Born Series and Higher-Order Corrections

Lippmann-Schwinger Equation closed its operator equation $\hat T=\hat V+\hat V\hat G_0^{(+)}\hat T$ with a promise: "Iterating in powers of $\hat V$ generates the Born series (convergence anatomy: Born Series and Higher-Order Corrections)." This is that anatomy. Substituting the equation into itself unrolls scattering into a ladder — scatter once; scatter, fly freely, scatter again; and so on — with each rung one power of the potential and one free propagation. The first rung is the Born approximation. The rest of the ladder is this article: what the higher rungs mean, what they repair, and above all when the ladder can be climbed at all. Three owned results organize everything. The ladder is a geometric series in the operator $K=G_0V$ , its fate decided by one number — the spectral radius — and on the solvable separable model we measure the convergence ratio and find it equal to $|\lambda g_0(E)|$ to three digits, rung after rung. The wall is physical: the series diverges exactly where the potential binds (Jost–Pais), and because the radius falls with energy, a well that defeats the ladder at threshold surrenders at speed — we track the crossing to $k^*=0.65$ . And second order is where probability conservation returns: we prove $\mathrm{Im}f^{(2)}(0)=\frac{k}{4\pi}\sigma_B$ in general and verify it two independent ways, cashing the reality deficit Born Approximation left on the table. Two base-text errors — a second-Born prefactor violating this very identity by $-2\pi^2$ , and a dimensionally inconsistent convergence bound — are machine-proven and repaired in place. The exact forward statement remains Optical Theorem's; the resummed denominator's pole is one line here and Bound States and Zero-Energy Resonances' story there.

Key takeaways

Iterate $|\psi\rangle=|\phi\rangle+G_0V|\psi\rangle$ : the Born series $|\psi\rangle=\sum_n(G_0V)^n|\phi\rangle$ , $f=\sum_nf^{(n)}$ — rung $n$ is $n$ scatterings stitched by free propagators; rung 1 is the Born approximation.
The resummation is $\hat T=\hat V+\hat V\hat G_0\hat T$ (Lippmann-Schwinger Equation's object); the series is its geometric expansion in $K=G_0V$ , converging iff the spectral radius of $K$ stays below 1.
Measured on the separable model with 1009's exact closed form: error ratios $0.197/0.525/0.788$ rung over rung $=|\lambda g_0|$ exactly; divergence beyond, with the radius crossing 1 at $k^*=0.65$ for a binding-strength coupling — series diverge at low energy exactly when a bound state exists (Jost–Pais).
Second order repairs the first order's reality deficit: $\mathrm{Im}f^{(2)}(0)=\frac{k}{4\pi}\sigma_B^{(1)}$ , proven via the $i\pi\delta$ (on-shell) part of $G_0$ and verified on separable (exact) and local (two routes, $10^{-9}$ ) models.
Base repairs, machine-proven: the second-Born prefactor is $-\frac{m^2}{8\pi^4\hbar^4}$ (the base's $+\frac{m^2}{4\pi^2\hbar^4}$ violates the $V^2$ optical identity by exactly $-2\pi^2$ ); the workable convergence bound is Rollnik's $\left(\frac{m}{2\pi\hbar^2}\right)^2\iint\frac{|V||V'|}{|\mathbf r-\mathbf r'|^2}<1$ (the base's single-power form is dimensionally inconsistent).
Key idea: iterating $|\psi\rangle=|\phi\rangle+G_0V|\psi\rangle$ turns the implicit Lippmann-Schwinger equation into an explicit power series in $V$ : term $n$ is $n$ successive scatterings linked by free propagators, with the leading term being the Born approximation.

Where this lives today

The Born series is the template every modern expansion answers to. Nuclear reaction codes run DWBA — the same ladder with distorted waves on the ends — auditing convergence exactly as Derivation C does, and switching to coupled-channels resummation where the radius creeps up. Electron–molecule and positron communities publish explicit second-Born calculations, the $\mathrm{Im}f^{(2)}$ of Derivation B carrying the absorption first order misses. In many-body physics the geometric structure returns wearing diagrams: ladder and RPA resummations are the statement "our $K$ has an eigenvalue near 1, sum that channel exactly" — Weinberg's quasiparticle logic in Feynman dress. And modern scattering numerics has largely conceded the point of this article: production codes solve the Lippmann–Schwinger equation by grid inversion (1009's fourteen-digit demonstration scaled up) precisely because the resummed equation is indifferent to the wall that kills its own series.

