Boundary Conditions, Alternate Quantization, and Deformations

For a scalar sufficiently close to the Breitenlohner–Freedman (BF) bound, the bulk equation admits two normalizable boundary falloffs. Fixing one falloff or the other does more than select a convenient solution: it selects which member of a canonical pair is the source and therefore which boundary CFT is being defined. A functional relation between the two falloffs can instead describe a multi-trace deformation. The decisive tests are finite norm, vanishing symplectic flux, a self-adjoint boundary-value problem, and a stable spectrum; none of those tests can replace the others.

The derivation starts in Lorentzian asymptotically AdS$_{d+1}$ with the site’s $(+---)$ convention. Correlators and the renormalization-group flow are then computed in the Euclidean Poincaré patch. Here $d\ge2$ is the boundary spacetime dimension, $L$ is the AdS radius, and the conformal boundary lies at $z=0$. The explicit response calculation treats a free scalar on a fixed AdS background at leading semiclassical order. Backreaction and nonlinear stability are separated out at the end.

Required background. Timelike Boundary, Causality, and Boundary-Value Problems supplies the symplectic-flux and self-adjointness criteria. Bulk Fields and Boundary Operators supplies the mass–dimension relation and the basic near-boundary dictionary.

Helpful background. Conformal Perturbation Theory and Beta Functions supplies the field-theory interpretation of the beta function. Timelike Boundaries, Self-Adjoint Extensions, and AdS Boundary Conditions gives the curved-spacetime extension framework.

The BF window is a boundary-condition window

With the boundary d’Alembertian $\Box_{\partial}$ in the $(+---)$ convention, the Lorentzian Klein–Gordon equation reduces near $z=0$ to

$$\left[ -z^2\partial_z^2+(d-1)z\partial_z +z^2\Box_{\partial}+m^2L^2 \right]\phi=0.$$

Substituting $\phi\sim z^\Delta$ gives the indicial equation

$$\Delta(\Delta-d)=m^2L^2.$$

Its two roots are

$$\Delta_\pm=\frac d2\pm\nu, \qquad \nu=\sqrt{\frac{d^2}{4}+m^2L^2},$$

so a nonresonant solution has the near-boundary form

$$\phi(z,x) =z^{\Delta_-}\alpha(x)+z^{\Delta_+}\beta(x)+\cdots.$$

Under a boundary dilation, $[\alpha]=\Delta_-$ and $[\beta]=\Delta_+$. The coefficient with the smaller exponent is the slow falloff. Define the BF mass by $m_{\rm BF}^2L^2=-d^2/4$. The BF bound,

$$m^2L^2\ge -\frac{d^2}{4},$$

is the requirement that $\nu$ be real. It is only the first admissibility test.

For a mode of finite boundary frequency, the radial part of the ordinary Klein–Gordon norm of the slow falloff behaves as

$$\int_0^\epsilon dz\,z^{1-d}\left|z^{\Delta_-}\right|^2 =\int_0^\epsilon dz\,z^{1-2\nu}.$$

The integral converges precisely when $\nu<1$. Combining that condition with distinct real powers, $\nu>0$, gives the open alternate-quantization window

$$-\frac{d^2}{4}<m^2L^2<-\frac{d^2}{4}+1, \qquad 0<\nu<1.$$

In this window, the fast falloff is normalizable as usual and the slow falloff is also eligible. Eligibility is not yet a quantization: one must still choose a positive self-adjoint domain. The same upper endpoint follows from the scalar CFT unitarity bound,

$$\Delta_-\ge\frac{d-2}{2},$$

which becomes $\nu\le1$. The open interval is the generic two-quantization case. Klebanov and Witten 1999, §2.1, especially Eqs. (2.2)–(2.8), Open PDF relate the two falloffs to the two CFT dimensions; Ishibashi and Wald 2004, §3.3 and Theorem 3.2, Open PDF give the positive self-adjoint-extension analysis.

On a narrow screen, each comparison table below reflows into labeled cards. At intermediate widths, it scrolls horizontally without shrinking the text.

