ABCD Matrix Beam Propagation

A Gaussian beam arrives from the previous article carrying everything it needs inside a single complex number. aq-optics-gaussian-beams showed that the lowest transverse mode is pinned at every plane by two real quantities — the spot size $w$ and the wavefront radius $R$ — folded into the complex beam parameter through $1/q = 1/R - i\lambda/(\pi w^2)$ . The question that makes $q$ worth defining is the one answered here: given $q$ at the input of an optical train — lens, curved mirror, air gap, laser crystal — what is $q$ at the output, without re-solving the diffraction integral each time? The answer rests on a surprising economy. Paraxial ray optics is linear: a ray is a two-vector of height and slope, every element is a $2\times2$ matrix on it, and a compound system is the ordered product of its parts — nineteenth-century Gaussian optics. The non-obvious content, due to Kogelnik, is that the same matrices propagate a diffracting Gaussian beam, provided it is carried by $q$ and updated by the bilinear (Möbius) rule $q'=(Aq+B)/(Cq+D)$ . One matrix, computed once, serves both the ray trace and the wave. That single law does the heavy lifting of an entire subject. It locates the waist a laser makes after any lens train, matches a beam into a fibre or external cavity, and — by demanding a mode reproduce its own $q$ after one round trip — selects a resonator's eigenmode and, through the trace of the round-trip matrix, its stability. Throughout, $\det M = n_1/n_2$ runs quietly as an error check on every product.

Key takeaways

Any paraxial element — free space, thin lens, curved mirror, dielectric interface — is a $2\times2$ ray-transfer (ABCD) matrix acting on the ray vector $(y,\theta)^{\mathrm T}$ , and a cascade of elements is a single matrix product taken in reverse of the light path.
Kogelnik's law: the very same matrix propagates a Gaussian beam through the complex beam parameter by the bilinear rule $q'=(Aq+B)/(Cq+D)$ — ray tracing and beam propagation share one object.
The complex parameter $1/q = 1/R - i\lambda/(\pi w^2)$ packs curvature $R$ and spot size $w$ into one number, so one matrix update returns both at once.
For a system that starts and ends in the same medium $\det M = AD-BC = n_1/n_2 = 1$ , a running check that catches ordering and algebra mistakes.
Requiring $q$ to map to itself after a round trip, $q=(Aq+B)/(Cq+D)$ , selects a resonator's eigenmode and reduces to the stability bound $-1\le(A+D)/2\le1$ , i.e. $0\le g_1g_2\le1$ .
Key idea: represent the Gaussian beam by $1/q = 1/R - i\lambda/(\pi w^2)$ ; then the ray matrix that transforms rays also transforms the beam by the single rule $q'=(Aq+B)/(Cq+D)$ .

The core insight

Represent the Gaussian beam by $1/q = 1/R - i\lambda/(\pi w^2)$ ; then the ray matrix $M=\begin{pmatrix}A&B\\C&D\end{pmatrix}$ that transforms rays also transforms the beam by the single bilinear rule $q'=(Aq+B)/(Cq+D)$ .

Where this lives today

Open the design file of any commercial laser and the ABCD law is running underneath. Resonator-design software builds the round-trip matrix from the mirrors, gain rod, Brewster windows, and thermal lens, then reports the eigen- $q$ (mode size) and trace (stability margin) as the designer nudges spacings and radii so the TEM $_{00}$ mode fills the gain without clipping. In ultrafast lasers the same matrix carries an intensity-dependent Kerr lens, and Kerr-lens mode-locking places the cavity near a stability edge where that lens favours the pulsed, high-peak-intensity mode — how mode-locked Ti:sapphire oscillators hand out few-femtosecond pulses. The law is equally the backbone of precision experiments that couple light into cavities. Gravitational-wave detectors such as LIGO mode-match their laser to kilometre-scale arm cavities with ABCD-designed telescopes, because even a small $q$ mismatch scatters power into higher-order modes. The same reasoning launches beams into single-mode fibre, matches a frequency comb to a reference cavity, and shapes the traps of an atom-tweezer array — each a $q$ propagated through a product of $2\times2$ matrices.

