
N
X ¡N ¢
+ UB [(N − n)τ ]UF (nτ ) n |ψ0 i
n=0
where t = N τ . However, as we have already seen, ex-
perimental evidence suggests that the universe is not fol-
lowing the corresponding set of paths represented by the
last term on the right side. In order for our system to
provide a physical model of the observable universe, it
is therefore sufficient to only consider seed states that
satisfy the nonzero eigenvalue condition in Eq. (13). We
assume this to be the case in the remainder of this work.
VI. SCHRÖDINGER’S EQUATION FOR
BIEVOLUTION
We can construct the differential form of the bievo-
lution equation of motion as follows. Increasing t in
Eq. (16) by a relatively small time interval δt yields
|Ψ(t + δt)i = [UB (δt)UB (t) + UF (δt)UF (t)] |ψ0 i and so
the rate of change of the state is given by
δ|Ψ(t)i
= [iHB UB (t) − iHF UF (t)] |ψ0 i + O(δt)
δt
where δ|Ψ(t)i = |Ψ(t + δt)i − |ΨBi (t)i. Taking the limit
δt → τ and ignoring a term of order τ gives another key
result, the Schrödinger equation for bievolution
d|Ψ(t)i d|ψF (t)i d|ψB (t)i
= −
dt dt dt
where (d/dt)|ψµ (t)i = −iHµ |ψµ (t)i and |ψµ (t)i =
Uµ (t)|ψ0 i for µ = F or B.
7
VII. DISCUSSION
To gauge the full impact of T violation processes, it is
instructive to compare the above analysis with a system
which obeys T invariance. In this case HF = HB = H
and it can be shown that
¡ ¢
SN −n,n = exp[i(N − 2n)τ H] Nn
which is never zero and so all possible paths are included
in Eq. (6). This means that the evolution of the system
follows a random walk in time. The most likely paths are
for n ≈ 21 N and so the most likely evolved state is the
original seed state |ψ0 i. Moreover, the direction of time
is ambiguous as there is no physical evidence of any kind
to single out one direction over the other. The fact that
this ambiguity is removed for a system with T violation
processes leads directly to the proposition that T vio-
unidirectionality of time that we observe in the universe.
Just as important is the implicit assertion here that in
the absence of a unidirectional nature, time is randomly
directed at any given instant. We allowed for the possi-
bility of a random direction when we added amplitudes
in Eq. (4). In hindsight, the analysis of the preceding sec-
k tions can be seen to be a study of the dichotomy between
randomly directed and unidirectional time evolution. In
this context the important question about the unidirec-
tional nature of time is not how one direction of time
is chosen over the other, but rather how each direction
is maintained consistently from one instant to the next.
This is essentially the question that is being addressed in
this article, with T violation providing a possible mech-
anism as the answer.
We can also elaborate on what the time reversal opera-
tion means in the presence of T violating processes. The
bievolution equation Eq. (16) shows that as the value of
t increases, the states UF (t)|ψ0 i and UB (t)|ψ0 i in each
branch of the superposition trace out respective trajecto-
ries in the state space. Decreasing values of t then simply
“back tracks” the states along their respective trajec-
tories in a consistent manner, essentially tracking back
over what might be called their “histories”. In contrast,
the time reversal operation essentially interchanges UF (t)
and UB (t) in Eq. (16). This implies that the time rever-
sal operation refers to switching between the branches of
the bievolution equation as opposed to tracking back over
histories.
The physical implications of T violation described here
depend the magnitudes of the eigenvalues λ. Our esti-
mate of these eigenvalues depends on the number of par-
ticles involved in T violations processes and this has been
fixed only at f 1080 where f < 1 is a parameter of unde-
termined value. However, our analysis does not depend
critically on the value of f and, indeed, a large range of
its values support the key results. Also, our analysis has
focused on relatively large time scales in order to draw
out the main consequences of T violation for the unidi-
rectionality of time. In doing so we have ignored terms
 8

