The AMPS Entanglement Argument

The AMPS argument is a contradiction among three entanglement requirements for an old evaporating black hole. A late exterior mode $B$ must carry correlations that lower the entropy of the early radiation $E$ on a descending Page branch; a smooth horizon requires $B$ to be nearly vacuum-entangled with an interior partner $A$; semiclassical effective field theory treats $E$, $B$, and $A$ as independent systems in one state. Strong subadditivity forbids all three to the asserted accuracy. The statement $S(EB)\lt S(E)$ says that $B$ helps purify the radiation; it does not say that one mode by itself purifies all of $E$.

The three names used throughout are early radiation $E$, late mode $B$, and interior partner $A$.

Required background. Quantum Information supplies purification, conditional entropy, trace distance, and strong subadditivity used in the proof.

Helpful background. Semiclassical Gravity supplies the old-hole regime; Quantum Fields in Curved Spacetime supplies the near-horizon vacuum pair; Black-Hole Information supplies the Page-curve and complementarity assumptions under test.

Systems and hypotheses

Work with a finite-dimensional wave-packet regulator and one normalized state $\rho_{EBA}$ on

$$\mathcal H_E\otimes\mathcal H_B\otimes\mathcal H_A.$$

$E$ is radiation emitted before the selected late step, $B$ is the outgoing late wave packet, and $A$ is its interior partner. The proof assumes:

1. Post-Page purification. Unitary evaporation makes the fine-grained radiation entropy decrease when $B$ is added, $$S(EB)\lt S(E).$$
2. Smooth horizon. The $AB$ state is close to the local infalling vacuum. In the ideal limit it is pure and entangled, so $S(AB)=0$ and $S(A)=S(B)\gt0$.
3. Near-horizon effective field theory. $E$, $B$, and $A$ are independent tensor factors, or commuting algebras with an entropy theorem of the same accuracy, on a common comparison slice.
4. Controlled interactions. Evolution between the state assignments and comparison does not produce entropy corrections comparable to the claimed gaps.

The first premise is stronger than “the full evaporation is unitary”: it asserts that this late mode participates in the descending Page branch. The second is stronger than low curvature: it asserts a quantitative vacuum-entanglement relation. The third is where gravitational dressing or encoding proposals may enter.

Entropy contradiction

Strong subadditivity implies weak monotonicity for any tripartite state,

$$S(AB)+S(BE)\geq S(A)+S(E).$$

Define the radiation entropy decrease

$$\Delta=S(E)-S(EB)\gt0.$$

Substitution into weak monotonicity gives the necessary condition

$$S(AB)\geq S(A)+\Delta.$$

The smooth-horizon limit instead has $S(AB)=0$ and $S(A)\gt0$. This is an immediate contradiction. Equivalently, smoothness gives $S(AB)\lt S(A)$ while late purification gives $S(BE)\lt S(E)$; adding those strict inequalities reverses weak monotonicity.

This is the core argument in Almheiri et al. 2013, § 2, with typicality, greybody, mining, and backreaction qualifications discussed in Almheiri et al. 2013, §§ 3.1–3.2 and Appendices A–C. Harlow presents the same entropy architecture in Harlow 2016, § 7.1.

Approximate version

Real modes are not exact Bell pairs and gravitational factorization is not exact. Let $\sigma_{AB}$ be the regulated vacuum pair and define the trace distance by

$$T(\rho_{AB},\sigma_{AB}) =\frac12\lVert\rho_{AB}-\sigma_{AB}\rVert_1 \leq\varepsilon.$$

Assume $\varepsilon\leq\min\{1-1/d_A,1-1/d_{AB}\}$. For either relevant dimension $d$, define

$$f_d(\varepsilon) =\varepsilon\ln(d-1)+h_2(\varepsilon), \qquad h_2(\varepsilon) =-\varepsilon\ln\varepsilon-(1-\varepsilon)\ln(1-\varepsilon).$$

Fannes established finite-dimensional entropy continuity Fannes 1973, theorem; Audenaert’s sharp form gives $|S(\rho)-S(\sigma)|\leq f_d(\varepsilon)$ Audenaert 2007, Theorem 1. Trace distance contracts under reduction, so if the vacuum pair is pure and has $s_A^{\mathrm{vac}}=S(\sigma_A)$, then

$$S(AB)_\rho\leq f_{d_{AB}}(\varepsilon), \qquad S(A)_\rho\geq s_A^{\mathrm{vac}}-f_{d_A}(\varepsilon).$$

