Anti-Reflection and High-Reflection Coatings

Look at a modern camera lens at an angle and you catch a faint purple-green sheen; look at a laser mirror and you see a slab of clear glass that nonetheless throws back 99.99% of one colour. Both are the same trick, run in opposite directions. A stack of transparent dielectric films, each a fraction of a wavelength thick, sets the phase of the light reflected at every interface. Arrange those reflected wavelets to cancel and the surface all but vanishes — an anti-reflection (AR) coating. Arrange them to add and the surface becomes a near-perfect mirror with no metal at all — a high-reflection (HR) coating. The single idea underneath both is the quarter-wave layer. A film whose optical thickness is exactly a quarter wavelength forces the wave returning from its far face to come back half a cycle out of step with the wave reflected from its near face. Whether that half-cycle offset produces cancellation or reinforcement is then just bookkeeping on the reflection signs. For a single AR layer the deepest cancellation needs one more condition — the film index must be the geometric mean of the media it sits between, which is nothing but optical impedance matching, the same physics that stops echoes on a transmission line. Stack many quarter-wave pairs of alternating high and low index and the reflections pile up instead, driving the reflectance toward unity. Everything in this article — invisible eyeglasses, Bragg laser mirrors, the blue of a solar panel, dichroic filters — is a choice about which way to aim those quarter-wave phasors.

Key takeaways

Thin dielectric films control reflection by making the light reflected at successive interfaces interfere: cancel it (anti-reflection) or reinforce it (high-reflection).
A single quarter-wave film ( $n_f d = \lambda_0/4$ ) of index $n_f = \sqrt{n_0 n_s}$ gives exactly zero reflectance at the design wavelength — the film is an optical impedance-matcher.
Real single-layer AR uses MgF₂ ( $n_f = 1.38$ , closest durable index to the ideal $1.23$ ), cutting glass reflection from $4.26\%$ to $1.26\%$ .
A high reflector stacks $N$ high/low quarter-wave pairs; the reflectance climbs toward $1$ as $(n_H/n_L)^{2N}$ — eight TiO₂/SiO₂ pairs reach $\approx 99.95\%$ with no absorption.
A half-wave film is an "absentee" layer: invisible at the design wavelength, useful only for shaping the spectrum.
The condition is sharp: it holds only near $\lambda_0$ and near-normal incidence; off design the reflectance rises and AR-coated optics show their coloured sheen.
Key idea: reflected amplitudes from successive interfaces add as complex phasors, and choosing each layer's optical thickness (usually quarter-wave) and index sets their relative phase and length so the coherent sum is near zero (AR) or near one (HR).

The core insight

Every interface reflects a small phasor whose angle is set by the layer's optical thickness and whose length is set by its Fresnel coefficient. Coating design is the art of aiming those phasors: make them equal and opposite for an anti-reflection surface, or all parallel for a mirror. A quarter-wave film is the universal unit because it supplies exactly the half-cycle of round-trip phase that flips a reflection from adding to cancelling.

Where this lives today

Anti-reflection and high-reflection coatings are among the most-manufactured optical components on Earth. Every consumer camera lens, eyeglass, binocular, and microscope objective carries multilayer AR — a modern zoom lens has twenty or more air–glass surfaces, and without coatings their reflections would sap a third of the light and flood the image with veiling glare and ghosts. The faint magenta-green blush you see on coated optics is the residual reflectance, minimised in the green and rising at the spectrum's ends. Silicon solar cells wear a single AR layer (usually silicon nitride tuned near $600$ nm, giving panels their blue colour) that lifts conversion efficiency by a few percent absolute. At the high-reflectance extreme, dielectric Bragg stacks are the enabling technology of the laser: cavity mirrors above $99.99\%$ , output couplers with a few tenths of a percent transmission, and the ultra-low-loss super-mirrors of reference cavities and the LIGO gravitational-wave detectors, whose coatings reflect all but a few parts per million and whose residual thermal noise now limits the measurement. Combine stacks tuned to different bands and you get the dichroic beam splitters and interference filters at the heart of fluorescence microscopy, laser-line separation, and astronomical photometry.

