Threshold asymptotics and decay for massive Maxwell on subextremal Reissner--Nordström

Threshold asymptotics and deca y for massiv e Maxw ell on sub extremal Reissner–Nordström Bobb y Eka Gunara Theoretical Ph ysics Lab oratory , Theoretical High Energy Ph ysics Research Division, F acult y of Mathematics and Natural Sciences, Institut T eknologi Bandung, Jl. Ganesha no. 10 Bandung, Indonesia, 40132 Email: bobby@itb.ac.id Abstract W e study the neutral massiv e Maxwell (Proca) equation on sub extremal Reissner–Nordström exteriors. After spherical-harmonic decomp osition, the o dd sector is scalar, while the even sector remains a gen uinely coupled 2 × 2 system. Our starting point is that this ev en system admits an exact asymptotic polarization splitting at spatial inﬁnity . The three resulting channels carry eﬀectiv e angular momen ta ℓ − 1 , ℓ , and ℓ + 1 , and these are precisely the indices that go v ern the late-time thresholds. F or each ﬁxed angular momentum w e develop a threshold sp ectral theory for the cut-oﬀ resolven t. W e prov e meromorphic contin uation across the massiv e branch cut, rule out upp er-half-plane mo des and threshold resonances, and obtain explicit small- and large-Coulom b expansions for the branch-cut jump. In verting this jump yields p olarization-resolved in termediate tails together with the univ ersal very-late t − 5 / 6 branc h-cut law. At the full-ﬁeld level, high-order angular regularit y allows us to sum the mo dewise leading terms on compact radial sets and obtain a tw o-regime asymptotic expansion for the radiative branch-cut comp onen t of the Pro ca ﬁeld, with explicit co eﬃcien t ﬁelds and quantitativ e remainders. W e also analyze the quasib ound resonance branches created b y stable timelike trapping, prov e residue and reconstruction b ounds, and deriv e a fully self-contained dy adic pac ket estimate. As a result, the unsplit full Pro ca ﬁeld ob eys logarithmic compact-region decay , while the radiativ e branch-cut contribution retains explicit p olynomial asymptotics and explicit leading co eﬃcien ts. Con ten ts 1 In tro duction and main results 3 1.1 Bac kground and scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2 Relation to the literature and the main innov ation . . . . . . . . . . . . . 4 1.3 Notation and conv entions . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1 1.4 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.5 Self-con tained summation and pro of architecture . . . . . . . . . . . . . . 15 1.6 Organization of the pap er . . . . . . . . . . . . . . . . . . . . . . . . . . 15 1.7 Pro of dep endency diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 16 1.8 App endix guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 2 Geometry , mo de reduction, and p olarization splitting 16 2.1 Geometry and the ﬁeld equation . . . . . . . . . . . . . . . . . . . . . . . 16 2.2 Reduced odd/even mo de equations . . . . . . . . . . . . . . . . . . . . . 17 2.3 Exact asymptotic p olarization diagonalization . . . . . . . . . . . . . . . 18 3 Op erator-theoretic framew ork and limiting absorption 19 3.1 Channel Hamiltonians and asymptotic op erator structure . . . . . . . . . 19 3.2 Selfadjoin t realizations and energy spaces . . . . . . . . . . . . . . . . . . 20 3.3 Radial curren ts, W ronskians, and cut-oﬀ resolven ts . . . . . . . . . . . . 21 3.4 Limiting absorption aw ay from the thresholds . . . . . . . . . . . . . . . 22 3.5 Con tour deformation and the branch-cut form ula . . . . . . . . . . . . . 22 4 Channel resolv en ts and threshold represen tation 23 4.1 F ar-zone Coulomb normal form . . . . . . . . . . . . . . . . . . . . . . . 23 4.2 Green k ernels and branch-cut decomp osition . . . . . . . . . . . . . . . . 24 5 Pro of of the ﬁxed-mo de sp ectral theorem 24 6 Mo de stabilit y and threshold resonance exclusion 34 7 Explicit branc h-cut tails 35 7.1 Oscillatory in version formula and decomp osition of frequency space . . . 35 7.2 Sc h warzsc hild-to-Reissner–Nordström corresp ondence . . . . . . . . . . . 35 7.3 In termediate tails . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 7.4 The univ ersal very-late tail . . . . . . . . . . . . . . . . . . . . . . . . . . 37 8 Large-angular-momen tum analysis and branch-cut full-ﬁeld deca y 38 8.1 Large-angular-momen tum scaling and barrier geometry . . . . . . . . . . 38 8.2 Uniform WKB transp ort, cut-oﬀ resolv ents, and reconstruction . . . . . . 39 9 Quasib ound branc hes and residue b ounds 44 9.1 Guide to the pro of of the quasib ound theorems . . . . . . . . . . . . . . 44 9.2 The trapped well and semiclassical c hannel sym b ols . . . . . . . . . . . . 44 9.3 Bohr–Sommerfeld quan tization and quasib ound branc hes . . . . . . . . . 46 9.4 Residues and reconstruction of the ph ysical co eﬃcien ts . . . . . . . . . . 48 10 Summation of quasib ound residues and unsplit full-ﬁeld deca y 49 10.1 Guide to the pro of of the summed deca y theorem . . . . . . . . . . . . . 49 10.2 Contour decomp osition and the fast remainder . . . . . . . . . . . . . . . 50 10.3 Radiative proﬁles and pro of of the asymptotic expansion theorem . . . . 50 10.4 Explicit leading co eﬃcien t ﬁelds . . . . . . . . . . . . . . . . . . . . . . . 51 10.5 Dyadic pack ets and a self-contained damping estimate . . . . . . . . . . . 54 11 Outlo ok: extremality and rotation 57 2 A Harmonic reduction and b oundary constructions 57 B Threshold mo dels and ﬁxed-mo de sp ectral complements 61 C Large angular momenta, residues, and full-ﬁeld estimates 66 D A uxiliary regimes, monop ole analysis, and cutoﬀ indep endence 76 1 In tro duction and main results 1.1 Bac kground and scop e Massiv e ﬁelds on blac k-hole exteriors b eha v e diﬀeren tly from massless ones in t w o cou- pled w ays. The mass creates branc h p oin ts at ω = ± µ , so late times are gov erned b y oscillatory threshold tails rather than the familiar massless picture. A t the same time, it pro duces stable timelike trapping and therefore a discrete family of exp onentially long-liv ed quasib ound states. F or scalar ﬁelds on sub extremal Reissner–Nordström, this com bined picture is no w understo o d both mode b y mode and after summation o ver an- gular momenta [1, 2]. The Pro ca equation is the next natural mo del, but it is also the ﬁrst gen uinely v ectorial one: after spherical-harmonic reduction, the o dd sector is scalar whereas the even sector remains a coupled 2 × 2 system. Our goal is to giv e a rigorous late-time theory for the neutral Proca equation on sub extremal Reissner–Nordström. W e work on a ﬁxed exterior M > 0 , 0 ≤ | Q | < M , (1.1) with ev ent and Cauch y radii r ± = M ± q M 2 − Q 2 . (1.2) Unless stated otherwise w e restrict to angular momen ta ℓ ≥ 1 , so that the o dd channel and b oth even p olarizations are present. All p oin twise statemen ts are uniform on compact radial sets K ⋐ ( r + , ∞ ) and are form ulated either for a ﬁxed mo de or after summing o ver all angular momenta. The exceptional monop ole ℓ = 0 supp orts only the even electric c hannel and is treated separately in App endix D. W e exclude the extremal case | Q | = M , where the red-shift degenerates, the near-horizon AdS 2 × S 2 throat changes the threshold structure, and Aretakis-type horizon phenomena en ter [3]. What mak es Pro ca harder than the scalar problem is not a single complication but a com bination of three. First, the mass remo v es gauge freedom, so the ﬁeld itself is the natural unknown. Second, the even sector is genuinely matrix v alued at threshold. Third, the same p olarization structure m ust remain compatible with the large-angular- momen tum semiclassical analysis needed to con trol the quasib ound family . A full deca y theory must therefore connect ﬁxed-mo de threshold analysis, trapping-driven resonance theory , and unsplit full-ﬁeld reconstruction in one coheren t framew ork. This is exactly what w e do on sub extremal Reissner–Nordström. F or each ﬁxed an- gular momentum we pro ve a threshold sp ectral theorem for the o dd and ev en sectors, deriv e explicit p olarization-resolved branc h-cut asymptotics, sum those leading terms into full-ﬁeld radiativ e proﬁles with quantitativ e remainders, construct the quasibound p oles generated by stable timelik e trapping, and ﬁnally combine the contin uous and discrete sp ectral con tributions to obtain a self-contained compact-region deca y theorem for the full Pro ca ﬁeld. 3 1.2 Relation to the literature and the main inno v ation Earlier w ork already p oin ts tow ard the picture one should exp ect. F or Sch warzsc hild, Rosa and Dolan wrote the odd/even reduction for Pro ca explicitly [4], while K onoplya, Zhidenk o, and Molina predicted p olarization-dep enden t intermediate tails and the uni- v ersal very-late t − 5 / 6 la w by formal and numerical argumen ts [5]. The static Reissner– Nordström problem studied here is also the charged zero-rotation limit of the sepa- rated Kerr–Newman Pro ca system of [6]. On the scalar side, Pasqualotto, Shlap entokh- Rothman, and V an de Mo ortel established exact ﬁxed-mo de massive tails for stationary spherically symmetric black holes [1], and Shlap entokh-Rothman and V an de Mo ortel later prov ed compact-region decay on sub extremal Reissner–Nordström after summing o v er angular momen ta [2]. What was still missing w as a single rigorous framew ork that could sim ultaneously handle the massive branch cut, the vectorial p olarization structure, the quasib ound res- onance family created by stable timelike trapping, and the unsplit full-ﬁeld problem for Pro ca on a c harged black-hole bac kground. T o the b est of our knowledge, the present pap er is the ﬁrst to pro vide suc h a framework. The key new p oint is structural. The even Pro ca sector admits an exact asymptotic p olarization splitting at spatial inﬁnit y: once one passes from the standard pair ( u 2 , u 3 ) to the constan t com binations (1.3), the leading in v erse-square part diagonalizes in to the three eﬀectiv e angular momenta L − 1 = ℓ − 1 , L 0 = ℓ, L +1 = ℓ + 1 . This is the step that makes the problem manageable. It identiﬁes the correct thresh- old indices, sho ws that the c harge-dep endent coupling is shorter range than the leading asymptotic dynamics, and pro vides exactly the v ariables in whic h the large- ℓ semiclassical analysis can b e closed. Seen from that p ersp ectiv e, the contribution of the pap er is not just a collection of estimates. It is a single mec hanism that links threshold analysis, resonance theory , and full-ﬁeld reconstruction. Concretely , the main results may b e summarized as follows: (i) W e establish a matrix-v alued ﬁxed-mo de threshold theory for neutral Proca on sub extremal Reissner–Nordström, including meromorphic con tinuation of the cut- oﬀ resolven t across the massive branch cut, exclusion of upp er-half-plane mo des and threshold resonances, and explicit small- and large-Coulomb threshold expansions. (ii) W e derive explicit p olarization-resolv ed late-time asymptotics for the branch-cut con tribution, including the intermediate tails and the univ ersal v ery-late t − 5 / 6 la w. (iii) W e sum the mo dewise leading branc h-cut amplitudes to obtain a tw o-regime asymp- totic expansion for the radiative part of the full Pro ca ﬁeld on compact radial sets, with explicit leading co eﬃcien t ﬁelds and quan titative remainders in b oth the in- termediate and very-late regimes. (iv) W e construct the quasib ound resonance branc hes generated by stable timelik e trap- ping, together with uniform residue and reconstruction b ounds strong enough for angular summation. (v) W e combine the con tinuous and discrete sp ectral pieces to prov e a fully self- con tained compact-region deca y theorem for the unsplit Pro ca ﬁeld. The branch- cut part deca ys p olynomially and admits explicit asymptotic proﬁles, while the 4 discrete quasib ound con tribution is con trolled b y a logarithmic pac ket summation argumen t based only on tunnelling widths and high-order angular regularit y . A t the global summation stage the presen t version uses no imp orted pack et theorem. Instead w e work only with ingredients established in the pap er itself: t w o-sided tunnelling- width b ounds for the quasib ound p oles and p olynomial residue/reconstruction estimates strong enough to conv ert angular regularity of the data into arbitrary negativ e p ow ers of the pac ket scale. Summing the resulting pack et b ounds yields logarithmic deca y for the quasib ound contribution. If one later reinstates the arithmetic pac ket theorem of [2], then the sharp er p olynomial rates from the earlier draft reapp ear as an optional upgrade. 1.3 Notation and con v entions F or the reader’s conv enience we collect here all recurring notation used throughout the pap er. Auxiliary symbols that app ear only inside one proof (for example a temp orary cutoﬀ, a lo cal co eﬃcient matrix, or a short-liv ed current) are rein tro duced at ﬁrst ap- p earance. Bac kground parameters and geometry . The parameters M > 0 , Q ∈ R , and µ > 0 denote respectively the black-hole mass, the electric c harge of the background, and the Pro ca mass. Throughout the pap er we assume the sub extremal condition 0 ≤ | Q | < M . The horizon radii are r ± = M ± q M 2 − Q 2 , and the static co eﬃcient is f ( r ) = 1 − 2 M r + Q 2 r 2 = ( r − r + )( r − r − ) r 2 . The exterior region is { r > r + } . The tortoise co ordinate r ∗ is deﬁned by d r ∗ / d r = f ( r ) − 1 , so r ∗ → −∞ at the future even t horizon and r ∗ → + ∞ at spatial inﬁnit y . The surface gra vit y of the even t horizon is κ + := r + − r − 2 r 2 + > 0 . Whenev er K ⋐ ( r + , ∞ ) app ears, it denotes a ﬁxed compact radial set in the exterior. Field v ariables and harmonic lab els. The unkno wn A is the neutral Pro ca one-form, F = d A is its ﬁeld strength, and the ﬁeld equation is ∇ β F αβ + µ 2 A α = 0 . The angular mo de lab els are ( ℓ, m ) with ℓ ∈ N and m = − ℓ, . . . , ℓ . W e write λ ℓ := ℓ ( ℓ + 1) , λ := q λ ℓ when it is con venien t to use either λ ℓ or λ 2 in the reduced equations. The o dd radial amplitude is denoted by u 4 , the even amplitudes b y ( u 2 , u 3 ) , and the exceptional monop ole electric mo de at ℓ = 0 b y u 0 . Scalar spherical harmonics are written Y ℓm , while Y ( P ) ℓm denotes the p olarization-adapted v ector harmonics used in the full-ﬁeld reconstruction. 5 When a p oin t of the unit sphere is needed in point wise estimates w e write ϑ ∈ S 2 ; in a few form ulas the same role is play ed by an unadorned ω , whic h should then b e read as an angular v ariable rather than as a sp ectral frequency . P olarization v ariables. W e set v 0 := u 4 for the o dd channel and deﬁne the even p olarization basis b y v − 1 := ℓu 2 + u 3 2 ℓ + 1 , v +1 := ( ℓ + 1) u 2 − u 3 2 ℓ + 1 . The p olarization index is P ∈ {− 1 , 0 , +1 } , so that v P means v − 1 , v 0 , or v +1 according to the v alue of P . The corresp onding eﬀectiv e angular momen ta are L − 1 = ℓ − 1 , L 0 = ℓ, L +1 = ℓ + 1 , and w e often write this compactly as L P = ℓ + P . The constan t matrix T ℓ is the even- sector c hange of basis from ( u 2 , u 3 ) to ( v − 1 , v +1 ) . The diagonal matrix D ℓ = diag  L − 1 ( L − 1 + 1) , L +1 ( L +1 + 1)  is the leading r − 2 ev en potential in the polarization basis, while C ℓ is the shorter-range ev en coupling matrix entering at orders r − 3 and r − 4 . The ﬁnite-order diﬀerential op erator B ℓ denotes the reconstruction map from the scalarized c hannel v ariables bac k to the ph ysical Pro ca co eﬃcients. Op erators, energy spaces, and cutoﬀs. F or each ﬁxed ℓ ≥ 1 , the reduced o dd and even problems deﬁne the channel Hamiltonians H odd ℓ = − ∂ 2 r ∗ + V odd ℓ , H ev ℓ = − ∂ 2 r ∗ Id 2 + V ev ℓ , and we write H ♯ ℓ when a statement applies to either channel. The corresp onding L 2 spaces are h odd ℓ = L 2 ( R r ∗ ) , h ev ℓ = L 2 ( R r ∗ ; C 2 ) , while h ℓ denotes the ﬁxed-mo de ﬁnite-energy space generically . The selfadjoin t ﬁrst- order generator of the reduced time evolution is denoted b y A ℓ . Compactly supp orted radial cutoﬀs are written χ, e χ ∈ C ∞ 0 ( r + , ∞ ) . High-order initial-data norms are denoted b y E N [ A [0]] ; these are the Sob olev-t yp e mo de energies controlling ⟨ ℓ ⟩ N -w eigh ted angular sums. Sp ectral and threshold parameters. The complex sp ectral frequency is ω . When the op erator-theoretic sp ectral v ariable is needed w e write λ := ω 2 . Near the massive thresholds w e use ϖ := q µ 2 − ω 2 , κ := M µ 2 ϖ , with the branch of ϖ c hosen by the condition ℜ ϖ > 0 on {ℑ ω > 0 } . The notation disc means the jump across the cut [ − µ, µ ] , and Res denotes residue. The quan tit y ν ℓ,P is the threshold index determined by the in v erse-square co eﬃcient in the far-zone normal 6 form. The cut-oﬀ channel resolven t is written R ℓ,P ( ω ) , its Green kernel is G ℓ,P ( ω ; r , r ′ ) , the horizon and inﬁnity basis solutions are mark ed b y the subscripts hor and out , W ℓ,P denotes the asso ciated W ronskian, and E ℓ is the Ev ans determinant whose zeros coincide with p oles of the meromorphically con tin ued resolv en t. Threshold amplitudes and time scales. The functions a ℓ,P ( r , r ′ ) and b ± ℓ,P ( r , r ′ ) are the leading amplitudes in the small- κ and large- κ expansions of disc G ℓ,P . After inv erse F ourier transform, A ℓ,P ( r , r ′ ; Q ) and B ℓ,P ( r , r ′ ; Q ) denote the corresponding in termediate- and v ery-late-time amplitudes, while δ ℓ,P ( Q ) and δ ℓ,P, 0 ( Q ) are the asso ciated oscillatory phases. The small-mass parameter is ε µ,Q := ( M µ ) 2 + ( Qµ ) 2 . The t wo threshold time scales are κ ∗ ( t ) = M µ 3 / 2 t 1 / 2 , ϖ 0 ( t ) =  2 π M µ 3 t  1 / 3 , κ 0 ( t ) = M µ 2 ϖ 0 ( t ) . Time-domain decomp osition. F or eac h ﬁxed ( ℓ, m, P ) , u bc ℓm,P denotes the branch-cut con tribution after subtraction of discrete pole residues, u qb ℓm,P denotes the quasib ound con tribution, and u fast denotes the exp onen tially decaying remainder coming from p oles and con tour pieces b ounded a w ay from the real axis. A t the full-ﬁeld lev el w e write A = A bc + A qb + A fast , and sometimes u poles ℓm = u qb ℓm + u fast ℓm . The mo dal co eﬃcien ts in the branch-cut harmonic expansion are written a bc ℓm,P . The smallest threshold index is ν ∗ := inf ℓ ≥ 1 , P ∈{− 1 , 0 , +1 } ν ℓ,P , and the decay exp onents used later are γ ∗ := min  ν ∗ + 1 , 5 6  , γ qb := 5 6 − 1 23 , γ full := min { γ ∗ , γ qb } . Semiclassical and quasib ound notation. In the large-angular-momentum analysis w e use the semiclassical parameter h = ( ℓ + 1 2 ) − 1 . The compact in terv al I trap ⋐ (0 , µ ) is the trapped energy window, and I ⋐ I trap denotes a smaller ﬁxed interv al. The scalarized semiclassical c hannel op erators are written P h,P ( ω ) , and U h denotes the analytic diagonalizer for the ev en 2 × 2 system near the trapp ed region. The quasibound p oles are ω ℓ,n,P , indexed by in tegers n ∈ N ℓ,P ( I ) . The quan tities S P , J P , and d P are resp ectiv ely the w ell action, the tunnelling action, and the Agmon distance; ϑ P is the subprincipal phase correction; Γ P is the prefactor in the width form ula; and Π ℓ,n,P := − Res ω = ω ℓ,n,P R ℓ,P ( ω ) is the residue pro jector. Dy adic pack ets of poles are denoted b y P j . 7 Asymptotic notation and lo cal pro of symbols. W e write ⟨ x ⟩ = (1 + x 2 ) 1 / 2 , in particular ⟨ ℓ ⟩ = (1 + ℓ 2 ) 1 / 2 . The notation A ≲ B means A ≤ C B for a constant dep ending only on the background parameters, the Pro ca mass µ , the chosen compact set K , and ﬁnitely many cutoﬀ norms. All p oint wise estimates are uniform for r, r ′ ∈ K . Our in verse F ourier transform uses the phase e − i ω t . The remaining pro of-lo cal sym b ols, such as the radial current J , the threshold cutoﬀs η t, ± , and the b ounded co eﬃcient matrices B ℓ,j and E ℓ,j , are deﬁned again at ﬁrst app earance and are not reused outside their lo cal arguments. 1.4 Main results The results come in t wo la y ers. A t ﬁxed angular momentum w e dev elop the threshold sp ectral theory and read oﬀ the late-time tails from the branch cut. W e then lift this mo d- ewise information to the full ﬁeld, where one m ust con trol both large angular momenta and the quasib ound family generated b y stable timelike trapping. The ﬁrst structural fact is the asymptotic p olarization splitting at spatial inﬁnity . If ( u 2 , u 3 ) denotes the standard ev en pair in the v ector-spherical-harmonic decomp osition, then the constant combinations v − 1 = ℓu 2 + u 3 2 ℓ + 1 , v +1 = ( ℓ + 1) u 2 − u 3 2 ℓ + 1 (1.3) diagonalize the leading r − 2 ev en matrix exactly . The three resulting eﬀective angular momen ta are L − 1 = ℓ − 1 , L 0 = ℓ, L +1 = ℓ + 1 . (1.4) In this basis the charge Q enters only one order shorter range, at r − 4 , so the leading threshold dynamics separate cleanly in to three c hannels. The natural late-time ob ject is the branch-cut piece after the discrete p oles hav e b een remo v ed: u ℓm ( t ) = u poles ℓm ( t ) + u bc ℓm ( t ) . This distinction is essential in the massive problem. The oscillatory tails come from the cut [ − µ, µ ] , while quasinormal and quasib ound states b elong to the meromorphic part. All explicit tail statements b elo w are therefore formulated for u bc ℓm,P . W e b egin with the ﬁxed-mo de sp ectral theorem, which drives ev ery later time-domain statemen t. Theorem 1.1 (Fixed-mo de sp ectral and threshold theorem) . Fix a sub extr emal R eissner– Nor dstr öm exterior (1.1) and an angular momentum ℓ ≥ 1 . F or e ach p olarization P ∈ {− 1 , 0 , +1 } ther e exists a thr eshold index ν ℓ,P > − 1 2 such that the fol lowing hold. (i) The ﬁxe d-mo de cut-oﬀ r esolvent extends mer omorphic al ly fr om {ℑ ω > 0 } to a slit strip {|ℑ ω | < η } \ [ − µ, µ ] for some η > 0 , with at most ﬁnitely many p oles in c omp act subsets. 8 (ii) Ther e is no r e al p ole and no thr eshold r esonanc e at ω = ± µ . (iii) In the smal l-Coulomb-p ar ameter r e gime κ = M µ 2 /ϖ ≪ 1 , with ϖ = √ µ 2 − ω 2 , the thr eshold disc ontinuity ob eys disc G ℓ,P ( ω ; r , r ′ ) = a ℓ,P ( r , r ′ ) ϖ 2 ν ℓ,P + O  κ ϖ 2 ν ℓ,P  + O  ϖ 2 ν ℓ,P +2  (1.5) uniformly for r , r ′ in c omp act subsets of ( r + , ∞ ) . (iv) In the lar ge-Coulomb-p ar ameter r e gime κ ≫ 1 , the thr eshold disc ontinuity ob eys disc G ℓ,P ( ω ; r , r ′ ) = b + ℓ,P ( r , r ′ ) e 2 π i κ + b − ℓ,P ( r , r ′ ) e − 2 π i κ + O ( κ − 1 ) + O ( ϖ ) (1.6) uniformly for r , r ′ in c omp act subsets of ( r + , ∞ ) . (v) The thr eshold indic es satisfy ν ℓ,P = L P + 1 2 + O  ( M µ ) 2 + ( Qµ ) 2  (1.7) in the smal l-mass r e gime ( M µ ) 2 + ( Qµ ) 2 ≪ 1 . The time-domain tail theorem is then obtained by oscillatory inv ersion of the branc h- cut jump. Theorem 1.2 (Mo dewise branc h-cut tails on sub extremal Reissner–Nordström) . Fix ℓ ≥ 1 and let u bc ℓm,P ( t, r, r ′ ) b e the br anch-cut c ontribution for one ﬁxe d ( ℓ, m ) and one p olarization P ∈ {− 1 , 0 , +1 } . (a) Interme diate r e gime. If κ ∗ ( t ) := M µ 3 / 2 t 1 / 2 → 0 , (1.8) then u bc ℓm,P ( t, r, r ′ ) = A ℓ,P ( r , r ′ ; Q ) t − ( ν ℓ,P +1) sin( µt + δ ℓ,P ( Q )) + O  κ ∗ ( t ) t − ( ν ℓ,P +1)  + O  t − ( ν ℓ,P +2)  . (1.9) (b) Smal l-mass explicit exp onents. If in addition ε µ,Q := ( M µ ) 2 + ( Qµ ) 2 ≪ 1 , then the thr eshold indic es ob ey (1.7) ; henc e the le ading interme diate de c ay exp onents r e duc e to ℓ + 1 2 , ℓ + 3 2 , ℓ + 5 2 for P = − 1 , 0 , +1 , r esp e ctively. (c) V ery-late r e gime. L et ϖ 0 ( t ) :=  2 π M µ 3 t  1 / 3 , κ 0 ( t ) := M µ 2 ϖ 0 ( t ) . (1.10) If κ 0 ( t ) → ∞ , then u bc ℓm,P ( t, r, r ′ ) = B ℓ,P ( r , r ′ ; Q ) t − 5 / 6 sin  µt − 3 2 (2 π M µ ) 2 / 3 ( µt ) 1 / 3 + δ ℓ,P, 0 ( Q ) + O ( κ 0 ( t ) − 1 + ϖ 0 ( t ))  + O  κ 0 ( t ) − 1 + ϖ 0 ( t )  t − 5 / 6 . (1.11) The exp onent 5 / 6 is indep endent of ℓ , of the p olarization, and of Q . 