Scalar mass ranges and admissible quantizations

Mass range	Near-boundary behavior	Ordinary unitary interpretation
m2 < m2BF	ν is imaginary	BF instability; no ordinary stable AdS vacuum
m2 = m2BF	the powers coalesce and a logarithmic branch appears	separate BF-endpoint analysis with scale mixing
0 < ν < 1	two distinct normalizable falloffs	standard and alternate quantizations are both possible before further stability tests
ν = 1	the slow norm is logarithmically divergent and Δ− = (d − 2)/2	exceptional free/singleton and logarithmic-counterterm issues; not the generic alternate theory
ν > 1	the slow branch has divergent ordinary norm	standard quantization only in an ordinary unitary CFT

The two admissible choices in the open window describe two CFTs obtained from the same bulk mass and Lagrangian. They do not mean that one fixed CFT contains the same single-trace operator with both dimensions at once.

Boundary flux selects admissible relations

Let $\mathcal N_\phi$ denote a positive boundary response normalization, with the conventional factor $L^{d-1}$ from the bulk action absorbed into it once and for all. For a boundary metric $g_{(0)}$, the renormalized Lorentzian flux through a boundary region $R$ has the form

$$\Omega_R(\delta_1\phi,\delta_2\phi) \propto 2\nu\mathcal N_\phi \int_R d^dx\,\sqrt{\left|g_{(0)}\right|}\, \left( \delta_1\alpha\,\delta_2\beta -\delta_2\alpha\,\delta_1\beta \right),$$

where the omitted overall sign depends on the orientation of the boundary normal. The variations are tangent to the space of allowed solutions. Consequently:

• fixing $\alpha$ gives $\delta\alpha=0$ and zero flux;
• fixing $\beta$ gives $\delta\beta=0$ and zero flux;
• imposing $\beta=\delta\mathcal B[\alpha]/\delta\alpha$ gives zero integrated flux when the Hessian of the real functional $\mathcal B$ is symmetric under the boundary inner product.

For the last case, write $\delta\beta=K\delta\alpha$. The two terms in $\Omega_R$ cancel when $K(x,y)=K(y,x)$. Geometrically, an admissible boundary condition selects a Lagrangian subspace of the canonical boundary data $(\alpha,\beta)$.

Flux conservation is necessary for Hamiltonian evolution, but it does not prove that the spatial operator is positive. A real Robin parameter can close the flux while still admitting a negative eigenvalue. Likewise, a frequency-dependent or nonlocal kernel needs a separate causal and self-adjointness analysis.

The Legendre transform exchanges source and response

Choose a definite Euclidean convention. Let $I_D$ be the renormalized on-shell saddle functional with $\alpha$ held fixed, and define the connected generator by $\mathcal W_+=-I_D$. Throughout the open window, choose a nonresonant finite-counterterm scheme in which $\beta$ denotes the renormalized response coefficient, with removable local response terms set to zero. Choose the radial orientation so that

$$\delta I_D =-C_\phi\int d^dx\,\beta\,\delta\alpha, \qquad C_\phi\equiv2\nu\mathcal N_\phi>0.$$

Standard quantization therefore has

$$J_+=\alpha, \qquad \langle\mathcal O_+\rangle =\frac{\delta\mathcal W_+}{\delta J_+} =C_\phi\beta.$$

To hold the conjugate variable fixed, add the finite boundary term

$$I_N=I_D+C_\phi\int d^dx\,\alpha\beta.$$

Its variation is

$$\delta I_N=C_\phi\int d^dx\,\alpha\,\delta\beta.$$

Defining the alternate source as $J_-=-\beta$ gives the positive-normalization dictionary

$$\mathcal W_-=-I_N, \qquad \langle\mathcal O_-\rangle =\frac{\delta\mathcal W_-}{\delta J_-} =C_\phi\alpha.$$

The minus sign in $J_-=-\beta$ is a convention, not new physics. One may instead call $\beta$ the source and $-\mathcal O_-$ the operator. Physical correlators agree after translating the source and operator definitions together. Klebanov and Witten 1999, §2.2, Eqs. (2.17)–(2.20), Open PDF show the inverse two-point functions produced by the transform, while Papadimitriou 2007, §3, Tables 2–3, Open PDF states the renormalized source and effective-action dictionaries.

