The problem area is geometric optics, namely refraction across homogeneous media with constant speed of light.
I explain the three steps of a methodological doubt that popped up.
Polite request: please use answers for answering and comments for commenting or asking.

Step 1: Where Snell's law comes from

I am aware that Snell's law is derived from Fermat's principle of least time.

Snell's law is derived upon applying Fermat's principle to an idealized ray between two points, say $A$ and $B$, at either side of a single refractive surface, $\sigma$, a plane for simplicity.
Once you determine the traversal time $T$ and minimize it with respect to the position of the point of incidence $R$ on the refractive surface, say with coordinate $r$, the statement $\delta T(r)=0$ tells us that the quantity $\sin \theta/c$ has to be the same across the refractive interface $\sigma$.
There is no need to develop the algebra further and determine the position of $R$, neither do the positions of $A$ and $B$ need to be specified. Snell's law is in fact a necessary and sufficient condition for a point of incidence $R$ to be found, so it applies to any ray and any point on the refractive interface $\sigma$. (Leaving aside the rays resulting into reflection, for simplicity.)

Broadly speaking, if I phrase it right, Snell's law is a corollary of Fermat's principle.
So far, plain sailing.

Step 2: A similar refraction problem with a single interface

Let's turn to drawing a real ray between an illuminated point $A$ and an eye point $B$ at either side of a single refractive surface $\sigma$. The positions of $A$ and $B$ are now meaningful for this problem. The objective is to determine the position of the point of incidence $R$.

Fermat's principle can be used for this problem of course. You need to move on from where Snell's reasoning stopped. The minimization problem $\delta T(r)=0$ is one algebraic equation in one unknown. You solve it and find the coordinate of $R$ on the refracting interface that is physically meaningful. Of course, Snell's law is implicit in this result.

This too is plain sailing.

Step 3: Refraction across two interfaces

Now, suppose that between $A$ and $B$ (known) there are two refractive interfaces, $\sigma_1$ and $\sigma_2$ (planes, for simplicity). The ray between $A$ and $B$ crosses three media with a different speed of light. I want to determine the ray path between $A$ and $B$ as before: the challenge is to determine the two points of incidence of the ray on the interfaces, $R_1$ and $R_2$.

Fermat's principle applies here too. The traversal time $T$ is the sum of the times to traverse the three media across the segments $AR_1$, $R_1 R_2$, $R_2 B$. The traversal time has to be minimized. Here, this statement gives one equation with two unknowns, $\delta T(r_1, r_2) = 0$. I then looked for one more geometric condition to link $r_1$ and $r_2$ and use the same algebra as the system with a single refractive interface.

Here, I thought of calling in Snell's law. Using the prior knowledge of the relationship between the angles of incidence and refraction at each interface, I could express the traversal time as $T(r_1)$ or $T(r_2)$, minimize it, and fix either $R_1$ or $R_2$ first.

This is where the methodological doubt kicked in.

  • Snell's law is in itself a corollary of Fermat's principle, that is the result of a minimization problem, $\delta T = 0$.
  • If I use Snell's law to anticipate the ray behaviour across two interfaces, I am basically feeding into the new $T$ the result of a prior $\delta T = 0$. The new $T$ has to be minimized yet, though. The new minimization problem should bring about, as a corollary, that the Snell's law applies on the same interfaces where I had imposed it.

Basically, I would be sneaking into the problem's givens a part of the solution, with a potential to corrupt the method that guarantees the correctness of the solution.

So, the main question is whether this concern is well posed in the first place.

  • If so, which other non-Snell condition should be imposed to solve for $r_1$ and $r_2$?
  • If not, what makes the trick or rather what debunks the concern?