9 R emark 1.1 . Theorem 1.2 is deliberately stated for the branc h-cut con tribution after subtraction of discrete p ole residues. This is the correct ﬁxed-mo de formulation in the massiv e problem, b ecause long-lived quasib ound p oles ma y dominate the full signal b efore the cut contribution b ecomes visible. Large angular momenta and branc h-cut full-ﬁeld deca y . The summed angular-momentum problem is the p oin t of con tact with [2]. In the scalar theory , one prov es a compact-region p olynomial deca y theorem for the full ﬁeld after o v ercoming the obstruction created by stable timelik e trapping. F or Pro ca, the argument naturally splits in to a con tin uous sp ectral part and a discrete quasib ound part. The ﬁrst ingredient is a uniform large- ℓ b ound for the branch-cut k ernels; the second is a semiclassical description and summation of the quasib ound resonance family . Theorem 1.3 (Uniform angular summabilit y of branch-cut k ernels) . F or every c omp act K ⋐ ( r + , ∞ ) ther e exist an inte ger N 0 ≥ 0 and a c onstant C K such that, for every admissible ( ℓ, m, P ) , sup r,r ′ ∈ K    u bc ℓm,P ( t, r, r ′ )    ≤ C K ⟨ ℓ ⟩ N 0    t − ( ν ℓ,P +1) , κ ∗ ( t ) ≤ 1 , t − 5 / 6 , κ 0 ( t ) ≥ 1 , (1.12) with the same interme diate and very-late r emainder structur e establishe d in Se ction 7, uniformly in ( ℓ, m, P ) . Mor e over, sup r,r ′ ∈ K  | A ℓ,P ( r , r ′ ; Q ) | + | B ℓ,P ( r , r ′ ; Q ) |  ≤ C K ⟨ ℓ ⟩ N 0 . (1.13) Theorem 1.4 (F ull-ﬁeld p oint wise branch-cut deca y) . L et A bc b e gener ate d by smo oth c omp actly supp orte d initial data. L et N 0 b e the inte ger fr om The or em 1.3. F or every c omp act K ⋐ ( r + , ∞ ) and every inte ger N > N 0 + 2 ther e exists C K,N such that sup r ∈ K , ω ∈ S 2    A bc ( t, r, ω )    ≤ C K,N E N [ A [0]] 1 / 2 t − ( ν ∗ +1) , κ ∗ ( t ) ≤ 1 . (1.14) wher e ν ∗ := inf ℓ ≥ 1 , P ∈{− 1 , 0 , +1 } ν ℓ,P . If κ 0 ( t ) ≥ 1 , then sup r ∈ K , ω ∈ S 2    A bc ( t, r, ω )    ≤ C K,N E N [ A [0]] 1 / 2 t − 5 / 6 . (1.15) In the Schwarzschild c ase, ν ∗ = 1 / 2 and the interme diate ful l-ﬁeld b ound is t − 3 / 2 ; in the smal l-mass R eissner–Nor dstr öm r e gime, ν ∗ = 1 / 2 + O (( M µ ) 2 + ( Qµ ) 2 ) . Corollary 1.5 (Polynomial time deca y for the radiative Pro ca ﬁeld) . Deﬁne γ ∗ := min  ν ∗ + 1 , 5 6  . Then for every c omp act K ⋐ ( r + , ∞ ) and every inte ger N > N 0 + 2 ther e exists C K,N such that sup r ∈ K , ω ∈ S 2    A bc ( t, r, ω )    ≤ C K,N E N [ A [0]] 1 / 2 t − γ ∗ . (1.16) In the Schwarzschild c ase one has γ ∗ = 5 / 6 . Mor e gener al ly, if ( M µ ) 2 + ( Qµ ) 2 is suﬃ- ciently smal l, then γ ∗ = 5 / 6 as wel l. 10 The corollary isolates the con tinuous spectral contribution. The unsplit full-ﬁeld theorem prov ed later in Theorem 1.10 adds the quasib ound resonance family and upgrades this branc h-cut estimate to a deca y theorem for the whole Pro ca ﬁeld. Quasib ound p oles, residue b ounds, and unsplit full-ﬁeld deca y . One of the cen tral lessons of [2] is that stable timelike trapping pro duces discrete bad frequencies and an un b ounded F ourier transform ev en though p oin twise decay still holds. The same geometric issue is present for massive spin- 1 ﬁelds. The present pap er prov es the corresp onding discrete sp ectral theorem as well: after mo de stability has ruled out upp er-half-plane and threshold pathologies, the lo wer-half-plane quasibound family is constructed semiclassically , its residues are con trolled uniformly , and the resulting time series is summed to recov er the unsplit full ﬁeld. The starting p oint is the following mo de-stabilit y theorem. Theorem 1.6 (Mo de stabilit y and threshold exclusion) . Fix ℓ ≥ 1 . Ther e is no nontrivial sep ar ate d solution of the form A = e − i ω t b A ( r , ω ) which is ingoing at the futur e event horizon and outgoing or de c aying at inﬁnity, with fr e quency ω in the close d upp er half-plane. Mor e pr e cisely: (i) if ℑ ω > 0 , no such mo de exists; (ii) if ω ∈ [ − µ, µ ] , ther e is no r e al cut p ole and no thr eshold r esonanc e at ω = ± µ . Conse quently the Evans determinant E ℓ is zer o-fr e e on {ℑ ω > 0 } and zer o-fr e e in punc- tur e d thr eshold neighb orho o ds of ω = ± µ . The con tinuous sp ectral results ab ov e are complemen ted by a discrete resonance pac k- age. Let A ( t ) = A bc ( t ) + A qb ( t ) + A fast ( t ) denote the decomp osition of the full Proca ﬁeld in to the branch-cut con tribution, the quasib ound p ole contribution, and the exp onen tially decaying remainder coming from p oles and contour terms b ounded a w ay from the real axis. Theorem 1.7 (Quasib ound branc hes and Bohr–Sommerfeld quan tization) . Fix a c om- p act ener gy interval I ⋐ (0 , µ ) c ontaine d in the stably tr app e d r e gime. Ther e exist ℓ sc ∈ N and c I > 0 such that for every ℓ ≥ ℓ sc and every p olarization P ∈ {− 1 , 0 , +1 } , the p oles of the ﬁxe d-mo de r esolvent with ℜ ω ∈ I , 0 > ℑ ω > − e − c I ℓ ar e simple and may b e indexe d by inte gers n ∈ N ℓ,P ( I ) as ω ℓ,n,P . W riting h = ( ℓ + 1 2 ) − 1 , one has the Bohr–Sommerfeld law S P ( ℜ ω ℓ,n,P ; h ) = 2 π h  n + 1 2  + h ϑ P ( ℜ ω ℓ,n,P ; h ) + O ( h 2 ) , with ϑ P = O (1) , and the tunnel ling width formula ℑ ω ℓ,n,P = − Γ P ( ℜ ω ℓ,n,P ; h ) exp  − 2 J P ( ℜ ω ℓ,n,P ; h ) h  (1 + O ( h )) . 11 Theorem 1.8 (Residue pro jectors and Agmon lo calization) . F or every c omp act r adial set K ⋐ ( r + , ∞ ) ther e exist inte gers N qb , N rec ≥ 0 and a c onstant C K such that every quasib ound r esidue pr oje ctor Π ℓ,n,P := − Res ω = ω ℓ,n,P R ℓ,P ( ω ) ob eys sup r,r ′ ∈ K | Π ℓ,n,P ( r , r ′ ) | ≤ C K ⟨ ℓ ⟩ N qb , and, for c omp actly supp orte d initial data, sup r ∈ K , ω ∈ S 2 | Π ℓ,n,P A [0]( r , ω ) | ≤ C K,N ⟨ ℓ ⟩ N rec E N [ A [0]] 1 / 2 . If K is disjoint fr om the classic al ly al lowe d wel l c orr esp onding to ℜ ω ℓ,n,P , then the r esidue kernel satisﬁes an A gmon-typ e exp onential b ound on K × K . Theorem 1.9 (Self-con tained summed quasib ound deca y) . F or smo oth c omp actly sup- p orte d initial data and every lo garithmic p ower L > 0 , ther e exists an inte ger N sum qb ( L ) such that for every c omp act K ⋐ ( r + , ∞ ) , sup r ∈ K , ω ∈ S 2    A qb ( t, r, ω )    ≤ C K,N ,L E N [ A [0]] 1 / 2 (log(2 + t )) − L for al l N ≥ N sum qb ( L ) and al l t ≥ 2 . Theorem 1.10 (Self-con tained full-ﬁeld deca y) . F or smo oth c omp actly supp orte d initial data, every lo garithmic p ower L > 0 , and every c omp act K ⋐ ( r + , ∞ ) , ther e exists N full ( L ) such that sup r ∈ K , ω ∈ S 2 | A ( t, r, ω ) | ≤ C K,N ,L E N [ A [0]] 1 / 2  t − γ ∗ + (log (2 + t )) − L  for al l N ≥ N full ( L ) and al l t ≥ 2 , wher e γ ∗ is the br anch-cut exp onent fr om Cor ol lary 1.5. In p articular, sup r ∈ K , ω ∈ S 2 | A ( t, r, ω ) | ≤ C K,N ,L E N [ A [0]] 1 / 2 (log(2 + t )) − L . Theorem 1.11 (T wo-regime asymptotic expansion of the full Pro ca ﬁeld) . Fix a c omp act set K ⋐ ( r + , ∞ ) and let A b e gener ate d by smo oth c omp actly supp orte d initial data. Then ther e exist an inte ger ℓ as ≥ 1 , ﬁnite-r ank ﬁelds A bc int , lo w ( t ) , A bc late , low ( t ) , c oming fr om the ﬁnitely many mo des 0 ≤ ℓ < ℓ as and dep ending line arly on A [ 0] , and an inte ger N as > N 0 + 2 such that for every N ≥ N as ther e exist c o eﬃcient functions, line ar in A [ 0] , A ℓm,P [ A [0]]( r ) , B ℓm,P [ A [0]]( r ) , ℓ ≥ ℓ as , | m | ≤ ℓ, P ∈ {− 1 , 0 , +1 } , with sup r ∈ K  |A ℓm,P [ A [0]]( r ) | + |B ℓm,P [ A [0]]( r ) |  ≤ C K,N ⟨ ℓ ⟩ N 0 E ℓm,P [ A [0]] 1 / 2 , (1.17) 12 wher e N 0 is the angular-loss exp onent fr om The or em 1.3. Deﬁne A bc int ( t, r, ϑ ) := A bc int , lo w ( t, r, ϑ ) + X ℓ ≥ ℓ as ℓ X m = − ℓ X P ∈{− 1 , 0 , +1 } A ℓm,P [ A [0]]( r ) Y ( P ) ℓm ( ϑ ) t − ( ν ℓ,P +1) sin( µt + δ ℓ,P ( Q )) , (1.18) A bc late ( t, r, ϑ ) := A bc late , low ( t, r, ϑ ) + t − 5 / 6 X ℓ ≥ ℓ as ℓ X m = − ℓ X P ∈{− 1 , 0 , +1 } B ℓm,P [ A [0]]( r ) Y ( P ) ℓm ( ϑ ) × sin  µt − 3 2 (2 π M µ ) 2 / 3 ( µt ) 1 / 3 + δ ℓ,P, 0 ( Q )  . (1.19) Then b oth series c onver ge in C 0 ( K × S 2 ) and satisfy sup r ∈ K , ϑ ∈ S 2    A bc int ( t, r, ϑ )    ≤ C K,N E N [ A [0]] 1 / 2 t − ( ν ∗ +1) , (1.20) sup r ∈ K , ϑ ∈ S 2    A bc late ( t, r, ϑ )    ≤ C K,N E N [ A [0]] 1 / 2 t − 5 / 6 . (1.21) If κ ∗ ( t ) ≤ 1 , then A ( t ) = A bc int ( t ) + A qb ( t ) + A fast ( t ) + R int ( t ) (1.22) on K , with r emainder b ound sup r ∈ K , ϑ ∈ S 2 | R int ( t, r, ϑ ) | ≤ C K,N E N [ A [0]] 1 / 2  κ ∗ ( t ) t − ( ν ∗ +1) + t − ( ν ∗ +2)  . (1.23) If κ 0 ( t ) ≥ 1 , then A ( t ) = A bc late ( t ) + A qb ( t ) + A fast ( t ) + R late ( t ) (1.24) on K , with r emainder b ound sup r ∈ K , ϑ ∈ S 2 | R late ( t, r, ϑ ) | ≤ C K,N E N [ A [0]] 1 / 2  κ 0 ( t ) − 1 + ϖ 0 ( t )  t − 5 / 6 . (1.25) Thus the ful l solution admits an explicit r adiative asymptotic exp ansion in b oth time r e gimes, with the only additional late-time obstruction c arrie d by the sep ar ately c ontr ol le d quasib ound term A qb . Corollary 1.12 (Explicit leading co eﬃcient ﬁelds) . Fix a c omp act set K ⋐ ( r + , ∞ ) and let A b e gener ate d by smo oth c omp actly supp orte d initial data. L et e A ℓm,P [ A [0]]( r ) , e B ℓm,P [ A [0]]( r ) , denote the ful l mo dal c o eﬃcient families furnishe d by the pr o of of The or em 1.11; these dep end line arly on A [0] , and for ℓ = 0 only the ele ctric monop ole P = +1 o c curs. Then ther e exist a ﬁnite nonempty set Σ ∗ := { ( ℓ, P ) : ℓ ≥ 1 , P ∈ {− 1 , 0 , +1 } , ν ℓ,P = ν ∗ } and a numb er ρ ∗ > 0 such that ν ℓ,P ≥ ν ∗ + ρ ∗ whenever ( ℓ, P ) / ∈ Σ ∗ . 13 Deﬁne the interme diate c o eﬃcient ﬁelds S ∗ ( r , ϑ ) := X ( ℓ,P ) ∈ Σ ∗ ℓ X m = − ℓ e A ℓm,P [ A [0]]( r ) cos δ ℓ,P ( Q ) Y ( P ) ℓm ( ϑ ) , (1.26) C ∗ ( r , ϑ ) := X ( ℓ,P ) ∈ Σ ∗ ℓ X m = − ℓ e A ℓm,P [ A [0]]( r ) sin δ ℓ,P ( Q ) Y ( P ) ℓm ( ϑ ) , (1.27) and the very-late c o eﬃcient ﬁelds S late ( r , ϑ ) := X ℓ ≥ 0 ℓ X m = − ℓ X P ∈P ℓ e B ℓm,P [ A [0]]( r ) cos δ ℓ,P, 0 ( Q ) Y ( P ) ℓm ( ϑ ) , (1.28) C late ( r , ϑ ) := X ℓ ≥ 0 ℓ X m = − ℓ X P ∈P ℓ e B ℓm,P [ A [0]]( r ) sin δ ℓ,P, 0 ( Q ) Y ( P ) ℓm ( ϑ ) , (1.29) wher e P 0 = { +1 } and P ℓ = {− 1 , 0 , +1 } for ℓ ≥ 1 . Then S ∗ and C ∗ ar e ﬁnite-r ank, the series deﬁning S late and C late c onver ge in C 0 ( K × S 2 ) , and ther e exists N lead such that for every N ≥ N lead , sup r ∈ K , ϑ ∈ S 2  |S ∗ ( r , ϑ ) | + |C ∗ ( r , ϑ ) | + |S late ( r , ϑ ) | + |C late ( r , ϑ ) |  ≤ C K,N E N [ A [0]] 1 / 2 , (1.30) and, on K , A ( t ) = t − ( ν ∗ +1)  S ∗ ( r , ϑ ) sin( µt ) + C ∗ ( r , ϑ ) cos( µt )  + A qb ( t ) + A fast ( t ) + e R int ( t ) , (1.31) A ( t ) = t − 5 / 6  S late ( r , ϑ ) sin Θ( t ) + C late ( r , ϑ ) cos Θ( t )  + A qb ( t ) + A fast ( t ) + e R late ( t ) , (1.32) wher e Θ( t ) := µt − 3 2 (2 π M µ ) 2 / 3 ( µt ) 1 / 3 , σ ∗ := min { 1 , ρ ∗ } . If κ ∗ ( t ) ≤ 1 , then sup r ∈ K , ϑ ∈ S 2    e R int ( t, r, ϑ )    ≤ C K,N E N [ A [0]] 1 / 2  κ ∗ ( t ) t − ( ν ∗ +1) + t − ( ν ∗ +1+ σ ∗ )  . (1.33) If κ 0 ( t ) ≥ 1 , then sup r ∈ K , ϑ ∈ S 2    e R late ( t, r, ϑ )    ≤ C K,N E N [ A [0]] 1 / 2  κ 0 ( t ) − 1 + ϖ 0 ( t )  t − 5 / 6 . (1.34) In p articular, the dominant interme diate c o eﬃcient is ﬁnite-r ank and is supp orte d exactly on the channels with thr eshold index ν ∗ . R emark 1.2 (Scope of the leading co eﬃcien t statemen t) . Corollary 1.12 iden tiﬁes the leading co eﬃcient ﬁelds of the radiativ e branch-cut con tribution after the quasibound term has b een separated. I t do es not claim that the display ed oscillatory term dominates the unsplit full ﬁeld for arbitrary data, since the self-con tained quasib ound estimate is logarithmic. Moreov er the co eﬃcient ﬁelds are data dep endent and ma y v anish identically for special initial data; in that case the ﬁrst non trivial radiativ e term is obtained b y rep eating the same argumen t with the smallest threshold index actually activ ated b y the data. 14 R emark 1.3 (Optional external arithmetic upgrade) . I f one supplements the present pap er with the abstract arithmetic pac ket theorem of [2], then the logarithmic quasib ound estimate abov e upgrades to the p olynomial bound t − 5 / 6+1 / 23 from the earlier draft, and under the exponent-pair conjecture to t − 5 / 6+ ε for every ε > 0 . This stronger reﬁnemen t is not used anywhere in the self-con tained core developed b elo w. 1.5 Self-con tained summation and pro of arc hitecture The logic of the paper is easiest to read in four passes. W e b egin by reducing the Pro ca equation to one o dd scalar channel and one ev en 2 × 2 system, and b y iden tifying the p o- larization basis in which the large- r b ehavior b ecomes transparent. W e then construct the ﬁxed-mo de resolv ent, contin ue it across the massiv e branch cut, and extract the thresh- old expansions that con trol the late-time signal. In parallel we develop the semiclassical theory of the quasib ound p oles created b y stable timelike trapping. The ﬁnal step is to put the contin uous and discrete pieces back together at the full-ﬁeld level. A useful feature of the present version is that the discrete summation is no w gen uinely self-con tained. The only inputs are ones prov ed in the pap er itself: tw o-sided tunnelling- width b ounds and p olynomial residue/reconstruction estimates. On a dyadic pac ket the tunnelling form ula con tributes an explicit damping factor exp( − ct e − C 2 j ) , while angular regularit y of the initial data pa ys for an y negative p ow er of the pack et scale 2 j . A short dy adic summation lemma then turns these pack et b ounds into logarithmic decay for the quasib ound contribution. The sharper polynomial rates from the earlier draft remain a v ailable only as an external arithmetic upgrade; see Remark 1.3. F or readers skimming the o verall argumen t, it is helpful to keep the pap er in ﬁve mo dules: (1) Section 2 constructs the o dd and even c hannel equations and identiﬁes the asymp- totic polarization basis. (2) Sections 3–5 dev elop the op erator-theoretic ﬁxed-mo de framework and pro ve the threshold sp ectral theorem. (3) Sections 6 and 7 con vert that sp ectral theorem into mo de stabilit y and explicit branc h-cut tails. (4) Section 8 carries out the large-angular-momentum analysis needed for branc h-cut summation on compact radial sets. (5) Sections 9 and 10 construct the quasib ound p ole family , prov e the self-con tained pac k et estimate, sum the discrete series, and establish the unsplit full-ﬁeld decay theorem. 1.6 Organization of the pap er The paper is written so that the ﬁxed-mode theory comes ﬁrst and the full-ﬁeld re- construction comes later. Sections 2–5 set up the o dd/even reduction, the selfadjoint c hannel framework, the Green kernels, and the threshold sp ectral theorem. Sections 6 and 7 then turn that sp ectral input in to mo de stability and explicit in termediate and v ery-late branc h-cut asymptotics. Section 8 provides the large-angular-momentum anal- ysis needed to sum the branch-cut con tribution on compact radial sets. Sections 9 and 15 10 dev elop the semiclassical quasib ound resonance theory , pro ve the residue and recon- struction b ounds, derive the self-con tained pack et estimate, and com bine the discrete and con tin uous pieces to obtain deca y for the unsplit Pro ca ﬁeld. Section 11 closes with the directions that lie b eyond the static sub extremal setting. 1.7 Pro of dep endency diagram The diagram b elow records the main logical dep endencies in a w ay that is easy to c hec k while reading or refereeing the pap er: Section 2 = ⇒ Sections 3–5 ⇓ Theorem 1.1 Theorem 1.1 = ⇒ Section 6 = ⇒ Theorem 1.6 Theorem 1.1 = ⇒ Section 7 = ⇒ Theorem 1.2 Section 8 = ⇒ Theorems 1.3,1.4 Section 9 = ⇒ Theorems 1.7,1.8 Theorems 1.4,1.7,1.8 = ⇒ Section 10 = ⇒ Theorems 1.9,1.10 . In other words, the pap er has four real b ottlenec ks: the ﬁxed-mode sp ectral theorem, the time-domain in version of the branc h cut, the large-angular-momentum uniformit y theorem, and the quasib ound summation theorem. Everything else is arranged to feed in to one of these p oints or to draw consequences from them. 1.8 App endix guide The app endices are group ed by function rather than by the order in whic h they are used. App endix A records the v ector-spherical-harmonic reduction together with the boundary constructions at the horizon and inﬁnit y . App endix B collects the threshold mo del equa- tions, sp ecial-function asymptotics, oscillatory lemmas, and ﬁxed-mo de technical comple- men ts. Appendix C gathers the large-angular-momentum, reconstruction, lo cal-energy , and dy adic b o okkeeping estimates used in the full-ﬁeld argumen t. App endix D treats the small-mass regime, the monop ole and static-threshold input, and the cutoﬀ-indep endence argumen t for the branch-cut ﬁeld on compact radial sets. 2 Geometry , mo de reduction, and p olarization split- ting 2.1 Geometry and the ﬁeld equation W e begin b y ﬁxing the background geometry and writing the Proca system in the form that will b e used throughout the pap er. g RN = − f ( r ) d t 2 + f ( r ) − 1 d r 2 + r 2 dΩ 2 , f ( r ) = 1 − 2 M r + Q 2 r 2 = ( r − r + )( r − r − ) r 2 , r > r + . (2.1) 16 with r ± = M ± q M 2 − Q 2 , κ + := r + − r − 2 r 2 + > 0 . The tortoise co ordinate is deﬁned b y d r ∗ = f − 1 d r . Near the ev en t horizon r ∗ = 1 2 κ + log( r − r + ) + O (1) , r ↓ r + , (2.2) whereas at inﬁnity r ∗ = r + 2 M log r + O (1) . (2.3) The massiv e Maxwell–Proca equation is ∇ µ F µν − µ 2 A ν = 0 , F = d A, µ > 0 . (2.4) T aking the div ergence and using antisymmetry of F gives the forced Lorenz constraint ∇ ν A ν = 0 , (2.5) so there is no gauge freedom in the massive theory . 2.2 Reduced o dd/ev en mo de equations Unless stated otherwise, we w ork here with ℓ ≥ 1 ; the monop ole is p ostp oned b ecause only the even electric c hannel surviv es there. W e use the standard v ector spherical harmonic decomp osition. F or each ( ℓ, m ) the o dd sector is carried b y one scalar amplitude u 4 ( t, r ) , while the ev en sector is enco ded b y a pair ( u 2 ( t, r ) , u 3 ( t, r )) . W riting λ 2 = ℓ ( ℓ + 1) , the neutral Pro ca equations on a static spherically symmetric background reduce to the same o dd/ev en pattern as in Sc h warzsc hild, with the curv ature term 1 − 3 M r replaced b y the Reissner–Nordström com bination f ( r ) − r f ′ ( r ) 2 = 1 − 3 M r + 2 Q 2 r 2 . The reduced time-domain equations are therefore  − ∂ 2 t + ∂ 2 r ∗ − f  µ 2 + λ 2 r 2  u 4 = 0 (2.6) and  − ∂ 2 t + ∂ 2 r ∗ − f  µ 2 + λ 2 r 2  u 2 − 2 f r 2  f − r f ′ 2  ( u 2 − u 3 ) = 0 , (2.7)  − ∂ 2 t + ∂ 2 r ∗ − f  µ 2 + λ 2 r 2  u 3 + 2 f λ 2 r 2 u 2 = 0 . (2.8) After F ourier transformation with time dep endence e − i ω t , these b ecome u ′′ 4 +  ω 2 − f µ 2 − f λ 2 r 2  u 4 = 0 (2.9) 17 and u ′′ 2 +  ω 2 − f µ 2 − f λ 2 r 2  u 2 − 2 f r 2  f − r f ′ 2  ( u 2 − u 3 ) = 0 , (2.10) u ′′ 3 +  ω 2 − f µ 2 − f λ 2 r 2  u 3 + 2 f λ 2 r 2 u 2 = 0 , (2.11) where primes denote ∂ r ∗ . The reduction is consistent with the Sch warzsc hild form ulas of Rosa–Dolan [4], and the fact that Pro ca separates in the full Kerr–Newman family sho ws that this static RN mo del is the correct zero-rotation limit of the general charged blac k-hole problem [6]. 2.3 Exact asymptotic p olarization diagonalization The o dd sector already giv es one ph ysical polarization; we denote it by P = 0 . The ev en sector is more interesting. The right v ariables are not the original pair ( u 2 , u 3 ) , but the constant com binations b elo w. In those co ordinates the leading inv erse-square term b ecomes diagonal, and the three eﬀective angular momenta ℓ − 1 , ℓ , and ℓ + 1 app ear transparen tly . v 0 := u 4 , L 0 := ℓ. (2.12) The even sector is diagonalized at leading order by the same constant transformation as in Sc hw arzschild: T ℓ = 1 1 ℓ + 1 − ℓ ! , T − 1 ℓ = 1 2 ℓ + 1 ℓ 1 ℓ + 1 − 1 ! , (2.13) with v − 1 v +1 ! = T − 1 ℓ u 2 u 3 ! = 1 2 ℓ + 1 ℓu 2 + u 3 ( ℓ + 1) u 2 − u 3 ! . (2.14) Equiv alen tly , u 2 = v − 1 + v +1 , u 3 = ( ℓ + 1) v − 1 − ℓv +1 . (2.15) Prop osition 2.1 (Reissner–Nordström ev en-sector p olarization form) . F or every ℓ ≥ 1 , the even fr e quency-domain system (2.10) – (2.11) takes the matrix form v ′′ +  ω 2 − f µ 2  v − f r 2 D ℓ v − 6 M f (2 ℓ + 1) r 3 C ℓ v + 4 Q 2 f (2 ℓ + 1) r 4 C ℓ v = 0 , (2.16) wher e v = v − 1 v +1 ! , D ℓ = diag  ℓ ( ℓ − 1) , ( ℓ + 1)( ℓ + 2)  , (2.17) and C ℓ = ℓ 2 − ℓ ( ℓ + 1) ℓ ( ℓ + 1) − ( ℓ + 1) 2 ! . (2.18) In p articular, the le ading r − 2 p art diagonalizes exactly, with eﬀe ctive angular momenta L − 1 = ℓ − 1 , L +1 = ℓ + 1 , (2.19) and the char ge Q ﬁrst enters the even c oupling at or der r − 4 . 18 Pr o of. W rite the ev en system as u ′′ +  ω 2 − f µ 2  u − f r 2 A ℓ ( r ) u = 0 , u = u 2 u 3 ! , (2.20) with A ℓ ( r ) = λ 2 + 2  f − rf ′ 2  − 2  f − rf ′ 2  − 2 λ 2 λ 2 ! . (2.21) Since f − r f ′ 2 = 1 − 3 M r + 2 Q 2 r 2 , a direct computation gives T − 1 ℓ A ℓ ( r ) T ℓ = D ℓ + 6 M (2 ℓ + 1) r C ℓ − 4 Q 2 (2 ℓ + 1) r 2 C ℓ . (2.22) Substituting u = T ℓ v in to (2.20) yields (2.16). The diagonal en tries of D ℓ are ℓ ( ℓ − 1) = L − 1 ( L − 1 + 1) , ( ℓ + 1)( ℓ + 2) = L +1 ( L +1 + 1) , whic h pro v es the claim. R emark 2.1 (Three asymptotic c hannels) . Sub extremal Reissner–Nordström Pro ca there- fore has three asymptotic p olarizations: the o dd channel P = 0 with L 0 = ℓ , and tw o ev en c hannels P = ± 1 with L ± 1 = ℓ ± 1 . This exact asymptotic triplet is the source of the small-mass intermediate exp onents. 3 Op erator-theoretic framew ork and limiting ab- sorption 3.1 Channel Hamiltonians and asymptotic op erator structure F or each ﬁxed angular momen tum ℓ ≥ 1 , the reduced equations become one scalar and one matrix Sc hrö dinger operator on the tortoise line. This is the natural setting for limiting absorption, meromorphic contin uation, and the threshold analysis carried out later. H odd ℓ = − ∂ 2 r ∗ + V odd ℓ ( r ) , H ev ℓ = − ∂ 2 r ∗ Id 2 + V ev ℓ ( r ) . (3.1) In the p olarization basis of Prop osition 2.1, the matrix p otential has the asymptotic form V ℓ,P ( r ) = µ 2 + L P ( L P + 1) r 2 − 2 M µ 2 r + O ( r − 3 ) , r → ∞ , (3.2) while at the future ev ent horizon V ℓ,P ( r ∗ ) = O (e 2 κ + r ∗ ) , r ∗ → −∞ . (3.3) Th us the left end of the tortoise line is short range, whereas the right end is long range with a Coulomb term and a matrix-v alued in verse-square correction. This step-like structure is the analytic origin of the massive branch p oints ω = ± µ . 19 The time-dependent reduced evolution is generated by the wa ve op erator − ∂ 2 t + H odd ℓ or − ∂ 2 t + H ev ℓ , dep ending on the p olarization. It is conv enient to separate the sp ectral parameter λ = ω 2 from the threshold parameter ϖ = √ µ 2 − ω 2 . The op erator-theoretic spectrum liv es naturally in the λ -plane, while the branch structure relev ant for late-time asymptotics liv es in the ω -plane. 3.2 Selfadjoin t realizations and energy spaces Let h odd ℓ = L 2 ( R r ∗ ) and h ev ℓ = L 2 ( R r ∗ ; C 2 ) . Deﬁne D ( H odd ℓ ) = H 2 ( R r ∗ ) , D ( H ev ℓ ) = H 2 ( R r ∗ ; C 2 ) . Because the co eﬃcient matrices are smo oth, real-symmetric, and b ounded together with all deriv atives on compact sets, and b ecause (3.3) and (3.2) imply relative b oundedness with resp ect to − ∂ 2 r ∗ , the following prop osition records the op erator-theoretic realization used later. Prop osition 3.1 (Selfadjoint c hannel Hamiltonians) . F or every ℓ ≥ 1 , the op er ators H odd ℓ and H ev ℓ ar e selfadjoint on their natur al H 2 domains. Their quadr atic forms ar e b ounde d b elow, and the c orr esp onding ﬁrst-or der ener gy gener ators A odd ℓ = 0 1 − H odd ℓ 0 ! , A ev ℓ = 0 Id − H ev ℓ 0 ! gener ate str ongly c ontinuous unitary gr oups on the o dd and even ener gy sp ac es. Pr o of. W rite H ♯ ℓ = H 0 + V ♯ ℓ , where H 0 = − ∂ 2 r ∗ on L 2 ( R r ∗ ) in the o dd c hannel and on L 2 ( R r ∗ ; C 2 ) in the even channel. By (3.3) and (3.2), eac h en try of V ♯ ℓ is real-v alued, b elongs to L ∞ loc , deca ys exponentially as r ∗ → −∞ , and is O ( r − 2 ) as r ∗ → + ∞ . Hence, for ev ery δ > 0 ,    ⟨ V ♯ ℓ u, u ⟩    ≤ δ ∥ u ′ ∥ 2 L 2 + C ℓ,δ ∥ u ∥ 2 L 2 , u ∈ H 1 , b y lo cal b oundedness on compact sets and the one-dimensional Hardy inequality on the p ositiv e end. Th us V ♯ ℓ is inﬁnitesimally form-bounded with resp ect to H 0 , the closed quadratic forms q ♯ ℓ [ u ] = ∥ u ′ ∥ 2 L 2 + ⟨ V ♯ ℓ u, u ⟩ are semib ounded on H 1 , and the asso ciated selfadjoint op erators coincide with − ∂ 2 r ∗ + V ♯ ℓ on H 2 b y elliptic regularit y . This prov es the ﬁrst assertion. Cho ose c ℓ > 0 so that H ♯ ℓ + c ℓ ≥ Id . Endow the energy space H ♯ ℓ = H 1 × L 2 with norm ∥ ( u, v ) ∥ 2 H ♯ ℓ = q ♯ ℓ [ u ] + c ℓ ∥ u ∥ 2 L 2 + ∥ v ∥ 2 L 2 . On the domain H 2 × H 1 , the op erator A ♯ ℓ ( u, v ) = ( v , − H ♯ ℓ u ) is sk ew-adjoint with resp ect to the induced energy inner product. Stone’s theorem therefore yields a strongly contin uous unitary group for b oth the o dd and even ﬁrst-order ev olutions. 20 R emark 3.1 (Why the massiv e thresholds do not coincide with the sp ectral edge) . The essen tial sp ectrum of the c hannel Hamiltonians is [0 , ∞ ) b ecause the horizon end of the tortoise line is asymptotically free. The distinguished frequencies ω = ± µ arise instead from the asymptotic op erator at spatial inﬁnit y , where the p oten tial tends to µ 2 . In the ω -plane, this right-end threshold creates the branc h points resp onsible for the late-time tails. This distinction b etw een the op erator sp ectrum in λ = ω 2 and the asymptotic branc h structure in ω is harmless once it is stated explicitly . 