At the semiclassical large-$N$ saddle, the change of boundary polarization reduces to the Legendre transform just displayed. Beyond saddle level, the corresponding operation is a functional Fourier transform, and loop corrections can modify both normalization and operator dimensions.

Standard, alternate, and mixed source–response dictionaries

Quantization	Fixed source	One-point response	Source-free condition	CFT dimension
Standard	J+ = α	⟨O+⟩ = Cφβ	α = 0	Δ+
Alternate	J− = −β	⟨O−⟩ = Cφα	β = 0	Δ−
Mixed from alternate	j = −β + U′(σ)	σ = Cφα	β = U′(σ)	generally an RG flow, not a CFT

A multi-trace deformation becomes a mixed condition

Start with the alternate CFT and deform its Euclidean action by a local potential $\mathcal U(\mathcal O_-)$. With an external source $j$, write

$$Z_{\mathcal U}[j] =\int\mathcal D\Phi\, \exp\left[ -S_-[\Phi]-\int d^dx\,\mathcal U(\mathcal O_-) +\int d^dx\,j\mathcal O_- \right].$$

At large $N$, factorization permits a saddle in the expectation value

$$\sigma\equiv\langle\mathcal O_-\rangle=C_\phi\alpha.$$

The deformed effective action is $\Gamma_-[\sigma]+\int\mathcal U(\sigma)-\int j\sigma$. Its stationary equation gives

$$J_-=j-\mathcal U'(\sigma).$$

Using $J_-=-\beta$ yields the normalized mixed-source relation

$$j=-\beta+\mathcal U'(C_\phi\alpha).$$

Thus zero external source imposes

$$\beta=\mathcal U'(C_\phi\alpha).$$

Because this is the derivative of a real functional, its linearized kernel is symmetric and the symplectic flux vanishes. This is the same variational relation as Witten 2001, §3, Eqs. (3.7)–(3.9), Open PDF and Papadimitriou 2007, §3, Eq. (3.15), Open PDF, after translating their coefficient conventions.

For the double-trace deformation

$$\mathcal U(\sigma)=\frac{\lambda_{\rm bc}}{2}\sigma^2,$$

the source becomes

$$j=-\beta+f_{\rm bc}\alpha, \qquad f_{\rm bc}\equiv C_\phi\lambda_{\rm bc}.$$

At $j=0$, the bulk condition is $\beta=f_{\rm bc}\alpha$. This normalization map matters: the coefficient of $\mathcal O_-^2/2$ in the CFT action is $\lambda_{\rm bc}=f_{\rm bc}/C_\phi$, not $f_{\rm bc}$, unless the operator has been rescaled accordingly.

Both coefficients have mass dimension

$$[\lambda_{\rm bc}]=[f_{\rm bc}] =d-2\Delta_-=2\nu>0,$$

so the deformation is relevant throughout the open window. More generally, an $n$-trace term $\lambda_n\mathcal O_-^n$ has $[\lambda_n]=d-n\Delta_-$ and is relevant, marginal, or irrelevant according to whether that number is positive, zero, or negative.

The RG fixed point depends on the coupling coordinate

The Robin coefficient $f_{\rm bc}$ is a dimensionful crossover scale in the finite boundary condition. In the direct dimensionless boundary-condition coordinate

$$q(\mu)\equiv f_{\rm bc}\mu^{-2\nu},$$

holding $f_{\rm bc}$ fixed gives

$$\beta_q\equiv\mu\frac{dq}{d\mu}=-2\nu q.$$

The alternate CFT is $q=0$. Along the stable positive trajectory, lowering $\mu$ sends $q\to+\infty$, which is the standard-quantization endpoint. Thus the infrared fixed point lies at coordinate infinity in $q$.