The physical picture

Start with rays, because the wave result is smuggled in on their backs. In the paraxial world a ray at any plane is two numbers: its height $y$ and slope $\theta$ (the rigorous invariant uses $n\theta$ ). Every element does one linear thing to that pair. Free space of length $L$ leaves the slope alone but raises the height by $L\theta$ — a shear of phase space. A thin lens leaves the height alone but bends the slope by $-y/f$ — a shear in the orthogonal direction. Curved mirrors, interfaces and slabs are variations on these two moves. Because each is linear, the whole train collapses to one $2\times2$ matrix, and because the elementary matrices conserve the optical invariant, that matrix always has $\det = n_1/n_2$ .

Figure

A single ray of height y and slope θ crosses free space then a thin lens: free space of length L = 0.30 m shears the height by Lθ while a lens of focal length f = 0.20 m kicks the slope by −y/f — the system matrix is the product M_lens · M_space, read right to left, in reverse of the order the light meets them

Now the leap. A Gaussian beam is a wave with diffraction, yet at each plane it too is fixed by two real numbers, its width $w$ and curvature $R$ — precisely the freedom of one complex number $q$ , which then rides the identical matrices as a ray. The intuition: a Gaussian is the analytic continuation of a spherical wave to a complex source position, so promoting the wavefront radius $R$ to the complex $q$ brings the diffraction term $-i\lambda/(\pi w^2)$ along for free. At a waist $R\to\infty$ , so $q = i z_R$ is purely imaginary with $z_R = \pi w_0^2/\lambda$ the Rayleigh range — the cleanest place to begin any calculation. The payoff is visible the moment you focus a laser. A thin lens acts at a point, so it leaves $w$ untouched but adds curvature, $1/R' = 1/R - 1/f$ — exactly $q' = q/(1-q/f)$ from the lens matrix. Feed a collimated beam into a short lens and the ABCD law hands you the focused spot size and location in two lines. The second figure carries that computation: a 1 mm helium–neon beam squeezed to a 20 µm waist.

Figure

Focusing a collimated 1 mm helium–neon beam (λ = 632.8 nm) with a 100 mm lens: q at the lens equals i z_R, and one bilinear update gives a waist w₀′ = 20 µm at z ≈ f = 100 mm with Rayleigh range z_R′ = 2 mm, so the depth of focus 2 z_R′ ≈ 4 mm and the spot is about 50× smaller than the input

The core equations

A paraxial ray is a two-vector and every element is a linear map on it. The definition below is the whole formalism in one line: $A,B,C,D$ read off directly from how an element mixes input height and slope into output height and slope.

\begin{pmatrix}y_2\\ \theta_2\end{pmatrix}=\begin{pmatrix}A&B\\C&D\end{pmatrix}\begin{pmatrix}y_1\\ \theta_1\end{pmatrix} \qquad (1)

where: $y$ — ray height at the reference plane; $\theta$ — ray slope (the rigorous invariant uses $n\theta$ ); $A$ — height-to-height (dimensionless); $B$ — slope-to-height (length); $C$ — height-to-slope (inverse length); $D$ — slope-to-slope (dimensionless). Physically $B=0$ is an imaging condition (output height independent of input slope) and $C=0$ is an afocal, telescope-like system (output slope independent of input height). The three workhorse elements follow from the shear picture: free space slides height along slope, a thin lens kicks slope in proportion to height, a flat interface rescales slope by the index ratio (paraxial Snell).

M_{\text{space}}=\begin{pmatrix}1&L\\0&1\end{pmatrix},\qquad M_{\text{lens}}=\begin{pmatrix}1&0\\-1/f&1\end{pmatrix},\qquad M_{\text{interface}}=\begin{pmatrix}1&0\\0&n_1/n_2\end{pmatrix} \qquad (2)

where: $L$ — propagation distance; $f$ — thin-lens focal length ( $f>0$ converging, $f<0$ diverging); $n_1,n_2$ — indices across the interface. Read these as verbs: space transports, lens bends, interface refracts. The two remaining staples are the curved mirror (a lens of focal length $R_m/2$ ) and the slab (a shortened air gap of reduced thickness $d/n$ ).

