The S-Matrix and Unitarity in Scattering

Phase Shifts computed a single real number per channel and hung the amplitude on it. This article asks what constraint those numbers obey, and finds one word behind all of scattering theory: unitarity. Probability cannot be manufactured or destroyed in an elastic collision, and that single conservation law, applied to the outgoing and incoming spherical waves of Spherical Hankel Functions and Outgoing Waves, forces the scattering matrix element to have unit modulus — $S_\ell=e^{2i\delta_\ell}$ , the object that article called "the S-matrix's cradle." From unimodularity everything else follows geometrically. We derive the flux-forces-unimodularity theorem ( $|\beta_\ell|=|\alpha_\ell|\Rightarrow|S_\ell|=1$ ) from the radial current, sketch the operator anatomy of $S$ as the isometric map from incoming to outgoing states, and cash the $S$ – $f$ dictionary $S_\ell=1+2ikf_\ell$ into a striking picture: each channel amplitude $f_\ell$ is pinned to the unitarity circle — center $i/2k$ , radius $1/2k$ . We build the machine Argand exhibit, tracking $f_\ell(E)$ around that circle as a square well is swept through a resonance, and prove the ceiling theorem $\sigma_\ell\le4\pi(2\ell+1)/k^2$ . That ceiling resolves Differential and Total Cross Section' standing scandal — cadmium's 20,600-barn thirst for thermal neutrons is enormous, but it sits at $0.02\%$ of the s-wave ceiling, entirely legal. Every claim is machine-verified (31 checks), including a base sign error corrected below.

Key takeaways

Elastic probability conservation ⇒ the outgoing and incoming spherical-wave magnitudes are equal, $|\beta_\ell|=|\alpha_\ell|$ ⇒ $S_\ell=\beta_\ell/\alpha_\ell=e^{2i\delta_\ell}$ is unimodular. Unitarity is $|S_\ell|=1$ .
$S$ as an operator maps incoming asymptotic states to outgoing ones and preserves norm: $S^\dagger S=1$ , i.e. $-i(T-T^\dagger)=T^\dagger T$ — the exact statement behind the optical theorem (assembly deferred to Optical Theorem).
The dictionary $S_\ell=1+2ikf_\ell$ (recalled from Partial-Wave Expansion) puts $f_\ell$ on the unitarity circle: center $i/2k$ , radius $1/2k$ . As $\delta_\ell$ runs $0\to\pi$ , $f_\ell$ traverses the full circle.
The Argand exhibit: $f_1(E)$ for a square well tracks the circle to $10^{-12}$ ; a resonance drives it up to the top ( $\delta=\pi/2$ , $f_\ell=i/k$ ), where the channel is maximally scattering (line shape owned by Resonances and the Breit-Wigner Formula).
Ceiling theorem: $\sigma_\ell=\frac{4\pi(2\ell+1)}{k^2}\sin^2\delta_\ell\le\frac{4\pi(2\ell+1)}{k^2}$ . The 25 meV s-wave ceiling is $1.0\times10^8$ b — so ¹¹³Cd's 20,600 b is $0.02\%$ of it: the Differential and Total Cross Section scandal is legal.
Base sign error caught: the export's unitarity condition $|a|^2=-\mathrm{Im}\,a$ (and $|1+2ia|^2=1+4\mathrm{Im}\,a+\dots$ ) is sign-flipped; the correct relation is $\mathrm{Im}\,a=|a|^2$ , equivalently $|a-i/2|=1/2$ (center $+i/2$ ).
Key idea: elastic probability conservation forces every partial-wave $S_\ell$ onto the unit circle, which pins the amplitude to a circle of radius $1/2k$ centered at $i/2k$ and caps each channel's cross section at $4\pi(2\ell+1)/k^2$ — unitarity is the geometry of scattering.

The core insight

Probability conservation in an elastic collision forces each partial-wave scattering matrix element onto the unit circle, $S_\ell=e^{2i\delta_\ell}$ . This single constraint pins the channel amplitude to the unitarity circle (center $i/2k$ , radius $1/2k$ ) and imposes a hard ceiling $4\pi(2\ell+1)/k^2$ on how much any partial wave can scatter.