The familiar finite quadratic fixed point uses a different, bounded CFT coupling coordinate. For example, define

$$g=g_*\frac{q}{1+q}.$$

The chain rule gives

$$\beta_g =-2\nu g+\kappa g^2, \qquad \kappa\equiv\frac{2\nu}{g_*},$$

so the same endpoints are

$$g_{\rm UV}=0, \qquad g_{\rm IR}=g_*=\frac{2\nu}{\kappa}.$$

The finite location $g_*$ is therefore coordinate and normalization dependent; it is not a second finite value of $f_{\rm bc}$. More general scheme changes alter the higher terms, and nonfactorizing correlators or operator mixing add model-dependent corrections. There is no universal $O(N^{-2})$ remainder because vector and adjoint large-$N$ limits have different counting.

Aharony, Gur-Ari, and Klinghoffer 2015, §4.1, Eqs. (4.4), (4.7)–(4.8), Open PDF make the nonresonant relevant coupling-coordinate distinction explicit. Gubser and Klebanov 2003, §2, Eqs. (2)–(12), Open PDF derive the relevant large-$N$ flow and its Legendre-transform endpoint. Witten 2001, §4, Eqs. (4.1)–(4.9), Open PDF gives the important marginal $\nu=0$ quadratic-beta example; it should not be mistaken for the direct running of $f_{\rm bc}$ at nonzero $\nu$.

Interior regularity makes the response solvable

In Euclidean Poincaré AdS, Fourier transform along the boundary and let $k=\sqrt{k_i k_i}>0$. The radial equation is

$$\left[ z^2\partial_z^2-(d-1)z\partial_z -k^2z^2-m^2L^2 \right]\phi(z,k)=0.$$

The solution regular as $z\to\infty$ is

$$\phi(z,k)=A(k)z^{d/2}K_\nu(kz).$$

For $0<\nu<1$, the small-$z$ expansion gives

$$\begin{aligned} \alpha(k)&=A(k)2^{\nu-1}\Gamma(\nu)k^{-\nu},\\ \beta(k)&=A(k)2^{-\nu-1}\Gamma(-\nu)k^\nu. \end{aligned}$$

Therefore

$$\frac{\beta(k)}{\alpha(k)} =2^{-2\nu}\frac{\Gamma(-\nu)}{\Gamma(\nu)}k^{2\nu} =-C_\nu k^{2\nu},$$

where

$$C_\nu \equiv-2^{-2\nu}\frac{\Gamma(-\nu)}{\Gamma(\nu)} =2^{-2\nu}\frac{\Gamma(1-\nu)}{\nu\Gamma(\nu)}>0.$$

Combining regularity with $j=-\beta+f_{\rm bc}\alpha$ gives

$$j(k)=\left(C_\nu k^{2\nu}+f_{\rm bc}\right)\alpha(k).$$

Since $\sigma=C_\phi\alpha$, the connected two-point response is

$$G_f(k) \equiv\frac{\delta\sigma(k)}{\delta j(k)} =\frac{C_\phi}{C_\nu k^{2\nu}+f_{\rm bc}}.$$

This one formula performs three independent checks:

1. Ultraviolet dimension. At $f_{\rm bc}=0$, $G_0(k)\propto k^{-2\nu}$. Since a scalar two-point function of dimension $\Delta$ scales as $k^{2\Delta-d}$, this is $\Delta=\Delta_-$.

2. Infrared dimension. For $f_{\rm bc}>0$ and $k\ll k_f$, where $k_f=(f_{\rm bc}/C_\nu)^{1/(2\nu)}$,

   $$G_f(k) =\frac{C_\phi}{f_{\rm bc}} -\frac{C_\phi C_\nu}{f_{\rm bc}^2}k^{2\nu} +\cdots.$$

   The constant is a local contact term. Equivalently, normalize the endpoint operator as $\mathcal O_{\rm IR}\propto f_{\rm bc}\mathcal O_-$ and discard its local term. The first remaining nonlocal term scales as $k^{2\nu}=k^{2\Delta_+-d}$. The infrared primary therefore has dimension $\Delta_+$.

3. Linear stability. For $f_{\rm bc}\ge0$, the denominator has no zero at positive Euclidean $k$. For $f_{\rm bc}<0$, it vanishes at

   $$k_*=\left(\frac{|f_{\rm bc}|}{C_\nu}\right)^{1/(2\nu)}.$$

   Analytic continuation gives a tachyonic mode at sufficiently small spatial momentum. Thus a real flux-preserving condition can still be unstable.