The physical picture

Start with one film on glass. A little light reflects at the top air–film boundary and a little more at the film–glass boundary underneath. If the film is a quarter-wavelength thick, the second reflection makes a half-wave round trip and returns inverted relative to the first. Make the two reflected amplitudes equal in size as well, and they annihilate: the surface goes black in reflection, and every photon that would have bounced off is instead handed through to the glass.

Figure

Single-layer anti-reflection coating: the air→film reflection and the film→glass reflection return antiphase (extra π from the λ/4 round trip) and, when equal in size, cancel

For a mirror you do the reverse: alternate high- and low-index quarter-wave layers so that every reflected wavelet returns in phase with the first. Each interface adds its little phasor pointing the same way, and after enough pairs the reflected amplitude saturates near unity while the transmitted beam is starved to almost nothing — a dielectric mirror above 99.9% with no lossy metal. The number that matters is the index contrast $n_H/n_L$ raised to the power of the layer count: reflectance approaches $1$ geometrically.

Figure

The punchline: HR reflectance climbs geometrically with the number of quarter-wave pairs, reaching ≈ 99.95% at 8 TiO₂/SiO₂ pairs — while single-layer AR only dips glass from 4.26% to 1.26%

Think in phasors throughout. Each interface contributes a reflected phasor whose angle is fixed by the layer's phase thickness $\delta = 2\pi n d\cos\theta/\lambda$ and whose length is fixed by the Fresnel coefficient there. The hidden piece of bookkeeping is the $\pi$ phase flip on reflection at a low-to-high-index interface (the half-wave loss owned by the Stokes-relations article): without it, a quarter-wave layer would double the reflection rather than erase it.

The core equations

A single bare interface between air and glass already reflects a few percent, and that Fresnel amplitude is the baseline every coating fights against or exploits. At normal incidence it depends only on the two indices.

r = \frac{n_0 - n_s}{n_0 + n_s}, \qquad R = r^2 \qquad (1)

where: $n_0$ — index of the incident medium (air $\approx 1$ ); $n_s$ — index of the substrate; $r$ — amplitude reflection coefficient (its sign carries the phase); $R$ — reflectance (fraction of power). For crown glass, $r = -0.206$ and $R = 4.26\%$ per surface — small, but a multi-element lens has dozens of surfaces. Reflections from two faces of a film interfere only through the phase the wave picks up crossing it. That phase thickness is what a designer actually controls, through the film's index and physical thickness.

\delta = \frac{2\pi}{\lambda}\, n_f d \cos\theta_f \qquad (2)

where: $n_f, d$ — film index and physical thickness; $\theta_f$ — ray angle inside the film; $\lambda$ — free-space wavelength. The quarter-wave film is defined by $\delta = \pi/2$ at the design wavelength $\lambda_0$ , i.e. $n_f d = \lambda_0/4$ , which is the round trip that inverts the second reflection. The deepest single-layer anti-reflection needs two things at once: the quarter-wave thickness (to invert) and an index that equalises the two reflected amplitudes (to cancel). The second condition is the geometric mean.

n_f = \sqrt{n_0\, n_s}, \qquad n_f d = \frac{\lambda_0}{4} \qquad (3)

where: $\lambda_0$ — design wavelength. Read $n_f = \sqrt{n_0 n_s}$ as optical impedance matching: the film's admittance is the geometric mean of air and glass, exactly the intermediate step that suppresses an echo, so no reflected power survives at $\lambda_0$ . For any number of layers, the workhorse is the $2\times2$ characteristic matrix: each homogeneous film maps the tangential fields on one face to those on the other, and multiplying the matrices gives the whole stack.

