A key relation in optical spectroscopy also used to derive natural bandwidth is:

$\Delta E \Delta t=ℏ$

$\Delta E$ is the energy uncertainty of a state, $\Delta t$ should be the uncertainty of the lifetime $τ_b$ of the state. I don't understand why $\Delta t$ can be confounded with the lifetime itself, $τ_b$, and the relation written as

$\Delta E τ_b=ℏ$

  • $\begingroup$ Where does it say $E$ and $t$ are not conjugated? AFAIK you can get the energy-time-uncertainty just by rearranging the position-momentum-uncertainty. $\endgroup$
    – Feodoran
    Feb 19 '19 at 10:36
  • $\begingroup$ @Alchimista It ($E$ and $t$ not being conjugated) is not part of the question, yes. But since you mentioned it, and I have not heard such statements before, I was hopping you had a reference where you found such statements. $\endgroup$
    – Feodoran
    Feb 19 '19 at 14:15
  • $\begingroup$ @fedoran here in Chem or Phys SE. I am trying to find it as I was curious already. I feel the discussion boil to be as the situation in Q but I am also going do delete the previous comments as for they can be detrimental and distracting. I could make a question from scratch, if necessary. $\endgroup$
    – Alchimista
    Feb 19 '19 at 15:14
  • $\begingroup$ Also related the discussion here and link therein physics.stackexchange.com/questions/365873/… $\endgroup$
    – Alchimista
    Feb 19 '19 at 15:30
  • $\begingroup$ I would understand such a treatment (lifetime = delta t) only assuming the excited state as a very short leaving ones, ie one that would decay "immediately" but it survives just because of uncertainty. $\endgroup$
    – Alchimista
    Feb 19 '19 at 15:38

For practical purposes the relevant aspect of the relation between the uncertainties in time and energy is that it is a Fourier relationship that explains how peak widths in a spectrum (with linewidths in frequency units) relate to the decay rates of time-dependent signals. A decay can be considered the statistical result of many independent events with an average probability of occurring described by a time constant $\tau$. It can further be shown via the Fourier transform that when a signal decays exponentially with time constant $\tau$, this leads to a Lorentzian peak shape with a width at half height proportional to $\tau^{-1}$. It is even simpler to show this in the case of a Gaussian shaped decay with linewidth ~$\sigma_t$. The resulting frequency peak is then Gaussian with linewidth ~$\sigma_t^{-1}$. When you multiply the linewidth parameters of the Gaussian functions you get:


If you use Planck's relation $E=h\nu$ if follows that


Now - as mentioned earlier - you can think of the width of the Gaussian or Lorentzian time-dependent decay as proportional (or described by) a time constant $\tau$ (which is the default description in the case of exponential decays). It follows that

$$\sigma_E~\tau \approx h$$


The time energy uncertainty principle arises from the fact that a function can be described as a sum of terms in a conjugate variable, as is the case of any fourier transform. In this case the variables are frequency and time. The time-energy uncertainty does not seem to have any more fundamental meaning than this, unlike the Heisenberg position-momentum uncertainty principle.

As an excited state decays it is necessary to write the wavefunction in a time dependent form. If a pure state has energy $E$ and wavefuntion $\psi$ we can write the time-dependent (or total) wavefunction as $\displaystyle \varphi=\psi e^{-iE_0/\hbar}$ (where $i=\sqrt{-1}$ and $\hbar =h/2\pi$ ) and is a stationary state because the probability of finding it in that state is $|\varphi|^2=|\psi|^2$ which is a constant.

However, if the state decays, because of interactions between electrons/nuclei within a molecule or with nearby molecules (collisions for example), and if the lifetime is $\tau$ then the probability of being in the excited state at time $t$ is $\displaystyle |\varphi|^2=|\psi|^2 e^{-t/\tau}$ which means that

$$ \varphi(t)=\psi e^{-iE_0t/\hbar-t/2\tau}$$

expressing this as a product $\displaystyle \varphi=\psi e^{-iE_0t/\hbar}e^{-t/2\tau}$ and using the mathematical relationship $e^{ix}=\cos(ix)+i\sin(ix)$ shows that the real part of $\varphi$ oscillates due to the cosine term and decays away due to the exponential term in $\tau$. i.e a damped oscillation.

The calculation is easier by making some substitutions, for example the frequency is $\omega_0=E_0/\hbar$ and decay rate constant $\gamma=1/\tau$ then $\displaystyle \varphi=\psi \cos(\omega_0 t)e^{-\gamma/2}$.