of order τ , the Planck time. These terms imply that
the directionality of time has finer details for relatively
small time intervals. The elaboration of these details is,
however, beyond the scope of this article.
There are also potentially important experimental im-
plications of the analysis presented here. A sufficiently
small value of f would produce incomplete destructive in-
terference and a deviation from the dynamics described
by the bievolution equation Eq. (16). This would be man-
ifest in a lower bound to the accuracy of time metrol-
ogy. A more extreme case occurs in the very early uni-
verse during the inflation and radiation-dominated peri-
ods [15, 16]. Baryogenesis would not have yet occurred
and CP and T violation would have been relatively rare
events. This implies that the value of f would have been
negligible during these periods and so the direction of
time evolution would have been uncertain. This has po-
tentially important ramifications for cosmological models
and their testing against observational data of the cosmic
microwave background radiation [15].
In conclusion, our general understanding of T violat-
ing processes has previously been hindered by the lack
of a formalism for accommodating their time asymmetry
in a single dynamical law. We have removed this ob-
stacle by deriving, from a first principles analysis based
on Feynman’s path integrals [11], a general method for
incorporating both versions of the Hamiltonian for a T
violating process, one for forwards and the other for back-
wards evolution, in a single dynamical equation of motion
as given by Eq. (16). Moreover the physical significance
of the relatively weak and rare T violation processes ex-
hibited by mesons was previously a puzzle, but we have
shown that these processes can affect the time evolution
of the universe on a grand scale. Indeed, we have shown
that T violation provides, at least in principle, a physical
mechanism for the phenomenological unidirectionality of
time. Finally, we have shown that with no T violating
processes the evolution of a universe-like system would
effectively be a random walk through time, and to this
extent we have glimpsed the quantum nature of time it-
self.
Acknowledgments
The author thanks Prof. D. T. Pegg for sharing his
wealth of knowledge on the nature of time and for the
generous advice he offered in the development of this ar-
ticle, Prof. S. M. Barnett for his encouragement and help
with kaon dynamics and Prof. H. M. Wiseman for helpful
discussions. This work was supported by the Australian
Research Council and the Queensland State Government.

APPENDIX A: REORDERING FACTORS OF Sm,n

The operator Sm,n is defined recursively in Eq. (7) as

m
X
Sm,n = Sm−k,n−1 UF (τ )UB (kτ ) (A1)
k=0

with Sm,0 = UB (mτ ) and so

m
X m
X
Sm,1 = Sm−k,0 UF (τ )UB (kτ ) = UB [(m − k)τ ]UF (τ )UB (kτ ) . (A2)
k=0 k=0

The Zassenhaus formula [17]

1 2
[A,B] O(²3 ) 1 2
[B,A] O(²3 )
eA² eB² e− 2 ² e = eB² eA² e− 2 ² e

(which is related to the Baker-Campbell-Hausdorff formula) where A and B are arbitrary operators and ² is a small
parameter, can be written as
2
[A,B] O(²3 )
eA² eB² = eB² eA² e² e

which shows that

2
[HF ,HB ] O(τ 3 )
e−ijτ HF eikτ HB = eikτ HB e−ijτ HF ejkτ e . (A3)

In terms of the operators UF (τ ) = e−iτ HF and UB (τ ) = eiτ HB this result reads

2
[HF ,HB ] O(τ 3 )
UF (jτ )UB (kτ ) = UB (kτ )UF (jτ )ejkτ e (A4)
 9

or equivalently
2
[HF ,HB ] O(τ 3 )
UB (kτ )UF (jτ ) = UF (jτ )UB (kτ )e−jkτ e . (A5)

Hence
m
X
Sm,n = Sm−k,n−1 UF (τ )UB (kτ )
k=0
Xm
2
[HF ,HB ] O(τ 3 )
= Sm−k,n−1 UB (kτ )UF (τ )ekτ e
k=0
m m−k
X X 2
[HF ,HB ] O(τ 3 )
= Sm−k−j,n−2 UF (τ )UB (jτ )UB (kτ )UF (τ )ekτ e
k=0 j=0
m m−k
X X 2
[HF ,HB ] O(τ 3 )
= Sm−k−j,n−2 UF (τ )UB [(j + k)τ ]UF (τ )ekτ e
k=0 j=0
m m−k
X X 2
[HF ,HB ] O(τ 3 )
= Sm−k−j,n−2 UF (2τ )UB [(j + k)τ ]e−jτ e
k=0 j=0
m m−k
X X 2
[HF ,HB ] O(τ 3 )
= Sm−k−j,n−2 UB [(j + k)τ ]UF (2τ )e(j+2k)τ e . (A6)
k=0 j=0