M_j = \begin{pmatrix} \cos\delta_j & \dfrac{i}{\eta_j}\sin\delta_j \\[6pt] i\,\eta_j\sin\delta_j & \cos\delta_j \end{pmatrix}, \qquad \begin{pmatrix} B \\ C \end{pmatrix} = \Big(\prod_j M_j\Big)\begin{pmatrix} 1 \\ \eta_s \end{pmatrix} \qquad (4)

where: $\delta_j$ — phase thickness of layer $j$ ; $\eta_j$ — optical admittance ( $\eta_j = n_j$ at normal incidence); $B, C$ — normalised tangential $E$ and $H$ at the front face; $\eta_s = n_s$ — substrate admittance. The single matrix product carries all the interference; no beam-by-beam sum is needed. The stack's effective input admittance $Y = C/B$ then plugs straight into the same reflection formula as a bare interface — that is the elegance of the method.

Y = \frac{C}{B}, \qquad r = \frac{\eta_0 - Y}{\eta_0 + Y}, \qquad R = |r|^2 \qquad (5)

where: $\eta_0 = n_0$ — admittance of the incident medium; $Y$ — the admittance the whole coated stack presents to the incoming wave. Anti-reflection means engineering $Y \to \eta_0$ (matched, $R\to 0$ ); high reflection means driving $Y$ far from $\eta_0$ (mismatched, $R\to 1$ ). For a mirror the layers are all quarter-wave, and each one simply inverts the running admittance, so $N$ high/low pairs compound the index ratio geometrically.

R_{\mathrm{HR}} = \left(\frac{\eta_0 - Y_N}{\eta_0 + Y_N}\right)^2, \qquad Y_N = \Big(\frac{n_H}{n_L}\Big)^{2N}\frac{n_H^2}{n_s} \qquad (6)

where: $n_H, n_L$ — high and low film indices; $N$ — number of high/low pairs (here terminated with a high-index layer against both air and substrate); $Y_N$ — the resulting input admittance, which runs away from $\eta_0$ as $N$ grows, forcing $R_{\mathrm{HR}}\to 1$ .

Derivation ASingle-layer anti-reflection as exact optical impedance matching

The phasor sketch says two equal, opposite reflections cancel, but it uses a weak-reflection approximation and hides the multiple bounces inside the film. Here we prove the cancellation is exact — reflectance identically zero, all bounces included — using the characteristic matrix, and in doing so read off the impedance-matching meaning of the geometric-mean index. Take one film of admittance $\eta_f = n_f$ on a substrate $\eta_s = n_s$ in air $\eta_0 = n_0$ , tuned to quarter-wave at $\lambda_0$ .

Step 1. Set $\delta_f = \pi/2$ (quarter-wave), so $\cos\delta_f = 0$ and $\sin\delta_f = 1$ , and the layer matrix collapses to $M_f = \begin{pmatrix} 0 & i/n_f \\ i n_f & 0 \end{pmatrix}$ . Why: at exactly a quarter wave the film has no "cosine" (in-phase) part left — it acts as a pure admittance inverter, which is the whole reason quarter-wave layers are special.

Step 2. Apply it to the substrate vector: $\begin{pmatrix} B \\ C \end{pmatrix} = M_f \begin{pmatrix} 1 \\ n_s \end{pmatrix} = \begin{pmatrix} i\,n_s/n_f \\ i\,n_f \end{pmatrix}$ . Why: the tangential $E$ and $H$ at the outer face are what the incoming wave actually sees; everything about the buried interface is now folded into these two numbers.

Step 3. Form the input admittance $Y = C/B = (i n_f)/(i n_s/n_f) = n_f^2/n_s$ ; the imaginary units cancel and $Y$ is real. Why: a real admittance means the coated surface looks like a lossless resistive match — there is a real index for which it equals air's, and that is where reflection dies.

Step 4. Insert into $r = (n_0 - Y)/(n_0 + Y)$ : reflection vanishes when $n_0 = Y = n_f^2/n_s$ , i.e. $n_f = \sqrt{n_0 n_s}$ . Why: this is impedance matching stated as an index — the film must be the geometric mean of the media it bridges, the optical analogue of a quarter-wave transformer on a transmission line. *Verified symbolically ( $Y = n_f^2/n_s$ and $r=0 \Leftrightarrow n_f=\sqrt{n_0 n_s}$ ) and numerically ( $n_0=1, n_s=1.52 \Rightarrow n_f^{\text{ideal}}=1.2329$ ; MgF₂ $n_f=1.38 \Rightarrow R=1.26\%$ vs bare $4.26\%$ ) in verify_aq-optics-antireflection-high-reflection-coatings.py.*