Now state the factorization uncertainty with its sign: suppose the usable weak-monotonicity relation is

$$S(AB)+S(BE)\geq S(A)+S(E)-\eta, \qquad \eta\geq0.$$

Together with $\Delta=S(E)-S(BE)\gt0$, consistency would require

$$f_{d_{AB}}(\varepsilon) \geq s_A^{\mathrm{vac}}-f_{d_A}(\varepsilon)+\Delta-\eta.$$

Consequently the assumptions are contradictory whenever

$$s_A^{\mathrm{vac}}+\Delta \gt f_{d_{AB}}(\varepsilon)+f_{d_A}(\varepsilon)+\eta.$$

This form names the trace-distance convention, finite dimensions, continuity constants, and allowed inequality error. A correction can resolve the contradiction only if it is large enough in the relevant entanglement measure or changes the subsystem premise. Calling a correction “nonperturbative” does not determine whether its entropy effect is sufficient.

The proof also assumes all quantities refer to one common state. Combining the $AB$ vacuum from one effective description with the $EB$ purification from a different, incompatible state would manufacture rather than solve a contradiction.

Finite-qubit check

Two explicit states isolate the premises. If

$$\rho_{EBA}=\lvert0\rangle\langle0\rvert_E \otimes\lvert\Phi^+\rangle\langle\Phi^+\rvert_{BA},$$

then the horizon is smooth in the toy sense, but $S(BE)=\ln2\gt S(E)=0$: adding $B$ raises radiation entropy. If instead

$$\rho_{EBA}=\lvert\Phi^+\rangle\langle\Phi^+\rvert_{EB} \otimes\lvert0\rangle\langle0\rvert_A,$$

then $S(BE)=0\lt S(E)=\ln2$, but $BA$ is not a smooth Bell pair. The models satisfy the same tensor factorization and show that either entanglement pattern is possible separately, not simultaneously.

A GHZ state supplies another adverse check: all one-qubit marginals are mixed, but no pair is pure and maximally entangled. “$B$ is correlated with both” is therefore not the same claim as the two near-maximal entanglements used by AMPS.

What each response changes

Response class	Premise changed	Remaining obligation
Fundamental loss or hidden final sector	Post-Page purification of complete radiation	Define the complete channel and correlation carrier
Firewall or drama	Smooth $AB$ vacuum entanglement	Specify excitation scale, formation time, conservation, and backreaction
Nonfactorization or exterior encoding	Independent $EBA$ factors	Construct the replacement algebra and reproduce semiclassical correlators
State-dependent interior	One state-independent $A$ operator on the whole state class	Prove linearity, overlap consistency, and Born-rule agreement on its domain
Operational complementarity alone	May change no entropy premise	Explain why a common-state inequality is inapplicable, not merely hard to measure

The table is not a verdict. AMPS proves that the assumption set is inconsistent; it does not prove that a unique firewall outcome follows. Each proposed response must be evaluated on the additional obligations it creates.

Scope and limitation

The proof is exact for finite factors and conditional in gravity. It does not establish that a factorization-violating completion exists, that a firewall is dynamically produced, or that an operational decoder can run. Generic entropy theory belongs to Quantum Information, the local vacuum to Quantum Fields in Curved Spacetime, and evaporation backreaction to Semiclassical Gravity. Explicit string or holographic constructions are separate.

Exercise

Suppose $S(A)=0.6$ nats, the radiation decrease is $\Delta=0.2$ nats, and smoothness permits only $S(AB)\leq0.1$ nats. Allow a weak-monotonicity error $\eta=0.02$ nats as defined above. Can the hypotheses hold together?

Solution

No. The allowed inequality requires $S(AB)\geq S(A)+\Delta-\eta=0.78$ nats, while smoothness allows at most $0.1$. A gap of $0.68$ nats remains after the stated error is included. To restore consistency, at least one input must change: the radiation need not decrease, the horizon pair need not be close to vacuum, or the common independent-factor inequality need not apply to the quoted accuracy.

References

• Almheiri, Ahmed, Donald Marolf, Joseph Polchinski, and James Sully. 2013. “Black Holes: Complementarity or Firewalls?” Journal of High Energy Physics 2013 (02): 062. Open text.
• Audenaert, Koenraad M. R. 2007. “A Sharp Continuity Estimate for the von Neumann Entropy.” Journal of Physics A: Mathematical and Theoretical 40: 8127–8136. DOI.
• Fannes, Mark. 1973. “A Continuity Property of the Entropy Density for Spin Lattice Systems.” Communications in Mathematical Physics 31: 291–294. DOI.
• Harlow, Daniel. 2016. “Jerusalem Lectures on Black Holes and Quantum Information.” Reviews of Modern Physics 88: 015002. Open text.