Say we have a density matrix for a spin-1 system given by:

$$ \hat \rho = \left[\begin{array} A \hat \rho_{00}& \hat \rho_{01} & \hat \rho_{02} \\ \hat \rho_{10} & \hat \rho_{11} & \hat \rho_{12} \\ \hat \rho_{20} & \hat \rho_{21} & \hat \rho_{22} \end{array}\right] $$

The diagonal entries $\rho_{00}$, $\rho_{11}$ and $\rho_{22}$ describe the probability of the system being in the $|0 \rangle$, $|1 \rangle$ and $|2 \rangle$ state.

The off-diagonal elements of a density matrix are referred to as coherences and they describe phase relationships between states that can develop.

Is there some relationship between the coherences of $\hat \rho$ and the diagonal entries of $\hat \rho$?

My intuition leads me to believe there is a relationship. For example, if a system has a non-zero coherence say $\rho_{01} = \langle 0 | \hat \rho | 1 \rangle$ (and its conjugate). Say the system underwent a process that reduced the probability of being in the $|0 \rangle$ and $|1 \rangle$ state, then $ \rho_{00}$ and $\rho_{11}$ would reduce, would we not then also expect $ \rho_{01}$ (and its conjugate) to reduce as well?

  • 1
    $\begingroup$ The relationships between the coefficients of $\rho$ are the ones which comes from the definition of a density matrix : $\rho^\dagger = \rho$, $\operatorname{Tr}(\rho) =\rho$ and $\rho \geq 0$. $\endgroup$ May 17, 2021 at 20:29
  • $\begingroup$ @SolubleFish Hmm, so the trace of the density matrix should be equal to 1 (probability conservation). And we know that off-diagonal elements are equal to their conjugates (i.e. $\hat \rho^{\dagger} = \hat \rho$) as you said. However, I don't see how either of these statements give you a relationship between the coherences of $\hat \rho$ and the diagonal elements of $\hat \rho$. Could you elaborate? Thanks $\endgroup$
    – NahPlsMan
    May 17, 2021 at 20:36

1 Answer 1


Consider a finite dimensional complex Hilbert space $H$ of dimension $d$ equipped with an inner product denoted by $(\cdot,\cdot)$ and let $\rho$ be a generic density operator, i.e. a positive semi-definite operator with unit trace. This enables us to write

$$\rho = (\rho^{1/2})^\dagger \,\rho^{1/2} \quad . $$

Define $\rho_{nm} \equiv (v_n,\rho\, v_m)$, where $V=\{v_n\}$ is a complete orthonormal basis in $H$. Using the Cauchy-Schwarz inequality, we can show that the off-diagonal elements of the density matrix are bounded by the diagonal elements: \begin{align} |\rho_{nm}| &= |(v_n,\,(\rho^{1/2})^\dagger \,\rho^{1/2}\, v_m) | =|(\rho^{1/2}\,v_n,\,\rho^{1/2}\, v_m) |\\ &\leq \sqrt{(\rho^{1/2}\, v_n,\,\rho^{1/2}\, v_n)\, (\rho^{1/2}\, v_m,\,\rho^{1/2}\, v_m)} \\ &= \sqrt{(v_n,\,\rho \,v_n) \, (v_m,\,\rho \,v_m)} \quad , \end{align}

which eventually yields

$$|\rho_{nm}|^2 \leq \rho_{nn}\, \rho_{mm} \quad. \tag{$*$}$$

The above inequality holds for all elements $v_n,v_m \in V$. Note that in particular the off-diagonal elements vanish if one of the corresponding diagonal elements is zero.

Regarding the purity, which we will define as $$\pi(\rho) \equiv \mathrm{Tr}\rho^2 \quad . $$ The purity shows how 'mixed' / 'pure' a density operator is and it is bounded by $d^{-1}\leq \pi(\rho) \leq 1$ for all density operators $\rho$. The lower bound is reached for maximally mixed states and the upper bound for pure states. Most importantly, the value of $\pi(\rho)$ is independent of the basis in which you express the density operator as a matrix.

