I was reading Chapter 12.1 in Hayt & Buck "Engineering Electromagnetics" 8-th edition. Here they discuss the reflection of uniform plane waves at normal incidence.

They derived the following expressions for the reflection and transmission coefficients:

$\Gamma = \frac{E_{x10}^{-}}{E_{x10}^{+}} = \frac{\eta_2 - \eta_1}{\eta_2 + \eta_1}$

$\tau = \frac{E_{x20}^{+}}{E_{x10}^{+}} = \frac{2\eta_2}{\eta_1+\eta_2} = 1 + \Gamma$

where $\eta_1, \eta_2$ are the intrinsic impedances of the two materials (which may be complex), and the electric field is uniform in the x direction, parallel to the interface.

Then they consider the power reflected and the power transmitted.

They use Poynting's theorem in phasor form: $\left<S\right> = \left|\frac{1}{2}\Re\left\{\mathbf{E}_s \times \mathbf{H}_s\right\}\right|$. From here, they use the reflection and transmission coefficients and the intrinsic impedances to conclude that

$\left<S_{1r}\right> = \left|\Gamma\right|^2\left<S_{1i}\right>$

$\left<S_{2}\right> = \frac{\Re\left\{1/\eta_2^*\right\}}{\Re\left\{1/\eta_1^*\right\}}\left|\tau\right|^2\left<S_{1i}\right>$

On the other hand, conservation of energy implies that the transmitted power must be the incident power minus the reflected power, so another expression is

$\left<S_{2}\right> = \left(1 - \left|\Gamma\right|^2\right)\left<S_{1i}\right>$

But the two forms of the coefficients obtained for the transmitted power are not in general equal. From my calculations, they would be equal if and only if the two intrinsic impedances have a ratio which is a real number. But I don't see any reason for this to be the case in general, and yet none of the steps seem to make that assumption.

So, where does the contradiction arise?

  • $\begingroup$ I would guess that with a complex impedance this implies some sort of dispersion, (i.e. work is been doing on charges, perhaps creating surface currents). $\endgroup$ Nov 10 '15 at 11:39
  • $\begingroup$ I also thought this, but doesn't Poyting's theorem take conduction into account? If I recall correctly, it contains the sum of conduction current energy, electric field energy and magnetic field energy. $\endgroup$
    – Tob Ernack
    Nov 10 '15 at 15:18
  • 1
    $\begingroup$ It does, but that then doesn't imply that the Poyting vector is conserved, think about a resistor, the Poyting vector goes into the resistor with the energy it carries been turned in to Joule heating. $\endgroup$ Nov 10 '15 at 20:15

If the impedances are complex then it means you have dissipative terms. For example, if the problem was normal incidence from vacuum into a conductor, then energy conservation is not as simple as saying the (magnitude of the) Poynting vector of the incident wave equals the sum of the Poynting vectors in the transmitted and reflected waves.

In a conductor, then there is an ${\bf E} \cdot {\bf J}$ term that must be included in any conservation of energy calculation, because the transmitted E-field does work on the conduction charges (not the case if the impedances are real).

For example, in a conducting material with a complex wavevector, the electric field could have the form $$ \vec{E} = E_0 \exp(-\alpha x) \exp[i(kx-\omega t)]\ \hat{j}$$ and the time-averaged Poynting vector will be proportional to $\exp(-2\alpha x)$, but is a vector along the x-axis.

The divergence of this is non-zero, which tells you that the Poynting vector is not a conserved quantity, so you cannot just equate the sums of the Poynting vectors.