2 Answers 2


$\newcommand{\bl}[1]{\boldsymbol{#1}} \newcommand{\e}{\bl=} \newcommand{\p}{\bl+} \newcommand{\m}{\bl-} \newcommand{\gr}{\bl>} \newcommand{\les}{\bl<} \newcommand{\greq}{\bl\ge} \newcommand{\leseq}{\bl\le} \newcommand{\plr}[1]{\left(#1\right)} \newcommand{\blr}[1]{\left[#1\right]} \newcommand{\lara}[1]{\langle#1\rangle} \newcommand{\lav}[1]{\langle#1|} \newcommand{\vra}[1]{|#1\rangle} \newcommand{\lavra}[2]{\langle#1|#2\rangle} \newcommand{\lavvra}[3]{\langle#1|\,#2\,|#3\rangle} \newcommand{\vp}{\vphantom{\dfrac{a}{b}}} \newcommand{\hp}[1]{\hphantom{#1}} \newcommand{\x}{\bl\times} \newcommand{\qqlraqq}{\qquad\bl{-\!\!\!-\!\!\!-\!\!\!\longrightarrow}\qquad} \newcommand{\qqLraqq}{\qquad\boldsymbol{\e\!\e\!\e\!\e\!\Longrightarrow}\qquad} \newcommand{\tl}[1]{\tag{#1}\label{#1}} \newcommand{\hebl}{\bl{=\!=\!=\!==\!=\!=\!==\!=\!=\!==\!=\!=\!==\!=\!=\!==\!=\!=\!==\!=\!=\!==\!=\!=\!==\!=\!=\!==\!=\!=\!==\!=\!=\!==\!=\!=\!=}}$

enter image description here

Consider that in Figure 01 above the givens are $v_1,v_2,v_3,\alpha,y_1,y_2,\beta$. The two equations of Snell's Law on the surfaces $\sigma_1\,$ and $\sigma_2$ will give a system with respect to the unknown variables $x_1,x_2$ and identify the points $\texttt C$ and $\texttt D$ respectively.

\begin{align} t & \e t_1\p t_2\p t_3 \tl{01}\\ t_1\plr{x_1} & \e \dfrac{\sqrt{x^2_1\p \plr{\alpha\m y_1}^2}}{v_1} \tl{02.1}\\ t_2\plr{x_1,x_2} & \e \dfrac{\sqrt{\plr{x_2\m x_1}^2\p \plr{y_2\m y_1}^2}}{v_2} \tl{02.2}\\ t_3\plr{x_2} & \e \dfrac{\sqrt{\plr{\beta\m x_2}^2\p y^2_2}}{v_3} \tl{02.3} \end{align}

Therefore: \begin{equation} t\plr{x_1,x_2}\e \dfrac{\sqrt{x^2_1\p \plr{\alpha\m y_1}^2}}{v_1}\p\dfrac{\sqrt{\plr{x_2\m x_1}^2\p \plr{y_2\m y_1}^2}}{v_2}\p\dfrac{\sqrt{\plr{\beta\m x_2}^2\p y^2_2}}{v_3} \tl{03} \end{equation}

\begin{align} \mathrm dt & \e \dfrac{\partial t}{\partial x_1}\mathrm dx_1\p\dfrac{\partial t}{\partial x_2}\mathrm dx_2\e 0 \quad\bl\implies \tl{04}\\ \dfrac{\partial t}{\partial x_1} & \e\left[\dfrac{1}{v_1}\dfrac{x_1}{\sqrt{x^2_1\p \plr{\alpha\m y_1}^2}}\m\dfrac{1}{v_2}\dfrac{x_2\m x_1}{\sqrt{\plr{x_2\m x_1}^2\p \plr{y_2\m y_1}^2}}\right]\e0 \tl{04.1}\\ \dfrac{\partial t}{\partial x_2} & \e\left[\dfrac{1}{v_2}\dfrac{x_2\m x_1}{\sqrt{\plr{x_2\m x_1}^2\p \plr{y_2\m y_1}^2}} \m\dfrac{1}{v_3}\dfrac{\beta\m x_2}{\sqrt{\plr{\beta\m x_2}^2\p y^2_2}} \right]\e 0 \tl{04.2} \end{align}

enter image description here

The surface in Figure 02 represents the function $t\plr{x_1,x_2}$ of Equation \eqref{03}. The graticule of $x_1$- and $x_2$-parameter curves is shown on the surface. The drawing was produced with the following data: \begin{equation} \begin{split} \alpha &\e 10 \qquad \beta \e 10\\ y_1 &\e 7 \qquad y_2 \e 4\\ v_1 &\e 5 \qquad v_2 \e 4 \qquad v_3 \e 3\\ x_1 &\bl\in \blr{-5,21} \qquad x_2 \bl\in \blr{-5,21} \end{split} \tl{05} \end{equation} Note that the solution with respect to $x_1,x_2$ depends on the ratios of the speeds and not on their values. For example, instead of the speeds in \eqref{05}, we could equivalently have used the following refraction indices:
\begin{equation} n_1 \e 1 \qquad n_2 \e \frac54\e 1.25 \qquad n_3 \e\frac53\approx 1.67 \tl{06} \end{equation}