But since every density operator is diagonalizable, there always exists a basis in which there are no off-diagonal elements. For example, a pure state as well as a maximally mixed state each have a basis in which all off-diagonal elements are zero, while their corresponding purities differ. So just from the fact that the off-diagonal elements (in a given basis) are zero or non-zero you don't know, in general, anything about the purity of a density operator.

In other words, the purity of a density operator is basis-independent, in contrast to the appearance of off-diagonal elements.

That being said, we still can conclude two things from equation $(*)$, namely:

$$\mathrm{Tr} \rho^2 \leq \left(\mathrm{Tr} \rho\right)^2 = 1 $$ and further that the equality in $(*)$ holds for all $v_n, v_m \in V$ if only if $\rho$ is a pure state.

To show the latter statement, we first suppose that $|\rho_{nm}|^2 = \rho_{nn}\, \rho_{mm}$ holds for all $v_n,v_m \in V$. Then

$$\mathrm{Tr}\rho^2 = \sum\limits_{nm} |\rho_{nm}|^2 = \sum\limits_{nm} \rho_{nn} \, \rho_{mm} = \left(\mathrm{Tr}\rho\right)^2 = 1 \quad ,$$

which shows that $\rho$ is pure. For the reverse direction, we recall that for a pure density operator $\rho^\psi$ there exists by definition a unit vector $ \psi \in H$ s.t. $\rho^\psi = \psi \psi^\dagger$. Expanding $\psi$ in the orthonormal basis $V$:

$$ \psi = \sum\limits_n c_n \, v_n$$

with $c_n \equiv (v_n,\psi)$ and $\sum\limits_n |c_n|^2 = 1$ shows that $\rho^\psi_{nm} = c_n\, \bar{c}_m $ and thus $$|\rho^\psi_{nm}|^2 = |c_n|^2\, |c_m|^2 = \rho^\psi_{nn}\, \rho^\psi_{mm} \quad , $$

which completes the proof.

Consequently, for a mixed density operator $\sigma$ we find that there exists at least one pair $v_i,v_j \in V$ such that:

$$ |\sigma_{ij}|^2 < \sigma_{ii}\, \sigma_{jj} \quad ,$$

while for a pure state $\rho$

$$|\rho_{nm}|^2 = \rho_{nn}\, \rho_{mm} \quad \forall\, v_n,v_m \in V $$ holds.

It is stressed that we chose an arbitrary orthonormal basis and hence the above (in)equalities hold for any such basis (specifically the eigenbasis). All in all, we see that equation $(*)$ also serves as a criterion to decide whether a given density operator expressed in an arbitrary complete orthonormal basis is pure or mixed.

  • 1
    $\begingroup$ Thank you, that was insightful. Just to clarify, is there some relationship between the inequality $|\rho_{nm}|^2 \leq \rho_{nn} \rho_{mm}$ and the pureness of the system? For example, a mixed state (or impure state) has the relation $\mathrm{Tr}[\rho^{2}] < 1$, where the value of $\mathrm{Tr}[\rho^{2}]$ tends closer towards 0 the more mixed the state is. Is there some relation between the mixedness of a state and the inequality that you have derived (i.e. does a more mixed state lead to smaller value of $|\rho_{nm}|^{2}$)? $\endgroup$
    – NahPlsMan
    May 17, 2021 at 22:20
  • 2
    $\begingroup$ @NahPlsMan I've edited the answer. Please double check everything and let me know if everything is fine and whether you have further questions. $\endgroup$ May 18, 2021 at 11:51
  • $\begingroup$ Is there a reference that I can refer to for the inequality you derived? $\endgroup$
    – NahPlsMan
    May 21, 2021 at 13:15
  • 2
    $\begingroup$ @NahPlsMan Sorry, I don't know any particular book... But this is very well-known, see for example this homework sheet. Exercise $6$ this answer is basically exercise 6. $\endgroup$ May 21, 2021 at 13:21
  • 1
    $\begingroup$ @NahPlsMan Another homework sheet. Both courses are on quantum optics... It seems that this is a fairly common exercise there. But as I said, I don't know any reference, sorry. Hope this helps, tho. $\endgroup$ May 21, 2021 at 13:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.