Holographic RG flows and wormholes from sinusoidal scalars

While a perturbative solution in the dimensionless source strength $\widetilde{J}$ can be found analytically, to incorporate full backreaction due to the scalar fields, we need to look for a numerical solution. In practice, we will implement these boundary conditions numerically by evaluating the near-boundary conditions at a UV cutoff $z=\epsilon\ll 1$, and the deep-bulk conditions at an IR cutoff $z=z_{\text{max}}\gg\epsilon$

We will justify why this is a reasonable approach a posteriori once we obtain the desired solutions.

To solve our boundary value problem, we will use the shooting method; swapping the boundary conditions at $z_{\text{max}}$ for two shooting parameters $A^{\prime}(\epsilon)=p$ and $\varphi^{\prime}(\epsilon)=q$ at $z=\epsilon$ turns it into a more tractable initial value problem. Numerically solving the coupled differential equations Eqs. (19) and (20) with the above initial conditions gives us a two-parameter family of solutions, $A(z;p,q)$ and $\varphi(z;p,q)$. Standard numerical root-finding can then give the desired solutions with parameters $p=p_{0}$ and $q=q_{0}$ such that $A^{\prime}(z_{\text{max}};p_{0},q_{0})=0$ and $\varphi^{\prime}(z_{\text{max}};p_{0},q_{0})=0$, satisfying the original boundary conditions.

Examples of one-boundary solutions obtained from the shooting method are shown in Fig. 1.

A general feature of these solutions is that the radial scalar profile $\varphi(z)$ grows as it moves away from the boundary, reaching a maximum value before decaying to zero as it moves deeper into the bulk. Simultaneously, the FG scale factor gets deformed away from Poincaré–AdS, with $A(z)$ decreasing accordingly as $\varphi(z)$ increases, before relaxing to the original Poincaré–AdS geometry as $\varphi(z)$ decays back to zero.⁹⁹9While it may appear that the minima of $A(z)$ and the maxima of $\varphi(z)$ coincide, in actuality the scalar peaks first, and the backreacting dip in the FG scale factor lags behind slightly. To see this last point explicitly, observe that $A(z)$ asymptotes to a value $0<A_{\infty}<1$ as $z\to\infty$. The metric deep in the bulk then becomes

This is just the Poincaré–AdS metric with a trivially rescaled boundary metric; the original AdS length is left unchanged. If we denote by $z_{\infty}$ the value of $z$ where the scalar field has completely died down and the FG scale factor remains constant, i.e.

we see that for $J\geq 1$, the solutions return to Poincaré–AdS at $z_{\infty}\sim O(1)$, with $z_{\infty}$ decreasing as $J$ increases. In particular, we have checked that increasing $z_{\text{max}}$ in Eq. (24) beyond $z_{\infty}$ does not change the numerical solutions obtained through our shooting algorithm. This justifies our approach in using an IR cutoff to numerically implement regularity as $z\to\infty$. We also expect this method to apply more generally for arbitrary boundary dimension $d$ and other relevant operators $\Delta<d$.

As Fig. 1 suggests, we find that one-boundary solutions exist for any source strength $J$. We will see in Section 4 that wormhole solutions only exist above a minimum critical value of $J$. The solutions in Fig. 1 were chosen in anticipation of this and for comparison reflects solutions below ($J=10$), around ($J=50$), and above ($J=200$) the critical threshold. At large $J$, the metric becomes strongly deformed closer to the boundary than for small $J$, but even in such cases $A(z)$ never fully reaches zero.¹⁰¹⁰10To give the reader a sense of the numbers involved, when $J=1000$ for example, the FG scale factor reaches a global minimum of $A(0.04)\approx 0.003$. We expect that the geometry only truly pinches off in the bulk when either $J\to\infty$, where the source is so strong that $A(z)$ hits zero before it has a chance to rebound, or in the infinite wavelength limit $k\to 0$, where the boundary source is just a constant and $\varphi(z)$ monotonically increases until $A(z)=0$. Indeed, the fact that $A_{\infty}$ decreases as $J$ increases is consistent with this limit where the source becomes homogeneous.

