Beruflich Dokumente
Kultur Dokumente
Arun K. Pati
Abstract. Origin of the speed-up in quantum algorithms is not yet fully understood. But there are
indications that entanglement does play an important role in quantum computation. Algorithms that
do not involve entanglement can be efficiently simulated on a classical computer. Here I make a
simple observation. I show that infinitesimal change in multi-particle pure product state always
gives rise to an entangled state. During quantum computation even though at each time instant the
state is not entangled, it does pass through entangled states at infinitesimal time steps. This applies
to multi-particle pure, pseudo-pure and general mixed states as well. This suggests that even though
unitary operators do not produce any entanglement, quantum entanglement guides the process of
quantum computation.
Keywords: Quantum entanglement, quantum computing, speed-up
PACS: <Missing classification>
INTRODUCTION:
Quantum entanglement is one of the much studied subject in recent years due to its
potential application in information processing. In early days, the notion of quantum
entanglement was much debated concept. Einstein was not in favor of such a notion as
it leads to spooky-action at a distance [1]. But, now we know that the spooky-action
is not so disturbing as it cannot be used on its own for faster than light communica-
tion. However, supplemented by classical communication, quantum entanglement can
become a resource for very useful and exotic information processing tasks. Many im-
portant tasks like super dense coding, quantum teleportation, remote state preparation,
quantum cryptography etc require quantum entanglement. It is also argued that quantum
entanglement may play an important role in quantum algorithms [2] and in giving extra
power to quantum computers [3].
Is it also at the heart of quantum computation? Yes, indeed. I make a simple yet
an important observation that could throw some light on the role of entanglement in
quantum evolution and this in turn may answer the question where from the extra power
comes for quantum computation.
Usual quantum computation paradigm involves preparation of initial logical states
and application of sequence of unitary evolution operators (prescribed by a particular
quantum mechanical algorithm) and then finally reading out the desired answer. In this
context an important question has been whether linear superposition alone is sufficient
to have the required speed-up or we need quantum entanglementthe weirdest feature
of quantum world. Though the existing quantum algorithms such as Deustch-Jozsa [4],
Grover [5] and Shor [6] require quantum entanglement it is not clear whether in general
115
qubits is taken to be an equal superposition of all possible bit strings, i.e.,
1 N 1
0 x
Nx 0
(1)
116
sy k 0
1
sz k cos2 k sin2 k (5)
N 1
The bipartite entanglement in the pure state may be characterized by calculating the
von Neumann entropy of this reduced state. Using the expansion formula
1 s s 1 2s
log sP log P log 1 (6)
2 2 1 s
which holds for any 0 s 1, the von Neumann entropy may be calculated to be given
by
S k tr k log k
1 sk
1 log 1 s k
2
1 sk
log 1 s k (7)
2
The right-hand-side of this expression is independent of the choice of the remaining
qubit . Therefore, (7) holds for any one qubit versus n 1 qubit partitioning. It shows
that the reduced density matrix of the single qubit does not arise from a maximally
entangled state of n qubits, as the von Neumann entropy is not exactly unity. Since the
reduced state of Eq. (3) is not pure the full state must be entangled. To see how impure
the state in Eq. (3) is one may calculate the linear entropy L of it which is given by
1 sk 2
L k tr k k 2
(8)
2
If the linear entropy is zero the state is pure and as it approaches 12 the state approaches
a completely random mixture. In the quantum search algorithm the parameter s k can
never be zero because that would mean that cos k and sin k are simultaneously zero,
which cannot be satisfied. So although the reduced density matrix of the qubit may lie
close to the completely mixed state it can never become the identity one.
Calculate the Hilbert-Schimdt norm of the difference of the completely mixed one
and the reduced state. This Hilbert-Schmidt distance for kth iteration during quantum
search algorithm is given by
I I
d k 2
k 2
HS tr k 2
2 2
1 sk 2
L k (9)
2 2
The distance d k provides an idea of how the reduced state of an individual qubit
behaves during the kth iteration. It shows that the reduced density matrix of the qubit
differs from a completely random mixture by an order of O s k . From Eq. (5) and (9)
that for 0 sin 1 1 N and for k 2 the reduced density matrix of any remaining
qubit is pure, implying that the whole state must have been non-entangled. Thus, we see
that although the initial and target states are separable, the intermediate states through
which the system evolves are always entangled.