3.3 Radial curren ts, W ronskians, and cut-oﬀ resolv ents Let u, v b e scalar channel solutions. Their radial curren t is Q [ u, v ] = u ′ v − u v ′ . F or vector solutions u , v ∈ C 2 w e set Q [ u , v ] = u ′ ∗ v − u ∗ v ′ . If u and v solv e the same scalar equation, or if u and v solv e the same symmetric matrix equation, then Q is indep endent of r ∗ . This conserv ed W ronskian curren t is the basic algebraic ob ject behind the resolv ent form ula, the Ev ans determinan t, and the real-frequency exclusion argument. Cho ose χ ∈ C ∞ 0 (( r + , ∞ )) . F or ℑ ω > 0 we deﬁne the cut-oﬀ resolven ts R odd ℓ,χ ( ω ) = χ ( H odd ℓ − ω 2 ) − 1 χ, R ev ℓ,χ ( ω ) = χ ( H ev ℓ − ω 2 ) − 1 χ. The o dd kernel is expressed in terms of one ingoing horizon solution and one decaying inﬁnit y solution. The ev en k ernel is analogous, except that the scalar W ronskian is replaced b y a 2 × 2 matc hing matrix. Lemma 3.2 (Green kernel form ula) . L et u hor b e the sc alar o dd solution ingoing at the futur e event horizon and let u ∞ b e the sc alar o dd solution de c aying for ℜ ϖ > 0 at inﬁnity. Then G odd ℓ ( ω ; r ∗ , r ′ ∗ ) = u hor ( r < , ω ) u ∞ ( r > , ω ) Q [ u hor , u ∞ ] , wher e r < := min { r ∗ , r ′ ∗ } and r > := max { r ∗ , r ′ ∗ } . In the even se ctor the same formula holds with u hor , u ∞ r eplac e d by fundamental solution matric es and the sc alar W r onskian r eplac e d by the matching matrix intr o duc e d in Step 4 of Se ction 5. Pr o of. In the scalar c hannel set W ( ω ) = Q [ u hor , u ∞ ] , which is indep enden t of r ∗ . F or ﬁxed r ′ ∗ deﬁne G ( r ∗ ) =    W ( ω ) − 1 u hor ( r ∗ , ω ) u ∞ ( r ′ ∗ , ω ) , r ∗ < r ′ ∗ , W ( ω ) − 1 u hor ( r ′ ∗ , ω ) u ∞ ( r ∗ , ω ) , r ∗ > r ′ ∗ . Then G solve s the homogeneous equation a w a y from r ′ ∗ , is contin uous at r ′ ∗ , and satisﬁes ∂ r ∗ G ( r ′ ∗ + 0) − ∂ r ∗ G ( r ′ ∗ − 0) = 1 precisely b ecause Q [ u hor , u ∞ ] = W ( ω ) . Hence ( H odd ℓ − ω 2 ) G = δ r ′ ∗ , whic h is the asserted Green k ernel formula. In the even sector one argues iden tically with fundamen tal matrices. W riting the k ernel on the t w o sides of the diagonal as linear combinations of the horizon and inﬁnit y bases, the con tinuit y condition and the unit jump of the ﬁrst deriv ative give a linear system whose co eﬃcient matrix is the matc hing matrix. Solving that system yields the stated matrix analogue. 21 3.4 Limiting absorption a w ay from the thresholds The cut-oﬀ resolven ts are analytic for ℑ ω > 0 , but for late-time asymptotics one must understand their boundary b eha vior as ω approac hes the real axis. A wa y from the dis- tinguished p oin ts ω = ± µ , the problem is no more singular than for a one-dimensional short-range scattering system with a step potential. The long-range diﬃculty app ears only at the massive thresholds themselv es. Prop osition 3.3 (Boundary v alues aw a y from ω = ± µ ) . Fix χ ∈ C ∞ 0 (( r + , ∞ )) and a c omp act interval I ⋐ R \ {± µ } . Then the limits R odd ℓ,χ ( ω ± i0) , R ev ℓ,χ ( ω ± i0) exist for ω ∈ I as b ounde d maps fr om L 2 to H 2 loc , and dep end c ontinuously on ω . Pr o of. Because I sta ys a p ositive distance a w ay from ± µ , either | ω | < µ with ℜ ϖ ( ω ) ≥ c I > 0 or | ω | > µ with k ( ω ) = √ ω 2 − µ 2 ≥ c I > 0 . Lemmas 5.3 and 5.5 therefore pro duce horizon and inﬁnit y bases that are contin uous in ω ∈ I and ob ey uniform b ounds on compact radial sets. If I ⊂ [ − µ, µ ] , Prop osition 5.13 sho ws that the matc hing determinant nev er v anishes on I ; if I ∩ [ − µ, µ ] = ∅ , the same non v anishing follows from selfadjoin tness of the channel Hamiltonians together with the ordinary limiting absorption principle for one-dimensional Sc hrö dinger systems with short-range p erturbations of a constant end state. Hence the denominator in the Green formula stays uniformly aw ay from zero on compact subin terv als of I . Substituting the b oundary bases into Lemma 3.2 yields a k ernel con tin uous in ω with v alues in H 2 loc a w ay from the diagonal, and the jump relation is uniform in ω . Standard ODE regularity on compact sets therefore gives con tinuit y of the cut-oﬀ resolv ent as an op erator L 2 → H 2 loc . 3.5 Con tour deformation and the branc h-cut form ula Let Γ be a con tour in the upper half-plane enclosing the real axis and the poles of the meromorphically con tinued resolv en t in a ﬁnite spectral windo w. F or compactly supp orted data, the ﬁxed-mo de solution is recov ered from the con tour form ula u ℓ ( t ) = 1 2 π i Z Γ e − i ω t R ℓ,χ ( ω ) d ω . Once the con tour is pushed through the slit strip constructed in Section 5, the solution splits in to a ﬁnite p ole sum and a branc h-cut in tegral o ver [ − µ, µ ] : u ℓ ( t ) = u poles ℓ ( t ) + 1 2 π i Z µ − µ e − i ω t disc R ℓ,χ ( ω ) d ω . (3.4) This is the representation used throughout Sections 7 and 8. The contour deformation is justiﬁed b ecause all nonph ysical-sheet con tributions either cross isolated p oles or can b e pushed in to regions where the oscillatory factor e − i ω t is exp onen tially decaying. R emark 3.2 (Wh y w e k eep the pole sum separate) . F or the scalar problem, the discrete frequencies generated by stable timelik e trapping are already subtle enough to force a separate analysis in the full-ﬁeld theorem. The Pro ca situation is even more delicate b ecause the even sector is matrix v alued and the quasib ound p oles carry polarization information. F ormula (3.4) therefore deﬁnes the branc h-cut ﬁeld as a mathematically precise object rather than as a heuristic subtraction. 22 4 Channel resolv en ts and threshold represen tation 4.1 F ar-zone Coulom b normal form The threshold analysis is naturally expressed in a scalar normal form. One ﬁrst passes from r ∗ to r , remov es the ﬁrst deriv ativ e b y a scalar conjugation, and then applies a near- iden tit y asymptotic diagonalization in the ev en sector. The result is a diagonal threshold system whose r − 2 co eﬃcien t deﬁnes the threshold indices ν ℓ,P . Lemma 4.1 (F ar-zone threshold normal form) . Fix ℓ ≥ 1 . Ther e exist R 0 > r + , η > 0 , a smo oth invertible tr ansformation U ( r, ω ) = S ( r, ω ) W ( r, ω ) , S ( r, ω ) = Id + O ( r − 1 ) , and channel-dep endent indic es ν ℓ,P > − 1 / 2 such that for r ≥ R 0 every o dd/even p olar- ization channel is r e duc e d to w ′′ P +  − ϖ 2 + 2 M µ 2 r − ν 2 ℓ,P − 1 4 r 2 + V sr ℓ,P ( r , ω )  w P = 0 , (4.1) wher e ϖ = √ µ 2 − ω 2 and ∂ j ω V sr ℓ,P ( r , ω ) = O ( r − 3 ) , j = 0 , 1 , uniformly for |ℑ ω | < η . Mor e over, ν ℓ,P = L P + 1 2 + O  ( M µ ) 2 + ( Qµ ) 2  (4.2) in the smal l-mass r e gime. Pr o of. W rite the o dd scalar equation and the even p olarization system in the r -v ariable using ∂ r ∗ = f ∂ r . In b oth cases one obtains an equation of the form f 2 U rr + f f ′ U r +  ω 2 − f µ 2 − r − 2 f D ℓ − R ℓ ( r , ω )  U = 0 , where D ℓ is the scalar v alue ℓ ( ℓ + 1) in the o dd channel or the diagonal matrix with en tries L P ( L P + 1) in the ev en p olarization basis, and where R ℓ = O ( r − 3 ) together with one ω -deriv ativ e. Conjugating by the usual Liouville factor remov es the ﬁrst deriv ative and giv es W ′′ +  − ϖ 2 + 2 M µ 2 r − D ℓ + 1 4 Id + E ℓ r 2 + O ( r − 3 )  W = 0 . Here E ℓ is uniformly b ounded in ℓ : ev ery contribution to E ℓ comes from the ﬁxed metric co eﬃcien ts, the Liouville conjugation, and the bounded polarization matrix T ℓ , while the en tire large angular-momen tum dep endence has already b een isolated in D ℓ . The co eﬃcien t of r − 1 is universal b ecause r ∗ = r + 2 M log r + O (1) as r → ∞ and therefore dep ends only on the ADM mass term in the metric expansion. In the o dd c hannel the equation is already scalar. In the even channel, Prop osition 2.1 sho ws that the exact r − 2 part is diagonal in the basis (1.3), while the remaining oﬀ- diagonal terms are O ( r − 3 ) and O ( r − 4 ) . Solving the ﬁrst homological equation for a near- iden tit y diagonalizer S ( r , ω ) = Id + r − 1 S 1 ( r , ω ) + O ( r − 2 ) remo ves the residual oﬀ-diagonal r − 3 con tribution and yields decoupled scalar equations with short-range remainder V sr ℓ,P = O ( r − 3 ) . The resulting in verse-square co eﬃcients deﬁne ν 2 ℓ,P − 1 4 . Since these co eﬃcien ts reduce to L P ( L P + 1) at M µ = Qµ = 0 and dep end smoothly on ( M µ, Qµ ) , T a ylor expansion yields (4.2). 23 Ignoring the short-range remainder in (4.1) giv es the Whittaker mo del equation d 2 w d z 2 +  − 1 4 + κ z + 1 / 4 − ν 2 ℓ,P z 2  w = 0 , z = 2 ϖ r, κ = M µ 2 ϖ . (4.3) The model solutions are the Whittak er functions M κ,ν ℓ,P (2 ϖ r ) , W κ,ν ℓ,P (2 ϖ r ) . (4.4) 4.2 Green k ernels and branc h-cut decomp osition Let f ℓ,P ( r , ω ) denote the horizon-regular solution normalized b y f ℓ,P ( r , ω ) ∼ e − i ω r ∗ as r ↓ r + , (4.5) and let g ℓ,P ( r , ω ) denote the outgoing Jost solution normalized by g ℓ,P ( r , ω ) ∼ W κ,ν ℓ,P (2 ϖ r ) as r → ∞ . (4.6) The ﬁxed-mode Green kernel is G ℓ,P ( ω ; r , r ′ ) = f ℓ,P ( r < , ω ) g ℓ,P ( r > , ω ) W ℓ,P ( ω ) , r < := min { r, r ′ } , r > := max { r , r ′ } , (4.7) where W ℓ,P is the scalar or c hannel-pro jected W ronskian. The retarded k ernel decomp oses as u ℓm,P ( t, r, r ′ ) = u poles ℓm,P ( t, r, r ′ ) + u bc ℓm,P ( t, r, r ′ ) , (4.8) with branc h-cut part u bc ℓm,P ( t, r, r ′ ) = 1 2 π Z µ − µ e − i ω t disc G ℓ,P ( ω ; r , r ′ ) d ω . (4.9) Because ϖ = √ µ 2 − ω 2 c hanges sign across [ − µ, µ ] , that in terv al is the threshold branc h cut. 5 Pro of of the ﬁxed-mo de sp ectral theorem This section prov es Theorem 1.1. The proof is mo dewise and pro ceeds in nine steps: threshold normal form, horizon F rob enius theory , inﬁnity Jost construction, Ev ans de- terminan t, discreteness of poles, exclusion of real cut p oles and threshold resonances, small- κ Bessel asymptotics, large- κ Whittak er asymptotics, and assembly . Throughout this section we ﬁx one angular mo de ℓ ≥ 1 . The o dd sector has dimension n = 1 and the even sector has dimension n = 2 . In the even sector all statements are made in the p olarization basis ( v − 1 , v +1 ) . Step 1. Coulom b normal form The ﬁrst task is to separate the universal long-range Coulomb interaction from the shorter-range remainder in a form uniform across the o dd and ev en p olarizations. In the p olarization basis the co eﬃcien t matrix has the asymptotic expansion f ( r ) µ 2 + f ( r ) r 2 D ℓ + 1 r 3 B ℓ, 1 ( r ) + 1 r 4 B ℓ, 2 ( r ) , 24 where D ℓ is diagonal with entries L P ( L P + 1) and the matrices B ℓ,j ( r ) remain bounded with all deriv atives for large r . Using f ( r ) = 1 − 2 M /r + Q 2 /r 2 , one rewrites each c hannel as u ′′ +  − ϖ 2 + 2 M µ 2 r − ν 2 ℓ,P − 1 4 r 2 + W ℓ,P ( r , ω )  u = 0 , (5.1) with W ℓ,P ( r , ω ) = O ( r − 3 ) and the threshold index ν ℓ,P deﬁned by the exact co eﬃcien t of r − 2 . The key p oint is that the Q -dep endent correction do es not alter the universal Coulomb co eﬃcien t 2 M µ 2 /r . Consequen tly the saddle resp onsible for the v ery-late tail is un- c hanged from the Sc h w arzschild case. What the c harge do es change is the exact v alue of the in verse-square co eﬃcient and hence the intermediate exp onent through ν ℓ,P . Lemma 5.1 (Co eﬃcien t b ounds in Coulom b normal form) . Fix ℓ ≥ 1 and one p olariza- tion P . F or ω in a c omp act subset of the slit strip and r suﬃciently lar ge,    ∂ j r ∂ k ω W ℓ,P ( r , ω )    ≤ C j,k,ℓ r − 3 − j , and the same estimate holds for the oﬀ-diagonal r emainder in the even se ctor b efor e ﬁnal sc alar diagonalization. Pr o of. The claim follo ws b y direct expansion of f ( r ) = 1 − 2 M /r + Q 2 /r 2 and the trans- formed even matrix of Prop osition 2.1. Ev ery term b eyond the diagonal Coulomb and in v erse-square co eﬃcien ts con tains at least one additional p ow er of r − 1 . Diﬀeren tiation in ω falls only on ϖ = √ µ 2 − ω 2 and on b ounded analytic coeﬃcient matrices, so no loss o ccurs aw ay from the endp oin ts. Step 1 is Lemma 4.1. It pro vides the threshold normal form (4.1) and deﬁnes the c hannel indices ν ℓ,P . Step 2. Horizon F rob enius bases Near r = r + , the function f has a simple zero and the tortoise coordinate satisﬁes r − r + ∼ e 2 κ + r ∗ . Consequently ev ery c hannel equation b ecomes a short-range p erturbation of u ′′ + ω 2 u = 0 as r ∗ → −∞ . The ingoing and outgoing horizon b eha viors are therefore e − i ω r ∗ and e +i ω r ∗ . T o make this precise, one writes u ± hor ( r , ω ) = e ∓ i ω r ∗ h ± ( r , ω ) , where h ± extend smo othly to r = r + and satisfy a regular singular system in the v ariable r − r + . Lemma 5.2 (Horizon F rob enius bases with parameter dep endence) . F or every c omp act ω -set in the slit strip, ther e exist unique sc alar o dd solutions and unique even matrix fundamental solutions of the form u ± hor ( r , ω ) = e ∓ i ω r ∗  1 + ∞ X n =1 a ± n ( ω )( r − r + ) n  , with normal ly c onver gent series ne ar r = r + . The c o eﬃcients dep end analytic al ly on ω and satisfy b ounds uniform on c omp act ω -sets. 25 Pr o of. After factoring out e ∓ i ω r ∗ , the remaining equation has co eﬃcien ts holomorphic in r − r + with a regular singular p oin t at the horizon. The indicial ro ots are 0 and therefore nonresonant once the oscillatory factor has been remov ed. Standard F rob enius recursion giv es existence and uniqueness of the series. Uniform conv ergence and analytic dep endence on ω follow from the recursive b ounds on the co eﬃcien ts. The horizon bases provide the left-hand b oundary data used later in the Ev ans de- terminan t. They also enco de the physical ingoing condition required in the deﬁnition of quasinormal and quasib ound mo des. Lemma 5.3 (Horizon-regular basis) . F or e ach ﬁxe d ℓ ther e exists a unique n × n matrix solution F hor ( r , ω ) of the r e duc e d system such that F hor ( r , ω ) = e − i ω r ∗  Id n + H ( r, ω )  , H ( r , ω ) = O ( r − r + ) , (5.2) as r ↓ r + . Mor e over F hor dep ends analytic al ly on ω in a strip |ℑ ω | < η . Pr o of. W rite the radial system in the co ordinate x = r − r + . Since f ( r ) = 2 κ + x + O ( x 2 ) , the reduced equations ha ve a regular singular p oin t at x = 0 with indicial ro ots ± i ω / (2 κ + ) . The ingoing ro ot is realized by the factor x − i ω / (2 κ + ) = e − i ω r ∗ . The remaining co eﬃcien t matrix is analytic in ( x, ω ) , so F rob enius theory yields a unique analytic matrix Id n + H ( x, ω ) with H (0 , ω ) = 0 . Analytic dep endence on ω follo ws from analytic dependence of the co eﬃcien ts. Step 3. Inﬁnit y Jost bases A t spatial inﬁnity the c hannel equation is long range rather than short range. After remo ving the Coulom b factor, the decaying inﬁnit y solution is constructed by a V olterra equation rather than by a simple pow er series. One writes u ∞ ( r , ω ) = exp( − ϖ r ) r − κ m ( r , ω ) , where κ = M µ 2 /ϖ and m ( r, ω ) → 1 as r → ∞ when ℜ ϖ > 0 . Substituting this ansatz in to (5.1) produces an integral equation of V olterra t yp e for m . Lemma 5.4 (Inﬁnit y V olterra construction) . F or ℜ ϖ > 0 and ω in a c omp act subset of the physic al she et, ther e exists a unique o dd de c aying Jost solution and a unique even de c aying Jost b asis such that u ∞ ( r , ω ) = exp( − ϖ r ) r − κ (1 + η ( r , ω )) , η ( r , ω ) → 0 , as r → ∞ . Mor e over, for every j, k ≥ 0 ,    ∂ j r ∂ k ω η ( r, ω )    ≤ C j,k,ℓ r − 1 − j for r suﬃciently lar ge. 26 Pr o of. Insert the Jost ansatz into the equation and integrate twice from inﬁnit y . Because the remainder W ℓ,P is O ( r − 3 ) , the resulting integral equation is V olterra with an integrable k ernel after the Coulom b factor has b een remov ed. A con traction argument on a large half-line gives existence and uniqueness. Diﬀeren tiating the in tegral equation with resp ect to r and ω yields the stated b ounds. The V olterra representation is the right to ol b oth for contin uation in ω and for thresh- old asymptotics. Near ω = ± µ , the deca y factor exp( − ϖr ) becomes w eak, but the normalization abov e still captures the exact singular dep endence on ϖ and κ . Lemma 5.5 (Whittaker–V olterra construction) . Fix a br anch of ϖ = √ µ 2 − ω 2 on the slit strip and assume ℜ ϖ ≥ 0 . Then for R ≥ R 0 suﬃciently lar ge ther e exists a unique outgoing matrix solution F out ( r , ω ) of the r e duc e d system such that F out ( r , ω ) = W 0 ( r , ω )  Id n + Q ( r, ω )  , Q ( r , ω ) = O ( r − 1 ) , (5.3) as r → ∞ , wher e W 0 ( r , ω ) = diag  W κ,ν j (2 ϖ r )  n j =1 . (5.4) Mor e over F out dep ends analytic al ly on ω on the slit strip. Pr o of. Let M 0 ( r , ω ) = diag ( M κ,ν j (2 ϖ r )) n j =1 and W 0 ( r , ω ) as in (5.4). Since the remainder in (4.1) is O ( r − 3 ) , v ariation of constan ts giv es the V olterra equation F out ( r ) = W 0 ( r ) − Z ∞ r G 0 ( r , s ; ω ) R ( s, ω ) F out ( s ) d s, (5.5) where G 0 is the diagonal mo del Green matrix built from ( M 0 , W 0 ) . Standard Whittaker asymptotics imply sup r ≥ R Z ∞ r ∥ G 0 ( r , s ; ω ) R ( s, ω ) ∥ d s ≤ C R − 1 for R large, uniformly on compact subsets of the slit strip. Hence the V olterra op erator is a con traction on L ∞ ([ R, ∞ )) , the Neumann series conv erges, and the resulting solution is analytic in ω . Step 4. Ev ans determinan t and meromorphic con tin uation The resolven t is meromorphically con tin ued b y matc hing the left and righ t bases con- structed in Steps 2 and 3. In the o dd sector the relev ant scalar quan tity is the W ronskian E odd ℓ ( ω ) = Q [ u − hor , u ∞ ] , while in the even sector one uses the 2 × 2 matching matrix formed by ev aluating the horizon basis and the inﬁnit y basis at one common radius. Because b oth bases solv e the same symmetric equation, the determinant of this matc hing matrix is independent of the matc hing radius. Prop osition 5.6 (Ev ans determinant and con tinuation) . F or e ach ﬁxe d ℓ ≥ 1 , the o dd W r onskian and the even matching determinant extend analytic al ly fr om the physic al half- plane into the slit strip. The cut-oﬀ r esolvent kernels extend mer omorphic al ly ther e, and their p oles ar e pr e cisely the zer os of the c orr esp onding Evans determinant. 27 Pr o of. The horizon basis is analytic by Lemma 5.2. The inﬁnit y basis is analytic for ℜ ϖ > 0 and extends across the slit by contin uation of the V olterra integral equation in the v ariable ϖ , keeping trac k of the c hosen branch. The Green kernel form ula of Lemma 3.2 then shows that every p ossible singularity of the con tinued resolv en t is caused b y failure of the tw o bases to span the solution space, equiv alently b y v anishing of the matching determinan t. Since the determinant is analytic, the singularities are meromorphic. This step is where the matrix nature of the ev en sector matters most. The contin uation argumen t has the same analytic structure as in the scalar case, but the matc hing ob ject is now a gen uine 2 imes 2 determinant and the threshold analysis m ust b e p erformed only after diagonalization of the leading matrix asymptotics. W e conv ert the second-order system to ﬁrst order: Y = W W ′ ! ∈ C 2 n , Y ′ = A ( r , ω ) Y , A ( r, ω ) = 0 Id n −V ( r , ω ) 0 ! . (5.6) Prop osition 5.7 (Ev ans determinant and matc hing matrix) . Fix r 0 ∈ ( R 0 , ∞ ) . L et Y hor ( r , ω ) b e the 2 n × n matrix obtaine d fr om F hor and its derivative, and let Y out ( r , ω ) b e the c orr esp onding matrix for F out . Set Y ℓ ( r , ω ) := h Y hor ( r , ω ) Y out ( r , ω ) i ∈ C 2 n × 2 n , (5.7) and deﬁne the Evans determinant E ℓ ( ω ) := det Y ℓ ( r 0 , ω ) . (5.8) Then: (a) E ℓ ( ω ) is analytic on the slit strip; (b) E ℓ ( ω ) = 0 if and only if ther e exists a nontrivial mo de satisfying the horizon and outgoing b oundary c onditions simultane ously; (c) the ﬁrst-or der Gr e en matrix, and henc e the ﬁxe d-mo de se c ond-or der Gr e en kernel, is mer omorphic on the slit strip, with p oles pr e cisely at the zer os of E ℓ . Pr o of. Because b oth blo cks in (5.7) solve the same ﬁrst-order system (5.6), Liouville’s form ula giv es ∂ r det Y ℓ ( r , ω ) = tr A ( r , ω ) det Y ℓ ( r , ω ) = 0 , since tr A = 0 . Th us the determinan t is indep endent of the matching radius. Analyticity in ω follo ws from Lemmas 5.3 and 5.5. If E ℓ ( ω ) = 0 , then the columns of Y hor ( r 0 , ω ) and Y out ( r 0 , ω ) are linearly dep enden t, so there exist nonzero v ectors a and b with Y hor ( r 0 , ω ) a = Y out ( r 0 , ω ) b. Uniqueness for ﬁrst-order ODEs implies that the corresp onding horizon and outgoing so- lutions agree for all r , hence determine a nontrivial global mo de satisfying b oth b oundary conditions. The con v erse implication is immediate from the same uniqueness argumen t, pro ving part (b). F or the Green matrix, ﬁx r ′ and write the solution on the tw o sides of r ′ as a linear com bination of the horizon and outgoing bases. Con tinuit y of the ﬁeld and the canonical 28 jump of the deriv ative lead to a linear system whose co eﬃcient matrix is exactly Y ℓ ( r ′ , ω ) . The co eﬃcien ts of the solution are therefore rational functions of the entries of Y ℓ ( r ′ , ω ) − 1 and hence meromorphic with p oles only at zeros of E ℓ . Pro jecting back to the second-order v ariables yields part (c). Corollary 5.8 (Pro of of Theorem 1.1(i)) . Item (i) of The or em 1.1 holds. Pr o of. By Proposition 5.7, the ﬁxed-mode resolv ent is meromorphic on the slit strip. Zeros of a nontrivial analytic function are discrete, so every compact subset contains only ﬁnitely man y p oles. Step 5. Discrete p oles Meromorphic contin uation alone does not y et sa y an ything about the location or m ulti- plicit y of p oles. One must also show that p oles are discrete and cannot accum ulate in compact subsets of the slit strip. This follo ws from analyticity of the Ev ans determinan t together with the fact that nontrivial kernel elemen ts solve a ﬁnite-dimensional matching problem. Lemma 5.9 (Local discreteness of poles) . F or e ach ﬁxe d ℓ ≥ 1 , the p ole set of the c ontinue d cut-oﬀ r esolvent in the slit strip is discr ete, with ﬁnite algebr aic multiplicity at every p ole and no ac cumulation in c omp act subsets. Pr o of. By Prop osition 5.6, poles coincide with zeros of the Ev ans determinant. The latter is an analytic scalar function in the o dd sector and an analytic determinan t in the ev en sector. The identit y theorem implies that its zeros are isolated unless it v anishes iden tically . The latter p ossibility is excluded b ecause for ℑ ω ≫ 1 the resolven t is b ounded and the matching determinan t is nonzero. The ﬁnite algebraic m ultiplicit y of p oles follows from the order of v anishing of an analytic function. The discrete p oles include quasinormal mo des and, in the low er half-plane near the real axis, quasib ound p oles asso ciated with stable timelik e trapping. Their presence is compatible with all the branch-cut tail theorems prov ed later b ecause the pole con tribu- tion is explicitly separated from the contin uous sp ectral con tribution. Lemma 5.10 (Discrete p ole set) . F or e ach ﬁxe d ℓ , the p ole set of the ﬁxe d-mo de r esolvent in the slit strip c onsists of isolate d p oints with no ac cumulation in c omp act subsets. Pr o of. The p oles are the zeros of E ℓ ( ω ) and are therefore isolated unless E ℓ v anishes iden tically . The latter is imp ossible b ecause for large p ositive imaginary ω the horizon and outgoing manifolds are transv erse. Step 6. Exclusion of real cut p oles and threshold resonances This is the dynamical core of the ﬁxed-mo de sp ectral theorem. Supp ose ﬁrst that ω ∈ ( − µ, µ ) is a real cut frequency and that a nontrivial separated solution is ingoing at the future horizon and decaying at inﬁnit y . Ev aluating the radial curren t b etw een t wo radial slices and taking limits at the horizon and at inﬁnit y yields a contradiction: the horizon ﬂux is nonzero unless the solution v anishes, while the deca ying inﬁnit y condition forces the curren t to v anish there. 