The Generalized Uncertainty Principle

Jun-Li Li and Cong-Feng Qiao∗

arXiv:2003.08705v2 [quant-ph] 19 Jan 2021

School of Physical Sciences, University of Chinese Academy of Sciences

YuQuan Road 19A, Beijing 100049, China
Center of Materials Science and Optoelectronics Engineering & CMSOT,
University of Chinese Academy of Sciences, YuQuan Road 19A, Beijing 100049, China
Key Laboratory of Vacuum Physics, University of Chinese Academy of Sciences
YuQuan Road 19A, Beijing 100049, China
∗
To whom correspondence should be addressed; E-mail: [email protected].

Abstract

The uncertainty principle lies at the heart of quantum physics, and is widely

thought of as a fundamental limit on the measurement precisions of incompatible

observables. Here we show that the traditional uncertainty relation in fact belongs

to the leading order approximation of a generalized uncertainty relation. That is, the

leading order linear dependence of observables gives the Heisenberg type of uncer-

tainty relations, while higher order nonlinear dependence may reveal more different

and interesting correlation properties. Applications of the generalized uncertainty

relation and the high order nonlinear dependence between observables in quantum

information science are also discussed.

1
1 Introduction

The proposition of the uncertainty principle may be attributed to the early efforts

devoted to incorporate the wave and particle natures to each individual quantum. Heisen-
berg stated that the canonically conjugate quantities, x and p, can be determined simul-
taneously only with a characteristic indeterminacy [1]. A well known formulation of the

uncertainty relation takes the following form [2]

1
∆X 2 ∆Y 2 ≥ |h[X, Y ]i|2 , (1)
4

where the variance ∆X 2 ≡ hX 2 i − hXi2 is a measure of the uncertainty for observable

X (similarly for Y ) and [X, Y ] ≡ XY − Y X is the commutator. In relation (1), X and
Y are no longer restricted to canonical variables but can be arbitrary observables. An

improvement of the uncertainty relation was made by Schrödinger [3]

1 1
∆X 2 ∆Y 2 ≥ |h[X, Y ]i|2 + |h{X, Y }i − 2hXihY i|2 . (2)
4 4

Here {X, Y } ≡ XY +Y X is the anticommutator. The uncertainty relations (1) and (2) are
normally explained as trade-off relations for the uncertainties of incompatible observables,
which is lower bounded by the expectation values of their (anti)commutator. In other

words, the variances of incompatible observables are not independent but correlated with
each other. In recent years, important progresses have been made in the study of variance
based uncertainty relations, e.g. [4, 5], but few studies concern about the higher order

moments of observables.
The lower bounds of the above variance based uncertainty relations depend on the
quantum state and may reach zero, in which case the uncertainty relations become trivial.

Partly due to this reason, the entropic uncertainty relation was introduced [6], with a

2
typical form of [7]
 
1
H(X) + H(Y ) ≥ ln , (3)
c
where c = maxi,j |hxi |yj i|2 is the maximum overlap of eigenbases |xi i and |yj i of X

and Y respectively and it turns out to be state independent. ln(·) here is the natural
logarithm. The Shannon entropy H(·) is another measure of uncertainty for observables
regarding to the probability distribution of the measurement outcomes. In the presence

of quantum memory, the bound of the entropic unceratinty relation may be reduced [8],
see Refs. [9] and [10] for recent developments. Note, recent study indicates that these
two superficially different forms of uncertainty relations (variance based and entropic) are

in fact interconvertible [11]. Moreover, it has been realized that the variance or entropy
alone may not be sufficient to fully characterize the properties of quantum uncertainty
[12], higher order moments of observables tend to be nontrivial [11]. The variance is

merely the second-order central moment.

In this work, by interpreting the incompatibility as the statistical dependence between
observables, we derive a generalized uncertainty relation (GUR) capable of embodying

the incompatibility (dependence) of observables to arbitrary order. In Section 2, after

a reinterpretation of complementary we employ the statistical quantity of cumulant to
measure the dependence between observables in different orders. Explicit form of GUR

is derived and expanded in terms of cumulants in Section 3. The first few terms in the
expansion are studied in the subsections of Section 3. The leading order nontrivial linear
dependence turns out to be the Heisenberg type uncertainty relation. The first nonlinear

dependence between incompatible observables gives the “Skewness uncertainty relation”.

In Section 4, concrete examples and simple applications of the new findings are given for
understanding the high order dependence. Section 5 remains for conclusion.

3
Figure 1. Two interpretations of complementarity. Regarding X ∪ Y as a whole,
we have that: (I) the completeness of X precludes the wholeness of Y ; (II) X and Y
cannot be as a whole simultaneously.

2 The uncertainty principle and dependence

To understand the wave-particle duality in atomic phenomena, a fundamental concept

“Complementarity Principle” was proposed by Bohr, which may be stated as “any given

application of classical concepts precludes the simultaneous use of other classical concepts
which in a different connection are equally necessary for the elucidation of the phenomena”
[13]. While Heisenberg put forward a more operational idea “that canonically conjugate

quantities can be determined simultaneously only with a characteristic indeterminacy”

[1]. These two statements about the complementarity of two incompatible observables, X

and Y , are illustrated in Figure 1. While Bohr’s interpretation implies that the precise de-
termination of one observable precludes the other, the Heisenberg’s interpretation reflects
the fact that the two observables cannot be determined simultaneously. Note that both

Heisenberg’s uncertainty principle and Bohr’s interpretation concern about the impaired
sections in measurement. In view of the overlapped section, we are encouraged to think

4
of a slightly different interpretation for the complementarity principle,

Observation 1 The generalized uncertainty principle: In quantum theory, if the

quantities are incompatible, they are mutually dependent.

