Relativistic Beam Dynamics in Accelerators

A proton at the top of the LHC's ramp is travelling at $0.999\,999\,990\,5$ of the speed of light. Give it $1\%$ more momentum — a colossal $68$ GeV, more energy than the entire Tevatron beam — and its speed increases by $1.9$ parts in $10^{10}$ . The velocity is frozen. And once you accept that, almost every design decision in a modern accelerator follows, because a machine cannot steer a beam by its speed if the speed will not move. It must steer it by geometry — by how long a path the particle takes. That substitution is the whole of relativistic beam dynamics. It is why a particle in a synchrotron has two different masses — $\gamma m$ when you push it sideways and $\gamma^3 m$ when you push it forwards. It is why an accelerator's bunches hold themselves together instead of smearing out. And it is why every proton machine has one energy at which it briefly goes blind — the transition energy, where the frozen-velocity term and the path-length term cancel exactly, longitudinal focusing vanishes, and the radio-frequency system must jump its phase by $180°$ in a few milliseconds or lose the beam.

Key takeaways

Signature $(+,-,-,-)$ : $\mathrm{d}s^2 = c^2\mathrm{d}t^2 - \mathrm{d}x^2$ , $p\!\cdot\!p = +m^2c^2$ , timelike positive.
Relativistic inertia is anisotropic: $\mathbf{F} = \gamma m\,\mathbf{a} + \gamma^3 m\,(\boldsymbol\beta\!\cdot\!\mathbf{a})\boldsymbol\beta$ . Push across the motion and the particle's inertia is $\gamma m$ ; push along it and the inertia is $\gamma^3 m$ . At high $\gamma$ it is $\gamma^2$ times harder to change a particle's speed than its direction.
Consequence, and it is the article's spine: the velocity freezes. $\dfrac{\mathrm{d}\beta}{\beta} = \dfrac{1}{\gamma^2}\dfrac{\mathrm{d}p}{p}$ . At the LHC's $\gamma = 7247$ , a $1\%$ momentum kick moves the speed by $1.9\times10^{-10}$ .
Steering is therefore all rigidity: $p[\mathrm{GeV}/c] = 0.299792\,B[\mathrm{T}]\,\rho[\mathrm{m}]$ . Focusing is all geometry: alternating quadrupoles, net-focusing by $1/f^* = L/f^2 > 0$ , betatron oscillations obeying Hill's equation.
Owned result: the transition energy is where relativity hands over to geometry. The phase-slip factor $\eta = \alpha_c - 1/\gamma^2$ is a race between a geometric term ( $\alpha_c$ : more momentum ⟹ longer orbit ⟹ arrives late) and a purely relativistic term ( $1/\gamma^2$ : more momentum ⟹ faster ⟹ arrives early). They cancel at $\boxed{\gamma_t = 1/\sqrt{\alpha_c}}$ , where longitudinal focusing vanishes and the stable RF phase flips from $\phi_s$ to $\pi - \phi_s$ . The CERN PS ( $\gamma_t = 6.1$ ) must cross it every cycle; the LHC ( $\gamma_t = 53.8$ , injecting at $\gamma = 480$ ) was designed never to.
Acceleration shrinks the beam for free: normalized emittance $\varepsilon_N = \beta\gamma\,\varepsilon$ is invariant, so $\varepsilon \propto 1/\beta\gamma$ . The LHC's beam shrinks by $\gamma$ -ratio $15$ from injection to collision with no cooling at all, reaching a $\mathbf{10\ \mu m}$ spot.
Key idea: at $\beta \to 1$ you cannot control a beam through its velocity, because its velocity will not move. Everything an accelerator does, it does through path length and rigidity.

The core insight

Newton's second law has one mass in it. Relativity's has two — and they differ by $\gamma^2$ , which at the LHC is $5\times10^{7}$ . A modern accelerator is best understood as a machine built entirely around that ratio: it gives up on controlling the beam's speed (hopeless), and instead controls where the beam is and how long its path is. Rigidity, betatron oscillation, phase stability and transition energy are four names for the same surrender.

Where this lives today

Transition crossing is still an operational scar. The CERN PS, injecting at $\gamma = 2.49$ and extracting at $\gamma = 28.7$ , must pass $\gamma_t = 6.09$ every cycle, and the beam loss and emittance blow-up around that instant have been fought for six decades with $\gamma_t$ -jump quadrupoles and RF gymnastics. Modern designs increasingly do what the LHC did: choose a lattice that puts the problem outside the machine's range.

