Group Velocity Dispersion and Pulse Broadening

Every bit that crosses an ocean travels as a pulse of light, and every pulse of light is a package of wavelengths. aq-optics-group-velocity-group-index established that the package's envelope moves at the group velocity $v_g$ , with inverse $\beta_1=n_g/c$ — one number, one arrival time. But $v_g$ is built from a dispersive index, so it disperses too: the red components of the pulse ride a slightly different group velocity than the blue ones, and after enough fibre they no longer arrive together. The pulse spreads. This second-order effect — the frequency dependence of the group velocity itself — is group velocity dispersion (GVD), the quantity that once capped every long-haul optical link. The organizing move is a Taylor expansion of the propagation constant $\beta(\omega)=n(\omega)\omega/c$ about the carrier: $\beta_0$ is phase, $\beta_1$ is group delay, and the new term this article owns is $\beta_2=\mathrm{d}^2\beta/\mathrm{d}\omega^2$ , the GVD coefficient. Engineers quote its wavelength-domain twin, the dispersion parameter $D=-(\lambda/c)\,\mathrm{d}^2n/\mathrm{d}\lambda^2=-(2\pi c/\lambda^2)\beta_2$ , whose sign vocabulary comes from aq-optics-normal-anomalous-dispersion and whose input $\mathrm{d}^2n/\mathrm{d}\lambda^2$ comes straight from a Sellmeier fit (aq-optics-sellmeier-cauchy-equations). Fused silica's material dispersion crosses zero near $1.27\ \mu\mathrm{m}$ and reaches $D=+21.9$ ps/(nm·km) at the 1550 nm telecom wavelength; standard SMF-28 fibre carries $D=17$ ps/(nm·km), so a source of 1 nm linewidth smears by $\Delta\tau=|D|\,L\,\Delta\lambda=1700$ ps — 1.7 ns — over 100 km, seventeen bit slots at 10 Gbit/s. A transform-limited 10 ps pulse holds its shape only for a dispersion length $L_D=T_0^2/|\beta_2|=4.61$ km. Fibre design itself (modes, dispersion management, solitons) belongs to later articles; here we build the broadening machinery those articles spend.

Key takeaways

A pulse is a band of frequencies; because $v_g=v_g(\omega)$ in any real medium, its spectral slices walk apart in time — group velocity dispersion, the second-order term $\beta_2=\mathrm{d}^2\beta/\mathrm{d}\omega^2$ of the expanded propagation constant.
The engineer's form is the dispersion parameter $D=-(\lambda/c)\,\mathrm{d}^2n/\mathrm{d}\lambda^2=-(2\pi c/\lambda^2)\beta_2$ , in ps/(nm·km); a source of linewidth $\Delta\lambda$ broadens by $\Delta\tau=|D|\,L\,\Delta\lambda$ .
Signs: normal GVD means $\beta_2>0$ , $D<0$ (red leads); anomalous GVD means $\beta_2<0$ , $D>0$ (blue leads). Fused silica flips sign at its zero-dispersion wavelength, $\sim1.27\ \mu\mathrm{m}$ for the bulk material, moved to $\sim1.31\ \mu\mathrm{m}$ in standard fibre by waveguide dispersion.
Anchor numbers: silica material $D(1550\ \mathrm{nm})=+21.9$ ps/(nm·km); SMF-28 total $D=17$ ps/(nm·km) $\Leftrightarrow$ $\beta_2=-21.7$ ps²/km; 100 km and 1 nm give $\Delta\tau=1700$ ps $=1.7$ ns.
A transform-limited Gaussian of width $T_0$ keeps its shape for a dispersion length $L_D=T_0^2/|\beta_2|$ — 4.61 km for 10 ps in SMF-28 — then grows as $T(z)=T_0\sqrt{1+(z/L_D)^2}$ , acquiring a linear chirp; the spectrum never changes, so the process is reversible.
Key idea: Taylor-expand $\beta(\omega)$ about the carrier — the second-order coefficient $\beta_2$ (equivalently $D=-2\pi c\beta_2/\lambda^2$ ) makes each spectral component travel at its own group velocity, stretching and chirping every pulse by $\Delta\tau=|D|L\Delta\lambda$ over a scale set by $L_D=T_0^2/|\beta_2|$ .