Go deeper · Derivation B · 1 minHigh reflection: admittance inversion compounded over a quarter-wave stack

Derivation BHigh reflection: admittance inversion compounded over a quarter-wave stack

Now aim the phasors the other way. The goal is to show that a periodic stack of high/low quarter-wave films drives the reflectance toward unity, and to get the closed form for how fast — the geometric growth that lets eight pairs beat any metal mirror. The key is that a quarter-wave layer inverts admittance, so a high/low pair multiplies it by a fixed ratio, and $N$ pairs raise that ratio to the $N$ -th power.

Step 1. From Step 1 above, a single quarter-wave layer of admittance $\eta$ maps a load admittance $Y$ to $\eta^2/Y$ (pure inversion about $\eta$ ). Why: this one fact — quarter-wave = inverter — is the entire mechanism; the rest is arithmetic.

Step 2. Send the substrate admittance $n_s$ through a low then a high layer: first $n_L^2/n_s$ , then $n_H^2/(n_L^2/n_s) = (n_H/n_L)^2\, n_s$ . One high/low pair multiplies the admittance by $(n_H/n_L)^2$ . Why: each pair steps the admittance by the same factor, so the stack is a geometric ladder — the signature of a Bragg structure.

Step 3. After $N$ pairs (with a final high-index layer facing air), $Y_N = (n_H/n_L)^{2N}\, (n_H^2/n_s)$ , which grows without bound as $N$ increases because $n_H > n_L$ . Why: a large admittance mismatch to air is exactly a large reflectance — the coating deliberately makes itself look nothing like the medium the light came from.

Step 4. Feed $Y_N$ into $R = ((n_0 - Y_N)/(n_0 + Y_N))^2$ ; as $Y_N \to \infty$ , $R \to 1$ . A half-wave layer, by contrast, has $\delta = \pi$ so $M = \mathrm{diag}(-1,-1)$ leaves $Y$ unchanged — an "absentee" film, invisible at $\lambda_0$ . Why: this both explains the mirror and warns you that only quarter-wave layers do work at the design wavelength — thickness errors that reach a half wave contribute nothing. *Verified: the $(n_H/n_L)^2$ per-pair inversion and the eight-pair value ( $n_H=2.35, n_L=1.46, n_s=1.52 \Rightarrow R=99.95\%$ ), plus the absentee half-wave identity ( $M=\mathrm{diag}(-1,-1)$ , $Y$ unchanged), in verify_aq-optics-antireflection-high-reflection-coatings.py.*

Limiting cases

Case	Result	Meaning
Ideal single-layer AR at $\lambda_0$	$n_f = \sqrt{n_0 n_s}$ , $n_f d = \lambda_0/4$ $\Rightarrow$ $R = 0$	needs BOTH the geometric-mean index AND the quarter-wave thickness — either alone fails
MgF₂ on crown glass	$n_f = 1.38$ vs ideal $1.23$ $\Rightarrow$ $R = 1.26\%$	closest durable index; residual because $1.38 \neq \sqrt{1.52}$
Half-wave ("absentee") layer	$\delta = \pi$ $\Rightarrow$ $M = \mathrm{diag}(-1,-1)$ , $Y$ unchanged	optically invisible at $\lambda_0$ ; only reshapes the spectrum off-design
$N$ -pair HR stack	$Y_N = (n_H/n_L)^{2N}(n_H^2/n_s) \to \infty$ $\Rightarrow$ $R \to 1$	reflectance rises geometrically with pair count and index contrast
Off-design wavelength / oblique angle	$\delta \neq \pi/2$ ; TE/TM split	AR minimum blue-shifts as $\lambda_0\cos\theta_f$ and lifts off zero — the coloured sheen

Common trap“Any thin transparent film on glass will reduce its reflection”

It will not, and often does the opposite. Anti-reflection needs two conditions met together: the film index close to $\sqrt{n_0 n_s}$ (to equalise the two reflected amplitudes) and an optical thickness of a quarter wave (to invert the second one). Get the thickness wrong by half a wave and you have an "absentee" layer that does nothing at $\lambda_0$ ; get the index wrong — say a high-index film on glass — and the two reflections add, turning your would-be AR coating into a weak mirror. A quarter-wave film of index above the glass raises the reflectance. The coating only subtracts reflection when the phasors are arranged to cancel, and that is a designed coincidence of index and thickness, not a property of "thinness."

