Sum Uncertainty Relation in Quantum Theory

A. K. Pati and P. K. Sahu

Institute of Physics, Bhubaneswar-751005, Orissa, India

We prove a new sum uncertainty relation in quantum theory which states that the uncertainty in the sum of two
or more observables is always less than or equal to the sum of the uncertainties in corresponding observables.
This shows that the quantum mechanical uncertainty in any observable is a convex function. We prove that if we
have a finite number
√ N of identically prepared quantum systems, then a joint measurement of any observable
gives an error N less than that of the individual measurements. This has application in quantum metrology that
aims to give better precision in the parameter estimation. Furthermore, this proves that a quantum system evolves
slowly under the action of a sum Hamiltonian than the sum of individuals, even if they are non-commuting.
arXiv:quant-ph/0608092v1 10 Aug 2006

PACS numbers: 03.67.-a, 03.65.Bz, 03.65.Ta, 03.67.Lx

Unlike in classical physics, there are restrictions in quan- ∆A, and ∆B? The following theorem answers this.
tum theory on how accurately one can measure observables
even in principle [1]. This is well documented by the famous Theorem: Quantum fluctuation in the sum of any two
Heisenberg uncertainty relation for position and momentum observables is always less than or equal to the sum of their
of a quantum particle [2]. Later on the uncertainty relation individual fluctuations, i.e., ∆(A + B) ≤ ∆A + ∆B.
was generalized for any two non-commuting observables [3].
It tells that the product of uncertainties in two non-commuting Proof: Let A and B are two observables which could be com-
observables in a given quantum states is greater than or equal muting or non-commuting. Let us define two unnormalized
to the average of their commutator in the corresponding quan- vectors |Ψ1 i = (A − hAi)|Ψi and |Ψ2 i = (B − hBi)|Ψi,
tum state. The Heisenberg uncertainty relation is then a spe- where hAi = hΨ|A|Ψi and hBi = hΨ|B|Ψi are quantum me-
cial case of this generalized uncertainty relation. chanical averages of A and B, respectively in the state |Ψi.
Consider the norm of sum of two vectors |Ψ1 i + |Ψ2 i. This is
Here we ask, given two or more observables of a quantum
given by
system if one measures their sum, will the uncertainty be more
or less than the sum of their individual uncertainties? It turns ||Ψ1 + Ψ2 ||2 = ||Ψ1 ||2 + ||Ψ2 ||2 + 2RehΨ1 |Ψ2 i
out that the error introduced in the sum of observables is al-
= ∆A2 + ∆B 2 + 2RehΨ1 |Ψ2 i, (1)
ways less than or equal to the sum of the errors introduced
by individual observables. This we term as the sum uncer- where ||Ψ1 ||2 = hΨ1 |Ψ1 i = hΨ|(A − hAi)2 |Ψi = ∆A2
tainty relation—which forms the basis to show that quantum and ||Ψ2 ||2 = hΨ2 |Ψ2 i = hΨ|(B − hBi)2 |Ψi = ∆B 2 . Us-
mechanical uncertainty in any observable is actually a convex ing the fact that RehΨ1 |Ψ2 i ≤ |hΨ1 |Ψ2 i| and further using
function. Furthermore, we show that if we have a finite num- the Schwartz inequality we have RehΨ1 |Ψ2 i ≤ ||Ψ1 || ||Ψ2 ||.
ber of identically prepared quantum systems, then the mea- Then the norm of sum of two vectors satisfy
surement
√ of the collective observable gives an error which is
N smaller than the one obtained via individual measure- ||Ψ1 + Ψ2 ||2 ≤ ∆A2 + ∆B 2 + 2∆A∆B. (2)
ments. We apply these ideas in quantum metrology that aims
to give better precision in the parameter estimation. More- On the other hand direct evaluation of ||Ψ1 + Ψ2 || gives
over, we will give some examples and illustrate the relation
for some simple quantum mechanical systems. One conse- ||Ψ1 + Ψ2 ||2 = hΨ|[(A − hAi) + (B − hBi)]2 Ψi
quence of the sum uncertainty relation is that a quantum sys- = hΨ|(A + B)2 |Ψi − hΨ|(A + B)|Ψi2
tem evolves more slowly under the action of a sum Hamil- = ∆(A + B)2 . (3)
tonian than the sum of either separately, i.e., mixing of even
non-commuting Hamiltonians slows down the system. Thus, (2) and (3) imply that
Sum uncertainty relation: Consider a quantum state |Ψi ∆(A + B) ≤ ∆A + ∆B (4)
in a Hilbert space H. Let A and B are two general ob-
servables (they could be commuting or non-commuting) which is the sum uncertainty relation. Hence, the proof.
that represent some physical quantities. Then, the quantum The physical meaning of the sum uncertainty relation is that
mechanical uncertainties associated with these observables in if we have an ensemble of quantum systems then the igno-
the state |Ψi are defined via ∆A2 = hΨ|A2 |Ψi − hΨ|A|Ψi2 rance in totality is always less than the sum of the individual
and ∆B 2 = hΨ|B 2 |Ψi − hΨ|B|Ψi2 . Similarly, we can ignorance. In case of two observables, if we prepare a large
define the uncertainty in the sum of two observables as number of quantum systems in the state |Ψi, and then per-
∆(A + B)2 = hΨ|(A + B)2 |Ψi − hΨ|(A + B)|Ψi2 . Here, we form the measurement of A on some of those systems and B
address the question: what is the relation between ∆(A + B), on some others, then the standard deviations in A plus B will
 2