The core insight

Expand $\beta(\omega)=n(\omega)\omega/c$ about the carrier: $\beta_1=n_g/c$ delays the pulse, and $\beta_2=\mathrm{d}^2\beta/\mathrm{d}\omega^2$ — the group velocity dispersion — spreads it, since each spectral slice rides its own $v_g$ . In wavelength language $D=-(\lambda/c)\,\mathrm{d}^2n/\mathrm{d}\lambda^2$ and the broadening is $\Delta\tau=|D|\,L\,\Delta\lambda$ , significant beyond $L_D=T_0^2/|\beta_2|$ .

Where this lives today

GVD management is a daily engineering discipline. Long-haul fibre links are built as dispersion maps — spans of standard fibre alternated with dispersion-compensating fibre or chirped fibre Bragg gratings of opposite $D$ — although modern coherent receivers increasingly undo the accumulated quadratic phase in digital signal processing, turning a physical problem into arithmetic. In ultrafast laboratories the same $\beta_2$ bookkeeping runs in femtoseconds and millimetres: mode-locked oscillators need net intracavity GVD near zero, so chirped mirrors and prism pairs are specified in fs² of group-delay dispersion, and any pulse sent through a microscope objective is pre-chirped to arrive short at the sample. Chirped-pulse amplification — stretch with one sign of dispersion, amplify safely, recompress with the other — rests entirely on the reversibility of the GVD phase and earned the 2018 Nobel Prize in Physics. And the anomalous-GVD soliton balance underpins supercontinuum sources and the frequency combs of optical clocks — all of it downstream of the one coefficient $\beta_2$ .

The physical picture

Picture the pulse as a bundle of runners, one per spectral component, released together. The group delay $\beta_1$ is the pace they share: the bundle arrives late but tight. GVD is a systematic pace difference across the bundle — in normal GVD ( $\beta_2>0$ ) the red runners are consistently faster, in anomalous GVD ( $\beta_2<0$ ) the blue ones are — so the bundle strings out along the track in wavelength order. Nobody leaves the race: the spectrum is untouched, energy is conserved, and only the arrival schedule changes. That ordering is the chirp: the instantaneous frequency sweeps monotonically from front to back — precisely what a device of opposite $\beta_2$ can unwind to recompress the pulse.

Figure

A transform-limited pulse of width T₀ enters a fibre of length L. Each wavelength rides its own group velocity, so the output envelope is stretched by Δτ = |D|·L·Δλ and chirped: with normal GVD (β₂ > 0, D < 0) the red components lead and the blue trail, and the instantaneous frequency rises across the pulse.

The single number that decides whether any of this matters is the dispersion length $L_D=T_0^2/|\beta_2|$ . It scales with the square of the pulse width: a nanosecond pulse shrugs off kilometres of glass, a 10 ps telecom pulse lasts 4.61 km in SMF-28, and a 100 fs oscillator pulse is reshaped within tens of centimetres of glass. The second figure is the curve every fibre engineer carries in their head — the dispersion parameter of fused silica against wavelength, with its zero crossing and the 1550 nm operating point.

Figure

The punchline — material dispersion D(λ) of fused silica from the Malitson Sellmeier fit: D crosses zero at 1.273 µm and reaches +21.9 ps/(nm·km) at 1550 nm (anomalous GVD, β₂ < 0). Standard SMF-28 fibre (total D = 17 ps/(nm·km), zero shifted to ~1.31 µm by waveguide dispersion) smears a 1 nm source by Δτ = 17×100×1 = 1700 ps = 1.7 ns over 100 km.

The core equations

Everything starts from the propagation constant of a mode of frequency $\omega$ , expanded in a Taylor series about the pulse's carrier $\omega_0$ . Each coefficient has a job: phase, delay, spread.

\beta(\omega)=\frac{n(\omega)\,\omega}{c}=\beta_0+\beta_1(\omega-\omega_0)+\tfrac12\beta_2(\omega-\omega_0)^2+\tfrac16\beta_3(\omega-\omega_0)^3+\cdots \qquad (1)

where: $\beta$ — propagation constant (rad/m); $n(\omega)$ — phase index; $\omega_0$ — carrier frequency; $\beta_0=\beta(\omega_0)$ — carrier phase per length; $\beta_1=\mathrm{d}\beta/\mathrm{d}\omega$ — group delay per length; $\beta_2=\mathrm{d}^2\beta/\mathrm{d}\omega^2$ — the GVD coefficient (s²/m, usually ps²/km); $\beta_3$ — third-order dispersion, relevant only near $\beta_2=0$ or for very short pulses. The first-order coefficient is recalled, not derived: aq-optics-group-velocity-group-index owns the group velocity and group index, and here they simply name the linear term.