And the spin gives you a thermometer. In a bending field a particle's spin precesses faster than its momentum by the anomalous factor, so the spin tune — extra precessions per turn — is $\nu = a\gamma$ , with $a = (g-2)/2$ a known constant. Since $\nu$ is a pure number and $a_e$ is known to parts in $10^{9}$ , measuring $\nu$ measures $\gamma$ , and hence the beam energy, absolutely. One unit of spin tune is exactly $m_ec^2/a_e = \mathbf{440.65\ MeV}$ of electron beam energy. At LEP this resonant depolarization calibrated the beam energy to better than $1$ MeV — precise enough that the measured energy visibly responded to the tides of the solid Earth and to leakage currents from the Geneva–Paris TGV line. It is this chapter's sharpest case of relativistic kinematics used as a metrological standard. The covariant spin-precession equation underwriting it is `aq-sr-bmt-equation`, and the kinematic rotation beneath that is `aq-sr-thomas-precession`; the same $a\gamma$ physics, turned around to measure $a$ rather than $\gamma$ , is `aq-sr-muon-g2-time-dilation`.

The physical picture

Take $\mathbf{p} = \gamma m\mathbf{v}$ and differentiate. The Lorentz factor depends on the speed, so the product rule produces two terms:

\mathbf{F} = \frac{\mathrm{d}\mathbf{p}}{\mathrm{d}t} = \gamma m\,\mathbf{a} \;+\; \gamma^3 m\,(\boldsymbol\beta\!\cdot\!\mathbf{a})\,\boldsymbol\beta .

The second term lives entirely along the direction of motion. So if you push sideways — $\boldsymbol\beta\!\cdot\!\mathbf{a} = 0$ — it vanishes, and the particle responds with an effective mass of $\gamma m$ . If you push forwards, both terms act along $\boldsymbol\beta$ and the effective mass is $\gamma^3 m$ . Force and acceleration are not even parallel in general. This is not a curiosity. At the LHC's top energy the ratio $\gamma^2$ is $5.3\times10^{7}$ . Bending a $6.8$ TeV proton through the full $27$ km ring is routine engineering with $8.33$ T dipoles; speeding it up by any noticeable amount is essentially impossible. The machine spends eight years of civil engineering and $1232$ superconducting dipoles to change the beam's direction, and adds only a few MeV per turn to its speed — and the speed barely notices. The cleanest statement of the freeze is a logarithmic derivative. From $\beta = pc/E$ and $E^2 = p^2c^2 + m^2c^4$ :

\frac{\mathrm{d}E}{E} = \beta^2\,\frac{\mathrm{d}p}{p} \qquad\Longrightarrow\qquad \boxed{\;\frac{\mathrm{d}\beta}{\beta} = \frac{1}{\gamma^2}\,\frac{\mathrm{d}p}{p}\;}

A $1\%$ momentum spread in an LHC bunch corresponds to a velocity spread of $1.9\times10^{-10}$ . Every proton in the beam, fast ones and slow ones alike, is travelling at the same speed to ten significant figures. The only thing that distinguishes them is the length of the path they take.

Figure

The whole article in one plot. The particle's TWO effective masses, as a function of γ. Transverse inertia (blue) is γm — push sideways and the particle responds like a body of mass γm. Longitudinal inertia (red) is γ³m — push forwards and it responds like a body γ² times heavier still. The gap between the two curves IS the design constraint of every accelerator: at the LHC's γ = 7247 the ratio is 5.3×10⁷, so bending the beam is easy and speeding it up is hopeless. Below, the consequence: the fractional velocity response dβ/β per unit dp/p collapses as 1/γ². At LHC injection (γ = 480) a 1% momentum kick moves the speed by 4.3×10⁻⁸; at top energy by 1.9×10⁻¹⁰; at LEP2's γ = 204,501 by 2.4×10⁻¹³. The velocity is frozen, and everything else in this article is the consequence

The core equations

Relativistic inertia, resolved into its two components:

\boxed{\;\mathbf{F} = \frac{\mathrm{d}\mathbf{p}}{\mathrm{d}t} = \gamma m\,\mathbf{a} + \gamma^3 m\,(\boldsymbol\beta\!\cdot\!\mathbf{a})\,\boldsymbol\beta\;} \qquad (1)

where: $\mathbf{a} = \mathrm{d}\mathbf{v}/\mathrm{d}t$ is the coordinate acceleration. Physical reading: transverse inertia $\gamma m$ , longitudinal inertia $\gamma^3 m$ . Force and acceleration are not parallel unless the push is purely along or purely across the motion. The frozen velocity, which everything below rests on:

\boxed{\;\frac{\mathrm{d}\beta}{\beta} = \frac{1}{\gamma^{2}}\,\frac{\mathrm{d}p}{p}\;}, \qquad \frac{\mathrm{d}E}{E} = \beta^{2}\,\frac{\mathrm{d}p}{p} \qquad (2)

where: exact, at all speeds. Physical reading: this is equation (1)'s longitudinal statement, in the currency an accelerator actually uses. As $\gamma \to \infty$ , momentum and energy track each other perfectly while the velocity refuses to move at all. Magnetic rigidity — how you steer something whose speed you cannot change:

\boxed{\;B\rho = \frac{p}{q}\;} \qquad\Longrightarrow\qquad p\,[\mathrm{GeV}/c] = 0.299792\;B\,[\mathrm{T}]\;\rho\,[\mathrm{m}] \qquad (3)

where: the coefficient is $c \times 10^{-9}$ , not a fitted constant. Physical reading: rigidity depends on momentum alone — not on energy, not on mass, not on speed. It is validated against a real machine in worked example 1. Transverse motion: Hill's equation, and its Courant–Snyder solution.

x''(s) + K(s)\,x(s) = 0,\qquad K_x = \frac{1}{\rho^2} - k,\quad K_y = +k \qquad (4)

x(s) = \sqrt{\varepsilon\,\beta(s)}\;\cos\!\big(\psi(s) + \phi_0\big),\qquad \psi(s) = \int\!\frac{\mathrm{d}s}{\beta(s)} \qquad (5)

where: $s$ is arc length (the independent variable — not time), $k = (q/p)\,\partial B_y/\partial x$ is the normalized quadrupole gradient, and $\beta(s)$ is the Twiss beta function, an envelope, entirely unrelated to $v/c$ . Physical reading: the beam does not travel down the centre of the pipe; it oscillates about it, and $\sqrt{\varepsilon\beta(s)}$ is how far out it swings. The relativity is hidden inside $k \propto 1/p$ : a stiffer beam is harder to focus, in exactly the way it is harder to bend. Emittance and adiabatic damping:

\boxed{\;\varepsilon_N = \beta\gamma\,\varepsilon = \text{invariant under acceleration}\;} \qquad\Longrightarrow\qquad \varepsilon \propto \frac{1}{\beta\gamma} \qquad (6)

where: $\varepsilon$ is the geometric emittance (phase-space area over $\pi$ ), $\varepsilon_N$ the normalized emittance. Physical reading: acceleration cools the beam for free. Adding longitudinal momentum without adding transverse momentum tilts every trajectory closer to the axis, so the beam's angular spread shrinks as $1/\beta\gamma$ while the invariant $\varepsilon_N$ is untouched. And the longitudinal master equation — the phase-slip factor and the transition energy:

\boxed{\;\eta \;=\; \alpha_c - \frac{1}{\gamma^{2}}\;},\qquad \boxed{\;\gamma_t = \frac{1}{\sqrt{\alpha_c}}\;} \qquad (7)

where: $\alpha_c \equiv (\mathrm{d}C/C)/(\mathrm{d}p/p)$ is the momentum compaction — a pure number fixed by the lattice geometry, typically small and positive. Physical reading: $\eta$ decides whether a higher-momentum particle arrives early or late. The two terms fight, and the article's owned result is what happens when they draw.

Why a bunch holds together — and the one energy at which it does not

Derivation AThe phase-slip factor and the transition energy where relativity takes over from geometry

A synchrotron does not accelerate a continuous stream; it accelerates bunches, and it holds each bunch together with a feedback loop so stable that a proton can survive $10^{8}$ turns. The loop works by making a particle's arrival time depend on its momentum — so that a particle with too much energy is punished with a badly-timed kick. This card derives the quantity that decides the sign of that punishment, shows it is a race between a geometric effect and a relativistic one, and locates the unique energy at which the race is a draw and the whole feedback loop switches off.

Step 1. Ask how the revolution time depends on momentum. With $C$ the orbit circumference and $\beta c$ the speed, $T = C/(\beta c)$ , so

\frac{\mathrm{d}T}{T} = \frac{\mathrm{d}C}{C} - \frac{\mathrm{d}\beta}{\beta} .

Why: two, and only two, things can make a particle late — a longer path, or a slower speed. The whole of longitudinal dynamics is the competition between those two terms, and nothing else enters.

Step 2. Evaluate the geometric term. A higher-momentum particle is stiffer, bends less, and swings out to a longer orbit. Define the momentum compaction by exactly that:

\frac{\mathrm{d}C}{C} \equiv \alpha_c\,\frac{\mathrm{d}p}{p},\qquad \alpha_c > 0 .

Why: $\alpha_c$ is pure lattice geometry — where the dipoles are, how the dispersion runs. It contains no relativity whatever, and for a typical ring it is small: $3.45\times10^{-4}$ for the LHC, $0.027$ for the CERN PS.

Step 3. Evaluate the relativistic term — and this is the one that carries the physics. From equation (2),

\frac{\mathrm{d}\beta}{\beta} = \frac{1}{\gamma^2}\,\frac{\mathrm{d}p}{p} .

