Theoretical formulation of finite-dimensional discrete phase spaces: II. On the uncertainty principle for Schwinger unitary operators

Definition (Schwinger). Let $\{{\bf U},{\bf V}\}$ be a pair of unitary operators defined in a $N$-dimensional state vector space, and $\{|u_{\alpha}\rangle,|v_{\beta}\rangle\}$ denote their respective orthonormal eigenvectors related by the inner product $\langle u_{\alpha}|v_{\beta}\rangle=\frac{1}{\sqrt{N}}\omega^{\alpha\beta}$ with $\omega\mathrel{\mathop{:}}=\exp\left(\frac{2\pi\mathrm{i}}{N}\right)$. The general properties

together with the fundamental relations

constitute an important set of mathematical rules that are related to the generalized Clifford algebra [17]. Here, the discrete labels $\{\alpha,\beta,\eta,\xi\}$ obey the arithmetic modulo $N$ and $\langle u_{\alpha}|v_{\beta}\rangle$ represents a symmetrical finite Fourier kernel. A compilation of results and properties associated with ${\bf U}$ and ${\bf V}$ can be found in Ref. [1].

In the following, let $\rho$ describe a set of physical systems labeled by a finite space of states, whereas $\mathscr{V}_{\mbox{\tiny${\rm U}$}}\mathrel{\mathop{:}}=1-|\langle{\bf U}\rangle|^{2}$ and $\mathscr{V}_{\mbox{\tiny${\rm V}$}}\mathrel{\mathop{:}}=1-|\langle{\bf V}\rangle|^{2}$ denote the variances related to the respective unitary operators ${\bf U}$ and ${\bf V}$ — in this case, $\langle{\bf U}\rangle$ and $\langle{\bf V}\rangle$ represent the mean values of ${\bf U}$ and ${\bf V}$ defined in a $N$-dimensional state vectors space. Since $\rho$ refers to a normalized density operator, the Cauchy-Schwarz inequality allows us to prove that $|\langle{\bf U}\rangle|^{2}$ and $|\langle{\bf V}\rangle|^{2}$ are restricted to the closed interval $[0,1]$; consequently, both the variances are trivially bounded by $0\leq\mathscr{V}_{\mbox{\tiny${\rm U}$}(\mbox{\tiny${\rm V}$})}\leq 1$. Indeed, the upper and lower bounds are promptly reached when one considers the localized bases $\left\{|u_{\alpha}\rangle\langle u_{\alpha}|\right\}_{0\leq\alpha\leq N-1}$ and $\left\{|v_{\beta}\rangle\langle v_{\beta}|\right\}_{0\leq\beta\leq N-1}$, i.e., for a given $\mbox{\boldmath$\rho$}=|u_{\alpha}\rangle\langle u_{\alpha}|\Rightarrow\mathscr{V}_{\mbox{\tiny${\rm U}$}}=0$ and $\mathscr{V}_{\mbox{\tiny${\rm V}$}}=1$; otherwise, if $\mbox{\boldmath$\rho$}=|v_{\beta}\rangle\langle v_{\beta}|\Rightarrow\mathscr{V}_{\mbox{\tiny${\rm U}$}}=1$ and $\mathscr{V}_{\mbox{\tiny${\rm V}$}}=0$. Furthermore, note that $\mathscr{V}_{\mbox{\tiny${\rm U}$}}$ and $\mathscr{V}_{\mbox{\tiny${\rm V}$}}$ are invariant under phase transformations, namely, ${\bf U}\rightarrow\mathrm{e}^{\mathrm{i}\varphi}{\bf U}$ and ${\bf V}\rightarrow\mathrm{e}^{\mathrm{i}\theta}{\bf V}$ for any $\{\varphi,\theta\}\in\mathbb{R}$.

This inequality is based on the Wiener-Kinchin theorem for signal processing and provides a constraint between the values of $\langle\overline{{\bf V}}^{\xi}\rangle$ (correlation function) and $\langle\overline{{\bf U}}^{\eta}\rangle$ (discrete Fourier transform of the intensity time series). According to Massar and Spindel: ‘This kind of constraint should prove useful in signal processing, as it constrains what kinds of signals are possible, or what kind of wavelet bases one can construct.’ We state this result in the theorem below (proved in the supplementary material from Ref. [14]), for then proceeding with a numerical study of its content and first implications.

