betaFeedback

WorksPhysics and Astronomy

Exact Stress Tensor Flows in Causal Electrodynamics: Auxiliary Field Formalism

Abstract

Causal self-dual nonlinear electrodynamics admits a simple auxiliary-field description: the Lagrangian is linear in the two Lorentz invariants of the field strength, and the dynamics is encoded in a potential W(φ)W(φ) of a single auxiliary variable. We study non-perturbative TTˉT\bar T-like flows of these theories in this formalism. The irrelevant (stress-tensor) deformation acts on the potential by a rescaling, while the marginal ModMax-type deformation translates its vacuum. We derive the resulting flow equations exactly, show that the irrelevant exact TTˉT\bar T-like flow takes the autonomous form ∂λL=X F(r)\partial_λ\mathcal L=X\,F(r), where the closed form kernel FF depends only on the ratio r=−Tμμ/4X−Yr=-T^μ_μ/\sqrt{4X-Y} of stress-tensor invariants. We also give an explicit reconstruction of the auxiliary potential from a given kernel. Causality imposes the differential inequality D(r)>1/(1−r)D(r)>1/(1-r), while a globally single-valued kernel requires the ratio rr to be monotonic along the auxiliary branch. We examine the relation between the evenness of the exact TTˉT\bar T-like flow , the φφ-parity of the potential, and the analyticity of the Lagrangian in the electromagnetic invariants, including the effect of a marginal ModMax parameter. We construct the linear auxiliary and Hamiltonian formulations and show that the Hamiltonian flow is governed by the same exact TTˉT\bar T-like flow with the opposite sign. We also express Born's Legendre self-duality, the two characteristic cones, and the energy conditions in terms of the kernel. In particular, r=(ε−p⊥)/(ε+p⊥)r=(\varepsilon-p_\perp)/(\varepsilon+p_\perp) has a direct physical interpretation. Exact TTˉT\bar T-like flow s are obtained for generalized Born-Infeld, logarithmic, q-deformed, and no-ττ-maximum electrodynamics.

Introduction and main of results

The TTˉT\bar T deformation of two-dimensional quantum field theories [11, 12, 48] (see [23, 24] for reviews) is a prototypical example of a deformation driven by a composite operator built from the energy-momentum tensor. Its classical version, in which the Lagrangian obeys a flow equation ∂λL=det⁡Tμν\partial_\lambda\mathcal L=\det T_{\mu\nu}, generates the Nambu–Goto action from a free scalar and, more generally, produces a large class of solvable irrelevant deformations. Two ideas have turned out to be particularly fruitful. The first is that the deformation operator need not be the determinant of the stress tensor: any function of its Lorentz invariants defines a “TTˉT\bar T-like” flow, and marginal members of this family, such as the root-TTˉT\bar T operator, preserve classical integrability of sigma models [6, 5, 45]. The second is that the construction is not tied to two dimensions: the flows can be defined in arbitrary dimension by a metric (or stress-tensor) approach [49, 7, 50], and in particular for four-dimensional nonlinear electrodynamics (NED), where Born–Infeld [46] and ModMax theories arise from Maxwell theory by stress-tensor flows in the same way as the Nambu–Goto action arises from a free boson [42, 4, 43, 19, 22].

Four-dimensional NED (see [47] for a review) has its own distinguished sub-class: theories that are invariant under electric–magnetic duality rotations [3, 2, 1, 25]. Duality invariance of a Lagrangian L(S,P)\mathcal L(S,P) of the two invariants SS and PP of the field strength is equivalent to a single nonlinear PDE, which is solved by Born–Infeld theory and by the conformal ModMax theory [16], and which is preserved by the stress-tensor flows [19]. The interplay of such flows with duality, in particular of the root-TTˉT\bar T operator, was studied in [22, 34], and the effect of such flows on birefringence in [44]. A parallel development, due to Russo-Townsend, is the systematic understanding of when a self-dual NED is causal, meaning that perturbations propagate inside or on the light cone, and which energy conditions it obeys [17, 13]. Self-dual causal theories are described by a function of one variable. This can be done either through a Courant-Hilbert (CH) construction, in which a self-dual Lagrangian is generated from a boundary function ℓ(τ)\ell(\tau) with ℓ˙≥1\dot{\ell}\ge1 and ℓ¨≥0\ddot{\ell}\ge0 [35, 18, 21], or, as shown most recently, through a simplified formulation with a single auxiliary field φ\varphi [15]1 (see also [37, 38]),

