12.2: Statistics of Ensembles

Time Dependent Quantum Mechanics and Spectroscopy 2025e · LibreTexts

0%

Contents

    Text

    Size
    100%
    Theme
    Typeface
    Line length
    Spacing

    Read aloud

    How

    Speed: 1×
    Light up
    Highlight

    Symbols: Mulberry Symbols, © 2018–2026 Steve Lee, CC BY-SA 4.0 (licence, the set), the site's own copies.

    From LibreTexts, under CC BY-NC-SA 4.0, as the source states it; its equations are written in TeX, which the PDF shows typeset. This page at LibreTexts.

    12.2: Statistics of Ensembles

    With the need to describe ensembles, let’s begin by reviewing the statistical description of a system. The variable \(A\), which can be a classical dynamical variable or quantum expectation value, is observed to take on a distribution of different values. The basic statistical properties of this distribution can be described in terms of the mean and mean-square values of \(A\), as determined from a large number of measurements:

    \[\left\langle A \right\rangle = \frac{1}{N}\sum\limits_{i = 1}^{N} A_{i} \label{eq12.2.01}\]\[\left\langle A^{2} \right\rangle = \frac{1}{N}\sum\limits_{i = 1}^{N} A_i^{2} \label{eq12.2.02}\]

    Here, the summation over \(i\) refers to averaging over \(N\) independent measurements. In fact, this is how the expectation value of a quantum operator is defined—as the average value of \(A\) or \(A^2\) obtained from many measurements on identically prepared pure states. For an ensemble, the states are drawn from a distribution of states. Alternatively, we can determine the same properties from:

    \[\left\langle A \right\rangle = \sum\limits_{j = 1}^{M} P_{j}A_{j} \label{eq12.2.03}\]\[\left\langle A^{2} \right\rangle = \sum\limits_{j = 1}^{M} P_{j}A_j^{2} \label{eq12.2.04}\]

    Now the sum over \(j\) refers to a sum over the \(M\) possible values that \(A\) can take, weighted by \(P_j\), the probability of observing a particular value \(A_j\). Naturally, \(\sum_{j}{P_j=1}\). While the result of these discrete averages is the same regardless of the method we use, it can be practical to use one form over the other. For instance, in calculating ensemble averages, we typically specify the probability of occupying a particular state \(P_j\). When we want to distinguish the methods, we will use \(\bar{A}\) to refer to that average from a sum of repeated measurements and \(\langle A\rangle\) to refer to an average over probability distributions.

    When the accessible values come from a continuous distribution—as opposed to a discrete distribution—we now describe the internal variable with a distribution function, \(P(A)\), which characterizes the probability of observing \(A\) between \(A\) and \(A+dA\) with \(\int P(A) \, dA=1\). \(P(A)\) is also known as the probability density since its units are \(A^{-1}\). The statistics of \(A\) are now described in terms of the moments of the distribution function. The first and second moment are just the mean and mean-squared values of \(A\)

    \[\left\langle A \right\rangle = \int dA \, A P( A ) \label{eq12.2.05}\]\[\left\langle A^{2} \right\rangle = \int dA \, A^{2} P( A ) \label{eq12.2.06}\]

    The breadth of the distribution function for \(A\) is captured in the variance of the distribution

    \[\sigma ^{2} = \left\langle A^{2} \right\rangle - \left\langle A \right\rangle ^{2} \label{eq12.2.07}\]

    \(\sigma\) is known as the standard deviation.

    Graph showing a probability distribution curve, labeled P(A) vs. ⟨A⟩, with arrows indicating parameters and an equation included.
    Figure \(\PageIndex{2}\): A simple normalized probability distribution function for the variable \(A\) is statistically described through its first and second moments, \(\left\langle A\right\rangle\) and \(\left\langle A^2\right\rangle\).

    These concepts can be generalized to a time-dependent probability density to describe the dynamical properties of the system. This function, written either \(P(A,t;A_0,t_0)\) or just \(P(A,t)\), is a conditional probability for observing the particle between \(A\) and \(A+dA\) at a time point \(t\), given that it was initially at \(A_0\) at an earlier time \(t_0\). We obtain the global probability density, \(P(A)\), by integrating \(P(A,t)\) over time. For instance, the average and mean-square displacement of a particle at time \(t^\prime\), given that it started at \(x=0\) at \(t=0\), are

    \[\begin{aligned}
    & \left\langle x(t') \right\rangle = \int dx \, x(t') P\left( x,t' \right) \\
    & \left\langle x^{2}(t') \right\rangle = \int dx \, x^{2}(t') P \left( x,t' \right) \\
    \end{aligned} \label{eq12.2.08}\]