{pf*} Theorem (Massar and Spindel). Let $\overline{{\bf U}}$ and $\overline{{\bf V}}$ denote two unitary operators such that $\overline{{\bf U}}\,\overline{{\bf V}}=\mathrm{e}^{\mathrm{i}\Phi}\overline{{\bf V}}\,\overline{{\bf U}}$ and $\overline{{\bf U}}^{\dagger}\overline{{\bf V}}=\mathrm{e}^{-\mathrm{i}\Phi}\overline{{\bf V}}\,\overline{{\bf U}}^{\dagger}$ with $\Phi\in[0,\pi)$. The variances $\mathscr{V}_{\overline{\mbox{\tiny${\rm U}$}}}$ and $\mathscr{V}_{\overline{\mbox{\tiny${\rm V}$}}}$ — here defined for a given quantum state $\rho$ and limited to the closed interval $[0,1]$ — satisfy the inequality $$(1+2A)\mathscr{V}_{\overline{\mbox{\tiny${\rm U}$}}}\mathscr{V}_{\overline{\mbox{\tiny${\rm V}$}}}\geq A^{2}(1-\mathscr{V}_{\overline{\mbox{\tiny${\rm U}$}}}-\mathscr{V}_{\overline{\mbox{\tiny${\rm V}$}}})$$ (1) where $A=\tan\left(\frac{\Phi}{2}\right)$. The saturation is reached for localized bases.

This theorem leads us, in principle, to consider the different possibilities of connections between the Schwinger unitary operators $\{{\bf U},{\bf V}\}$ (defined in the previous subsection) and $\{\overline{\bf U},\overline{\bf V}\}$. A first immediate link yields the relation $\overline{\bf U}={\bf U}$ and $\overline{\bf V}={\bf V}$ for $\Phi=-\frac{2\pi}{N}$, which implies in the apparent violation of Eq. (1) since $\Phi\notin[0,\pi)$. The second connection establishes the alternative relation $\overline{\bf U}={\bf V}$ and $\overline{\bf V}={\bf U}$ with $\Phi=\frac{2\pi}{N}$, this result being responsible for validating the Massar-Spindel inequality. It is important to mention that the apparent problem detected in the first situation can be properly circumvented making $\Phi\rightarrow-\Phi$ and $A\rightarrow-A$ with $\Phi=\frac{2\pi}{N}$ fixed. That is, the modified bound

holds for any $A=\tan\left(\frac{\pi}{N}\right)$ and integer $N\in[2,\infty)$.

For the sake of simplicity and convenience, let us now introduce the shift operator $\Delta\mbox{\boldmath$\mathcal{O}$}\equiv\mbox{\boldmath$\mathcal{O}$}-\langle\mbox{\boldmath$\mathcal{O}$}\rangle$ with $\langle\mbox{\boldmath$\mathcal{O}$}\rangle\neq 0$ for a given arbitrary unitary operator $\mathcal{O}$ and density operator $\rho$, which leads to define¹¹ 1 It is important to note a certain level of arbitrariness in this definition because there is no especification whatsoever of which density operator $\rho$ is used to evaluate the mean value in the denominator of Eq. (2). Henceforth, this arbitrariness will be removed by exploring the fact that, in our computations, the operator $\delta\mbox{\boldmath$\mathcal{O}$}$ will always appear as an argument of a variance function — in such a situation, we will understand that the expectation value $\langle\mbox{\boldmath$\mathcal{O}$}\rangle$ is computed with respect to the same state used in the computation of $\langle\delta\mbox{\boldmath$\mathcal{O}$}\rangle$. its non-unitary counterpart as follows [25]:

Following, it is straighforward to show that both the variances $\mathscr{V}_{\mbox{\tiny$\mathcal{O}$}}$ and $\mathscr{V}_{\delta\mbox{\tiny$\mathcal{O}$}}$ are related through the expressions

which substituted into inequality (1) for $\overline{\bf U}={\bf V}$ and $\overline{\bf V}={\bf U}$ gives