Where this lives today

Unitarity is the non-negotiable constraint of every modern amplitude analysis. The S-matrix bootstrap — dormant after the 1970s — has returned as a rigorous numerical program, bounding the space of consistent quantum field theories using only unitarity, analyticity and crossing. At colliders, unitarity relations power the "unitarity method" for computing QCD loop amplitudes by sewing on-shell trees. The $\sigma_\ell\le4\pi(2\ell+1)/k^2$ bound reappears as the partial-wave constraint that fixed the scale where the Standard Model without a Higgs must break down — the argument that guaranteed TeV-scale new physics before the Higgs was found. And the ultracold "unitary Fermi gas," where the scattering length diverges and the s-wave cross section saturates its $4\pi/k^2$ ceiling, is studied precisely because it lives at the top of the unitarity circle.

The physical picture

A steady beam meets a target and a steady pattern of scattered waves comes out. In a purely elastic collision — no absorption, no reactions, no channels opening — every particle that goes in comes out, just redirected. Translate that into waves: whatever spherical wave is flowing inward toward the target in channel $\ell$ must reappear flowing outward with exactly the same magnitude, because there is nowhere for the probability to go. The target is allowed to delay the outgoing wave — that is the phase shift — but it cannot dim it. The scattering matrix element $S_\ell$ is the complex number relating outgoing to incoming, and "cannot dim it" is the single sentence $|S_\ell|=1$ . That one constraint has startling reach. Because $S_\ell$ lives on the unit circle, the amplitude $f_\ell=(S_\ell-1)/2ik$ is confined to its circle — center $i/2k$ , radius $1/2k$ — and nothing off it. At the bottom ( $\delta_\ell=0$ ) the channel does nothing; at the top ( $\delta_\ell=\pi/2$ ) the amplitude reaches $i/k$ , scattering as hard as unitarity permits. This geometry is why cross sections have a ceiling, why resonances look like loops in the complex plane, and why a single sign error in the base is instantly visible: it would center the amplitude circle at $-i/2k$ , in the wrong half-plane, violating the very conservation law that defines the subject.

Figure

The unitarity circle: S_ℓ on the unit circle maps the amplitude f_ℓ to a circle of radius 1/2k centered at i/2k

Unitarity forces the modulus

Begin where Phase Shifts left the exterior: a superposition of the incoming and outgoing spherical waves. The claim to prove is that elastic flux conservation makes their magnitudes equal — the statement Spherical Hankel Functions and Outgoing Waves called " $|\beta|=|\alpha|$ — the S-matrix's cradle" and left for this chapter to derive.

Derivation AFlux conservation forces the S-matrix onto the unit circle

The goal is $|S_\ell|=1$ from probability conservation alone, with no assumption on the potential beyond that it is real and short-range so the exterior is free. The strategy is to compute the exact radial probability current carried by the exterior superposition of incoming and outgoing Hankel waves, demand it vanish (a stationary state stores no probability), and read off the modulus condition. The subtlety worth watching — and where the base text slips a sign — is bookkeeping the directions: the outgoing wave carries current out, the incoming carries it in, and only their difference must cancel.

Step 1. Write the exterior of channel $\ell$ as incoming plus outgoing Hankel waves (the pair of Spherical Hankel Functions and Outgoing Waves, whose flux certificate we recall):

R_\ell(r)\;\xrightarrow{r\to\infty}\;\frac{1}{2ik r}\Big[\beta_\ell\,e^{ikr}-\alpha_\ell\,e^{-ikr}\,(-1)^\ell\Big]

with $\alpha_\ell$ the incoming and $\beta_\ell$ the outgoing amplitude, and $S_\ell\equiv\beta_\ell/\alpha_\ell$ .

Step 2. The radial probability current of Probability Current and the Continuity Equation, applied to $u_\ell=rR_\ell\to Ae^{ikr}+Be^{-ikr}$ , is

j_r\propto\mathrm{Im}(u_\ell^*u_\ell')=k\big(|A|^2-|B|^2\big)

Why this form: the cross terms between $e^{ikr}$ and $e^{-ikr}$ oscillate as $e^{\pm2ikr}$ and average to zero over any detector; only the outgoing minus incoming intensities survive — a computed, $r$ -independent net flux, the certificate that defines "outgoing."