The physical picture

Read the Lippmann–Schwinger equation as a feedback loop. The beam enters; the potential converts some of it into scattered wave; the scattered wave flies freely — that is $G_0$ — and re-enters the potential, which converts it again. The Born series is this loop unrolled: rung $n$ makes $n$ visits. Whether unrolling is legitimate is the feedback question: after one scattering and one free flight, does the wave return weaker than it came? If every cycle attenuates — $\|G_0V\|<1$ — the ladder converges geometrically and each rung is a genuine correction. If a cycle amplifies, the wave builds up inside the potential and no finite number of single-scattering stories describes it: the potential is trying to trap the particle. A bound state is feedback that closed on itself; divergence of the Born series is its scattering-room echo.

Figure

The ladder: one, two, three scatterings stitched by free propagators — and the feedback question that decides convergence

Energy tilts the balance. At high energy $G_0$ oscillates between visits, scrambling the phases of successive scatterings — the ladder converges even for binding potentials. At low energy the propagator is patient; a well that binds feeds the wave back constructively and the series is lost. One coupling, two fates — we will measure the crossover. The rungs are not decoration even when tiny. The first is real for a real potential — Born Approximation proved it cannot dim the beam it scatters. The second rung's on-shell piece is where the beam's shadow begins: its imaginary forward part comes from the $i\epsilon$ pole of $G_0$ — intermediate states that conserve energy — and balances the first rung's cross section to the digit. Unitarity, in perturbation theory, is paid one order late.

The series, assembled

Write the Lippmann–Schwinger equation abstractly, $|\psi\rangle=|\phi\rangle+G_0^{(+)}V|\psi\rangle$ , with $G_0^{(+)}=(E-H_0+i\epsilon)^{-1}$ the outgoing resolvent whose credentials — pole prescription, position-space kernel — belong to Green Functions in Quantum Mechanics and Green's Function for Scattering and are consumed here whole.

|\psi^{(+)}\rangle=\sum_{n=0}^{\infty}\left(G_0^{(+)}V\right)^n|\phi\rangle \qquad (1)

where: the $n$ -th term makes $n$ visits to $V$ stitched by $n$ free flights (the last flight is the trip to the detector, supplied by the asymptotics of Green's Function for Scattering). Projecting the ladder onto the far field converts it into a series for the amplitude, $f=\sum_{n\ge1}f^{(n)}$ , whose first term is Born Approximation's Fourier transform:

f^{(1)}=-\frac{m}{2\pi\hbar^2}\,\tilde V(\mathbf q),\qquad f^{(2)}=-\frac{m^2}{8\pi^4\hbar^4}\int\mathrm d^3k''\,\frac{\tilde V(\mathbf k'-\mathbf k'')\,\tilde V(\mathbf k''-\mathbf k)}{k^2-k''^2+i\epsilon} \qquad (2)

where: $\mathbf k''$ runs over all intermediate momenta — off-shell ones included: between visits the particle borrows energy freely, and only the $i\epsilon$ -tagged shell $k''=k$ contributes an imaginary part. ⚠ Base repair (machine-proven): the base text printed this prefactor as $+\frac{m^2}{4\pi^2\hbar^4}$ . Track the constants — $f^{(2)}=-\frac{m}{2\pi\hbar^2}(2\pi)^3\langle\mathbf k'|VG_0V|\mathbf k\rangle$ with $\langle\mathbf k'|V|\mathbf k''\rangle=\tilde V/(2\pi)^3$ and $E-E''=\frac{\hbar^2}{2m}(k^2-k''^2)$ — and the prefactor is $-\frac{m^2}{8\pi^4\hbar^4}$ : negative, and smaller by $2\pi^2$ . The error is not cosmetic: Derivation B's identity $\mathrm{Im}f^{(2)}(0)=\frac{k}{4\pi}\sigma_B$ — which the base itself states — is violated by exactly $-2\pi^2$ with the printed constant (verify script, symbolic). An equation that contradicts its own consistency check three lines later is the AllQFT corpus in one lesson: rederive, then trust. The whole ladder resums into the transition operator of Lippmann-Schwinger Equation — "keeping $\hat T\approx\hat V$ is the Born approximation"; keeping everything is

