Michelson Interferometer · Guide · glidepathscience.org
GuideMichelson Interferometer
Contents
This Guide explains the ideas and equations behind every Walkthrough step, Concept Challenge question and Predict round, in the order a textbook takes them. Part 1 prepares the Intro track and Part 2 the Advanced track; each takes about a quarter of an hour to read. Each section ends with a worked example that is not one of the sim’s own problems, and the last section of each part takes the same instrument to LIGO.
Part 1 · The essentials
Math you’ll need: algebra and a wave’s phase; no calculus.
One beam, two arms, one screen
In 1881 Albert Michelson built an instrument to measure tiny differences in the time light takes along two paths. It is simple to describe. A beam of light meets a beamsplitter, a glass plate with a thin, half-silvered coating, which reflects half the light and lets half through. The two halves travel at right angles down two arms, reflect from two mirrors, and come back to the beamsplitter, which sends part of each on to a screen. There the two beams overlap and add, as waves do.
Because the two beams come from one source, they keep a fixed phase relation, and the screen shows interference fringes: bright where the two waves arrive in step, dark where a crest of one meets a trough of the other. The fringes are exquisitely sensitive to the arms’ lengths. A change of a fraction of a wavelength of light, a few hundred nanometers, rearranges them. Michelson used the instrument to measure the wavelengths of spectral lines, to measure the standard meter bar in wavelengths of light, and, with Edward Morley, to look for the motion of the Earth through a supposed ether. More than a century later, the same layout with arms four kilometers long detected gravitational waves.
In this sim, as in most lab instruments, one mirror, M1, is fixed and the other, M2, rides on a micrometer screw so it can be moved along its arm. The letter $d$ stands for M2’s offset from the position where the two arms’ light paths are equal.
Two paths from one beam
Everything the instrument does comes from the difference between the two beams’ paths. If you block one arm with a card, the fringes vanish: the screen is lit evenly by the other beam alone, because there is no second wave for it to add to. Fringes are never the image of anything; they are two waves of one light meeting with different histories.
The difference in history is set by the mirrors. Suppose M2 moves a distance $d$ farther from the beamsplitter. Light in that arm travels out to M2 and back again, so it crosses the extra gap twice. The path difference between the arms grows by twice the mirror’s move:
$Δ = 2d$
This round-trip factor of 2 is the first thing to remember about every Michelson measurement, and the most common thing to forget. A move of a quarter of a wavelength changes the path by half a wavelength, enough to take the screen from bright to dark.
Whether the center is bright or dark depends on how many wavelengths fit into the path difference. When $Δ$ is a whole number of wavelengths the waves arrive in step; when it is a whole number plus a half they cancel. With this sim’s beamsplitter, whose coating treats the two beams alike,
bright at the center where $Δ = mλ$, dark where $Δ = (m + ½)λ$
where $m$ is a whole number. A real beamsplitter may give one beam an extra fixed phase: a bare glass surface, by the flip rule for reflections, flips one beam by half a wave and swaps bright for dark. That only changes which mirror position is bright. It never changes the counting that follows, which is why it matters so little in practice. The other output, back toward the laser, always carries the complement: when the screen is dark, all the light returns toward the source. The two ports share the light, and none is destroyed.
Rings, and counting fringes
With a perfectly parallel beam and perfectly aligned mirrors the whole screen would be one fringe, brightening and darkening together as M2 moves. A short lens before the beamsplitter makes the beam diverge instead, so light reaching the screen away from its center crossed the arms at a small angle θ. For that light the path difference is a little less than at the center,
$Δ = 2d\cos θ$
and the screen shows circular fringes: each ring is a cone of light that crossed the arms at the same angle, where $2d\cos θ$ equals a whole number of wavelengths. Watch what happens as M2 moves away from the equal-path position, so $d$ grows. For a ring of a fixed order to keep its path difference, $\cos θ$ must fall, so every ring moves outward, and at the center the path difference passes new whole numbers of wavelengths: new rings are born there and spread. Move M2 back toward equal paths and the rings shrink into the center and vanish there. The direction depends only on whether the paths are becoming more or less unequal, not on which arm is longer. Two ways of watching are easy to confuse: follow one ring, one order, and it moves outward as $|d|$ grows; but count the third ring from the center and you are looking at a newer ring each time, closer in, because new rings keep arriving at the center. Both are true at once.