29 A t the threshold ω = ± µ one m ust work slightly harder b ecause the inﬁnity b ehavior is no longer exp onen tially deca ying. A threshold resonance would corresp ond to a nontrivial b ounded solution in the threshold normal form. The absence of such a resonance is pro v ed b y matching the threshold asymptotics to the channel energy identit y and using the static no-hair input at ω = 0 when required. Prop osition 5.11 (Real-frequency exclusion via the radial current) . Fix ℓ ≥ 1 . Ther e is no nontrivial o dd or even channel solution which is ingoing at the futur e horizon and either de c aying at inﬁnity for | ω | < µ or b ounde d in the thr eshold sense at ω = ± µ . In p articular, ther e is no r e al cut p ole and no thr eshold r esonanc e. Pr o of. Let u denote the scalar or vector solution. The conserv ed current Q [ u , u ] is purely imaginary and indep endent of r ∗ . At the future horizon, the ingoing b ehavior giv es Q [ u , u ] = − 2i ω | a hor | 2 for some nonzero amplitude vector a hor unless u ≡ 0 . F or | ω | < µ , the decaying condition at inﬁnit y gives Q [ u , u ] → 0 as r ∗ → + ∞ , con tradiction. A t threshold, one uses the explicit threshold normal form to show that every resonan t solution has v anishing curren t at inﬁnity and is therefore again trivial. If ω = 0 , the argumen t is com bined with the static Proca no-hair theorem to exclude nontrivial static b ound states. The same identit y will later imply mo de stabilit y in the op en upper half-plane once com bined with unitarit y of the exact c hannel evolution. Lemma 5.12 (Separated radial current) . Fix ℓ ≥ 1 . F or every r e al fr e quency ω and every sep ar ate d solution of the r e duc e d o dd/even system, ther e exists a sc alar r adial curr ent J ℓ,ω [ U ]( r ) with the fol lowing pr op erties: (a) ∂ r ∗ J ℓ,ω [ U ] = 0 ; (b) if U ( r ) = a e − i ω r ∗ + o (1) as r ∗ → −∞ , then J ℓ,ω [ U ]( r ) = − ω q ℓ ( a ) + o (1) , wher e q ℓ is a p ositive deﬁnite Hermitian form on the horizon data; (c) if U is exp onential ly de c aying for | ω | < µ or thr eshold-sub or dinate for | ω | = µ , then lim r →∞ J ℓ,ω [ U ]( r ) = 0 . Pr o of. The curren t is the separated stationary energy ﬂux asso ciated with the Killing ﬁeld ∂ t . After pro jection on to vector spherical harmonics one obtains, for each ﬁxed ℓ , a p ositiv e deﬁnite quadratic form on the reduced mode v ariables and their r ∗ -deriv atives. Its imaginary p olarization yields a conserv ed sesquilinear ﬂux J ℓ,ω . At the horizon, the asymptotic U ( r ) = a e − i ω r ∗ (1 + o (1)) giv es the stated limit with a p ositive deﬁnite form q ℓ ; p ositivit y is exactly the absence of sup erradiance for a neutral ﬁeld on a static Reissner– Nordström bac kground. A t inﬁnit y , exp onen tially deca ying or threshold-sub ordinate so- lutions carry zero ﬂux b ecause b oth U and U ′ are nonoscillatory and v anish suﬃcien tly fast. 30 Prop osition 5.13 (No real cut p oles and no threshold resonances) . Ther e is no non- trivial sep ar ate d Pr o c a mo de of the form A = e − i ω t b A ( r , ω ) with r e al ω ∈ [ − µ, µ ] that is ingoing at the futur e horizon and exp onential ly de c aying or thr eshold-sub or dinate at inﬁnity. Conse quently E ℓ ( ω )  = 0 for every r e al ω ∈ [ − µ, µ ] , in p articular at ω = ± µ . Pr o of. Assume ﬁrst that ω ∈ [ − µ, µ ] \ { 0 } . By Lemma 5.12, the radial current is constant. The inﬁnit y b oundary condition giv es lim r →∞ J ℓ,ω [ U ]( r ) = 0 , while the ingoing horizon asymptotic giv es lim r ∗ →−∞ J ℓ,ω [ U ]( r ) = − ω q ℓ ( a ) . Since q ℓ is p ositiv e deﬁnite and ω  = 0 , this forces a = 0 and hence U ≡ 0 by uniqueness at the regular singular horizon. It remains to treat the static case ω = 0 . In the o dd c hannel, u ′′ 4 − f  µ 2 + ℓ ( ℓ + 1) r 2  u 4 = 0 . Multiplying b y u 4 , in tegrating ov er r ∗ ∈ ( −∞ , ∞ ) , and integrating b y parts yields Z ∞ −∞  | u ′ 4 | 2 + f  µ 2 + ℓ ( ℓ + 1) r 2  | u 4 | 2  d r ∗ = 0 , so u 4 ≡ 0 . In the ev en sector, a static regular deca ying solution w ould deﬁne a smooth ﬁnite- energy static Pro ca ﬁeld on an asymptotically ﬂat static blac k-hole exterior. Bekenstein’s static Pro ca no-hair theorem excludes such a ﬁeld [7, 8]. Hence the even static mo de also v anishes. Therefore E ℓ ( ω )  = 0 on [ − µ, µ ] . Corollary 5.14 (Pro of of Theorem 1.1(ii)) . Item (ii) of The or em 1.1 holds. Step 7. Small- κ threshold asymptotics The intermediate tails come from the regime κ = M µ 2 /ϖ ≪ 1 . In this range the Coulom b factor is p erturbative and the leading mo del is an in verse-square equation. After scaling x = ϖ r , the c hannel equation b ecomes u xx +  − 1 + 2 κ x − ν 2 ℓ,P − 1 4 x 2 + O ( ϖ x − 3 )  u = 0 . A t κ = 0 the decaying solution is a mo diﬁed Bessel function of order ν ℓ,P . The jump across the cut therefore comes from the classical discontin uit y of the Bessel mo del, and the ﬁrst correction is linear in κ . Lemma 5.15 (Small- κ threshold expansion) . F or e ach ﬁxe d ℓ ≥ 1 and p olarization P , disc G ℓ,P ( ω ; r , r ′ ) = a ℓ,P ( r , r ′ ) ϖ 2 ν ℓ,P + O  κ ϖ 2 ν ℓ,P  + O  ϖ 2 ν ℓ,P +2  , uniformly on c omp act r , r ′ -sets as ϖ → 0 with κ → 0 . 31 Pr o of. The deca ying inﬁnity solution is matc hed to the Bessel mo del on the scale x = ϖ r . Since the Coulomb term carries one explicit factor of κ , a p erturbativ e V olterra argument around the Bessel equation yields the ﬁrst error term. The next correction comes from the shorter-range remainder of the channel p otential and is quadratic in the threshold scale. The horizon side is analytic in ϖ and therefore con tributes only to the amplitude a ℓ,P ( r , r ′ ) . This is the precise sp ectral statemen t b ehind the p olarization-resolved in termediate exp onen ts. The exp onen t is not guessed from formal dimensional analysis; it is the exact p o w er determined b y the inv erse-square threshold index ν ℓ,P . In tro duce the scaled v ariable x = ϖ r . Then (4.1) b ecomes ∂ 2 x w P +  − 1 − ν 2 ℓ,P − 1 4 x 2 + 2 κ x + B P ( x ; ϖ , κ )  w P = 0 , (5.9) with B P ( x ; ϖ , κ ) = O  ϖ x 3  . F or κ ≪ 1 , the Coulomb term is p erturbative and the mo del equation is the mo diﬁed Bessel equation. Lemma 5.16 (Small- κ outgoing basis) . A ssume κ ≤ κ 1 and ϖ ≤ ϖ 1 with κ 1 , ϖ 1 suﬃ- ciently smal l. Then the outgoing b asis has the r epr esentation g ℓ,P ( r , ω ) = √ ϖ r K ν ℓ,P ( ϖ r )  1 + E K ( r , ω )  + √ ϖ r I ν ℓ,P ( ϖ r ) E I ( r , ω ) , (5.10) wher e E K ( r , ω ) = O ( κ ) + O ( ϖ 2 ) , E I ( r , ω ) = O ( κ ) + O ( ϖ 2 ) uniformly for r in c omp act subsets of ( r + , ∞ ) . Pr o of. W rite the exact equation as the Bessel mo del plus the p erturbation 2 κ x + B P ( x ; ϖ , κ ) and use a V olterra equation around the basis { √ x I ν ℓ,P ( x ) , √ x K ν ℓ,P ( x ) } . Since κ ≪ 1 and B P = O ( ϖ x − 3 ) , the V olterra op erator is small on compact r -sets and the represen tation follo ws. Prop osition 5.17 (Small- κ threshold discon tinuit y) . In the r e gime κ ≪ 1 one has disc G ℓ,P ( ω ; r , r ′ ) = a ℓ,P ( r , r ′ ) ϖ 2 ν ℓ,P + O  κ ϖ 2 ν ℓ,P  + O  ϖ 2 ν ℓ,P +2  (5.11) uniformly for r , r ′ in c omp act subsets of ( r + , ∞ ) . Pr o of. The discon tin uity comes from analytic con tinuation of the outgoing basis across the cut. Since K ν (e π i z ) − K ν (e − π i z ) = π i sin( π ν ) sin( π ν ) I ν ( z ) , the leading jump is carried by the co eﬃcient of I ν ℓ,P in (5.10). The prefactor is precisely ϖ 2 ν ℓ,P , while the V olterra corrections produce the stated O ( κ ϖ 2 ν ℓ,P ) and O ( ϖ 2 ν ℓ,P +2 ) errors. 32 Step 8. Large- κ threshold asymptotics The very-late tail is go v erned by the opp osite regime κ ≫ 1 . Here the Coulom b term is dominan t and the correct mo del is the Whittaker equation u xx +  − 1 4 + κ x + 1 4 − ν 2 ℓ,P x 2  u = 0 , x = 2 ϖ r . Its distinguished decaying solution is W κ,ν ℓ,P ( x ) , whose mono drom y in κ and asymptotics as x → 0 are resp onsible for the oscillatory factors e ± 2 π i κ app earing in the cut jump. Lemma 5.18 (Large- κ Whittak er expansion) . A s ϖ → 0 with κ → ∞ , disc G ℓ,P ( ω ; r , r ′ ) = b + ℓ,P ( r , r ′ ) e 2 π i κ + b − ℓ,P ( r , r ′ ) e − 2 π i κ + O ( κ − 1 ) + O ( ϖ ) , uniformly on c omp act r , r ′ -sets. Pr o of. The inﬁnity basis is matc hed to the Whittak er mo del on the scale x = 2 ϖ r . The analytic con tinuation across the slit c hanges κ by sign and pro duces the t wo mono drom y factors e ± 2 π i κ . The shorter-range remainder con tributes only lo wer-order corrections. Uniformit y on compact radial sets follows b ecause the near-zone transfer matrix betw een the Whittak er region and the compact set is analytic and bounded. The large- κ expansion is the exact p oin t where the universal t − 5 / 6 exp onen t enters. Ev ery p olarization dep endence is conﬁned to the amplitudes b ± ℓ,P and to constant phases, while the oscillatory exp onential carries the same saddle structure as in the scalar case. When κ ≫ 1 , the Whittak er connection form ulas carry the Coulomb mono dromy e ± 2 π i κ . Prop osition 5.19 (Large- κ threshold discon tinuit y) . In the r e gime κ ≫ 1 one has disc G ℓ,P ( ω ; r , r ′ ) = b + ℓ,P ( r , r ′ ) e 2 π i κ + b − ℓ,P ( r , r ′ ) e − 2 π i κ + O ( κ − 1 ) + O ( ϖ ) (5.12) uniformly for r , r ′ in c omp act subsets of ( r + , ∞ ) . Pr o of. Use the Whittaker basis (4.4) and the Whittak er connection form ulas relating W κ,ν and M κ,ν . Up on contin uation around ϖ = 0 , the coeﬃcients pic k up the mon- o drom y factors e ± 2 π i κ . The V olterra correction around the exact equation contributes only O ( κ − 1 ) + O ( ϖ ) on compact r -sets. Step 9. Assem bly With the ingredients of Steps 1–8 in place, the pro of of Theorem 1.1 b ecomes a b o okkeep- ing exercise. The horizon and inﬁnity constructions deﬁne the Ev ans determinant and the con tinued Green k ernel. Lemma 5.9 giv es the lo cal discreteness of p oles. Prop osi- tion 5.11 excludes real cut p oles and threshold resonances. Lemmas 5.15 and 5.18 provide the explicit threshold discontin uity form ulas in the intermediate and very-late regimes. It is useful to stress that no hidden case distinction remains. The o dd sector is scalar and the even sector is matrix v alued, but the pro of is organized so that ev ery scalar statemen t has a matrix coun terpart with the same logical role. Likewise, the charge parameter Q aﬀects the threshold index and the amplitudes, but it do es not alter the univ ersal Coulom b co eﬃcien t or the short-range nature of the horizon end. 33 Completion of the pr o of of The or em 1.1. Assertions (i) and (ii) follow from Proposi- tion 5.6 together with Lemma 5.9 and Prop osition 5.11. Assertion (iii) is Lemma 5.15, and assertion (iv) is Lemma 5.18. Finally , assertion (v) follo ws b y expanding the exact in v erse-square co eﬃcient in the polarization basis and comparing it to the Sc hw arzschild v alues L P + 1 2 ; the Q -dep endence enters only at order ( Qµ ) 2 and the geometric mass correction en ters at order ( M µ ) 2 . Corollary 5.20 (Proof of Theorem 1.1) . Items (iii) – (v) of The or em 1.1 fol low fr om Pr op ositions 5.17 and 5.19 to gether with the index exp ansion (4.2) . Henc e The or em 1.1 is pr ove d. 6 Mo de stabilit y and threshold resonance exclusion This section isolates the dynamical meaning of Step 6 of Section 5. F or the ﬁxed- ℓ reduced Hamiltonian, unstable separated mo des in the op en upp er half-plane are ruled out b y unitarity of the exact energy ev olution, while Prop osition 5.13 excludes real cut p oles and threshold resonances. Prop osition 6.1 (Selfadjoint ﬁxed-mo de energy evolution) . F or e ach ﬁxe d angular mo- mentum ℓ , the r e duc e d Pr o c a system deﬁnes a p ositive c onserve d ener gy on the ﬁnite- ener gy sp ac e h ℓ , and the c orr esp onding time evolution is gener ate d by a selfadjoint op er- ator A ℓ . Pr o of. By Prop osition 3.1, eac h reduced c hannel Hamiltonian H ♯ ℓ is selfadjoint and semi- b ounded on H 2 . After reconstructing the ph ysical o dd and even v ariables from the c hannel v ariables, the ﬁxed- ℓ energy ma y be written as E ℓ [( u, v )] = 1 2 X ♯  ⟨ ( H ♯ ℓ + c ℓ ) u ♯ , u ♯ ⟩ + ∥ v ♯ ∥ 2 L 2  , with c ℓ c hosen so that ev ery summand is p ositive. This norm is equiv alent to the natural ﬁnite-energy norm on the reduced mo de space h ℓ b ecause the Lorenz co nstraint has already been eliminated in the reduction. On the dense domain H 2 × H 1 , the ﬁrst-order generator is the orthogonal sum of the c hannel generators from Prop osition 3.1, conjugated b y the ﬁxed reconstruction matrices. It is therefore selfadjoin t, and the corresp onding unitary group preserv es E ℓ . This is the ﬁxed-mo de energy evolution used in the mo de-stability argument. Pr o of of The or em 1.6. By Prop osition 6.1, the ﬁxed- ℓ reduced dynamics is generated by a selfadjoin t op erator A ℓ on the ﬁnite-energy space h ℓ . If A = e − i ω t b A w ere a non trivial ﬁnite-energy separated mo de with ℑ ω > 0 , then e − i t A ℓ b A = e − i ω t b A w ould hav e norm e ℑ ω t    b A    h ℓ , con tradicting unitarity of the exact evolution. Hence no upp er-half-plane mo de exists. The statement on the real cut and at threshold is exactly Prop osition 5.13. Since E ℓ is analytic on the slit strip b y Prop osition 5.7, zero-freeness at ω = ± µ yields zero-free punctured neighborho o ds of the thresholds. 34 Corollary 6.2 (Threshold zero-free neighborho o ds) . Ther e exist δ > 0 and η th > 0 such that n ω : 0 < | ω ∓ µ | < δ, − η th < ℑ ω < η th o \ [ − µ, µ ] c ontains no zer o of E ℓ . Pr o of. This is the last assertion of Theorem 1.6. 7 Explicit branc h-cut tails This section turns the threshold discontin uity formulas in to explicit oscillatory late-time tails and makes the Sc hw arzschild-to-Reissner–Nordström transition precise. 7.1 Oscillatory in v ersion form ula and decomp osition of fre- quency space Let χ ∈ C ∞ 0 (( r + , ∞ )) be identically 1 on the compact radial set under consideration. After con tour deformation, the ﬁxed-mo de branc h-cut contribution can b e written as u bc ℓm,P ( t, r, r ′ ) = 1 2 π i Z µ − µ e − i ω t χ ( r ) χ ( r ′ ) disc G ℓ,P ( ω ; r , r ′ ) d ω . (7.1) T o extract asymptotics, w e split the cut in tegral in to three regions: (i) an upper-endp oint region µ − ω ≲ t − 1+ σ ; (ii) a lo wer-endpoint region µ + ω ≲ t − 1+ σ ; (iii) a cen tral region separated from b oth thresholds. The cen tral region contributes sup er-p olynomially after rep eated integration by parts b e- cause the phase deriv ativ e do es not v anish there. The endp oint regions are then analyzed in the v ariable ϖ = √ µ 2 − ω 2 and split further according to the size of κ = M µ 2 /ϖ . This decomp osition is exactly what separates the intermediate and very-late regimes. In the in termediate regime one sta ys in the endp oint zone where κ ≪ 1 and the Bessel expansion of Step 7 is v alid. In the very-late regime the dominant contribution comes from the part of the endp oin t zone where κ ≫ 1 and the Whittaker mono dromy of Step 8 pro duces a saddle p oint of the phase. 7.2 Sc h w arzsc hild-to-Reissner–Nordström corresp ondence Theorem 7.1 (Sc hw arzschild-to-Reissner–Nordström tail correspondence) . Fix ℓ ≥ 1 and one p olarization P ∈ {− 1 , 0 , +1 } . (a) Schwarzschild. A ssume Q = 0 and κ ∗ ( t ) → 0 . Then the thr eshold index is exactly ν ℓ,P = L P + 1 2 , so the interme diate br anch-cut tails ar e explicitly u bc ℓm, − 1 ( t, r, r ′ ) = A ℓ, − 1 ( r , r ′ ) t − ( ℓ +1 / 2) sin( µt + δ ℓ, − 1 ) + O  κ ∗ ( t ) t − ( ℓ +1 / 2)  + O  t − ( ℓ +3 / 2)  , u bc ℓm, 0 ( t, r, r ′ ) = A ℓ, 0 ( r , r ′ ) t − ( ℓ +3 / 2) sin( µt + δ ℓ, 0 ) + O  κ ∗ ( t ) t − ( ℓ +3 / 2)  + O  t − ( ℓ +5 / 2)  , u bc ℓm, +1 ( t, r, r ′ ) = A ℓ, +1 ( r , r ′ ) t − ( ℓ +5 / 2) sin( µt + δ ℓ, +1 ) + O  κ ∗ ( t ) t − ( ℓ +5 / 2)  + O  t − ( ℓ +7 / 2)  . 35 (b) R eissner–Nor dstr öm. F or gener al sub extr emal Q , the same formulas hold with the half-inte ger exp onents r eplac e d by ν ℓ,P + 1 . In the smal l-mass r e gime ( M µ ) 2 + ( Qµ ) 2 ≪ 1 , these exp onents r e duc e to the Schwarzschild values up to O (( M µ ) 2 + ( Qµ ) 2 ) c orr e ctions. (c) Universal very-late tail. If κ 0 ( t ) → ∞ , then in every sub extr emal R eissner– Nor dstr öm exterior u bc ℓm,P ( t, r, r ′ ) = B ℓ,P ( r , r ′ ; Q ) t − 5 / 6 sin  µt − 3 2 (2 π M µ ) 2 / 3 ( µt ) 1 / 3 + δ ℓ,P, 0 ( Q ) + o (1)  , and the exp onent 5 / 6 is indep endent of ℓ , of the p olarization, and of Q . Pr o of. When Q = 0 , Prop osition 2.1 reduces the ev en sector to an exact diagonal r − 2 matrix with eigen v alues L ± 1 ( L ± 1 + 1) and short-range oﬀ-diagonal remainder O ( r − 3 ) ; the o dd channel has the exact scalar r − 2 co eﬃcien t L 0 ( L 0 + 1) . Hence the threshold normal form (4.1) has ν ℓ,P = L P + 1 2 exactly . The in termediate formulas then follow from Theorem 7.3, and the very-late form ula is Theorem 7.6. The general Reissner–Nordström statemen t is exactly Theorems 7.3 and 7.6 together with Corollary 7.4. 7.3 In termediate tails The in termediate asymptotics come from the endp oint singularity disc G ℓ,P ∼ ϖ 2 ν ℓ,P . Lemma 7.2 (Endp oint oscillatory in tegral) . L et α > − 1 and a > 0 . Then, as t → ∞ , Z ε 0 e i aϖ 2 t ϖ α d ϖ = 1 2 Γ  α + 1 2  e π i 4 ( α +1) a − ( α +1) / 2 t − ( α +1) / 2 + O  t − ( α +3) / 2  . (7.2) Pr o of. Set s = aϖ 2 t . The claim follows from the standard con tour represen tation of the gamma function. Theorem 7.3 (Intermediate branch-cut asymptotics) . If κ ∗ ( t ) = M µ 3 / 2 t 1 / 2 → 0 , then u bc ℓm,P ( t, r, r ′ ) = A ℓ,P ( r , r ′ ; Q ) t − ( ν ℓ,P +1) sin( µt + δ ℓ,P ( Q )) + O  κ ∗ ( t ) t − ( ν ℓ,P +1)  + O  t − ( ν ℓ,P +2)  , (7.3) uniformly for r , r ′ in c omp act subsets of ( r + , ∞ ) . Pr o of. Insert (1.5) into (4.9). Near the upp er endp oin t, ω = µ − ϖ 2 2 µ + O ( ϖ 4 ) , d ω = − ϖ µ d ϖ + O ( ϖ 3 )d ϖ . Hence the leading upp er-endp oint con tribution is a ℓ,P ( r , r ′ ) 2 π µ e − i µt Z ε 0 e i t 2 µ ϖ 2 ϖ 2 ν ℓ,P +1 d ϖ . By Lemma 7.2 with α = 2 ν ℓ,P + 1 , this contributes t − ( ν ℓ,P +1) . The O ( κ ϖ 2 ν ℓ,P ) term pro duces the stated κ ∗ ( t ) t − ( ν ℓ,P +1) error, and the O ( ϖ 2 ν ℓ,P +2 ) term yields O ( t − ( ν ℓ,P +2) ) . A dding the lo wer endp oint gives the sine form. 36 Corollary 7.4 (Small-mass explicit exp onen ts) . L et ε µ,Q := ( M µ ) 2 + ( Qµ ) 2 . F or ℓ ≥ 1 , if ε µ,Q ≪ 1 then ν ℓ, − 1 = ℓ − 1 2 + O ( ε µ,Q ) , ν ℓ, 0 = ℓ + 1 2 + O ( ε µ,Q ) , ν ℓ, +1 = ℓ + 3 2 + O ( ε µ,Q ) . Henc e the le ading smal l-mass interme diate exp onents ar e P = − 1 : ℓ + 1 2 , P = 0 : ℓ + 3 2 , P = +1 : ℓ + 5 2 . F or ℓ = 0 , only the ele ctric monop ole P = +1 r emains and its le ading smal l-mass inter- me diate exp onent is 5 / 2 . Pr o of. This is immediate from (4.2) and the v alues (1.4). 7.4 The univ ersal v ery-late tail Prop osition 7.5 (Mo del saddle estimate) . L et I ( t ) = Z ε 0 ( β ϖ + O ( ϖ 2 ))e iΨ t ( ϖ ) d ϖ , (7.4) wher e Ψ t ( ϖ ) = tϖ 2 2 µ + 2 π M µ 2 ϖ . (7.5) Then, as t → ∞ , I ( t ) = β ϖ 0 ( t ) s 2 π µ 3 t e iΨ t ( ϖ 0 ( t ))+ π i / 4 + O  ( κ 0 ( t ) − 1 + ϖ 0 ( t )) t − 5 / 6  , (7.6) wher e ϖ 0 ( t ) = (2 π M µ 3 /t ) 1 / 3 . Pr o of. Set ϖ = t − 1 / 3 y . Then the phase b ecomes i t 1 / 3 Φ( y ) with Φ( y ) = y 2 2 µ + 2 π M µ 2 y . The function Φ has a unique nondegenerate critical p oin t at y 0 = (2 π M µ 3 ) 1 / 3 . Stationary phase with large parameter t 1 / 3 yields an extra factor t − 1 / 6 on top of the substitution factor t − 2 / 3 , giving the scale t − 5 / 6 . Theorem 7.6 (Universal very-late asymptotics) . If κ 0 ( t ) → ∞ , then u bc ℓm,P ( t, r, r ′ ) = B ℓ,P ( r , r ′ ; Q ) t − 5 / 6 sin  µt − 3 2 (2 π M µ ) 2 / 3 ( µt ) 1 / 3 + δ ℓ,P, 0 ( Q ) + O ( κ 0 ( t ) − 1 + ϖ 0 ( t ))  + O  κ 0 ( t ) − 1 + ϖ 0 ( t )  t − 5 / 6 . (7.7) The exp onent 5 / 6 is indep endent of ℓ , of the p olarization, and of the black-hole char ge Q . 37 Pr o of. By (1.6), the branc h-cut jump is a sum of t w o oscillatory pieces with phases ± 2 π κ . Restricting to one endp oint and using ω = µ − ϖ 2 2 µ + O ( ϖ 4 ) giv es an in tegral of the form e − i µt Z ε 0 a ∞ ℓ,P ( ϖ ; r, r ′ )e iΨ t ( ϖ ) d ϖ , where a ∞ ℓ,P ( ϖ ; r, r ′ ) = β ℓ,P ( r , r ′ ; Q ) ϖ + O (( κ − 1 + ϖ ) ϖ ) . Prop osition 7.5 applies directly and yields the stated asymptotic. The blac k-hole c harge modiﬁes only the amplitude and the constan t phase shift, not the saddle exp onent. 8 Large-angular-momen tum analysis and branc h-cut full-ﬁeld deca y 8.1 Large-angular-momen tum scaling and barrier geometry T o sum the branc h-cut contribution o v er all angular momen ta, w e need uniform con trol as ℓ → ∞ on compact radial sets. There are really tw o tasks here. One is analytic: sho w that the reduced c hannel kernels gro w at most p olynomially in ℓ . The other is geometric: explain wh y , for large ℓ , the compact radial region lies w ell inside the forbidden zone where an elliptic/WKB argument applies. F or ℓ ≥ 1 we set h = ( ℓ + 1 2 ) − 1 . On a compact radial set K ⋐ ( r + , ∞ ) , the principal in verse-square term b eha ves lik e h − 2 and dominates the b ounded mass term. After conjugation b y the asymptotic p olarization basis, eac h channel therefore tak es the semiclassical form P ℓ,P ( h, ω ) = h 2 D 2 r ∗ + V 0 ,P ( r ) + hV 1 ,P ( r , ω ) + h 2 V 2 ,P ( r , ω ) , with V 0 ,P ( r ) = f ( r ) L P ( L P + 1) ( ℓ + 1 2 ) 2 r 2 . Since K stays aw ay from both the horizon and inﬁnit y , V 0 ,P is uniformly p ositiv e on K for large ℓ . The next lemma isolates the corresp onding barrier geometry . Lemma 8.1 (Large- ℓ barrier geometry and turning p oints) . Fix a c omp act fr e quency interval J ⋐ [ − µ, µ ] and a c omp act r adial set K ⋐ ( r + , ∞ ) . Ther e exist h 0 > 0 , a cutoﬀ neighb orho o d f K ⋐ ( r + , ∞ ) c ontaining K , and smo oth functions r − ,P ( ω ; h ) < r + ,P ( ω ; h ) , ( ω , h ) ∈ J × (0 , h 0 ) , P ∈ {− 1 , 0 , +1 } , such that the fol lowing hold uniformly in P : (i) the sc alarize d channel symb ol q P ( r , ξ ; ω , h ) := ξ 2 + V 0 ,P ( r ) + hV 1 ,P ( r , ω ) + h 2 V 2 ,P ( r , ω ) − h 2 ω 2 vanishes at ξ = 0 exactly at the two turning p oints r − ,P ( ω ; h ) and r + ,P ( ω ; h ) ; 38 (ii) the turning p oints ar e simple and satisfy r − ,P ( ω ; h ) − r + ≍ h 2 , r + ,P ( ω ; h ) ≍ h − 1 ; (iii) one has q P ( r , ξ ; ω , h ) ≥ c K (1 + ξ 2 ) for r ∈ f K , so no turning p oint interse cts the c omp act r e gion supp orting the cut-oﬀ r esolvent. Pr o of. The principal barrier V 0 ,P ( r ) = f ( r ) L P ( L P + 1) / (( ℓ + 1 2 ) 2 r 2 ) is strictly p ositive on every compact subset of the exterior and v anishes only at the horizon and at spatial inﬁnit y . Near r = r + one has f ( r ) = 2 κ + ( r − r + ) + O (( r − r + ) 2 ) , so solving V 0 ,P ( r ) = h 2 ω 2 giv es the inner turning point asymptotic r − ,P ( ω ; h ) − r + ≍ h 2 . F or large r , one has V 0 ,P ( r ) = r − 2 (1 + O ( r − 1 )) uniformly in the p olarization, so V 0 ,P ( r ) = h 2 ω 2 yields r + ,P ( ω ; h ) ≍ h − 1 . The corrections hV 1 ,P + h 2 V 2 ,P are lo wer order, hence the implicit- function theorem preserv es the tw o simple ro ots and their asymptotics for all suﬃciently small h . Since K is ﬁxed a wa y from the horizon and inﬁnity , shrinking h 0 if necessary giv es the uniform p ositivity of q P on f K . 8.2 Uniform WKB transp ort, cut-oﬀ resolv en ts, and recon- struction The t wo turning p oints from Lemma 8.1 separate the horizon and inﬁnity asymptotic zones from the compact radial region where the large- ℓ summation is carried out. T o compare the threshold bases with the compact cut-oﬀ resolv ent one needs a uniform Liouville–Green transp ort through the forbidden region and a uniform Airy matc hing across the simple turning p oin ts. Prop osition 8.2 (Uniform WKB transp ort and Airy matching) . Fix J ⋐ [ − µ, µ ] and K ⋐ ( r + , ∞ ) . F or e ach p olarization P ∈ {− 1 , 0 , +1 } and e ach suﬃciently smal l h , ther e exist exact channel solutions w mid P, ± ( r , ω ; h ) deﬁne d on the close d interval b etwe en the two turning p oints such that w mid P, ± ( r , ω ; h ) = exp  ± h − 1 Φ P ( r , ω ; h )  a 0 , ± ( r , ω ; h ) + ha 1 , ± ( r , ω ; h )  , wher e Φ P ( r , ω ; h ) = Z r r − ,P ( ω ; h ) q q P ( s, 0; ω , h ) d s ∗ and the amplitudes satisfy uniform symb ol b ounds sup r ∈ K    ∂ α r ∂ β ω a j, ± ( r , ω ; h )    ≤ C αβ j . Mor e over, the tr ansfer matric es fr om the horizon F r ob enius b asis and fr om the inﬁnity V olterr a b asis to the b asis ( w mid P, + , w mid P, − ) ar e O ( ⟨ ℓ ⟩ N ) , to gether with one ω -derivative, uni- formly for ω ∈ J . Pr o of. On eac h subinterv al av oiding the turning p oints, the c hannel equations are scalar or diagonally scalarized second-order equations with co eﬃcien ts admitting complete semi- classical symbol expansions in h . Since the turning p oints are simple b y Lemma 8.1, the standard Liouville–Green construction pro duces exact WKB bases with the stated phase 39 and amplitude b ounds. In neighborho o ds of the turning p oints, the rescaled equations reduce to Airy normal form with uniformly controlled errors; consequently the transfer matrices are the universal Airy connection matrices plus O ( h ) corrections. Comp osing the Airy matching with the horizon F rob enius basis from Section 5 and the inﬁnit y V olterra basis from Section 4 gives the polynomial con trol of the transfer matrices. The polyno- mial loss in ℓ comes only from the constant p olarization change of basis and from one application of Cramer’s rule to the matching matrices. Prop osition 8.3 (Uniform cut-oﬀ resolv ent b ounds on compact radial sets) . Fix J ⋐ [ − µ, µ ] and K ⋐ ( r + , ∞ ) . Ther e exist inte gers M 0 , M 1 ≥ 0 such that for every p olariza- tion P ∈ {− 1 , 0 , +1 } , sup r,r ′ ∈ K    ∂ j ω G ℓ,P ( ω ; r , r ′ )    ≤ C K,J,j ⟨ ℓ ⟩ M j , j = 0 , 1 , uniformly for ω ∈ J and ℓ ≥ 1 . The same b ound holds for the jump disc G ℓ,P ( ω ; r , r ′ ) acr oss the massive cut. Pr o of. Cho ose cutoﬀs χ, e χ ∈ C ∞ 0 (( r + , ∞ )) with χ ≡ 1 on K and e χ ≡ 1 on a neighborho o d of supp χ . By Lemma 8.1, the symbol is uniformly elliptic on supp e χ , so semiclassical el- liptic regularity gives a cut-oﬀ parametrix for the c hannel resolven t there. Prop osition 8.2 transp orts the horizon and inﬁnit y bases to the b oundary of the cut-oﬀ region with only p olynomial loss in ℓ , and Theorem 1.6 excludes real p oles and threshold resonances, so the matc hing determinant is uniformly b ounded a w ay from zero on J . Applying Cramer’s rule to the Green kernel represen tation therefore yields the stated polynomial k ernel b ounds, together with one ω -deriv ative. Prop osition 8.4 (Uniform reconstruction of the ph ysical co eﬃcients) . F or every c omp act K ⋐ ( r + , ∞ ) ther e exists an inte ger N rec ≥ 0 such that the physic al br anch-cut c o eﬃcients satisfy sup r ∈ K    a bc ℓm,P ( t, r )    ≤ C K ⟨ ℓ ⟩ N rec X Q ∈{− 1 , 0 , +1 } X j ≤ 1 sup r ∈ K    ∂ j r v bc Q ( t, r )    , and the same estimate holds for the r esidue pr oje ctors acting on c omp actly supp orte d initial data. Pr o of. The o dd c hannel is already one ph ysical co eﬃcien t. In the even sector one inv erts the constan t matrix T ℓ from (2.13), then uses the Lorenz constraint to recov er the remain- ing algebraic co eﬃcien t. Ev ery co eﬃcient in this reconstruction is a rational function of ℓ with at most p olynomial gro wth, and only one radial deriv ative app ears. Since K sta ys a w ay from the horizon, the co eﬃcients are uniformly b ounded in r , whence the stated p olynomial reconstruction estimate. Let A bc denote the branch-cut part of the full Proca ﬁeld. Expanding in v ector spherical harmonics gives A bc ( t, r, ω ) = ∞ X ℓ =0 ℓ X m = − ℓ X P a bc ℓm,P ( t, r ) Y ( P ) ℓm ( ω ) , (8.1) where for ℓ = 0 only the even electric channel is presen t. T o pass from the ﬁxed-mo de theorems to p oin t wise full-ﬁeld estimates one needs quantitativ e control of the ﬁxed-mo de constan ts as ℓ → ∞ . W e no w pro ve the required uniformity . 