The generalized uncertainty principle indicates that measuring one observable may pro-

vide you some information about its incompatible partners. To ascertain its physical
consequences, we need to quantify the generalized uncertainty principle. For this pur-
pose, we first define the statistical dependence of physical observables.

Given a random variable X, the moment generating function takes the following form
∞
X sn
hesX i = hX n i , s∈C. (4)
n=0
n!

Here hXi means the expectation value of a variable X and the parameter s is a complex
number. The logarithm (with base e if not specified) of equation (4) then generates the

cumulants [14], that is

s2 2 s3 3
 
sX
K(sX) ≡ log(he i) = log 1 + shXi + hX i + hX i + · · ·
2! 3!
∞
X sm
= κm (X) , (5)
m=1
m!

where the sum runs over a power series of s whose coefficients κm (X) are collections of
different orders of moments. κm (X) is called the cumulant of order m which exists if the
mth and lower orders of moments of X exist [14]. The first few orders of cumulants are

Mean : κ1 = hXi , (6)

Variance : κ2 = hX 2 i − hXi2 , (7)

Skewness : κ3 = hX 3 i − 3hX 2 ihXi + 2hXi3 , (8)

Kurtosis : κ4 = hX 4 i − 4hX 3 ihXi − 3hX 2 i2 + 12hX 2 ihXi2 − 6hXi4 . (9)

5
Here κ1 is the expectation value of an observable with given distribution; the variance κ2
measures the spread of the distribution; the skewness κ3 reflects the distribution asym-

metry; the kurtosis κ4 measures the tailedness.

For two random variables X and Y , the cumulants generating function is
∞
X sm tn
K(sX + tY ) ≡ log(hesX+tY i) = κmn , (10)
m+n=1
m!n!

where κmn are named cross cumulants [15]. While κm0 and κ0n have similar expressions

as equations (6)-(9), the first two terms of cross cumulants κmn read
1
κ11 = h{X, Y }i − hXihY i , (11)
2
1
κ12 = h{X, Y, Y }i − (hXihY 2 i + h{X, Y }ihY i) + 2hXihY i2 . (12)
3
Here {X, Y, Y } ≡ XY Y + Y XY + Y Y X is defined as the 3rd order anticommutator and
subscripts in κmn indicate that there are m Xs and n Ys in the expansion of moments.
The cross cumulants for multiple variables can be similarly defined

~
K(s1 X1 + s2 X2 + · · · + sN XN ) ≡ log(he~s·X i) . (13)

According to the Corollary of Theorem I in Ref. [16] we further have:

Observation 2 The cross cumulant is nonzero, if and only if its variables are statistically

connected.

Observation 2 implies that, the cumulants are capable of quantifying the statistical depen-
dence of physical observables. Considering the simplest non-trivial case of two variables
X and Y , there exists [15]

Corollary 1 The cumulants linearize addition of independent random variables

X, Y independent =⇒ κn (sX + tY ) − [sn κn (X) + tn κn (Y )] = 0 , ∀n ≥ 1 , (14)

where we have used the fact κn (sX) = sn κn (X).

6
The contrapositive of Corollary 1 is: if there is any n that κn (sX + tY ) − [κn (sX) +
κn (tY )] 6= 0, then X and Y are nth order dependent. That is, the difference between the

cumulant of the sums and the sum of the cumulants signifies the statistical dependence
between observables.

3 The generalized uncertainty relation

Observation 2 indicates that the dependence in the generalized uncertainty principle

may be demonstrated by the cross cumulants which are generated via the expansion of

exponential operator sums. Consider the Cauchy-Schwarz inequality for the two state
∗X
vectors of es |ψi and etY |ψi, we have the following GUR:

Theorem 1 For arbitrary observables X and Y , there exists a generalized uncertainty

relation

K[(s + s∗ )X] + K[(t + t∗ )Y ] ≥ K(Zst ) + K ∗ (Zst ) , s, t ∈ C . (15)

Here K(·) signifies the generating function of cumulants defined in equation (5); * means
the complex conjugation; Zst = log(esX etY ) = Z1 + Z11 + · · · is defined as

1
Z1 = sX + tY , Z11 = [sX, tY ] , · · · , (16)
2

in light of the well-known Baker-Campbell-Hausdorff (BCH) formula.