Why: this is the frozen velocity, and it is the entire relativistic content of the argument. Non-relativistically, $\gamma \to 1$ and this term is $\mathrm{d}p/p$ at full strength — more momentum means proportionally more speed. Relativistically it is throttled by $1/\gamma^2$ and dies. The particle's speed stops responding while its orbit keeps growing.

Step 4. Subtract, and name the result.

\frac{\mathrm{d}T}{T} = \left(\alpha_c - \frac{1}{\gamma^2}\right)\frac{\mathrm{d}p}{p} \;\equiv\; \eta\,\frac{\mathrm{d}p}{p},\qquad \boxed{\;\eta = \alpha_c - \frac{1}{\gamma^2}\;}

Why: read the sign. Below transition ( $\gamma$ small, $1/\gamma^2 > \alpha_c$ , $\eta < 0$ ): the speed-up wins, and a higher-momentum particle arrives early. Above transition ( $\gamma$ large, $1/\gamma^2 \to 0$ , $\eta \to \alpha_c > 0$ ): the longer path wins, and a higher-momentum particle arrives late. The particle's response to being given energy reverses sign, purely because of relativity.

Step 5. Find the draw, and see what breaks. Setting $\eta = 0$ :

\boxed{\;\gamma_t = \frac{1}{\sqrt{\alpha_c}}\;}

At this energy, and only here, a momentum error produces no timing error at all. The RF cavity can no longer tell an off-energy particle from an on-energy one, longitudinal focusing vanishes, the bunch stops being held together, and the synchrotron oscillation frequency $\Omega_s \propto \sqrt{\eta}$ goes to zero. Crossing it requires the stable phase to jump from $\phi_s$ to $\pi - \phi_s$ — a $180°$ flip of the entire RF system, executed in milliseconds while the beam is briefly unfocused. Why: this is the article's owned result and it is a genuinely relativistic accident. $\alpha_c$ is geometry; $1/\gamma^2$ is relativity; and $\gamma_t$ is the energy at which relativity finishes taking over from geometry. In a Newtonian world $1/\gamma^2 \equiv 1 > \alpha_c$ always, $\eta$ would be negative for every machine ever built, and transition energy would not exist.

Step 6. Check it against real machines. The CERN PS has $\alpha_c = 0.027$ , hence $\gamma_t = 6.09$ ; it injects at $\gamma = 2.49$ and extracts at $\gamma = 28.7$ , so it crosses transition on every single cycle, and has done since 1959 — the phase jump is routine operations, and the beam loss around it is a permanent nuisance. The LHC has $\alpha_c = 3.45\times10^{-4}$ , hence $\gamma_t = 53.8$ ; it injects at $\gamma = 480$ , already far above. Why: that was a design choice, not luck. The LHC never crosses transition because a lattice was chosen to put $\gamma_t$ below the injection energy. Equation (7) is not a curiosity to be admired — it is a constraint that shapes the machine. *Verified in verify_aq-sr-relativistic-beam-dynamics.py: sympy derives $(\mathrm{d}\beta/\beta)/(\mathrm{d}p/p) = 1/\gamma^2$ and $(\mathrm{d}E/E)/(\mathrm{d}p/p) = \beta^2$ directly from $E^2 = p^2c^2 + m^2c^4$ ; $\eta = \alpha_c - 1/\gamma^2$ and $\gamma_t = \alpha_c^{-1/2}$ are solved symbolically; the sign of $\eta$ is asserted negative below $\gamma_t$ and positive above; and the PS/LHC transition arithmetic is recomputed at 50 dps and grepped against this article.*

Go deeper · Derivation B · 2 minWhy alternating a focusing lens with a defocusing one focuses

Derivation BWhy alternating a focusing lens with a defocusing one focuses

A quadrupole magnet cannot focus a beam. Maxwell forbids it: a field that converges a beam in $x$ necessarily diverges it in $y$ , and the sum of the two focusing strengths is exactly zero. For twenty years this looked like a hard limit, and it forced early synchrotrons to use weak, gentle bending fields and beam pipes the size of a person. The escape — found by Christofilos in 1950 and published by Courant, Livingston and Snyder in 1952 — is that a sequence of alternating lenses can focus even though every individual lens is neutral. The goal here is to prove it in three lines of matrix algebra, and to see exactly where the free lunch comes from.

Step 1. Write the thin-lens transfer matrices, acting on the vector $(x, x')$ :

M_F = \begin{pmatrix} 1 & 0 \\ -1/f & 1\end{pmatrix},\qquad M_D = \begin{pmatrix} 1 & 0 \\ +1/f & 1\end{pmatrix},\qquad M_L = \begin{pmatrix} 1 & L \\ 0 & 1\end{pmatrix}.

Why: $M_F$ kicks a ray back toward the axis in proportion to its displacement; $M_D$ kicks it away, by exactly as much. They are inverses. Multiplied together with nothing between them they give the identity — no focusing, as Maxwell demands.