L(S,P;φ)=cosh⁡φ S+sinh⁡φ S2+P2−W(φ),\begin{equation} \mathcal{L}(S,P;\varphi)=\cosh\varphi\,S+\sinh\varphi\,\sqrt{S^2+P^2}-W(\varphi), \tag*{(1)} \end{equation}

where the equation of motion for φ\varphi is algebraic, W(φ)W(\varphi) is the potential satisfying dL/dφ=0d\mathcal{L}/d\varphi=0, and the two electromagnetic invariants are S=−14FμνFμν,S=-\frac14F_{\mu\nu}F^{\mu\nu}, and P=−14FμνF~μνP=-\frac14F_{\mu\nu}\widetilde F^{\mu\nu}. In this formulation duality invariance is manifest, since the Lagrangian is linear in SS and S2+P2\sqrt{S^2+P^2} once φ\varphi is treated as independent. Moreover, weak-field causality implies strong-field causality, causality is equivalent to convexity-type conditions on WW, and the dominant and strong energy conditions reduce to simple inequalities for WW and W′W' [15, 18]. Auxiliary-field formulations of this type have also been central to the construction of integrable sigma models [8, 9, 29, 30, 31, 53, 10, 27, 52, 26], and have been related to four-dimensional duality-invariant electrodynamics and to root-TTˉT\bar T flows in [28, 32, 33]. These results suggest that the one-variable potential should be the natural object on which TTˉT\bar T-like flows act.

Both descriptions of causal self-dual electrodynamics encode the theory in a function of a single variable, while a TTˉT\bar T-like flow is expressed in terms of the stress tensor, which depends on the fields. For two-parameter families involving an irrelevant flow and a marginal ModMax flow, one therefore needs a precise dictionary between (i) the flow of the single-variable function, (ii) the flow of the Lagrangian in terms of the stress-tensor invariants, X=TμνTμνX=T_{\mu\nu}T^{\mu\nu} and Y=TμμTννY=T^\mu{}_\mu T^\nu{}_\nu, and (iii) the physical requirements of causality and positivity of energy. Stress-tensor flows of causal self-dual theories have previously been studied in the CH language [34, 22, 33]. The classification of causal NED by φ\varphi-parity and irrelevant deformations was initiated in [36], and the Courant–Hilbert counterpart of the present analysis will be presented in [51]. In Ref. [54], we derive an exact nonperturbative TTˉT\bar{T}-like flow for the class of one-scale, self-dual nonlinear electrodynamics theories admitting a Courant-Hilbert representation without γ\gamma-coupling. Here we develop the corresponding dictionary in the auxiliary-field formalism and use it to study the reconstruction of the potential, the Hamiltonian formulation, characteristic cones, and energy conditions.

In this paper, we consider the two-parameter family of potentials

W(φ;λ,γ)=1λ w(φ−γ),\begin{equation} W(\varphi;\lambda,\gamma)=\frac{1}{\lambda}\,w(\varphi-\gamma), \tag*{(2)} \end{equation}

with a marginal (ModMax) parameter γ≥0\gamma\ge0 and an irrelevant coupling λ\lambda. Our main results are the following.

  • Auxiliary dictionary and flow equations. We map the CH data (ℓ,τ)(\ell,\tau) to the auxiliary potential ww (§ 2), express causality as w′′>w′w''>w' for λ>0\lambda>0, and show that the stress-tensor invariants and the derivatives ∂λL=−Tμμ/(4λ)\partial_\lambda\mathcal L=-T^\mu{}_\mu/(4\lambda) and ∂γL=144X−Y\partial_\gamma\mathcal L=\tfrac14\sqrt{4X-Y} are determined by ww and w′w'.

  • Exact TTˉT\bar T-like flow and reconstruction. The irrelevant flow is ∂λL=X F(r)\partial_\lambda\mathcal L=X\,F(r) with r=−Tμμ/4X−Y=w/w′r=-T^\mu{}_\mu/\sqrt{4X-Y}=w/w'. We give the parametric formula for FF and the inverse reconstruction of ww and ψ\psi from FF, rewrite causality as the condition D(r)>1/(1−r)D(r)>1/(1-r) and derive the bound r<1−e−ψr<1-e^{-\psi} (§ 3).

  • Global structure. Causality ensures a unique auxiliary solution but not a single-valued kernel; the latter is equivalent to w′2−ww′′>0w'^2-ww''>0. We characterise the obstruction and rewrite the flow in terms of the principal invariant ρ=Y/X\rho=\sqrt{Y/X} (§ 4).

  • φ\varphi-Parity and analyticity. We separate vacuum-centred reflection, full φ\varphi-parity and analyticity of L\mathcal L in S,P2S,P^2, show that they coincide only for γ=0\gamma=0, and identify the first odd kernel coefficient F′(0)F'(0) as the leading obstruction to analyticity (§ 5).

  • Linear and Hamiltonian formulations. We give the linear (angular, complex and two-real-auxiliary) Lagrangians, the auxiliary Hamiltonian and its Courant–Hilbert data, and prove ∂λH=−X F(r)\partial_\lambda\mathcal H=-X\,F(r) and ∂γH=−144X−Y\partial_\gamma\mathcal H=-\tfrac14\sqrt{4X-Y} (§ 6).

  • Duality, characteristic cones and energy conditions. We show that Born’s Legendre self-duality holds for every reconstructed theory, determine the two birefringence coefficients A±A_\pm in terms of the kernel, show that constant A−/A+A_-/A_+ singles out generalized Born–Infeld theory, and prove that r=(ε−p⊥)/(ε+p⊥)r=(\varepsilon-p_\perp)/(\varepsilon+p_\perp) (§ 7).