Each time the center goes from bright to dark and back to bright, the path difference has grown by one wavelength, so the mirror has moved half a wavelength. Counting $N$ fringes past the center for a move $d$ therefore gives
$N = 2d/λ$
This turns the interferometer into a ruler marked in half wavelengths. If you know how far the micrometer moved the mirror, the count gives the wavelength, $λ = 2d/N$; if you know the wavelength, it gives the distance, $d = Nλ/2$. The count is good to about one fringe, so counting more fringes makes either measurement finer: a few hundred fringes measure a wavelength to a fraction of a percent. A motor driving M2 at a steady speed $v$ sends fringes past a detector at a steady rate, $f = 2v/λ$, which is the same factor of 2 per second.
The usual mistake is to count one fringe per wavelength of mirror travel, which halves a measured wavelength or doubles a measured distance. Brightness and beam width never enter: a brighter laser makes brighter fringes, not more of them.
Alignment, white light and the zero
A real interferometer starts misaligned. With a narrow, parallel laser beam you first see two spots on the screen, one from each mirror; if M2 is tilted by an angle α, its reflected beam turns by $2α$, and its spot lands away from M1’s. Turning M2’s tilt screws brings the spots together. Once they overlap, the remaining tilt shows as straight fringes: the two beams now cross at a small angle $2α$, so their path difference changes steadily across the screen, by one wavelength every $λ/(2α)$. Less tilt means wider fringes, until a single fringe fills the overlap; then a diverging beam gives rings, and the last of the tilt shows as rings whose center sits off the middle of the screen.
Fringes in laser light appear over a wide range of arm lengths, because a laser’s light stays coherent over millimeters to meters, depending on the laser. White light is different. Each color in it makes its own fringes, spaced according to its own wavelength. At exactly equal paths every color is bright at once, and the center is white. A little away, red and blue fall out of step and the fringes take on color. A micrometer or two away, the fringes of all the colors overlap at random, each filling in the others’ dark places, and the screen is evenly white. White-light fringes therefore appear only within about a micrometer of equal paths, and that makes them valuable: the white central fringe marks the instrument’s zero to a fraction of a wavelength, a reference no laser can give.
Finding it is the last step of alignment. With rings in laser light, turn M2 toward equal paths and the rings collapse into the center and grow fewer and wider; then switch to white light and creep the fine screw until the colored fringes sweep into view.
Changing the medium: a gas cell
Path difference is really optical path: the distance times the refractive index $n$, because light travels $n$ times slower in a medium and fits $n$ times as many wavelengths into the same length. Put a cell of length ℓ with flat windows in one arm, pump it empty, and let gas in. The cell’s length does not change, but its optical path grows by $(n − 1)ℓ$ each way, $2ℓ(n − 1)$ in all, and fringes stream past the center as it fills:
$N = 2ℓ(n − 1)/λ$
For air at room conditions $n − 1$ is under 3 × 10⁻⁴: the index differs from 1 only in its fourth decimal place. Yet over a few centimeters it amounts to tens of wavelengths, easy to count. The usual slip is to forget that the light crosses the cell twice. The same idea measures the index of any transparent gas, and in Part 2 it becomes a measurement of how the index depends on pressure.
From the bench to LIGO
Everything so far happens on a lab bench. The same instrument, scaled up, made one of the great measurements of physics. The Laser Interferometer Gravitational-Wave Observatory, LIGO, is a Michelson interferometer with arms 4 km long. On 14 September 2015 it recorded the gravitational waves from two black holes spiraling together more than a billion light years away, the event called GW150914 (the discovery paper is cited in the Sources below).
A gravitational wave is a ripple in the geometry of space. Passing straight through the detector from overhead, it stretches distances along one direction across its path while it squeezes them along the direction at right angles, then reverses, many times a second. Describe its size by the strain $h$, the fractional change of length. One arm of length $L$ grows and the other shrinks by
$ΔL = hL/2$
so the arms’ difference changes by $hL$. This holds while the wave’s wavelength, $c/f$, is far longer than the arms: at 100 Hz it is about 3,000 km against LIGO’s 4 km. A wave that stretched both arms alike would leave their difference alone and a Michelson would see nothing; it is the opposite motion of the two arms that the instrument is built to catch.
The strains are fantastically small: GW150914 reached a peak of about 10⁻²¹. That is why the arms are long. The same strain over a longer arm is a larger change of length, and the interferometer measures length changes in units of the light’s wavelength, exactly as on the bench. The output port is held on a dark fringe, so with no wave almost no light reaches the photodetector; a passing wave lets out a little light that rises and falls with the wave. As the two black holes spiral inward their orbit speeds up, and the wave’s frequency and strength rise together toward the merger: a chirp.