Step 2. Put a drift between them and multiply, defocusing last:

M = M_D\,M_L\,M_F = \begin{pmatrix} 1 - L/f & L \\ -L/f^2 & 1 + L/f\end{pmatrix}.

Why: the order matters, and the drift is the whole trick. It is what stops the two lenses from cancelling.

Step 3. Read the net focal length off the lower-left element, which is by definition $-1/f^*$ :

\boxed{\;\frac{1}{f^{*}} = \frac{L}{f^{2}} \;>\; 0\;}

Net focusing — and, remarkably, whichever lens comes first, since $L/f^2$ is even in $f$ . Why: here is the mechanism, and it is pure geometry. The focusing lens bends the ray toward the axis. It therefore arrives at the defocusing lens closer in, where a quadrupole's kick — proportional to displacement — is weaker. The defocusing lens pushes back, but from a shorter lever arm. It cannot undo what was done. The net is convergent, second-order in $L/f$ , and it costs nothing but drift space.

Step 4. Note what it bought. A strong-focusing lattice confines the beam to millimetres instead of tens of centimetres, so the magnet aperture — and with it the iron, the power and the cost — collapses. Every synchrotron built since 1952 uses it. Why: this is why the transverse half of the subject is written in $\beta(s)$ and $\varepsilon$ rather than in force laws. Once the lattice alternates, the motion is a driven oscillation and the useful variables are amplitude and phase. *Verified in verify_aq-sr-relativistic-beam-dynamics.py: sympy multiplies the three matrices symbolically, confirms $M_{21} = -L/f^2$ exactly, confirms $\det M = 1$ (symplectic, as Liouville requires), confirms the result is unchanged when $M_F$ and $M_D$ are swapped, and confirms that with no drift the two lenses cancel to the identity — the drift is the entire mechanism.*

Figure

THE OWNED RESULT. The phase-slip factor η = α_c − 1/γ², plotted against γ, is a RACE between two curves. The blue line is the geometric term α_c — pure lattice, no relativity, a constant: more momentum means a longer orbit, so the particle arrives LATE. The red curve is the relativistic term 1/γ², which starts at 1 (Newton) and dies as the velocity freezes: more momentum means more speed, so the particle arrives EARLY. Below transition the red curve wins and η is negative. Above it the blue line wins and η is positive. They cross at γ_t = 1/√α_c, where η = 0, the RF can no longer tell an off-momentum particle from an on-momentum one, longitudinal focusing VANISHES, and the stable phase must jump by 180°. The CERN PS (α_c = 0.027, γ_t = 6.09) injects at γ = 2.49 and extracts at γ = 28.7 — it crosses transition on every cycle. The LHC (α_c = 3.45×10⁻⁴, γ_t = 53.8) injects at γ = 480, already far above, and never crosses at all. And note the Newtonian limit at the far left: at γ = 1 the red curve sits at 1, above α_c for ANY real lattice, so η is negative always — TRANSITION ENERGY DOES NOT EXIST IN NEWTONIAN PHYSICS. It is made entirely of the frozen velocity

Limiting cases

Case	Result	Meaning
$\gamma \to 1$	$\mathbf{F} \to m\mathbf{a}$ , both inertias $\to m$	Newton. Longitudinal and transverse mass coincide; there is no anisotropy to exploit
$\gamma \to 1$ in (7)	$\eta \to \alpha_c - 1$	for any real lattice $\alpha_c \ll 1$ , so $\eta < 0$ always. Transition energy does not exist in a Newtonian world
$\gamma \gg 1$	$\eta \to \alpha_c$	the velocity is fully frozen; timing is pure geometry. All high-energy rings live here
$\gamma = \gamma_t$	$\eta = 0$ , $\Omega_s = 0$	no longitudinal focusing. The bunch is unheld; the RF must jump phase by $180°$
$\gamma \to \infty$	$\varepsilon = \varepsilon_N/\beta\gamma \to 0$	adiabatic damping: the beam shrinks for free, without cooling
$c \to \infty$	$\gamma \to 1$ , $\mathrm{d}\beta/\beta \to \mathrm{d}p/p$	speed tracks momentum perfectly. Every effect in this article vanishes

Common trapThe two betas are not the same beta

Equation (5) contains the worst symbol collision in the subject. The $\beta(s)$ in $x(s) = \sqrt{\varepsilon\beta(s)}\cos\psi$ is the Twiss beta function, an envelope with units of length — metres, varying around the ring, squeezed to $\beta^* = 0.30$ m at the LHC's interaction point. It has nothing whatever to do with $\beta = v/c$ , which is dimensionless and does not vary at all. Both appear in equation (6): $\varepsilon_N = \beta\gamma\,\varepsilon$ uses the velocity. Likewise emittance $\varepsilon$ is an area, not a length, and the geometric $\varepsilon$ shrinks under acceleration while the normalized $\varepsilon_N$ does not. Quoting one for the other is off by $\beta\gamma$ , which at the LHC is $7247$ .