40 Pr o of of The or em 1.3. Fix K ⋐ ( r + , ∞ ) and c ho ose cutoﬀs χ, e χ ∈ C ∞ 0 (( r + , ∞ )) suc h that χ ≡ 1 on K and e χ ≡ 1 on a neighborho o d of supp χ . Lemma 8.1, Prop osition 8.2, Prop osition 8.3, and Proposition 8.4 isolate the new large- ℓ input. The remaining argu- men t assem bles these ingredients with the ﬁxed-mo de threshold formulas. The pro of has ﬁv e steps. Step 1. L ow angular momenta. Cho ose ℓ 0 ≥ 1 large enough that the semiclassical argu- men t below applies for ev ery ℓ ≥ ℓ 0 . F or the ﬁnitely man y mo des 0 ≤ ℓ < ℓ 0 , including the monop ole electric channel at ℓ = 0 , the b ound follows b y taking the maxim um of the ﬁxed-mo de constan ts in Theorems 7.3 and 7.6. Hence only the regime ℓ ≥ ℓ 0 needs analysis. Step 2. Semiclassic al high-angular-momentum el lipticity on c omp act r -sets. Set h = ( ℓ + 1 2 ) − 1 . In the o dd channel, P h, 0 ( ω ) := − h 2 ∂ 2 r ∗ + h 2 ℓ ( ℓ + 1) f ( r ) r 2 + h 2  f ( r ) µ 2 − ω 2  . (8.2) By Prop osition 2.1 and Lemma 4.1, each even channel can b e written on supp e χ as P h,P ( ω ) := − h 2 ∂ 2 r ∗ + V 0 ,P ( r ) + hV 1 ,P ( r , ω ; h ) + h 2 V 2 ,P ( r , ω ; h ) , P = ± 1 , (8.3) where V 0 , − 1 ( r ) = f ( r ) ℓ ( ℓ − 1) ( ℓ + 1 2 ) 2 r 2 , V 0 , +1 ( r ) = f ( r ) ( ℓ + 1)( ℓ + 2) ( ℓ + 1 2 ) 2 r 2 , and V j,P together with ∂ ω V j,P are uniformly b ounded on supp e χ for j = 1 , 2 . Because f ( r ) /r 2 is strictly p ositiv e on the compact set supp e χ , there exists c K > 0 suc h that f ( r ) r 2 ≥ c K on supp e χ. F or ℓ ≥ ℓ 0 one then has, uniformly in | ω | ≤ µ , ξ 2 + 1 2 c K ≤ ℜ σ ( P h,P ( ω ))( r, ξ ) for ( r, ξ ) ∈ T ∗ supp e χ, (8.4) for every channel P . Thus the cut-oﬀ channel op erators are semiclassically elliptic on supp e χ . Standard matrix-v alued semiclassical elliptic regularit y therefore gives a cut-oﬀ parametrix Q h,P ( ω ) ∈ Ψ − 2 h suc h that for every N , χ R ℓ,P ( ω ± i0) χ = Q h,P ( ω ) + O ( h N ) : L 2 → H N h , (8.5) uniformly for | ω | < µ , where R ℓ,P ( ω ± i0) denotes the b oundary-v alue resolv ent in the c hannel P . The absence of real cut p oles and threshold resonances from Theorem 1.6 guaran tees that these b oundary v alues are uniquely deﬁned. Since k ernels of op erators in Ψ − 2 h on a one-dimensional manifold are O ( h − 1 ) on compact sets, and since ∂ ω P h,P ( ω ) = O ( h 2 ) , there exist integers M 0 , M 1 suc h that sup r,r ′ ∈ K    ∂ j ω G ℓ,P ( ω ; r , r ′ )    ≤ C K,j ⟨ ℓ ⟩ M j , j = 0 , 1 , (8.6) 41 uniformly for ℓ ≥ ℓ 0 , | ω | < µ , and ev ery admissible P . Step 3. Uniform c ontr ol of the thr eshold matching c o eﬃcients. Fix a matching radius R > sup K . The map taking Cauc h y data at r = R to v alues on K is gov erned b y (8.6); hence it contributes at most p olynomial factors in ℓ . At r = R , the small- κ outgoing basis is given b y Lemma 5.16, and the large- κ contin uation is gov erned b y Prop osition 5.19. Because R is ﬁxed, the large-order b ounds I ν ( z ) = O  ( z / 2) ν Γ( ν + 1)  , K ν ( z ) = O  Γ( ν )( z / 2) − ν  , v alid for z in compact subsets of (0 , ∞ ) , together with the identit y W  √ z I ν ( z ) , √ z K ν ( z )  = 1 , sho w that after the explicit factor ϖ 2 ν ℓ,P is extracted, the remaining co eﬃcien ts are p olynomially b ounded in ν ℓ,P ∼ ℓ . Likewise, the Whittaker connection co eﬃcients are ratios of gamma functions; Stirling’s form ula in vertical strips sho ws that, after the oscillatory mono dromy factors e ± 2 π i κ are remov ed, their dep endence on ν ℓ,P is also at most p olynomial. Consequen tly , for some in teger N th , sup r,r ′ ∈ K | disc G ℓ,P ( ω ; r , r ′ ) | ≤ C K ⟨ ℓ ⟩ N th  ϖ 2 ν ℓ,P + κ ϖ 2 ν ℓ,P + ϖ 2 ν ℓ,P +2  , κ ≤ 1 , (8.7) sup r,r ′ ∈ K | disc G ℓ,P ( ω ; r , r ′ ) | ≤ C K ⟨ ℓ ⟩ N th  1 + κ − 1 + ϖ  , κ ≥ 1 , (8.8) with the same polynomial b ound for one ω -deriv ative and for the co eﬃcien ts a ℓ,P ( r , r ′ ) , b ± ℓ,P ( r , r ′ ) in Prop ositions 5.17 and 5.19. In particular, sup r,r ′ ∈ K  | A ℓ,P ( r , r ′ ; Q ) | + | B ℓ,P ( r , r ′ ; Q ) |  ≤ C K ⟨ ℓ ⟩ N th . (8.9) Step 4. Uniform oscil latory inversion. The proofs of Theorems 7.3 and 7.6 use only the threshold form ulas, one ω -deriv ative of the amplitudes, and the oscillatory estimates of Lemma 7.2 and Prop osition 7.5. By (8.7), (8.8), and (8.9), ev ery constant in those argu- men ts is p olynomially b ounded in ℓ . Rep eating the same endp oint and saddle calculations therefore giv es sup r,r ′ ∈ K    u bc ℓm,P ( t, r, r ′ )    ≤ C K ⟨ ℓ ⟩ N 0    t − ( ν ℓ,P +1) , κ ∗ ( t ) ≤ 1 , t − 5 / 6 , κ 0 ( t ) ≥ 1 , with the same remainder bounds as in Theorems 7.3 and 7.6. Because the reduced equations are indep endent of m , the constants are uniform in | m | ≤ ℓ . Step 5. Monop ole and r e c onstruction of the physic al c o eﬃcients. The monop ole electric c hannel at ℓ = 0 w as already absorb ed in Step 1. The passage from the channel v ariables ( v − 1 , v 0 , v +1 ) bac k to the ph ysical ev en/o dd co eﬃcien ts uses only the constant matrices T ± 1 ℓ , the near-iden tit y diagonalizer of Lemma 4.1, and at most one radial deriv ative. On compact r -sets these op erators ha ve co eﬃcients b ounded b y C ⟨ ℓ ⟩ , so the same p olynomial estimate holds for the physical branch-cut propagators. Enlarging N 0 if necessary and com bining this with (8.9) prov es (1.12) and (1.13). 42 F or eac h mo de, let E ℓm,P [ A [0]] denote the ﬁxed-mo de energy of the initial data. F or N ∈ N set E N [ A [0]] := X | α |≤ N E [Ω α A ](0) , where Ω denotes the rotation generators on S 2 . Parsev al on the sphere gives X ℓ,m,P ⟨ ℓ ⟩ 2 N E ℓm,P [ A [0]] ≲ E N [ A [0]] . Pr o of of The or em 1.4. Let K 0 ⋐ ( r + , ∞ ) contain the radial supp ort of the initial data. By Theorem 1.3, Cauch y–Sch warz in the source v ariable r ′ , and equiv alence of the ﬁxed- mo de energy with the H 1 × L 2 norm of the reduced initial data on K 0 , eac h mo dal co eﬃcien t satisﬁes sup r ∈ K    a bc ℓm,P ( t, r )    ≤ C K,K 0 ⟨ ℓ ⟩ N 0 E ℓm,P [ A [0]] 1 / 2    t − ( ν ℓ,P +1) , κ ∗ ( t ) ≤ 1 , t − 5 / 6 , κ 0 ( t ) ≥ 1 . No w let Ω 1 , Ω 2 , Ω 3 b e the rotation ﬁelds on S 2 . Because the bac kground is spherically symmetric, Ω α comm utes with the Pro ca operator and with the branch-cut spectral pro jector. F or ev ery ﬁxed r ∈ K , Sob olev on S 2 giv es sup ω ∈ S 2    A bc ( t, r, ω )    ≲ 2 X j =0    ∇ j S 2 A bc ( t, r, · )    L 2 ( S 2 ) . In the vector spherical harmonic basis, ∇ j S 2 con tributes the factor ⟨ ℓ ⟩ j , so P arsev al and the previous mo dal estimate yield    ∇ j S 2 A bc ( t, r, · )    L 2 ( S 2 ) ≤ C K  X ℓ,m,P ⟨ ℓ ⟩ 2( j + N 0 ) E ℓm,P [ A [0]]  1 / 2 ×    t − ( ν ∗ +1) , κ ∗ ( t ) ≤ 1 , t − 5 / 6 , κ 0 ( t ) ≥ 1 . The w eighted modal energy sum is b ounded b y C N E N [ A [0]] whenev er N > N 0 + 2 . Summing o v er j = 0 , 1 , 2 prov es (1.14) and (1.15). The explicit Sc hw arzschild and small- mass Reissner–Nordström v alues of ν ∗ follo w from Theorem 7.1 and Corollary 7.4. The monop ole ℓ = 0 do es not aﬀect the leading intermediate rate, since its only surviving c hannel is electric and decays faster. Pr o of of Cor ol lary 1.5. Theorem 1.4 giv es t wo estimates, one in the in termediate regime and one in the v ery-late regime. By deﬁnition, γ ∗ := min  ν ∗ + 1 , 5 6  . If κ ∗ ( t ) ≤ 1 , then (1.14) implies sup r ∈ K , ω ∈ S 2    A bc ( t, r, ω )    ≤ C K,N E N [ A [0]] 1 / 2 t − ( ν ∗ +1) ≤ C K,N E N [ A [0]] 1 / 2 t − γ ∗ . If κ 0 ( t ) ≥ 1 , then (1.15) implies sup r ∈ K , ω ∈ S 2    A bc ( t, r, ω )    ≤ C K,N E N [ A [0]] 1 / 2 t − 5 / 6 ≤ C K,N E N [ A [0]] 1 / 2 t − γ ∗ . This pro v es (1.16). In Sc hw arzschild one has ν ∗ = 1 / 2 , hence γ ∗ = 5 / 6 . In the suﬃcien tly small-mass Reissner–Nordström regime, Theorem 1.4 gives ν ∗ = 1 / 2 + O (( M µ ) 2 + ( Qµ ) 2 ) , so for small enough ( M µ ) 2 + ( Qµ ) 2 one again has ν ∗ > − 1 / 6 and therefore γ ∗ = 5 / 6 . 43 9 Quasib ound branc hes and residue b ounds Up to this point w e ha ve analyzed the con tin uous spectral con tribution. T o reco v er the full ﬁeld, one must also understand the long-liv ed p oles created by stable timelike trapping. This section dev elops the semiclassical structure of that quasib ound family and extracts the residue b ounds that later feed into the dy adic pac k et summation. 9.1 Guide to the pro of of the quasib ound theorems The pro of follows the familiar trapp ed-w ell strategy , but with one extra step forced b y the v ector c haracter of Proca. Near the trapped set we ﬁrst microlo cally diagonalize the even 2 × 2 system, so that all three p olarizations reduce to scalar semiclassical w ell problems. Once this is done, the argumen t lo oks classical: we iden tify the w ell geometry and the turning p oints, construct WKB bases, match them across the turning p oints, factor the Ev ans determinant, and then use Rouché’s theorem together with the analytic implicit-function theorem to lo cate the p oles and compute their widths. The same pac kage also yields the residue b ounds and Agmon lo calization estimates needed later. W e k eep the full c hain in the pap er because it is used t wice: ﬁrst to con- struct the quasibound resonances themselv es, and then again when v erifying the pac ket estimates in Section 10. 9.2 The trapp ed w ell and semiclassical c hannel sym b ols W e now pass from the ﬁxed-mo de threshold analysis to the semiclassical description of the trapp ed well. The p olarization splitting from Section 2 remains the key simpliﬁcation: on a compact neigh b orho o d of the trapp ed region, each c hannel can b e treated as a scalar semiclassical w ell problem up to lo w er-order corrections. Set h = ( ℓ + 1 2 ) − 1 . On a ﬁxed compact neighborho o d of the trapp ed region, Proposition 2.1, Proposi- tion 9.4, and the diagonalization argumen t already used in Section 8 giv e channel op era- tors of the form P h,P ( ω ) = − h 2 ∂ 2 r ∗ + V P ( r ; h ) − ω 2 , P ∈ {− 1 , 0 , +1 } , where V P ( r ; h ) = f ( r )  µ 2 + L P ( L P + 1) r 2  + hW 1 ,P ( r ) + h 2 W 2 ,P ( r ) , with L − 1 = ℓ − 1 , L 0 = ℓ , L +1 = ℓ + 1 , and W j,P together with ﬁnitely many deriv a- tiv es uniformly bounded on compact sets. A t h = 0 this is precisely the scalar massive Reissner–Nordström p oten tial with angular parameter L P . The stable timelike trapping analysis from the scalar problem therefore applies branc h wise, but we record the relev ant steps here in a theorem c hain b ecause they are used t wice: once to construct the p oles and once again to v erify the pac k et h yp otheses in Section 10. Lemma 9.1 (T rapping window and stable well geometry) . Ther e exist h 0 > 0 , a c omp act interval I trap ⋐ (0 , µ ) , and smo oth turning p oints r 1 ,P ( E ; h ) < r 2 ,P ( E ; h ) < r 3 ,P ( E ; h ) , E ∈ I trap , 0 < h < h 0 , with the fol lowing pr op erties for every p olarization P ∈ {− 1 , 0 , +1 } : 44 (i) V P ( r j,P ( E ; h ); h ) = E 2 and e ach r o ot is simple; (ii) P h,P ( E ) is classic al ly al lowe d on ( r + , r 1 ,P ( E ; h )) ∪ ( r 2 ,P ( E ; h ) , r 3 ,P ( E ; h )) and for- bidden on ( r 1 ,P ( E ; h ) , r 2 ,P ( E ; h )) ∪ ( r 3 ,P ( E ; h ) , ∞ ) ; (iii) V P ( · ; h ) has one nonde gener ate lo c al maximum b etwe en r + and r 1 ,P ( E ; h ) and one nonde gener ate lo c al minimum b etwe en r 2 ,P ( E ; h ) and r 3 ,P ( E ; h ) ; (iv) the interval I trap may b e chosen uniformly in the p olarization. Pr o of. F or h = 0 this is the standard stable timelike trapping picture for the scalar p oten- tial f ( r )( µ 2 + λ/r 2 ) on sub extremal Reissner–Nordström with large angular parameter λ ; see the geometric analysis underlying [2]. The channel p otentials V P ( · ; h ) are C ∞ - small perturbations of that principal scalar p oten tial on compact sets, uniformly in the p olarization, b ecause the ev en-sector coupling is one order lo wer and the diagonalizer is near identit y . The simple ro ots and the nondegenerate critical p oints therefore p ersist b y the implicit-function theorem for all suﬃcien tly small h , and compactness of the ﬁnite p olarization set allows one common in terv al I trap . Lemma 9.2 (T urning p oints and Airy co ordinates) . F or e ach P ∈ {− 1 , 0 , +1 } and e ach E ∈ I trap ther e exist neighb orho o ds U j,P ( E ; h ) of the turning p oints r j,P ( E ; h ) and smo oth A iry c o or dinates ζ j,P ( r , E ; h ) such that P h,P ( E ) = h 2 / 3  ∂ 2 ζ j,P − ζ j,P  + h 4 / 3 B j,P ( r , E ; h ) on U j,P ( E ; h ) , with B j,P uniformly b ounde d to gether with ﬁnitely many derivatives. In p articular, the A iry matching c onstants ar e uniform in ( E , h, P ) . Pr o of. Since each turning p oint is simple, the Langer c hange of v ariables deﬁned by 2 3 ζ j,P ( r , E ; h ) 3 / 2 = ± Z r r j,P ( E ; h ) q | V P ( s ; h ) − E 2 | d s ∗ reduces the scalarized equation to Airy normal form. Uniformity follows b ecause I trap is compact and the deriv atives of the channel p otentials are uniformly b ounded in the p olarization. F or E ∈ I trap deﬁne the well action, the inner tunnelling action, and the outer Agmon distance b y S P ( E ; h ) = 2 Z r 3 ,P ( E ; h ) r 2 ,P ( E ; h ) q E 2 − V P ( r ; h ) d r ∗ , (9.1) J P ( E ; h ) = Z r 2 ,P ( E ; h ) r 1 ,P ( E ; h ) q V P ( r ; h ) − E 2 d r ∗ , (9.2) and, for r ≥ r 3 ,P ( E ; h ) , d P ( r ; E , h ) = Z r r 3 ,P ( E ; h ) q V P ( s ; h ) − E 2 d s ∗ . (9.3) By smooth dep endence on ( E , h ) and the nondegeneracy of the w ell, ∂ E S P ( E ; h ) is b ounded abov e and b elo w b y p ositiv e constan ts on I trap . 45 Prop osition 9.3 (Uniform WKB bases and transfer matrices) . F or e ach p olarization P ∈ {− 1 , 0 , +1 } and e ach ener gy E ∈ I trap , ther e exist exact WKB b ases on the clas- sic al ly al lowe d and forbidden subintervals determine d by L emma 9.1, with phases given by the actions (9.1) – (9.3) . After A iry matching thr ough the thr e e turning p oints fr om L emma 9.2, the tr ansfer matrix fr om the exact ingoing horizon b asis to the exact de c ay- ing inﬁnity b asis admits the factorization T P ( E ; h ) = T hor P ( E ; h ) M 1 ,P ( E ; h ) M 2 ,P ( E ; h ) M 3 ,P ( E ; h ) T ∞ P ( E ; h ) , wher e e ach M j,P ( E ; h ) e quals the universal A iry c onne ction matrix plus an O ( h ) err or, uniformly in ( E , h, P ) . Pr o of. A wa y from the turning p oin ts, the c hannel equations are scalar or diagonally scalarized scalar equations with smooth coeﬃcients, so the Liouville–Green construc- tion pro duces exact oscillatory or exp onential WKB bases with sym b ol expansions in h . Lemma 9.2 supplies uniform Airy co ordinates at eac h turning p oin t. Matc hing the WKB bases to the Airy solutions then yields the factorization ab ov e. The error terms are uni- form b ecause the turning p oin ts remain simple and the Airy co ordinates are uniformly con trolled on the compact trapp ed in terv al. Prop osition 9.4 (Microlo cal diagonalization in the trapp ed region) . On a ﬁxe d neigh- b orho o d of the tr app e d wel l and for ener gies E ∈ I trap , the even 2 × 2 channel system admits an analytic semiclassic al diagonalizer U h such that U − 1 h P even h ( ω ) U h = diag( P h, − 1 ( ω ) , P h, +1 ( ω )) + O ( h ∞ ) micr olo c al ly ne ar the tr app e d set. The o dd channel is alr e ady sc alar. Pr o of. The leading in v erse-square matrix is exactly diagonal in the p olarization basis and its t w o eigen v alues diﬀer b y 2 ℓ + 1 . Hence the principal ev en eigen v alues are separated b y a positive m ultiple of h − 1 in the trapp ed region. Standard iterativ e semiclassical diagonalization for matrix Schrödinger op erators then remo ves the oﬀ-diagonal terms to arbitrary order in h . The o dd c hannel needs no further reduction. 9.3 Bohr–Sommerfeld quan tization and quasib ound branc hes Let E ℓ,P ( ω ) denote the scalar Ev ans determinan t in the o dd channel and the diagonalized Ev ans determinant in the even c hannels. By Prop osition 5.7, p oles of the meromorphically con tin ued resolv ent coincide with zeros of E ℓ,P ( ω ) . Prop osition 9.5 (Ev ans determinant factorization in the well) . Fix I ⋐ I trap . F or e ach p olarization P ∈ {− 1 , 0 , +1 } ther e exist smo oth functions C P , ϑ P , and G P on I × (0 , h 0 ) , with C P and G P b ounde d away fr om zer o and inﬁnity, such that for ω in a c omplex O ( h ) -neighb orho o d of I , E ℓ,P ( ω ) = C P ( ω ; h )   sin  S P ( ω ; h ) 2 h + π 4 + ϑ P ( ω ; h )  + i e − J P ( ω ; h ) /h G P ( ω ; h )   + O  e − 2 J P ( ω ; h ) /h  . (9.4) 46 Pr o of. F or eac h scalarized channel, one constructs exact ingoing solutions near the hori- zon and exact deca ying solutions near inﬁnity by the F rob enius and V olterra metho ds from Sections 5 and 4. In the three regions determined b y Lemma 9.1, one then builds WKB bases with phase functions given by (9.1) and (9.2). Airy matc hing across the three simple turning p oints transfers the outer deca ying solution in to a linear combination of the tw o oscillatory w ell modes, and the transfer across the inner barrier contributes the tunnelling factor e − J P /h . The well oscillation con tributes the phase S P / (2 h ) + π / 4 + ϑ P . T aking the determinant with the exact ingoing horizon solution yields (9.4). Every co ef- ﬁcien t is smo oth in ( ω , h ) b ecause all turning p oin ts are simple and dep end smo othly on the parameters. Theorem 9.6 (Bohr–Sommerfeld quantization in one channel) . Fix I ⋐ I trap and one p olarization P ∈ {− 1 , 0 , +1 } . F or every suﬃciently smal l h and every inte ger n with 2 π h  n + 1 2  ∈ S P ( I ; h ) + O ( h ) , ther e exists a unique c omplex p ole ω ℓ,n,P with ℜ ω ℓ,n,P ∈ I satisfying S P ( ℜ ω ℓ,n,P ; h ) = 2 π h  n + 1 2  + hϑ P ( ℜ ω ℓ,n,P ; h ) + O ( h 2 ) and ℑ ω ℓ,n,P = − Γ P ( ℜ ω ℓ,n,P ; h ) exp  − 2 J P ( ℜ ω ℓ,n,P ; h ) h  (1 + O ( h )) . Mor e over the zer o is simple. Pr o of. Prop osition 9.3 iden tiﬁes the exact transfer matrix across the well, and Prop osi- tion 9.5 conv erts that transfer matrix in to the factorization of the Ev ans determinan t. Since ∂ E S P ( E ; h ) is b ounded ab ov e and b elo w b y p ositiv e constan ts on I , the ro ots of the principal Bohr–Sommerfeld equation are simple and v ary smo othly in ( n, h ) . The exp onen tially small tunnelling term in (9.4) is then handled by the analytic implicit- function theorem or Rouc hé’s theorem, yielding the unique nearb y complex zero together with the tunnelling-width form ula. Simplicit y follows from diﬀeren tiating the factorized Ev ans determinan t at the zero. Corollary 9.7 (Branc h enumeration and W eyl la w) . Fix I ⋐ I trap . F or e ach p olarization P ∈ {− 1 , 0 , +1 } and every suﬃciently smal l h , the p oles of the r esolvent with ℜ ω ∈ I and 0 > ℑ ω > − exp( − c I /h ) form a family { ω ℓ,n,P : n ∈ N ℓ,P ( I ) } , wher e # N ℓ,P ( I ) = 1 2 π h  S P ( I ; h )  + O (1) , and every such p ole lies on exactly one smo oth br anch n 7→ ω ℓ,n,P . Pr o of. Theorem 9.6 giv es uniqueness of the p ole attached to eac h admissible integer n . Summing ov er the admissible v alues of n giv es the W eyl count, and uniqueness sho ws that the p oles organize in to smo oth branc hes. 47 Prop osition 9.8 (F requency deriv ativ e and spacing b ounds) . F or every c omp act I ⋐ I trap ther e exists C I > 0 such that, for al l admissible ( ℓ, n, P ) with ℜ ω ℓ,n,P ∈ I , C − 1 I h ≤ ∂ n ℜ ω ℓ,n,P ≤ C I h, and, after smo oth interp olation in n , at le ast one derivative among ∂ 2 n ℜ ω ℓ,n,P and ∂ 3 n ℜ ω ℓ,n,P is b ounde d fr om b elow by C − 1 I h 2 on e ach dyadic p acket. In addition,    ∂ a h ∂ b n ω ℓ,n,P    ≤ C I ,ab for al l a + b ≤ 2 . Pr o of. Diﬀeren tiate the quantization la w from Theorem 9.6. Since ∂ E S P ( E ; h ) is b ounded ab ov e and b elow on I , one obtains the monotone spacing estimate ∂ n ℜ ω ℓ,n,P ≍ h . A second diﬀeren tiation sho ws that the curv ature is controlled by the ﬁrst nonv anishing deriv ativ e of the action S P , which is nonzero on dyadic pac kets b ecause the well is nondegenerate. The bounds inv olving h follo w from diﬀeren tiating the same implicit relation and using smo othness of S P , ϑ P , and Γ P . Pr o of of The or em 1.7. Apply Theorem 9.6 separately to the three p olarizations P ∈ {− 1 , 0 , +1 } and then inv oke Corollary 9.7 to index the p oles in the strip ℜ ω ∈ I , 0 > ℑ ω > − exp( − c I /h ) , b y integers n ∈ N ℓ,P ( I ) . The Bohr–Sommerfeld law and the tunnelling-width formula are exactly the conclusions of Theorem 9.6, while simplicity of the p oles is part of the same theorem. Since h = ( ℓ + 1 2 ) − 1 , this gives the stated form of the theorem. 9.4 Residues and reconstruction of the ph ysical co eﬃcients Prop osition 9.9 (Deriv ative of the Ev ans determinan t and normalized modes) . F or every c omp act tr app e d interval I ⋐ I trap ther e exist c onstants c I , C I > 0 such that, for every admissible quasib ound p ole, c I h − 1 ≤ | ∂ ω E ℓ,P ( ω ℓ,n,P ) | ≤ C I h − 1 . Mor e over the c orr esp onding ingoing and de c aying mo de r epr esentatives may b e normalize d so that on e ach c omp act r adial set K ⋐ ( r + , ∞ ) , sup r ∈ K    u hor ℓ,n,P ( r )    + sup r ∈ K    u ∞ ℓ,n,P ( r )    ≤ C K h − 1 / 2 , with the A gmon impr ovement    u hor ℓ,n,P ( r )    +    u ∞ ℓ,n,P ( r )    ≤ C K h − 1 / 2 exp  − d P ( r ; ℜ ω ℓ,n,P , h ) h  whenever K is disjoint fr om the classic al ly al lowe d wel l. Pr o of. The lo w er and upp er b ounds for ∂ ω E ℓ,P ( ω ℓ,n,P ) follo w from diﬀerentiating the fac- torization of Prop osition 9.5 at a simple zero, together with the p ositivity of ∂ E S P on the trapp ed interv al. The mode b ounds come from the WKB normalization in the classically allo w ed well and from standard Agmon estimates in the forbidden region, using the exact WKB bases from Prop osition 9.3. The compactness of I ensures that the constan ts are uniform. 48 Pr o of of The or em 1.8. Because ev ery quasib ound p ole is simple, the corresp onding residue projector has the separated form Π ℓ,n,P ( r , r ′ ) = u hor ℓ,n,P ( r ) u ∞ ℓ,n,P ( r ′ ) ∂ ω E ℓ,P ( ω ℓ,n,P ) , (9.5) where u hor ℓ,n,P is the exact ingoing mo de and u ∞ ℓ,n,P the exact decaying mo de, b oth nor- malized compatibly with the Ev ans determinan t. Prop osition 9.9 gives the t wo-sided b ound | ∂ ω E ℓ,P ( ω ℓ,n,P ) | ≍ h − 1 (9.6) uniformly on compact trapp ed in terv als. The same prop osition giv es the WKB and Agmon b ounds for the n umerator mo des. Substituting those estimates in to (9.5) yields sup r,r ′ ∈ K | Π ℓ,n,P ( r , r ′ ) | ≤ C K h − N qb ≤ C K ⟨ ℓ ⟩ N qb for some in teger N qb , together with the Agmon improv ement whenever K is disjoin t from the allo wed well. T o reconstruct the ph ysical Pro ca co eﬃcien ts, one uses the same constant p olarization matrices and at most one radial deriv ativ e as in Prop osition 8.4. On compact r -sets these op erators cost at most one further p olynomial p ow er of ℓ . Finally , Cauch y–Sch w arz in the source v ariable together with the high-order angular energy b ound giv es sup r ∈ K , ω ∈ S 2 | Π ℓ,n,P A [0]( r , ω ) | ≤ C K,N ⟨ ℓ ⟩ N rec E N [ A [0]] 1 / 2 . This pro ves the theorem. 