\beta_1=\frac{1}{v_g}=\frac{n_g}{c}=\frac{1}{c}\left(n+\omega\frac{\mathrm{d}n}{\mathrm{d}\omega}\right) \qquad (2)

where: $v_g$ — group velocity; $n_g$ — group index. A pulse of one spectral slice arrives after $\beta_1 L$ ; GVD is the statement that $\beta_1$ itself depends on $\omega$ . Differentiating once more expresses the GVD coefficient through the index — the two-derivative structure is why GVD is so much more sensitive to the shape of $n(\omega)$ than the group delay is.

\beta_2=\frac{\mathrm{d}^2\beta}{\mathrm{d}\omega^2}=\frac{1}{c}\left(2\frac{\mathrm{d}n}{\mathrm{d}\omega}+\omega\frac{\mathrm{d}^2n}{\mathrm{d}\omega^2}\right) \qquad (3)

where: $\mathrm{d}n/\mathrm{d}\omega$ , $\mathrm{d}^2n/\mathrm{d}\omega^2$ — slope and curvature of the index in frequency. Normal GVD has $\beta_2>0$ (higher frequencies delayed more), anomalous GVD $\beta_2<0$ . Practitioners work in wavelength, because sources are specified by linewidth in nm. The dispersion parameter $D$ is the wavelength derivative of the group delay, and it inherits a sign flip from $\mathrm{d}\omega/\mathrm{d}\lambda<0$ .

D=\frac{\mathrm{d}\beta_1}{\mathrm{d}\lambda}=-\frac{2\pi c}{\lambda^2}\,\beta_2=-\frac{\lambda}{c}\,\frac{\mathrm{d}^2n}{\mathrm{d}\lambda^2} \qquad (4)

where: $D$ — dispersion parameter, ps/(nm·km); $\lambda$ — vacuum wavelength. The last form takes $\mathrm{d}^2n/\mathrm{d}\lambda^2$ directly from a Sellmeier fit (recall aq-optics-sellmeier-cauchy-equations). Watch units: with $\lambda$ in metres and $\mathrm{d}^2n/\mathrm{d}\lambda^2$ in m⁻², $D$ lands in s/m², and $1\ \mathrm{s/m^2}=10^6$ ps/(nm·km). The working broadening formula follows at once: a source occupying a wavelength band $\Delta\lambda$ accumulates a spread of group delays across a length $L$ .

\Delta\tau=|D|\;L\;\Delta\lambda \qquad (5)

where: $\Delta\tau$ — pulse-broadening (spread of arrival times); $L$ — propagation length; $\Delta\lambda$ — source spectral width. This is the linewidth-dominated regime, the right formula whenever $\Delta\lambda$ far exceeds the pulse's transform-limited bandwidth — the usual case for modulated laser diodes. For a transform-limited pulse the relevant scale is instead set by its own bandwidth $\sim1/T_0$ , giving the dispersion length and the Gaussian broadening law.

L_D=\frac{T_0^2}{|\beta_2|},\qquad T(z)=T_0\sqrt{1+\left(\frac{z}{L_D}\right)^2} \qquad (6)

where: $T_0$ — initial $1/e$ half-width of the field envelope; $L_D$ — the distance at which broadening becomes significant ( $T=\sqrt2\,T_0$ at $z=L_D$ ); for $z\gg L_D$ , $T\to|\beta_2|z/T_0$ , linear in distance and inversely proportional to $T_0$ . GVD adds a quadratic spectral phase only — $|\tilde A(\omega)|$ is preserved, so the broadening is deterministic and reversible.

Derivation AFrom the Taylor-expanded propagation constant to the broadening law

The goal is to show that the second-order Taylor coefficient of $\beta(\omega)$ is precisely what turns a spread of frequencies into a spread of arrival times, and to derive the two working results — $\Delta\tau=|D|L\Delta\lambda$ for a broadband source and the scale $L_D=T_0^2/|\beta_2|$ for a transform-limited pulse. We recall the envelope result by name and differentiate the group delay across the spectrum.