enter image description here

In Figure 03 the two curves $\mathcal C_1$ and $\mathcal C_2$ on the surface are the geometrical loci of the minimum with respect to $t$ of the $\,x_1$- and $\,x_2$-parameter curves respectively. Their projections $\mathcal C'_{\,\!1}$ and $\mathcal C'_{\,\!2}$ on the $\,x_1x_2$-plane intersect at a point $\texttt P$, the coordinates of which are the solution $(x_1,x_2)$ of the system of Equations \eqref{04.1}-\eqref{04.2}.

As an example we determine the values of $(x_1,x_2)$ for the configuration in Figure 01 given the data in Equation \eqref{05}.

Squaring the Snell's Law equations \eqref{04.1} and \eqref{04.2} the pair of unknowns $(x_1,x_2)$ must satisfy the following non-linear system: \begin{align} v^2_2x^2_1\left[\plr{x_2\m x_1}^2\p \plr{y_2\m y_1}^2\vp\right]\m v^2_1\plr{x_2\m x_1}^2\left[x^2_1\p \plr{\alpha\m y_1}^2\vp\right] &\e0 \tl{07.1}\\ \!\!\!\!\!\!\!\!\!\!\!\!\!\!v^2_3\plr{x_2\m x_1}^2\left[\plr{\beta\m x_2}^2\p y^2_2\vp\right]\m v^2_2\plr{\beta\m x_2}^2\left[\plr{x_2\m x_1}^2\p \plr{y_2\m y_1}^2\vp\right]&\e0 \tl{07.2} \end{align} After inserting the data of \eqref{05}, these equations are represented by the curves on the $x_1x_2$-plane shown in Figure 04. More precisely, Equation \eqref{07.1} is represented by the two-branch curve $\mathcal C'_{\,\!1}\bl\cup\mathcal C''_{\,\!1}$, while Equation \eqref{07.2} is represented by the two-branch curve $\mathcal C'_{\,\!2}\bl\cup\mathcal C''_{\,\!2}$.

The branches $\mathcal C'_{\,\!1}$ and $\mathcal C'_{\,\!2}$ are accepted while the branches $\mathcal C''_{\,\!1}$ and $\mathcal C''_{\,\!2}$ are rejected based on the following reasoning.
Equation \eqref{04.1} is identical with the Snell's Law at point $\texttt C$ in Figure 01: \begin{equation} n_1\sin\theta_1\plr{x_1,x_2}\e n_2\sin\theta_2\plr{x_1,x_2} \tl{08} \end{equation} Equation \eqref{07.1} is equivalent to the square of Equation \eqref{08}: \begin{equation} n^2_1\sin^2\theta_1\plr{x_1,x_2}\e n^2_2\sin^2\theta_2\plr{x_1,x_2} \tl{09} \end{equation} On the one hand, the solution pairs $(x_1,x_2)$ of Equation \eqref{09} represented by the branch $\mathcal C'_{\,\!1}$ satisfy \eqref{08} and are accepted. On the other hand, those represented by the branch $\mathcal C''_{\,\!1}$ satisfy Equation \eqref{10} below and are rejected.
\begin{equation} n_1\sin\theta_1\plr{x_1,x_2}\e \m n_2\sin\theta_2\plr{x_1,x_2} \tl{10} \end{equation} Similarly, Equation \eqref{04.2} is identical with the Snell's Law at point $\texttt D\,$ in Figure 01: \begin{equation} n_2\sin\theta_2\plr{x_1,x_2}\e n_3\sin\theta_3\plr{x_1,x_2} \tl{11} \end{equation} Equation \eqref{07.2} is equivalent to the square of Equation \eqref{11}: \begin{equation} n^2_2\sin^2\theta_2\plr{x_1,x_2}\e n^2_3\sin^2\theta_3\plr{x_1,x_2} \tl{12} \end{equation} On the one hand, the solution pairs $(x_1,x_2)$ of Equation \eqref{12} represented by the branch $\mathcal C'_{\,\!2}$ satisfy Equation \eqref{11} and are accepted. On the other hand, those represented by the branch $\mathcal C''_{\,\!2}$ satisfy \eqref{13} below and are rejected.
\begin{equation} n_2\sin\theta_2\plr{x_1,x_2}\e \m n_3\sin\theta_3\plr{x_1,x_2} \tl{13} \end{equation}