Setting ` = j + k and rearranging the order the terms are summed to eliminate j then yields
m X̀
X 2
[HF ,HB ] O(τ 3 )
Sm,n = Sm−`,n−2 UB (`τ )UF (2τ )e(`+k)τ e
`=0 k=0
m
X X̀
2 2
[HF ,HB ] O(τ 3 )
= Sm−`,n−2 UB (`τ )UF (2τ )e`τ [HF ,HB ]
ekτ e
`=0 k=0
m m−`
X X X̀
2 2
[HF ,HB ] O(τ 3 )
= Sm−`−r,n−3 UF (τ )UB [(` + r)τ ]UF (2τ )e`τ [HF ,HB ]
ekτ e
`=0 r=0 k=0

m m−`
X X X̀
2 2
[HF ,HB ] O(τ 3 )
Sm,n = Sm−`−r,n−3 UB [(` + r)τ ]UF (3τ )e(r+2`)τ [HF ,HB ]
ekτ e . (A7)
`=0 r=0 k=0

Rearranging as before but with j = ` + r and eliminating r yields

X j
m X X̀
2 2
[HF ,HB ] O(τ 3 )
Sm,n = Sm−j,n−3 UB (jτ )UF (3τ )e(j+`)τ [HF ,HB ]
ekτ e
j=0 `=0 k=0
m
X j X̀
X
2 2
[HF ,HB ] O(τ 3 )
= Sm−j,n−3 UB (jτ )UF (3τ )ejτ [HF ,HB ]
e(`+k)τ e . (A8)
j=0 `=0 k=0

Continuing in this way gives, after w uses of the recursive relation Eq. (A1),
m
X v
X s X̀
X
vτ 2 [HF ,HB ] 2
[HF ,HB ] O(τ 3 )
Sm,n = Sm−v,n−w UB (vτ )UF (wτ )e ··· e(r+···+`+k)τ e (A9)
v=0 r=0 `=0 k=0

in which there are w summations. Letting w = n then yields

m
X v
X s X̀
X
2 2
[HF ,HB ] O(τ 3 )
Sm,n = Sm−v,0 UB (vτ )UF (nτ )evτ [HF ,HB ]
··· e(r+···+`+k)τ e
v=0 r=0 `=0 k=0
 10

m
X v
X s X̀
X
2 2
[HF ,HB ] O(τ 3 )
= UB (mτ )UF (nτ )evτ [HF ,HB ]
··· e(r+···+`+k)τ e
v=0 r=0 `=0 k=0
m X
X v s X̀
X 2
[HF ,HB ] O(τ 3 )
= UB (mτ )UF (nτ ) ··· e(v+r+···+`+k)τ e . (A10)
v=0 r=0 `=0 k=0

Hence we arrive at Eq. (8)

m
X s X̀
X 2
[HF ,HB ] O(τ 3 )
Sm,n = UB (mτ )UF (nτ ) ··· e(v+···+`+k)τ e (A11)
v=0 `=0 k=0
in which there are n summations.

APPENDIX B: SIMPLIFYING Im,n (λ)

Consider the following manipulations of nested summations of the kind

X j X̀
w X s X
X n
··· ei(j+`+m+···+n+k)θ . (B1)
j=0 `=0 m=0 n=0 k=0

We call the summations “nested” because the upper limit n of the last summation over k is the index of the second-last
summation and so on. The nested property means that summations cannot be evaluated independently of each other.
However it is possible to reduce the number of nested summations by performing certain operations which we now
describe.
First, by changing the order in which the indices j and ` are summed, we find
X j X̀
w X s X
X n w X
X w X̀ s X
X n
··· ei(j+`+m+···+n+k)θ = ··· ei(j+`+m+···+n+k)θ , (B2)
j=0 `=0 m=0 n=0 k=0 `=0 j=` m=0 n=0 k=0

next, by cyclically interchanging the indices in the order j → k → n → s → · · · → m → ` → j on the right-hand side,
we get
X j X̀
w X s X
X n w X
X j X̀
w X s
X
··· ei(j+`+m+···+n+k)θ = ··· ei(k+j+`+m+···+n)θ , (B3)
j=0 `=0 m=0 n=0 k=0 j=0 k=j `=0 m=0 n=0

and finally, bringing the sum over k to the extreme right on the right-hand side gives
X j X̀
w X s X
X n X j X̀
w X s X
X w
··· ei(j+`+m+···+n+k)θ = ··· ei(j+`+m+···+n+k)θ . (B4)
j=0 `=0 m=0 n=0 k=0 j=0 `=0 m=0 n=0 k=j