M_{\text{mirror}}=\begin{pmatrix}1&0\\-2/R_m&1\end{pmatrix},\qquad M_{\text{slab}}=\begin{pmatrix}1&d/n\\0&1\end{pmatrix} \qquad (3)

where: $R_m$ — mirror radius (concave $R_m>0$ , $f=R_m/2$ ); $d$ — slab thickness; $n$ — slab index. The reduced thickness $d/n$ is why a glass block looks optically thinner — light covers the geometric distance but accrues the phase of a shorter air path. Every elementary matrix conserves the optical invariant, and the bookkeeping of that conservation is a determinant law holding for the whole product — the single most useful sanity check in the subject.

\det M = AD-BC = \frac{n_1}{n_2} \qquad (4)

where: $n_1,n_2$ — indices of the first (input) and last (output) media. In a system beginning and ending in air, $n_1=n_2$ and $\det M = 1$ exactly; any other value flags a dropped term or mis-ordered product. This is the ray-optics face of Liouville's theorem — phase-space area (étendue) is conserved. Now the bridge to waves. Kogelnik's ABCD law states that the identical matrix propagates a Gaussian beam if it is carried by $q$ and updated by a bilinear — not linear — map. It is proved below from a wavefront-slope argument.

q' = \frac{Aq+B}{Cq+D} \qquad (5)

where: $q$ — complex beam parameter entering; $q'$ — the same leaving. The same $A,B,C,D$ that steered the ray now steer the beam, and because $q$ carries an imaginary part, one application transports curvature and spot size together. The trap: $q$ transforms bilinearly, $1/q$ does not transform linearly. Splitting the reciprocal of $q$ into curvature and diffraction is what makes it worth carrying.

\frac{1}{q(z)} = \frac{1}{R(z)} - i\,\frac{\lambda}{\pi w^2(z)} \qquad (6)

where: $R(z)$ — wavefront radius (positive diverging); $w(z)$ — beam radius where the field falls to $1/e$ ; $\lambda$ — wavelength in the medium, $\lambda_0/n$ . The real part is geometry (where the beam seems to come from), the imaginary part diffraction (how tightly it is confined). Near a waist $q$ takes its most transparent form, and from any $q$ the width and curvature are recovered as an imaginary and a real part.

q(z) = (z-z_0) + i z_R,\quad z_R = \frac{\pi w_0^2}{\lambda},\qquad w^2 = -\frac{\lambda}{\pi\,\mathrm{Im}(1/q)},\quad R = \frac{|q|^2}{\mathrm{Re}(q)} \qquad (7)

where: $z_0$ — waist position; $z_R$ — Rayleigh range; $w_0$ — waist radius. At the waist $q=iz_R$ is pure imaginary; free space adds $z-z_0$ to the real part, which is why $q'=q+L$ reproduces the free-beam expansion $w(z)=w_0\sqrt{1+(z/z_R)^2}$ .

Derivation AProof of the ABCD law from the wavefront slope

The goal is to show, without touching a diffraction integral, that the complex parameter must obey $q'=(Aq+B)/(Cq+D)$ whenever rays obey $M$ . Treat a Gaussian as a spherical wave whose radius has been promoted to the complex $q$ , so a ray on its wavefront has slope $\theta=y/q$ ; the ray matrix acts on that ray, and forming the output radius from the transformed height and slope forces the bilinear law. Because $q$ carries an imaginary part, the argument that images curvature also transports the diffracting spot size.

Step 1. Near a reference plane the paraxial field is $\exp[-ik(x^2+y^2)/2q]$ , a spherical wave whose (complex) radius is $q$ . A ray on that wavefront at height $y$ has local slope $\theta = y/q$ . Why: this is the one definition coupling $q$ to a ray — the wavefront radius is $q$ and a ray is perpendicular to the wavefront, so slope is height over radius.