Part 2 · Going further
Math you’ll need: trigonometric identities and sums of waves; the Fourier transform is described and used, not derived.
The two-beam intensity and visibility
Treat each beam at the screen as a wave of amplitude $E_k$, with intensity $I_k ∝ E_k^2$. The two waves differ in phase by $δ = 2πΔ/λ$ (plus any fixed phase from the beamsplitter, here zero). Adding them as arrows, phasors of lengths $\sqrt{I_1}$ and $\sqrt{I_2}$ at an angle δ, and squaring the sum gives
$I = I_1 + I_2 + 2\sqrt{I_1 I_2}\cos δ$ $δ = 2πΔ/λ$
The first two terms are what the beams would give if they did not interfere; the last, the cross term, moves light from the dark fringes into the bright ones. When the arms are unequal, the intensity swings between $(\sqrt{I_1} + \sqrt{I_2})^2$ and $(\sqrt{I_1} − \sqrt{I_2})^2$, never reaching zero. The contrast of the fringes is measured by their visibility,
$V = 2\sqrt{I_1 I_2}/(I_1 + I_2)$
which is 1 for equal arms and falls slowly as they become unbalanced. The fringes stay in the same places; only their contrast changes. A common slip is to compute $(I_1 − I_2)/(I_1 + I_2)$, which is a statement about intensities, while fringes are made by amplitudes.
Ring orders, tilt fringes and the compensator
The ring geometry follows from $2d\cos θ = mλ$. At the center the order is $m_0 = 2d/λ$. If M2 is set so the center is bright, $m_0$ is a whole number, and the $p$th bright ring out from it has order $m_0 − p$:
$\cos θ_p = (m_0 − p)λ/(2d)$
For small angles $1 − θ^2/2 ≈ \cos θ$ gives $θ_p ≈ \sqrt{pλ/d}$: the rings crowd outward as $\sqrt p$, and grow as $1/\sqrt d$ when the paths approach equality. A lens of focal length $f$ in front of the screen sends every ray at angle θ to radius $r = f\tan θ$, so the rings can be measured on the screen. If the center is not exactly bright, the first ring is the next whole order below $m_0$, and the fraction $m_0 − ⌊m_0⌋$ shows in the first ring’s size.
A tilted mirror makes the fringes straight. M2 tilted by α returns its beam at $2α$, so the path difference changes across the screen at the rate $2α$ and the fringes are spaced
spacing $λ/(2α)$
(exactly $λ/\sin 2α$). Halve the tilt and they double in width; moving M2 along its arm only slides them sideways.
The beamsplitter’s glass makes one more correction necessary. One beam crosses the plate once, the other three times. For a single wavelength the extra glass is just extra path, and moving M2 cancels it. But glass slows each color by a different amount, so no single mirror position equalizes the paths for every color, and white-light fringes smear out. A compensator, an uncoated plate of the same glass at the same angle in the other arm, gives both beams the same glass, color by color. Without it the white-light fringes do not vanish entirely: they move to where the beams’ group delays match, millimeters away, and spread into a long, faint, colored smear.
Coherence and a doublet’s beats
No real source is perfectly monochromatic. A spectral line of width Δλ is a band of wavelengths, each making its own fringes; as the path difference grows they drift out of step, and the fringes fade. The path difference over which they last is the coherence length,
$ℓ_c = λ^2/Δλ$
a convention rather than a sharp edge. For a line with a Gaussian shape the visibility falls as $\exp[−(πΔ\,Δλ/λ^2)^2/(4\ln 2)]$: to one half at $Δ = 0.44ℓ_c$ and to 2.8 % at $ℓ_c$. Because $ℓ_c$ is a path difference, the mirror can move only half of it, $ℓ_c/2$, from equal paths before the fringes are gone. Lasers have coherence lengths of millimeters to meters, a lamp’s single line millimeters to centimeters, an LED some micrometers, white light about one micrometer.
Two close lines behave differently: their fringe sets fall out of step and back again. Sodium’s yellow light is two lines 0.6 nm apart. As M2 moves, the fringes fade and return, and the period of these beats in mirror travel, $D$, is the move over which the two lines’ path differences differ by one more wavelength:
$Δλ = λ^2/(2D)$
with λ the lines’ mean. If the two lines are equally strong the fringes vanish at the faintest; if one has strengths $a$ and the other $b$, the visibility falls only to $|a − b|/(a + b)$. Here again the factor 2 hides: $D$ is mirror travel, and the path changes twice as much.