To illustrate Eq. (4) and corroborate the analytic results obtained by Massar and Spindel, let us introduce the parameters $\mathcal{S}_{\delta\mbox{\tiny${\rm U}$}}=\varepsilon^{-1}\mathscr{V}_{\delta\mbox{\tiny${\rm U}$}}$ and $\mathcal{S}_{\delta\mbox{\tiny${\rm V}$}}=\varepsilon^{-1}\mathscr{V}_{\delta\mbox{\tiny${\rm V}$}}$ such that $\mathcal{S}_{\delta\mbox{\tiny${\rm U}$}}\mathcal{S}_{\delta\mbox{\tiny${\rm V}$}}\geq 1$. This particular inequality defines a region in the two-dimensional space limited by the rectangular hyperbola $\mathcal{S}_{\delta\mbox{\tiny${\rm V}$}}=\mathcal{S}_{\delta\mbox{\tiny${\rm U}$}}^{-1}$ that preserves the original equation $\mathscr{V}_{\delta\mbox{\tiny${\rm U}$}}\mathscr{V}_{\delta\mbox{\tiny${\rm V}$}}\geq\varepsilon^{2}$. Figure 1 shows the plots of $\mathcal{S}_{\delta\mbox{\tiny${\rm V}$}}$ versus $\mathcal{S}_{\delta\mbox{\tiny${\rm U}$}}$ for approximately $10^{6}$ states of randomly generated $\rho$, within the visualization window $0<\mathcal{S}_{\delta\mbox{\tiny${\rm U}$}},\mathcal{S}_{\delta\mbox{\tiny${\rm V}$}}\leq 10$, with (a) $N=2$ $(\varepsilon\rightarrow 1)$, (b) $N=3$, (c) $N=4$, and (d) $N=5$. Note that, excepting picture (a), all the subsequent cases exhibit a gap between the distribution of states and the rectangular hyperbola, the size of such a gap being dependent on the value of $N$ (such an evidence motivates the search for a tightest bound that corroborates the numerical calculations). In fact, the dashed lines in (b,c,d) describe hyperbolic curves of the form $\mathcal{S}_{\delta\mbox{\tiny${\rm U}$}}\mathcal{S}_{\delta\mbox{\tiny${\rm V}$}}=x$, where the value of $x$ was chosen as the smallest value of $\mathcal{S}_{\delta\mbox{\tiny${\rm U}$}}\mathcal{S}_{\delta\mbox{\tiny${\rm V}$}}$ amongst the ones computed with the randomly sampled states. Later in this paper, a more rigorous procedure for obtaining such values will be outlined. Finally, let us briefly mention that the dot-dashed lines exhibited in these pictures correspond to the intermediate result $\mathcal{S}_{\delta\mbox{\tiny${\rm U}$}}\mathcal{S}_{\delta\mbox{\tiny${\rm V}$}}\geq 1+\sin\left(\frac{2\pi}{N}\right)$, which will be properly demonstrated in the next section.

Let $\{{\bf A},{\bf B}\}$ denote a pair of Hermitian operators defined in a $N$-dimensional state vectors space which obey the RS uncertainty principle [20]

where $\mathscr{C}({\bf A},{\bf B})\mathrel{\mathop{:}}=\langle\frac{1}{2}\{{\bf A},{\bf B}\}\rangle-\langle{\bf A}\rangle\langle{\bf B}\rangle$ represents the covariance function, and $[{\bf A},{\bf B}]$ $\left(\{{\bf A},{\bf B}\}\right)$ corresponds to the commutator (anticommutator) between ${\bf A}$ and ${\bf B}$.²² 2 It is important to emphasize that, according to Cauchy-Schwarz inequality, the covariance function $\mathscr{C}({\bf A},{\bf B})$ is restricted to the closed interval $\left[-\sqrt{\mathscr{V}_{\mbox{\tiny${\rm A}$}}\mathscr{V}_{\mbox{\tiny${\rm B}$}}},\sqrt{\mathscr{V}_{\mbox{\tiny${\rm A}$}}\mathscr{V}_{\mbox{\tiny${\rm B}$}}}\,\right]$, namely, $\left|\mathscr{C}({\bf A},{\bf B})\right|\leq\sqrt{\mathscr{V}_{\mbox{\tiny${\rm A}$}}\mathscr{V}_{\mbox{\tiny${\rm B}$}}}$. Indeed, for a given operator ${\bf C}\mathrel{\mathop{:}}={\bf A}-\frac{\mathscr{C}({\bf A},{\bf B})}{2\mathscr{V}_{\mbox{\tiny${\rm B}$}}}{\bf B}$ with $\mathscr{V}_{\mbox{\tiny${\rm B}$}}\neq 0$ and $\mathscr{V}_{\mbox{\scriptsize${\rm C}$}}\geq 0$, it turns immediate to obtain the relation $\mathscr{V}_{\mbox{\scriptsize${\rm C}$}}=\mathscr{V}_{\mbox{\tiny${\rm A}$}}-\frac{\mathscr{C}^{2}({\bf A},{\bf B})}{\mathscr{V}_{\mbox{\tiny${\rm B}$}}}$, which demonstrates the previous result for $\mathscr{C}({\bf A},{\bf B})$. Moreover, let us also state a very useful result for both the commutation and anticommutation relations between ${\bf A}$ and ${\bf B}$, that is $|\langle[{\bf A},{\bf B}]\rangle|^{2}+|\langle\{{\bf A},{\bf B}\}\rangle|^{2}=4|\langle{\bf A}{\bf B}\rangle|^{2}$. Next, since ${\bf U}$ and ${\bf V}$ are generally non-Hermitian operators, let us consider the cartesian decomposition of an arbitrary non-Hermitian operator $\mathcal{O}$ into its ‘real’ and ‘imaginary’ parts as follows [28]:

Note that $\mathrm{Re}[\mbox{\boldmath$\mathcal{O}$}]$ and $\mathrm{Im}[\mbox{\boldmath$\mathcal{O}$}]$ represent two Hermitian operators that comply with the RS uncertainty principle and allow to introduce, in particular, the cosine and sine operators through the Schwinger unitary operators.

