• Embed Doc
  • Copy Link
  • Readcast
  • Collections
  • CommentGo Back
Download
 
  a  r   X   i  v  :   0   8   0   2 .   0   4   3   8  v   3   [  q  u  a  n   t  -  p   h   ]   2   5   A  u  g   2   0   0   9
A quantum solution to the arrow-of-time dilemma
Lorenzo Maccone
∗
QUIT, Dip. A. Volta, 27100 Pavia, and Institute for Scientific Interchange, 10133 Torino, Italy.
The arrow of time dilemma: the laws of physics are invariant for time inversion, whereas thefamiliar phenomena we see everyday are not (i.e. entropy increases). I show that, within a quantummechanical framework, all phenomena which leave a trail of information behind (and hence canbe studied by physics) are those where entropy necessarily increases or remains constant. Allphenomena where the entropy decreases must not leave any information of their having happened.This situation is completely indistinguishable from their not having happened at all. In the light of this observation, the second law of thermodynamics is reduced to a mere tautology: physics cannotstudy those processes where entropy has decreased, even if they were commonplace.
PACS numbers: 03.67.-a,03.65.Ud
Paradoxeshave alwaysbeen very fruitful in stimulatingadvances in physics. One which still lacks a satisfactoryexplanation is the Loschmidt paradox [1]. Namely, howcan we obtain irreversible phenomena from reversibletime-symmetric physical laws [2]? The irreversibility inPhysics is summarized by the second law of thermody-namics: entropy, which measures the degradation of theusable energy in a system, never decreases in isolatedsystems. Many approaches have been proposed to solvethis conundrum, but most ultimately resort to postulat-ing low entropy initial states (see e.g. [3,4]), which is clearly an
ad hoc
assumption [5]. Others suggest that thethermodynamic arrow of time is in some way connectedto the cosmological one [6], that physical laws must bemodified to embed irreversibility [7], that irreversibilityarises from decoherence [8], or from some time-symmetricmechanism embedded in quantum mechanics [9],
etc.
Re-cent reviews on this problem are given in Ref. [10].Here I propose a different approach, based on exist-ing laws of physics (quantum mechanics). I show thatentropy in a system can both increase and decrease (astime reversal dictates), but that all entropy-decreasingtransformations cannot leave
any 
trace of their havinghappened. Since no information on them exists, this isindistinguishable from the situation in which such trans-formations do not happen at all: “The past exists onlyinsofar as it is recorded in the present” [11]. Then thesecond law is forcefully valid: the only physical evolu-tions we see in our past, and which can then be studied,are those where entropy has not decreased.I start by briefly relating the thermodynamic entropywith the von Neumann entropy, and introducing the sec-ond law. I then present two thought experiments, whereentropy is deleted together with all records of the entropyincreasing processes: even though at some time the en-tropy of the system had definitely increased, afterwardit is decreased again, but none of the observers can beaware of it. I conclude with a general derivation throughthe analysis of the entropy transfers that take place inphysical transformations.
Entropy and the second law.