We can abbreviate this general summation property as

w
X t X
X s X
n w
X t X
X s X
w
··· ei(j+···+s+n+k)θ = ··· ei(j+···+s+n+k)θ . (B5)
j=0 s=0 n=0 k=0 j=0 s=0 n=0 k=j

Consider the product

³ ´X
s X
n s X
X n s X
X n
iθ i(n+k)θ iθ i(n+k)θ
e +1 e =e e + ei(n+k)θ . (B6)
n=0 k=0 n=0 k=0 n=0 k=0

Using Eq. (B5) for the two summations in the first term on the right-hand side gives
³ ´Xs Xn X s X s Xs Xn
eiθ + 1 ei(n+k)θ = eiθ ei(n+k)θ + ei(n+k)θ
n=0 k=0 n=0 k=n n=0 k=0
s X
X s s X
X n
= ei(n+k+1)θ + ei(n+k)θ
n=0 k=n n=0 k=0
s
X s+1
X s X
X n s X
X s+1
= ei(n+k)θ + ei(n+k)θ = ei(n+k)θ . (B7)
n=0 k=n+1 n=0 k=0 n=0 k=0
 11

The two nested summations on the left-hand side have been reduced to two un-nested summations on the right-hand
side. Similarly, using Eq. (B5) again as well as Eq. (B7) we find
³ ´X
t X
s X
n t X
X s X
n ³ ´X
t X
s X
n
i2θ iθ i(s+n+k)θ i2θ i(s+n+k)θ iθ
e +e +1 e = e e + e +1 ei(s+n+k)θ
s=0 n=0 k=0 s=0 n=0 k=0 s=0 n=0 k=0
t X
X s X
t t X
X s X
s+1
= ei(s+n+k+2)θ + ei(s+n+k)θ
s=0 n=0 k=s s=0 n=0 k=0
t X
X s t+2
X t X
X s X
s+1
= ei(s+n+k)θ + ei(s+n+k)θ
s=0 n=0 k=s+2 s=0 n=0 k=0
t X
X s X
t+2
= ei(s+n+k)θ . (B8)
s=0 n=0 k=0

Here the three nested summations on the left-hand side have been reduced to two nested summations and one un-
nested summation on the right-hand side. This implies that, for example, p + 1 nested summations can be reduced
to p nested and one un-nested summations as follows:
³X
p ´X
w t X
X s X
n ³X
w t X
X s ´ w+p
X
eirθ ··· ei(j+···+s+n+k)θ = ··· ei(j+···+s+n)θ eikθ . (B9)
r=0 j=0 s=0 n=0 k=0 j=0 s=0 n=0 k=0
| {z } | {z }
p+1 nested sums p nested sums
Moreover, by repeating this process once more, the p nested summations of the right-hand side can be further reduced
to p − 1 nested and one un-nested summations. Continuing in this way eventually leads to the following result
p ³X
Y q ´X
w t X
X s X
n Y³ X
p−1 q ´³ X
w t X
X s ´w+p
X
eirθ ··· ei(j+···+s+n+k)θ = eirθ ··· ei(j+···+s+n)θ eikθ
q=0 r=0 j=0 s=0 n=0 k=0 q=0 r=0 j=0 s=0 n=0 k=0
| {z } | {z }
p+1 nested sums p nested sums
³X
w ´ ³ w+p−1
X ´³ w+p
X ´
= eijθ · · · einθ eikθ
j=0 n=0 k=0
| {z }
p+1 un-nested sums
p ³ w+q
Y X ´
= eijθ . (B10)
q=0 j=0