  • $\begingroup$ Indeed, conductors will have energy dissipated into the induced currents. But what puzzles me is why these are not accounted for by the Poynting vector. In the book, the theorem states that the flux of the E x H vector into a closed surface is the sum of: the volume integral of E dot J; the time derivative of the volume integral of 1/2 D dot E; the time derivative of the volume integral of 1/2 B dot H. So I would expect the dissipative term to be part of the E * J integral... $\endgroup$
    – Tob Ernack
    Nov 11 '15 at 23:13
  • $\begingroup$ @TobErnack I don't follow your confusion. Take the example of a wave propagating in a metal. Its amplitude and its Poynting vector decay exponentially. If I consider a volume in the metal, the Poynting vector in does not equal the Poynting vector out. The divergence of $S$ is non-zero, so the flux of $S$ is not conserved. I think that is all that is going on here. $\endgroup$
    – ProfRob
    Nov 11 '15 at 23:26
  • $\begingroup$ Hm I think I see what you mean. In the metal, if we go deep enough, the amplitude of the electric and magnetic fields will become very small and so the Poynting vector there will also be small, so it doesn't take into account all the power that was dissipated in the charges while getting there. But what about the fields right at the boundary? They don't yet have the time to decay, but their amplitudes are still different because of the transmission coefficient. $\endgroup$
    – Tob Ernack
    Nov 11 '15 at 23:54
  • $\begingroup$ So if we calculated the Poynting vector at a point just below the interface, the exponential decay term would be negligibly different from 1, and so the average power there should be close to the total power? But it seems that the magnitude of the vector is instead discontinuous right at the boundary. $\endgroup$
    – Tob Ernack
    Nov 11 '15 at 23:58
  • $\begingroup$ @TobErnack The divergence is also maximised! ${\bf E}\cdot {\bf J}$ is a sink of Poynting vector. Immediately beyond the boundary the difference between incident and reflected power is used to power the transmitted wave and power a current density. $\endgroup$
    – ProfRob
    Nov 12 '15 at 0:48

You only expect transmitted power to be equal to the incident power minus the reflected power when no energy is supplied to charges (which can then lose energy to heat).

The Poynting vector does not have a divergence free energy flow, it can gain or lose energy by getting or supplying energy to charges. And the charges can give or get energy from the fields but can also gain or lose energy to heat.

An energy conservation argument tells you nothing unless you know there is nothing else the energy is going to.


I came back to this problem a few years later, and I finally figured out what was wrong.

First of all, the Poynting vectors are indeed conserved across the boundary (placed at $z = 0$), but you cannot use the $\vec{S}_{1i}$ and $\vec{S}_{1r}$ vectors independently in the energy conservation equation. This is because there is interference between the incident and reflected waves, which causes a standing wave pattern in medium 1, whose average intensity is position-dependent.

In general, interference effects cause the intensity of the sum of two waves to be different from the sum of the intensities of the two waves. The interference means that you cannot simply say $\langle S_{1i}\rangle = \langle S_{1r}\rangle + \langle S_2\rangle$ or $(1 - |\Gamma|^2)\langle S_{1i}\rangle = \langle S_2\rangle$

Let $\vec{S}_1$ be the total Poynting vector in medium 1, and $\vec{S}_2$ the same for medium 2.

By definition, $\vec{S}_1 = \vec{E}_1 \times \vec{H}_1$ where $\vec{E}_1$, $\vec{H}_1$ are the total electric and magnetic fields in medium 1. Assuming the interface between medium 1 and 2 has no current sheets ($\vec{K}_s = \vec{0}$), then the boundary conditions imply that $\vec{E}_1 = \vec{E}_2$ and $\vec{H}_1 = \vec{H}_2$. Therefore we clearly must also have $\vec{S}_2 = \vec{S}_1$, and intensity is conserved across the interface (at $z = 0$).

This means we should still be able to say that $\langle S_1\rangle = \langle S_2\rangle$ (at $z = 0$, again). So how do we correctly find $\langle S_1\rangle$ given $\langle S_{1i}\rangle$?

The incident wave in medium 1 propagates in the $\hat{z}$ direction, with $\vec{E}_{1i}$ in $\hat{x}$ and $\vec{H}_{1i}$ in $\hat{y}$. Moreover, in phasor form we know that $\mathbf{H}_{1i} = \frac{\mathbf{E}_{1i}}{\eta_1}$

The reflected wave in medium 1 propagates in the $-\hat{z}$ direction, with $\vec{E}_{1r}$ again in $\hat{x}$. This implies $\vec{H}_{1r}$ is in $-\hat{y}$, because the direction of the wave has been flipped. In phasor form, we thus have $\mathbf{H}_{1r} = -\frac{\mathbf{E}_{1r}}{\eta_1}$. In addition, we know that $\mathbf{E}_{1r} = \Gamma\mathbf{E}_{1i}$, and hence $\mathbf{H}_{1r} = -\Gamma\mathbf{H}_{1i}$.