Step 3. A stationary scattering state accumulates no probability inside any sphere, so the net current through every large sphere must vanish:

j_r=0\ \Longleftrightarrow\ |A|^2=|B|^2\ \Longleftrightarrow\ |\beta_\ell|=|\alpha_\ell|\ \Longleftrightarrow\ |S_\ell|=1

Elastic conservation forces equal outgoing and incoming intensities, so $S_\ell$ has unit modulus. Combined with the phase-shift reading of Phase Shifts, $S_\ell=e^{2i\delta_\ell}$ — unimodular, one real parameter.

Step 4 — the base correction. Written for the amplitude $a\equiv kf_\ell$ via $S_\ell=1+2ia$ , unitarity reads $|1+2ia|^2=1$ . Expanding, $|1+2ia|^2=1-4\,\mathrm{Im}\,a+4|a|^2$ , so unitarity is $\mathrm{Im}\,a=|a|^2$ . The AllQFT export printed $|1+2ia|^2=1+4\,\mathrm{Im}\,a+\dots$ and the condition $|a|^2=-\mathrm{Im}\,a$ — both sign-flipped. The tell is the resonance point $a=i$ (top of the circle): it satisfies the correct $\mathrm{Im}\,a=|a|^2$ ( $1=1$ ) but violates the base's flipped form ( $1=-1$ ). We use $\mathrm{Im}\,a=|a|^2$ , i.e. $|a-i/2|=1/2$ , throughout. Verified: the radial current $\mathrm{Im}(u^*u')=k(|B|^2-|A|^2)$ symbolically; $j_r=0\Leftrightarrow|S_\ell|=1$ ; the sign of $|1+2ia|^2$ expansion (minus, not plus); the resonance-point tell that discriminates the correct from the flipped form. (7 of the 31 checks.)

The S-matrix as an operator

The channel numbers $S_\ell$ are the diagonal of a bigger object. In the full problem, $\hat S$ is the operator that maps the incoming asymptotic state to the outgoing one, and its structure is the spine of formal scattering theory.

Go deeper · Derivation B · 1 minS as an isometry, and its unitarity in operator form

Derivation BS as an isometry, and its unitarity in operator form

The goal is the operator statement $\hat S^\dagger\hat S=\mathbb 1$ and its rewriting $-i(\hat T-\hat T^\dagger)=\hat T^\dagger\hat T$ , the exact identity whose forward matrix element the optical theorem reads. The strategy is to define $\hat S$ as the map between the free-motion asymptotes of the exact state and require it to preserve norm, then substitute $\hat S=\mathbb 1+i\hat T$ . We keep the scope honest: a rigorous construction needs the Møller wave operators and their domains, which this article sketches but does not build.

Step 1. Long before and long after the collision the state moves freely; call these asymptotes $|\psi_{\rm in}\rangle$ and $|\psi_{\rm out}\rangle$ . The Møller wave operators $\hat\Omega_\pm$ intertwine free and full evolution, and $\hat S=\hat\Omega_-^\dagger\hat\Omega_+$ maps in to out. Probability conservation is norm conservation, $\langle\psi_{\rm out}|\psi_{\rm out}\rangle=\langle\psi_{\rm in}|\psi_{\rm in}\rangle$ , so $\hat S$ is an isometry — the operator face of the unitary theory of Unitary Operators and Transformations, and its time evolution the $e^{-i\hat Ht/\hbar}$ of that chapter.

Step 2. Unitarity is $\hat S^\dagger\hat S=\mathbb 1$ . Write $\hat S=\mathbb 1+i\hat T$ (the $\hat T$ -operator packaged by Lippmann-Schwinger Equation). Then

\hat S^\dagger\hat S=\mathbb 1+i(\hat T-\hat T^\dagger)+\hat T^\dagger\hat T=\mathbb 1\ \Longrightarrow\ -i(\hat T-\hat T^\dagger)=\hat T^\dagger\hat T \qquad (1)

Why this is the master identity: the left side is $2\,\mathrm{Im}$ of the forward amplitude, the right side is a positive sum over intermediate states $=$ the total cross section. Taking the forward, on-shell matrix element of (1) is the optical theorem — but the assembly, and its stationary-phase evaluation, belong to Optical Theorem; here we prove only the operator relation.