Asymptotically AdS spacetimes were first understood in Girardello et al. (1998) and Freedman et al. (1999a) as the geometrization of RG flows induced by relevant deformations of the dual CFT. Given that we have the same asymptotic geometry and similar boundary conditions, it is natural to want to interpret our one-boundary solutions through the lens of holographic RG flows. The main difference here is that the putative flow dual to our geometry stems from sinusoidal sources on the boundary, unlike the usual case where the sources are homogeneous.

To better appreciate this difference, let us briefly recall what happens for homogeneous deformations. In the framework of the holographic renormalization group, the radial coordinate is identified with the renormalization scale as $\mu\sim 1/z$ Peet and Polchinski (1999); Porrati and Starinets (1999); Balasubramanian and Kraus (1999). The boundary CFT defined on $\mathbb{R}^{d}$ at $z=0$ therefore corresponds to a UV fixed point of the RG flow. Turning on sources for scalar operators with dimension $\Delta<d$ induces a relevant deformation of the UV CFT, the physics of which is captured by the geometry away from the boundary. A central result which extends the celebrated $c$-theorem of Zamolodchikov Zamolodchikov (1986) is the holographic $c$-theorem established in Freedman et al. (1999a). It guarantees that the holographically defined $c$-function must monotonically decrease under nontrivial RG flow under the assumption of Poincaré invariance and the null energy condition (NEC). Additionally, it is well known that for homogeneous deformations, if the geometry cannot flow to another AdS vacuum, i.e. an IR CFT, then the geometry terminates at some finite depth $z_{\text{max}}$ (Fig. 2, right), where the dual field theory is gapped or confining in the IR Girardello et al. (1999, 2000); Polchinski and Strassler (2000). In our setup where the scalar potential is quadratic so that no other AdS extrema are available, this is precisely the expectation alluded to at the end of Section 3.1.

In contrast, we find that for our one-boundary solutions, relevant deformations induced by sinusoidal boundary sources are not gapped as the geometry extends smoothly to $z\to\infty$ (Fig. 2, left). One way to see this explicitly is to compute the holographic $\beta$- and $c$-functions introduced in Freedman et al. (1999a), which quantitatively probe how the CFT evolves under RG flow. Standard applications of this framework compute $\beta$-functions for spatially independent couplings (sources) as a gradient flow of the corresponding homogeneous bulk scalar fields. Since our couplings $J_{I}(\vec{x})$ in Eq. (16) are spatially dependent, however, we emphasize that what we calculate here is not the $\beta$-function for the fully spatially-dependent sources, but rather an effective holographic $\beta$-function obtained from the spatially homogeneous radial mode $\varphi(z)$ common to our bulk scalar fields.¹¹¹¹11Several studies have explored aspects of a $\beta$-function for spatially-dependent couplings, but this subject remains somewhat underdeveloped. See e.g. Hoyos et al. (2013); Hartnoll et al. (2015); Aharony et al. (2016). Nevertheless, given our sinusoidal scalar ansatz, one reasonable interpretation of the resulting $\beta$-function is that it describes a spatially-dependent running coupling of the form

with $J$ setting the UV value for the running coupling.

Under this interpretation, the holographic $\beta$- and $c$-functions in the metric Eq. (18) are

where $-\ln z$ is the RG time. Applying Eq. (28) to our one-boundary solutions, we see in Fig. 3 that both $\beta(\varphi)$ and $c(z)$ exhibit non-monotonic, cyclic behavior under RG flow. Indeed, the $\beta$-function is multivalued with a $\beta(\varphi)<0$ and a $\beta(\varphi)>0$ branch. For each flow, $\beta(\varphi)$ leaves the UV fixed point located at $\varphi=0$, traversing the negative branch until it reaches a maximum value of the scalar field, $\varphi_{\text{max}}$. The $\beta$-function then switches to the positive branch and flows back toward $\varphi=0$; the fixed point where the flow originated is now an IR attractor. If we go by usual RG lore, the zero of $\beta(\varphi)$ at $\varphi_{\text{max}}$ belies the fact that it is not a fixed point of the RG transformation. Unlike single-valued $\beta$-functions encountered in standard quantum field theories, multivalued $\beta$-functions have richer renormalization structures, and a more suitable diagnostic is the RG “acceleration” Curtright and Zachos (2011)