The inverse kernel and gap equation agree with Papadimitriou 2007, §4.1, Eqs. (4.5)–(4.8), Open PDF. Troost 2004, §§3.3–4, Eqs. (12)–(16), Open PDF identifies the bound-state pole for the unstable sign. Local finite counterterms can change analytic contact terms, but they cannot remove the nonanalytic scaling or a physical pole after all source definitions are translated consistently.

The AdS₄ mass −2/L² scalar is the diagnostic example

Set $d=3$ and $m^2L^2=-2$. Then

$$\nu=\frac12, \qquad \Delta_-=1, \qquad \Delta_+=2.$$

The half-integer Bessel function is elementary:

$$K_{1/2}(kz)=\sqrt{\frac{\pi}{2kz}}e^{-kz}.$$

Consequently the regular mode is, up to normalization,

$$z^{3/2}K_{1/2}(kz) \propto z e^{-kz} =z-kz^2+O(z^3).$$

Hence

$$C_{1/2}=1, \qquad \beta(k)=-k\alpha(k), \qquad G_f(k)=\frac{\mathcal N_\phi}{k+f_{\rm bc}}.$$

Every abstract statement is now visible:

• at $f_{\rm bc}=0$, $G_0\propto k^{-1}$ has the $\Delta_-=1$ scaling;
• for $f_{\rm bc}>0$, the flow scale is $k_f=f_{\rm bc}$, and the infrared nonlocal term $-\mathcal N_\phi k/f_{\rm bc}^2$ has the $\Delta_+=2$ scaling;
• as $f_{\rm bc}\to+\infty$, the zero-source condition $\beta=f_{\rm bc}\alpha$ forces $\alpha\to0$ for a finite solution, approaching standard quantization;
• for $f_{\rm bc}<0$, the Euclidean pole is at $k_*=|f_{\rm bc}|$.

For Lorentzian frequency $\omega$ and spatial momentum $\mathbf p$, take the retarded principal branch

$$k=\sqrt{\mathbf p^2-(\omega+i0)^2}.$$

Squaring the continued pole condition then gives

$$\omega^2=\mathbf p^2-|f_{\rm bc}|^2.$$

Modes with $|\mathbf p|<|f_{\rm bc}|$ grow exponentially. This is the adversarial case: the mass remains inside the two-quantization window and the mixed condition remains real and flux preserving, yet the sign $f_{\rm bc}<0$ fails the spectral-stability test. In the pure quadratic boundary model, the corresponding CFT potential $\mathcal U=\lambda_{\rm bc}\sigma^2/2$ is also unbounded below. Stabilizing higher-trace terms can change the nonlinear boundedness question, but they define a different deformation and do not erase the linear pole about the same vacuum without changing its quadratic Hessian.

Flux conservation is not nonlinear gravitational stability

The pole test proves or disproves linear stability for this fixed-background scalar problem. It does not establish a nonlinear positive-energy theorem for a scalar coupled to gravity. On global AdS, even the allowed range of a linear Robin parameter depends on the geometry, angular mode, and coefficient convention; Ishibashi and Wald 2004, §3.3, Theorem 3.2 and Eqs. (169)–(186), Open PDF give the precise positive-extension criterion.

In “designer gravity,” the boundary condition is specified by a nonlinear potential, the scalar backreacts, and stability depends on the complete holographic effective potential and on properties of the bulk scalar potential. Amsel and Marolf 2006, §VI, especially Eq. (6.19), and §VII, Open PDF obtain an energy lower bound under additional hypotheses: the boundary potential has a global minimum, the BF bound is not saturated, and the bulk potential admits the required superpotential. A formal zero-flux condition supplies none of those hypotheses.