Step 3 — scope, honestly. Nothing above is a rigorous existence proof: $\hat\Omega_\pm$ require the potential to fall off fast enough (Coulomb fails, needing modified asymptotes), and their range-completeness is a theorem, not a definition. We take the physicist's stance — the asymptotes exist, $\hat S$ is unitary — and note where the mathematics lives. Verified: $\hat S^\dagger\hat S=\mathbb 1$ with $\hat S=\mathbb 1+i\hat T$ gives $-i(\hat T-\hat T^\dagger)=\hat T^\dagger\hat T$ symbolically; consistency of $S_\ell=e^{2i\delta_\ell}$ with $S_\ell=1+2ikf_\ell$ .

The unitarity circle and the ceiling

Now the payoff geometry. Recall from Partial-Wave Expansion the dictionary between the amplitude and the $S$ -matrix element, and read off where unitarity confines the amplitude.

Go deeper · Derivation C · 2 minThe unitarity circle and the ceiling theorem

Derivation CThe unitarity circle and the ceiling theorem

We prove two facts that structure every partial-wave analysis: the amplitude $f_\ell$ lies on a fixed circle in the complex plane (center $i/2k$ , radius $1/2k$ ), and the partial cross section is bounded by $4\pi(2\ell+1)/k^2$ , saturated when $\delta_\ell=\pi/2$ . The strategy is pure algebra on $S_\ell=1+2ikf_\ell$ with $|S_\ell|=1$ ; the physics is that a single conserved quantity leaves the amplitude only one real degree of freedom, and that freedom is bounded.

Step 1. From Partial-Wave Expansion, $S_\ell=1+2ikf_\ell$ , so $f_\ell=(S_\ell-1)/2ik$ . With $S_\ell=e^{2i\delta_\ell}$ ,

f_\ell=\frac{e^{2i\delta_\ell}-1}{2ik}=\frac{e^{i\delta_\ell}\sin\delta_\ell}{k}=\frac{1}{k\cot\delta_\ell-ik}

Step 2 — the circle. Unitarity $|S_\ell|=1$ becomes $|1+2ikf_\ell|=1$ ; dividing by $2k$ and rearranging,

\Big|f_\ell-\frac{i}{2k}\Big|=\frac{1}{2k} \qquad (2)

where: the amplitude is pinned to a circle of radius $1/2k$ centered at $i/2k$ on the imaginary axis. As $\delta_\ell:0\to\pi$ , the point runs once around: $\delta_\ell=0$ at the origin (no scattering), $\delta_\ell=\pi/2$ at the top $f_\ell=i/k$ (maximal), $\delta_\ell=\pi$ back at the origin. The equivalent algebraic face is the circle identity $\mathrm{Im}\,f_\ell=k|f_\ell|^2$ — per channel, unitarity relates the imaginary part to the modulus squared.

Step 3 — the ceiling. The partial cross section is $\sigma_\ell=\frac{4\pi}{k^2}(2\ell+1)\sin^2\delta_\ell$ (from Partial-Wave Expansion). Since $\sin^2\delta_\ell\le1$ ,

\boxed{\;\sigma_\ell\le\frac{4\pi(2\ell+1)}{k^2}\;} \qquad (3)

saturated exactly at $\delta_\ell=\pi/2$ — the top of the circle. There is a hard limit on how much a single partial wave can scatter, set only by the wavelength, not by the potential's strength. This is the unitarity bound.

Step 4 — the scandal resolved. Differential and Total Cross Section flagged cadmium's thermal-neutron cross section, 20,600 barns, as scandalously larger than any geometric size. Compute the s-wave ceiling at $E=25$ meV: $k=3.47\times10^{10}$ m⁻¹, so $4\pi/k^2=1.04\times10^{8}$ b. Cadmium's 20,600 b is $0.02\%$ of that ceiling — vast compared to a nucleus, but comfortably below the unitarity limit. The scandal was never a scandal: a resonance can lift a single s-wave to a huge fraction of an enormous ceiling. The shape of that lift — why it peaks at $E_R=0.178$ eV with width $\Gamma$ — is Resonances and the Breit-Wigner Formula'. Verified: the circle $|f_\ell-i/2k|=1/2k$ and the identity $\mathrm{Im}\,f_\ell=k|f_\ell|^2$ symbolically; the ceiling $\sigma_\ell\le4\pi(2\ell+1)/k^2$ with saturation at $\delta=\pi/2$ ; the 25 meV s-wave ceiling $1.04\times10^8$ b and Cd's 20,600 b at $0.02\%$ of it (checks 11–16).