\hat T=\hat V+\hat V\hat G_0^{(+)}\hat T=\sum_{n=0}^{\infty}\hat V\left(\hat G_0^{(+)}\hat V\right)^n \qquad (3)

where: the closed equation on the left is stronger than the series on the right — 1009 solved it exactly where we will shortly watch the series die. One contrast is worth a line: the time-dependent Dyson expansion of Time-Ordered Exponential and the Dyson Series carries $1/n!$ factors tamed by time-ordering; here there is no time, no ordering, no factorial — the resolvent ladder is a bare geometric series, which is exactly why a single number will control it.

Derivation AUnrolling the equation and reading each rung

The goal is the Born series (1)–(3) with its structure made inevitable rather than notational: why iteration produces a geometric ladder, what a rung means physically, and why the partial sums obey $S_N=(1-K^N)(1-K)^{-1}|\phi\rangle$ -form bookkeeping that Derivation C will convert into a sharp convergence verdict. The strategy is two substitutions and one projection, with the operator $K\equiv G_0^{(+)}V$ isolated as the object everything depends on.

Step 1. Substitute the equation into itself once: $|\psi\rangle=|\phi\rangle+K|\phi\rangle+K^2|\psi\rangle$ . The remainder after $N$ substitutions is exact: $|\psi\rangle=\sum_{n=0}^{N-1}K^n|\phi\rangle+K^N|\psi\rangle$ . Why keep the remainder: truncation error is not mysterious — it is $K^N|\psi\rangle$ , the exact state after $N$ feedback cycles. If repeated application of $K$ shrinks vectors, the remainder dies and the series converges to $\psi$ ; if not, no amount of patience helps. Convergence of the Born series is a statement about the operator $K$ , full stop.

Step 2. Read rung $n$ : $K^n|\phi\rangle=(G_0V)\cdots(G_0V)|\phi\rangle$ — the beam hits $V$ ( $V|\phi\rangle$ ), propagates freely ( $G_0$ ), hits again, $n$ times. In momentum space each $G_0$ is diagonal, $\frac{1}{E-E_{k''}+i\epsilon}$ , and each $V$ is a Fourier kernel $\tilde V(\Delta\mathbf k)/(2\pi)^3$ — inserting complete sets gives Eq. (2) for $n=2$ , with all constants pinned in the verify script (the base's prefactor fails that pinning; see above).

Step 3. Project onto the far field. Applying the asymptotic reduction of Green's Function for Scattering to the last propagator of every rung turns the state ladder into the amplitude ladder $f=\sum f^{(n)}$ , $f^{(n)}=-\frac{m}{2\pi\hbar^2}\langle\mathbf k'|V(G_0V)^{n-1}|\mathbf k\rangle(2\pi)^3$ . Defining $T|\phi\rangle=V|\psi\rangle$ and applying $V$ to Step 1's identity delivers $T=V+VG_0T$ and its iterates — Eq. (3), the resummation. Why "geometric" is not a metaphor: on the separable model $\hat V=\lambda|g\rangle\langle g|$ , the operator $K$ has a single relevant eigenvalue $\lambda g_0(E)$ , and the amplitude ladder is literally $f_B\sum_n(\lambda g_0)^n$ — the partial-sum identity $\sum_{n<N}x^n=(1-x^N)/(1-x)$ (verified symbolically) becomes the entire convergence story, with $x$ measurable. Derivation C measures it. Verified: partial-sum identity symbolic; rung-2 constants pinned against the $(2\pi)^{-3/2}$ conventions of Lippmann-Schwinger Equation; the ladder's first term reproduces Born Approximation's $f_B$ (verify checks 4–8).

Second order pays the unitarity debt

Born Approximation ended on a deficit: $f^{(1)}$ is real, yet $\sigma_B>0$ — the beam scatters without dimming. The debt falls due at $O(V^2)$ , and the second rung pays it exactly.