Step 2. Apply the matrix with $\theta_1=y_1/q_1$ : $y_2 = Ay_1+B\theta_1 = (A+B/q_1)\,y_1$ and $\theta_2 = Cy_1+D\theta_1 = (C+D/q_1)\,y_1$ . Why: (1) is the linear map; substituting $\theta_1=y_1/q_1$ expresses both outputs in the single height $y_1$ , which is about to cancel.

Step 3. The output wavefront has radius $q_2 = y_2/\theta_2 = (A+B/q_1)/(C+D/q_1) = (Aq_1+B)/(Cq_1+D)$ ; the factor $y_1$ cancels and the bilinear law falls out. Why: identifying $q_2$ with the height-to-slope ratio reverses Step 1 and turns a linear ray map into a Möbius map on $q$ .

Step 4. Check against free space, $A=D=1$ , $B=L$ , $C=0$ : starting at a waist $q_1=iz_R$ gives $q_2 = L+iz_R$ , exactly the free Gaussian beam whose spot grows as $w(z)=w_0\sqrt{1+(z/z_R)^2}$ . Why: the simplest non-trivial element must reproduce the known diffracting beam, confirming the imaginary part rides along correctly.

Go deeper · Derivation B · 1 minSelf-consistent resonator mode and the stability condition

Derivation BSelf-consistent resonator mode and the stability condition

The goal is to turn "a cavity supports a stable Gaussian mode" into an algebraic test on the round-trip matrix. A true mode reproduces itself after one round trip, so its $q$ is a fixed point of the bilinear map; that condition yields a quadratic whose roots have a finite, real waist only when the matrix trace lies in a bounded window. That window is the stability region, and for two mirrors it collapses to the $g_1g_2$ diagram of aq-optics-resonator-stability-g-parameters.

Step 1. Let $M=\begin{pmatrix}A&B\\C&D\end{pmatrix}$ be the round-trip matrix at a reference plane with $\det M=1$ , and demand $q=(Aq+B)/(Cq+D)$ . Cross-multiplying gives the fixed-point quadratic $Cq^2+(D-A)q-B=0$ . Why: self-consistency defines a mode — after one round trip the beam must look identical, so $q$ is a fixed point of the map, not merely mapped to something similar.

Step 2. Solve for the reciprocal: $1/q = (D-A)/2B \mp (1/B)\sqrt{[(A+D)/2]^2 - 1}$ , using $\det M=1$ to tidy the discriminant. The real part is $1/R$ , the imaginary part $-\lambda/\pi w^2$ . Why: working with $1/q$ splits the answer cleanly into curvature and spot size, which is what we read off.

Step 3. A confined beam needs $\mathrm{Im}(1/q)$ real and non-zero, so the root must be imaginary, giving $\lambda/\pi w^2 = (1/|B|)\sqrt{1-[(A+D)/2]^2}>0$ , real only when $|(A+D)/2|<1$ . Why: a finite spot needs positive $\lambda/\pi w^2$ ; the moment the trace leaves the window the root turns real, $w$ becomes imaginary, and the mode diffracts out of the cavity — instability.

Step 4. For a two-mirror cavity of length $L$ with mirror radii $R_{m,1},R_{m,2}$ , the round-trip trace reduces to $(A+D)/2 = 2g_1g_2-1$ with $g_i = 1 - L/R_{m,i}$ , so $|(A+D)/2|\le1$ becomes the classic $0\le g_1g_2\le1$ . Why: this identity is the bridge from the abstract trace test to the engineering diagram every laser designer keeps on the wall.

Go deeper · Derivation C · 1 minFocusing a Gaussian beam — waist location and size after a lens

Derivation CFocusing a Gaussian beam — waist location and size after a lens

The goal is a closed formula for where a laser actually focuses and how small the spot gets — differing from the textbook lens law by a diffraction correction. The strategy is pure ABCD: write $q$ just before the lens, apply the thin-lens update, propagate an unknown distance, and demand a new waist (real part of $q$ zero). That distance and the surviving imaginary part give the focal shift and focused waist; the geometric result returns only as $z_R\to0$ .