Thus
Qp ´
³P
w+q
w
X t X
X s X
n
q=0 eijθ j=0
··· ei(j+···+s+n+k)θ = Q ³P ´ . (B11)
p q irθ
j=0 s=0 n=0 k=0 q=0 r=0 e
| {z }
p+1 nested sums
Evaluating the two geometric series on the right-hand side then gives
Qp ³ i(w+q+1)θ
´ Qp+1 ³ i(w+q)θ
´
Xw Xt X s X n
q=0 1 − e q=1 1 − e
··· ei(j+···+s+n+k)θ = Q ³ ´ = Q ³ ´ . (B12)
p i(q+1)θ p+1 iqθ
j=0 s=0 n=0 k=0 q=0 1 − e q=1 1 − e
| {z }
p+1 nested sums
Adjusting the number of nested summations of the left-hand side then gives the useful result that
Qp ³ ´
Xw Xt X s X n
q=1 1 − ei(w+q)θ
··· ei(j+···+s+n+k)θ = Qp ³ ´ . (B13)
iqθ
j=0 s=0 n=0 k=0 q=1 1 − e
| {z }
p nested sums
 12

We can now use this result to simplify the n nested summations in Eq. (11) as follows. First we note that
Qn ³ −i(m+q)τ 2 λ
´
Xm Xs X̀
2 q=1 1 − e
Im,n (λ) = ··· e−i(v+···+`+k)τ λ = Qn ³ ´ (B14)
1 − e −iqτ 2 λ
v=0 `=0 k=0 q=1

and then by redefining the index q in the numerator we arrive at

Qn−1 ³ −i(n+m−q)τ 2 λ
´
q=0 1 − e
Im,n (λ) = Qn ³ −iqτ 2 λ
´ (B15)
q=1 1 − e

which is Eq. (12) of the paper.

APPENDIX C: ESTIMATING EIGENVALUES

(1) (1)
We estimate the eigenvalues of the commutator [HF , HB ] on the state space of a kaon field mode that represents
a single particle. However, to avoid unnecessary clutter, we will omit the superscript “(1)” on the single particle
Hamiltonians in the remainder of this section.
It is convenient to invoke the CPT theorem and use the CP operator instead of the T operator as follows

HB = T HF T −1 = (CP )−1 HF (CP ) . (C1)

Consider the matrix representation in the kaon state space where the vectors (1, 0) and (0, 1) represent the kaon and
anti-kaon states |K 0 i and |K 0 i, respectively. In this basis the general form of the CP operator is
· ¸
0 eiη
CP = (C2)
eiξ 0

where η and ξ are possible phase angles. By redefining the complex phase of the kaon state where |K 0 i is multiplied
by ei(η−ξ)/2 , the CP operator becomes
· ¸ · ¸
0 eiθ iθ 0 1
CP = =e (C3)
eiθ 0 1 0

where θ = 12 (η + ξ). (This is a more useful transformation for us here compared to the usual one, which multiplies
|K 0 i by e−iξ .) Let Mij be the matrix elements of HF in the kaon state basis where M11 = hK 0 |HF |K 0 i etc. To find
the corresponding matrix elements of HF we use Eq. (C1), that is

HB = (CP )−1 HF (CP )

· ¸· ¸ · ¸ · ¸
0 1 M11 M12 0 1 M22 M21
= e−iθ eiθ = .
1 0 M21 M22 1 0 M12 M11

Hence we immediately find

· ¸
M11 M22 + (M12 )2 M11 (M21 + M12 )
HF HB =
M22 (M21 + M12 ) M11 M22 + (M21 )2
· ¸
M11 M22 + (M21 )2 M22 (M21 + M12 )
HB HF = ,
M11 (M21 + M12 ) M11 M22 + (M12 )2

and so the commutator is

· ¸
(M12 )2 − (M21 )2 (M11 − M22 )(M21 + M12 )
[HF , HB ] = HF HB − HB HF = (C4)
−(M11 − M22 )(M21 + M12 ) −[(M12 )2 − (M21 )2 ]

which is in the form

· ¸
A B
[HF , HB ] = . (C5)
−B −A
 13
√
The eigenvalues of the last matrix are ± A2 − B 2 . Note that A is imaginary and B is real, and so the eigenvalues
are imaginary. We now determine empirical values for the variables Mi,j in Eq. (C4).
The phenomenological model given in Lee and Wolfenstein [13] describes the evolution of the amplitudes a(t) and
b(t) for the single-particle kaon field to be in the states |K 0 i and |K 0 i, respectively, by the differential equation

dψ(t) 1
= − (Γ + iM )ψ(t) (C6)
dt h̄
where ψ = (ab ). The matrix on the right-hand side is defined by