The proof of Theorem 1 is presented in the Appendix A. The primary merit of equation
(15) is that the quantum and classical (say communitive hereafter) dependence between
observables can be distinguished via high order powers of s, t in Zst , i.e., the commutators

in the BCH formula. For commutative observables, equation (15) predicts

K[(s + s∗ )X] + K[(t + t∗ )Y ] ≥ K(sX + tY ) + c.c. , (17)

7
where “c.c.” means the complex conjugation of the previous term. The GUR of equation
(15) then may be expressed in a more distinct form

{K[(s + s∗ )X] − K(sX) − K ∗ (sX)} + {K[(t + t∗ )Y ] − K(tY ) − K ∗ (tY )}

≥ [K(Zst ) − K(sX) − K(tY )] + c.c. . (18)

The right hand side of inequality (18) is composed of (anti)commutators of different orders
and reveals the dependence between observables according to Corollary 1. We may say
that, the sum of the statistical properties of each observable on the left hand side of

inequality (18) is lower bounded by their statistical dependence on the right hand side.
The dependence would be zero for observables without correlations. We shall show what
new the GUR may tell by expanding it in s and t in the infinitesimal limit.

3.1 The first order: Trivial identity for mean values

The first order cumulant is the mean value. Comparing the coefficients of s, t, and

their complex conjugates on both sides of equation (15) gives

(s + s∗ )hXi + (t + t∗ )hY i = hsX + tY i + hsX + tY i∗ . (19)

The equality (19) is established on: 1. The expectation values of Hermitian operators
are real, hXi = hXi∗ ; 2. The superposition principle holds for Hermitian operators,
hX + Y i = hXi + hY i. The equality means that there is no contribution from the mean

value to the GUR.

3.2 The second order: Variance uncertainty relations

Expanding equation (18) to the second order and neglecting the high order contribu-
tions of O(sm tn ) for n + m ≥ 3, we have

8
Corollary 2 For two observables X and Y , there exists the following uncertainty relation
for cumulant κ2

|s|2 κ2 (X) + |t|2 κ2 (Y ) ≥ [κ11 (sX, tY ) + hZ11 i] + c.c. , (20)

st
where s, t ∈ C and |s|, |t|  1. Note, hZ11 i = h[X, Y ]i may not be purely imaginary for
2
complex s and t.

The proof Corollary 2 is presented in Appendix B. The right hand side of equation (20) is
composed of the expectation values of a commutator and an anticommutator (see equation

(11)) and reflects the linear dependence between X and Y . To exemplify this, we write
the second order cumulants in terms of variance
q
|s|2 ∆X 2 + |t|2 ∆Y 2 ≥ |st| |h[X, Y ]i|2 + |h{X, Y }i − 2hXihY i|2

= 2|st||h(X − hXi)(Y − hY i)i| . (21)

q q
∆Y ∆X
Let |s| = ε ∆X
and |t| = ε ∆Y
, relation (21) implies the uncertainty relation (2),

and gives out a constraint on Pearson correlation coefficient ρX,Y which is a measure of
correlation (linear dependence) between two variables X and Y

|h(X − hXi)(Y − hY i)i|

ρX,Y = ≤1. (22)
∆X∆Y

From equation (22) we know that the maximal dependence occurs when (X − hXi)|ψi ∝

(Y − hY i)|ψi, viz. ρX,Y = 1.

We may regard any nontrivial statistical constraint on two or more correlated physical
observables as the uncertainty relation, i.e., equation (20) in Corollary 2 is an uncertainty

relation for the statistical quantity of variance (the 2nd order cumulant). The GUR
exhibits the statistical constraint between observables in full order dependence. In this
sense, the GUR may be regarded as a superset of inequivalent uncertainty relations,

9
which is applicable to both classical and quantum quantities. According to (20), the
uncertainty relation (UR) for classical commutative observables and quantum UR for

non-commutative observables then turn to be (up to the second order)

Classical UR : |s|2 κ2 (X) + |t|2 κ2 (Y ) ≥ κ11 (sX, tY ) + c.c. , (23)

Quantum UR : |s|2 κ2 (X) + |t|2 κ2 (Y ) ≥ [κ11 (sX, tY ) + hZ11 i] + c.c. . (24)

When the cross cumulant κ11 = 0 the classical theory predicts no constraint on variances,
while in quantum mechanics there remains a constraint induced by Z11 (here hZ11 i is not
purely imaginary). In the case of κ11 = 0 and hZ11 i = 0, the Heisenberg uncertainty

relation would become trivial, whereas the GUR remains meaningful since hereby X
and Y are only linearly uncorrelated but may be nonlinearly dependent, i.e., nonlinear
dependence may appear in high order expansions.

3.3 The third order: Skewness uncertainty relation

While correlation is broadly employed to indicate the statistical linear dependence of

random variables, GUR enables us to explore the higher order nonlinear dependence, a
peculiar character of GUR. Expand equation (15) or (18) to the third order, i.e., keeping
those sm tn terms with m + n ≤ 3 while neglecting the higher order ones in the small s

and t limits, we have the following Corollary:

Corollary 3 For two observables X and Y , following skewness uncertainty relation holds:

s + s∗ t + t∗
   
2 2
|s| κ2 (X) + κ3 (X) + |t| κ2 (Y ) + κ3 (Y )
2 2

κ12 (sX, tY ) + κ21 (sX, tY )
≥ κ11 (sX, tY ) + +
2

h{Z1 , Z11 }i − 2hZ1 ihZ11 i
hZ11 + Z12 + Z21 i + + c.c. . (25)
2!

10
Figure 2. The skewness of different distributions. Given a probabiity distribu-
tion function p(x) of the random variable X, the skewness κ3 describes the distribution
asymmetry. κ3 is zero when distribution is symmetric.