Definition. Let $\left\{{\bf C}_{\mbox{\tiny${\rm U}$}},{\bf S}_{\mbox{\tiny${\rm U}$}},{\bf C}_{\mbox{\tiny${\rm V}$}},{\bf S}_{\mbox{\tiny${\rm V}$}}\right\}$ denote four Hermitian operators written in terms of simple combinations associated with ${\bf U}$ and ${\bf V}$, i.e., ${\bf C}_{\mbox{\tiny${\rm U}$}}\mathrel{\mathop{:}}=\mathrm{Re}[{\bf U}]$, ${\bf S}_{\mbox{\tiny${\rm U}$}}\mathrel{\mathop{:}}=\mathrm{Im}[{\bf U}]$, ${\bf C}_{\mbox{\tiny${\rm V}$}}\mathrel{\mathop{:}}=\mathrm{Re}[{\bf V}]$, and ${\bf S}_{\mbox{\tiny${\rm V}$}}\mathrel{\mathop{:}}=\mathrm{Im}[{\bf V}]$. The commutation relations involving these operators exhibit a direct connection with certain anticommutation relations:

where the parameter $A$ was previously defined in Eq. (1). These results lead us, in principle, to conclude that partial information on two particular elements of the set is not complete, since the complementary elements are also necessary to fully characterize the commutator $[{\bf U},{\bf V}]$. In this sense, let us now consider the four RS uncertainty principles below:

In addition, the extra result ${\bf C}_{\mbox{\tiny${\rm U}$}(\mbox{\tiny${\rm V}$})}^{2}+{\bf S}_{\mbox{\tiny${\rm U}$}(\mbox{\tiny${\rm V}$})}^{2}={\bf 1}$ resembles a well-known mathematical property associated with the trigonometric functions cosine and sine. For this reason, these Hermitian operators will be henceforth termed ‘cosine’ and ‘sine’ operators, whose respective variances can be shown to obey the mathematical identity

To make complete this definition, it is interesting to observe that certain combinations of $\langle{\bf C}_{\mbox{\tiny${\rm U}$}(\mbox{\tiny${\rm V}$})}\rangle^{2}$ and $\langle{\bf S}_{\mbox{\tiny${\rm U}$}(\mbox{\tiny${\rm V}$})}\rangle^{2}$ also yield the additional result

which proves itself useful in our subsequent calculations.

Next, by means of mathematical remarks, we establish an important set of other useful results for the sine and cosine operators, whose relevance is intrinsically connected with the hierarchy relations involving the uncertainty principles related to ${\bf U}$ and ${\bf V}$ which generalize the Massar-Spindel inequality. It is worth mentioning that some proofs demand a logical sequence of algebraic manipulations to be detailed in the text.

Remark 1. Although the sine and cosine operators are genuinely defined for unitary operators, we shall employ, in due time, the same definition for a general operator ${\bf X}$ as well. In this general case, it can be easily demonstrated that the following properties hold: $\langle{\bf C}_{\mbox{\tiny${\rm X}$}}^{2}\rangle+\langle{\bf S}_{\mbox{\tiny${\rm X}$}}^{2}\rangle=\langle{\textstyle\frac{1}{2}}\{{\bf X},{\bf X}^{\dagger}\}\rangle$, $\langle{\bf C}_{\mbox{\tiny${\rm X}$}}\rangle^{2}+\langle{\bf S}_{\mbox{\tiny${\rm X}$}}\rangle^{2}=|\langle{\bf X}\rangle|^{2}$, and $\mathscr{V}_{\mbox{\scriptsize${\rm C}$}_{\mbox{\tiny${\rm X}$}}}+\mathscr{V}_{\mbox{\scriptsize${\rm S}$}_{\mbox{\tiny${\rm X}$}}}=\mathscr{C}({\bf X},{\bf X}^{\dagger})$. Note that for a normal operator ${\bf N}$ (which satisfies $[{\bf N},{\bf N}^{\dagger}]=0$), the covariance function between ${\bf N}$ and ${\bf N}^{\dagger}$ matches the variance of ${\bf N}$, namely, $\mathscr{V}_{\mbox{\scriptsize${\rm C}$}_{\mbox{\tiny$\mathrm{N}$}}}+\mathscr{V}_{\mbox{\scriptsize${\rm S}$}_{\mbox{\tiny$\mathrm{N}$}}}=\mathscr{V}_{\mbox{\tiny$\mathrm{N}$}}$.