The interferogram as a Fourier transform
The fading and beating are clues to a deeper fact. Record the detector at the center while M2 scans; the record against the path difference, the interferogram, is a sum of one cosine for every wavelength in the light, each weighted by its share of the spectrum. A sum of cosines is a Fourier series, so the Fourier transform of the interferogram gives back the spectrum. This is Fourier-transform spectroscopy, the method of most infrared spectrometers.
Two numbers describe such a measurement. The longest path difference recorded, $Δ_{max}$ (twice the mirror’s scan), sets how finely the spectrum is resolved: the transform’s bins are $1/Δ_{max}$ apart in wavenumber $σ = 1/λ$, and since $δσ = δλ/λ^2$,
$δλ = λ^2/Δ_{max}$
The step between samples sets the shortest wavelength recorded: sampling every δΔ of path reaches down to $2δΔ$ (the Nyquist limit), so samples every 100 nm of mirror travel, 200 nm of path, cover the visible down to 400 nm. The bin spacing is a property of the scan, an analysis quantity: it is not an uncertainty on where a line lies, which a strong, clean line can fix far more finely. Padding the record with zeros before the transform draws the spectrum more smoothly but resolves nothing finer than $λ^2/Δ_{max}$.
The index of a gas
For a dilute gas, $n − 1$ is proportional to the number of molecules per unit volume, each polarized a little by the light’s field. At a fixed temperature that makes it proportional to the pressure, so as a cell fills the fringe count grows in a straight line with pressure. Counting fringes against a pressure gauge measures the slope, and with $N$ the count from vacuum to a pressure $P$,
$n − 1 = Nλ/(2ℓ)$
at that pressure. Different gases differ greatly: air bends light about eight times more than helium, whose two electrons are bound tightly and hardly respond to the light’s field. Measuring at several pressures and fitting a line also tests the proportionality itself.
The detector in outline: strain, the dark port, noise and a chirp
Return to LIGO with the tools of Part 2. A strain $h$ changes the arms’ difference by $hL$, and the light crosses that change twice, so a plain Michelson’s output phase shifts by
$Δφ = 4πhL/λ$
in the long-wavelength limit, while the wave is far longer than the arms (and, once the arms are cavities, than the distance the light travels while it is stored in them). For a strain of 10⁻²¹ on arms 2.5 km long it is 2.95 × 10⁻¹¹ rad, far too small to see. Three ideas, here only in outline, make it measurable.
First, the output port is held near a dark fringe. With no wave almost no light reaches the photodetector, so a passing wave’s small change in the arms lets out a little light against almost none. Because the two ports share the light, a dark output sends nearly all of it back toward the laser, and a partly transmitting mirror there, the power-recycling mirror, sends it back in: LIGO’s 20 W laser builds to about 700 W on the beamsplitter. Real detectors sit a little off dark, so the signal is proportional to $h$; exactly at dark it would grow only as $h^2$.
Second, each arm is a Fabry–Pérot cavity: a second mirror near the beamsplitter makes the light bounce back and forth many times before it returns. A cavity of finesse $F$ multiplies the phase by about $2F/π$, as if the arm were that many times longer, until the wave changes faster than the light’s storage time, above a frequency of about $c/(4FL)$.
Third, noise. The light arrives as photons, randomly, and this shot noise sets a strain floor that falls as more photons arrive each second:
$h_{shot} ∝ 1/\sqrt P$
where $P$ is the power in the interferometer (a scaling, here, not a full formula). Ground vibration, seismic noise, walls off the lowest frequencies, below about 10 Hz; the random thermal motion of the mirrors and their suspensions rules the middle of the band; shot noise rules the top, where the cavities’ response falls. The sim’s sensitivity curve shows these shapes with textbook scalings; it is illustrative, not LIGO’s measured noise.
Finally, the signal itself. At leading order in the orbit’s speed, two masses spiraling together radiate a wave whose frequency rises at a rate set by a single combination of the masses, the chirp mass $M_c$:
$\dot f = (96/5)π^{8/3}(GM_c/c^3)^{5/3}f^{11/3}$
Reading the frequency and its rate of change from the data gives $M_c$ directly. The formula holds while the orbit is wide; near the merger the full theory takes over, and the sim’s model chirp simply ends at an illustrative merger frequency.