10 Summation of quasib ound residues and unsplit full-ﬁeld deca y A t this stage both spectral pieces are on the table. The branch-cut contribution w as handled in Section 8; what remains is to sum the long-liv ed quasibound residues and add them bac k in. The outcome is a deca y theorem for the unsplit Pro ca ﬁeld together with the full-ﬁeld asymptotic expansions stated in the introduction. 10.1 Guide to the pro of of the summed deca y theorem The summation argument is inten tionally elementary once the sp ectral input is in place. W e deform the in verse Laplace con tour into three pieces: the massive branch cut, the quasib ound p oles, and a remainder that stays a ﬁxed distance below the real axis. The branc h-cut part is already under control. F or the p oles, w e group large angular momen ta in to dy adic pac k ets indexed b y the semiclassical parameter h = ( ℓ + 1 2 ) − 1 . On each pac ket w e use only three ingredients: the W eyl coun t from Corollary 9.7, the tunnelling-width form ula from Theorem 9.6, and the residue/reconstruction b ounds from Theorem 1.8. T ogether they give a pack et estimate with explicit exp onen tial damping and a con trollable algebraic loss in the pack et scale. A short dy adic summation lemma then yields logarithmic decay for the full quasib ound con tribution. This is exactly the p oin t at which the present v ersion b ecomes self-con tained. 49 10.2 Con tour decomp osition and the fast remainder W e ﬁrst record the clean sp ectral splitting that underlies the ﬁnal argument: branc h cut, quasib ound residues, and a remainder that deca ys exponentially b ecause it sta ys uniformly b elo w the real axis. Prop osition 10.1 (Contin uous/discrete decomposition with fast remainder) . F or c om- p actly supp orte d initial data and every c omp act r adial set K ⋐ ( r + , ∞ ) one has on K A ( t ) = A bc ( t ) + A qb ( t ) + A fast ( t ) , (10.1) wher e A bc is the br anch-cut c ontribution studie d in Se ctions 7 and 8, A qb is the sum of r esidues at the quasib ound p oles fr om The or em 1.7, and the r emainder satisﬁes sup r ∈ K , ω ∈ S 2    A fast ( t, r, ω )    ≤ C K,N e − η t E N [ A [0]] 1 / 2 for some η > 0 and every suﬃciently lar ge N . Pr o of. By Theorem 1.1, the cut-oﬀ resolv ent extends meromorphically to a slit strip {|ℑ ω | < η } \ [ − µ, µ ] . Start from the in v erse Laplace represen tation of the solution with con tour in {ℑ ω = σ } , σ > 0 , and deform the con tour to {ℑ ω = − η } while keeping small detours around the branch cut [ − µ, µ ] . The detours around the cut pro duce A bc . Residues of the p oles with real parts in the trapp ed in terv al and imaginary parts tending to zero pro duce A qb . Ev ery other p ole lies a deﬁnite distance b elow the real axis, and the remaining deformed contour also lies at imaginary part − η ; b oth contributions therefore satisfy the stated exp onential b ound. 10.3 Radiativ e proﬁles and pro of of the asymptotic expansion theorem W e now turn the modal asymptotics in to full-ﬁeld proﬁles. The main p oin t is that the large- ℓ bounds are strong enough to sum b oth the leading terms and the remainders in C 0 ( K × S 2 ) , while the ﬁnitely many lo w modes can b e absorbed in to ﬁnite-rank proﬁle ﬁelds. Pr o of of The or em 1.11. Choose ℓ as ≥ 1 large enough that the large- ℓ analysis of Section 8 applies for ev ery ℓ ≥ ℓ as . F or the ﬁnitely man y mo des 0 ≤ ℓ < ℓ as , including the exceptional electric monop ole, the ﬁxed-mo de asymptotic analysis from Section 7 together with the monop ole discussion in Appendix D pro vide explicit leading terms of the same form, and the lo w-mo de part of the pro of of Theorem 1.3 supplies the uniform constants needed to absorb them into ﬁnite-rank ﬁelds. Summing those ﬁnitely many contributions deﬁnes the ﬁnite-rank ﬁelds A bc int , lo w and A bc late , low . No w ﬁx one high mode ( ℓ, m, P ) with ℓ ≥ ℓ as . Let K 0 ⋐ ( r + , ∞ ) contain the radial supp ort of the initial data. Pair the k ernel asymptotics of Theorems 7.3 and 7.6 with the reduced initial data on K 0 exactly as in the pro of of Theorem 1.4. This pro duces co eﬃcien t functions A ℓm,P [ A [0]] and B ℓm,P [ A [0]] on K such that the mo dal branch-cut co eﬃcien t ob eys a bc ℓm,P ( t, r ) = A ℓm,P [ A [0]]( r ) t − ( ν ℓ,P +1) sin( µt + δ ℓ,P ( Q )) + r int ℓm,P ( t, r ) , (10.2) a bc ℓm,P ( t, r ) = B ℓm,P [ A [0]]( r ) t − 5 / 6 sin  µt − 3 2 (2 π M µ ) 2 / 3 ( µt ) 1 / 3 + δ ℓ,P, 0 ( Q )  + r late ℓm,P ( t, r ) , (10.3) 50 where, uniformly for r ∈ K , |A ℓm,P [ A [0]]( r ) | + |B ℓm,P [ A [0]]( r ) | ≤ C K ⟨ ℓ ⟩ N 0 E ℓm,P [ A [0]] 1 / 2 , (10.4)    r int ℓm,P ( t, r )    ≤ C K ⟨ ℓ ⟩ N 0 E ℓm,P [ A [0]] 1 / 2  κ ∗ ( t ) t − ( ν ℓ,P +1) + t − ( ν ℓ,P +2)  , (10.5)    r late ℓm,P ( t, r )    ≤ C K ⟨ ℓ ⟩ N 0 E ℓm,P [ A [0]] 1 / 2  κ 0 ( t ) − 1 + ϖ 0 ( t )  t − 5 / 6 . (10.6) Here (10.4) follows from (1.13), Cauch y–Sc hw arz in the source v ariable, and equiv alence of the ﬁxed-mo de energy with the lo cal H 1 × L 2 norm of the reduced data. In the v ery- late regime, Theorem 7.6 contains a phase error O ( κ 0 ( t ) − 1 + ϖ 0 ( t )) inside the sine; using sin( x + η ) = sin x + O ( η ) and the same co eﬃcient b ound absorbs that phase error into (10.6). Deﬁne the high-mo de pieces of A bc int and A bc late b y the series in (1.18) and (1.19). By Sob olev on S 2 , the vector-harmonic addition theorem, and Parsev al exactly as in the pro of of Theorem 1.4, (10.4) implies sup r ∈ K , ϑ ∈ S 2    A bc int ( t, r, ϑ )    ≤ C K,N E N [ A [0]] 1 / 2 t − ( ν ∗ +1) , and sup r ∈ K , ϑ ∈ S 2    A bc late ( t, r, ϑ )    ≤ C K,N E N [ A [0]] 1 / 2 t − 5 / 6 . for ev ery N > N 0 + 2 . In particular, the deﬁning series con v erge in C 0 ( K × S 2 ) and (1.20)–(1.21) follo w after absorbing the ﬁnitely many low mo des in to the constan t. Next sum the remainders. Applying the same Sob olev–Parsev al argument to (10.5) yields sup r ∈ K , ϑ ∈ S 2    A bc ( t, r, ϑ ) − A bc int ( t, r, ϑ )    ≤ C K,N E N [ A [0]] 1 / 2  κ ∗ ( t ) t − ( ν ∗ +1) + t − ( ν ∗ +2)  whenev er κ ∗ ( t ) ≤ 1 , because ν ℓ,P ≥ ν ∗ for ev ery channel. This is exactly (1.23). Likewise, (10.6) giv es sup r ∈ K , ϑ ∈ S 2    A bc ( t, r, ϑ ) − A bc late ( t, r, ϑ )    ≤ C K,N E N [ A [0]] 1 / 2  κ 0 ( t ) − 1 + ϖ 0 ( t )  t − 5 / 6 whenev er κ 0 ( t ) ≥ 1 , whic h is (1.25). Finally , Prop osition 10.1 giv es A ( t ) = A bc ( t ) + A qb ( t ) + A fast ( t ) . Substituting the t w o branc h-cut expansions just pro ved yields (1.22) and (1.24). This completes the pro of. 10.4 Explicit leading co eﬃcien t ﬁelds W e no w isolate the coeﬃcient ﬁelds that actually app ear in the radiativ e branc h-cut asymptotics. No new sp ectral input is needed here; the p oint is simply to reorganize 51 Theorem 1.11 so that the dominant oscillatory term is visible and the faster c hannels are clearly separated. F or later reference we ﬁx the following notation. In the pro of of Theorem 1.11 the ﬁnitely man y low mo des were absorb ed into the ﬁelds A bc int , lo w and A bc late , low . Cho ose once and for all co eﬃcient functions e A ℓm,P [ A [0]]( r ) , e B ℓm,P [ A [0]]( r ) , for ev ery admissible mo de ( ℓ, m, P ) , so that for ℓ ≥ ℓ as they agree with the co eﬃcient functions from Theorem 1.11, while for the ﬁnitely man y lo w er mo des they are the ﬁxed- mo de coeﬃcients used to deﬁne A bc int , lo w and A bc late , low . Enlarging the constan t from (1.17) if necessary , one has sup r ∈ K     e A ℓm,P [ A [0]]( r )    +    e B ℓm,P [ A [0]]( r )     ≤ C K,N ⟨ ℓ ⟩ N 0 E ℓm,P [ A [0]] 1 / 2 (10.7) for ev ery admissible mo de, with the understanding that for ℓ = 0 only the electric channel P = +1 occurs. Lemma 10.2 (Dominant threshold set and sp ectral gap) . Ther e exists C ν > 0 such that ν 2 ℓ,P ≥  L P + 1 2  2 − C ν for every ℓ ≥ 1 and every P ∈ {− 1 , 0 , +1 } . In p articular, ν ℓ,P → ∞ as ℓ → ∞ , uniformly in P . Conse quently the set Σ ∗ := { ( ℓ, P ) : ℓ ≥ 1 , P ∈ {− 1 , 0 , +1 } , ν ℓ,P = ν ∗ } is ﬁnite and nonempty, and ther e exists ρ ∗ > 0 such that ν ℓ,P ≥ ν ∗ + ρ ∗ for every ( ℓ, P ) / ∈ Σ ∗ . Pr o of. Return to the pro of of Lemma 4.1. After the Liouville conjugation one obtains W ′′ +  − ϖ 2 + 2 M µ 2 r − D ℓ + 1 4 Id + E ℓ r 2 + O ( r − 3 )  W = 0 , where D ℓ is diagonal with en tries L P ( L P + 1) and E ℓ is uniformly bounded in ℓ . After the near-iden tit y diagonalization used there, the exact channel co eﬃcient of r − 2 therefore has the form ν 2 ℓ,P − 1 4 = L P ( L P + 1) + e ℓ,P , | e ℓ,P | ≤ C ν , for some constant C ν indep enden t of ℓ and P . This gives the ﬁrst inequalit y . Since L P ∈ { ℓ − 1 , ℓ, ℓ + 1 } , one has L P + 1 2 ≥ ℓ − 1 2 , hence ν 2 ℓ,P ≥  ℓ − 1 2  2 − C ν . 52 Therefore ν ℓ,P → ∞ uniformly in P as ℓ → ∞ . Cho ose ℓ ∗ so large that ν ℓ,P ≥ ν ∗ + 1 for every ℓ ≥ ℓ ∗ and every p olarization. Then ν ∗ is attained among the ﬁnitely man y c hannels with 1 ≤ ℓ < ℓ ∗ , so Σ ∗ is ﬁnite and nonempty . On the ﬁnite complemen t { ( ℓ, P ) : 1 ≤ ℓ < ℓ ∗ , ( ℓ, P ) / ∈ Σ ∗ } the p ositive minim um of ν ℓ,P − ν ∗ deﬁnes a n umber ρ 0 > 0 . Setting ρ ∗ := min { 1 , ρ 0 } yields the stated gap for all channels. Pr o of of Cor ol lary 1.12. Set P 0 := { +1 } and P ℓ := {− 1 , 0 , +1 } for ℓ ≥ 1 . Fix the full co eﬃcien t families e A ℓm,P and e B ℓm,P ab o v e. The b ound (10.7), together with Sobolev on S 2 and P arsev al exactly as in the pro of of Theorem 1.11, shows that the series deﬁning S late and C late con v erge in C 0 ( K × S 2 ) and satisfy sup r ∈ K , ϑ ∈ S 2  |S late ( r , ϑ ) | + |C late ( r , ϑ ) |  ≤ C K,N E N [ A [0]] 1 / 2 for every N > N 0 + 2 . Since Σ ∗ is ﬁnite, the same b ound for S ∗ and C ∗ is immediate, and (1.30) follo ws after enlarging the constan t. F or the intermediate regime, split the proﬁle from Theorem 1.11 into the dominan t part and the strictly faster remainder: A bc int = A bc ∗ , lead + A bc ∗ , rem , where A bc ∗ , lead ( t, r, ϑ ) := t − ( ν ∗ +1) X ( ℓ,P ) ∈ Σ ∗ ℓ X m = − ℓ e A ℓm,P [ A [0]]( r ) Y ( P ) ℓm ( ϑ ) sin( µt + δ ℓ,P ( Q )) . Using sin( µt + δ ) = sin( µt ) cos δ + cos( µt ) sin δ , this b ecomes exactly A bc ∗ , lead ( t, r, ϑ ) = t − ( ν ∗ +1)  S ∗ ( r , ϑ ) sin( µt ) + C ∗ ( r , ϑ ) cos( µt )  . F or every c hannel with ℓ ≥ 1 o ccurring in A bc ∗ , rem , Lemma 10.2 gives ν ℓ,P ≥ ν ∗ + ρ ∗ . The only contribution not co v ered b y that lemma is the electric monop ole. By the monop ole analysis in Appendix D—compare also the ﬁnal sen tence of the pro of of Theorem 1.4— its intermediate decay exp onen t is strictly larger than ν ∗ + 1 . Since the low-mode sector is ﬁnite, after decreasing ρ ∗ > 0 if necessary we ma y therefore assume that ev ery term con tained in A bc ∗ , rem , including the monop ole if present, deca ys at least like t − ( ν ∗ +1+ ρ ∗ ) . Hence the same Sob olev–Parsev al summation as ab ov e giv es sup r ∈ K , ϑ ∈ S 2    A bc ∗ , rem ( t, r, ϑ )    ≤ C K,N E N [ A [0]] 1 / 2 t − ( ν ∗ +1+ ρ ∗ ) for ev ery N > N 0 + 2 . The intermediate expansion (1.22) and remainder b ound (1.23) no w giv e A ( t ) = A bc ∗ , lead ( t ) + A qb ( t ) + A fast ( t ) + e R int ( t ) , with e R int := A bc ∗ , rem + R int . Since σ ∗ = min { 1 , ρ ∗ } and t − ( ν ∗ +2) ≤ t − ( ν ∗ +1+ σ ∗ ) , the b ounds for A bc ∗ , rem and R int imply (1.33). This prov es (1.31). 53 F or the v ery-late regime, write the full proﬁle from (1.19) as A bc late ( t, r, ϑ ) = t − 5 / 6 X ℓ ≥ 0 ℓ X m = − ℓ X P ∈P ℓ e B ℓm,P [ A [0]]( r ) Y ( P ) ℓm ( ϑ ) sin(Θ( t ) + δ ℓ,P, 0 ( Q )) . Expanding the sine once more yields A bc late ( t, r, ϑ ) = t − 5 / 6  S late ( r , ϑ ) sin Θ( t ) + C late ( r , ϑ ) cos Θ( t )  . Substituting this iden tit y in to (1.24) and using (1.25) giv es (1.32) and (1.34). The pro of is complete. 10.5 Dy adic pac k ets and a self-con tained damping estimate The remaining task is to sum the quasib ound contribution in a wa y that uses only in- formation established in this pap er. The natural unit is a dy adic pac ket in angular momen tum. On such a pack et, the widths are exp onen tially small but quantitativ ely con trolled, and the residue bounds con v ert angular regularity of the data in to algebraic deca y in the pack et scale. F or high angular momen ta, group the quasib ound p oles in to dy adic angular pack ets P j := { ( ℓ, n, m, P ) : 2 − j − 1 < h ≤ 2 − j } , h = ( ℓ + 1 2 ) − 1 , and write the corresp onding pac ket con tribution as A qb P j ( t ) := X ( ℓ,n,m,P ) ∈ P j e − i ω ℓ,n,P t Π ℓ,n,P A [0] . F or a ﬁxed angular mo de, let a qb ℓm,P ( t, r ) := X n ∈N ℓ,P ( I ) e − i ω ℓ,n,P t Π ℓ,n,P A ℓm,P [0]( r ) , where A ℓm,P [0] denotes the ( ℓ, m, P ) -comp onent of the initial data in the reduced v ari- ables. The pack et estimate prov ed b elow uses only the tunnelling-width formula and the mo dewise residue b ounds. Lemma 10.3 (T wo-sided tunnelling widths on a trapp ed pac ket) . Fix a c omp act tr app e d interval I ⋐ I trap . Ther e exist c onstants c wd , C wd > 0 such that every quasib ound p ole with ℜ ω ℓ,n,P ∈ I and 2 − j − 1 < h ≤ 2 − j satisﬁes c wd exp( − C wd 2 j ) ≤ −ℑ ω ℓ,n,P ≤ C wd exp( − c wd 2 j ) . Pr o of. By Theorem 9.6, −ℑ ω ℓ,n,P = Γ P ( ℜ ω ℓ,n,P ; h ) exp  − 2 J P ( ℜ ω ℓ,n,P ; h ) h  (1 + O ( h )) . On the compact trapp ed in terv al I , the functions Γ P and J P are smo oth and strictly p ositiv e, uniformly in the ﬁnite p olarization set P ∈ {− 1 , 0 , +1 } . Hence there exist n um b ers Γ ± , J ± > 0 suc h that 0 < Γ − ≤ Γ P ≤ Γ + , 0 < J − ≤ J P ≤ J + , for all ( E , h, P ) ∈ I × (0 , h 0 ) × {− 1 , 0 , +1 } . Since h ≃ 2 − j on the pack et and 1 + O ( h ) is b ounded abov e and b elo w for h small, the claimed t w o-sided b ound follo ws. 54 Lemma 10.4 (Mo dewise quasibound sum) . F or every c omp act K ⋐ ( r + , ∞ ) ther e exists an inte ger N mod such that sup r ∈ K    a qb ℓm,P ( t, r )    ≤ C K ⟨ ℓ ⟩ N mod exp( − c wd t e − C wd ℓ ) E ℓm,P [ A [0]] 1 / 2 for al l admissible ( ℓ, m, P ) with ℓ ≥ ℓ sc . Pr o of. The pro of of Theorem 1.8 giv es, b efore the ﬁnal angular summation step, sup r ∈ K | Π ℓ,n,P A ℓm,P [0]( r ) | ≤ C K ⟨ ℓ ⟩ N res E ℓm,P [ A [0]] 1 / 2 for some integer N res , uniformly in n and in the ﬁnite p olarization lab el P . By Corol- lary 9.7, the num b er of admissible integers in N ℓ,P ( I ) is O ( h − 1 ) = O ( ℓ ) . Lemma 10.3 giv es    e − i ω ℓ,n,P t    = e t ℑ ω ℓ,n,P ≤ exp( − c wd t e − C wd ℓ ) . Summing o v er n and absorbing the extra factor ℓ in to the p olynomial loss pro ves the claim. Prop osition 10.5 (Self-con tained dy adic pac ket estimate) . F or every c omp act K ⋐ ( r + , ∞ ) and every lo garithmic p ower L > 0 , ther e exists an inte ger N pkt ( L ) such that, for every dyadic p acket P j , sup r ∈ K , ω ∈ S 2    A qb P j ( t, r, ω )    ≤ C K,N ,L 2 − j L exp( − c wd t e − C wd 2 j ) E N [ A [0]] 1 / 2 for al l N ≥ N pkt ( L ) and al l t ≥ 2 . Pr o of. Fix r ∈ K . Sobolev on S 2 giv es sup ω ∈ S 2    A qb P j ( t, r, ω )    ≲ 2 X s =0    ∇ s S 2 A qb P j ( t, r, · )    L 2 ( S 2 ) . On the pac ket 2 − j − 1 < h ≤ 2 − j one has ⟨ ℓ ⟩ ≃ 2 j . Therefore Lemma 10.4 and P arsev al imply    ∇ s S 2 A qb P j ( t, r, · )    L 2 ( S 2 ) ≤ C K 2 j ( s + N mod ) exp( − c wd t e − C wd 2 j )   X 2 − j − 1 0 and c, C > 0 . Then ther e exists C L > 0 such that X j ≥ 0 2 − j L exp( − ct e − C 2 j ) ≤ C L (log(2 + t )) − L , t ≥ 2 . Pr o of. Cho ose J ( t ) ∈ N so that 2 J ( t ) ≤ 1 2 C log(2 + t ) < 2 J ( t )+1 . If j ≤ J ( t ) , then e − C 2 j ≥ (2 + t ) − 1 / 2 , so exp( − ct e − C 2 j ) ≤ exp( − c t (2 + t ) − 1 / 2 ) ≤ exp( − c ′ t 1 / 2 ) for some c ′ > 0 . Thus X j ≤ J ( t ) 2 − j L exp( − ct e − C 2 j ) ≤ C exp( − c ′ t 1 / 2 ) . If j > J ( t ) , then a geometric sum gives X j >J ( t ) 2 − j L exp( − ct e − C 2 j ) ≤ X j >J ( t ) 2 − j L ≤ C L 2 − J ( t ) L ≤ C L (log(2 + t )) − L . Finally , exp( − c ′ t 1 / 2 ) ≤ C L (log(2 + t )) − L for t ≥ 2 , so the t wo pieces combine to pro v e the lemma. Pr o of of The or em 1.9. Split the quasib ound contribution into lo w and high angular mo- men ta. F or the ﬁnitely man y low mo des ℓ < ℓ sc , Prop osition 10.1 sho ws that ev ery discrete p ole is b ounded a wa y from the real axis, hence contributes exp onen tially decay- ing terms. F or the high mo des ℓ ≥ ℓ sc , decomp ose the p ole sum into dyadic pac k ets P j . Prop osition 10.5 and Lemma 10.6 giv e X j ≥ 0 sup r ∈ K , ω ∈ S 2    A qb P j ( t, r, ω )    ≤ C K,N ,L E N [ A [0]] 1 / 2 X j ≥ 0 2 − j L exp( − c wd t e − C wd 2 j ) ≤ C K,N ,L E N [ A [0]] 1 / 2 (log(2 + t )) − L . A dding the ﬁnitely man y exp onentially decaying lo w-mo de contributions pro ves the the- orem. Pr o of of The or em 1.10. By Prop osition 10.1, A ( t ) = A bc ( t ) + A qb ( t ) + A fast ( t ) . The branc h-cut term satisﬁes Corollary 1.5, namely sup r ∈ K , ω ∈ S 2    A bc ( t, r, ω )    ≤ C K,N E N [ A [0]] 1 / 2 t − γ ∗ . The quasibound term satisﬁes Theorem 1.9, sup r ∈ K , ω ∈ S 2    A qb ( t, r, ω )    ≤ C K,N ,L E N [ A [0]] 1 / 2 (log(2 + t )) − L . Finally , A fast is exp onen tially deca ying by Prop osition 10.1, hence it is b ounded b y the same logarithmic rate after enlarging the constant. Summing the three pieces prov es the theorem. 56 11 Outlo ok: extremalit y and rotation T wo natural directions remain just b eyond the reac h of the present framework. The ﬁrst is extremalit y . When | Q | = M , the obstruction is geometric rather than algebraic: the surface gra vit y v anishes, the red-shift estimate degenerates, and the near- horizon region dev elops an AdS 2 × S 2 throat. One then has to analyze a second threshold at ω = 0 in addition to ω = ± µ , construct a degenerate horizon basis, and understand ho w Aretakis-type horizon quantities en ter the late-time asymptotics. The sub extremal argumen t dev elop ed here do es not directly con trol those eﬀects. The second is rotation. The static Reissner–Nordström problem treated here is the c harged zero-rotation precursor to the fully separated Kerr–Newman Pro ca system of [6]. Muc h of the ﬁxed-mo de machinery should survive in that setting—horizon F rob enius the- ory , inﬁnity V olterra theory , Ev ans determinants, and threshold sp ecial-function asymp- totics. The real new diﬃcult y is the sim ultaneous con trol of the angular and radial sp ectral parameters across the coupled family . Extending the presen t threshold-and- resonance picture to Kerr–Newman w ould therefore require new ideas, not just a longer v ersion of the same argument. A Harmonic reduction and b oundary constructions Harmonic con v en tions on the round sphere Let Y ℓm denote the standard scalar spherical harmonics normalized by ∆ S 2 Y ℓm = − ℓ ( ℓ + 1) Y ℓm . W e write Y A := ∇ A Y ℓm , S A := ϵ A B ∇ B Y ℓm , for the even and o dd v ector harmonics, where ϵ AB is the volume form of the round sphere. The basic identities are ∇ A Y A = − ℓ ( ℓ + 1) Y ℓm , (A.1) ∇ A S A = 0 , (A.2) ∆ S 2 Y A = (1 − ℓ ( ℓ + 1)) Y A , (A.3) ∆ S 2 S A = (1 − ℓ ( ℓ + 1)) S A . (A.4) These form ulas are rep eatedly used in the angular reduction. W e also record the L ∞ addition-theorem bound sup ω ∈ S 2 ℓ X m = − ℓ | Y ℓm ( ω ) | 2 = 2 ℓ + 1 4 π , (A.5) and its vector analogue sup ω ∈ S 2 ℓ X m = − ℓ  | Y ℓm ( ω ) | 2 + | Y A ( ω ) | 2 + | S A ( ω ) | 2  ≲ ⟨ ℓ ⟩ 2 . (A.6) Mo de decomp osition of the Pro ca p otential 57 Ev ery suﬃcien tly regular Pro ca p otential may b e written as A = X ℓ,m  a ℓm 0 ( t, r ) Y ℓm d t + a ℓm 1 ( t, r ) Y ℓm d r + a ℓm 2 ( t, r ) Y A d x A + a ℓm 3 ( t, r ) S A d x A  . (A.7) The o dd and even sectors decouple b ecause S A is divergence-free and orthogonal to the ev en sector generated b y Y ℓm and Y A . The o dd sector con tains only the co eﬃcient a ℓm 3 , while the even sector is generated b y the triple ( a ℓm 0 , a ℓm 1 , a ℓm 2 ) . A direct calculation gives the ﬁeld strength comp onents F tr = ∂ t a 1 − ∂ r a 0 , (A.8) F even tA = ( ∂ t a 2 − a 0 ) Y A , (A.9) F even rA = ( ∂ r a 2 − a 1 ) Y A , (A.10) F odd AB = 2 a 3 ∇ [ A S B ] , (A.11) together with the ob vious o dd-electric comp onents inv olving a 3 . Substituting these ex- pressions in to ∇ µ F µν − µ 2 A ν = 0 and using (A.1)–(A.4) yields the reduced equations. Elimination of constrained v ariables The forced Lorenz condition ∇ ν A ν = 0 b ecomes, mode by mo de, − f − 1 ∂ t a 0 + 1 r 2 ∂ r ( r 2 f a 1 ) − ℓ ( ℓ + 1) r 2 a 2 = 0 . (A.12) In the massive theory this is not a gauge condition but an equation implied b y the Pro ca system. It can therefore be used to eliminate one ev en v ariable. A con v enient choice is to k eep a 2 together with the electric com bination z := r 2 ( ∂ t a 1 − ∂ r a 0 ) , whic h is prop ortional to the radial electric ﬁeld. After F ourier transformation in time, the remaining v ariable ma y b e eliminated algebraically , pro ducing a second-order 2 × 2 system. The algebra is length y but completely explicit. Prop osition A.1 (Detailed o dd/even reduction) . F or every ℓ ≥ 1 , the o dd c o eﬃcient a ℓm 3 satisﬁes a single sc alar e quation of R e gge–Whe eler typ e, while the even c o eﬃcients may b e arr ange d into a p air ( u 2 , u 3 ) ob eying (2.7) – (2.8) . The only diﬀer enc e b etwe en Schwarzschild and R eissner–Nor dstr öm is the r eplac ement 1 − 3 M r − → 1 − 3 M r + 2 Q 2 r 2 = f − r f ′ 2 in the c oupling c o eﬃcient. Pr o of. The o dd sector follows by substituting the o dd ansatz A odd = a 3 S A d x A in to the Pro ca equation and using (A.4). The ev en sector requires solving the ( t, r ) -comp onents of the Pro ca equation together with (A.12) for a 0 and a 1 in terms of the electric ﬁeld com bination and a 2 . After substitution in to the angular comp onen t, one arriv es at the pair (2.7)–(2.8). The c harged background aﬀects only the static co eﬃcien t f − rf ′ 2 , whic h is precisely the expression displa yed ab o ve. 58 Asymptotic p olarization basis The constant transformation (2.13) diagonalizes the exact in verse-square part of the ev en system. This is a sp ecial feature of the neutral Pro ca system on spherically symmet- ric backgrounds and is the basic reason why the threshold analysis may still b e organized in terms of three eﬀectiv e scalar c hannels. W riting u = u 2 u 3 ! , v = T − 1 ℓ u , one ﬁnds T − 1 ℓ ℓ ( ℓ + 1) − 2  f − rf ′ 2  2  f − rf ′ 2  − 2 ℓ ( ℓ + 1) ℓ ( ℓ + 1) ! T ℓ = D ℓ + 1 r E ℓ, 1 + 1 r 2 E ℓ, 2 , with D ℓ = diag( ℓ ( ℓ − 1) , ( ℓ + 1)( ℓ + 2)) . The RN charge ﬁrst enters the error matrix E ℓ, 2 and is therefore one order shorter range than the diagonal r − 2 term. This precise asymptotic hierarc h y is what makes the Sch warzsc hild-to-RN transition conceptually transparent. The monop ole sector F or ℓ = 0 the o dd c hannel disappears and the even sector collapses to a single electric mo de. The reduced monop ole equation has the same general horizon and inﬁnity struc- ture as the higher angular momenta, but there is only one p olarization and the eﬀective in v erse-square co eﬃcien t corresponds to L = 1 in the small-mass regime. Consequently the leading small-mass in termediate exponent is 5 / 2 rather than 1 / 2 or 3 / 2 . W e return to this p oint in App endix D. Detailed horizon and inﬁnity constructions Horizon expansions and recursion form ulas W rite z = r − r + and factor the oscillatory term according to u ( z , ω ) = z − i ω / (2 κ + ) h ( z , ω ) or u ( z , ω ) = z +i ω / (2 κ + ) h ( z , ω ) , dep ending on whether one wan ts the ingoing or outgoing solution. Since r ∗ = 1 2 κ + log z + O (1) , these factors are exactly the horizon oscillations e ∓ i ω r ∗ . The remaining amplitude solv es a regular-singular equation with analytic co eﬃcien ts. W riting h ( z , ω ) = ∞ X n =0 a n ( ω ) z n , a 0 ( ω ) = 1 , giv es the recursion α n ( ω ) a n ( ω ) = n − 1 X j =0 β n,j ( ω ) a j ( ω ) , n ≥ 1 , (A.13) with analytic co eﬃcients α n , β n,j determined b y the T aylor series of the c hannel p otential. Since α n ( ω ) nev er v anishes for n ≥ 1 , the recursion uniquely deﬁnes all co eﬃcients. A similar construction applies to the even matrix system. There one writes U ( z , ω ) = z − i ω / (2 κ + ) H ( z , ω ) , 59 where H is matrix v alued and H ( z , ω ) = ∞ X n =0 A n ( ω ) z n , A 0 ( ω ) = Id 2 . The recursiv e equations are linear matrix equations with uniformly inv ertible co eﬃcients and ma y b e solved term b y term. V olterra construction at inﬁnity T o construct the inﬁnit y solution, introduce the renormalized unkno wn u ( r , ω ) = exp( − ϖ r ) r − κ m ( r , ω ) . Substituting in to the scalar c hannel equation yields m ′′ − 2 ϖ m ′ − 2 κ r m ′ = f W ℓ,P ( r , ω ) m, where f W ℓ,P ( r , ω ) = O ( r − 3 ) . Integrating twice from inﬁnit y pro duces m ( r , ω ) = 1 + Z ∞ r K ( r, s ; ω ) f W ℓ,P ( s, ω ) m ( s, ω ) d s. (A.14) The k ernel K inherits one exp onential decay factor when ℜ ϖ > 0 and one integrable algebraic factor from the Coulom b normalization. In particular, sup r ≥ R Z ∞ r    K ( r, s ; ω ) f W ℓ,P ( s, ω )    d s ≤ 1 2 for R suﬃciently large, uniformly on compact ω -sets. This gives a con traction on L ∞ ([ R, ∞ )) and hence a unique solution. The ev en matrix system is handled by the same method after conjugation by the asymptotic polarization basis. The V olterra equation then takes the matrix form M ( r, ω ) = Id 2 + Z ∞ r K ( r , s ; ω ) W ( s, ω ) M ( s, ω ) d s, and the con traction argumen t applies in the Banac h space of b ounded 2 × 2 matrix functions. Analytic dep endence on the sp ectral parameter The dep endence on ω enters through ϖ = √ µ 2 − ω 2 and κ = M µ 2 /ϖ . A w a y from the slit, b oth are analytic. Diﬀeren tiating the V olterra equation with resp ect to ω shows recursiv ely that m and M are analytic as w ell. Near the thresholds, the singular de- p endence on ϖ is completely explicit and captured by the prefactor exp( − ϖ r ) r − κ . The renormalized amplitudes remain b ounded and admit the deriv ative estimates required in the main b o dy . F or later use w e record the identit y    ∂ k ω m ( r , ω )    +    ∂ k ω M ( r, ω )    ≤ C k,ℓ r − 1 , (A.15) uniformly for r ≥ R and for ω in compact subsets of the physical sheet b ounded aw ay from the slit endp oints. W ronskians and matching matrices 60 Let u − hor , u + hor b e the horizon basis and u − ∞ , u + ∞ the inﬁnity basis. In the odd sector one deﬁnes W ℓ ( ω ) = Q [ u − hor , u ∞ ] , where u ∞ is the decaying inﬁnity solution on the physical sheet. In the even sector, letting U hor and U ∞ b e 2 × 2 fundamen tal matrices, one sets M ℓ ( ω ) = U ′ hor ( r , ω ) ∗ U ∞ ( r , ω ) − U hor ( r , ω ) ∗ U ′ ∞ ( r , ω ) . Because the p oten tial matrix is symmetric, ∂ r ∗ M ℓ ( ω ) = 0 . Hence det M ℓ ( ω ) is indep en- den t of r and is the correct matrix-v alued Ev ans determinan t. Meromorphic con tin uation across the slit Con tin uation to the nonphysical sheet is p erformed b y con tinuing the v ariable ϖ across the cut [ − µ, µ ] . The horizon basis is en tire in ω and unaﬀected b y this step. The inﬁnit y basis changes b ecause ϖ c hanges sign and the Coulomb factor acquires mono drom y . The con tin ued basis is still well deﬁned in the slit strip, and the Green kernel form ula remains v alid with the same matching determinant. This gives the meromorphic con tinuation used in the main b o dy . B Threshold mo dels and ﬁxed-mo de sp ectral com- plemen ts Bessel mo del for small Coulom b parameter When κ ≪ 1 , one freezes the Coulom b term and considers the mo del equation u ′′ +  − ϖ 2 − ν 2 − 1 4 r 2  u = 0 . (B.1) In the v ariable x = ϖ r , the deca ying solution is u ( r ) = √ r K ν ( x ) , where K ν is the mo diﬁed Bessel function. Its small-argument asymptotics are K ν ( x ) = 2 ν − 1 Γ( ν ) x − ν + 2 − ν − 1 Γ( − ν ) x ν + O ( x 2 − ν ) + O ( x 2+ ν ) , (B.2) pro vided ν / ∈ 1 2 Z ≤ 0 . The discon tin uit y across the slit is therefore of order ϖ 2 ν , whic h is the origin of the in termediate time exp onen t ν + 1 after oscillatory inv ersion. P erturbation b y the Coulomb term Restoring the Coulomb term giv es u ′′ +  − ϖ 2 + 2 M µ 2 r − ν 2 − 1 4 r 2  u = 0 . In the small- κ regime the Coulomb term is treated as a p erturbation of the Bessel mo del. The correction enters linearly in κ and therefore do es not change the principal p ow er of ϖ in the cut discontin uity . The shorter-range remainder W ℓ,P = O ( r − 3 ) con tributes one additional factor of ϖ 2 after rescaling. 