Here s, t ∈ C, |s|, |t|  1; Z1 = sX + tY , Z11 = 21 [sX, tY ], Z21 = 1

12
[sX, [sX, tY ]], and
1
Z12 = 12
[tY, [tY, sX]].

The procedure on how to truncate power series to the desired order is demonstrated in

the Appendix C. The third order cumulant κ3 names “skewness”, which characterizes the
distribution asymmetry, as shown in Figure 2.
In a special case of s = t = ε being real and [X, Y ] = 0, equation (25) gives the

following variance-skewness relation

κ2 (X) + κ2 (Y ) + ε [κ3 (X) + κ3 (Y )]

≥h{X, Y }i − 2hXihY i+
 
1 2 2
ε h{X, Y, Y }i − (hXihY i + h{X, Y }ihY i) + 2hXihY i + (X ↔ Y ) , (26)
3

where X ↔ Y means the exchange of the observables in previous terms. The right hand
side of equation (26) is zero for independent observables X and Y .

11
4 The full order dependence and examples

In view of the second and third order expansions of GUR, it is clear that there are

two types of terms on the right hand side of (18). One is established on the commutating
operators that signifies the classical dependence between physical observables, the other is
established on the nontrivial commutators from the BCH formula signifying the quantum

dependence. Now we reformulate the GUR in a form that full orders of dependence
appears coherently:

Proposition 1 For two physical observables, there exists a compatible uncertainty rela-
tion

∗X ∗ 
2
hesX+s ihetY +t Y i ≥  esX+tY  , (27)

and an incompatible one

∗X ∗
D 1
E2
hesX+s ihetY +t Y i ≥  esX+tY + 2 [sX,tY ]+···  . (28)
 

The former amounts to commutative classical observables, while the latter to noncommu-
1
tative quantum observables. Here esX+tY + 2 [sX,tY ]+··· = esX etY , and the parameters s, t can

be arbitrary complex numbers.

The equations (27) and (28) hold for all s and t, and any observables. As equations (27)
and (28) are two special cases of the GUR, Proposition 1 indicates that the existence of

uncertainty relation does not stem from the noncommutativity of the observables. This
can also be seen from equation (2), where the bound on the right hand side may be
nontrivial even if [X, Y ] = 0. This further rationalizes the “dependence” interpretation

of the GUR.
Example 1 To exhibit the nonclassical dependence, we shall show the relation (27) may
be violated in quantum theory through an example. For observables of X = σx and

12
Figure 3. The quantum violation of the classical uncertainty relation. The
meshed surface represents the quantum prediction for the left hand side of relation (27),
while the lower surface is the right hand side of it. The quantum state is taken to be
|ψ1 i = cos 2θ |+i + eiφ sin 2θ |−i with two observables of σx and σy . The violation happens
in two circled regions around (θ, φ) = {( π2 , π), ( π2 , 3π
2
)}, where the two observables have
statistical dependence that cannot be explained by classical theory.

Y = σy with state |ψ1 i = cos 2θ |+i + eiφ sin 2θ |−i, the left hand side of relation (27) is

(cosh 2 + cos φ sin θ sinh 2)(cosh 2 + sin φ sin θ sinh 2) , (29)

while the right hand side is

1 h√ √ √ i2
2 cosh 2 + sin θ(cos φ + sin φ) sinh 2 . (30)
2

Here, we have assumed s = t = 1 for simplicity. Expressions (29) and (30) are numeri-
cally plotted in Figure 3, where violations of equation (27) evidently exist. According to
the generalized uncertainty principle, the two observables possess nonclassical statistical

13
dependence in the violation region.
Example 2 There also exist the quantum constraints on two incompatible observables

that can be detected by the uncertainty relation (28), while failed by the uncertainty
relation (2). We consider the two observables of angular momentums Lx and Ly where
[Lx , Ly ] = i~Lz 6= 0. For the quantum state |ψ2 i = √1 (|1i + |0i + | − 1i) of L = 1 system,
3

it is easy to find that the right hand of equation (2) is zero, which gives

∆L2x ∆L2y ≥ 0 . (31)

Since the variance is greater than or equal to zero, (31) tends to be trivial, which for
example cannot rule out the possibility of ∆L2x = ∆L2y = 0, the precise measurements of
Lx and Ly simultaneously. Nevertheless, for the same state and observables, the following

equivalent form of equation (28) for real s and t


 2
hesX+sX i hetY +tY i  esX etY 
≥  sX tY  (32)
|hesX i|2 |hetY i|2  he ihe i 

gives a nontrivial constraint. It is evident that for s, t > 0 the right hand side of equation
(32) is always larger than 1, the accessible minimum of the left hand side. That means
the (32) may enforce certain constraint on the simultaneous measurements of Lx and Ly .

Therefore, we may conclude that in the framework of GUR Lx and Ly are linearly uncorre-
lated, but are higher order dependent. This is a novel insight regarding the compatibility
of physical observables.