Worked examples

1 · Magnesium-fluoride AR on a camera lens (confidence builder)

A crown-glass surface ( $n_s = 1.52$ ) in air is coated with a single quarter-wave MgF₂ film ( $n_f = 1.38$ ) for green light, $\lambda_0 = 550$ nm. Find the film thickness and the residual reflectance, and compare with the bare surface and the ideal film.

Step 1. Physical thickness from the quarter-wave condition: $d = \lambda_0/(4 n_f) = 550\,\text{nm}/(4\times 1.38) = \mathbf{99.6\ nm}$ — about a hundred atoms of MgF₂, deposited to nanometre tolerance.

Step 2. Residual reflectance from the matched-layer formula $R = \big((n_0 n_s - n_f^2)/(n_0 n_s + n_f^2)\big)^2$ with $n_0 n_s = 1.52$ and $n_f^2 = 1.9044$ : $R = ((1.52 - 1.9044)/(1.52 + 1.9044))^2 = (-0.3844/3.4244)^2 = \mathbf{1.26\%}$ .

Step 3. Compare. Bare glass reflects $R_{\text{bare}} = ((1-1.52)/(1+1.52))^2 = \mathbf{4.26\%}$ , so the single MgF₂ layer cuts the per-surface loss by a factor of $3.4$ . The ideal film would be $n_f = \sqrt{1.52} = 1.233$ , giving exactly $0$ — but no hard, stable, adherent coating has that index, which is why MgF₂ (durable at $1.38$ ) is the classic compromise.

Step 4. The residual is why coated optics show a faint purple-green reflection: $1.26\%$ is the minimum at $550$ nm, but it rises toward the red and blue ends, leaving those colours over-represented in the leftover reflection. $d = 99.6$ nm, residual $R = 1.26\%$ versus $4.26\%$ bare — a $3.4\times$ reduction per surface.

2 · An eight-pair dielectric high reflector (exam level)

A laser mirror is built from TiO₂ ( $n_H = 2.35$ ) and SiO₂ ( $n_L = 1.46$ ) quarter-wave layers on fused-silica glass ( $n_s = 1.52$ ) at $\lambda_0 = 550$ nm, using $N = 8$ high/low pairs terminated with high-index layers. Find the layer thicknesses and the reflectance, and hence the transmitted fraction.

Step 1. Quarter-wave thicknesses: $d_H = \lambda_0/(4 n_H) = 550/(4\times 2.35) = \mathbf{58.5\ nm}$ (TiO₂) and $d_L = \lambda_0/(4 n_L) = 550/(4\times 1.46) = \mathbf{94.2\ nm}$ (SiO₂). The stack is $17$ films, about $1.2\ \mu$ m thick in all.

Step 2. Input admittance $Y_8 = (n_H/n_L)^{16}\,(n_H^2/n_s)$ . Here $(2.35/1.46)^{16} = (1.6096)^{16} = 2.03\times 10^{3}$ and $n_H^2/n_s = 5.5225/1.52 = 3.633$ , so $Y_8 = 2.03\times10^3 \times 3.633 = 7.38\times 10^{3}$ — a huge mismatch to air's admittance of $1$ .

Step 3. Reflectance $R = ((1 - Y_8)/(1 + Y_8))^2 = ((1-7380)/(1+7380))^2 = \mathbf{99.95\%}$ , so the transmitted fraction is only $T = 1 - R \approx \mathbf{0.05\%}$ — five parts in ten thousand leak through, with essentially zero absorption because the dielectrics are transparent.