be more than the standard deviation in the measurement of consists of N -identically prepared quantum systems. Let each
(A + B) on those systems. Hence, it is always advisable to go system be in the state |Ψi. Therefore, the combined state vec-
for ‘total measurement’ if we want to minimize the error. tor of N -particle is given by
In fact, it is not difficult to see that if we have more than
two observables (say three observables A, B, and C), then the |Ψi⊗N = |Ψi1 ⊗ |Ψi2 ⊗ · · · ⊗ |ΨiN . (9)
sum uncertainty relation will read as
Let us first measure an observable A on each particle in-
∆(A + B + C) ≤ ∆A + ∆B + ∆C. (5) dividually (not collectively). The individual observables of
interest are A1 = A ⊗ I ⊗ · · · I, A2 = I ⊗ A ⊗ · · · I, .... and
In general for N observables A1 , A2 , · · · , AN , we will have AN = I ⊗ I ⊗ · · · A. Then, one can check that the average of
the sum uncertainty relation as Ai (i = 1, 2, . . . N ) in the state |Ψi⊗N = hΨ|A|Ψi. Similarly,
X X the uncertainty in each Ai is ∆A. Therefore, Pthe sum of uncer-
∆( Ai ) ≤ ∆Ai , (i = 1, 2, · · · , N ). (6) tainties in the individual measurements is i ∆Ai = N ∆A.
i i Now, suppose we perform measurement of the sum observ-
Convexity of quantum uncertainty: The above inequality able on N -copies. The sum observables is given by
brings out an important property of the quantum uncertainty X
with convexity of a function. To be specific, we will show that AS = Ai = A ⊗ I ⊗ · · · I + I ⊗ A ⊗ · · · I
i
the quantum mechanical uncertainty in any observable is ac-
tually a convex function. Recall that f is a convex function if + · · · + I ⊗ I ⊗ · · · A. (10)