Step 1. Put the input waist $w_0$ a distance $d_1$ before the lens, so at the lens $q_1 = d_1 + iz_R$ with $z_R=\pi w_0^2/\lambda$ . Why: calculations are cleanest referenced to a waist, where $q$ is purely imaginary and free space just adds the real $d_1$ .

Step 2. Pass through the lens with $1/q_2 = 1/q_1 - 1/f$ — spot continuous, curvature gains $-1/f$ . Why: a thin lens acts at a point, so $w$ cannot jump; only the wavefront curvature changes, exactly the shift encoded in $1/f$ .

Step 3. Propagate $d_2$ and require a waist, $\mathrm{Re}(q_2+d_2)=0$ , i.e. $d_2=-\mathrm{Re}(q_2)$ ; the surviving imaginary part is $z_R'=\mathrm{Im}(q_2)$ and $w_0'=\sqrt{\lambda z_R'/\pi}$ . Why: a waist is where $q$ is pure imaginary, so setting the real part to zero locates it and the imaginary part sizes it.

Step 4. The algebra collapses to $d_2 = f + (d_1-f)f^2/[(d_1-f)^2+z_R^2]$ and $w_0' = w_0 f/\sqrt{(d_1-f)^2+z_R^2}$ . As $z_R\to0$ this becomes the thin-lens conjugate law $1/d_1+1/d_2=1/f$ ; the extra $z_R^2$ is the Gaussian focal shift pulling the waist inside the geometric focus. Why: $z_R^2$ is exactly what geometric optics omits, so a real laser never focuses quite at $f$ unless tightly Rayleigh-confined.

Limiting cases

Every familiar result of ray and beam optics is a corner of the ABCD law, and each is a check on the matrix.

Case	Result	Meaning
Geometric limit $\lambda\to0$	$\mathrm{Im}(1/q)\to0,\;\; R'=\dfrac{AR+B}{CR+D}$	The diffraction term vanishes, $q\to R$ becomes real, and the bilinear law reduces to ordinary wavefront imaging — plain lens optics.
Free space, length $L$	$M=\begin{pmatrix}1&L\\0&1\end{pmatrix},\;\; q'=q+L$	$q$ simply advances; from a waist $q_0=iz_R$ one gets $q=z+iz_R$ , reproducing $w(z),R(z)$ of a free Gaussian beam.
Thin lens, focal length $f$	$M=\begin{pmatrix}1&0\\-1/f&1\end{pmatrix},\;\; \dfrac1{q'}=\dfrac1q-\dfrac1f$	Spot continuous, curvature gains $-1/f$ ; $\det M=1$ ; $f>0$ converges, $f<0$ diverges.
Collimated input into a lens	$q=iz_R\;\to\;$ waist at $z=\dfrac{f z_R^2}{f^2+z_R^2}\approx f$	For $z_R\gg f$ (large input beam) the waist sits at the focal plane; for $z_R\lesssim f$ it shifts inward — the Gaussian focal shift.
Marginal stability	$(A+D)/2=\pm1$ , i.e. $g_1g_2=0$ or $1$	The eigen- $q$ discriminant vanishes; the mode is on the edge of the stability window (confocal or concentric/plane limits).

Common trapThe reciprocal 1/q does not transform linearly, and matrix order is the reverse of the light path

Two traps sink most first attempts. First, although $1/q$ is the transparent combination of $R$ and $w$ , it does not transform by a simple sum: only $q$ obeys the clean rule $q'=(Aq+B)/(Cq+D)$ . Writing " $1/q'=1/q$ plus something" is right only for a thin lens (where it reduces to $1/q'=1/q-1/f$ ) and wrong for free space, where the beam diffracts — always transform $q$ and invert at the very end. Second, order runs backwards: if light meets element 1 then element 2, the system matrix is $M_2M_1$ , not $M_1M_2$ , because element 1 acts first and stands rightmost, nearest the input.