Go deeper · Derivation B · 1 minThe second Born term restores the optical identity at its own order

Derivation BThe second Born term restores the optical identity at its own order

The goal is the theorem $\mathrm{Im}f^{(2)}(0)=\frac{k}{4\pi}\sigma_B^{(1)}$ — the order- $V^2$ face of the forward-consistency relation whose exact statement belongs to Optical Theorem — derived in general and verified by machine two independent ways. The strategy: only the on-shell part of $G_0$ can generate an imaginary part; isolate it with the Sokhotski–Plemelj split that Green Functions in Quantum Mechanics owns, and watch the delta function rebuild $\sigma_B$ out of $|\tilde V|^2$ .

Step 1. In the forward element ( $\mathbf k'=\mathbf k$ ) of Eq. (2), split the propagator: $\frac{1}{k^2-k''^2+i\epsilon}=\mathrm P\frac{1}{k^2-k''^2}-i\pi\delta(k^2-k''^2)$ . The principal value is real and dresses $\mathrm{Re}f^{(2)}$ ; every drop of imaginary part comes from the delta — from intermediate states that conserve energy. Virtual visits shift; real visits absorb.

Step 2. Collapse the delta: $\int\mathrm d^3k''\,F(\mathbf k'')\,\delta(k''^2-k^2)=\frac{k}{2}\int F(k\hat n)\,\mathrm d\Omega''$ . With $F=|\tilde V(\mathbf k-\mathbf k'')|^2$ (forward element, real central $V$ ),

\mathrm{Im}f^{(2)}(0)=\frac{m^2}{8\pi^4\hbar^4}\cdot\pi\cdot\frac{k}{2}\int|\tilde V(\mathbf k-k\hat n)|^2\,\mathrm d\Omega''=\frac{k}{4\pi}\left(\frac{m}{2\pi\hbar^2}\right)^2\int|\tilde V|^2\,\mathrm d\Omega''=\frac{k}{4\pi}\,\sigma_B^{(1)}

Why this is inevitable: the delta puts the intermediate particle on shell in direction $\hat n$ — the second-order forward amplitude interferes the beam with a wave that really scattered into $\hat n$ and back. Summing over $\hat n$ counts every particle first order removed from the beam. The shadow is multiple scattering's first installment.

Step 3. Machine verification, twice. Separable model: $\mathrm{Im}f^{(2)}=\mathrm{Im}[f_B\cdot\lambda g_0]=k|f_B|^2$ exactly (the $\mathrm{Im}\,g_0=-k|g(k)|^2/2\pi$ of 1009's served card doing the work) — residual $10^{-14}$ . Local square well: left side assembled from channel squares $\sum_\ell(2\ell+1)(\delta_\ell^{B})^2/k$ , right side from the direct angular integral of $|f_B|^2$ — two routes sharing no code, equal to $10^{-9}$ . Verified: the symbolic prefactor identity (and the base constant's $-2\pi^2$ violation of it), the separable exactness, and the two-route local check are verify checks 19–25.

Order bookkeeping closes: at $O(V)$ the amplitude is real and nothing needs balancing; at $O(V^2)$ , $\mathrm{Im}f^{(2)}(0)$ balances $\sigma_B^{(1)}$ ; at $O(V^3)$ , $\mathrm{Im}f^{(3)}$ balances the $f^{(1)*}f^{(2)}$ cross terms — each order's imaginary forward part audits the previous order's counting. Truncate consistently and no experiment at that accuracy catches you. The all-orders statement — one line, exact, no series — is Optical Theorem's.

The convergence wall, located and measured

Go deeper · Derivation C · 2 minThe spectral radius, the bound-state wall, and a bound you can actually check

Derivation CThe spectral radius, the bound-state wall, and a bound you can actually check

The goal is the convergence verdict: the Born series converges iff the spectral radius of $K(E)=G_0^{(+)}(E)V$ is below one; the physical translation (divergence ⟺ binding, energy-dependently); and a checkable sufficient condition, repaired from the base's dimensionally inconsistent form. The strategy: Step 1's exact remainder needs $K^N\to0$ , which for a compact kernel is precisely spectral radius $<1$ ; the separable model makes every abstraction a number; the Hilbert–Schmidt norm gives the practical certificate.

Step 1. From Derivation A, the truncation error is $K^N|\psi\rangle$ . $K^N\to0$ iff every eigenvalue $\kappa_i(E)$ of $K$ has $|\kappa_i|<1$ — the spectral radius condition. On the separable model the spectrum is one number, $\kappa=\lambda g_0(E)$ , so the theory predicts error ratios of exactly $|\lambda g_0|$ rung over rung. Measured (verify, $k=1$ ): ratios $0.197,\ 0.525,\ 0.788$ at $\lambda/\lambda_c=0.3,\ 0.8,\ 1.2$ — three digits of agreement with $|\lambda g_0|$ , rung after rung.