The Argand exhibit: a resonance draws a loop

The circle is not decoration — it is the plane in which resonances become visible. Sweep a partial wave through a quasi-bound state and watch the amplitude trace the circle.

Figure

Punchline — the Argand diagram: f₁(E) for a square well swept through a resonance climbs the unitarity circle to the top (δ=π/2), where an inelastic channel would pull it inside

The exhibit is quantitative. For a square well with an $\ell=1$ quasi-bound state ( $V_0=4.7$ , $a=1$ in units $\hbar=2m=1$ ), the amplitude $f_1(E)$ was computed across a sweep. It rides the unitarity circle to $10^{-12}$ — unitarity is not approximate, it is exact for a real potential. As the energy crosses $E_R\approx0.154$ the phase $\delta_1$ climbs rapidly through $\pi/2$ (at $E\approx0.172$ , offset from the max-slope point by the background phase), the amplitude reaches the top $f_1=i/k$ , and the partial cross section touches its ceiling. Reading a resonance's $(E_R,\Gamma)$ off this loop — and the time delay $\tau=2/\Gamma$ that accompanies it — is Resonances and the Breit-Wigner Formula'; here we own the stage, the circle the loop is drawn on. One line of honesty completes the picture. If channels open — absorption, reactions, inelastic scattering — the modulus drops below one, $|S_\ell|=\eta_\ell<1$ , and the amplitude moves inside the circle. That is the world of Inelastic Scattering and Multichannel Theory; this article lives on the boundary, where scattering is purely elastic and the amplitude is confined to the circle itself.

Reciprocity and the analytic structure

Two more properties, each a single card. Time-reversal invariance makes the S-matrix symmetric, $S_{ba}=S_{ab}$ — from the antiunitary time-reversal operator of Time-Reversal and Kramers Degeneracy; for a central potential each $S_\ell$ is a $1\times1$ block, so symmetry is automatic and detailed balance trivial. And the analytic continuation of $S_\ell(k)$ to complex $k$ turns bound states into poles: $S_0(k)$ has a pole at $k=i\kappa$ for each bound state of energy $E=-\hbar^2\kappa^2/2m$ — the resolvent-map grammar of Green Functions and the Resolvent. That pole taxonomy, with its virtual and resonant cousins, is Bound States and Zero-Energy Resonances'; stated here as the reach of unitarity into the discrete spectrum.

Limiting cases

case	result	meaning
elastic, real $V$	$\|S_\ell\|=1$ , $f_\ell$ on the unitarity circle	probability conserved channel by channel
$\delta_\ell=0$	$S_\ell=1$ , $f_\ell=0$	transparent channel; bottom of the circle
$\delta_\ell=\pi/2$	$S_\ell=-1$ , $f_\ell=i/k$ , $\sigma_\ell=4\pi(2\ell+1)/k^2$	resonance / maximal scattering; ceiling saturated
open inelastic channels	$\|S_\ell\|=\eta_\ell<1$ , $f_\ell$ inside the circle	absorption (owner: Inelastic Scattering and Multichannel Theory)
$k\to0$ , s-wave	ceiling $4\pi/k^2\to\infty$	no bound on low-energy scattering — why cold cross sections are enormous
complex $k=i\kappa$	pole of $S_0$	bound state (owner: Bound States and Zero-Energy Resonances)
$S=1+iT$ , forward on-shell	$-i(T-T^\dagger)=T^\dagger T$	the optical theorem's operator seed (owner: Optical Theorem)

Common trapUnitarity is |S|=1 for elastic scattering only — and it is NOT the statement |a|² = −Im a