Step 1. Recall from aq-optics-group-velocity-group-index that a narrowband packet centred at $\omega_0$ has its envelope travel at $v_g=(\mathrm{d}\beta/\mathrm{d}\omega)^{-1}$ , so its arrival time over length $L$ is the group delay $\tau(\omega_0)=\beta_1(\omega_0)\,L$ . Why: this is the owned first-order result — one frequency slice, one delay; GVD begins where this delay starts depending on which slice you ask about.

Step 2. Expand $\beta(\omega)$ about $\omega_0$ as in equation (1). Each spectral slice at offset $\Omega=\omega-\omega_0$ has its own inverse group velocity $\beta_1(\omega_0+\Omega)=\beta_1+\beta_2\Omega+\cdots$ , so its delay is $\tau(\Omega)=(\beta_1+\beta_2\Omega)L$ . Why: truncating at second order is the whole GVD approximation — the delay becomes linear across the spectrum, which is what produces a clean, invertible linear chirp rather than distortion.

Step 3. The spread of delays across a source of bandwidth $\Delta\omega$ is $\Delta\tau=|\beta_2|\,L\,\Delta\omega$ . Why: the constant $\beta_1 L$ shifts every slice equally and cannot broaden anything; only the differential term survives the subtraction — broadening is a second-order effect.

Step 4. Convert to wavelength. From $\omega=2\pi c/\lambda$ , $\Delta\omega=(2\pi c/\lambda^2)\Delta\lambda$ , so $\Delta\tau=|\beta_2|(2\pi c/\lambda^2)L\,\Delta\lambda=|D|\,L\,\Delta\lambda$ using $D=-(2\pi c/\lambda^2)\beta_2$ . Why: sources are specified in nm of linewidth, so this is the form used at the bench; the modulus hides the sign because a spread is a spread whichever colour leads.

Step 5. For a transform-limited pulse there is no independent $\Delta\lambda$ : its bandwidth is $\Delta\omega\sim1/T_0$ . Setting the accumulated spread equal to the original width, $|\beta_2|\,z\,(1/T_0)\sim T_0$ , defines the distance $z=L_D=T_0^2/|\beta_2|$ at which broadening becomes order one; the exact Gaussian solution gives $T(z)=T_0\sqrt{1+(z/L_D)^2}$ , with $T(L_D)=\sqrt2\,T_0$ . Why: the $T_0^2$ scaling is the punchline — halving the pulse width doubles its bandwidth and halves its tolerance, quartering $L_D$ . *Verified: $\Delta\tau=|D|L\Delta\lambda=1700$ ps for SMF-28 (17 ps/(nm·km), 100 km, 1 nm), $L_D=4.61$ km for $T_0=10$ ps, and $T(L_D)=\sqrt2 T_0$ in verify_aq-optics-group-velocity-dispersion-pulse-broadening.py.*

Go deeper · Derivation B · 2 minThe D to beta-2 conversion and the zero-dispersion wavelength

Derivation BThe D to beta-2 conversion and the zero-dispersion wavelength

The aim is to connect the two currencies of dispersion — the physicist's $\beta_2$ in frequency and the engineer's $D$ in wavelength — including the unit conversion that trips real calculations, and to locate the zero-dispersion wavelength where fused silica's $\mathrm{d}^2n/\mathrm{d}\lambda^2$ changes sign. We differentiate the group delay in wavelength, chain-rule to frequency, and evaluate on the Malitson Sellmeier fit.

Step 1. Define $D\equiv\mathrm{d}\beta_1/\mathrm{d}\lambda$ , the change of group delay per unit length per unit wavelength. Write $\beta_1=n_g/c=(n-\lambda\,\mathrm{d}n/\mathrm{d}\lambda)/c$ , the group index recalled from aq-optics-group-velocity-group-index. Why: $D$ is defined directly in the measured variables — nanometres of source width, kilometres of fibre — which is why data sheets quote it instead of $\beta_2$ .