We have verified that at the branch point $\varphi_{\text{max}}$, the RG acceleration is finite and positive. $\varphi_{\text{max}}$ is therefore a genuine “turning point” of the RG flow where $\beta$ changes sign as it passes through; there is no scale invariant theory at the branch point.¹²¹²12True fixed points have all higher $\ln z$-derivatives of $\beta$ vanishing such that the flow lands sufficiently “softly” at the zero Curtright and Zachos (2011). In a similar vein, the holographic $c$-function initially decreases from $c_{0}$ as expected, but reverses course and overshoots past $c_{0}$, before settling down back to $c_{0}$ as $z\to\infty$. The point where $c(z)$ momentarily returns to $c_{0}$ in the middle of the flow coincides with the point where $A(z)$ reaches its global minimum.¹³¹³13Incidentally, this is not at the same point of the flow as the turning point of the $\beta$-function because $A^{\prime}(z)$ and $\varphi^{\prime}(z)$ reaches their respective extrema at different $z$. This type of flow where the theory leaves the UV repeller only to return to it as an IR attractor is suitably known as a boomerang RG flow Donos et al. (2017, 2018).

Are these one-boundary boomerang flows pathological given that they appear to violate the $c$-theorem? Monotonicity of the holographic $c$-function requires Poincaré invariance and the NEC Freedman et al. (1999a). Since we work in Euclidean signature and our scalar fields are sinusoidal, neither condition holds and there is no guarantee that $c(z)$ should be monotonic. Indeed, many known examples of consistent single-coupling RG flows that exhibit non-monotonic, chaotic, or other exotic behaviors have multivalued $\beta$-functions Curtright et al. (2012); Ilderton (2020); Kiritsis et al. (2017). The non-monotonicity of the holographic $c$-function and the two-branched $\beta$-function in Fig. 3 therefore indicate that our boomerang flows fall under this class of exotic RG flows that are physically sensible. While this does mean we can no longer interpret $c(z)$ as a measure of effective degrees of freedom at intermediate scales, in the spirit of the $c$-theorem, we can nevertheless identify $c(z)$ with the central charge at fixed points of these boomerang flows.¹⁴¹⁴14To be precise, in the same vein as Casini et al. (2011), because our main example is in $d=3$ boundary dimensions where the trace anomaly is absent, the holographic $c$-function is expected to be proportional to the $F$-coefficient proposed by Jafferis et al. (2011) which was proven to obey an $F$-theorem in Casini and Huerta (2012). The fact that the $\beta$-function returns to the origin, $c(z)$ returns to $c_{0}$, and the geometry deep in the bulk as given by Eq. (25) is conformally equivalent to the boundary geometry are three diagnostics that point towards a single fact: the UV and IR fixed points describe one and the same CFT!

To get an intuition for why the RG flow for one-boundary solutions return, we can examine the effective action for $\varphi(z)$

Because the bulk scalar mass is related to the dual operator dimension by $m^{2}=\Delta(\Delta-d)$, from the scale-dependent effective mass term $m^{2}_{\mathrm{eff}}(z)$ for $\varphi$, it is clear that even if the initial deformation was relevant in the UV ($m^{2}_{\mathrm{eff}}(0)<0$), it effectively becomes irrelevant in the IR ($m^{2}_{\mathrm{eff}}(z)>0$). Unlike the homogeneous case where the scalar mass remains unchanged along the RG flow and the scalar field grows monotonically, here $\varphi(z)$ begins to decay once its effective mass becomes positive, allowing the gravitational backreaction to settle, and $A(z)$ to eventually relax back to Poincaré–AdS. It is crucial to note that our multivalued $\beta$-functions are a direct consequence of the gradient energy of the sinusoidal scalars, giving rise to both the non-monotonicity and cyclic nature of the boomerang RG flows.