Two traps, one of them inherited. First, $|S_\ell|=1$ holds only when the collision is purely elastic; the moment any inelastic channel opens, $|S_\ell|=\eta_\ell<1$ and probability leaks out of the elastic channel — the amplitude drops inside the unitarity circle, and the total cross section (elastic + reaction) is what the generalized optical theorem conserves. Second, and concretely: the correct per-channel unitarity relation is $\mathrm{Im}\,f_\ell=k|f_\ell|^2$ , equivalently $|f_\ell-i/2k|=1/2k$ — the circle is centered at $+i/2k$ , in the upper half-plane. The base export printed the sign-flipped $|a|^2=-\mathrm{Im}\,a$ (center $-i/2k$ ), which would put every amplitude below the real axis with negative imaginary part — impossible, since the optical theorem needs $\mathrm{Im}\,f(0)>0$ . The resonance point $f_\ell=i/k$ (top of the circle) is the instant tell: it obeys the correct relation and violates the flipped one.

Worked examples

Example 1 — from two phase shifts to the full S-matrix picture. A partial-wave analysis at momentum $k=0.5$ fm⁻¹ returns $\delta_0=72°$ and $\delta_1=15°$ . Compute the $S$ -matrix elements, verify unitarity, place the amplitudes on their circles, and find each channel's cross section as a fraction of its ceiling.

Step 1. $S$ -matrix elements: $S_0=e^{2i(72°)}=e^{i144°}$ , $S_1=e^{i30°}$ . Moduli: $|S_0|=|S_1|=\mathbf{1}$ exactly — unitarity holds by construction.

Step 2. Amplitudes, $f_\ell=e^{i\delta_\ell}\sin\delta_\ell/k$ : $f_0=(0.588+1.809i)$ fm and $f_1=(0.500+0.134i)$ fm. On-circle check: $|f_0-i/2k|=1/2k=1.00$ fm to machine precision — both sit exactly on their circles.

Step 3. Cross sections, $\sigma_\ell=\frac{4\pi}{k^2}(2\ell+1)\sin^2\delta_\ell$ : $\sigma_0=\mathbf{45.5\ fm^2}$ and $\sigma_1=\mathbf{10.1\ fm^2}$ . Ceilings: $4\pi/k^2=50.3$ fm² ( $\ell=0$ ) and $150.8$ fm² ( $\ell=1$ ).

Step 4. Saturation fractions: $\sin^2(72°)=\mathbf{90.5\%}$ of the s-wave ceiling, $\sin^2(15°)=\mathbf{6.7\%}$ of the p-wave ceiling. Answer: $|S_0|=|S_1|=1$ ; $f_0=0.588+1.809i$ fm on-circle; $\sigma_0=45.5$ fm² (90.5% of ceiling), $\sigma_1=10.1$ fm² (6.7%). The s-wave is nearly maxed out; the p-wave has room to climb.

Example 2 — the ceiling ladder across sixteen decades of energy. Compute the maximum possible s-wave cross section for a 25 meV thermal neutron, a 1 MeV fast neutron, and a 100 GeV proton, and read what the ceiling teaches.

Step 1. Thermal neutron, $E=25$ meV: $k=3.47\times10^{10}$ m⁻¹, ceiling $4\pi/k^2=\mathbf{1.04\times10^8\ b}$ . This is why cold neutrons can have monstrous cross sections — the ceiling itself is astronomical.

Step 2. Fast neutron, $E=1$ MeV: $k$ is $\sqrt{4\times10^7}\approx6300\times$ larger, so the ceiling drops by that factor squared to $\mathbf{2.6\ b}$ — nuclear-geometric scale. The same physics, six orders of magnitude of ceiling apart.

Step 3. 100 GeV proton: $k=511$ fm⁻¹, ceiling $\mathbf{4.8\times10^{-4}\ mb}$ . At collider energies a single partial wave scatters almost nothing; real hadronic cross sections come from summing thousands of partial waves ( $\ell_{\max}\approx kR$ ).

Step 4. The lesson, in one number: the ceiling collapses as $1/k^2$ , so every enormous low-energy cross section is a low- $k$ ceiling being approached, never a violation of unitarity. Answer: ceilings $1.04\times10^8$ b / 2.6 b / $4.8\times10^{-4}$ mb — the $1/k^2$ collapse is why "large cross section" always means "low energy." Cadmium's 20,600 b (Example resolved above) is one thermal s-wave riding $0.02\%$ of the first ceiling. Every number in both examples is computed in the verify script (checks 12, 24–31).