Remark 2. Let us initially consider the sum of all aforementioned RS uncertainty principles for the sine and cosine operators, as well as the connection between the commutation and anticommuation relations of such Hermitian operators. Adequate algebraic manipulations allow, in principle, to obtain an inequality for the product $\mathscr{V}_{\mbox{\tiny${\rm U}$}}\mathscr{V}_{\mbox{\tiny${\rm V}$}}$ with remarkable mathematical features, i.e.,

where

Note that $\mathcal{F}$ and $\mathcal{H}$ show a nontrivial dependence on the unitary operators ${\bf U}$ and ${\bf V}$, which will be properly discussed in the subsequent remarks. Moreover, if compared with Massar-Spindel inequality (1), such a result yields subtle additional corrections that will depend explicitly on the dimension of the state space.

For completeness sake, let us now establish a first numerical evaluation on the functions $\mathcal{F}$ and $\mathcal{H}$. Figure 2 exhibits the plots of $\mathcal{F}$ versus $\mathcal{H}$ for (a) $N=4$ and (b) $N=5$ with the same number of normalized random states used in the previous figure. Since the solid line represents $\mathcal{F}=\mathcal{H}$ in both the situations, this preliminary numerical search demonstrates a greater contribution coming from $\mathcal{H}$ than from $\mathcal{F}$ for most states of low dimension $N$. This fact appeals for a detailed formal investigation on the origins of such contributions, in which each term of $\mathcal{F}$ and $\mathcal{H}$ would be examined separately.

Remark 3. Important mathematical properties of $\mathcal{H}({\bf U},{\bf V})$ are easily attained examining each of its terms separately: for instance, to calculate the modulus squared of $\langle{\bf C}_{\mbox{\tiny${\rm U}$}}{\bf C}_{\mbox{\tiny${\rm V}$}}\rangle$, we initially expand the cosine operators ${\bf C}_{\mbox{\tiny${\rm U}$}}$ and ${\bf C}_{\mbox{\tiny${\rm V}$}}$ in terms of the Schwinger unitary operators ${\bf U}$ and ${\bf V}$; the next step consists in decomposing the resulting terms into their real and imaginary Hermitian parts, whose final expression assumes the form

Once the remaining terms are obtained in the same way, some algebraic simplification gives rise to the following expression for $\mathcal{H}$:

However, this result does not represent a convenient form of $\mathcal{H}$ since the proper summation of $\langle{\bf C}_{\mbox{\tiny$\mathcal{O}$}}\rangle^{2}+\langle{\bf S}_{\mbox{\tiny$\mathcal{O}$}}\rangle^{2}=1-\mathscr{V}_{\mbox{\tiny$\mathcal{O}$}}$ with $\langle{\bf C}_{\mbox{\tiny$\mathcal{O}$}}^{2}\rangle+\langle{\bf S}_{\mbox{\tiny$\mathcal{O}$}}^{2}\rangle=1$ (in this situation, one considers all possible matches of $\mbox{\boldmath$\mathcal{O}$}\equiv{\bf U}{\bf V},{\bf U}{\bf V}^{\dagger},{\bf U}^{\dagger}{\bf V},{\bf U}^{\dagger}{\bf V}^{\dagger}$) permits us to derive a simplified expression for such a function that includes the variances $\mathscr{V}_{\mbox{\tiny${\rm U}$}\mbox{\tiny${\rm V}$}}$ and $\mathscr{V}_{\mbox{\tiny${\rm U}$}\mbox{\tiny${\rm V}$}^{\dagger}}$. Indeed, since $\mathscr{V}_{\mbox{\tiny${\rm U}$}^{\dagger}\mbox{\tiny${\rm V}$}}\equiv\mathscr{V}_{\mbox{\tiny${\rm U}$}\mbox{\tiny${\rm V}$}^{\dagger}}$ and $\mathscr{V}_{\mbox{\tiny${\rm U}$}^{\dagger}\mbox{\tiny${\rm V}$}^{\dagger}}\equiv\mathscr{V}_{\mbox{\tiny${\rm U}$}\mbox{\tiny${\rm V}$}}$, it turns immediate to prove that

which is invariant under the transformations ${\bf U}\rightarrow\mathrm{e}^{\mathrm{i}\varphi}{\bf U}$ and ${\bf V}\rightarrow\mathrm{e}^{\mathrm{i}\theta}{\bf V}$ for $\{\varphi,\theta\}\in\mathbb{R}$.