To compare the one-boundary and wormhole solutions on equal footing, we need to understand what values of $J$ give rise to each wormhole solution. Since our wormholes were found by fixing $a_{0}$, we have to extract the corresponding $J$ by applying the holographic dictionary to the asymptotics of $\varphi(\eta)$. Recalling the exposition surrounding Eq. (14), near the boundary, the leading behavior of $\varphi(\eta)$ is given by its non-normalizable and normalizable modes, the coefficients of which correspond to the boundary source and VEV of the dual scalar operator Gubser et al. (1998); Witten (1998a). Changing to near-boundary FG coordinates where $z\simeq 1/a(\eta)\simeq\eta_{*}-\eta$, for $d=3$ and $\Delta=2$ we have¹⁶¹⁶16More generally, we have Eq. (84) for any $d$ and $\Delta$ where conformal anomalies are absent.

Extracting the source is then simply a matter of fitting $\varphi(\eta)$ in a small window around $\eta_{*}$ to the power law approximation above. From the initial conditions in Eq. (36), it is apparent that the family of solutions is controlled by the ratio $a_{0}/k$ that appears in the denominator of $\varphi_{0}$. Thus, all physically distinct wormhole solutions can be characterized by $(J/k,~a_{0}/k)$. For a given throat size $a_{0}/k$, the corresponding dimensionless source strength $J/k$ that supports the wormhole solution is shown in Fig. 5.

It turns out that for a given source, there are generally two connected solutions: a small and a large wormhole. As $J/k$ decreases and reaches a critical value $(J/k)_{\text{crit}}$, both branches of wormholes merge into a single wormhole solution we dub the “critical” wormhole. Below this threshold however, no wormhole solutions exist.¹⁷¹⁷17Presumably, the solution curve moves to somewhere in the complex $(a_{0}/k)$-plane. Indeed, complex saddles to the gravitational path integral arising from this type of bifurcation have been found even in the standard Hawking–Page transition Mahajan and Singhi (2025). What role such complex saddles may play for wormholes and factorization is an interesting question that will be left for future work. In the $d=3,\Delta=2$ example shown above in Fig. 5, this critical wormhole occurs at $(J/k)_{\text{crit}}\simeq 22.5$ and has a throat size of $(a_{0}/k)_{\text{crit}}\simeq 0.493$. The $a_{0}=1.0$ solution (orange curve) is therefore a close representation of the critical wormhole. This bifurcating structure of wormhole saddles is in accordance with the results of Marolf and Santos (2021) for various bottom-up models with compact boundary manifolds.

Unlike Marolf and Santos (2021) however, our $\mathbb{R}^{3}$ wormholes only exist in a finite band between a minimum and maximum value of $a_{0}/k$. In particular, $a_{0}/k$ is upper bounded by the denominator of $\varphi_{0}$ in Eq. (36) reaching zero. For $d=3,\Delta=2$, this is $(a_{0}/k)_{\text{max}}=\sqrt{2}/2\simeq 0.707$. As $a_{0}/k$ approaches this value from below, $\varphi(0)$ goes from being local minimum to a global maximum, and the asymptotic boundary gets pushed in towards the throat at $\eta=0$, eventually reaching it in the limit $a_{0}/k\to(a_{0}/k)_{\text{max}}$, or equivalently as $J/k\to\infty$ for the large branch of wormholes. On the other hand, we have a lower bound on the throat size of $(a_{0}/k)_{\text{min}}\simeq 0.373$ determined numerically. As $a_{0}/k$ approaches this from above, the global maximum of $\varphi(\eta)$ gets pushed out towards the boundary at $\eta_{*}$, reaching it as $a_{0}/k\to(a_{0}/k)_{\text{min}}$, or as $J/k\to\infty$ in the small branch of wormholes. We therefore see that in Fig. 4, the $a_{0}=1.0$ solution (orange curve) serves to demarcate the small and large branches of wormholes as represented by the $a_{0}=0.8$ solution (blue curve) and $a_{0}=1.2$ solution (green curve), which themselves are examples of wormholes with throat sizes close to $(a_{0}/k)_{\text{min}}$ and $(a_{0}/k)_{\text{max}}$ respectively. Attempts to find solutions with initial conditions outside $(a_{0}/k)_{\text{min}}<a_{0}/k<(a_{0}/k)_{\text{max}}$ result in geometries that do not satisfy AdS boundary conditions; $a(\eta)$ goes to zero at finite $\eta$.