Distinct boundary-condition and stability tests

Test	Question answered	What it does not prove
Norm	Is the falloff eligible for the phase space?	Self-adjointness or positivity
Symplectic flux	Is the canonical form conserved on allowed variations?	Absence of negative modes
Self-adjointness	Does the spatial operator have a real self-adjoint domain supporting conservative classical evolution?	Positivity needed for a stable positive-frequency quantization
Pole or spectrum	Are there linear tachyons or negative modes in the stated background?	Nonlinear gravitational energy bounds
Bounded effective energy	Is the backreacting theory stable under the theorem's hypotheses?	Validity outside those hypotheses

Endpoints and nonlocal conditions require new analyses

At the BF point $\nu=0$, the two powers coalesce:

$$\phi=z^{d/2}\left[\alpha\log(z\mu)+\beta+\cdots\right].$$

Changing $\mu$ mixes the two coefficients. Source and response therefore run into one another, and the nominal double-trace interaction is marginal with logarithmic running. Without introducing an additional scale, only one natural conformally invariant source-free choice survives. One cannot obtain this endpoint by simply inserting $\nu=0$ into formulas that divided by $2\nu$.

At $\nu=1$, the alternate dimension saturates the scalar unitarity bound. The slow norm diverges logarithmically, and null descendants, logarithmic counterterms, and a possible singleton boundary degree of freedom make this another separate endpoint. For $\nu>1$, the slow branch violates the ordinary scalar unitarity bound and its norm diverges, so the Legendre algebra alone does not manufacture a unitary alternate CFT.

A relation such as $\beta(k)=F(\omega,\mathbf p)\alpha(k)$ can formally cancel flux if its kernel is symmetric, yet poles or nonanalytic frequency dependence may introduce instabilities, acausality, or extra degrees of freedom. Such a kernel is a new boundary theory and needs a new causal spectral analysis; it is not justified by the local double-trace calculation above.

What the calculation establishes

For a scalar with $0<\nu<1$, this page has established five linked results:

1. both falloffs have finite norm;
2. $\alpha$ and $\beta$ are conjugate boundary data;
3. fixing either member produces the $\Delta_+$ or $\Delta_-$ CFT dictionary;
4. a normalized double-trace deformation imposes $j=-\beta+f_{\rm bc}\alpha$;
5. interior regularity yields a response that flows from $k^{-2\nu}$ to the nonlocal $k^{2\nu}$ scaling, while the wrong sign produces an explicit pole.

The calculation is linear in the bulk scalar, leading order in the semiclassical/large-$N$ expansion, and tied to the displayed source convention. Interactions can mix operators and multi-trace couplings. Finite local counterterms can shift contact terms and coupling coordinates. Gauge fields, gravity, and mixed-symmetry fields carry additional constraints and global data.

Conformal Perturbation Theory and Beta Functions develops the field-theory beta-function machinery. Scalar Counterterms and the Renormalized Action derives the counterterms underlying $I_D$, while Finite Counterterms, Schemes, and Multi-Trace Data distinguishes a scheme change from a change of quantization or theory. Dictionary Normalization and Global-Data Audit supplies the cross-convention checklist.

Common pitfalls

Treating normalizability as the whole boundary-value problem. A finite norm only makes a falloff eligible. Flux conservation, self-adjointness, positivity, and causality remain separate checks.

Assigning both dimensions in one fixed CFT. Standard and alternate quantization are different boundary theories. A double-trace deformation connects their fixed points; it does not give one operator two simultaneous scaling dimensions.

Identifying the Robin coefficient with every CFT coupling coordinate. The dimensionful $f_{\rm bc}$ fixes the mixed boundary condition and has its standard endpoint at infinity. A finite quadratic fixed point belongs to a nonlinear, scheme-dependent coordinate such as $g$.

Calling every finite $f_{\rm bc}$ a conformal boundary condition. The dimensionful parameter introduces the scale $k_f$. Generic finite $f_{\rm bc}$ describes an RG flow, whereas the scale-invariant endpoints are the two CFT quantizations.

Inferring the infrared dimension from the constant term. The leading small-$k$ term in $G_f$ is a contact term and vanishes at separated points. The first nonanalytic term determines the infrared scaling.

Naming a stable sign without declaring conventions. Reversing the definition of $\beta$, the source, or the deformation coupling reverses verbal sign labels. The invariant diagnostics are the pole location, reflection positivity, and bounded energy after a complete translation.