Example 3 The generalized uncertainty relation can be applied to the detection of en-
tanglement as other uncertainty relations [17]. In qubit system, equation (20) gives (see
Appendix D)

κ2 (σx ) + κ2 (σy ) + κ2 (σz ) ≥ 1 . (33)

14
Here σi are Pauli matrices. For local observables A = σx ⊗ 1 + 1 ⊗ σx , B = σy ⊗ 1 + 1 ⊗ σy ,
and C = σz ⊗ 1 + 1 ⊗ σz of two-qubit system, the separable states predicts [18]

κ2 (A) + κ2 (B) + κ2 (C) ≥ 2 , (34)

where the lower bound 2 is just the sum of the lower bound of equation (33) for both

sides. While for the singlet state |ψ3 i = √1 (| + −i − | − +i), we have

2

κ2 (A) + κ2 (B) + κ2 (C) = 0 . (35)

This violates equation (34) which indicates the non-separability of the state.
Example 4 The third order cumulant κ3 can be used to exhibit new type of nonlocality
[19]. We consider the following operator for two-qubit system

S = X ⊗ Y − X ⊗ Y 0 + X0 ⊗ Y + X0 ⊗ Y 0 . (36)

Here we may choose X = σz , X 0 = σx , Y = sin θσx + cos θσz , and Y 0 = cos θσx − sin θσz .

In classical correlated systems where |hSi| ≤ 2, there exists the following bound [19]

|κ3 (S)| =  (S − hSi)3  ≤ 8 .


(37)

√
√1 (| 64 6
The maximal value of κ3 (S) in the singlet state |ψ3 i = 2
+ −i − | − +i) is 9
,
which violates the equation (37). This nonclassical phenomenon is called the “skewness
1−η
nonlocality” [19]. For the Werner state ρw = ⊗ 1 + η|ψ3 ihψ3 | with − 31 ≤ η ≤ 1,
4
1
1 1
it can be checked that the skewness nonlocality exists if η > √ 3
− ∼ 0.46. While
2 3
the Bell nonlocality for projective measurements exists if and only if η > KG1(3) ∼ 0.66,
where KG (3) is the Grothendieck’s constant of order three [20]. Clearly, the skewness

nonlocality is fundamentally different from the Bell nonlocality, and it arises purely from
the higher order nonlinear dependence between the local observables in κ3 (S).

15
5 Conclusion

We proposed a generalized uncertainty principle based on an alternative interpretation

of Bohr’s concept of complementarity. That is, the incompatibility between observables

is interpreted as the statistical dependence between them. To elucidate the generalized
principle, a GUR was derived which exhibits full-order statistical dependence between the

physical observables. We found the dependence of observables can be well characterized

by the statistical quantity of cumulant, by which the classical dependence turns out to be
distinguishable from the quantum one. The second (first nontrivial) order approximation

of GUR expansion yields the linear dependence and gives out the Heisenberg type uncer-
tainty relation, while the third order skewness uncertainty relation predicts a constraint
on the distribution asymmetries in different measurements. Concrete examples are also

given to demonstrate the higher order dependence between observables and its possible
applications in quantum information theory.

Acknowledgements

This work was supported in part by the National Natural Science Foundation of China
under the Grants 11975236 and 11635009; and by the University of Chinese Academy of

Sciences.

References

[1] W. Heisenberg, Über den anschaulichen Inhalt der quantentheoretischen Kinematik

und Mechanik. Z. Phys. 43, 172 (1927); in Quantum theory and Measurement, edited
by J. A. Wheeler and W. H. Zurek, (Princeton University press, Princeton, NJ, 1983),
pp. 62-84.

16
 [2] H. P. Robertson, The uncertainty principle, Phys. Rev. 34, 163-164 (1929).

[3] E. Schrödinger, About Heisenberg uncertainty relation, Sitzungsber. Preuss. Akad.

Wiss. Berlin (Math. Phys.) 19, 296-303 (1930) (see also, arXiv: quant-ph/9903100).

[4] L. Maccone and P. K. Arun, Stronger uncertainty relations for all incompatible ob-

servables, Phys. Rev. Lett. 113, 260401 (2014).

[5] Jun-Li Li and Cong-Feng Qiao, Reformulating the quantum uncertainty relation, Sci.
Rep. 5, 12708 (2015).

[6] D. Deutsch, Uncertainty in quantum measurements, Phys. Rev. Lett. 50, 631-633
(1983).

[7] H. Maassen and J. B. M. Uffink, Generalized entropic uncertainty relations, Phys.

Rev. Lett. 60, 1103-1106 (1988).

[8] M. Berta, M. Christandl, R. Colbeck, J. M. Renes, and R. Renner, The uncertainty

principle in the presence of quantum memory, Nat. Phys. 6, 659-662 (2010).

[9] Dong Wang, Fei Ming, Ming-Liang Hu, and Liu Ye, Quantum-memory-assisted en-
tropic uncertainty relations, Ann. Phys. (Berlin) 531, 1900124 (2019).

[10] Fei Ming, Dong Wang, Xiao-Gang Fan, Wei-Nan Shi, Liu Ye, and Jing-Ling Chen,

Improved tripartite uncertainty relation with quantum memory, Phys. Rev. A 102,
012206 (2020).

[11] Jun-Li Li and Cong-Feng Qiao, Equivalence theorem of uncertainty relations, J. Phys.
A: Math. Theor. 50, 03LT01 (2017).

17
[12] Jun-Li Li and Cong-Feng Qiao, The optimal uncertainty relation, Ann. Phys. (Berlin)
531, 1900143 (2019).

[13] N. Bohr, Atomic theory and the description of nature, (Cambridge University Press,

1934), pp.10.