It is natural to imagine that a wormhole describes two RG flows that start separately in the UV and join in the IR at the throat (Fig. 6, left). This is certainly one of several interpretations that have been ascribed to wormholes as holographic RG flows (see Ghodsi et al. (2023); Chandra (2024) for discussions). Here we adopt the perspective of Van Raamsdonk (2022); Antonini et al. (2024, 2022) and make this intuition precise. In a similar spirit as the folding trick for boundary CFTs Wong and Affleck (1994), we can view our $\mathbb{Z}_{2}$-symmetric wormholes not as a two-sided flow for two CFTs on $\mathbb{R}^{d}$, but instead as a one-sided holographic RG flow for a CFT defined on $\mathbb{R}^{d}\times S^{0}$, where the flow terminates at the wormhole throat after traversing a finite radial depth (Fig. 6, right). The IR theory is therefore gapped¹⁸¹⁸18Since our theory is defined primarily in Euclidean signature, strictly speaking we do not have a Hamiltonian as in Lorentzian signature. Our usage of the term “gapped” here refers to the exponential decay of two-point correlators for large separations with a weight determined by the lowest eigenvalue of the scalar wave operator on the wormhole geometry. This plays the role analogous to that of the Lorentzian mass gap.; the throat of the wormhole is where the $S^{0}$ “pinches off”, much like the confinement mechanism of Witten (1998b) where the radius of the $S^{1}$ smoothly contracts to zero. In the following, we will study the consequences of interpreting the wormhole geometry as a gapped holographic RG flow.

To fold our wormhole in half, let us define $\tilde{\eta}=|\eta|$ as the radial coordinate so that the entire geometry lives between $\tilde{\eta}\in(0,\eta_{*})$. This allows us to find FG coordinates with which we can interpret the folded wormhole as a holographic RG flow. Given a wormhole solution, Eqs. (18) and (31) tell us that a one-to-one mapping from $\tilde{\eta}$ to $z$ can be found by solving

with the boundary condition

In practice, this amounts to numerically integrating from zero to some UV cutoff $\eta_{*}-\epsilon$. The result is a one-to-one function $z(\tilde{\eta})\in(0,z_{0})$, where the boundary is now at $z=0$, and the throat is at some finite radial depth $z=z_{0}$. To see how the wormhole solutions compare with the disconnected solution, we fix $J/k=50$ as an example and show them all on the same plot in Fig. 7.

Unsurprisingly, the near-UV flow of all three saddles coincides because they share the same boundary source. The key difference between the connected and disconnected solutions lies in the IR: the wormhole saddles terminate at finite $z$ whereas the disconnected saddle extends infinitely deep into the bulk. The small wormhole and disconnected geometry share the same extrema within numerical precision for $A(z)$ and $\varphi(z)$. On the other hand, the extrema for the large wormhole are shifted relative to the other saddles; the minimum of $A(z)$ occurs at smaller $z$, and the maximum of $\varphi(z)$ occurs at larger $z$.

Having obtained $A(z)$ and $\varphi(z)$ for all three saddles, we can now study how the holographic flow functions for the folded wormholes compare with their disconnected cousin. In terms of the FLRW scale factor in FG coordinates $a(z)$, they are

where for the reader’s convenience we have included the $\beta$-function which is unchanged from before. The flow functions for all three saddles are plotted together in Fig. 8 for comparison.