61 Whittak er mo del for large Coulom b parameter F or κ ≫ 1 it is instead natural to write the mo del in the v ariable x = 2 ϖ r : u xx +  − 1 4 + κ x + 1 4 − ν 2 x 2  u = 0 . (B.3) The deca ying solution is the Whittak er function W κ,ν ( x ) . Its b eha vior as x → 0 is go verned b y Gamma factors: W κ,ν ( x ) = Γ( − 2 ν ) Γ( 1 2 − ν − κ ) x ν + 1 2 + Γ(2 ν ) Γ( 1 2 + ν − κ ) x − ν + 1 2 + · · · . (B.4) A cross the branch cut, the recipro cal Gamma factors pro duce the oscillatory mono drom y e ± 2 π i κ after Stirling asymptotics. Gamma-function asymptotics A standard Stirling expansion giv es, uniformly in ν on compact sets, Γ( 1 2 + ν + i κ ) Γ( 1 2 − ν + i κ ) = (i κ ) 2 ν  1 + O ( κ − 1 )  , κ → + ∞ . (B.5) Com bined with the reﬂection formula Γ( z )Γ(1 − z ) = π sin π z , this is enough to deriv e the t wo exp onential factors in the large- κ jump form ula. The imp ortan t point is that the phase comes en tirely from the Coulom b monodromy; the shorter-range remainder aﬀects only lo wer-order corrections. T ransfer to the physical c hannel solutions The mo del asymptotics are transferred to the exact channel equation b y matc hing at an in termediate radius R ( ϖ ) satisfying 1 ≪ R ( ϖ ) ≪ ϖ − 1 in the small- κ regime , and R ( ϖ ) ∼ ϖ − 1 / 2 in the large- κ regime . On such scales the model equation dominates, while the exact remainder is p erturbativ e. This giv es the precise threshold form ulas used in Steps 7 and 8 of Section 5. Oscillatory in v ersion and phase analysis Endp oin t in tegrals The endpoint mechanism is captured b y in tegrals of the form I α ( t ) = Z ε 0 e i atϖ 2 ϖ α d ϖ , α > − 1 . Set s = atϖ 2 . Then I α ( t ) = 1 2 a − ( α +1) / 2 t − ( α +1) / 2 Z aε 2 t 0 e i s s ( α − 1) / 2 d s. 62 Deforming the contour to the p ositiv e imaginary axis gives Z ∞ 0 e i s s ( α − 1) / 2 d s = e π i 4 ( α +1) Γ  α + 1 2  , whic h is the explicit leading term in Lemma 7.2. The remainder comes from truncating the con tour and integrating b y parts once more. The Coulom b saddle and the univ ersal exp onent The v ery-late tail is go verned b y the phase Ψ t ( ϖ ) = tϖ 2 2 µ + 2 π M µ 2 ϖ . Its deriv ative v anishes at ϖ 0 ( t ) =  2 π M µ 3 t  1 / 3 . After the scaling ϖ = t − 1 / 3 y , the phase takes the form t 1 / 3 Φ( y ) with Φ( y ) = y 2 2 µ + 2 π M µ 2 y . The second deriv ative Φ ′′ ( y 0 ) is nonzero, so a single stationary-phase step yields a factor t − 1 / 6 . Since the c hange of v ariables contributes t − 2 / 3 , the net decay rate is t − 5 / 6 . Remainders from the central part of the cut Once both endp oin t regions hav e b een separated, the remaining central frequency in terv al is treated by rep eated integration b y parts. Indeed, if ω sta ys aw ay from ± µ , then the phase deriv ative of e − i ω t is constan t and the resolv ent jump is smo oth in ω . Hence ev ery integration b y parts gains one factor of t − 1 . The compactly supp orted cutoﬀs in r ensure that all diﬀeren tiated amplitudes remain b ounded. This justiﬁes neglecting the cen tral frequency region in every asymptotic theorem. Com bination of the tw o endp oints The upp er endp oin t contributes a phase e − i µt times an oscillatory integral in ϖ . The lo w er endpoint contributes the complex conjugate phase e +i µt , up to c hannel-dep endent constan t phases coming from the threshold amplitudes. Summing the t w o endp oint con- tributions therefore giv es a real oscillatory sine or cosine. The pap er uses the sine con- v en tion A ℓ,P ( r , r ′ ; Q ) sin( µt + δ ℓ,P ( Q )) b ecause it is inv arian t under the change of sign of the threshold amplitude. Uniformit y on compact radial sets All oscillatory argumen ts are carried out with r and r ′ conﬁned to a ﬁxed compact subset K ⋐ ( r + , ∞ ) . This remo ves three tec hnical diﬃculties at once: there are no turning points on K for large angular momen tum, the c hannel transfer matrices are uniformly bounded there, and diﬀerentiation of the resolv ent kernel with respect to the radial v ariables costs only a polynomial factor in ℓ . The compact-radial-set form ulation is therefore the natural one for b oth the ﬁxed-mo de and full-ﬁeld theorems. T ec hnical complemen ts to the ﬁxed-mo de sp ectral theorem Detailed co eﬃcien t expansions at inﬁnit y 63 In the main b o dy w e used only the schematic large- r form of the c hannel equations. F or sev eral error estimates it is conv enient to record the ﬁrst few terms explicitly . W rite f ( r ) = 1 − 2 M r + Q 2 r 2 , f ( r ) µ 2 = µ 2 − 2 M µ 2 r + Q 2 µ 2 r 2 . In the o dd sector one therefore has V ℓ, 0 ( r ) = µ 2 − 2 M µ 2 r + ℓ ( ℓ + 1) + Q 2 µ 2 r 2 − 2 M ℓ ( ℓ + 1) r 3 + Q 2 ℓ ( ℓ + 1) r 4 . (B.6) In the even sector, after conjugation b y the p olarization basis, the diagonal entries are V ℓ, − 1 ( r ) = µ 2 − 2 M µ 2 r + ℓ ( ℓ − 1) + Q 2 µ 2 r 2 + O ( r − 3 ) , (B.7) V ℓ, +1 ( r ) = µ 2 − 2 M µ 2 r + ( ℓ + 1)( ℓ + 2) + Q 2 µ 2 r 2 + O ( r − 3 ) , (B.8) while the oﬀ-diagonal term is O ( r − 3 ) in Sch warzsc hild and improv es to O ( r − 4 ) for the c harge-dep enden t correction. This is the quan titative version of the statemen t that the RN c harge do es not c hange the univ ersal Coulomb co eﬃcient. The threshold index ν ℓ,P is deﬁned by ν 2 ℓ,P − 1 4 = L P ( L P + 1) + Q 2 µ 2 + ρ ℓ,P , where ρ ℓ,P is pro duced by the r − 2 diagonalization error and v anishes iden tically in the Sc h warzsc hild limit. In particular, ν ℓ,P = L P + 1 2 + O (( M µ ) 2 + ( Qµ ) 2 ) in the p erturbative-mass regime, uniformly for every ﬁxed ℓ . Cutoﬀ comm utators and lo cal resolv en t identities Let χ, e χ ∈ C ∞ 0 (( r + , ∞ )) with e χ ≡ 1 on a neighborho o d of supp χ . The cutoﬀ resolven t ob eys the identit y χ ( H ℓ − ω 2 ) − 1 χ = χ ( H ℓ − ω 2 0 ) − 1 χ + ( ω 2 − ω 2 0 ) χ ( H ℓ − ω 2 ) − 1 e χ ( H ℓ − ω 2 0 ) − 1 χ + C χ, e χ , (B.9) where the comm utator correction C χ, e χ is supp orted where the deriv atives of e χ are nonzero and is therefore harmless on compact radial sets. The crucial algebraic p oin t is that [ H ℓ , e χ ] = − 2 e χ ′ ∂ r ∗ − e χ ′′ , indep enden tly of the scalar or matrix c haracter of the channel. By iterating (B.9), one gains any ﬁnite n um b er of lo cal deriv atives at the cost of multiplying b y cutoﬀ resolven ts on sligh tly larger compact sets. A second identit y used implicitly in the time-domain inv ersion is the deriv ative formula ∂ ω ( H ℓ − ω 2 ) − 1 = 2 ω ( H ℓ − ω 2 ) − 2 . (B.10) After inserting compact cutoﬀs and using (B.9), one obtains polynomial local bounds for ∂ j ω R ℓ,χ ( ω ) a wa y from poles and thresholds. These estimates justify the repeated diﬀeren tiations in ω used in the con tour and oscillatory arguments. 64 Realit y symmetries and jump structure Because the channel co eﬃcients are real, the Jost solutions ob ey u hor ( r , ¯ ω ) = u hor ( r , ω ) , u ∞ ( r , ¯ ω ) = u ∞ ( r , ω ) , whenev er b oth sides are deﬁned. Consequen tly the Ev ans determinant satisﬁes E ℓ ( ¯ ω ) = E ℓ ( ω ) , and the contin ued resolven t k ernel ob eys the Sch warz reﬂection symmetry G ℓ ( ¯ ω ; r , r ′ ) = G ℓ ( ω ; r , r ′ ) . (B.11) Restricting (B.11) to the cut immediately giv es disc G ℓ ( ω ; r , r ′ ) = G ℓ ( ω + i0; r, r ′ ) − G ℓ ( ω − i0; r, r ′ ) = 2i ℑG ℓ ( ω + i0; r, r ′ ) . Hence the cut discon tinuit y is purely imaginary up to the common conv ention factor 2i . This is the sp ectral reason that the t w o endp oint con tributions combine in to a real sine term in the time domain. A more detailed pro of of the real-frequency exclusion The pro of of Prop osition 5.11 ma y b e ampliﬁed as follo ws. Assume ﬁrst that 0 < | ω | < µ and let u b e a non trivial channel solution ingoing at the future horizon and deca ying at inﬁnit y . The current J ( r ∗ ) = 1 2i Q [ u , u ] is constan t. At inﬁnity , exp onential decay gives J (+ ∞ ) = 0 . A t the horizon, the ingoing expansion yields J ( −∞ ) = ω | a hor | 2 . Th us a hor = 0 , so by uniqueness of the horizon Cauch y problem the solution v anishes iden tically . F or ω = ± µ , the inﬁnity asymptotics no longer decay exp onen tially , but a threshold resonan t state is b ounded after remov al of the asymptotic oscillation. Such a b ounded state still has v anishing curren t at inﬁnit y b ecause the leading threshold asymptotic coeﬃcient is real. The same con tradiction follo ws. A t ω = 0 , the curren t argumen t is insuﬃcient b ecause the horizon ﬂux v anishes iden- tically . One then returns to the full static Proca equation and inv ok es the no-hair iden- tit y of Appendix D. This separates the gen uinely static obstruction from the oscillatory threshold problem and makes the logic of the exclusion transparen t. Threshold neigh b orho o ds and quan titativ e zero-free regions The argument ab o ve giv es qualitativ e zero-freeness of the Ev ans determinant near ω = ± µ . F or contour deformation one also needs a quan titative v ersion. Let δ > 0 b e small and consider Ω ± δ = { ω : 0 < | ω ∓ µ | < δ, |ℑ ω | < η } \ [ − µ, µ ] . Since the contin ued Ev ans determinan t is analytic on Ω ± δ and nonv anishing on the b ound- ary for δ and η suﬃcien tly small, the minimum mo dulus principle yields inf Ω ± δ |E ℓ ( ω ) | ≥ c ℓ,δ,η > 0 . (B.12) 65 This lo wer b ound is not uniform in ℓ and is not used in the summed angular-momen tum theorem. It is, ho wev er, enough to justify the ﬁxed-mo de contour deformations and the threshold localizations employ ed in Section 7. Mo dewise lo cal energy expansion Although the resonance-expansion section w as remo v ed from the streamlined version of the pap er, the ﬁxed-mo de contour argumen t still gives a lo cal energy decomp osition whic h is useful conceptually . Let Γ be a contour surrounding the ﬁnitely man y p oles of the contin ued resolv ent in a compact sp ectral window and in tersecting the slit only along [ − µ, µ ] . Then χu ℓ ( t ) = X ω j ∈ Poles ℓ e − i ω j t Π ℓ,j χu ℓ [0] + 1 2 π i Z µ − µ e − i ω t χ disc R ℓ ( ω ) χ u ℓ [0] d ω + Rem ℓ ( t ) , where the remainder comes from the upp er and low er contour segmen ts and is exp onen- tially small whenever the contour is c hosen a wa y from the real axis. The main bo dy fo cuses on the branch-cut integral b ecause that is the source of the explicit oscillatory tails. The formula ab o v e clariﬁes once more wh y the discrete p ole con tribution m ust b e treated separately in the massiv e problem. C Large angular momen ta, residues, and full-ﬁeld estimates Semiclassical rescaling Set h = ( ℓ + 1 2 ) − 1 and write the channel op erator sc hematically as P ℓ,P ( h, ω ) = h 2 D 2 r ∗ + V 0 ,P ( r ) + hV 1 ,P ( r , ω ) + h 2 V 2 ,P ( r , ω ) . On a compact radial set K ⋐ ( r + , ∞ ) , the leading p otential V 0 ,P is strictly positive for all suﬃciently small h , uniformly in ω in a compact frequency window near the cut. This is the compact ellipticity input. Lemma C.1 (Compact ellipticity for large ℓ ) . F or every c omp act K ⋐ ( r + , ∞ ) ther e exist c K > 0 and ℓ K such that V 0 ,P ( r ) ≥ c K for al l r ∈ K , al l ℓ ≥ ℓ K , and al l p olarizations P . Pr o of. Since K sta ys aw ay from the horizon, f ( r ) is b ounded b elow b y a p ositive constan t on K . The quantities L P ( L P + 1) / ( ℓ + 1 2 ) 2 con v erge to 1 as ℓ → ∞ , uniformly in P ∈ {− 1 , 0 , +1 } . Hence V 0 ,P ( r ) is bounded below by a p ositive constan t on K for ℓ large. Resolv en t b ounds on compact sets Compact ellipticit y implies lo cal resolv ent estimates. Let χ ∈ C ∞ 0 (( r + , ∞ )) b e identi- cally 1 on K . A semiclassical parametrix giv es    χP ℓ,P ( h, ω ) − 1 χ    L 2 → H 2 h ≤ C K h − N 66 for some in teger N indep endent of ℓ . The point is not to optimize N but to pro ve that the loss is p olynomial and uniform. The same argumen t applies to deriv atives of the k ernel with resp ect to r and r ′ . By Sob olev em b edding on the compact set K , one obtains p oint wise kernel b ounds p olyno- mial in ℓ . Uniform con trol of threshold amplitudes The amplitudes A ℓ,P ( r , r ′ ; Q ) and B ℓ,P ( r , r ′ ; Q ) en tering the time-domain asymptotics are combinations of horizon transfer matrices, inﬁnit y transfer matrices, and threshold co eﬃcien ts of the mo del equations. Eac h factor is p olynomially b ounded in ℓ on compact radial sets. Consequently sup r,r ′ ∈ K  | A ℓ,P ( r , r ′ ; Q ) | + | B ℓ,P ( r , r ′ ; Q ) |  ≤ C K ⟨ ℓ ⟩ N 0 for some integer N 0 . Reconstruction of physical coeﬃcients The reduced channel v ariables are not themselves the ph ysical co eﬃcients of the full Pro ca p oten tial. The o dd c hannel is iden tical to the odd physical amplitude, but the ev en channels m ust b e con v erted back to ( u 2 , u 3 ) and then to the original angular coeﬃ- cien ts. Every suc h reconstruction is algebraic-diﬀerential of uniformly b ounded order. In particular, on compact radial sets,    a bc ℓm,P ( t, r )    ≤ C K ⟨ ℓ ⟩ N rec X j ≤ J sup r ∈ K    ∂ j r v bc ℓm,P ( t, r )    . This is the only place where the full v ector nature of the ﬁeld re-enters after the channel analysis. Summation o v er angular mo des and Sob olev on the sphere Using (A.5)–(A.6), one has sup ω ∈ S 2 ℓ X m = − ℓ    Y ( P ) ℓm ( ω )    2 ≲ ⟨ ℓ ⟩ 2 , uniformly in the p olarization lab el P . Therefore, by Cauc hy–Sc hw arz, ℓ X m = − ℓ    a bc ℓm,P ( t, r ) Y ( P ) ℓm ( ω )    ≲ ⟨ ℓ ⟩  X m    a bc ℓm,P ( t, r )    2  1 / 2 . The high-angular-momentum p olynomial b ound from the preceding app endix estimates is then summable provided the initial data hav e suﬃcien tly man y angular deriv atives, exactly as enco ded in the high-order energy norm E N [ A [0]] . Completion of the full-ﬁeld theorem Com bining the uniform k ernel b ounds, the reconstruction estimate, and the spherical- harmonic summation giv es Theorem 1.3. The full-ﬁeld p oint wise deca y theorem follo ws immediately b y summing the in termediate and very-late b ounds ov er ℓ and m . The ﬁnal result is p olynomial b ecause ev ery loss in the argument is polynomial in ℓ and the initial data are assumed to ha ve suﬃcien tly man y angular deriv atives. T ec hnical complemen ts to the full-ﬁeld theorem 67 Dy adic decomp osition near the branc h p oints A conv enient w ay to organize the angular summation is to decomp ose the threshold v ariable dy adically . Let ψ ∈ C ∞ 0 ((1 / 2 , 2)) satisfy X j ∈ Z ψ (2 − j x ) = 1 , x > 0 , and deﬁne ψ j ( ϖ ) = ψ (2 − j ϖ ) , j ∈ Z . Near the upp er endp oint one writes u bc , + ℓm,P ( t, r, r ′ ) = X j ≥ j 0 1 2 π i Z ε 0 e − i ω ( ϖ ) t ψ j ( ϖ ) disc G ℓ,P ( ω ( ϖ ); r , r ′ ) d ω d ϖ d ϖ . The dy adic lo calization isolates the scales on which the phase and the amplitude hav e comparable size. When 2 j ≪ ϖ 0 ( t ) , the contribution is in the pre-saddle region and is treated by rep eated in tegration by parts. When 2 j ∼ ϖ 0 ( t ) , one is at the stationary scale and the saddle analysis applies. When 2 j ≫ ϖ 0 ( t ) but still close to threshold, one is in the endpoint-Bessel regime. Compact ellipticit y with parameters F or the summed theorem, one needs not just a qualitativ e large- ℓ bound but a b ound uniform in the spectral localization and in ﬁnitely man y radial deriv atives. Let K ⋐ ( r + , ∞ ) and let χ ∈ C ∞ 0 (( r + , ∞ )) equal 1 on K . Semiclassical ellipticity gives ∥ χu ∥ H 2 h ≤ C K  ∥ χP ℓ,P ( h, ω ) u ∥ L 2 + ∥ e χu ∥ L 2  , (C.1) uniformly for large ℓ and ω in a ﬁxed compact threshold windo w. The proof is the standard p ositive-comm utator estimate with a compactly supp orted symbol equal to 1 on K . Since the principal sym b ol is elliptic there, the estimate may b e iterated to gain an y ﬁnite n umber of h -deriv ativ es. A useful consequence of (C.1) is that diﬀerentiation of the branch-cut kernel with resp ect to r and r ′ costs only a p olynomial factor in ℓ . This is why the reconstruction from the c hannel v ariables to the ph ysical ﬁeld co eﬃcien ts do es not destro y summabilit y . Angular regularit y and high-order energies Let Ω 1 , Ω 2 , Ω 3 b e the standard rotation vector ﬁelds on the round sphere. Because the bac kground is spherically symmetric, these commute with the Pro ca equation. The high-order energy norm used in Theorem 1.4 may therefore b e tak en to b e E N [ A [0]] = X | α | + j ≤ N E h ∂ j t Ω α A [0] i . After angular decomp osition, this con trols p olynomial w eights in ℓ : X ℓ,m,P ⟨ ℓ ⟩ 2 N E ℓm,P [ A [0]] ≲ E N [ A [0]] . This estimate is the analytic bridge b etw een the p olynomial losses in ℓ coming from the sp ectral argumen t and the ﬁnal absolute con v ergence of the full spherical-harmonic series. A mo del pro of of the summed intermediate b ound 68 T o illustrate the logic, consider only the in termediate regime. Using the ﬁxed-mode estimate of Theorem 7.3 and the amplitude b ound of Theorem 1.3, one obtains    a bc ℓm,P ( t, r )    ≤ C K ⟨ ℓ ⟩ N 0 t − ( ν ℓ,P +1) E ℓm,P [ A [0]] 1 / 2 . Since ν ℓ,P ≥ ν ∗ , one ma y replace t − ( ν ℓ,P +1) b y t − ( ν ∗ +1) . Summing o v er m with the v ector- harmonic addition theorem gives ℓ X m = − ℓ    a bc ℓm,P ( t, r ) Y ( P ) ℓm ( ω )    ≤ C K ⟨ ℓ ⟩ N 0 +1 t − ( ν ∗ +1)  X m E ℓm,P [ A [0]]  1 / 2 . A ﬁnal Cauc h y–Sch warz summation in ℓ is conv ergent as so on as N > N 0 + 2 in the high-order energy . This is the core of the full-ﬁeld argument; the very-late estimate is iden tical except that the common deca y rate is t − 5 / 6 . Near-horizon compact sets The compact region K ma y approach the ev en t horizon, pro vided it remains strictly inside the exterior and do es not cross r = r + . All argumen ts ab o v e still apply b ecause the tortoise co ordinate sends the horizon to −∞ and the c hannel co eﬃcien ts are smo oth on ev ery compact subset of ( r + , ∞ ) . In particular, the large- ℓ compact ellipticit y is compatible with taking K of the form K = [ r + + δ, R ] for arbitrary ﬁxed δ > 0 . Thus the p olynomial-decay theorem controls the ﬁeld uniformly up to any prescrib ed distance from the horizon. W eigh ted lo cal energy , radial Sob olev b ounds, and p oint wise reconstruction Lo cal energy norms on compact radial sets Fix a compact radial set K ⋐ ( r + , ∞ ) and let χ K ∈ C ∞ 0 (( r + , ∞ )) equal 1 on a neigh b orho o d of K . F or a time slice { t = const } , deﬁne the lo calized energy norm E K [ A ]( t ) := Z R r ∗ × S 2 χ K ( r )  | ∂ t A | 2 + | ∂ r ∗ A | 2 + |∇ S 2 A | 2 + µ 2 | A | 2  d r ∗ d ω . Because the metric co eﬃcients are smo oth and b ounded ab o v e and b elo w on K , this norm is equiv alen t to any other lo cal H 1 -based energy norm constructed from the stationary Killing ﬁeld ∂ t . The high-order lo cal energies are obtained by commuting with ∂ t and the rotation ﬁelds Ω i : E K,N [ A ]( t ) := X j + | α |≤ N E K [ ∂ j t Ω α A ]( t ) . The purpose of the branc h-cut analysis is to pro ve precise deca y estimates for these lo calized quantities after angular decomp osition. Radial Sob olev on compact in terv als Let I ⋐ R b e a compact in terv al in the tortoise co ordinate. The standard one- dimensional Sobolev inequality gives sup r ∗ ∈ I | u ( r ∗ ) | 2 ≤ C I Z I  | u ( r ∗ ) | 2 + | u ′ ( r ∗ ) | 2  d r ∗ . (C.2) 69 Applied to eac h angular co eﬃcient and summed ov er ( ℓ, m, P ) , (C.2) conv erts control of lo calized radial H 1 norms into p oint wise control in r on compact radial sets. Since all compact subsets of the exterior ma y b e cov ered b y ﬁnitely many suc h interv als, the estimate globalizes immediately . A second useful inequality is the diﬀeren tiated version sup r ∗ ∈ I | u ′ ( r ∗ ) | 2 ≤ C I Z I  | u ′ ( r ∗ ) | 2 + | u ′′ ( r ∗ ) | 2  d r ∗ . (C.3) When com bined with the c hannel equations, (C.3) shows that p oint wise control of the ﬁrst radial deriv ativ e is ultimately reducible to the same lo cal energy quantities plus one application of the channel op erator. Comm uting the channel equations with radial deriv atives On a compact radial set K , the c hannel equations imply ∂ 2 r ∗ u ℓ,P =  V ℓ,P ( r ) − ω 2  u ℓ,P + F ℓ,P , where F ℓ,P v anishes for homogeneous channel solutions and is compactly supp orted when cutoﬀ comm utators are present. Rep eated diﬀerentiation yields, for ev ery in teger j ≥ 2 , ∂ j r ∗ u ℓ,P = X q ≤ j − 2 c j,q ( r , ω , ℓ ) ∂ q r ∗ u ℓ,P + commutator terms , (C.4) with co eﬃcients c j,q p olynomial in ℓ on compact sets. Consequently all high radial deriv a- tiv es of the c hannel k ernels are con trolled b y ﬁnitely man y lo w radial deriv ativ es at the cost of a p olynomial factor in ℓ . This is the mechanism b ehind the reconstruction estimate used in the full-ﬁeld theo- rem. The full Pro ca ﬁeld is built from the reduced c hannel v ariables b y applying a ﬁnite family of radial diﬀeren tial op erators of b ounded order. Because eac h additional radial deriv ativ e costs only a p olynomial loss in ℓ , the en tire reconstruction remains compatible with angular summability . Sob olev on the sphere and tensorial harmonics Let f ( ω ) = P ℓ,m f ℓm Y ℓm ( ω ) . The usual Sob olev em b edding on S 2 giv es sup ω ∈ S 2 | f ( ω ) | 2 ≤ C X ℓ,m ⟨ ℓ ⟩ 2 s | f ℓm | 2 , s > 1 . (C.5) F or v ector and one-form v alued spherical harmonics the same estimate holds comp onen- t wise in an y ﬁxed orthonormal frame, with the same threshold s > 1 . This is equiv alent to the addition-theorem b ounds (A.5)–(A.6), but the Sob olev form ulation is often more con v enient when comm uting with angular deriv atives. Com bining (C.2) and (C.5) giv es a compact-cylinder p oin t wise estimate sup ( r,ω ) ∈ K × S 2 | A ( t, r, ω ) | ≲ X j + | α |≤ 