Common trapAdiabatic damping is not cooling, and it does not beat Liouville

Equation (6) says the geometric emittance falls as $1/\beta\gamma$ : a beam $240\ \mu$ m across at LHC injection is $10\ \mu$ m across at collision, with no cooling system anywhere. It is tempting — and wrong — to call this cooling. Liouville is not violated, because the conserved phase-space volume is the one built from the canonical momenta, and that is exactly $\varepsilon_N$ , which does not change. Acceleration adds longitudinal momentum without adding transverse momentum, so every trajectory's angle $x' = p_x/p_s$ shrinks: the beam becomes more parallel, not more ordered. Genuine cooling — stochastic, electron, radiation damping — reduces $\varepsilon_N$ itself and needs a dissipative process. Acceleration alone never can.

Worked examples

1 · Steer a 6.8 TeV proton — and check the formula against the real machine first

The LHC's dipoles produce $8.33$ T. What bending radius do they need for a $7$ TeV proton, and does the ring fit in the $26.7$ km LEP tunnel it inherited?

Step 1 — identify. This is rigidity, equation (3): $p = 0.299792\,B\rho$ , with the coefficient $c \times 10^{-9}$ and no $\gamma$ anywhere. Momentum is all a magnet sees.

Step 2 — validate the formula before trusting it. The LHC's actual bending radius is $\rho = 2804$ m. Then

p = 0.299792 \times 8.33 \times 2804 = \mathbf{7.00\ TeV}/c ,

which is the design energy to three figures. The formula reproduces a real machine, so we may use it.

Step 3 — check the geometry. The dipoles occupy only the arcs, so $2\pi\rho = 2\pi(2804) = 17.6$ km of bending, inside a $26.7$ km ring — the remaining $9.1$ km is straight sections for the RF, the focusing lattice and the four experiments. The bending radius is not the ring radius, and confusing them is a standard error.

Step 4 — the boundary number. Run 3 operates at $6.8$ TeV, not $7.0$ , because the dipoles were never fully trained; from (3), $6.8$ TeV needs $B = 8.09$ T. The energy of the LHC is set by one number: how much field its magnets will hold without quenching. Everything else — the tunnel, the optics, the physics reach of equation (6) in `aq-sr-collider-fixed-target-energetics` — followed from that.

2 · Free cooling: why the LHC beam is 23 times narrower at collision than at injection

The LHC injects protons at $450$ GeV with a normalized emittance $\varepsilon_N = 2.5\ \mu$ m, and collides them at $6.8$ TeV. No cooling system is applied. How much does the beam shrink, and what is the spot size at the interaction point?

Step 1 — identify. Equation (6): $\varepsilon_N = \beta\gamma\,\varepsilon$ is invariant, so the geometric emittance falls as $1/\beta\gamma$ . With $\beta \approx 1$ throughout, $\varepsilon \propto 1/\gamma$ .

Step 2 — the two Lorentz factors. $\gamma_{\text{inj}} = 450/0.9383 = 480$ and $\gamma_{\text{coll}} = 6800/0.9383 = \mathbf{7247}$ . The ratio is $\mathbf{15.1}$ : the geometric emittance shrinks by a factor of $15$ during the ramp, and nothing was done to it.

Step 3 — the emittances. $\varepsilon_{\text{inj}} = 2.5\times10^{-6}/480 = 5.21\times10^{-9}$ m·rad; $\varepsilon_{\text{coll}} = 2.5\times10^{-6}/7247 = \mathbf{3.45\times10^{-10}}$ m·rad.

Step 4 — the spot. The beam radius is $\sigma = \sqrt{\varepsilon\,\beta(s)}$ from (5). At injection the optics are relaxed, $\beta \approx 11$ m, giving $\sigma = \sqrt{5.21\times10^{-9} \times 11} = \mathbf{240\ \mu m}$ . At the collision point the optics are squeezed to $\beta^* = 0.30$ m, giving

\sigma^{*} = \sqrt{3.45\times10^{-10} \times 0.30} = \mathbf{10.2\ \mu m} .

Step 5 — the boundary number. A $10\ \mu$ m spot — a fifth the width of a human hair — carrying $6.8$ TeV per proton, held for hours. And the total narrowing of $\mathbf{23.5}$ in radius came from two sources only: acceleration (free, relativistic, factor $\sqrt{15.1} = \mathbf{3.89}$ ) and optics (paid for, factor $\sqrt{11/0.3} = \mathbf{6.06}$ ). Relativity did more than a third of the work and sent no bill.