Like the boomerang flows of Section 3.2, the wormhole flows also have multivalued $\beta$-functions with one positive and one negative branch. Both small and large wormholes share their near-UV trajectory with the boomerang flow before differing in the IR. Instead of returning to the origin, after switching branches at $\varphi_{\text{max}}$ all wormhole flows end at the throat, $\varphi_{0}$ defined in Eq. (36). This is how the “pinching off” of the $S^{0}$ in $A(z)$ manifests itself in the RG perspective. A crucial observation is that $\varphi_{0}$ is again a turning point of the RG flow, not a fixed point; the RG acceleration in Eq. (29) is finite and negative at $\varphi_{0}$.¹⁹¹⁹19In fact, if we extend the RG flow by working with the unfolded wormhole, we will find that $\varphi_{0}$ is another branch point where the $\beta$-function continues past the throat to another negative branch.

How do we see that the IR theory is “gapped”? In the conformal FLRW metric of Eq. (31), any $\mathbb{Z}_{2}$-symmetric wormhole has finite range in the $\eta$ direction. It follows from Antonini et al. (2022) that the corresponding Sturm–Liouville problem for a probe scalar field on the wormhole background has a discrete spectrum of eigenvalues. If we were working in Lorentzian signature, we could immediately conclude that this implies a discrete particle spectrum for the dual CFT and call it a day. But because we have a Euclidean wormhole, we need to work a little harder; the Euclidean theory does not come equipped with a Hilbert space a priori.²⁰²⁰20One has to specify a Euclidean “time” direction to define a Hamiltonian operator and construct a Hilbert space, provided the Osterwalder–Schrader axioms Osterwalder and Schrader (1975) are satisfied. The Hilbert space of the Lorentzian quantum field theory after analytic continuation therefore depends on how the Euclidean theory is time-sliced.

The AdS/CFT extrapolate dictionary tells us that two-point correlators of CFT operators can be obtained by extrapolating bulk vacuum correlators to the boundaries at $\pm\eta_{*}$ Gubser et al. (1998); Witten (1998a); Freedman et al. (1999b). Denoting an insertion of the operator $\mathcal{O}(\vec{x})$ on the left boundary at $-\eta_{*}$ by $\mathcal{O}_{-}(\vec{x})$ and an insertion on the right boundary at $+\eta_{*}$ by $\mathcal{O}_{+}(\vec{x})$, for $d=3$ the two-point functions are (see Section 2.1 of Antonini et al. (2022) for details)

for same-boundary insertions and

for cross-boundary insertions. In the above expressions, the mode sums are over the normalizable basis of solutions for the Klein–Gordon equation on the wormhole background.²¹²¹21By a normalizable basis, we mean in the same sense as standard quantization on an AdS background, e.g. as in Eq. (38). These are $L^{2}$-normalizable solutions to the Klein–Gordon equation. Given a wormhole scale factor $a(\eta)$, the coefficients $u_{n}$ of the normalizable modes and the eigenvalues $\lambda_{n}$ can always be determined by solving the corresponding Sturm–Liouville problem Antonini et al. (2022). As $|\vec{x}-\vec{y}|\to\infty$, the leading contribution for both types of correlators come from the $n=0$ term

and similarly for $\braket{\mathcal{O}_{-}(\vec{x})\mathcal{O}_{\mp}(\vec{y})}$. This exponential falloff of a two-point function is the hallmark of a gapped theory. In Euclidean signature, this implies that the correlation length of the system is finite and determined by the lowest eigenvalue, $\xi=1/\sqrt{\lambda_{0}}$, i.e. the “mass gap”.