The total phasors in medium 1 are then $\mathbf{E}_1 = \mathbf{E}_{1i} + \mathbf{E}_{1r} = (1 + \Gamma)\mathbf{E}_{1i}$ and $\mathbf{H}_1 = \mathbf{H}_{1i} + \mathbf{H}_{1r} = (1 - \Gamma)\mathbf{H}_{1i}$.

Therefore the Poynting vector in phasor form is $\mathbf{S}_1 = \frac{1}{2}\mathbf{E}_1\times\mathbf{H}_1^* = \frac{1}{2}(1 + \Gamma)(1 - \Gamma^*)\mathbf{E}_{1i}\times\mathbf{H}_{1i} = (1 + \Gamma)(1 - \Gamma^*)\mathbf{S}_{1i}$

Note that $\mathbf{S}_1 = (1 - |\Gamma|^2)\mathbf{S}_{1i} + 2j\Im\{\Gamma\}\mathbf{S}_{1i}$, so there is an additional term besides the $(1 - |\Gamma|^2)$ factor that comes from the interference effect. When $\Gamma$ and $\mathbf{S}_{1i}$ have imaginary components (i.e. when $\frac{\eta_2}{\eta_1}$ is not real and $\eta_1$ is not real) we get a nonzero contribution to the real part of the Poynting phasor. So we can only use $\langle S_{1i}\rangle = \langle S_{1r}\rangle + \langle S_2\rangle$ when $\eta_1$ is real or when $\frac{\eta_2}{\eta_1}$ is real.

We can substitute $\Gamma = \frac{\eta_2 - \eta_1}{\eta_2 + \eta_1}$ and $\mathbf{S}_{1i} = \frac{|\mathbf{E}_{1i}|^2}{2\eta_1}$ to obtain $$\mathbf{S}_1 = \frac{2\eta_2}{\eta_1 + \eta_2}\frac{2\eta_1^*}{\eta_1^* + \eta_2^*}\frac{|\mathbf{E}_{1i}|^2}{2\eta_1} = \frac{2\eta_2}{|\eta_1 + \eta_2|^2}|\mathbf{E}_{1i}|^2 = |\tau|^2\frac{|\mathbf{E}_{1i}|^2}{2\eta_2^*} = \frac{|\mathbf{E}_2|^2}{2\eta_2^*} = \mathbf{S}_2$$

which confirms that $\vec{S}_1 = \vec{S}_2$.

Moreover, $\langle S_1\rangle = \langle S_2\rangle = \Re\left\{|\tau|^2\frac{|\mathbf{E}_{1i}|^2}{2\eta_2^*}\right\} = \frac{|\tau|^2|\mathbf{E}_{1i}|^2}{2}\Re\{1/\eta_2^*\} = \frac{\Re\{1/\eta_2^*\}}{\Re\{1/\eta_1^*\}}|\tau|^2\langle S_{1i}\rangle$ and so we confirmed the formula for $\langle S_2\rangle$.

The comments about power dissipation due to $\vec{E}\cdot\vec{J}$ terms are valid, but they do not imply that $\vec{S}$ is discontinuous across the interface. The dissipative terms will produce an absorption coefficient which causes $\vec{S}$ to decay as the distance increases. But right across an interface, the enclosed volume is zero, so the volume integral of $\vec{E}\cdot\vec{J}$ will be nil (assuming the current densities are distributed over a volume, and there are no current sheets), and hence the inward flux equals the outward flux across the interface (i.e. the Poynting vector is continuous).

Additionally, absorption coefficients can arise even if the medium is non-conductive, i.e. even if $\vec{J}_f = \vec{0}$. This is because power can also be dissipated by bound currents and movement of bound charges which do not get accounted for explicitly in Poynting's theorem. Instead they show up by causing complex electric and magnetic permittivities.

Another way to understand this is that although $\vec{S}$ has generally nonzero divergence, the divergence is still finite, so fluxes of $\vec{S}$ can only vary continuously across a volume, and they cannot just change right across an interface. For this to happen you would need infinite volume current densities, or basically something like current sheets or current wires.


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.