Worked examples

1 · Focusing a helium–neon beam to a spot (the concrete case)

A collimated helium–neon laser ( $\lambda = 632.8$ nm), beam radius $w_0 = 1.0$ mm, passes through a thin lens $f = 100$ mm. Find the focused spot size, its location, and the depth of focus.

Step 1. Collimated input means the waist is at the lens, so $q_1 = iz_R$ with $z_R = \pi w_0^2/\lambda = \pi(10^{-3})^2/(632.8\times10^{-9}) = \mathbf{4.96\ m}$ . Since $z_R\gg f$ , the focus will land very near the focal plane.

Step 2. Apply the lens: $1/q_2 = 1/q_1 - 1/f = -10 - i\,0.201\ \mathrm{m^{-1}}$ . Inverting, $q_2 = -0.09996 + i\,0.00201$ m.

Step 3. The new waist lies $d_2 = -\mathrm{Re}(q_2) = \mathbf{99.96\ mm}$ downstream — indistinguishable from $f=100$ mm — with new Rayleigh range $z_R' = \mathrm{Im}(q_2) = 2.0$ mm.

Step 4. The waist is $w_0' = \sqrt{\lambda z_R'/\pi} = \sqrt{(632.8\times10^{-9})(2.0\times10^{-3})/\pi} = \mathbf{20\ \mu m}$ , and the depth of focus is $2z_R' = \mathbf{4.0\ mm}$ — a 50× reduction held over a 4 mm span, the numbers behind every laser-cutting or microscopy focus. Focus at 100 mm, spot w₀′ = 20 µm, depth of focus 4 mm.

2 · A two-lens beam expander, built by matrix product

Two thin lenses $f_1 = 25$ mm, $f_2 = 100$ mm, separated by $d = f_1+f_2 = 125$ mm (a Keplerian telescope). Build the system matrix, check the determinant, find the magnification.

Step 1. Light meets lens 1, the gap, then lens 2, so the matrix is the reverse-order product $M = M_{L2}\,M_{\text{space}}\,M_{L1}$ with $M_{L1}=\begin{pmatrix}1&0\\-40&1\end{pmatrix}$ ( $1/f_1=40\ \mathrm{m^{-1}}$ ), $M_{\text{space}}=\begin{pmatrix}1&0.125\\0&1\end{pmatrix}$ , $M_{L2}=\begin{pmatrix}1&0\\-10&1\end{pmatrix}$ .

Step 2. Multiply right to left: $M_{\text{space}}M_{L1}=\begin{pmatrix}-4&0.125\\-40&1\end{pmatrix}$ , then $M_{L2}$ times that gives $M = \begin{pmatrix}-4 & 0.125\\ 0 & -0.25\end{pmatrix}$ .

Step 3. Determinant check: $\det M = (-4)(-0.25) - (0.125)(0) = \mathbf{1.0}$ — as required in air, so no term was dropped. Note $C=0$ : an afocal system, exactly what "telescope" means.

Step 4. For a collimated input ( $q_1=iz_R$ ) the update gives $q'=(Aq_1+B)/D$ with imaginary part $z_R'=A^2 z_R = 16 z_R$ , so $w_0'^2 = 16 w_0^2$ and the beam is magnified by $|A| = f_2/f_1 = \mathbf{4\times}$ . M has det = 1, C = 0, and magnification f₂/f₁ = 4×.

3 · A symmetric two-mirror cavity: stability and mode size

A Nd:YAG laser ( $\lambda = 1064$ nm) uses two identical concave mirrors of radius $R_m = 1.0$ m separated by $L = 0.50$ m. Is it stable, and what is the mode waist?

Step 1. The $g$ -parameters: $g_1 = g_2 = 1 - L/R_m = 0.5$ , so $g_1g_2 = 0.25$ . Since $0 < 0.25 < 1$ , the cavity is stable.

Step 2. Cross-check via the round-trip trace, $(A+D)/2 = 2g_1g_2 - 1 = -0.5$ ; indeed $|-0.5| < 1$ , consistent with the stability window above.