Γ + iM = D + i(E1 σ1 + E2 σ2 + E3 σ3 ) (C7)

where the matrices Γ and M are Hermitian and

· ¸ · ¸ · ¸
0 1 0 −i 1 0
σ1 = , σ2 = , σ3 = . (C8)
1 0 i 0 0 −1

The analysis below shows that the matrix M in Eq. (C6) represents a CP violating Hamiltonian. For our purposes,
it provides a sufficient estimate of the T violating Hamiltonian for a neutral kaon. Accordingly, we use the empirical
values of the elements of M as estimates of the corresponding matrix elements Mi,j of the Hamiltonian HF that
appear in Eq. (C4). We should also mention that we could use M as an estimate of the matrix elements of HB
instead, however the only difference this would make is that the matrix elements of the commutator [HF , HB ] would
be multiplied by −1, and this has no physical consequences because the two eigenvalues are the negatives of each
other.
If the CPT theorem holds, which we assume to be the case, then E1 = E cos φ, E2 = E sin φ and E3 = 0. Here φ is
a complex parameter [13] which is related to the CP violation parameter ² by ² = (1 − eiφ )/(1 + eiφ ), or equivalently,
1−²
eiφ = . (C9)
1+²
The values of D and E comprise the decay rates γi and masses mi of the short and long-lived components, |K10 i and
|K20 i respectively, of |K 0 i as follows
1
D = 4 (γ1 + γ2 ) + 12 i(m1 + m2 ) (C10)
1 1
iE = 4 (γ1 − γ2 ) + 2 i(m1 − m2 ) . (C11)

Here γi and mi are the energies γi = h̄γi0 and mi = m0i c2 , respectively, where γi0 and m0i are the actual decay rates
and masses, respectively. Also note [18] that |²| ≈ 2.3 × 10−3 and arg(²) ≈ 45◦ and so from Eq. (C9) we find
eiφ = 1 − 2² + O(²2 ). We can therefore make the following approximations
√
eiφ = 1 − 2|²|(1 + i) + O(²2 ) ,
∗ √
eiφ = 1/(eiφ )∗ = 1 + 2|²|(1 − i) + O(²2 ) ,
√
e−iφ = 1/(eiφ ) = 1 + 2|²|(1 + i) + O(²2 ) ,
∗ √
e−iφ = (eiφ )∗ = 1 − 2|²|(1 − i) + O(²2 ) .

This allows us to write down the following useful results

1 iφ ∗ √
2 (e + eiφ ) = 1 − i 2|²| + O(²2 ) ,
−iφ ∗ √
1
2 (e + e−iφ ) = 1 + i 2|²| + O(²2 ) ,
1 iφ iφ∗
√
2 (e − e ) = − 2|²| + O(²2 ) ,
1 −iφ ∗ √
2 (e − e−iφ ) = 2|²| + O(²2 ) .

From Eqs. (C7), (C8), (C10) and (C11) we find

Γ + iM = D + i(E cos φ σ1 + E sin φ σ2 )

 h i 
1 1 1 1 −iφ
4 (γ1 + γ 2 ) + 2 i(m 1 + m 2 ) 4 (γ 1 − γ2 ) + 2 i(m 1 − m2 ) e
= h i  . (C12)
1 1 iφ 1 1
(γ
4 1 − γ 2 ) + 2 i(m 1 − m 2 ) e (γ
4 1 + γ 2 ) + 2 i(m1 + m2 )
 14

The Hermitian transpose of this equation is

 h i 
1 1 1 1 −iφ∗
(γ
4 1 + γ 2 ) − 2 i(m1 + m 2 ) (γ 1 − γ2 ) − i(m 1 − m2 ) e
Γ − iM =  h i ∗ 4 2  . (C13)
1 1 iφ 1 1
4 (γ1 − γ2 ) − 2 i(m1 − m2 ) e 4 (γ1 + γ2 ) − 2 i(m1 + m2 )