Thermodynamic en-tropy is a quantity that measures how the usable en-ergy in a physical process is degraded into heat. Itcan be introduced in many ways from different axiom-atizations of thermodynamics. The von Neumann en-tropy of a quantum system in the state
ρ
is definedas
S 
(
ρ
)
≡ −
Tr[
ρ
log
2
ρ
]. When applicable, these twoentropies coincide (except for an inconsequential mul-tiplicative factor). This derives from an argument in-troduced by Einstein [12] and extended by Peres [13] (e.g. both the canonical and the microcanonical ensem-ble can be derived from quantum mechanical consider-ations [14,15]). For our purposes, however, it is suffi- cient to observe that thermodynamic and von Neumannentropies can be inter-converted, employing Maxwell-demons [16,17] or Szilard-engines [18,19]: useful work can be extracted from a single thermal reservoir by in-creasing the von Neumann entropy of a memory space.There are many different formulations of the secondlaw, but we can summarize them by stating that, inany process in which an isolated system goes from onestate to another, its thermodynamic entropy does notdecrease [20]. There is a hidden assumption in this state-ment. Whenever an isolated system is obtained by join-ing two previously isolated systems, then the second lawis valid only if the two systems are initially uncorrelated,i.e. if their initial joint entropy is the sum of their individ-ual entropies. It is generally impossible to exclude thattwo systems might be correlated in some unknown wayand there is no operative method to determine whethera system is uncorrelated from all others (e.g. given a boxcontaining some gas it is impossible to exclude that thegas particles might be correlated with other systems).Thus, in thermodynamics all systems are considered un-correlated, unless it is known otherwise. Without thisassumption, it would be impossible to assign an entropyto any system unless the state of the whole universeis known: a normal observer is limited in the informa-tion she can acquire and on the control she can apply.This implies that thermodynamic entropy is a subjectivequantity [21], even though
for all practical situations
thisis completely irrelevant: the eventual correlations in allmacroscopic systems are practically impossible to con-trol and exploit. Even though they are ignored by thenormal observer, correlations between herself and othersystems do exist. Until they areeliminated, the other sys-
 
2tems cannot decrease their entropy. A physical processmay either reduce or increase these correlations. Whenthey are reduced, this may seem to entail a diminish-ing of the entropy, but the observer will not be aware of it as her memories are correlations and will have beenerased by necessity (each bit of memory is one bit of correlation and, until her memory has been erased, thecorrelations are not eliminated). Instead, when the phys-ical process increases these correlations, she will see it asan increase in entropy. The observer will then only beaware of entropy non-decreasing processes. [Not even asuper-observer that can keep track of all the correlationswould ever see any entropy decrease. In fact, since he candiscover and take advantage of all correlations betweenmicroscopic degrees of freedom, all processes are alwayszero-entropy processes from his super-observer point of view.]The above analysis is limited to systems that are some-how correlated with the observer. One might then expectthat she could witness entropy decreasing processes insystems that are completely factorized from her. That isindeed the case: statistical microscopic fluctuations canoccasionally decrease the entropy of a system (the secondlaw has only a statistical valence). However, an observeris macroscopic by definition, and all remotely interact-ing macroscopic systems become correlated very rapidly(e.g. Borel famously calculated that moving a gram of material on the star Sirius by one meter can influencethe trajectories of the particles in a gas on earth on atime-scale of 
µ
s [22]). This is the same mechanism at thebasis of quantum decoherence [8], and it entails that inpractice the above analysis applies to all situations: noentropy decrease in macroscopic systems is ever observed.In what follows I will make these ideas rigorous.Since the two above definitions of entropy are equiv-alent, the von Neumann entropy also obeys the secondlaw. In fact, isolated systems evolve with unitary evolu-tions, which leave the von Neumann entropy invariant.There may be an increase if the evolution is not exactlyknown or if it creates unknown correlations among sub-systems. In the first case, the coarse-grained evolution isof the form
ρ
′
=
n
p
n
U 
n
ρU 
n
†
, where
p
n
is a probabil-ity and
U 
n
are unitary operators. Then the final entropy
S 
(
ρ
′
) may be larger than the initial entropy
S 
(
ρ
):
S 
(
ρ
′
) =
S 
(
n
 p
n
U 
n
ρU 
n
†
)
n
 p
n
S 
(
U 
n
ρU 
n
†
) =
S 
(
ρ
)(1)(the inequality follows from the concavityof the entropy).In the second case, the entropy of subsystems can in-crease, as some unknown correlations between them maybuild up:
S 
(
ρ
′
1
) +
S 
(
ρ
′
2
)
S 
(
U 
(
ρ
1
⊗
ρ
2
)
U 
†
) =
S 
(
ρ
1
) +
S 
(
ρ
2
)
,
(2)where
ρ
′
i
and
ρ
i
are the final and initial states of thesubsystems,
U 
is the evolution coupling them, and thelast equality holds if they are initially uncorrelated (theinequality follows from the subadditivity of the entropy).
Thought experiments.