117
MULTIPARTICLE QUANTUM EVOLUTION NEEDS
ENTANGLEMENT
In this section, I make a simple yet an important observation that could throw some light
on the role of entanglement in quantum evolution and this in turn answers the ques-
tion where from the extra power comes for quantum computation. We show that for any
multi-particle state infinitesimal change in the pure product state gives rise to an entan-
gled state. In the language of differential geometry given any manifold of multiparticle
quantum states (product or entangled), the tangent space is an entangled manifold. This
shows that even if at each instant of time the state is non-entangled, at infinitesimal time
steps the state is entangled. Thus, any multiparticle continuous evolution requires entan-
glement. This I call the hidden power of quantum entanglement. Applied to quantum
computation, this implies that during computation even though at each time instant the
state is a product state, it does pass through entangled states at infinitesimal time steps.
In other words, even non-entangling evolution needs entanglement at infinitesimal time
steps.
Consider a composite quantum system consisting of two or more subsystems. (For
simplicity we consider bi-partite systems in finite dimensional Hilbert space, but our
results hold for any number of particles and in any dimension). Let be a set of
vectors in 1 2 . If these vectors are not normalized we can consider a set
of vectors of norm one in . The set of rays of is called the projective
Hilbert space 1 2 . If dim 1 N 1 and dim 2 N 2 then
, CN1 N2 . The
projective Hilbert space space is CN1 N2 0 U 1 which is a complex manifold
of dimension N1 N2 1 . This can also be considered as a real manifold of dimension
2 N1N2 1 . Any quantum state (product or entangled) at a given instant of time can
be represented as a point in . The evolution of the state vector can be represented
by a curve : t t in whose projection lies in . Here, smooth
mappings : 0 t of an interval into a differentiable manifold are called smooth
curves in the manifold.
Let be a differentiable manifold embedded in R, and . A vector v is
called a tangent vector to at if there is a smooth curve passing through
d
such that v dt . The tangent space T of at is the set of all tangent
vectors to at . The tangent space to a differentiable manifold at the point
is a linear space having same dimension as that of . What we will prove is that
given any multiparticle pure state , the tangent vector in infinitesimal time step
d T is entangled.
118
T U T 0 , where U T exp iHT and H is the total Hamiltonian of
the system. The same U T can be obtained from infinitesimal changes via U T
limN I iH T N N . Now, if U t 0 t T is capable of producing entanglement,
then the state can be written as t i1 i2 in Ci1 i2 in t 1i2 in . There is clearly
entanglement present at any stage of quantum evolution as well as during infinitesimal
time steps. The tangent vector d is also an entangled one.
The surprising thing is that even if U t does not produce any entanglement, to be
able to have a continuous evolution we need quantum entanglement. Suppose we have a
composite system that consists of two subsystems. The state of the combined system is
then 1 2 1 2 , where 1 1 and 2 2 . Now consider the
infinitesimal change in the state vector (i.e., the tangent vector at ). This is a linear
mapping d : d and can be thought of as a derivation at on a differentiable
manifold . The infinitesimal change in is given by
1 d 2 d 1 2 T (10)
The above state is clearly entangled for generic changes in the subsystems 1 and 2 as
we cannot write d as tensor product of two infinitesimal changes in the respective
Hilbert spaces. Once we choose coordinates for 1 and 2 in , then there are
no coordinates which can express (1) as product states unless d i i , i 1 2 .