Step 3. The waist sits at the cavity centre, with Rayleigh range $z_R = \tfrac12\sqrt{L(2R_m - L)} = \tfrac12\sqrt{0.75} = \mathbf{0.433\ m}$ .

Step 4. Hence $w_0 = \sqrt{\lambda z_R/\pi} = \mathbf{0.38\ mm}$ at the centre, and on each mirror ( $L/2 = 0.25$ m from the waist) the spot has grown to $w = w_0\sqrt{1+(L/2z_R)^2} = 0.38\sqrt{1.33} = \mathbf{0.44\ mm}$ . These tell the designer whether the mode fills the gain rod without clipping on the mirror apertures. Stable ( $g_1g_2 = 0.25$ ); waist 0.38 mm, mirror spot 0.44 mm.

Physics of the portrait

The matrix idea for rays is nineteenth-century Gaussian optics, but the leap that made it the daily tool of laser physics — that one matrix propagates a diffracting beam — belongs to the founding era of the laser and to one engineer in particular.

Origins · 1 minHerwig Kogelnik and how this was discovered

Herwig Kogelnik

b. 1932 · co-author of the 1966 theory of laser beams and resonators that unified ray matrices with Gaussian-beam propagation

Herwig Kogelnik, an Austrian-born electrical engineer, arrived at Bell Laboratories in the early 1960s just as the laser was turning from curiosity into instrument, and his 1966 review with Tingye Li, "Laser Beams and Resonators" (Applied Optics 5, 1550), became one of the most cited papers in all of optics. Its lasting gift is the ABCD law: the recognition that the complex beam parameter $q$ , carrying a Gaussian's width and curvature together, propagates through any paraxial system by the same $2\times2$ matrix ray optics had used for a century — updated by a bilinear rather than a linear map. That single observation collapsed beam propagation, mode matching, and resonator design into matrix multiplication and one self-consistency solve, which is why " $q$ -parameter" and "ABCD" are spoken in the same breath today. His reach did not stop there — he co-invented the distributed-feedback laser and laid foundations for holography and coupled-mode theory — but the through-line is the ABCD instinct: find the compact linear structure hiding inside a wave problem, and the engineering falls out.

History

1840

— Gauss formulates paraxial optics and the cardinal-point description of imaging

1961

— Boyd and Gordon solve the confocal resonator modes as Fox and Li compute cavity modes numerically

1964

— Halbach gives the matrix representation of Gaussian paraxial optics

1966

— Kogelnik and Li publish Laser Beams and Resonators, stating the ABCD law and g-parameter stability

1986

— Siegman's Lasers codifies ray-matrix and beam-matrix methods as standard laser-design tools

Going further

Texts. Saleh and Teich, Fundamentals of Photonics, 2nd ed., Ch. 1 (ray optics) and Ch. 3 (beam optics) · Kogelnik and Li, "Laser Beams and Resonators," Applied Optics 5, 1550 (1966) · Siegman, Lasers, Ch. 15–20 (ray matrices, beam and resonator analysis) · Hecht, Optics, 5th ed., §6.2 (matrix methods) · Demtröder, Laser Spectroscopy, Vol. 1, Ch. 5 (Gaussian beams and resonators) · Halbach, "Matrix Representation of Gaussian Optics," Am. J. Phys. 32, 90 (1964). The reduced thickness $d/n$ , the mirror focal length $f=R_m/2$ , and the Rayleigh range $z_R=\pi w_0^2/\lambda$ are standard results. Key consistency checks: each elementary matrix carries the stated determinant; the free-space and thin-lens sub-cases reproduce the standard $w(z),R(z)$ ; the stability derivation reduces to $0\le g_1g_2\le1$ via $(A+D)/2 = 2g_1g_2-1$ ; and the Gaussian-focusing formulas collapse to the thin-lens conjugate equation as $z_R\to0$ . The three worked examples have been numerically confirmed: HeNe focus (20 µm at 100 mm), telescope ( $\det M=1$ , $4\times$ ), and Nd:YAG cavity ( $g_1g_2=0.25$ , waist 0.38 mm).