Step 2. Differentiate: $\mathrm{d}\beta_1/\mathrm{d}\lambda=\tfrac1c\bigl(n'-n'-\lambda n''\bigr)=-\tfrac{\lambda}{c}\,n''$ , where primes are $\lambda$ -derivatives. Hence $D=-(\lambda/c)\,\mathrm{d}^2n/\mathrm{d}\lambda^2$ : the first derivatives cancel identically. Why: the cancellation says group delay dispersion cares only about the curvature of $n(\lambda)$ — a straight-line index, however steep, broadens nothing.

Step 3. Chain-rule to frequency: $D=(\mathrm{d}\beta_1/\mathrm{d}\omega)(\mathrm{d}\omega/\mathrm{d}\lambda)=\beta_2\times(-2\pi c/\lambda^2)$ , so $D=-(2\pi c/\lambda^2)\beta_2$ . Why: the Jacobian is negative, so $D$ and $\beta_2$ always carry opposite signs — the single most common sign error in the field: "normal" is $\beta_2>0$ but $D<0$ .

Step 4. Mind the units. From the Malitson Sellmeier fit for fused silica, $\mathrm{d}^2n/\mathrm{d}\lambda^2=-0.004238\ \mu\mathrm{m}^{-2}=-4.238\times10^9\ \mathrm{m}^{-2}$ at $\lambda=1.55\ \mu\mathrm{m}$ . Then $D=-(1.55\times10^{-6}/2.998\times10^8)\times(-4.238\times10^9)=+2.19\times10^{-5}\ \mathrm{s/m^2}$ , and since $1\ \mathrm{s/m^2}=10^6$ ps/(nm·km), $D=+21.9$ ps/(nm·km). Why: the ×10⁶ conversion (1 ps per nm per km $=10^{-12}\,\mathrm{s}/(10^{-9}\,\mathrm{m}\times10^{3}\,\mathrm{m})$ ) is where hand calculations most often lose three orders of magnitude.

Step 5. Locate the zero. $D=0$ exactly where $\mathrm{d}^2n/\mathrm{d}\lambda^2=0$ ; on the Malitson fit the curvature changes sign at $\lambda=1.273\ \mu\mathrm{m}$ — the material zero-dispersion wavelength. Below it silica is normal ( $D<0$ , $\beta_2>0$ ); above it anomalous ( $D>0$ , $\beta_2<0$ ). In a real fibre a waveguide contribution adds to the material term and shifts the total zero to $\sim1.31\ \mu\mathrm{m}$ — a fact of fibre design recalled here by name only. Why: the existence of a sign change is generic — $n(\lambda)$ decelerates its fall between the UV and IR bands — and it hands engineers a knob: operate near the zero, or exploit the sign for compensation and solitons. *Verified: $D=-(\lambda/c)n''=-(2\pi c/\lambda^2)\beta_2$ symbolically, $D(1550)=+21.9$ ps/(nm·km) with the explicit ×10⁶ conversion, and the 1.273 µm zero in verify_aq-optics-group-velocity-dispersion-pulse-broadening.py.*

Limiting cases

Case	Result	Meaning
Short distance, $z\ll L_D$	$T\approx T_0\bigl[1+\tfrac12(z/L_D)^2\bigr]$	broadening is quadratic and negligible; the pulse just translates at $v_g$
One dispersion length, $z=L_D$	$T=\sqrt2\,T_0$	the canonical "significant broadening" milestone
Long distance, $z\gg L_D$	$T\approx\|\beta_2\|z/T_0$	width grows linearly with $z$ and inversely with $T_0$ — shorter pulses fare worse
Monochromatic limit, $\Delta\lambda\to0$	$\Delta\tau\to0$ (floor at transform limit)	a single frequency cannot disperse; real pulses always carry bandwidth $\gtrsim1/T_0$
Zero-dispersion point, $\beta_2=0$	third-order $\beta_3$ dominates	at $\sim1.27\ \mu\mathrm{m}$ (material) the quadratic term dies; pulses distort asymmetrically instead
Anomalous GVD + Kerr nonlinearity	soliton possible when $L_D=L_{NL}$	GVD chirp cancels self-phase-modulation chirp — owned by the future optical-solitons article

Common trap“Anomalous GVD at 1550 nm means the fibre is operating inside an absorption band.”