Substituting directly at $\nu=0$ or $\nu=1$. Both endpoints contain logarithmic or null-state structure absent from the open window. They require their own counterterms and boundary Hilbert-space analysis.

Exercises

1. Recover the alternate-quantization window

Starting from the radial Klein–Gordon norm, show that the slow mode is normalizable only for $\nu<1$. Explain why the generic two-quantization interval is open at both ends.

Solution

On a constant-time slice, the radial measure and unit normal contribute $z^{1-d}$. For $\phi\sim z^{\Delta_-}$,

$$\lVert\phi\rVert_{\rm KG}^2 \sim\int_0^\epsilon dz\,z^{1-d+2\Delta_-} =\int_0^\epsilon dz\,z^{1-2\nu}.$$

An integral $\int_0^\epsilon dz\,z^a$ converges when $a>-1$, so $1-2\nu>-1$ and hence $\nu<1$. Reality of the exponents requires $\nu\ge0$. At $\nu=0$ the two roots coincide and generate a logarithmic solution; at $\nu=1$ the norm is logarithmically divergent and the alternate operator saturates the unitarity bound. Removing both exceptional endpoints leaves $0<\nu<1$.

2. Test a nonlinear mixed condition

Let $\beta(x)=\delta\mathcal B[\alpha]/\delta\alpha(x)$ for a real twice-differentiable functional $\mathcal B$. Prove that the integrated boundary flux vanishes for tangent variations. Why does the proof not establish stability?

Solution

Define

$$K(x,y)=\frac{\delta^2\mathcal B} {\delta\alpha(x)\delta\alpha(y)}.$$

Reality and differentiability give $K(x,y)=K(y,x)$ under the boundary inner product. Since

$$\delta_i\beta(x)=\int d^dy\,K(x,y)\delta_i\alpha(y),$$

interchanging $x$ and $y$ in the second term shows

$$\int d^dx\, \left(\delta_1\alpha\,\delta_2\beta -\delta_2\alpha\,\delta_1\beta\right)=0.$$

This proves only that the allowed data form a flux-free canonical subspace. The corresponding self-adjoint operator can still have a negative eigenvalue, as the $f_{\rm bc}<0$ pole demonstrates.

3. Check the source–response exchange

Starting from $\delta I_D=-C_\phi\int\beta\,\delta\alpha$, vary

$$I_N=I_D+C_\phi\int\alpha\beta$$

and recover both one-point functions in the convention $J_+=\alpha$ and $J_-=-\beta$.

Solution

The product variation cancels the $\beta\,\delta\alpha$ term:

$$\begin{aligned} \delta I_N &=-C_\phi\int\beta\,\delta\alpha +C_\phi\int \left(\beta\,\delta\alpha+\alpha\,\delta\beta\right)\\ &=C_\phi\int\alpha\,\delta\beta. \end{aligned}$$

Because $\mathcal W_+=-I_D$,

$$\frac{\delta\mathcal W_+}{\delta\alpha}=C_\phi\beta.$$

Because $\mathcal W_-=-I_N$ and $\delta J_-=-\delta\beta$,

$$\delta\mathcal W_- =-C_\phi\int\alpha\,\delta\beta =C_\phi\int\alpha\,\delta J_-,$$

so $\langle\mathcal O_-\rangle=C_\phi\alpha$.

4. Solve the AdS₄ response and locate the instability

For $d=3$ and $m^2L^2=-2$, use the regular mode to derive $\beta=-k\alpha$. Then compute $G_f(k)$, extract both endpoint dimensions, and continue the $f_{\rm bc}<0$ pole to Lorentzian signature.