Remark 4. Note that Eq. (11) represents a lower bound for any element of³³ 3 Let $\left\|{\bf U}\right\|_{\mathrm{HS}}=\left\|{\bf V}\right\|_{\mathrm{HS}}=\sqrt{N}$ and $\left\|{\bf C}_{\mbox{\tiny${\rm U}$}(\mbox{\tiny${\rm V}$})}\right\|_{\mathrm{HS}}=\left\|{\bf S}_{\mbox{\tiny${\rm U}$}(\mbox{\tiny${\rm V}$})}\right\|_{\mathrm{HS}}=\sqrt{\frac{N}{2}}$ characterize the Hilbert-Schmidt norms associated with the Schwinger unitary operators and their respective related trigonometric operators, where $\left\|{\bf X}\right\|_{\mathrm{HS}}\mathrel{\mathop{:}}=\sqrt{\mathrm{Tr}\left[{\bf X}^{\dagger}{\bf X}\right]}$ defines the aforementioned norm [28]. The further mathematical property $$\Bigl\{\left\|{\bf C}_{\mbox{\tiny${\rm U}$}}{\bf C}_{\mbox{\tiny${\rm V}$}}\right\|_{\mathrm{HS}},\left\|{\bf C}_{\mbox{\tiny${\rm U}$}}{\bf S}_{\mbox{\tiny${\rm V}$}}\right\|_{\mathrm{HS}},\left\|{\bf S}_{\mbox{\tiny${\rm U}$}}{\bf C}_{\mbox{\tiny${\rm V}$}}\right\|_{\mathrm{HS}},\left\|{\bf S}_{\mbox{\tiny${\rm U}$}}{\bf S}_{\mbox{\tiny${\rm V}$}}\right\|_{\mathrm{HS}}\Bigr\}\leq\frac{N}{2}$$ represents an effective gain in this stage, since it brings out relevant information on the different products of cosine and sine operators used in the text.

this fact being discussed by Massar and Spindel [14] through different mathematical arguments. In fact, the authors demonstrated that for a given choice of phases of the Schwinger unitary operators, the restrictions $\langle{\bf C}_{\mbox{\tiny${\rm U}$}(\mbox{\tiny${\rm V}$})}\rangle\in\mathbb{R}_{+}\Rightarrow\langle{\bf S}_{\mbox{\tiny${\rm U}$}(\mbox{\tiny${\rm V}$})}\rangle=0$ select a set $\{\mbox{\boldmath$\rho$}\}$ of finite quantum states for which the inequality

can be formally verified and also numerically tested. Despite the correlations between ${\bf U}$ and ${\bf V}$ do not appear in such an expression, the comparison with $\mathcal{H}$ is unavoidable in this case, since correlations represent an important quantum effect that deserve our attention. The saturation is reached in both the situations for localized bases.

Remark 5. Let us now decompose $\mathcal{F}$ into three terms $\mathcal{F}_{1}$, $\mathcal{F}_{2}$ and $\mathcal{F}_{3}$, whose different contributions will be formally calculated with the help of the mathematical procedure sketched in the previous remarks. We initially consider the term

responsible for contributions associated with squared mean values of all anticommutation relations involved in $\mathcal{F}$ through the covariance functions. So, let $\langle\{{\bf C}_{\mbox{\tiny${\rm U}$}},{\bf C}_{\mbox{\tiny${\rm V}$}}\}\rangle$ admit the form

in complete analogy with $\langle{\bf C}_{\mbox{\tiny${\rm U}$}}{\bf C}_{\mbox{\tiny${\rm V}$}}\rangle$. Writing the remaining mean values in an analogous way, subsequent computations allow to prove that $\mathcal{F}_{1}$ is connected with $\mathcal{H}({\bf U},{\bf V})$ through the relation $\mathcal{F}_{1}=\cos^{2}\left(\frac{\Phi}{2}\right)\mathcal{H}$ for $\Phi=\frac{2\pi}{N}$ fixed.

Following, in what concerns the second term

it can be expressed by means of the convenient form

which represents an important formal result in our calculations.