So, the solution pair $(x_1,x_2)$ is found as the coordinates of the point $\texttt P\equiv \mathcal C'_{\,\!1}\bl\cap\mathcal C'_{\,\!2}$. Note that the implicit curves $\mathcal C'_{\,\!1},\mathcal C'_{\,\!2}$ and their intersection point $\texttt P$ of Figure 04 are also shown in the $x_1x_2$-plane of Figure 03.

enter image description here

The coordinates $(x_1,x_2)$ with accuracy $10^{\m10}$ are: \begin{equation} x_1\e4.8443568078\,,\qquad x_2\e7.6277175213 \tl{14} \end{equation} We determine the angles from the geometry of the configuration: \begin{equation} \begin{split} \!\!\!\!\!\!\!\!\!\!\!\!\!\!\sin\theta_1 & \e\dfrac{x_1}{\sqrt{x^2_1\p \plr{\alpha\m y_1}^2}}\e0.8501777598\quad \bl\implies\theta_1\e58.2310087871^{\bl\circ}\\ \!\!\!\!\!\!\!\!\!\!\!\!\!\!\sin\theta_2 & \e\dfrac{x_2\m x_1}{\sqrt{\plr{x_2\m x_1}^2\p \plr{y_2\m y_1}^2}}\e0.6801422079 \quad \bl\implies\theta_2\e42.8547566685^{\bl\circ}\\ \!\!\!\!\!\!\!\!\!\!\!\!\!\!\sin\theta_3 & \e\dfrac{\beta\m x_2}{\sqrt{\plr{\beta\m x_2}^2\p y^2_2}}\e0.51010665593\quad \bl\implies\theta_3\e30.6709342993^{\bl\circ}\\ \end{split} \tl{15} \end{equation}

enter image description here

The resulting solution is shown in Figure 05. As a check, we have the following running constant along the light path $\texttt{ACDB}$: \begin{equation} \dfrac{\sin\theta_1}{v_1} \e \dfrac{\sin\theta_2}{v_2} \e \dfrac{\sin\theta_3}{v_3} \e 0.1700355520 \tl{16} \end{equation} or \begin{equation} n_1\sin\theta_1\e n_2\sin\theta_2 \e n_3\sin\theta_3 \e 0.8501777598 \tl{17} \end{equation}

  • 2
    $\begingroup$ Reconnecting to the question, the differential of the traversal time, properly worked out as in Eq. 04, provides two conditions, which make the search for a constraint unnecessary. The Snell's law at the two refracting planes is embedded in the two conditions (04.1) and (04.2). $\endgroup$ Dec 24, 2021 at 6:21
  • $\begingroup$ @XavierStuvw : Precisely. $\endgroup$
    – Frobenius
    Dec 24, 2021 at 7:00

It seems the core of OP's question is the following.

Question: When we derived Snell's law for a single interface from Fermat's principle, we held 2 points fixed. How can we then use repeatedly Snell's law for a double interface, since we are not allowed to hold 3 points fixed during the variation?

Answer: Although it is true that we're only allowed to hold 2 points fixed in Fermat's principle [namely the source and the target], Snell's law is a local condition. In fact, Snell's law holds at each point along the optical path(s), not just at the interface point(s). [In the bulk of a medium, Snell's law just says that the angles of incidence and refraction are the same relative to an arbitrary direction, i.e. the light continues along a straight line.]

  • $\begingroup$ The OP was considering to insert a constraint in the expression of the traverse time, so that its minimization had provided one equation in one unknown and a refracted ray could have been traced. This constraint was Snell's law. However, he sensed this was a conceptual mistake since he would have introduced a (local) solution in the set up of the problem. A sort of begging the question. In fact, differentiating the traverse time as a functional of n variables provides n relationships. The Fermat problem of refraction across multiple interfaces leads to a system of Snell statements $\endgroup$ Dec 24, 2021 at 14:10

Your Answer

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

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