The quantum mechanical uncertainty in the observable AS in

the state |Ψi⊗N is given by
X X
f( pi xi ) ≤ pi f (xi ), (i = 1, 2, · · · , N ) (7)
i i
∆A2S = ⊗N hΨ|A2S |Ψi⊗N − (⊗N hΨ|AS |Ψi⊗N )2 . (11)
P
where pi ’s satisfy 0 < pi < 1, i pi = 1 and xi is in the set
S [4]. Note that under scaling transformation of an operator Note that ⊗N hΨ|A2S |Ψi⊗N = N hΨ|A2 |Ψi + 2N hΨ|A|Ψi2
A → λA, the quantum mechanical uncertainty transforms as and ⊗N hΨ|AS |Ψi⊗N = N hΨ|A|Ψi. Hence, ∆A2S =
∆A → λ∆A. Using this fact we can have a more general sum N ∆A2 . This implies that the quantum√mechanical uncer-
uncertainty relation. For example, if we have a sum observ- tainty in the sum observable is ∆AS = N ∆A. Therefore,
P
able i pi Ai , with all pi ’s as positive numbers, then we have the sum uncertainty relation reads as
the following sum uncertainty relation √
N ∆A ≤ N ∆A (12)
X X
∆( pi Ai ) ≤ pi ∆Ai , (i = 1, 2, · · · , N ). (8) which is clearly satisfied. This analysis √ also suggests that the
i i error in the total measurement goes as N ∆A, whereas the
The meaning of this general sum uncertainty relation is that sum of errors in the individual
√ measurement goes as N ∆A.
‘mixing of commuting or non-commuting operators’ always Thus, there is an overall N improvement in the error of mea-
decreases the uncertainty. surement of sum observable with N -copies.
Now, note that inP(8) if the positive numbers pi ’s satisfy Parameter estimation and quantum metrology: Precision
0 < pi < 1 and i pi = 1, then ∆ is indeed a convex measurement which requires estimation of some parameter to
function. Furthermore, it is known that if f1 , f2 , . . . fn are its highest accuracy is an important problem. If one uses laws
convex functions on R and pi ≥ 0, (i = 1, 2, . . . n), then of quantum theory, then in the measurement of some parame-
ter the precession can be enhanced. This a topic of great study
P
f (x) = i pi fi (x) is also a convex function on R. By us-
ing this property of convex function, we can draw the fol- in quantum metrology. In this scheme one prepares a probe
lowing conclusion. Suppose we have several quantum states state |ψ0 i, applies a unitary operator U (θ) that depends on the
|Ψ1 i, |Ψ2 i, . . . |Ψn i and the quantum mechanical uncertain- parameter θ to be estimated and then measures some observ-
ties in A, in theP above states are ∆1 , ∆2 , . . . ∆n . Then, it fol- able X on the resulting state |ψ(θ)i. If U (θ) = exp(−iθH)
lows that ∆ = i pi ∆i is a convex function. The usefulness where H is a Hermitian operator, then using the Mandelstam-
of convexity of quantum uncertainty may be similar to the en- Tamm uncertainty relation [5, 6], we have
tropy of a quantum system. It may be mentioned that entropy 1  
of a quantum state is a concave function. Also, we know that ∆X∆H ≥ ∂hXi/∂θ. (13)
2
entropy signifies the information content of a quantum state.
So, in that sense one can think of negative of quantum uncer- The precision with which one can estimate θ is given by
tainty as a concave function and it may represent as a ‘mea-   1
sure of information’ (ignorance). Thus, this property unravels δθ = ∆X/∂hXi/∂θ ≥ , (14)
2∆H
another feature of quantum mechanical uncertainty.
Error in collective and individual measurements: We can test where ∆X, ∆H, and hXi have their usual meaning in the
the inequality (6) explicitly by considering an ensemble that quantum state |ψ(θ)i. Therefore, if we want to minimize the
 3

error in estimating the parameter, we have to minimize ∆X Braunstein [9]. But there is no further interaction between
or maximize ∆H. How to achieve that goal is the subject of these M -copies. Then, the combined state of M N probe state
quantum metrology [7]. It turns out that using quantum en- is given by
tangled probe states or entangling unitary operator one can
achieve better and better precision in the parameter estima- |ψH (θ)i⊗M = |ψH (θ)i1 ⊗|ψH (θ)i2 ⊗· · ·⊗|ψH (θ)iM . (19)
tion.
Recently, Giovannetti et al [8] have shown that that using On these collection of M N probe states we measure a sum
entangled probe state one can achieve an enhancement that observable. The observable of interest is
scales as 1/N . More recently, it was shown by Roy and ΠS = Π1 ⊗ I2 ⊗ · · · IM + I1 ⊗ Π2 ⊗ · · · ⊗ IM + · · ·
Braunstein [9] that if one exploits entangling unitary opera-
tor then one obtains an exponential enhancement in the pa- + I1 ⊗ I2 ⊗ · · · ⊗ ΠM , (20)
rameter estimation. In particular, by choosing an appropriate
where Πj , (j = 1, 2, . . . M ) are product of projection opera-
Hamiltonian one can apply the unitary operator U = e−iθH
tors on jth copy of the N -probe state. To be clear, Π1 = P =
and generate a N -qubit state given by (for details see [9])
⊗N N
i=1 Pi on the 1st N -probe state, Π2 = P = ⊗i=1 Pi on the
|ψH (θ) = e−iθH |00 . . . 00i (15) 2nd N -probe state and so on. The precision in the measure-
N −1 N −1 ment of the sum observable ΠS is given by
= cos(2 θ)|00 . . . 00i − i sin(2 θ)|11 . . . 11i.
 
Then, by measuring the observable X = ⊗N δθ = ∆ΠS /∂hΠS i/∂θ. (21)
i=1 Pj , where Pj =
|0ij h0| one can estimate θ as given by δθ = 1/2N which is
One can check that the quantum uncertainty and average in
the exponential enhancement in the precision.
ΠS for the M N probe state |ψH (θ)i⊗M are given by
Here, we show that there are other class of measurement
√
strategies also which can give the same precision. Suppose, M
instead of measuring the product observable we measure the ∆ΠS = sin(2N θ), hΠS i = M cos2 (2N −1 θ). (22)
P 2
sum observable, i.e., measure X = i Pi = PS = P1 ⊗ I ⊗
· · · I + I ⊗ P2 ⊗ · · · ⊗ I + · · · + I ⊗ I ⊗ · · · ⊗ PN . Then, the Therefore, the precision in the estimation of the parameter θ
precision in the parameter estimation is given by is given by
√
δθ = 1/ M 2N .
 