The quantum information the-ory mantra “Information is physical” [23] implies thatany record [24] of an occurred event can be decorrelatedfrom such event by an appropriate physical interaction.If 
all 
the records of an event are decorrelated from it,then by definition there is no way to know whether thisevent has ever happened. This situation is indistinguish-able from its not having happened. If this event has in-creased the entropy, the subsequent erasing of all recordscan (
will 
) produce an entropy decrease without violationof any physical law. We now analyze two such situations,an imperfect transmission of energy and a quantum mea-surement.Alice’s lab is perfectly isolated, so that to an outsideobserver (Bob), its quantum evolution is unitary. Ana-lyze the situation in which Bob sends Alice some energyin the form of light, a multimode electromagnetic fieldin a zero-entropy pure state. We suppose that, to se-cure the energy Bob is sending her, she uses many de-tectors which are not matched to his modes. Given asystem in almost any possible pure state, all its subsys-tems which are small enough are approximately in thecanonical state [14]. This implies that, if each of Alice’sdetectors is sensitive to only a small part of Bob’s modes,the detectors mostly see thermal radiation, and she feelsthem warming up. She will then be justified in assigninga nonzero thermodynamic entropy to her detectors, asshe sees them basically as thermal-equilibrium systems.One might object that she is mistaken, since the statesof the detectors are not uncorrelated. However, since sheignores the correlations, she cannot use such correlationsto extract energy from the detectors. Alice concludesthat most of the energy Bob sent her has been wastedas heat, raising the thermodynamic entropy of her lab.Suppose now that Bob has complete control of all thedegrees of freedom in her lab. He knows and can exploitthe correlations to recover all the energy he had initiallygiven Alice. Of course, although possible in principle,he needs a dauntingly complex transformation, which re-quires him to be able to control a huge number of herlab’s degrees of freedom (including the brain cells whereher memories are, and the notepads where she wrote thetemperatures!). To extract the energy, since it was ini-tially locked in a pure state of the field, he must returnit to a system in a zero-entropy pure state, i.e. factor-ized from all the other degrees of freedom of Alice’s lab.Then he must erase
all 
the correlations between them: atthe end of Bob’s recovery, Alice cannot remember feelingher detectors warm up, they are cool again, her notepadscontain no temperature information, and all the energyinitially in the electromagnetic field is again available,even though (from Alice’s point of view) most of it wasdefinitely locked into thermal energy at one time.The second though-experiment [25]is a prototypicalquantum measurement. Bob prepares a spin-1/2 parti-cle oriented along the
x
axis, e.g. in a spin
| →
stateand hands it to Alice. She sends it through a Stern-Gerlach apparatus oriented along the
z
axis [13]. The
 
3
BBAA
00000000001111111111
0000000000000000000011111111111111111111
S=0S=0
CNOT
S=1S=0b)a)
FIG. 1: a) Alice in her isolated lab performs a Stern-Gerlachmeasurement on a spin 1/2 particle initially oriented paral-lel to the
x
axis, i.e. in a state
| →
. Since the apparatus isoriented along the
z 
axis, this measurement creates one bitof entropy for Alice (not for Bob, who is isolated from herlab). b) Bob flips a switch that “cancels” Alice measurementby decorrelating from the spin all those degrees of freedom of her lab that have recorded the measurement outcome. Nowthe spin is returned to its initial state
| →
and Alice can-not have any memory of what her measurement result was.Her entropy has decreased, but she cannot remember it everhaving increased.