It does not — and this is the vocabulary trap of the whole chapter. Anomalous dispersion in the sense of aq-optics-normal-anomalous-dispersion means $\mathrm{d}n/\mathrm{d}\omega<0$ , the index slope reversing inside an absorption resonance. Anomalous GVD means only $\beta_2<0$ (equivalently $D>0$ ) — a statement about the second derivative, the curvature of $\beta(\omega)$ . Fused silica at 1550 nm is exquisitely transparent and its index slope is perfectly normal ( $\mathrm{d}n/\mathrm{d}\lambda<0$ ); what changes sign at the zero-dispersion wavelength is $\mathrm{d}^2n/\mathrm{d}\lambda^2$ , not $\mathrm{d}n/\mathrm{d}\lambda$ . The two "anomalous" regimes are different orders of derivative and different physics. Related: a large group index does not imply strong broadening — $n_g$ delays the whole pulse uniformly; only the dispersion of $n_g$ (i.e. $\beta_2$ ) broadens it. And keep the signs straight: normal GVD is $\beta_2>0$ and $D<0$ ; the pair never share a sign, because $\mathrm{d}\omega/\mathrm{d}\lambda<0$ .

Worked examples

1 · The bit-rate limit of 100 km of standard fibre (the real number)

SMF-28 fibre at 1550 nm has $D=17$ ps/(nm·km). A transmitter with 1 nm spectral width drives a 100 km link. Find the pulse broadening and the equivalent $\beta_2$ .

Step 1. Apply equation (5) in its native units: $\Delta\tau=|D|\,L\,\Delta\lambda=17\ \mathrm{ps/(nm\cdot km)}\times100\ \mathrm{km}\times1\ \mathrm{nm}=\mathbf{1700\ ps=1.7\ ns}$ . No unit gymnastics needed — this is exactly why $D$ is quoted in ps/(nm·km).

Step 2. Read the consequence. At 10 Gbit/s a bit slot is 100 ps; a 1.7 ns smear spreads each pulse across seventeen slots, obliterating the data. Hence dispersion compensation (or narrow-linewidth sources and coherent receivers) is mandatory on such links — the number that shaped fibre-system architecture.

Step 3. Convert to $\beta_2=-D\lambda^2/(2\pi c)$ . In SI, $D=17\ \mathrm{ps/(nm\cdot km)}=1.7\times10^{-5}\ \mathrm{s/m^2}$ , $\lambda^2=(1.55\times10^{-6})^2=2.4025\times10^{-12}\ \mathrm{m^2}$ , $2\pi c=1.884\times10^9$ m/s. So $\beta_2=-(1.7\times10^{-5})(2.4025\times10^{-12})/(1.884\times10^9)=-2.17\times10^{-26}\ \mathrm{s^2/m}=\mathbf{-21.7\ ps^2/km}$ . Negative: 1550 nm sits in the anomalous-GVD regime of silica. Δτ = 1700 ps = 1.7 ns; β₂ = −21.7 ps²/km.

2 · Dispersion length of a 10 ps pulse, and material versus total dispersion

A transform-limited Gaussian pulse with $T_0=10$ ps propagates in the same fibre ( $\beta_2=-21.7$ ps²/km). Find $L_D$ , the width after 100 km, and compare the fibre's total $D$ with the bulk-silica material value.

Step 1. Dispersion length: $L_D=T_0^2/|\beta_2|=(10\ \mathrm{ps})^2/(21.7\ \mathrm{ps^2/km})=100/21.7=\mathbf{4.61\ km}$ . After only 4.61 km the pulse is already $\sqrt2\times10=14.1$ ps wide.

Step 2. A 100 km link is $z/L_D=21.7$ dispersion lengths, deep in the linear regime, so $T\approx T_0\,z/L_D=10\times21.7=\mathbf{217\ ps}$ . (Note this transform-limited pulse fares far better than the 1 nm source of example 1 — its bandwidth is only $\sim0.2$ nm; equation (5) with the actual bandwidth and equation (6) tell one consistent story.)

Step 3. Material versus total. Bulk fused silica has material $D(1550)=+21.9$ ps/(nm·km) (derivation B), but SMF-28's measured total is $+17$ : the difference, $17-21.9=-4.9$ ps/(nm·km), is the waveguide dispersion of the guided mode, which also drags the zero from the material's 1.273 µm to the fibre's $\sim1.31\ \mu\mathrm{m}$ . How that waveguide term arises is owned by the future fibre-optics articles — here it is a named fact. L_D = 4.61 km; T(100 km) ≈ 217 ps; waveguide contribution ≈ −4.9 ps/(nm·km).