[14] A. Stuart and J. Keith Ord, Kendall’s Advanced Theory of Statistics, Vol 1: Distri-
bution Theory, 6th edition (Wiley, 2010).

[15] J. Novak and M. LaCroix, Three lectures on free probability, arXiv:1205.2097.

[16] R. Kubo, Generalized cumulant expansion method, J. Phys. Soc. Jpn. 17, 1100-1120
(1962).

[17] H. F. Hofmann and S. Takeuchi, Violation of local uncertainty relations as a signature

of entanglement, Phys. Rev. A 68, 032103 (2003).

[18] O. Gühne, Characterizing entanglement via uncertainty relations, Phys. Rev. Lett.
92, 117903 (2004).

[19] Jun-Li Li and Cong-Feng Qiao, The bedrock of quantum nonlocality, arXiv:
2008.06393.

[20] A. Acı́n, N. Gisin, and B. Toner, Grothendieck’s constant and local models for noisy

entangled quantum states, Phys. Rev. A 73, 062105 (2006).

18
 Appendix
A Proof of Theorem 1
∗X
For two physical obervables X and Y , we consider two state vectors es |ψi and
etY |ψi, where the Cauchy-Schwarz inequality tells

∗X ∗
hψ|esX es |ψihψ|et Y etY |ψi ≥ hψ|esX etY |ψihψ|esX etY |ψi∗ . (S1)

Both the left and right hand sides of the inequality (S1) are positive definite. The loga-

rithm function is monotone for positive real numbers, and we have

∗ ∗
log he(s+s )X i + log he(t+t )Y i ≥ log heZst i + log heZst i∗ .
   


(S2)

Here eZst = esX etY and Zst = log(esX etY ) = Z1 + Z11 + · · · . Using the cumulants
∗ ∗
generating functions K[(s + s∗ )X] = log he(s+s )X i , K[(t + t∗ )Y ] = log he(t+t )Y i , and
   

K(Zst ) = log heZst i , we arrive the Theorem 1.

Now we give a detailed expansion for terms in the generalized uncertainty relation
(GUR)

K[(s + s∗ )X] + K[(t + t∗ )Y ] ≥ K(Zst ) + c.c. , s, t ∈ C , (S3)

where “c.c.” means the complex conjugation of the previous term, i.e., K ∗ (Zst ). For

19
Zst = log(esX etY ), the BCH formula gives

1 1
Zst = sX + tY + [sX, tY ] + ([sX, [sX, tY ]] + [tY, [tY, sX]])
2 12
1
− [tY, [sX, [sX, tY ]]] −
24
1
([[[[sX, tY ], tY ], tY ], tY ] + [[[[tY, sX], sX], sX], sX]) +
720
1
([[[[sX, tY ], tY ], tY ], sX] + [[[[tY, sX], sX], sX], tY ]) +
360
1
([[[[tY, sX], tY ], sX], tY ] + [[[[sX, tY ], sX], tY ], sX]) + · · ·
120
= Z1 + Z11 + Z21 + Z12 + Z22 + Z14 + Z41 +
(1) (1) (2) (2)
Z23 + Z32 + Z23 + Z32 + · · · . (S4)

The terms in the expansion of Zst represent

1
Z1 = sX + tY , Z11 = [sX, tY ] , (S5)
2
1 1
Z21 = [sX, [sX, tY ]] , Z12 = [tY, [tY, sX]] , (S6)
12 12
1
Z22 = − [tY, [sX, [sX, tY ]]] , (S7)
24
1 1
Z14 = − [[[[sX, tY ], tY ], tY ], tY ] , Z41 = [[[[tY, sX], sX], sX], sX] , (S8)
720 720
(1) 1 (1) 1
Z23 = [[[[sX, tY ], tY ], tY ], sX] , Z32 = [[[[tY, sX], sX], sX], tY ] , (S9)
360 360
(2) 1 (2) 1
Z23 = [[[[tY, sX], tY ], sX], tY ] , Z32 = [[[[sX, tY ], sX], tY ], sX] . (S10)
120 120

The left hand side of equation (S3) goes as

(s + s∗ )2 (s + s∗ )3
K[(s + s∗ )X] = (s + s∗ )κ1 (X) + κ2 (X) + κ3 (X) + · · · , (S11)
2! 3!
(t + t∗ )2 (t + t∗ )3
K[(t + t∗ )Y ] = (t + t∗ )κ1 (Y ) + κ2 (Y ) + κ3 (Y ) + · · · . (S12)
2! 3!