Finally, let us briefly mention that $\mathcal{F}_{3}$ coincides with $(1-\mathscr{V}_{\mbox{\tiny${\rm U}$}})(1-\mathscr{V}_{\mbox{\tiny${\rm V}$}})$ since it is equivalent to the product $\left(\langle{\bf C}_{\mbox{\tiny${\rm U}$}}\rangle^{2}+\langle{\bf S}_{\mbox{\tiny${\rm U}$}}\rangle^{2}\right)\left(\langle{\bf C}_{\mbox{\tiny${\rm V}$}}\rangle^{2}+\langle{\bf S}_{\mbox{\tiny${\rm V}$}}\rangle^{2}\right)$; consequently, the function

can be immediately obtained. Note that the nontrivial dependence of Eq. (10) on the Schwinger unitary operators is completely justified in these remarks.

This set of mathematical remarks establishes a first solid algebraic framework for the unitary operators ${\bf U}$ and ${\bf V}$, whose intrinsic virtues lead us to formulate a theorem which generalizes the Massar-Spindel inequality (1) in order to include the quantum correlation effects between the aforementioned operators. In fact, this theorem consists of an initial compilation of efforts in our future search for the tightest bound, where the correlation function has occupied an important place in the investigative process.

{pf*} Theorem. Let ${\bf U}$ and ${\bf V}$ be two unitary operators defined in a $N$-dimensional state vector space which satisfy the commutation relations $\left[{\bf U},{\bf V}\right]=(1-\omega){\bf U}{\bf V}$ and $\left[{\bf U},{\bf V}^{\dagger}\right]=(1-\omega^{\ast}){\bf U}{\bf V}^{\dagger}$ for $\omega\mathrel{\mathop{:}}=\exp\left(\frac{2\pi\mathrm{i}}{N}\right)$. Moreover, let $\mathscr{V}_{\mbox{\tiny${\rm U}$}}\mathrel{\mathop{:}}=1-|\langle{\bf U}\rangle|^{2}$ and $\mathscr{V}_{\mbox{\tiny${\rm V}$}}\mathrel{\mathop{:}}=1-|\langle{\bf V}\rangle|^{2}$ represent the respective variances such that $0\leq\mathscr{V}_{\mbox{\tiny${\rm U}$}(\mbox{\tiny${\rm V}$})}\leq 1$. The inequality $$\mathscr{V}_{\mbox{\tiny${\rm U}$}}\mathscr{V}_{\mbox{\tiny${\rm V}$}}\geq\mathcal{F}({\bf U},{\bf V})+\sin^{2}\left(\frac{\pi}{N}\right)\mathcal{H}({\bf U},{\bf V})$$ (13) yields a new bound for $\mathscr{V}_{\mbox{\tiny${\rm U}$}}\mathscr{V}_{\mbox{\tiny${\rm V}$}}$, where the quantum correlation effects related to the unitary operators are taken into account. Note that $\mathcal{F}$ and $\mathcal{H}$ were precisely defined and formally studied in the previous mathematical remarks, the saturation $\mathcal{F}=\mathcal{H}=0$ being attained for localized bases.

Next, let us determine some additional results focused on the Hermitian operators $\left\{{\bf C}_{\delta\mbox{\tiny${\rm U}$}},{\bf S}_{\delta\mbox{\tiny${\rm U}$}},{\bf C}_{\delta\mbox{\tiny${\rm V}$}},{\bf S}_{\delta\mbox{\tiny${\rm V}$}}\right\}$ in order to yield a set of specific intermediate inequalities whose hierarchical relations correspond to a solid bridge towards the tightest bound.

We start this subsection stating a first important result for the sine and cosine operators $\left\{{\bf C}_{\delta\mbox{\tiny${\rm U}$}},{\bf S}_{\delta\mbox{\tiny${\rm U}$}},{\bf C}_{\delta\mbox{\tiny${\rm V}$}},{\bf S}_{\delta\mbox{\tiny${\rm V}$}}\right\}$ defined, in turn, in terms of the non-unitary counterparts $\delta{\bf U}$ and $\delta{\bf V}$. This particular result shows how the mean values of their commutation and anticommutation relations are linked with determined correlation functions, namely,

Following, let us complete this set of results with relations that connect all the covariance functions previously defined with the original framework:

These equations enable us to rewrite $\mathcal{F}$ in the compact form

while $(\mathscr{V}_{\mbox{\scriptsize${\rm C}$}_{\delta\mbox{\tiny${\rm U}$}}}+\mathscr{V}_{\mbox{\scriptsize${\rm S}$}_{\delta\mbox{\tiny${\rm U}$}}})(\mathscr{V}_{\mbox{\scriptsize${\rm C}$}_{\delta\mbox{\tiny${\rm V}$}}}+\mathscr{V}_{\mbox{\scriptsize${\rm S}$}_{\delta\mbox{\tiny${\rm V}$}}})=\mathscr{V}_{\delta\mbox{\tiny${\rm U}$}}\mathscr{V}_{\delta\mbox{\tiny${\rm V}$}}$ brings out a completely analogous expression to that established for $\mathscr{V}_{\mbox{\tiny${\rm U}$}}\mathscr{V}_{\mbox{\tiny${\rm V}$}}$; consequently,

represents an alternative form of Eq. (13) since $\mathscr{V}_{\mbox{\tiny${\rm U}$}}\mathscr{V}_{\mbox{\tiny${\rm V}$}}\equiv\mathcal{F}_{3}\mathscr{V}_{\delta\mbox{\tiny${\rm U}$}}\mathscr{V}_{\delta\mbox{\tiny${\rm V}$}}$.

Now, let us rewrite Eq. (15) according to definitions employed for $\mathcal{S}_{\delta\mbox{\tiny${\rm U}$}}$ and $\mathcal{S}_{\delta\mbox{\tiny${\rm V}$}}$ in the previous section,

From this result, it is relatively easy to derive a number of simpler (but looser) inequalities: for instance, given the non-negativity of $\frac{\mathcal{F}}{\mathcal{F}_{3}}$ [cf. Eq. (14)], the first term in the second bracket can be dropped to give

which corresponds, in such a case, to the HKR uncertainty principle; alternatively, $\frac{\mathcal{H}}{\mathcal{F}_{3}}$ can also be dropped (for the same reason, i.e., its non-negativity) in order to yield

Consequently, such inequalities are then combined in order to give a tighter one,

At this point, one is led to ask if there exists a general ordering relation between $\csc^{2}\left(\frac{\pi}{N}\right)\frac{\mathcal{F}}{\mathcal{F}_{3}}$ and $\frac{\mathcal{H}}{\mathcal{F}_{3}}$. To answer this question, let us establish a set of numerical calculations associated with the different values of dimension $N$, where, for each specific case, one has approximately $10^{6}$ randomly generated states. In order to make the presentation of these numerical results more self-contained, Fig. 3 depicts the plots of $\csc^{2}\left(\frac{\pi}{N}\right)\frac{\mathcal{F}}{\mathcal{F}_{3}}$ versus $\frac{\mathcal{H}}{\mathcal{F}_{3}}$ for $N=2,3,4,6$ and $20$. In this case, note that the maximum can indeed arise from either terms, depending on the particular chosen state $\rho$. As a rule of thumb, one has the following: for low dimensional states (e.g., $N=2,3$), the term $\frac{\mathcal{H}}{\mathcal{F}_{3}}$ usually dominates; however, as the dimension $N$ increases ($N\geq 6$), $\csc^{2}\left(\frac{\pi}{N}\right)\frac{\mathcal{F}}{\mathcal{F}_{3}}$ becomes the usually dominant term. Finally, if one considers $N=4$, it is visible that there is not a usually dominant term — in this case, the result of the optimization is strongly dependent on the particular input state.

As it turns out, inequality (17) can be further strengthened by the addition of yet another (less trivial) lower bound of $\csc^{2}\left(\frac{\pi}{N}\right)\frac{\mathcal{F}}{\mathcal{F}_{3}}+\frac{\mathcal{H}}{\mathcal{F}_{3}}$ to the maximization at hand. Just as $\frac{\mathcal{F}}{\mathcal{F}_{3}}$ could be written as in Eq. (14), we can also show that $\frac{\mathcal{H}}{\mathcal{F}_{3}}$ admits the form

With that in mind, some straightforward manipulation yields

for $x\equiv\mathscr{C}({\bf C}_{\delta\mbox{\tiny${\rm U}$}},{\bf C}_{\delta\mbox{\tiny${\rm V}$}})$, where the inequality follows from fact that $\frac{\mathcal{F}}{\mathcal{F}_{3}}\geq x^{2}$ [cf. Eq. (14)]. A mathematical virtue of this lower bound is that it depends only on the covariance $\mathscr{C}({\bf C}_{\delta\mbox{\tiny${\rm U}$}},{\bf C}_{\delta\mbox{\tiny${\rm V}$}})$, whereas $\frac{\mathcal{F}}{\mathcal{F}_{3}}$ and $\frac{\mathcal{H}}{\mathcal{F}_{3}}$ depend on the covariances of all combinations among ${\bf C}_{\delta\mbox{\tiny${\rm U}$}}$, ${\bf C}_{\delta\mbox{\tiny${\rm V}$}}$, ${\bf S}_{\delta\mbox{\tiny${\rm U}$}}$ and ${\bf S}_{\delta\mbox{\tiny${\rm V}$}}$. As a result, we may now write