But the later corresponds to stationary states, whereby the subsystems do not undergo
generic change. Now, if 0 1 1 1 N1 1 are homogeneous coordinates for 1 and
1
0 2 1 2 N2 1
are homogeneous coordinates for 2 , then the tangent vector can
2
be written as
2 1
d 1 d i 2 d i 1 2 (11)
i 2 2 i 1 1
2 1
119
Thus, we can say that when the state passes through infinitesimal changes entangle-
ment is necessary. This is because the tangent vector which tells us how the state vector
changes is typically entangled as given in (12). This is potentially one hidden power
of quantum entanglement. In any quantum universe, even if there is no direct or indirect
interaction between constituents, mere infinitesimal changes in two or more implies that
the infinitesimal change in the combined state is entangled! For product states one would
think that the global change can always be described as local changes. However, our ob-
servation shows, somewhat surprisingly, that whether the composite system is entangled
or non-entangled, global change cannot always be described as local changes.
Remarks: Note that the infinitesimal change cannot be applied as a bi-local operation,
i.e., an operation taking d d is an impossible one. This violates the
norm preservation. For bi-partite systems this would mean 1 2 1 d 2
cannot happen. We can prove this by contradiction. Suppose we have the mapping
f : d d . Then we have 1 d 1 2 d 2 . For any
normalized state we must have d as a purely imaginary number. This implies
that on lhs we have which is a purely imaginary number and on rhs we have
product of two purely imaginary numbers which is a real number. Since this cannot
hold, there is no bi-local infinitesimal changes. The proof can be generalized for more
than two subsystems. However, if one of of the subsystem does not undergo infinitesimal
change then it is possible to satisfy the norm preservation or isometric evolution. This
means we can have d I and I d but not d d .
120
d
entangled state given by d cos sin T . This implies that
two identical, non-entangled qubits however far separated, when we look at the change
in the combined state through infinitesimal time steps, then the infinitesimal change in
the state is a highly non-local state. For example, the quantum mechanical correlation in
the state d (up to local unitaries) is given by
d a b d a a E a b d 2 (16)
where E a b a b is the standard quantum mechanical correlation is a maximally
entangled state. This means if one looks at change in infinitesimal steps, one may
observe violation of Bells inequality even for product states.
121
an evolution given by
2 2 2 2 tr1 d d d 2 d 2
2 d 2 d 2 2 1 d 1
2 2 d 1 1 (19)
This shows that when d , then 2 transforms to d 2 along with noise
terms. Eqs(18) and (19) clearly show the entangling nature of infinitesimal change. If
it has no ability to create entanglement, then 1 would have gone to d 1 and 2
would have gone to d 2 under global infinitesimal changes.
At any stage of the computation (say kth step) we can write the n-qubit state generically
as
2n 1
1
k UkUk 1 U1
2
n xi (21)
i 0
122
optimize these paths. If i 0 is the initial state and f is the final state then the
total distance between them is given by [13, 14]
d 2 i f 41 i f 2
d 2 i U 41 i U 2
The success of an algorithm depends on how to minimize the number of steps. If the
system moves along geodesic paths (shortest paths) in the projective Hilbert space then
it can reach the desired state much faster. For example, in Grovers algorithm it can be
shown that the states indeed pass through geodesics and the number of steps calculated
using the above formula is exactly O N [16].
k u1 k u2 k un k 0 (22)
where each of these ui s are some local unitaries acting on single qubit Hilbert space
2 . But the infinitesimal change in the above state T is given by
k k
dk du1 k u2 k un k
u1 k u2 k dun k 0 (23)
which is an entangled state. This shows that the weirdest feature of quantum world
plays its role in every computation in a very subtle way. This is truly a hidden power
of quantum entanglement in quantum computation. It is hidden because, we do not look
at infinitesimal steps; we always consider finite time steps. Quantum entanglement is
necessary for any evolution (entangling or not) and hence for quantum computation.
We can say that during quantum computation possible directions in which one can pass
through a n-qubit register state is guided by entanglement.
123
One may ask does entanglement also plays any role when mixed state are involved
during quantum computation? This is exactly the case when one deals with NMR
implementations. There one typically encounters pseudo-pure states which comes as
a convex combination of a random mixture and a pure state. For n-qubits this is given
by 1 2In 0 0 , where is the purity parameter. After application of
sequence of unitary operators during certain computation the state changes as
U U 1 2In . It has been claimed that the states produced thus are
still not entangled even though the pure state component U 0 is entangled.