Physics of the portrait

Origins · 2 minEmittance and how this was discovered

Emittance

1950–1958 · Christofilos invents strong focusing and does not publish · Courant, Livingston and Snyder publish it · Courant and Snyder give the theory its language

The quantity that named this article is the one nobody set out to invent. Emittance is the area a beam occupies in phase space — position against angle — and it matters because Liouville's theorem says you cannot shrink it without dissipation. It is the beam's irreducible untidiness, and it is what a machine designer spends their life fighting. The fight was transformed in 1952. Ernest Courant, M. Stanley Livingston and Hartland Snyder published The Strong-Focusing Synchrotron (Phys. Rev. 88, 1190 (1952)), showing that alternating quadrupole gradients confine a beam far more tightly than the weak, gentle focusing then in use. The magnet apertures collapsed, and with them the cost of a high-energy machine. The idea had in fact been found two years earlier by Nicholas Christofilos, a Greek-American elevator engineer working alone, who filed a patent application on 10 March 1950 and never published a paper; the Brookhaven group acknowledged his priority when they learned of it. Then Courant and Snyder gave it a language. Theory of the Alternating-Gradient Synchrotron (Annals of Physics 3, 1 (1958)) is one of the most-used papers in accelerator physics: it introduced the Twiss parameters, the betatron amplitude function $\beta(s)$ , the phase advance, the tune, and the Courant–Snyder invariant that is the emittance. Every accelerator built since has been designed in their variables — which is why the beam envelope in a $27$ km superconducting proton ring is still described by a function called $\beta$ , defined in a paper about a machine a thousand times weaker.

History

1928

— Rolf Widerøe's linear resonance accelerator; the multi-pass principle that all later machines share.

1944

— Vladimir Veksler publishes phase stability (Compt. Rend. Acad. Sci. U.R.S.S. 43, 444); the RF bucket becomes possible.

1945

— Edwin McMillan independently proposes the synchrotron on the same principle (Phys. Rev. 68, 143).

1950

— Nicholas Christofilos conceives alternating-gradient focusing and files a patent application; he does not publish.

1952

— Courant, Livingston and Snyder publish the strong-focusing synchrotron (Phys. Rev. 88, 1190).

1958

— Courant and Snyder's Theory of the Alternating-Gradient Synchrotron (Annals of Physics 3, 1) gives the field its permanent language: Twiss parameters, beta function, tune, emittance.

1959

— The CERN Proton Synchrotron begins operating at 26 GeV; with a transition gamma of 6.1 it must cross transition on every cycle, and has done ever since.

1989

— LEP begins operation; resonant depolarization later calibrates its beam energy through the spin tune, and proves sensitive enough to detect Earth tides and a nearby railway.

2008

— The LHC is commissioned in LEP's tunnel; its lattice puts the transition gamma at 53.8, below the injection gamma of 480, so it never crosses transition at all.

Going further

Texts. Wiedemann, Particle Accelerator Physics, 4th ed. · S.Y. Lee, Accelerator Physics, 4th ed. · Sands, The Physics of Electron Storage Rings, SLAC-121 (1970) · Courant & Snyder, Ann. Phys. 3, 1 (1958) · Courant, Livingston & Snyder, Phys. Rev. 88, 1190 (1952) · Chao & Tigner (eds.), Handbook of Accelerator Physics and Engineering.

Provenance. Signature $(+,-,-,-)$ . Every number is recomputed at 50 dps in verify_aq-sr-relativistic-beam-dynamics.py, which reads this file and asserts each printed value against what it computed. Base status: the base's eq[6] ( $\eta = \alpha_c - 1/\gamma^2$ , $\gamma_t = 1/\sqrt{\alpha_c}$ ), its FODO result $1/f^* = L/f^2$ , its rigidity coefficient $0.299792$ , and its spin tune $\nu = a\gamma$ were each re-derived symbolically and are correct. Its eq[8], the practical synchrotron-loss formula $U_0[\mathrm{keV}] = 88.5\,E^4[\mathrm{GeV}]/\rho[\mathrm{m}]$ , was independently recomputed from first principles as $\mathbf{88.46}$ and is correct — and it is the anchor that convicts the neighbouring base: `aq-sr-synchrotron-radiation`'s per-turn loss formula disagrees with it by a factor of $c$ . That defect is repaired in that article, not this one.

Scope. This article owns the relativistic dynamics of a beam: the inertia anisotropy, the frozen velocity, rigidity, the phase-slip factor and the transition energy. The radiation the beam emits while being bent — its $\gamma^4$ scaling, its per-turn cost, and why it killed LEP — is `aq-sr-synchrotron-radiation`. The energetics of what you do with the beam once you have it is `aq-sr-collider-fixed-target-energetics`. The covariant spin-precession equation behind the spin tune is `aq-sr-bmt-equation`, and the kinematic rotation underneath it is `aq-sr-thomas-precession`; both are used here only through the single result $\nu = a\gamma$ , which is quoted, not derived.