Physics of the portrait

Group velocity dispersion is the rare concept whose portrait is a coefficient: $\beta_2$ , the second derivative that decides whether a pulse survives its journey.

Origins · 1 mingroup velocity dispersion and how this was discovered

group velocity dispersion

β₂ = d²β/dω², the second-order dispersion coefficient that broadens and chirps every pulse · fibre and ultrafast optics

Group velocity dispersion is what the twentieth century found waiting one derivative beyond Rayleigh. The group velocity itself was nineteenth-century physics — Hamilton, then Rayleigh — but as long as light meant lamps and mirrors, nobody cared that $v_g$ depended on frequency. The laser and the optical fibre changed that within a decade of each other: once information rode picosecond pulses down kilometres of glass, the dispersion of the group velocity became the resource-limiting quantity of a global industry. The formalism crystallized in the fibre era — expand $\beta(\omega)$ , keep $\beta_2$ , quote its wavelength twin $D=-(2\pi c/\lambda^2)\beta_2$ — and the numbers became household figures in photonics: $+17$ ps/(nm·km) at 1550 nm, zero near 1.31 µm, $-21.7$ ps²/km. What makes $\beta_2$ deeper than a nuisance parameter is its sign. Hasegawa and Tappert saw in 1973 that anomalous GVD ( $\beta_2<0$ ) could balance the Kerr nonlinearity to make solitons — pulses that refuse to broaden — and the same quadratic spectral phase, being deterministic and reversible, underlies chirped-pulse amplification, pulse compressors, and dispersion-managed links. GVD is the textbook example of physics at second order: invisible in any single measurement of arrival time, yet compounding along every kilometre until it either destroys the signal or, harnessed, carries it intact across an ocean.

History

1965

— Malitson publishes the fused-silica Sellmeier fit still used to compute material dispersion

1966

— Kao and Hockham propose low-loss silica fibre for optical communication

1970

— Kapron, Keck and Maurer at Corning reach 20 dB/km, making fibre links practical

1973

— Hasegawa and Tappert predict optical solitons balancing anomalous GVD against the Kerr nonlinearity

1979

— Miya and colleagues hit 0.2 dB/km at 1550 nm, fixing the telecom window where D = +17 ps/(nm·km)

1980

— Mollenauer, Stolen and Gordon observe fibre solitons, confirming the sign-of-β₂ physics

Going further

Texts. Agrawal, Nonlinear Fiber Optics, 5th ed., Ch. 3 (GVD, dispersion parameter) · Saleh & Teich, Fundamentals of Photonics, 3rd ed., Ch. 5 (pulsed optics) and Ch. 24 (fiber optics) · Hecht, Optics, 5th ed., §7.2 (dispersion, group velocity) · Loudon, The Quantum Theory of Light, 3rd ed., Ch. 1 · Demtröder, Laser Spectroscopy, Vol. 1, Ch. 6 (short-pulse propagation).

Provenance. Equations for $\beta_1$ , $\beta_2$ , $L_D$ , Gaussian broadening and the $D$ – $\beta_2$ conversion follow Agrawal (Nonlinear Fiber Optics, Ch. 3), Saleh & Teich (Ch. 5, 24) and Loudon (Ch. 1); a verifier should recompute $\beta_2$ by twice differentiating $\beta=n\omega/c$ , confirm the Gaussian broadening via the complex-width propagator, and check the standard silica values ( $D\approx17\,\mathrm{ps\,nm^{-1}\,km^{-1}}$ , $\beta_2\approx-21.7\,\mathrm{ps^2\,km^{-1}}$ at $1.55\,\mu\mathrm{m}$ ; zero-dispersion near $1.31\,\mu\mathrm{m}$ ) against Agrawal Table 3.1. The group velocity and group index are recalled by name from aq-optics-group-velocity-group-index, the sign vocabulary from aq-optics-normal-anomalous-dispersion, and the Sellmeier machinery for $\mathrm{d}^2n/\mathrm{d}\lambda^2$ from aq-optics-sellmeier-cauchy-equations; the material curve uses Malitson's fused-silica coefficients (material $D(1550)=+21.9$ ps/(nm·km), material zero at 1.273 µm). No non-standard numbers introduced.