Step 4. Contrast this with a metal mirror: even protected silver absorbs $1$ – $2\%$ per bounce and tarnishes. The dielectric stack beats it and can be pushed further — each added pair multiplies the admittance by $(n_H/n_L)^2 = 2.59$ , so a few more pairs reach the $99.99\%$ + needed for a laser cavity or a gravitational-wave mirror. $d_H = 58.5$ nm, $d_L = 94.2$ nm, $R = 99.95\%$ , $T \approx 0.05\%$ .

Physics of the portrait

The hero here is not a person but a structure — a repeating pair of quarter-wave films — that turns out to be the single most versatile object in thin-film optics, reading as an anti-reflector, a mirror, or a filter depending only on how you terminate and tune it.

Origins · 1 minquarter-wave stack and how this was discovered

quarter-wave stack

periodic dielectric structure · Rayleigh 1886 → Smakula 1935 → Blodgett 1939 · thin-film optics

The quarter-wave stack is the idea that a periodic sequence of transparent films, each a quarter wavelength thick, can engineer reflection almost arbitrarily. Its lineage is a chain of quiet, practical discoveries. In 1886 Lord Rayleigh noticed that tarnished old glass reflected less than fresh glass — a thin surface layer of lower index was accidentally anti-reflecting. In 1904 H. Dennis Taylor deliberately chemically tarnished lenses to exploit it. The modern vacuum-deposited coating was patented by Alexander Smakula at Zeiss in 1935 (kept a military secret for years, since it sharpened optics for rangefinders and periscopes), and in 1939 Katharine Blodgett and Irving Langmuir at General Electric built "invisible glass" from stacked molecular monolayers, the first controllable multilayer AR. Run the same quarter-wave periodicity with a large index contrast and it becomes a Bragg mirror — the reflectance-approaching-unity workhorse of every laser cavity — and put a half-wave spacer between two such mirrors and it becomes a narrowband interference filter. One structure, three instruments, all governed by where you aim the reflected phasors.

History

1886

— Lord Rayleigh reports that tarnished glass reflects less than fresh glass

1904

— H. Dennis Taylor deliberately tarnishes lens surfaces to reduce reflection

1935

— Alexander Smakula (Zeiss) patents vacuum-deposited anti-reflection coatings

1939

— Katharine Blodgett and Irving Langmuir demonstrate multilayer "invisible glass"

1950

s — dielectric Bragg mirrors and interference filters become standard laser and spectroscopy components

Going further

Texts. Hecht, Optics, 5th ed., §9.7–9.9 (thin films and multilayer coatings) · Saleh & Teich, Fundamentals of Photonics, 3rd ed., Ch. 7 (photonic layered media, transfer matrices) · Born & Wolf, Principles of Optics, 7th ed., Ch. 1 (characteristic matrix of stratified media) · Macleod, Thin-Film Optical Filters, 4th ed. · Foot, Atomic Physics, §7 (optical cavities and mirrors).

Provenance. Equations checked against Hecht Optics 5th ed. §9.7–9.9, Born & Wolf 7th ed. Ch. 1 (characteristic matrix), and Saleh & Teich 3rd ed. Ch. 7. The verifier reproduces: (1) the single-layer AR matrix algebra $Y = n_f^2/n_s$ and hence $n_f = \sqrt{n_0 n_s}$ ; (2) $R = 1.26\%$ for MgF₂ ( $n_f = 1.38$ ) on glass ( $n_s = 1.52$ ) via $R = ((n_0 n_s - n_f^2)/(n_0 n_s + n_f^2))^2$ ; (3) bare-glass $R = 4.26\%$ from Fresnel; (4) the eight-pair HR reflectance $99.95\%$ from the admittance recursion; (5) the half-wave absentee identity. All numbers are standard textbook values; no fabricated constants. This article owns the coating application; the underlying two-surface interference is recalled from thin-film interference, and the multi-reflection summation that governs a Bragg stack's sharpness is recalled from multiple-beam interference and the Airy function.

Builds on. Thin-Film Interference · Optical Path Difference and Phase Retardation · Wavefront-Splitting and Amplitude-Splitting Interferometers · Superposition Principle of Electromagnetic Waves. Leads to. Fringes of Equal Inclination and Equal Thickness · Fringe Visibility and Contrast · Temporal Coherence and Coherence Length.