Step 2. The wall is a bound state. An eigenvalue $\kappa(E)=1$ at negative energy means $G_0V|\chi\rangle=|\chi\rangle$ , i.e. $(E-H_0)|\chi\rangle=V|\chi\rangle$ — a genuine bound state at that energy (1009's "denominator heads for zero... a bound state being born," now read as a convergence obituary). Jost and Pais made the connection a criterion (1951): for attractive potentials, the low-energy Born series converges only if the potential is too weak to bind. But the radius is energy-dependent: $|g_0(E)|$ falls as $E$ grows (the oscillating propagator scrambles feedback), so the same coupling that kills the series at threshold releases it at speed. Measured on the model at $\lambda=1.2\lambda_c$ — a coupling that binds: radius $1.20$ at $k=0.05$ , crossing $1$ at $k^*=0.653$ , down to $0.14$ by $k=2.5$ . Below $k^*$ the partial sums visibly blow up ( $S_{10}$ error exceeds $S_5$ 's, check 18); above, geometric convergence resumes. Divergence is not a pathology — it is the series honestly reporting physics (a bound state) that no finite sum of single-scattering stories contains.

Step 3. A certificate you can integrate. The spectral radius is bounded by the Hilbert–Schmidt norm of the symmetrized kernel $|V|^{1/2}G_0|V|^{1/2}$ , giving the Rollnik condition (at $E=0$ , the worst case for attraction):

\left(\frac{m}{2\pi\hbar^2}\right)^2\iint\frac{|V(\mathbf r)|\,|V(\mathbf r')|}{|\mathbf r-\mathbf r'|^2}\,\mathrm d^3r\,\mathrm d^3r'\;<\;1

⚠ Base repair (machine-proven): the base printed this bound with a single power of $\frac{m}{2\pi\hbar^2}$ and a single power of $|\mathbf r-\mathbf r'|$ in the denominator — an expression with dimensions energy×volume, which no inequality against 1 can survive (dimensional audit in the verify script). The repaired form is dimensionless and standard. For the uniform ball the double integral evaluates to exactly $4\pi^2a^4$ (a small delight: measured $39.4784$ ), so the certificate reads $\frac{2m|V_0|a^2}{\hbar^2}<2$ — conservative against the true binding threshold $\pi^2/4\approx2.47$ , as a sufficient condition should be. Verified: measured ratios vs $|\lambda g_0|$ (checks 12–14), the $k^*$ crossing and low- $k$ divergence (15–18), both dimensional verdicts and the ball integral $=4\pi^2$ (26–28).

Figure

Punchline — the ladder measured: geometric error collapse below the wall, divergence above it, and the energy where the wall moves out of the way

Limiting cases

case	result	meaning
weak coupling, any $E$	radius $\ll1$ ; $f\approx f^{(1)}$	Born approximation's home turf; error ratio $=$ radius (measured 0.197)
high energy, any fixed $V$	$\|g_0(E)\|\to0$ ⇒ radius $\to0$	oscillating $G_0$ scrambles feedback; series converges even for binding wells
coupling $\to$ binding threshold at $E\to0$	radius $\to1$ ; convergence dies at low $E$	Jost–Pais wall; the resummed $T$ stays finite — the series fails, not the physics
coupling binds, $E$ raised	radius crosses 1 at $k^*$ (measured 0.653)	energy-dependent convergence: diverges at threshold, converges at speed
truncation at order $N$ , radius $r<1$	error $\sim r^N$ geometric	measured collapse: $52.5\%\to27.6\%\to4.0\%\to0.16\%$ at $r=0.525$
$O(V^2)$ forward	$\mathrm{Im}f^{(2)}(0)=\frac{k}{4\pi}\sigma_B$	unitarity restored order by order; exact statement is Optical Theorem's

Worked examples

Example 1 — climb the ladder where it converges. Separable model at $k=1$ , coupling $\lambda=0.8\lambda_c$ (strong but not binding): radius $r=|\lambda g_0|=0.525$ .