20
While on the right hand side, we have
*∞ +!
X Zn 
1 2 1 3

st
K(Zst ) = log = log 1 + hZst i + hZst i + hZst i + · · ·
n=0
n! 2! 3!
1 2 1 3 1 4 1 5
= hZst i + hZst i + hZst i + hZst i + hZst i + ···
2! 3! 4! 5!
 2
1 1 2 1 3 1 4
− hZst i + hZst i + hZst i + hZst i + · · ·
2 2! 3! 4!
 3
1 1 2 1 3 1 4
+ hZst i + hZst i + hZst i + hZst i + · · ·
3 2! 3! 4!
 4
1 1 2 1 3 1 4
− hZst i + hZst i + hZst i + hZst i + · · · + ··· . (S13)
4 2! 3! 4!
Here Zst is defined in equation (S4). For the second order, the coefficients of the terms
s2 , t2 , and st on right hand side can be read from the following
1 2 1 h[sX, tY ]i h(sX + tY )2 i − hsX + tY i2
K (2) (Zst ) = hZ11 i + hZ1 i − hZ1 i2 = +
2! 2 2 2
sth[X, Y ]i κ2 (sX) + κ2 (tY ) + 2κ11 (sX, tY )
= + . (S14)
2 2
For the third order, the coefficients of s3 , t3 , s2 t, and st2 on the right hand side are
hZ1 Z11 + Z11 Z1 i hZ13 i hZ1 i hZ12 i + 2hZ11 i


(3) hZ1 i3
K (Zst ) = hZ12 + Z21 i + + − +
2! 3! 2 3!
h[sX, [sX, tY ]]i + h[tY, [tY, sX]]i h{sX + tY, [sX, tY ]}i h(sX + tY )3 i
= + +
12 4 3!
hsX + tY i(h(sX + tY )2 i + h[sX, tY ]i) hsX + tY i3
− + , (S15)
2 3
which may be simplified into
h[sX, [sX, tY ]]i + h[tY, [tY, sX]]i h{sX + tY, [sX, tY ]}i
K (3) (Zst ) = +
12 4
hsX + tY ih[sX, tY ]i κ3 (sX + tY )
− +
2 3!
h[sX, [sX, tY ]]i + h[tY, [tY, sX]]i h{sX + tY, [sX, tY ]}i − 2hsX + tY ih[sX, tY ]i
= +
12 4
κ3 (sX + tY )
+ . (S16)
3!
Here κ3 (sX + tY ) = κ3 (sX) + 3κ12 (sX, tY ) + 3κ21 (sX, tY ) + κ3 (tY ).

21
B Proof of Corollary 2

The equation (20) in Corollary 2 is quite straightforward from the expansions of equa-

tions (S11), (S12), and (S14). Because for |s| ∼ |t|  1, s, t ∈ C, we have
 
2 2 hXY + Y Xi − 2hXihY i h[X, Y ]i
|s| κ2 (X) + |t| κ2 (Y ) ≥ st + st + c.c.
2 2
=Re[st](hXY + Y Xi − 2hXihY i) + Im[st]ih[X, Y ]i , (S17)

where “c.c.” means the complex conjugate of the previous term in the bracket; “Re”
and “Im” stand for the real and image parts of the parameters. The Cauchy-Schwarz

inequality further implies

Re[st](hXY + Y Xi − 2hXihY i) + Im[st]ih[X, Y ]i

p p
≤ Re(st)2 + Im(st)2 |hXY + Y Xi − 2hXihY i|2 + |h[X, Y ]i|2
p
= |st| |hXY + Y Xi − 2hXihY i|2 + |h[X, Y ]i|2 , (S18)

and the equality is satisfied for appropriately chose real and image parts of st. In the case

of the equality being satisfied, the right hand side of equation (S18) becomes the right
hand side of equation (S17), and equation (21) in the main text is arrived.

C The truncated cumulant expansions

The power series of K(sX) can be written as

s2 2 s3 3
 
sX
K(sX) = log(he i) = log 1 + hsXi + hX i + hX i + · · ·
2! 3!
2 3 4
f (s) f (s) f (s)
=f (s) − + − + ··· . (S19)
2 3 4
s2 s3
Here f (s) = hsXi + 2!
hX 2 i + 3!
hX 3 i + · · · = hesX i − 1 with f (0) = 0. The convergence
condition for f (s) around f (s) = 0 is
 
|f (s) − 0| = hψ|esX |ψi − 1 < 1 , ∀ |ψi. (S20)

22
Consider the special case of s ∈ R, we have the convergence region for s around s = 0

log 2 log 2
0 < hesX i < 2 =⇒ − <s< , (S21)
σmax σmax

where σmax is the largest singular value of Hermitian matrix X. For continuous observable
X whose probability distribution has the typical width of σ around x0 , we have
p p
s2 σ 2
+sx − 2σ 2 log 2 + x20 − x0 2σ 2 log 2 + x20 − x0
e 2 0
< 2 =⇒ < s < . (S22)
σ2 σ2

Here we have assumed that the moments hX n i exist for all orders n. Approximately, there

exist
√ √
− 2 log 2 2 log 2
x0  σ : <s< , (S23)
σ σ
− log 2 log 2
σ  x0 : <s< . (S24)
|x0 | |x0 |

Now we study some of the trivial implications of equation (26) for infinitesimal real

values of s ∼ t → ε. For independent observables, the right hand side of equation (26) is
zero, and by expressing the cumulants in terms of moments we have

κ2 (X) + κ2 (Y ) + ε [κ3 (X) + κ3 (Y )]

=h(X − hXi)2 i + h(Y − hY i)2 i + ε h(X − hXi)3 i + h(Y − hY i)3 i ≥ 0 .

 
(S25)

Or equivalently,

(X − hXi)2 [1 + ε(X − hXi)] + (Y − hY i)2 [1 + ε(Y − hY i)] ≥ 0 .