measurement consists in coupling the quantum systemwith some macroscopic degrees of freedom (a reservoir),not all of which are under the control of the exper-imenter [26], whence the irreversibility. Notice that,
| →
= (
| ↑
+
| ↓
)
/
√
2, where
| ↑
and
| ↓
are the eigen-states of a
z
measurement operator. Hence, this appara-tus will increase the entropy of the spin system by onebit [16]: Before the readout, the spin state will be in themaximally mixed state (
| ↑↑ |
+
| ↓↓ |
)
/
2. After Alicehas looked at the result, she has transferred this one bitof entropy, created by the measurement, to her memory.From the point of view of Bob, outside her isolated lab,Alice’s measurement is simply a (quantum) correlationof her measurement apparatus to the spin. [A thoroughanalysis of the microscopic details and of the thermody-namics of this type of measurement is given in Ref. [16].]The initial state of the spin
| →
= (
| ↑ 
+
| ↓ 
)
/
√
2evolves into the correlated (entangled) state(
| ↑ |
Alice sees “up”
+
| ↓ |
Alice sees “down”
)
/
√
2
,
(3)where the first ket in the two terms refers to the spinstate, whereas the second ket refers to the rest of Al-ice’s lab. Thus, from the point of view of Bob, Alice’smeasurement is an evolution similar to a controlled-NOTunitary transformation of the type
U 
cnot
(
|
0
+
|
1
)
|
0
=
|
0
|
0
+
|
1
|
1
. Such a transformation can be easily in-verted, as it is its own inverse. Analogously, Bob canflip a switch and invert Alice’s measurement. At the endof his operation, all records of the measurement result(Alice’s notepad, her brain cells, the apparatus gauges)will have been decorrelated from the spin state. She willremember having performed the measurement, but shewill be (
must 
be) unable to recall what the measurementresult was. In addition, the spin has become uncorre-lated from the measurement apparatus, so it is returnedto a pure state. I emphasize that Bob’s transformationis not necessarily a reversion of the dynamics of Alice’slab. [Notice that a “quantum eraser” [27] only permitsto decide
a posteriori 
which of two complementary mea-surements to perform using previouslycollected data: themeasurement process is not actually erased, and the en-tropy does not decrease.]In both the above experiments, from Alice’s point of view, entropy definitely has been created after she hasinteracted with Bob’s light or his spin. However, thisentropy is subsequently coherently erased by Bob. At theend of the process, looking back at the evolution in herlab, she cannot see any violation of the second law: shehas no (cannot have any) record of the fact that entropyat one point had increased.
Entropic considerations.
The above thought exper-iments exemplify a general situation: entropy can de-crease, but its decrease is accompanied by an erasure of any memory that the entropy-decreasing transformationhas occurred. In fact, any interaction between an ob-server
A
and a system
C 
which decreases their entropy bya certain quantity, must also reduce their quantum mu-tual information by the same amount (unless, of course,the entropy is dumped into a reservoir
R
). The quantummutual information
S 
(
A
:
C 
)
≡
S 
(
ρ
A
)+
S 
(
ρ
C 
)
−
S 
(
ρ
AC 
)measures the amount of shared quantum correlations be-tween the two systems
A
and
C 
(
ρ
AC 
being the state of the system
AC 
, and
ρ
A
and
ρ
C 
its partial traces, i.e. thestates of 
A
and
C 
).Taking the cue from [28], I now prove the above asser-tion, namely I show that∆
S 
(
A
) + ∆
S 
(
C 
)
−
∆
S 
(
R
)
−
∆
S 
(
A
:
C 
) = 0
,
(4)where ∆
S 
(
X 
)
≡
S 
t
(
ρ
X
)
−
S 
0
(
ρ
X
) is the entropy differ-ence between the final state at time
t
and the initial stateof the system
X 
, and where ∆
S 
(
A
:
C 
) =
S 
t
(
A
:
C 
)
−
S 
0
(
A
:
C 
) is the quantum mutual information difference.Choose the reservoir
R
so that the joint state of the sys-tems
ACR
is pure and so that the evolution maintains thepurity (
R
is a purification space). Then the initial and fi-nal entropies are
S 
0
(
AC 
) =
S 
0
(
R
) and
S 
t
(
AC 
) =
S 
t
(
R
),respectively. Thus we find
S 
0
(
AB
) =
S 
t
(
AB
)
−
∆
S 
(
R
)which, when substituted into the left-hand-side term of (4), shows that this term is null. [This proof is valid alsoif the evolution is not perfectly known, i.e. if it is givenby a random unitary map, see Eq. (1).]Now, to prove that the above reduction of entropy en-tails a memory erasure, I show that this erasure mustfollow from the elimination of quantum mutual informa-tion. A memory of an event is a physical system
A
whichhas nonzero classical mutual information on a system
C 
that bears the consequences of that event. Then, theerasure of the memory follows from an elimination of thequantum mutual information
S 
(
A
:
C 
) if this last quan-tity is an upper bound to the classical mutual information
of 00

Leave a Comment

You must be to leave a comment.
Submit
Characters: ...
You must be to leave a comment.
Submit
Characters: ...