Solution

Since $\nu=1/2$,

$$z^{3/2}K_{1/2}(kz)\propto z e^{-kz} =z-kz^2+\cdots.$$

Comparison with $z\alpha+z^2\beta+\cdots$ gives $\beta=-k\alpha$. The mixed source is therefore $j=(k+f_{\rm bc})\alpha$, while $\sigma=\mathcal N_\phi\alpha$, so

$$G_f(k)=\frac{\mathcal N_\phi}{k+f_{\rm bc}}.$$

At $f_{\rm bc}=0$, $G_0\propto k^{-1}=k^{2(1)-3}$, giving $\Delta_-=1$. For $f_{\rm bc}>0$,

$$G_f(k)=\frac{\mathcal N_\phi}{f_{\rm bc}} -\frac{\mathcal N_\phi}{f_{\rm bc}^2}k+\cdots.$$

The first term is a contact term; the nonlocal $k=k^{2(2)-3}$ term gives $\Delta_+=2$ after the infrared operator is normalized. If $f_{\rm bc}=-|f_{\rm bc}|$, the pole lies at Euclidean $k=|f_{\rm bc}|$. With $k^2\to\mathbf p^2-(\omega+i0)^2$, the pole condition is $\omega^2=\mathbf p^2-|f_{\rm bc}|^2$, so modes with $|\mathbf p|<|f_{\rm bc}|$ are tachyonic.

5. Translate between RG coordinates

Let $q=f_{\rm bc}\mu^{-2\nu}$ and $g=g_*q/(1+q)$. Derive both beta functions and identify the same ultraviolet and infrared endpoints in the two coordinates.

Solution

Holding the dimensionful Robin coefficient fixed gives

$$\beta_q=\mu\frac{dq}{d\mu}=-2\nu q.$$

Since $dg/dq=g_*/(1+q)^2$,

$$\begin{aligned} \beta_g &=\frac{g_*}{(1+q)^2}(-2\nu q)\\ &=-2\nu g\left(1-\frac{g}{g_*}\right)\\ &=-2\nu g+\frac{2\nu}{g_*}g^2. \end{aligned}$$

Thus $\kappa=2\nu/g_*$. The alternate fixed point is $q=0$, equivalently $g=0$. Lowering $\mu$ at fixed positive $f_{\rm bc}$ sends $q\to\infty$, equivalently $g\to g_*$. The standard endpoint is at infinity in the direct boundary-condition coordinate and finite in the bounded coordinate.

The chapter overview contains the structure diagram and validity and failure diagram. They are embedded there once so that their shared chapter-level context is not repeated on every article.

For the chapter-wide comparison of assumptions, counterevidence, falsifiers, and claim ceilings, see the claim-domain table.

References

• Aharony, Ofer, Guy Gur-Ari, and Nizan Klinghoffer. “The Holographic Dictionary for Beta Functions of Multi-Trace Coupling Constants.” Journal of High Energy Physics 05 (2015): 031. doi:10.1007/JHEP05(2015)031. Open PDF.
• Amsel, Aaron J., and Donald Marolf. “Energy Bounds in Designer Gravity.” Physical Review D 74 (2006): 064006; erratum 75 (2007): 029901. doi:10.1103/PhysRevD.74.064006. Open PDF.
• Gubser, Steven S., and Igor R. Klebanov. “A Universal Result on Central Charges in the Presence of Double-Trace Deformations.” Nuclear Physics B 656 (2003): 23–36. doi:10.1016/S0550-3213(03)00056-7. Open PDF.
• Ishibashi, Akihiro, and Robert M. Wald. “Dynamics in Non-Globally-Hyperbolic Static Spacetimes III: Anti-de Sitter Spacetime.” Classical and Quantum Gravity 21 (2004): 2981–3014. doi:10.1088/0264-9381/21/12/012. Open PDF.
• Klebanov, Igor R., and Edward Witten. “AdS/CFT Correspondence and Symmetry Breaking.” Nuclear Physics B 556 (1999): 89–114. doi:10.1016/S0550-3213(99)00387-9. Open PDF.
• Papadimitriou, Ioannis. “Multi-Trace Deformations in AdS/CFT: Exploring the Vacuum Structure of the Deformed CFT.” Journal of High Energy Physics 05 (2007): 075. doi:10.1088/1126-6708/2007/05/075. Open PDF.
• Troost, Jan. “A Note on Causality in the Bulk and Stability on the Boundary.” Physics Letters B 578 (2004): 210–214. doi:10.1016/j.physletb.2003.10.003. Open PDF.
• Witten, Edward. “Multi-Trace Operators, Boundary Conditions, and AdS/CFT Correspondence.” arXiv:hep-th/0112258 (2001). Open PDF.