(S26)

To ensure the convergence of both K(2εX) and K(2εY ), the real parameter ε shall be

2|ε|σmax < log 2 . (S27)

Here σmax is the largest singular value for X and Y . The spectrums of (X − hXi) and
(Y − hY i) lie in [−2σmax , 2σmax ], so the minimal eigenvalue λmin of 1 + ε(X − hXi) and

23
1 + ε(Y − hY i) is

λmin > 1 − 2|ε|σmax > 1 − log 2 > 0 . (S28)

The equation (S25) is trivially right for independent variables, i.e.,

κ2 (X) + κ2 (Y ) + ε [κ3 (X) + κ3 (Y )] ≥ 0 . (S29)

D Derivations of the Examples

Example 3 In qubit system, equation (20) predicts that the sum of the second order

cumulants of the observables pairs (σx , σy ), (σy , σz ), and (σz , σx ) are

|s1 |2 κ2 (σx ) + |t1 |2 κ2 (σy ) + |s2 |2 κ2 (σy ) + |t2 |2 κ2 (σz ) + |s3 |2 κ2 (σz ) + |t3 |2 κ2 (σx )
 1 1
≥ 2 |s1 t1 |(|hσx ihσy i|2 + |hσz i|2 ) 2 + |s2 t2 |(|hσy ihσz i|2 + |hσx i|2 ) 2
1

+|s3 t3 |(|hσz ihσx i|2 + |hσy i|2 ) 2 , (S30)

where we have used the anti-commutativity of the Pauli operators and appropriately
choosed phases of si and ti as that in Appendix B. If we set |s1 | = |t1 | = |s2 | = |t2 | =

|s3 | = |t3 |, then

κ2 (σx ) + κ2 (σy ) + κ2 (σz ) ≥ ζ1 + ζ2 + ζ3 ≥ 1 . (S31)

1 1
Here ζ1 = (|hσx ihσy i|2 + |hσz i|2 ) 2 , ζ2 = (|hσy ihσz i|2 + |hσx i|2 ) 2 , and ζ3 = (|hσz ihσx i|2 +
1
|hσy i|2 ) 2
Generally, we may set |s1 t1 | = ε21 , |s2 t2 | = ε22 , |s3 t3 | = ε23 where εi are real parameters.

The right hand side of equation (S30) can be made as large as

q
2 (ε41 + ε42 + ε43 )(ζ12 + ζ22 + ζ32 ) , (S32)

24
 ε2 ζi
at the condition of p 4 i 4 = p , i = 1, 2, 3. Equation (S30) now can
ε1 + ε2 + ε43 ζ12 + ζ22 + ζ32
be reexpressed as
   
|s1 | |t1 | |s2 | |t2 |
ζ1 |κ2 (σx ) + |κ2 (σy ) + ζ2 |κ2 (σy ) + |κ2 (σz )
|t1 | |s1 | |t2 | |s2 |
 
|s3 | |t3 |
q
+ζ3 |κ2 (σz ) + |κ2 (σx ) ≥ 2 ζ12 + ζ22 + ζ32 . (S33)
|t3 | |s3 |

This gives a variance uncertainty relation for qubit states.

Example 4 The quantum mechanical prediction for S = X ⊗Y −X ⊗Y 0 +X 0 ⊗Y +X 0 ⊗Y 0

is

hSi = E(X, Y ) − E(X, Y 0 ) + E(X 0 , Y ) + E(X 0 , Y 0 ) , (S34)

where the correlation is evaluated as E(X, Y ) = hX ⊗ Y i, and X = σz , X 0 = σx ,

Y = sin θσx + cos θσz , and Y 0 = cos θσx − sin θσz . In the local hidden variable theories

(LHVTs), there is the following inequality related to S

S = A(λ, X)[B(λ, Y ) − B(λ, Y 0 )] + A(λ, X 0 )[B(λ, Y ) + B(λ, Y 0 )] ∈ [−2, 2] . (S35)

Z Z
Here hSi ≡ ξλ S dλ and the correlation becomes E(X, Y ) = ξλ A(λ, X)B(λ, Y ) dλ

with ξλ being an unknown distributions of hidden variable(s). For random variable −2 ≤

S ≤ 2, there is
Z  Z
3
 
|κ3 (S)| =  (S − hSi) ξλ dλ ≤ |S − hSi|3 ξλ dλ ≤ 8 , (S36)

where we have used the fact that, for −2 ≤ S ≤ 2, the 3th central moment is less than 8
[S1].

The two-qubit Werner state is defined as

1−η
ρw ≡ 1 ⊗ 1 + η|ψ3 ihψ3 | . (S37)
4

25
Here |ψ3 i = √1 (| + −i − | − +i) and η ∈ [−1/3, 1]. The quantum mechanical result of
2

κ3 (S) for ρw is

κ3 (S) = −8η(cos θ + sin θ)[−1 − 3η + 2η 2 (cos θ + sin θ)2 ] . (S38)

It can be checked that for singlet state of η = 1, κ3 (S) may reach a maximal value of
√ 1 1
64 6
9
∼ 17.42 > 8. The violation of equation (S36) remains for η > √3
− ∼ 0.46.
2 3

References

[S1] M. Egozcue, L. F. Garcı́a, Wing-Keung Wong, and R. Zitikis, The smallest upper

bound for the pth absolute central moment of a class of random variables, Math.
Scientist 37, 125-131 (2012).

26