By subtracting Eq. (C13) from Eq. (C12) and dividing the result by 2i we get
· √ √ ¸
√
1
(m1 + m2 ) √ −i 41 (γ1 − γ2 ) 2|²| + 21 (m1 − m2 )(1 + i 2|²|)
M = 2
i 14 (γ1 − γ2 ) 2|²| + 12 (m1 − m2 )(1 − i 2|²|) 1
2 (m1 + m2 )
+O(²2 ) . (C14)

This shows that M11 = M22 and so B = 0 in Eqs. (C4) and (C5). The eigenvalues of the commutator [HF , HB ] are
therefore simply ±A where
2 2
A = M12 − M21
h √ √ i2
= − i 14 (γ1 − γ2 ) 2|²| + 12 (m1 − m2 )(1 + i 2|²|)
h √ √ i2
− i 14 (γ1 − γ2 ) 2|²| + 21 (m1 − m2 )(1 − i 2|²|) + O(²2 )
√ h √ √ i
= −i 14 (γ1 − γ2 ) 2|²| 21 (m1 − m2 ) 2(1 + i 2|²|) + 2(1 − i 2|²|)
h √ √ i
+ 14 (m1 − m2 )2 (1 + i 2|²|)2 − (1 − i 2|²|)2 + O(²2 )
h √ √ i
= i − (γ1 − γ2 ) 2|²| 12 (m1 − m2 ) + (m1 − m2 )2 2|²| + O(²2 )
√ h i
= i 2|²| (γ1 − γ2 ) 12 (m2 − m1 ) + (m1 − m2 )2 + O(²2 )
√ h i
= i 2|²| 12 ∆γ∆m + (∆m)2 + O(²2 ) (C15)

where ∆m = (m2 − m1 ) and ∆γ = (γ1 − γ2 ).

Finally, the empirical values [18]

∆m ≈ 0.56 × 1010 h̄s−1 ,

∆γ ≈ 1.1 × 1010 h̄s−1 ,

show that the eigenvalue A in Eq. (C15) can be approximated as

A ≈ i1017 h̄2 s−2 . (C16)

This is the value used in the paper for λ(1) where λ(1) = ±iA. Note that in the main text, we use units in which h̄ = 1.
This is equivalent to replacing the Hamiltonians HF and HB in this section with HF /h̄ and HB /h̄, respectively. In that
case the commutator in Eq. (C5) has the form [HF , HB ]/h̄2 which has a dimension of (time)2 . Hence λ(1) ≈ ±1017 s−2 .

[1] H. Price, Time’s Arrow and Archimedes’ Point (Oxford
Uni. Press, New York, 1996).
[2] J. H. Christenson, J. W. Cronin, V. L. Fitch, and R.
Turlay, Phys. Rev. Lett. 13, 138-140 (1964).
[3] A. D. Sakharov, JETP Lett. 5, 24-26 (1967).
[4] P. Pavlopoulos, Nucl. Phys. B 99, 16-23 (2001).
[5] A. Lusiani, J. Phys.: Conf. Ser. 171, 012037 (2009).
[6] A. Angelopoulos et al. (CPLEAR Collaboration), Phys.
Lett. B 444, 43-51 (1998).
[7] N. Cabibbo, Phys. Rev. Lett. 10, 531-533 (1963).
[8] M. Kobayashi and T. Maskawa, Prog. Theor. Phys. 49,
652-657 (1973).
[9] E. P. Wigner, Group theory and its application to the
quantum mechanics of atomic spectra (Academic Press,
New York, 1959).
[10] The Belle Collaboration, Nature 452, 332-335 (2008).
[11] R. P. Feynman, Rev. Mod. Phys. 20, 367-387 (1948).
[12] S.M. Barnett and J.A. Vaccaro, The Quantum Phase Op-
erator: A Review (Taylor and Francis, London, 2007).
[13] T.D. Lee and L. Wolfenstein, Phys. Rev. 138, B1490-

B1496 (1965).
[14] T. Rosenband et al. Science 319, 1808-1812 (2008).
[15] C.L. Bennett, Nature 440, 1126-1131 (2006).
[16] L. Kofman, A. Linde and A.A. Starobinsky, Phys. Rev.
Lett. 73, 3195-3198 (1994).
15
[17] M. Suzuki, Commun. Math. Phys. 57, 193-200 (1977).
[18] W-M. Yao et al. J. Phys. G: Nucl. Part. Phys. 33, 666-
684 (2006).