2    χ K ∂ j r ∗ Ω α A ( t )    L 2 ( R × S 2 ) . Th us, once the branch-cut part of the channel kernels has b een con trolled in a high-order lo cal energy norm, p oint wise full-ﬁeld deca y follows automatically . The reconstruction of the Pro ca co eﬃcients 70 Returning to the original decomp osition (A.7), the angular co eﬃcien ts a 0 , a 1 , a 2 , a 3 are reco vered from the reduced v ariables by an algebraic-diﬀeren tial map of the form      a 0 a 1 a 2 a 3      = B ℓ ( r , ∂ t , ∂ r ∗ )    v − 1 v 0 v +1    , (C.6) where B ℓ is a matrix of diﬀeren tial op erators whose coeﬃcients are rational functions of r smo oth on the exterior. On compact radial sets, ev ery co eﬃcient of B ℓ is p olynomially b ounded in ℓ . Moreov er, b ecause the Lorenz constrain t has already b een built in to the reduced system, no loss of deriv atives o ccurs b ey ond the ﬁnite order enco ded in B ℓ . It follo ws from (C.6) and (C.4) that sup r ∈ K | a ℓm,P ( t, r ) | ≤ C K ⟨ ℓ ⟩ N rec X q ≤ Q sup r ∈ K    ∂ q r ∗ v ℓm,P ( t, r )    . This estimate is rep eatedly used, sometimes implicitly , in the main b o dy whenever the c hannel theorems are translated back in to full Pro ca statements. P oin t wise con trol from high-order lo cal energies Com bining the radial and angular Sob olev inequalities with the reconstruction esti- mate yields sup ( r,ω ) ∈ K × S 2    A bc ( t, r, ω )    ≤ C K,N  E K,N [ A bc ]( t )  1 / 2 , (C.7) pro vided N is suﬃcien tly large. The high-order energy on the righ t-hand side is then b ounded by the sp ectral estimates of Section 8. F ormula (C.7) is the exact analytic bridge b et w een the c hannel k ernel b ounds and the ﬁnal full-ﬁeld p oint wise decay theorem. A lo cal energy formulation of the main deca y theorem F or applications it is sometimes con venien t to state the full-ﬁeld theorem directly in lo cal energy form. The pro ofs in the main b o dy imply that for every compact K ⋐ ( r + , ∞ ) and ev ery suﬃciently large in teger N , E K,N [ A bc ]( t ) 1 / 2 ≤ C K,N E N [ A [0]] 1 / 2    t − ( ν ∗ +1) , κ ∗ ( t ) ≤ 1 , t − 5 / 6 , κ 0 ( t ) ≥ 1 . P oin twise deca y is then an immediate corollary of (C.7). This formulation emphasizes once more that the pap er prov es a compact-region decay theorem for the radiative branch- cut ﬁeld, not merely an asymptotic formula for isolated spherical mo des. P arameter-dep enden t oscillatory estimates and dy adic summation across the crosso v er scale A uniﬁed mo del integral A con venien t common mo del for the endp oint and v ery-late analyses is I ( t ; β , γ , a ) := Z ε 0 exp  i tϖ 2 2 µ + i γ ϖ  ϖ β a ( ϖ ) d ϖ , (C.8) where β > − 1 , γ ≥ 0 , and a is a smo oth amplitude. The in termediate regime corresp onds formally to γ = 0 , while the very-late regime corresp onds to γ = 2 π M µ 2 and an ampli- tude with one additional factor of ϖ coming from the Jacobian and the threshold jump. 71 W orking with (C.8) has the adv an tage that a single dyadic decomp osition captures b oth mec hanisms. Let ψ be a dy adic partition of unity and deﬁne I j ( t ; β , γ , a ) := Z ε 0 exp  i tϖ 2 2 µ + i γ ϖ  ψ j ( ϖ ) ϖ β a ( ϖ ) d ϖ . Then I ( t ; β , γ , a ) = X j ≥ j 0 I j ( t ; β , γ , a ) . The dy adic sum is split according to the relativ e size of 2 j and the stationary scale ϖ 0 ( t, γ ) :=  µγ t  1 / 3 , with the conv ention ϖ 0 = 0 when γ = 0 . Pre-saddle and p ost-saddle dyadic estimates Supp ose ﬁrst that γ > 0 and 2 j ≤ c ϖ 0 ( t, γ ) with c suﬃciently small. Then the deriv ativ e of the phase Φ ′ t,γ ( ϖ ) = tϖ µ − γ ϖ 2 is b ounded b elow in magnitude b y c 1 γ 2 − 2 j . Integration by parts with the op erator L t,γ := 1 iΦ ′ t,γ ( ϖ ) ∂ ϖ therefore yields | I j ( t ; β , γ , a ) | ≤ C N  2 2 j γ  N 2 j ( β +1) X q ≤ N sup ϖ ∼ 2 j | ∂ q ϖ a ( ϖ ) | (C.9) for ev ery integer N ≥ 0 . The same argument applies in the p ost-saddle regime 2 j ≥ C ϖ 0 ( t, γ ) , where the phase deriv ative is instead b ounded below by c 2 t 2 j /µ and gives | I j ( t ; β , γ , a ) | ≤ C N ( t 2 j ) − N 2 j ( β +1) X q ≤ N sup ϖ ∼ 2 j | ∂ q ϖ a ( ϖ ) | . (C.10) These b ounds show that all dyadic scales except the stationary ones are negligible after summation. Stationary dy adic blo cks Assume no w that 2 j ∼ ϖ 0 ( t, γ ) . Set ϖ = 2 j y and write the phase as t (2 j ) 2 2 µ y 2 + γ 2 j y . When 2 j ∼ ϖ 0 ( t, γ ) , b oth terms are of the common size Λ( t, γ ) := t 1 / 3 γ 2 / 3 µ − 2 / 3 . The rescaled phase has a nondegenerate critical p oin t y = y 0 ∼ 1 , so stationary phase yields I j ( t ; β , γ , a ) = e iΘ t,γ Λ( t, γ ) − 1 / 2 (2 j ) β +1  c 0 a (2 j y 0 ) + O  Λ( t, γ ) − 1   , (C.11) 72 where c 0  = 0 is univ ersal and Θ t,γ is the critical v alue of the phase up to the standard π / 4 correction. Substituting 2 j ∼ ϖ 0 ( t, γ ) into (C.11) giv es the generic p ow er la w | I j ( t ; β , γ , a ) | ≲ t − 1 / 6 ϖ 0 ( t, γ ) β +1 . F or the Pro ca late-time tail one has β = 1 , γ = 2 π M µ 2 , and therefore t − 1 / 6 ϖ 2 0 = t − 1 / 6  M µ 3 t  2 / 3 ∼ t − 5 / 6 . The endp oin t case as a degenerate stationary problem When γ = 0 , there is no in terior stationary p oint and the dominant contribution comes from the boundary p oin t ϖ = 0 . The same dy adic decomposition still w orks. The p ost-saddle estimate (C.10) remains v alid, while the pre-saddle regime disapp ears. Summing the remaining dyadic blo c ks giv es I ( t ; β , 0 , a ) ∼ t − ( β +1) / 2 , whic h is precisely the endp oin t law used for the in termediate tail. In this sense the in termediate and very-late asymptotics are t w o faces of the same dyadic oscillatory theory: when γ = 0 the critical point collapses in to the endp oint, and when γ > 0 it mo ves into the in terior at the scale ϖ 0 ( t, γ ) . Uniform b ounds for parameter-dep enden t amplitudes In the applications, the amplitude a ( ϖ ) dep ends on ( ℓ, P , r , r ′ ) and may itself con tain con trolled p o w ers of κ − 1 and ϖ . A conv enient uniform assumption is that, on the dy adic supp ort ϖ ∼ 2 j , | ∂ q ϖ a ( ϖ ) | ≤ C q 2 − j q 2 j σ (C.12) for some gro wth exponent σ independent of j . Then (C.9) and (C.10) remain summable after c ho osing N suﬃcien tly large, and the stationary blo ck estimate b ecomes | I j ( t ; β , γ , a ) | ≲ t − 1 / 6 ϖ 0 ( t, γ ) β +1+ σ . This is exactly the form needed when the threshold jump is m ultiplied b y polynomially b ounded transfer co eﬃcients or b y ﬁnite n um b ers of radial deriv atives. Dy adic pro of of the univ ersal time-decay exponent W e now sk etc h a dyadic pro of of the universal Pro ca exp onen t whic h is indep endent of p olarization. In the large- κ regime, the cut jump has the form disc G ℓ,P ( ω ; r , r ′ ) =  b + ℓ,P e 2 π i κ + b − ℓ,P e − 2 π i κ  ϖ + O ( ϖ 2 + κ − 1 ϖ )  . Insert this in to the branc h-cut integral and lo calize dyadically . All blo c ks aw ay from the stationary scale are negligible b y (C.9) and (C.10). The ﬁnitely man y stationary blo cks satisfy (C.11) with β = 1 . Hence each con tributes O ( t − 5 / 6 ) , and the sum of all stationary blo c ks is also O ( t − 5 / 6 ) . Since none of these estimates dep ends on the p olarization b ey ond the b ounded amplitudes b ± ℓ,P , the exp onent is univ ersal. Dy adic pro of of the in termediate exp onent Similarly , in the small- κ regime the jump has the form disc G ℓ,P ( ω ; r , r ′ ) = a ℓ,P ( r , r ′ ) ϖ 2 ν ℓ,P  1 + O ( κ + ϖ 2 )  . 73 After m ultiplying by the Jacobian d ω / d ϖ ∼ − ϖ /µ , the mo del integral has exp onen t β = 2 ν ℓ,P + 1 . The endp oint law therefore giv es t − ( β +1) / 2 = t − ( ν ℓ,P +1) . The imp ortan t p oin t is that the dy adic decomp osition again reco v ers the exact exp onen t without appealing to any heuristic matc hing argumen t. Crosso v er summation and uniform remainders The dyadic decomp osition also clariﬁes the structure of the remainder terms. Let j ∗ ( t ) b e the integer suc h that 2 j ∗ ∼ ϖ 0 ( t, γ ) . Then I ( t ; β , γ , a ) = X j ≤ j ∗ − C I j + X | j − j ∗ |≤ C I j + X j ≥ j ∗ + C I j . The ﬁrst and third sums are con trolled by rep eated integration by parts and pro duce the algebraic remainders in κ 0 ( t ) − 1 and ϖ 0 ( t ) . The middle sum is ﬁnite and handled b y stationary phase. This trichotom y is exactly the frequency-space counterpart of the informal statemen t that the con tour in tegral is divided into pre-saddle, saddle, and p ost- saddle regions. Application to full-ﬁeld angular summation Finally , b ecause the dyadic estimates are uniform under the amplitude assumption (C.12), they ma y b e combined with the p olynomial large- ℓ b ounds of App endices C and C. One ﬁrst applies the dyadic oscillatory theory at ﬁxed ( ℓ, m, P ) , then sums in m using the addition theorem, and ﬁnally sums in ℓ using the high-order energy weigh ts. This giv es an alternativ e route to Theorems 1.3 and 1.4, no w written en tirely in terms of dy adic oscillatory in tegrals rather than direct endpoint and saddle calculations. A uxiliary regime lemmas, contour b o okk eeping, and remainder estimates Relations b et w een the threshold scales The paper uses several time-dep enden t scales: κ ∗ ( t ) = M µ 3 / 2 t 1 / 2 , ϖ 0 ( t ) =  2 π M µ 3 t  1 / 3 , κ 0 ( t ) = M µ 2 ϖ 0 ( t ) . It is o ccasionally conv enient to note the algebraic relations κ 0 ( t ) ϖ 0 ( t ) = M µ 2 , κ 0 ( t ) ∼ ( M µ 3 t ) 1 / 3 , ϖ 0 ( t ) ∼ ( M µ 3 ) 1 / 3 t − 1 / 3 . Th us κ 0 ( t ) → ∞ if and only if t → ∞ , while κ ∗ ( t ) → 0 is a gen uinely separate regime restriction corresp onding to times short compared with ( M µ 3 ) − 1 . In particular, the in termediate and v ery-late theorems describ e t w o diﬀeren t asymptotic windo ws rather than t wo diﬀerent notions of large time. Threshold cutoﬀs Let η ∈ C ∞ 0 ([0 , ∞ )) satisfy η ≡ 1 on [0 , 1] and η ≡ 0 on [ 2 , ∞ ) . F or a time parameter t deﬁne the upp er-endp oint cutoﬀ η t, + ( ω ) := η  µ − ω c 0 t − 1+ σ  , 74 and similarly the low er-endp oint cutoﬀ η t, − . Then the branc h-cut in tegral decomp oses as Z µ − µ = Z µ − µ η t, + + Z µ − µ η t, − + Z µ − µ (1 − η t, + − η t, − ) . The third term is the cen tral part of the cut and is handled by rep eated in tegration b y parts. The ﬁrst tw o are the endp oint pieces treated b y the threshold asymptotic expan- sions. The exact v alues of c 0 and σ are irrelev ant as long as the endp oin t neighborho o ds shrink p olynomially in time and remain disjoint. Cen tral-frequency in tegration by parts Supp ose a ( ω ) is C N on a compact interv al I ⋐ ( − µ, µ ) a wa y from the thresholds. Then Z I e − i ω t a ( ω ) d ω = ( it ) − N Z I e − i ω t a ( N ) ( ω ) d ω after integrating by parts N times, since the b oundary terms v anish if the amplitude is cutoﬀ a wa y from the ends of I . Therefore     Z I e − i ω t a ( ω ) d ω     ≤ C I ,N t − N sup ω ∈ I    a ( N ) ( ω )    . This simple fact is rep eatedly used, often without commen t, whenev er the branch-cut in tegral is split into endp oint and cen tral pieces. A b o okkeeping lemma for endp oin t remainders Consider an endp oint integral of the form Z ε 0 e i tϖ 2 / (2 µ ) ϖ α  b 0 + b 1 κ + b 2 ϖ 2  d ϖ , α > − 1 . The leading term con tributes t − ( α +1) / 2 . The κ -term equals M µ 2 ϖ − 1 times the leading in tegrand and therefore contributes M µ 2 t − α/ 2 = κ ∗ ( t ) t − ( α +1) / 2 , once α = 2 ν ℓ,P + 1 is substituted. The ϖ 2 -term contributes one further factor of t − 1 . This is the algebraic origin of the tw o remainder terms in Theorem 7.3. The lemma is elementary , but it is exactly the kind of b o okkeeping iden tit y one needs to keep the remainder structure logically consistent throughout the pap er. Con tour segmen ts aw a y from the cut Let Γ R denote the semicircular arc ω = Re i θ , θ ∈ [0 , π ] , in the upp er half-plane. F or compactly supp orted data and ﬁxed ℓ , the cutoﬀ resolven t satisﬁes the high-frequency b ound ∥ χ R ℓ ( ω ) χ ∥ L 2 → L 2 ≤ C χ | ω | − 2 uniformly on Γ R for R suﬃcien tly large. Therefore     Z Γ R e − i ω t χ R ℓ ( ω ) χ d ω     ≤ C χ R − 1 sup θ ∈ [ 0 ,π ] e − Rt sin θ , whic h tends to 0 as R → ∞ . This justiﬁes discarding the large semicircle in the con tour deformation leading to the branc h-cut form ula. 75 Similarly , the horizontal contour segments in the slit strip con tribute exp onentially deca ying terms whenever they are placed at ﬁxed imaginary height ± η . These are the con tour pieces referred to as exp onen tially small remainders in the b o dy of the pap er. A b o okkeeping lemma for the saddle remainders In the very-late regime one writes a ( ϖ ) = β ϖ + ϖ ρ ( ϖ ) , where | ρ ( ϖ ) | ≤ C ( κ − 1 + ϖ ) on the supp ort of the threshold lo calization. Substituting into the model saddle in tegral sho ws that the error term is O  ( κ 0 ( t ) − 1 + ϖ 0 ( t )) t − 5 / 6  , b ecause ev ery factor of κ − 1 or ϖ may b e ev aluated at the stationary scale ϖ 0 ( t ) after the stationary phase lo calization. This explains why the remainder in Theorem 7.6 has exactly the form recorded there. Final b o okkeeping principle A recurring theme in the pap er is that every time-deca y exp onen t is obtained by ﬁrst iden tifying the correct scale in the sp ectral v ariable and then ev aluating the size of the amplitude at that scale. The in termediate exponent is pro duced b y the b oundary scale ϖ ∼ t − 1 / 2 and the p o w er ϖ 2 ν ℓ,P +1 . The ve ry-late exp onent is pro duced b y the stationary scale ϖ ∼ t − 1 / 3 and the linear amplitude factor ϖ . Once this principle is made explicit, the seemingly diﬀerent estimates in the pap er become instances of the same general oscillatory bo okkeeping. D A uxiliary regimes, monop ole analysis, and cutoﬀ indep endence Exact Sc h w arzsc hild limit When Q = 0 , the RN co eﬃcient simpliﬁes to f ( r ) = 1 − 2 M r , and the transformed even matrix loses its r − 4 c harge contribution. In this limit the leading r − 2 diagonalization in the p olarization basis is exact to the order needed for the threshold analysis, and the threshold indices reduce to ν ℓ, − 1 = ℓ − 1 2 , ν ℓ, 0 = ℓ + 1 2 , ν ℓ, +1 = ℓ + 3 2 . Consequen tly the in termediate branch-cut exp onen ts are exactly ℓ + 1 2 , ℓ + 3 2 , ℓ + 5 2 . 76 This repro duces, in rigorous mo dewise form, the p olarization pattern long exp ected from the Sc hw arzschild Pro ca literature. P erturbation of the threshold indices F or general sub extremal RN, the threshold indices are deﬁned b y the exact r − 2 co eﬃ- cien ts of the diagonalized channel equations. The dep endence of these co eﬃcien ts on the parameters ( M , Q, µ ) is analytic in Q 2 and smo oth in µ 2 near the massless and uncharged limits. W riting ν ℓ,P = L P + 1 2 + δ ℓ,P ( M , Q, µ ) , one ma y derive the p erturbativ e equation  2 L P + 1  δ ℓ,P = Q 2 µ 2 + ρ ℓ,P ( M , Q, µ ) + O ( δ 2 ℓ,P ) , (D.1) where ρ ℓ,P collects the shorter-range diagonalization corrections. Since ρ ℓ,P = O (( M µ ) 2 + ( Qµ ) 2 ) , one obtains δ ℓ,P = O (( M µ ) 2 + ( Qµ ) 2 ) for ev ery ﬁxed ℓ and p olarization. This p erturbative statement is conceptually imp ortan t. It sho ws that the c harge af- fects the intermediate exp onents only through the threshold index and only p erturbativ ely in the small-mass regime. By contrast, the very-late exp onen t is en tirely insensitive to the p erturbation b ecause it is go v erned b y the universal Coulomb saddle. Con tin uit y of amplitudes and phases The amplitudes A ℓ,P ( r , r ′ ; Q ) and B ℓ,P ( r , r ′ ; Q ) , as well as the constan t phases δ ℓ,P ( Q ) and δ ℓ,P, 0 ( Q ) , are contin uous in the charge parameter Q on ev ery compact radial set. This follows from the con tinuit y of the horizon and inﬁnity transfer matrices and of the threshold co eﬃcien ts in the c hannel equations. In particular, as Q → 0 one recov ers the Sc h warzsc hild amplitudes and phases. Therefore the entire RN ﬁxed-mo de theorem may b e read as a deformation of the exact Sch w arzschild p olarization-resolv ed picture. Crosso v er b etw een the intermediate and v ery-late regimes The parameter separating the t wo asymptotic regimes is κ = M µ 2 /ϖ . A t the lev el of time scales, the crossov er is describ ed by the condition κ ∼ 1 , equiv alently t ∼ ( M µ 3 ) − 1 . Before this time scale, the inv erse-square threshold index determines the decay rate and the leading oscillation is essen tially sin( µt + δ ) . After this time scale, the Coulomb phase b ecomes nonlinear in time and the universal t − 5 / 6 saddle dominates. The ﬁxed- mo de branc h-cut signal therefore has t wo asymptotic faces: a p olarization-dep enden t in termediate face and a p olarization-indep endent v ery-late face. A regime diagram It is useful to summarize the hierarch y as follows. (i) Pr e-thr eshold times: the branch p oin t is not yet dominant and other parts of the con tour ma y con tribute comparably . 77 (ii) Interme diate times: κ ∗ ( t ) ≪ 1 and the endp oin t/Bessel analysis yields the rate t − ( ν ℓ,P +1) . (iii) V ery-late times: κ 0 ( t ) ≫ 1 and the saddle/Whittak er analysis yields the univ ersal rate t − 5 / 6 . In the small-mass Sch warzsc hild-like regime, the ﬁrst non trivial exp onent among the three p olarizations is alw a ys ℓ + 1 2 , so the slo w est in termediate branch-cut decay comes from the P = − 1 c hannel. Once the very-late regime is reached, all three p olarizations synchronize to the same exp onent. Small-mass full-ﬁeld consequence The full-ﬁeld theorem simpliﬁes dramatically when ( M µ ) 2 + ( Qµ ) 2 is suﬃciently small. In that case ν ∗ = 1 2 + O (( M µ ) 2 + ( Qµ ) 2 ) , so the intermediate full-ﬁeld rate is appro ximately t − 3 / 2 and the very-late full-ﬁeld rate is exactly t − 5 / 6 . Since 3 / 2 > 5 / 6 , the o verall p olynomial decay rate is then go verned b y the univ ersal very-late tail: sup r ∈ K ,ω ∈ S 2    A bc ( t, r, ω )    ≲ t − 5 / 6 . This is the direct spin- 1 analogue of the late-time scalar picture in the p erturbativ e mass regime. What the p erturbative statemen t do es and do es not imply The perturbative formulas ab ov e concern the exact threshold indices and the branch- cut amplitudes. They do not eliminate the discrete quasib ound family , whose widths are exp onen tially sensitiv e to the trapping geometry . In the fully self-con tained core of the paper this exp onen tial sensitivit y is exactly what leads to the logarithmic full- ﬁeld estimate pro ved in Section 10. Th us the small-mass regime mak es the contin uous asymptotics fully explicit, but it do es not b y itself pro duce a p olynomial treatment of the discrete quasib ound sum. An y sharp er p olynomial upgrade would require additional arithmetic input b eyond the self-con tained pac ket argumen t dev elop ed here. The monop ole, static thresholds, and no-hair input The reduced monop ole equation When ℓ = 0 , the o dd c hannel and the magnetic ev en channel v anish identically . The remaining electric mo de may b e represen ted b y a single radial function u 0 satisfying u ′′ 0 +  ω 2 − f µ 2 − 2 f r 2  f − r f ′ 2  u 0 = 0 . A t inﬁnit y this has the same Coulomb co eﬃcien t 2 M µ 2 /r and an in verse-square co eﬃcien t corresp onding to the eﬀective angular momentum L = 1 in the small-mass regime. Con- sequen tly the monopole supports the same universal v ery-late t − 5 / 6 la w and the leading small-mass in termediate exp onent 5 / 2 . Static solutions and no-hair A t ω = 0 , the Pro ca equation b ecomes an elliptic system on the static RN exterior. The static no-hair theorem implies that there is no non trivial smo oth solution whic h is 78 regular at the horizon and decays at inﬁnity . This statemen t is used in the main b o dy to exclude static threshold pathologies. The pro of is classical: one in tegrates the static Pro ca energy iden tit y o ver the exterior region and uses p ositivity of the mass term to force v anishing of the ﬁeld. Prop osition D.1 (No static Pro ca hair on sub extremal RN) . L et A b e a static Pr o c a solution on a sub extr emal RN exterior which is r e gular at the event horizon and de c ays suﬃciently fast at inﬁnity. Then A ≡ 0 . Pr o of. Multiply the static equation b y A , integrate b y parts o ver the exterior region truncated at r = R , and let R → ∞ . The b oundary terms v anish b y regularity at the horizon and decay at inﬁnit y . The bulk identit y is the sum of ∥ F ∥ 2 and µ 2 ∥ A ∥ 2 , b oth nonnegativ e. Hence b oth v anish and the ﬁeld is iden tically zero. Consequences for threshold analysis The no-hair statement rules out one p ossible loophole in the real-frequency exclusion argumen t, namely the existence of a static zero-mo de hidden inside the threshold normal form. T ogether with the radial curren t identit y , it closes the threshold-resonance exclusion needed in Theorem 1.6. Cutoﬀ indep endence and canonical meaning of the branch-cut ﬁeld Wh y cutoﬀ indep endence matters Throughout the pap er the cut-oﬀ resolven t is written with a compactly supp orted radial cutoﬀ χ . This is natural analytically b ecause the horizon and inﬁnit y ends ha v e diﬀeren t asymptotics, but it raises an obvious conceptual question: does the branc h-cut ﬁeld dep end on the particular cutoﬀ used to deﬁne it? On a ﬁxed compact radial set the answ er is no, pro vided the cutoﬀ is chosen identically 1 on that set. The present app endix records the simple argument. A lo cal resolven t identit y Let χ 1 , χ 2 ∈ C ∞ 0 (( r + , ∞ )) b oth equal 1 on a neigh b orho o d of the same compact set K ⋐ ( r + , ∞ ) . Then, on K × K , χ 1 ( H ℓ − ω 2 ) − 1 χ 1 − χ 2 ( H ℓ − ω 2 ) − 1 χ 2 = 0 as kernels, because b oth sides solv e the same inhomogeneous equation with the same delta singularity on the diagonal and v anish after multiplication b y (1 − χ j ) near K . More concretely , the diﬀerence factors through cutoﬀ commutators supp orted aw ay from K , and those comm utators are annihilated when the k ernel is restricted back to K . Consequence for the branch-cut in tegral Applying the previous identit y to the contin ued resolv ent and taking the discon tin uity across the cut shows that the branc h-cut kernel χ ( r ) χ ( r ′ ) disc G ℓ,P ( ω ; r , r ′ ) is independent of the c hoice of cutoﬀ as long as the cutoﬀ equals 1 on the compact radial set under consideration. The same is therefore true of the time-domain in tegral deﬁning u bc ℓm,P ( t, r, r ′ ) on that set. In particular, the branc h-cut ﬁeld used in the main theorems is a canonical lo cal ob ject, not an artifact of a particular cutoﬀ construction. 79 Global in terpretation The cutoﬀ-indep endence statemen t ma y b e summarized as follo ws: the branc h-cut ﬁeld is canonically deﬁned as an elemen t of the lo cal solution space on ev ery compact radial set, and diﬀerent cutoﬀs merely pro vide diﬀerent global representativ es of the same lo cal ob ject. This is exactly the right level of inv ariance for the compact-region p oint wise theorems pro ved in the pap er. A c kno wledgemen ts The author wishes to express his deep est gratitude to his family for their un w av er- ing supp ort and encouragement. The initial stage of this w ork, namely , Fixed-mo de sp ectral and threshold theorem, is is supported by Riset Unggulan ITB 2024 No. 959/IT1.B07.1/T A.00/2024. The ﬁnal stage of this work, namely , T w o-regime asymp- totic expansion of the full Pro ca ﬁeld, is supp orted by Riset Unggulan ITB 2025 No. 841/IT1.B07.1/T A.00/2025. References [1] F. Pasqualotto, Y. Shlap entokh-Rothman, and M. V an de Mo ortel, The asymptotics of massive ﬁelds on stationary spheric al ly symmetric black holes for al l angular mo- menta , arXiv:2303.17767, 2024. [2] Y. Shlap entokh-Rothman and M. V an de Mo ortel, Polynomial time de c ay for solu- tions of the Klein–Gor don e quation on a sub extr emal R eissner–Nor dstr öm black hole , Duk e Math. J. 175 (2026), no. 1, 1–134. [3] J. Lucietti, K. Murata, H. S. Reall, and N. T anahashi, On the horizon instability of an extr eme R eissner–Nor dstr öm black hole , JHEP 03 (2013), 035. [4] J. G. Rosa and S. R. Dolan, Massive ve ctor ﬁelds on the Schwarzschild sp ac etime: quasinormal mo des and b ound states , Phys. Rev. D 85 (2012), 044043. [5] R. A. K onoplya, A. Zhidenk o, and C. Molina, L ate time tails of the massive ve ctor ﬁeld in a black hole b ackgr ound , Phys. Rev. D 75 (2007), 084004. [6] R. Ca yuso, O. J. C. Dias, F. Gray , D. Kubizňák, A. Margalit, J. E. San tos, R. G. Souza, and L. Thiele, Massive ve ctor ﬁelds in K err–Newman and K err–Sen black hole sp ac etimes , JHEP 04 (2020), 159. [7] J. D. Bekenstein, Nonexistenc e of b aryon numb er for static black holes , Phys. Rev. D 5 (1972), 1239–1246. [8] J. D. Bekenstein, Nonexistenc e of b aryon numb er for black holes. II , Ph ys. Rev. D 5 (1972), 2403–2412. 80

Threshold asymptotics and decay for massive Maxwell on subextremal Reissner--Nordström

Original Paper

Comments & Academic Discussion

Leave a Comment

Original Paper

Related Papers

Comments & Academic Discussion

Leave a Comment