δθ = ∆PS /∂hPS i/∂θ. (16) (23)
The quantum uncertainty and average for PS in the state This result apparently may give an impression that this is bet-
|ψH (θ)i are given by ter than exponential [9]. But, if we re-express the result in
N terms of actual resource√used, i.e., the number K = M N ,
M −1
∆PS = sin(2N θ), hPS i = N cos2 (2N −1 θ). (17) then the precision δθ = N 2K( M ) √K2 1
, which is lower
2 K

than the exponential. Therefore, it is always not the case that

Therefore, the precision in the estimation of the parameter θ is by using more resources one can enhance the precision.
1/2N . Similarly, if we measure individually these projectors Now, we give few further applications of the sum uncer-
then the corresponding precision in the parameter θ will be tainty relation in quantum theory. First, we apply to Hamil-
X  X  tonian systems and second we apply to the speed of quantum
δθ = ∆Pi /∂( hPi i)/∂θ. (18)
mechanical systems.
i i
P Uncertainty in the Hamiltonian: One immediate applica-
One can check that for the state |ψH (θ)i, we have i ∆Pi = tion is that for any quantum mechanical system, the total
N N 2 N −1
P
2 sin(2 θ) and i hPi i = N cos (2 θ). Thus, again we Hamiltonian H consists of kinetic and potential energy, i.e.,
N
see that δθ = 1/2 . Hence, for the entangling unitary, joint H = T + V and using the sum uncertainty relation we have
measurement and individual measurements give the same pre- ∆H ≤ ∆T + ∆V . Thus, the uncertainty in the total energy
cision as obtained in [9]. The physical explanation is now in any state is bounded by the sum of uncertainties in the ki-
clear. This is happening because, the sum uncertainty relation netic and potential energy. This result is interesting, in the
is saturated for these observables and the equality holds. sense that if we want to do energy measurement with minimal
One may wonder if by increasing the resources and by ex- error, then do not measure kinetic and potential energy sepa-
ploiting the sum uncertainty relation (i.e. the idea that the rately. Always measure the total energy because the quantum
measurement of sum observable minimizes the error leading mechanical uncertainty is less in that case. Also, this shows
to better precision) one can enhance the precision in the pa- that the uncertainty in the total energy is less than the uncer-
rameter better than the exponential [9]. However, as we will tainty in the kinetic energy in the position basis. However
see this is not the case. Let us imagine that we have an en- in the momentum basis, the uncertainty in the total energy is
semble of N -probe states. The number of copies of N -probe less than the uncertainty in the potential energy. These obser-
states are finite (say) M . On each of the N -probe state we ap- vations may have many applications in the complex quantum
ply the entangling unitary operator as suggested by Roy and systems.
 4