Prerequisites. Group Velocity and Group Index · Phase Velocity · The Refractive Index · Dispersion: Normal and Anomalous · Sellmeier and Cauchy Dispersion Equations. Leads to. Chirped-Pulse Amplification · Optical Solitons · Dispersion Compensation and Management · Frequency Combs and Ultrafast Metrology.

Why do shorter pulses broaden so much faster?

Because bandwidth and duration are Fourier conjugates. A transform-limited pulse of width $T_0$ necessarily spans a frequency band $\Delta\omega\sim1/T_0$ ; halve $T_0$ and you double the band of group velocities the pulse samples, while simultaneously halving the width it must exceed to count as "broadened". The two effects multiply, giving $L_D=T_0^2/|\beta_2|$ — quadratic, not linear. Concretely: a 10 ps pulse in SMF-28 lasts 4.61 km, a 1 ps pulse only 46 m, and a 100 fs pulse about 46 cm. This is why femtosecond laboratories obsess over centimetres of glass while telecom systems worry over kilometres of fibre.

Does GVD destroy information or energy?

Neither. GVD is a purely dispersive, lossless operation: it multiplies the spectrum by a quadratic phase $e^{i\beta_2\Omega^2 z/2}$ and leaves $|\tilde A(\Omega)|$ — hence the energy and the spectral content — untouched. The broadening is a rearrangement of energy in time, deterministic and exactly invertible: pass the chirped pulse through an element of opposite $\beta_2$ (a grating pair, a chirped mirror, a compensating fibre) and the original pulse re-emerges. Information is lost only operationally, when overlapping pulses confuse a receiver that reads arrival times; a coherent receiver that measures the field can undo the known phase digitally.

Is anomalous GVD the same as the anomalous dispersion near an absorption line?

No — same adjective, different derivative. Anomalous dispersion (owned by the normal/anomalous-dispersion article) is $\mathrm{d}n/\mathrm{d}\omega<0$ , a first-derivative reversal that happens only inside an absorption band. Anomalous GVD is $\beta_2<0$ , a second-derivative statement that occurs in perfectly transparent material — silica beyond 1.27 µm — simply because the curvature of $n(\lambda)$ changes sign between the UV and IR resonances. At 1550 nm silica shows normal first-order dispersion and anomalous second-order dispersion at once. Reading one for the other is the commonest way to get the physics of the telecom window wrong.

Why do engineers quote D in ps/(nm·km) instead of just using β₂?

Because $D$ 's units match the measurables. A system designer knows the fibre length in km and the source linewidth in nm; multiplying by $D$ in ps/(nm·km) gives the broadening in ps with no conversions — equation (5) at a glance, as in $17\times100\times1=1700$ ps. $\beta_2$ in ps²/km is the natural theory variable (it sits in the envelope equation and in $L_D=T_0^2/|\beta_2|$ ), so both circulate, linked by $D=-(2\pi c/\lambda^2)\beta_2$ . The price of the convenience is the opposite-sign convention, which must be re-checked every time: positive $D$ is anomalous GVD, negative $\beta_2$ .

If β₂ = 0 at the zero-dispersion wavelength, why not run every link there?

Two reasons. First, $\beta_2=0$ does not end dispersion — it promotes the next Taylor term, the third-order $\beta_3$ , which broadens pulses asymmetrically with oscillatory tails and cannot be compensated by a simple opposite-sign element. Second, near-zero local dispersion is actually harmful in multi-channel (WDM) systems: with all channels travelling in lock-step, nonlinear four-wave mixing between them phase-matches efficiently and cross-talk explodes. Practical links prefer finite local dispersion to de-phase the nonlinearities, then cancel the total elsewhere — the art of dispersion management, owned by a later article.

Where does the chirp physically come from?

From time-ordering by group velocity. After a length $z$ , the slice at frequency offset $\Omega$ arrives delayed by $\beta_2\Omega z$ relative to the carrier. With $\beta_2>0$ the low-frequency (red) slices arrive early and the high-frequency (blue) slices late, so an observer at fixed $z$ sees the instantaneous frequency rise linearly through the pulse — an up-chirp; $\beta_2<0$ gives the mirror image. The chirp is not an added modulation but the bookkeeping of who arrived when, and its linearity is exactly what makes it removable by one quadratic phase of the opposite sign.