Step 1. Exact answer from the resummed denominator (1009's closed form): $f=-\lambda g^2/2\pi(1-\lambda g_0)$ .

Step 2. Partial sums against it: $S_1$ (Born) errs by 52.5% — half wrong, as $r\simeq0.5$ promises. $S_2$ : 27.6%. $S_5$ : 4.0%. $S_{10}$ : 0.16%.

Step 3. Each rung multiplies the error by $0.525$ (measured to three digits) — geometric, exactly as the spectral analysis dictates. Ten rungs buy three digits; the denominator buys them all at once.

Example 2 — audit a deuteron-class well before summing anything. Neutron–proton scattering at thermal energy, modeled as a square well $V_0=35$ MeV, $a=2.1$ fm, reduced mass $\mu c^2=469.5$ MeV.

Step 1. Binding audit: $s=2\mu V_0a^2/\hbar^2=3.72>\pi^2/4=2.47$ — the well binds (it is the deuteron). Jost–Pais verdict: the low-energy Born series diverges. No rung count helps.

Step 2. Certificate audit: Rollnik/HS norm $=\frac{2\mu V_0a^2}{\hbar^2}\cdot\frac{\sqrt{4\pi^2}a^2}{4\pi a^2}\cdot1=\frac{s}{2}=1.86>1$ — the sufficient condition refuses to certify, consistently.

Step 3. The escape routes, by name: resum (the separable/T-matrix route of Lippmann-Schwinger Equation), or change basis — at these energies one partial wave carries everything, and Phase Shifts plus Low-Energy Scattering and the Scattering Length solve in that basis exactly. Answer: for np at thermal energy the ladder is the wrong tool, provably, before any term is computed. All numbers machine-computed (verify checks 29–31).

Origins · 1 minBorn Series and how this was discovered

Born Series

1877–1951 · Göttingen, Princeton

The series is older than quantum mechanics. Carl Neumann's 1877 method solved integral equations by iterating the kernel — the geometric series on operators — decades before anyone had a wave function to iterate; Fredholm's theory then said when such ladders converge. Born's 1926 collision papers imported the scheme into the new mechanics: his first-order term became "the Born approximation," and the tower above it inherited his name, mathematics included. The physics question — when does the ladder converge for an actual potential? — waited until 1951, when Res Jost and Abraham Pais tied divergence at low energy to the existence of bound states: a well that traps cannot be summed into. That result reframed the series from a computational convenience into a diagnostic — divergence means something — and the postwar decades built accordingly: resummations (the $T$ -matrix denominators of this chapter), rearrangements like Weinberg's quasiparticle method that repair convergence by extracting the offending eigenvalue, and the modern habit, inherited by every diagrammatic expansion in many-body and field theory, of asking first where the geometric ladder breaks.

History

1877

— Neumann's iterated-kernel series solves integral equations by geometric expansion

1926

— Born's collision papers make the first rung of the iteration the working tool of quantum scattering

1932

— Feenberg's forward-consistency analysis points at what any single rung must miss

1951

— Jost and Pais prove the low-energy series diverges precisely when the potential binds

1963

— Weinberg's quasiparticle rearrangement repairs divergent ladders by extracting the guilty eigenvalue

Going further

References. Griffiths 3rd ed. Ch. 11 · Sakurai & Napolitano 2nd ed. Ch. 6 · Shankar 2nd ed. Ch. 19 · Cohen-Tannoudji, Diu & Laloë Vol. II Ch. VIII · Newton, Scattering Theory of Waves and Particles, Ch. 9 · Jost & Pais, Phys. Rev. 82, 840 (1951).