Why is the ideal AR index the geometric mean and not the average of the two indices?

Because anti-reflection needs the two reflected amplitudes to be equal, and the Fresnel amplitude at each interface depends on the index ratio, not the difference. The air–film reflection scales like $(n_0 - n_f)/(n_0 + n_f)$ and the film–glass one like $(n_f - n_s)/(n_f + n_s)$ ; setting their magnitudes equal gives $n_f^2 = n_0 n_s$ , the geometric mean. The transfer-matrix derivation makes it exact: a quarter-wave film presents input admittance $n_f^2/n_s$ , which matches air's admittance $n_0$ precisely when $n_f = \sqrt{n_0 n_s}$ . The arithmetic mean would match differences, but reflection is governed by ratios, so geometry wins.

If a single MgF₂ layer only reaches 1.26%, how do modern lenses get below 0.5%?

By stacking more layers with different indices and thicknesses. A single quarter-wave film has just two free parameters (index and thickness) and one target (zero at $\lambda_0$ ), so it can null reflection only at one wavelength and only if a material of the right index exists. Two- and three-layer designs (V-coats and broadband AR) add degrees of freedom that let a designer flatten the reflectance below $0.5\%$ across the whole visible band, at the cost of a slightly higher minimum. The characteristic-matrix method handles any number of layers in the same formalism — you optimise the layer indices and thicknesses numerically to shape $R(\lambda)$ .

Does the substrate reflection "disappear" in an AR coating, and where does that energy go?

It is not absorbed — it is transmitted. Anti-reflection is an interference effect, not an absorption effect: the reflected wavelets from the two interfaces cancel each other, and because energy is conserved in a lossless dielectric, the light that is no longer reflected must go forward into the glass. That is exactly the point of AR on a camera lens or solar cell: suppressing the reflection increases the transmission. The coating redistributes energy between the reflected and transmitted channels by controlling phase; it does not destroy it.

Why does a dielectric mirror beat a metal one if metals are "more reflective"?

Bare aluminium or silver reflects broadly across the spectrum, which metals do well, but always with a few percent of absorption per bounce and a surface that tarnishes. A dielectric quarter-wave stack is transparent — no free electrons to absorb — so at its design wavelength it can exceed $99.99\%$ reflectance with losses at the parts-per-million level, which is unreachable with metal. The trade-off is bandwidth: the dielectric mirror only works in a stop band around $\lambda_0$ and near-normal incidence, whereas a metal reflects all colours and all angles. You choose dielectric when you need extreme reflectance at one wavelength (a laser), metal when you need broadband (a telescope).

What actually happens to the coating when I tilt it — why the bluish sheen at an angle?

Two things. First, the quarter-wave condition involves $n_f d\cos\theta_f$ , so tilting shortens the effective optical thickness and the reflection minimum shifts to a shorter (bluer) wavelength, roughly $\lambda_{\min}(\theta) = \lambda_0\cos\theta_f$ . Second, at oblique incidence the TE and TM polarizations acquire different Fresnel amplitudes, so the two reflected phasors no longer cancel perfectly and the minimum reflectance lifts off zero. The result is that a coating optimised for green at normal incidence looks increasingly blue-reflecting and brighter as you tilt it — the coloured glare on eyeglasses seen edge-on.

Is the π phase flip really necessary, or just a sign convention?

It is physical and essential. At a low-to-high-index interface the reflected wave flips sign (a $\pi$ phase shift, the "half-wave loss"); at a high-to-low interface it does not. For a film on glass, the top (air→film) reflection flips and the bottom (film→glass) reflection also flips, but the bottom one additionally carries the $\pi$ of the quarter-wave round trip — the net is that the two arrive antiphase and can cancel. If you (wrongly) ignored the reflection-sign bookkeeping, you would predict a quarter-wave layer reinforces the reflection. The sign rule is owned by the Stokes-relations treatment of phase change on reflection, and it is what makes single-layer AR possible at all.