As the excited state is decaying the frequency of the emitted radiation is not monochromatic, which it would be if the state did not decay, but has a frequency distribution that is related to the exponential decay by a fourier transform. Put another way, the exponential decay can be constructed out of a set of waves each with a different frequency and amplitude, just as any normal function can be constructed as a fourier series. This means that we can find a function $A$ such that

$$\varphi(t) =\int A(\omega)e^{i\omega t}d\omega$$


$$A(\omega) \approx \frac{1}{i(\omega-\omega_o)+\gamma/2}+\frac{1}{i(\omega+\omega_o)+\gamma/2}$$

The second fraction approaches zero as $\omega \to \omega_0$, i.e close to the transition frequency (or transition energy) and so is insignificant compared to the first term where the frequency difference tends to zero. The intensity of the transition is $A(\omega)^*A(\omega)$ where * indicates the complex conjugate giving,

$$I(\omega)= \frac{\gamma/2}{(\omega-\omega_0)^2+(\gamma/2)^2}$$

where the intensity is normalised to unity over all frequencies. The full width at half maximum of the transition is

$$ \Delta\omega =\gamma , \text{ or} \qquad \Delta E\tau =\hbar $$

which is the 'time-energy' uncertainty after substituting back for the energy.


I've finally found the answer in Griffiths' Quantum Mechanics Ch.3; The $\Delta t$ in the energy-time uncertainty principle is not the standard deviation of a collection of time measurements (like $\Delta E$) but it is the time the system takes to change substantially;

As a mesure of how fast the system is changing, let us compure the time derivative of the expectation value of an Oobservable $Q(x,p,t)$:

$\frac{d}{dt} \langle Q \rangle = \frac{d}{dt} \langle \Psi | Q \Psi \rangle = \langle \frac{\partial \Psi}{\partial t} | Q \Psi \rangle +\langle \Psi| \frac{\partial Q}{\partial t} \Psi \rangle + \langle \Psi | Q \frac{\partial \Psi}{\partial t} \rangle$

Now, the Schroedinger equation says

$i\hbar\frac{\partial\Psi}{\partial t} = H \Psi$


$\frac{d}{dt} \langle Q \rangle = -\frac{1}{i\hbar} \langle H \Psi | Q \Psi \rangle +\frac{1}{i\hbar} \langle \Psi | H Q \Psi \rangle + \langle \frac{\partial Q}{\partial t}\rangle$

Given the fact that H is hermitian $\langle H\Psi | Q \Psi \rangle = \langle \Psi |HQ\Psi \rangle$, hence

$\frac{d}{dt} \langle Q \rangle = \frac{i}{\hbar} \langle [\hat A,\hat B] \rangle + \langle \frac{\partial Q}{\partial t}\rangle $

If you consider the operator $\hat{Q}$ time independent (Operators that depend explicitly by $t$ are quite rare), and the general form of the Uncertainty principle:

$\sigma_A^2 \sigma_B^2 ≥ (\frac{1}{2i} \langle [\hat A,\hat B] \rangle)^2$

If A=H and B=Q you have eventually:

$\sigma_H^2 \sigma_Q^2 ≥ (\frac{1}{2i} \langle [\hat H,\hat Q] \rangle)^2 = \left(\frac{1}{2i} \frac{\hbar}{i} \frac{d\langle Q \rangle}{dt}\right)^2 =\left(\frac{\hbar}{2}\right)^2 \left(\frac{d \langle Q \rangle}{dt}\right)^2$

and more simply

$\sigma_H^2 \sigma_Q^2 ≥ \frac{\hbar}{2}\left|\frac{d \langle Q \rangle}{dt}\right| $

By defining $\Delta E \equiv \sigma_H$ and

$\Delta t \equiv \frac{\sigma_Q}{d \langle Q \rangle/dt}$


$\Delta E \Delta t ≥ \frac{\hbar}{2}$

The important point to understand (which confused me) is that $\Delta t$ represents the amount of time it takes the expectation value of Q to change by one standard deviation (thus to $\tau_b$) and not the standard deviation of a collection of time measurements

  • $\begingroup$ Regarding your last sentence, this is like distinguishing between homogenenous versus inhomogenenous sources of line broadening. In fact, for a collection of time measurements both will occur, but the linewidth will reflect the lifetime you derive with the Heisenberg relation only in the limit of no inhomogeneous broadening mechanisms. $\endgroup$
    – Buck Thorn
    Mar 2 '19 at 10:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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