Provenance. Equations cross-checked against Griffiths (Ch. 11), Sakurai & Napolitano (Ch. 6), Shankar (Ch. 19), Cohen-Tannoudji (Vol. II, Ch. VIII), and Newton's Scattering Theory. A verifier should recompute: (i) the Yukawa Fourier transform $4\pi/(q^2+\mu^2)$ and the first Born amplitude $-2mV_0/[\hbar^2(q^2+\mu^2)]$ ; (ii) the Coulomb/Rutherford limit $\mathrm{d}\sigma/\mathrm{d}\Omega=(2mV_0/\hbar^2 q^2)^2$ ; (iii) the Neumann-series convergence bound $\|K\|^N/(1-\|K\|)$ ; (iv) consistency of $\mathrm{Im}\,f^{(2)}(0)$ with the optical theorem at order $V^2$ . No experimental numbers are quoted; only standard textbook closed forms appear. (Lane note: items (iii) and (iv) are executed and machine-verified here — (iii) as the measured geometric ratios and remainder bookkeeping, (iv) as Derivation B two independent ways. Items (i)–(ii), the Yukawa/Rutherford evaluations, are chapter property of Scattering by a Screened Coulomb Potential and are performed there. TWO base errors repaired in this article, both machine-proven: the second-Born prefactor and the convergence bound's form — see the marked cards.)

Builds on: Lippmann-Schwinger Equation · Green's Function for Scattering · Born Approximation · Scattering Amplitude · Asymptotic Form of the Scattering Wave Function Leads to: The S-Matrix and Unitarity in Scattering · Optical Theorem · Scattering by a Screened Coulomb Potential · Resonances and the Breit-Wigner Formula · Phase Shifts

Is the Born series just a Taylor series in the potential strength?

Formally yes — but the useful reading is operator-geometric: it is the Neumann series of the kernel $K(E)=G_0^{(+)}V$ , and convergence is governed by $K$ 's spectrum at the energy you scatter at, not by any coupling's smallness alone. That is why one potential can have a convergent series at one energy and a divergent one at another (measured: $k^*=0.653$ ) — behavior a naive "small $\lambda$ " picture cannot express.

Does the series converge for the Coulomb potential?

No — and it fails at every order, not just in the sum. The Coulomb tail makes each Born term beyond the first ill-defined (the intermediate integrals diverge at small momentum transfer), reflecting the fact that a $1/r$ potential never stops distorting the wave, so the plane-wave in/out states of the expansion are the wrong asymptotics — Asymptotic Form of the Scattering Wave Function's logarithmic-phase honesty card. Screen the potential and everything is finite order by order; the screened evaluation and its Rutherford limit live in Scattering by a Screened Coulomb Potential.

If the second Born term fixes unitarity, why not always stop at two terms?

Second order fixes unitarity at second order only — the $O(V^3)$ ledger opens immediately, $\mathrm{Im}f^{(3)}$ balancing $f^{(1)*}f^{(2)}$ cross terms. Two rungs are excellent when the radius is small (error $\sim r^2$ : measured $27.6\%$ at $r=0.525$ ), but near the wall no fixed rung count survives — the measured $S_{10}$ worse than $S_5$ below $k^*$ is what that looks like. Near the wall: resum, or change basis (phase shifts).

The convergence criterion involves an operator norm I can't compute. What do I actually check?

Three practical layers. Cheapest: the binding audit — if the potential holds (or nearly holds) a bound state, assume the low-energy series is lost (Jost–Pais); for a square well that is one inequality, $2\mu V_0a^2/\hbar^2$ vs $\pi^2/4$ . Next: the Rollnik integral of Derivation C — two quadratures, dimensionless, sufficient; for the uniform ball it collapses to $2\mu V_0a^2/\hbar^2<2$ . Sharpest: compute the eigenvalues of the discretized kernel $G_0V$ on a momentum grid — the same grid 1009 used to invert the equation — and read the spectral radius directly. The verify script does all three.

Why doesn't the time-dependent Dyson series have these convergence problems?

It has its own, but of a different species. The time-ordered exponential of Time-Ordered Exponential and the Dyson Series carries $1/n!$ from the shrinking time-simplex volume, so for bounded $H'$ on finite times it converges like an exponential series — always. The Born series has no simplex and no factorial: it is a bare geometric series over infinite time (stationary scattering), and infinite time is exactly what lets a would-be bound state feed back without limit. Same iteration idea; the factorial is the difference between an entire function and a geometric wall.

What does the divergence physically correspond to if the exact amplitude stays finite?

The series treats multiple scattering as a small correction to single scattering; divergence certifies that ordering is false — the wave's residence in the potential is dominated by arbitrarily many revisits. The denominator $1-\lambda g_0$ reorganizes those revisits into one object whose zero is the bound state (1009's pole line, developed in Bound States and Zero-Energy Resonances). Nothing observable blows up — what blows up is the pretense that the beam's story can be told one scattering at a time.