However, now we see how does entanglement play a role during quantum computation?
What we say is that even though the the pure state component is not entangled, the
infinitesimal change in , i.e., d indeed is entangled. This is because d d
d is an entangled one. Thus any continuous evolution of does require quantum
entanglement.
Our observation not only applies to multi-particle pure and pseudo-pure states but to
any mixed states as well. Consider a separable multi-particle (again for simplicity say
bi-partite) mixed state given by
pi i 1 i 2 (24)
i
where i 1 1 and i
2
2 are pure state components of subsystems 1
and 2, respectively with pi 0 i pi 1. We can show that even this separable state
when evolves in time, the infinitesimal change in the state is an entangled one. The
infinitesimal change in is given by
d pi i 1 d i 2 d i 1 i 2 (25)
i
To prove that d is entangled, let us assume that it is separable and then arrive at a
contradiction. If d is separable then there must be a decomposition such that we can
write this as
d wid i 1 d i 2 (26)
i
124
One may ask since any classical computer state can be written as a separable state
would that have the same behavior? The answer is no, because to have entangled tangent
vector we need derivative behavior and tensor product structure on the linear space. Also
one may ask if we have ordinary probabilistic description for a system comprising two
subsystems would we say that non-product states are necessary for any probabilistic
evolution as well? The answer will depend on whether we have tensor product structure
on a linear space as a form of description.
CONCLUSION
We have shown that entanglement is necessary for any quantum evolution. In short
any generic change in a quantum universe does require entanglement. Geometrically,
given a manifold of multi-particle quantum states if there are changes in two or more
subsystems then the tangent space is an entangled manifold. Since the tangent space
vectors tell the state-space vectors how to change so our result tells us something deep
about motion or change in general. Also we have shown that infinitesimal change cannot
be a bi-local operation. We have studied the reduced dynamics of the subsystem under
infinitesimal operation. This has immediate implication in quantum computation, where
one can argue that even though there is no entanglement generated during any stage
of computation, the evolution of multi-qubit state is guided by entanglement. This result
applies to pure state, pseudo-pure state and mixed state implementations as well. Though
the result of this paper may appear simple, it is nevertheless non-trivial. We hope that
this observation will unfold many other results in quantum theory and in the fast growing
field of quantum information theory.
In the spirit of the statement Space-time tells matter how to move and matter tells
space-time how to curve, I would like to conclude by saying that entanglement tells the
computation how to move and computation tells the tangent vector how to entangle.
REFERENCES
1. A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. 47, 777 (1935).
2. R. Jozsa, Geometric Issues in the Foundations of Science, Eds. S. Huggett et al, (Oxford University
Press, Oxford, 1997).
3. R. Fitzgerlad, Phys. Today, 53(1), 20 (2000).
4. D. Deutsch and R. Proc. R. Soc. London 439, 553 (1992).
5. L. K. Grover, Phys. Rev. Lett. 79, 325 (1997).
6. P. W. Shor, Symposium on Fundamentals of Computer Science (FOCS) 56 (1994).
7. S. L. Braunstein, C. M. Caves, R. Jozsa, N. Linden, S. Popescu, and R. Schack, Phys. Rev. Lett. 83,
1054 (1999).
8. D. Collins, K. W. Kim and W. C. Holton, Phys. Rev. A 58, R1633 (1998).
125
9. Arvind, Pramana J. of Phys. 56, 357 (2001).
10. S. L. Braunstein and A. K. Pati, Quantum Information and Computation (QIC), 2(2), 399 (2002).
11. N. Linden and S. Popescu, Phys. Rev. Lett. 87, 047901 (2001).
12. R. Jozsa and N. Linden, Proc. R. Soc. Lond. A 459, 2011 (2003).
13. A. K. Pati, Phys. Lett. A 159, 105 (1991).
14. A. K. Pati, Phys. Rev. A 52, 2576 (1995).
15. M. A. Nielsen and I. Chuang, Quantum Computation and Information, (Cambridge University Press,
Cambridge, 2000)
16. A. K. Pati, Quant-Ph/9807067, (1998).
.
126