Origins · 1 minS-Matrix and how this was discovered

S-Matrix

1937–1961 · Princeton, Leipzig

The scattering matrix was christened by John Wheeler in 1937, in a paper on the resonating-group structure of light nuclei, where "scattering matrix" organized the amplitudes connecting incoming and outgoing channels. Werner Heisenberg made it a program: in 1943 he argued that a fundamental theory should be built only from observable scattering data — the S-matrix — bypassing the unobservable wavefunction, an observables-only manifesto that echoed for decades. Christian Møller supplied the machinery in 1945, defining the wave operators $\hat\Omega_\pm$ that give $\hat S=\hat\Omega_-^\dagger\hat\Omega_+$ its rigorous meaning. The program peaked around 1961 with Geoffrey Chew's bootstrap — the claim that unitarity and analyticity alone could fix the strong interaction. The bootstrap failed as a theory of everything but left the S-matrix as the permanent language of scattering; today's amplitude-bootstrap revivals are its direct heirs.

History

1937

— Wheeler introduces the "scattering matrix" in his resonating-group analysis of nuclei (Phys. Rev. 52, 1107)

1943

— Heisenberg proposes the S-matrix as the foundation of a theory built only from observables (Z. Phys. 120)

1945

— Møller defines the wave operators, giving S = Ω₋†Ω₊ its rigorous meaning

1961

— Chew's bootstrap program attempts to fix the strong interaction from unitarity and analyticity alone

Going further

References. Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §6.4 · Shankar, Principles of Quantum Mechanics, 2nd ed., §19.5 · Griffiths, Introduction to Quantum Mechanics, 3rd ed., §11.4 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. II, Ch. VIII · Taylor, Scattering Theory, Ch. 3 · Newton, Scattering Theory of Waves and Particles, 2nd ed., Ch. 9.

Provenance. All equations are standard scattering-theory results. A verifier should confirm: (1) $S^\dagger S=\mathbb{1}$ and $S=1+iT$ giving $-i(T-T^\dagger)=T^\dagger T$ (Sakurai §6.4, Taylor Ch.3); (2) $S_\ell=e^{2i\delta_\ell}$ and $f_\ell=(S_\ell-1)/(2ik)=e^{i\delta_\ell}\sin\delta_\ell/k$ (Shankar §19.5, Griffiths §11.4); (3) optical theorem $\sigma_{\mathrm{tot}}=(4\pi/k)\mathrm{Im}\,f(0)$ (all texts); (4) inelastic formula $\sigma_\ell^{\mathrm{reac}}=(\pi/k^2)(2\ell+1)(1-\eta_\ell^2)$ (Cohen-Tannoudji Vol.II Ch.VIII); (5) unitarity limit $\sigma_\ell^{\max}=(4\pi/k^2)(2\ell+1)$ at $\delta_\ell=\pi/2$ ; (6) bound-state poles at $k=i\kappa$ , $E=-\hbar^2\kappa^2/2m$ (Newton, Taylor). Numerical relations (black-disk $\sigma_{\mathrm{el}}=\sigma_{\mathrm{reac}}=\pi R^2$ , $\sigma_{\mathrm{tot}}=2\pi R^2$ ) are standard. No nonstandard numbers introduced. (Lane note: this article owns the flux-forces-unimodularity theorem, the operator $S^\dagger S=\mathbb 1$ statement, the unitarity circle and the ceiling theorem, all machine-verified. A base sign error was found and fixed: the export's unitarity condition $|a|^2=-\mathrm{Im}\,a$ and the expansion $|1+2ia|^2=1+4\mathrm{Im}\,a$ are sign-flipped — the correct relation is $\mathrm{Im}\,a=|a|^2$ , $|a-i/2|=1/2$ , verified by the resonance-point test $a=i$ . The forward assembly of item (3) is Optical Theorem's; the inelastic item (4) and black-disk numbers are Inelastic Scattering and Multichannel Theory's; the pole taxonomy of item (6) is Bound States and Zero-Energy Resonances'.)