Builds on. Relativistic Momentum · Relativistic Force · The Lorentz Force. Leads to. Synchrotron Radiation · Muon Storage-Ring Time Dilation · Collider vs Fixed-Target Energetics.

Why is longitudinal inertia γ³m but transverse inertia only γm?

Because changing a particle's speed changes $\gamma$ , and changing its direction does not. Differentiate $\mathbf{p} = \gamma m \mathbf{v}$ : the product rule gives $\dot{\gamma} m\mathbf{v} + \gamma m \dot{\mathbf{v}}$ , and $\dot\gamma = \gamma^3(\boldsymbol\beta\!\cdot\!\dot{\boldsymbol\beta})$ . For a purely transverse push, $\boldsymbol\beta\!\cdot\!\mathbf{a} = 0$ , so $\dot\gamma = 0$ , the first term dies, and only $\gamma m\mathbf{a}$ survives. For a purely longitudinal push, both terms act, and they sum to $\gamma^3 m\mathbf{a}$ . The extra $\gamma^2$ is precisely the cost of the energy that a speed change requires and a direction change does not. It is not two masses in any deep sense — it is one $\mathrm{d}\mathbf{p}/\mathrm{d}t$ , resolved into components.

If the velocity is frozen, how does an accelerator accelerate anything?

It does not accelerate the speed; it accelerates the momentum and the energy, which are unbounded even though the speed is not. An RF cavity adds energy every turn; $E$ and $p$ climb without limit; $\beta$ creeps toward $1$ and stops mattering. Everything the machine subsequently does is a response to the momentum: the dipoles must be ramped in step with $p$ to keep the orbit fixed (equation 3), the quadrupole gradients must be ramped too because $k \propto 1/p$ , and the RF must track a revolution frequency that changes almost not at all. The word "accelerator" is a historical accident. It is a momentum-raiser.

Is the beta function in Hill's equation the same beta as v/c?

No, and this is the single worst notation collision in accelerator physics. The Twiss beta function $\beta(s)$ is an envelope with units of length — the LHC squeezes it to $\beta^* = 0.30$ m at the collision point — and it describes how far the beam swings off-axis. The relativistic beta $\beta = v/c$ is dimensionless, close to $1$ , and effectively constant. They appear in the same equation in (6): $\varepsilon_N = \beta\gamma\,\varepsilon$ uses the velocity. Nothing but context distinguishes them, and both are universal. Read carefully, and check the units.

What actually goes wrong at transition, physically?

The bunch stops being able to correct itself. Normally a particle with too much energy is nudged back by the RF, because its momentum error translates into a timing error, and the RF punishes bad timing. At $\gamma = \gamma_t$ the translation coefficient $\eta$ is zero: a particle can have a large momentum error and still arrive exactly on time. The RF is therefore blind to it, the restoring force vanishes ( $\Omega_s \propto \sqrt{|\eta|} \to 0$ ), and the bunch — held together by nothing — begins to lengthen and to spread in momentum. Worse, $\eta$ then reverses sign, so the phase that was stable becomes unstable: the RF must jump from $\phi_s$ to $\pi - \phi_s$ within milliseconds. The CERN PS has done this on every cycle since 1959, and it still costs beam.

Why does the LHC not have this problem?

Because its designers made $\gamma_t$ irrelevant by construction. The transition gamma depends only on the lattice — $\gamma_t = 1/\sqrt{\alpha_c}$ — and the LHC's momentum compaction $\alpha_c = 3.45\times10^{-4}$ puts $\gamma_t = 53.8$ . But the LHC does not build its own beam from rest: it inherits one from the SPS at $450$ GeV, where $\gamma = 480$ , already nine times above transition. The LHC therefore spends its entire operating life above transition and never crosses it. The problem was not solved. It was placed outside the machine's range — which is usually the better engineering.

Does the beam radiate while all this is going on?

Yes, and for electrons it dominates everything. A bent relativistic charge radiates with a power scaling as $\gamma^4/\rho^2$ , so a per-turn energy loss of $U_0[\mathrm{keV}] = 88.5\,E^4[\mathrm{GeV}]/\rho[\mathrm{m}]$ — which at LEP2's $104.5$ GeV and $\rho = 3026$ m came to $3.49$ GeV per turn, or $3.3\%$ of the entire beam energy, thrown away every $89$ microseconds and replaced by a vast RF plant. That $E^4/m^4$ law is what ended LEP and what put protons in its tunnel instead. It is the subject of `aq-sr-synchrotron-radiation`, and it is deliberately not derived here: this article is about how you steer a beam, not about what it emits while you do.