Reference
Common mistakes
Part 1
- Taking a mirror move $d$ as a path change $d$: the light goes there and back, so $Δ = 2d$.
- Counting one fringe per wavelength of mirror travel: one fringe is $λ/2$ of travel.
- Expecting one arm alone to show fringes, or light from two arms never to cancel: fringes are two waves adding, and a dark fringe sends its light out the other port.
- Thinking a brighter laser or a wider beam gives more fringes: the count is $2d/λ$.
- Expecting white-light fringes everywhere, or blaming their color on the beamsplitter: they appear only within about a micrometer of equal paths.
- Reversing the rings: moving away from equal paths, rings are born at the center and spread.
- Thinking a gravitational wave stretches both arms alike, or that a dark output means no signal: the arms move oppositely, and a dark port is where a small signal shows best.
Part 2
- Taking the half-visibility point, or the mirror travel, for the coherence length: $ℓ_c = λ^2/Δλ$ is a path difference, and the mirror moves half of it.
- Treating a spectrum’s bin spacing as the uncertainty of a line’s wavelength: it is a property of the scan’s length.
The equation sheet
| Part 1 | |
|---|---|
| round trip | $Δ = 2d$ |
| at the center | bright $Δ = mλ$, dark $Δ = (m + ½)λ$ |
| rings | $Δ = 2d\cos θ$ |
| counting | $N = 2d/λ$; $λ = 2d/N$; $d = Nλ/2$ |
| a motor | $f = 2v/λ$ |
| a gas cell | $N = 2ℓ(n − 1)/λ$ |
| strain | $ΔL = hL/2$ |
| Part 2 | |
|---|---|
| two beams | $I = I_1 + I_2 + 2\sqrt{I_1 I_2}\cos δ$, $δ = 2πΔ/λ$ |
| visibility | $V = 2\sqrt{I_1 I_2}/(I_1 + I_2)$ |
| rings | $\cos θ_p = (m_0 − p)λ/(2d)$; $r = f\tan θ$ |
| tilt | spacing $λ/(2α)$ |
| coherence | $ℓ_c = λ^2/Δλ$; $Δλ = λ^2/(2D)$ |
| spectra | $δλ = λ^2/Δ_{max}$ |
| a gas | $n − 1 = Nλ/(2ℓ)$ |
| LIGO | $Δφ = 4πhL/λ$; $h_{shot} ∝ 1/\sqrt P$ |
| a chirp | $\dot f = (96/5)π^{8/3}(GM_c/c^3)^{5/3}f^{11/3}$ |
Symbols: $d$ M2’s offset from equal paths (or a move of it); $Δ$ the path difference; $λ$ the wavelength in vacuum (air is close enough); $m$ an order and $m_0$ the center’s; $N$ a count of fringes; $v$ M2’s speed; $f$ a frequency, or a lens’s focal length in $r = f\tan θ$; $θ$ the angle in the arms; $p$ a ring’s number from the center; $α$ M2’s tilt; $ℓ$ a cell’s length; $n$ an index; $I_1$, $I_2$ the arms’ intensities; $δ$ the phase difference; $V$ the visibility; $ℓ_c$ the coherence length; $D$ a beat period in mirror travel; $Δ_{max}$ the longest path difference recorded; $h$ the strain; $L$ an arm’s length; $P$ the power; $M_c$ the chirp mass. Other books may write some of these with other letters; here each keeps one meaning.
Sources and further reading
- OpenStax, University Physics Volume 3, section 3.5, “The Michelson Interferometer,” CC BY-NC-SA 4.0 (openstax.org): the Intro track follows its order of topics.
- B. P. Abbott et al. (LIGO Scientific Collaboration and Virgo Collaboration), “Observation of Gravitational Waves from a Binary Black Hole Merger,” Physical Review Letters 116, 061102 (2016): LIGO’s arms, laser and powers, and GW150914’s strain, sweep and masses, used here as round values.
- E. Hecht, Optics, for the Michelson, coherence and Fourier-transform spectroscopy; P. R. Saulson, Fundamentals of Interferometric Gravitational Wave Detectors, for the detector in outline.
- This sim’s Field Manual (H) for the model and its limits; Interference & Diffraction for slits, gratings and thin films; the Quantum Eraser Lab for interference one photon at a time.
Michelson Interferometer v1.1 · © 2026 Dr. David Meier · CC BY-NC-SA 4.0 · glidepathscience.org