Builds on: Asymptotic Form of the Scattering Wave Function · Scattering Amplitude · Partial-Wave Expansion · Phase Shifts · Lippmann-Schwinger Equation Leads to: Optical Theorem · Resonances and the Breit-Wigner Formula · Bound States and Zero-Energy Resonances · Born Series and Higher-Order Corrections

Why is unitarity the same thing as probability conservation?

Because the S-matrix maps the incoming state to the outgoing one, and total probability is the norm of the state. If $\hat S$ preserves norm for every input — $\langle\psi|\hat S^\dagger\hat S|\psi\rangle=\langle\psi|\psi\rangle$ — then $\hat S^\dagger\hat S=\mathbb 1$ , which is the definition of unitary. Per channel this collapses to $|S_\ell|=1$ : the amount of wave flowing out equals the amount flowing in. When it fails, $|S_\ell|<1$ , probability has gone somewhere the elastic channel cannot see — into an inelastic channel or an absorptive process — and the deficit is exactly the reaction cross section.

If S_ℓ = e^{2iδ_ℓ} has only one real parameter, why bother with the S-matrix — isn't δ_ℓ enough?

For a single elastic channel, yes — $\delta_\ell$ and $S_\ell$ carry identical information, which is exactly this article's point: unitarity leaves one real degree of freedom per channel. The S-matrix earns its keep with several channels, where $S$ is a genuine matrix, unitarity $S^\dagger S=\mathbb 1$ couples them, and $|S_{\ell\ell}|<1$ measures the leak. It also earns it analytically — $S_\ell(k)$ continued to complex $k$ turns bound, virtual and resonant states into poles, information the real phase shift only hints at.

Does the unitarity circle mean the amplitude can never be large?

It bounds each partial wave: $|f_\ell|\le1/k$ , so $\sigma_\ell\le4\pi(2\ell+1)/k^2$ . The total amplitude can still be large by summing many channels — at high energy $\ell_{\max}\approx kR$ partial waves contribute and the geometric cross section $\sim\pi R^2$ emerges from thousands of terms each near its own small ceiling. What cannot happen is a single channel scattering more than its wavelength-set maximum; a computed $\sigma_\ell$ exceeding $4\pi(2\ell+1)/k^2$ is a sure sign of an error (a common one: forgetting to unitarize the low-energy $4\pi a_s^2$ formula once $ka_s\sim1$ ).

How does a resonance show up on the unitarity circle?

As a rapid counter-clockwise loop. Off resonance the amplitude sits near the bottom of the circle ( $\delta_\ell$ small); as the energy crosses $E_R$ the phase sweeps quickly through $\pi/2$ , driving the amplitude up to the top ( $f_\ell=i/k$ ) where scattering is maximal, then down the other side. The speed of the transit is the width: $\mathrm d\delta_\ell/\mathrm dE=2/\Gamma$ at the top. A slow, monotone crawl around the circle is a strong potential, not a resonance — the Argand loop's rate is what distinguishes them, and Resonances and the Breit-Wigner Formula turns that rate into the line shape and the Wigner time delay.

Where do bound states appear in the S-matrix?

As poles of $S_\ell(k)$ on the positive imaginary $k$ -axis: a bound state of binding energy $E_B=\hbar^2\kappa^2/2m$ is a pole at $k=+i\kappa$ , where the "outgoing" wave $e^{ikr}=e^{-\kappa r}$ becomes a normalizable tail. A pole at $k=-i\kappa$ (second energy sheet) is a virtual state. This analytic bridge between the continuous spectrum (real $k$ ) and the discrete spectrum (imaginary $k$ ) is a deep structural fact of scattering theory, and Bound States and Zero-Energy Resonances develops the full pole atlas.

Why is the S-matrix symmetric, and does it matter for central potentials?

Time-reversal invariance makes $S_{ba}=S_{ab}$ : the amplitude for $a\to b$ equals that for the reversed process $b\to a$ , the microscopic root of detailed balance. It follows from the antiunitary time-reversal operator (owned by Time-Reversal and Kramers Degeneracy). For a central potential each $S_\ell$ is a single number — a $1\times1$ block — so symmetry is automatic and carries no extra content. It becomes powerful with spin or multiple channels, where it relates genuinely different matrix elements and halves the number of independent amplitudes a phase-shift analysis must fit.