Prerequisites. Gaussian Beams · Optical Resonators and Cavity Modes · Transverse Cavity Modes (Hermite–Gaussian). Leads to. Resonator Stability and the g-Parameter Diagram · Longitudinal Modes and Free Spectral Range · Cavity Q Factor and Photon Lifetime.

Ask the tutor: Give me your lens spacings, mirror radii, or a beam you want to focus, and I will build the ABCD matrix, run the $q$ -parameter, and tell you exactly where the waist lands and how big it is.

Why does q transform bilinearly while a ray transforms linearly?

Because $q$ is not a ray — it is the ratio of a wavefront ray's height to its slope, $q=y/\theta$ . The matrix maps $(y,\theta)$ linearly, but a ratio of two linear combinations is a Möbius (fractional-linear) transformation, exactly $(Aq+B)/(Cq+D)$ . So the bilinear form is not an extra postulate; it is the automatic consequence of building $q$ as a height-over-slope ratio and letting the matrix act on numerator and denominator separately.

When can I use the ordinary thin-lens equation instead of the full q calculation?

Whenever the beam's Rayleigh range is short compared with every distance, $z_R\ll d_1,d_2,f$ . Then the imaginary part of $q$ is negligible, $q\to R$ becomes a real wavefront radius, and the ABCD law reduces to $1/d_1+1/d_2=1/f$ — pure geometric optics. The correction you drop is the $z_R^2$ term in the focusing formula, the Gaussian focal shift that pulls the waist inside the geometric image plane. For a large collimated beam through a short lens the shift is small; when $z_R$ is comparable to $f$ it can move the focus by tens of percent, and you must keep $q$ complex.

What is the physical meaning of each ABCD entry?

Read them as "what does the output depend on." $A$ (height-to-height) is the transverse magnification when $B=0$ . $B$ (slope-to-height, length) couples input angle into output position; $B=0$ is the imaging condition. $C$ (height-to-slope, inverse length) is the system's power — for a thin lens $-1/f$ — and $C=0$ marks an afocal (telescope) system whose output rays stay parallel. $D$ (slope-to-slope) is the angular magnification. All four are constrained by $\det M = n_1/n_2$ , so only three are independent in air.

The determinant came out to 1 but my answer is still wrong. What else could be off?

The determinant is necessary but not sufficient. A swapped multiplication order can still yield $\det=1$ , since $\det(M_1M_2)=\det(M_2M_1)$ — only the off-diagonal entries change — so it will not flag a reversed product. Sign conventions are the other culprit: a diverging lens needs $f<0$ , a concave mirror has $R_m>0$ with $f=R_m/2$ , and $q$ must use $\lambda$ in the medium, $\lambda_0/n$ , not the vacuum wavelength. When $\det=1$ but the number is wrong, check ordering, signs, and which wavelength you used before suspecting the algebra.

How does the round-trip matrix know about diffraction if it was built from ray optics?

It does not know directly — the matrix is purely geometric. Diffraction enters through the object it acts on: the complex $q$ , whose imaginary part is the term $-i\lambda/\pi w^2$ . Demanding that $q$ reproduce itself after a round trip gives a fixed-point equation with an imaginary part precisely because $q$ has one, and solving it forces a specific finite spot size — the balance between diffraction spreading the beam and the mirrors refocusing it. The ray matrix supplies the focusing bookkeeping, the imaginary part of $q$ supplies the spreading, and self-consistency is where the two meet to fix the mode.

Does the same formalism handle astigmatic or non-round beams?

Yes, with a small extension. A single $q$ assumes circular symmetry, but real systems — a Brewster-cut crystal, an off-axis mirror, cylindrical lenses — focus differently in the two transverse planes. The standard fix carries two independent parameters $q_x$ and $q_y$ , each propagated by the ABCD matrix for its plane (tangential and sagittal matrices differ for tilted elements). For general astigmatism where the axes rotate, one promotes $q$ to a $2\times2$ complex curvature tensor and $M$ to a $4\times4$ ray matrix (Arnaud, Siegman). The bilinear structure survives — it just acts on matrices instead of scalars.