Sub-additivity of quantum speed: Here, we ask whether the and H is the Hamiltonian. Then, the magnitude of the av-
speed of evolution of a state vector through Hilbert space erage of the velocity operator will obey |hvi| ≤ ~2 ∆x∆H.
behaves like the classical speed. In what follows, we will Now, if we have a Hamiltonian H1 , then the velocity opera-
show that the speed with which a quantum system evolves tor will obey |hv1 i| ≤ ~2 ∆x∆H1 . Similarly, for a Hamilto-
under two Hamiltonians (commuting or non-commuting) are nian H2 , the velocity |hv2 i| ≤ ~2 ∆x∆H2 . This implies that
not added up. (Note that classically, if a particle is subjected |hv1 i|max = ~2 ∆x∆H1 and |hv2 i|max = ~2 ∆x∆H2 . Now,
to two force fields, then the speed of a particle is added up.) using the sum uncertainty relation we have
In quantum theory, when a system evolves under some
Hamiltonian H, then the state undergoes a continuous time 2
evolution, i.e., |Ψ(0)i → |Ψ(t)i = exp(−iHt)|Ψ(0)i. One |hvi| ≤ ∆x(∆H1 + ∆H2 ). (27)
~
can ask how fast does the system evolve in time. Then, the
rate at which it evolves is nothing but the speed of transporta-
tion of the state vector in the projective Hilbert space [10, 11]. This suggests that |hvi| ≤ |hv1 i|max + |hv2 i|max . This is an-
This is defined as v = dD other interesting application of the sum uncertainty relation.
dt , where dD is the infinitesimal dis-
tance between nearby quantum states |Ψ(t)i and |Ψ(t + dt)i. Conclusion: We have proved a new sum uncertainty relation
The distance function is given by for general observables in quantum theory which shows that
quantum mechanical uncertainty in the sum of two or more
dt2
dD2 = (1 − |hΨ(t)|Ψ(t + dt)i|2 ) = ∆H 2 , (24) observables is always less than or equal to the sum of quan-
~2 tum uncertainties in the individual observables. We have also
where ∆H is the usual uncertainty in the Hamiltonian in the proved that the quantum mechanical uncertainty is indeed a
state |Ψi. Therefore, the speed at which a quantum system convex function. This property suggests that there is some
evolves is nothing but the uncertainty in the Hamiltonian of analogy between quantum uncertainty and entropy of a quan-
the system, i.e.,v = ∆H/~. This is the geometric meaning of tum mechanical system. We have shown that if we have a fi-
quantum fluctuation: more the fluctuation in the Hamiltonian, nite number of
√ identically prepared quantum states, then there
faster the system will evolve. is an overall N improvement in the error of measurement of
Now, imagine that a quantum system evolves under a the sum observable with N -copies. As an important applica-
Hamiltonian H1 , then the speed is given by v1 = ∆H1 /~. tion we have explained why the measurement of the sum and
Similarly, if this evolves under a Hamiltonian H2 , then the individual observables can give the same exponential preci-
speed is given by v2 = ∆H2 /~. Suppose, now the system sion. Also, we have shown that using more resources one can-
evolves under the Hamiltonian H = H1 + H2 . What will be not have a precision better than the exponential one. In addi-
the speed? Will the total speed be v = v1 + v2 ? The an- tion, we prove that in general a quantum system evolves more
swer is no. Using the sum uncertainty relation we see that slowly under the action of a sum Hamiltonian than the sum of
v = ∆H/~ ≤ ∆H1 /~ + ∆H2 /~. In other words, the quan- either separately. It is expected that the sum uncertainty rela-
tum speed obeys the relation tion will have wider applications in a variety of context like
quantum computation, quantum information theory and many
v ≤ v1 + v2 . (25)
body quantum systems.
The meaning of this equation is that in general a quantum sys- Acknowledgment: We are thankful to S. L. Braunstein for
tem will evolve more slowly under the action of a sum Hamil- useful remarks.
tonian than the sum of either separately. This is a non-trivial
result, in the sense that this holds for generic Hamiltonians
be they commuting or non-commuting. This is something
counter intuitive which arise due to quantum mechanical na-
ture of the associated observables and also the fact that quan- [1] P. A. M. Dirac, Principles of Quantum mechanics, Cambridge
tum systems obey the Schrödinger equation and not the New- University Press, 1930.
[2] W. Heisenberg, Z. Phys. 43, 172 (1927).
ton equation. Also, it may be noted that if we have a Hamil-
[3] H. P. Robertson, Phys. Rev. 35, 667 (1930).
tonian H = H1 − H2 , the speed will obey the relation [4] M. Marcus and H. Minc, A Survey of Matrix Theory and Matrix
v ≤ v1 + v2 . (26) inequalities, Dover Publications, New York, (1992).
[5] L. Mandelstam and I. G. Tamm, J. Phys. USSR 9, 249 (1945).
This is due to the fact that ∆(−A) = ∆A, i.e., quantum me- [6] S. L. Braunstein and C. M. Caves, Phys. Rev. Lett. 72, 3439
chanical uncertainty is an even function. (1994).
[7] V. Giovannetti, S. Lloyd and L. Macone, Science, 306, 1330
One may ask whether the velocity operator of a quantum
(2004).
system obeys the sub-additivity condition. We will show that [8] V. Giovannetti, S. Lloyd and L. Macone, Phys. Rev. Lett. 96,
in general this may not. But the average of the velocity op- 010401 (2006).
erator may obey a kind of sub-additivity condition. Note that [9] S. M. Roy and S. L. Braunstein, quant-ph/0607152.
using the Heisenberg equation of motion, the velocity opera- [10] Y. Aharonov and J. Anandan, Phys. Rev. Lett. 65, 1697 (1990).
1 [11] A. K. Pati, Phys. Lett. A 159, 105 (1991).
tor can be defined as v = i~ [x, H], where x is the position