Can the presence of the wormhole throat be detected in the UV? A geometric argument illustrated in Fig. 9 highlights the fact that the pinching off of the $S^{0}$ and thus the gapping of the boundary field theory is strictly an IR phenomenon. As $|\vec{x}-\vec{y}|\to 0$, scalar two-point functions in standard quantum field theory diverge as $|\vec{x}-\vec{y}|^{-2\Delta}$. This is the leading UV behavior that one would expect from the same-boundary correlators in Eq. (42). The story for cross-boundary correlators is somewhat more subtle.²²²²22The absence of short-distance singularities in wormhole cross-correlators, and more generally the analytic structure of wormhole two-point functions, were first studied in Betzios et al. (2019) and further developed in Betzios et al. (2022). For two disconnected boundaries, the cross-correlator trivially vanishes; the two boundary theories do not talk to each other. For two boundaries connected via a bulk wormhole, however, not only are cross-correlators nonzero, but they also remain finite as $\vec{x}$ approaches $\vec{y}$. From the perspective of the folded wormhole, this is due to the fact that the operator insertions are on opposite points of $S^{0}$, so we can have $\vec{x}=\vec{y}$ without insertions being physically on top of each other. Because the UV divergence of same-boundary correlators drowns out any finite IR contribution from the wormhole geometry, same-boundary correlators alone are insufficient to probe bulk connectedness.²³²³23Another way to understand this is that there is always a unique bulk geodesic connecting two points on the same boundary. Such a geodesic extends into the wormhole, but never passes the throat. From this point of view, same-boundary correlators are inherently incapable of detecting the presence of the wormhole throat. Only cross-correlations or large separations can “see” the wormhole throat; bulk connectedness is invisible in the UV.

We first calculate the renormalized Euclidean action for the one-boundary geometry of Section 3. Evaluating the canonical action Eq. (5) on the sinusoidal scalar ansatz in FG coordinates given by Eqs. (12) and (18), we get

The counterterm action is

Note that the curvature counterterm from Eq. (46) vanishes identically since the boundary metric is flat. Because none of the integrals above depend on $\vec{x}$, we can define $\int\mathrm{d}^{3}x\equiv\text{Vol}(\mathbb{R}^{3})$ and factor out the formally infinite integral over $\mathbb{R}^{3}$ from Eq. (45) to work with the renormalized action density $s_{\text{ren}}^{\text{(FG)}}$ instead.²⁴²⁴24In the following, we will refer to the action density simply as the action. Its precise meaning should hopefully be clear from context.

At this point, we can insert the numerical solutions for $A(z)$ and $\varphi(z)$ found in Section 3.1 to obtain $s_{\text{ren}}^{\text{(FG)}}(\widetilde{J})$ for particular solutions. We emphasize that because one-boundary solutions are parametrized by the dimensionless source strength $\widetilde{J}\equiv J/k^{d-\Delta}$ (or equivalently, $J$ and $k$), the renormalized action is ultimately a function of $\widetilde{J}$. To compute the remaining $z$ integrals, in practice we will numerically integrate between $\epsilon<z<z_{\text{max}}$. This is sensible to do since we have already seen from Eq. (26) that the metric quickly relaxes to Poincaré-AdS in the bulk. So long as we choose $z_{\text{max}}$ such that the behavior in Eq. (26) persists, we can analytically estimate the truncated part of the action (density) integral to be just the last term of Eq. (47),

Since $0<A_{\infty}<1$ always, even for seemingly small values like $z_{\text{max}}=2$, the error is at most $s_{\text{error}}<1/4$, with typical values of $A_{\infty}\sim 0.7$ giving $s_{\text{error}}\sim 0.18$. As the on-shell action itself is at least $O(10^{3})$, the error due to truncating the integral at finite $z$ is negligible. Given a choice of $\widetilde{J}$, once we combine all the components of $s_{\text{ren}}^{\text{(FG)}}$ for a single copy of the one-boundary geometry, we can simply double it to get the full action for the disconnected saddle, $s^{\text{(disc.)}}(\widetilde{J})\equiv 2s_{\text{ren}}^{\text{(FG)}}(\widetilde{J})$.

Next we turn to the wormhole geometries of Section 4. Working with the sinusoidal scalar ansatz in conformal FLRW coordinates given by Eqs. (12) and (31), each component of the canonical action Eq. (5) is given by

The two terms in $S_{\text{GHY}}^{\text{(conn.)}}$ have opposite sign due to the outward normal pointing in opposite directions. The corresponding holographic counterterms are (one set for each boundary),

Because our wormholes are $\mathbb{Z}_{2}$-symmetric, $a(\eta)$ and $\varphi(\eta)$ are even functions, so