    When discussing quantum systems, the key distinction between calculating statistical averages over classical probability distributions \(P(x)\) and the quantum expectation values for pure states is that the quantum expectation value is the intrinsic uncertainty in a perfectly defined state, whereas the classical average is over the uncertainty in defining the state. When we deal with quantum statistical mechanics of mixed states, we need to be concerned about both types of these types of statistical descriptions. The first level is the inherent uncertainty in quantum mechanical observables. A fundamental postulate is that the expectation value of an operator \(\langle \hat A \rangle = \left\langle \psi \right|\hat A\left| \psi \right\rangle \) is the mean value of \(A\) obtained over many observations on identically prepared systems \(|\psi\rangle\). \(\left\langle A\right\rangle\) represents the fundamental quantum uncertainty in a measurement of a precisely defined pure state with wave-like character. Given a basis set \({|n\rangle}\), we can calculate it as

    \[\left\langle A \right\rangle = \left\langle \psi \right|\hat A\left| \psi \right\rangle = \sum\limits_{m,n} c_{n}c_m^{*}\left\langle m \right|\hat A\left| n \right\rangle \label{eq12.2.09}\]

    where \(c_n\) describes the complex expansion coefficients. The second layer of statistical averaging is the uncertainty in defining the state of a system in an ensemble. A probability density \(P\) can be used to define possible outcomes of expectation value measurements on different initial systems \(\left|\psi^{(j)}\right\rangle\):

    \[\left\langle \left\langle A \right\rangle \right\rangle = \sum\limits_{j} P_{j} \left\langle \psi ^{(j)} \right|\hat A\left| \psi ^{(j)} \right\rangle \label{eq12.2.10}\]

    Here the double bracket notation is written to emphasize the two layers of averaging, but we take this to be understood moving forward. In calculating averages over probability densities, it is more common to assume that a common basis set applies to all realizations of the wavefunction, and therefore one can apply the statistical averages to the expansion coefficients. Then eq. (\ref{eq12.2.10}) is written

    \[\left\langle A \right\rangle = \sum\limits_{m,n} \left\langle c_{n}c_m^{*} \right\rangle \left\langle m \right|\hat A\left| n \right\rangle \label{eq12.2.11}\]

    where \(\left\langle c_{n}c_m^{*} \right\rangle \) are the statistically averaged expansion coefficients, which we will see are the elements of a density matrix.

    The statistical relationships above can be extended to two variables using a joint probability distribution, \(P(A,B)\), which characterizes the probability of observing both \(A\) between \(A\) and \(A+dA\) and \(B\) between \(B\) and \(B+dB\). The statistical relationship between the variables can also be quantified by calculating moments of \(P(A,B)\), either by summing or integrating over \(A\) and \(B\). The most important measure is the correlation function

    \[C_{AB} = \left\langle AB \right\rangle - \left\langle A \right\rangle \left\langle B \right\rangle \label{eq12.2.12}\]

    This is the covariance—the variance for a bivariate distribution—which measures the correlation between the variables \(A\) and \(B\). That is, for a specific value of \(A\), what are the associated statistics for \(B\)? To interpret this, it helps to define a correlation coefficient

    \[r = \frac{C_{AB}}{\sigma _{A}\sigma _{B}} \label{eq12.2.13}\]

    where \(\sigma_A\) and \(\sigma_B\) are the standard deviations for the probability densities \(P(A)\) and \(P(B)\). The correlation coefficient \(r\) can take on values from \(+1\) to \(-1\). If \(r = 1\) then there is perfect correlation between the two distributions. If the variables \(A\) and \(B\) depend on a common internal variable in the same way, then they are correlated. If no statistical relationship exists between the two distributions, then they are uncorrelated, \(r = 0\), and \(\left\langle AB \right\rangle = \left\langle A \right\rangle \left\langle B \right\rangle \). It is also possible that the distributions depend in an equal and opposite manner on an internal variable, in which case we call them anti-correlated with \(r = -1\).


    This page titled 12.2: Statistics of Ensembles was last modified on Thu, 04 Jun 2026 01:16:14 GMT and is shared under a CC BY-NC-SA 4.0 license and was authored, remixed, and/or curated by Andrei Tokmakoff via source content that was edited to the style and standards of the LibreTexts platform.