  • Exact models. We obtain exact TTˉT\bar T-like flow for generalized Born–Infeld, logarithmic, qq-deformed and no-τ\tau-maximum electrodynamics (§ 8).

The paper is organised as follows. In § 2 we set up the auxiliary-field description and its relation to the Courant–Hilbert description. § 3 derives the exact TTˉT\bar T-like flow and the reconstruction formulas; § 4 studies the global properties of the kernel; § 5 treats φ\varphi-parity and analyticity. § 6 contains the linear and Hamiltonian formulations, § 7 the duality, characteristic cones and energy conditions, and § 8 the explicit models. We conclude in § 9.

Auxiliary fields and the Courant–Hilbert representation

We relate the auxiliary-field formulation of nonlinear electrodynamics to the Courant–Hilbert (CH) description, in which the theory is encoded by a function ℓ(τ)\ell(\tau). The CH causality conditions are

ℓ˙(τ)≥1,ℓ¨(τ)≥0,\begin{equation} \dot\ell(\tau)\geq1,\qquad \ddot\ell(\tau)\geq0, \tag*{(3)} \end{equation}

where dots denote derivatives with respect to τ\tau.

Exponential variables and the Legendre map

We define two exponential variables,

y=eφ=eγ+ψ,y^=eψ,y=eγy^.\begin{equation} y=e^\varphi=e^{\gamma+\psi},\qquad \hat y=e^\psi,\qquad y=e^\gamma\hat y. \tag*{(4)} \end{equation}

The corresponding potential functions are

Ω(y)=1λw(log⁡y−γ),Ω(y^)=1λw(log⁡y^).\begin{equation} \Omega(y)=\frac1\lambda w(\log y-\gamma),\qquad \Omega(\hat y)=\frac1\lambda w(\log\hat y). \tag*{(5)} \end{equation}

In each expression, the derivative of Ω\Omega is taken with respect to its displayed argument. The two forms encode the same function ww, with the shift by γ\gamma accounted for by y=eγy^y=e^\gamma\hat y.

The Legendre relations are

ℓ(τ)=τy−Ω(y),τ=dΩ(y)dy,ℓ˙=y.\begin{equation} \ell(\tau)=\tau y-\Omega(y),\qquad \tau=\frac{d\Omega(y)}{dy},\qquad \dot\ell=y. \tag*{(6)} \end{equation}

Substituting y=eγ+ψy=e^{\gamma+\psi} gives

τ(ψ)=e−(γ+ψ)λw′(ψ),ℓ(τ(ψ))=w′(ψ)−w(ψ)λ,\begin{equation} \tau(\psi)=\frac{e^{-(\gamma+\psi)}}{\lambda}w'(\psi),\qquad \ell\bigl(\tau(\psi)\bigr)=\frac{w'(\psi)-w(\psi)}{\lambda}, \tag*{(7)} \end{equation}

where primes on ww denote derivatives with respect to ψ\psi. Hence

τℓ˙=w′(ψ)λ,ℓ−τℓ˙=−w(ψ)λ.\begin{equation} \tau\dot\ell=\frac{w'(\psi)}{\lambda},\qquad \ell-\tau\dot\ell=-\frac{w(\psi)}{\lambda}. \tag*{(8)} \end{equation}

Equivalently,

λΩ(y)=w(ψ),λΩ(y^)=w(ψ),\begin{equation} \lambda\Omega(y)=w(\psi),\qquad \lambda\Omega(\hat y)=w(\psi), \tag*{(9)} \end{equation}

and

λydΩ(y)dy=w′(ψ),λy^dΩ(y^)dy^=w′(ψ).\begin{equation} \lambda y\frac{d\Omega(y)}{dy}=w'(\psi),\qquad \lambda\hat y\frac{d\Omega(\hat y)}{d\hat y}=w'(\psi). \tag*{(10)} \end{equation}

Causality and regularity

Differentiating Ω(y)\Omega(y) twice gives

d2Ω(y)dy2=w′′(ψ)−w′(ψ)λy2.\begin{equation} \frac{d^2\Omega(y)}{dy^2} =\frac{w''(\psi)-w'(\psi)}{\lambda y^2}. \tag*{(11)} \end{equation}

Since dτ/dy=d2Ω/dy2d\tau/dy=d^2\Omega/dy^2 and ℓ˙=y\dot\ell=y, we obtain

dτdψ=e−(γ+ψ)λ[w′′(ψ)−w′(ψ)],ℓ¨=λe2(γ+ψ)w′′(ψ)−w′(ψ).\begin{equation} \frac{d\tau}{d\psi} =\frac{e^{-(\gamma+\psi)}}{\lambda} \bigl[w''(\psi)-w'(\psi)\bigr],\qquad \ddot\ell=\frac{\lambda e^{2(\gamma+\psi)}} {w''(\psi)-w'(\psi)}. \tag*{(12)} \end{equation}

The condition ℓ˙≥1\dot\ell\geq1 follows from ℓ˙=eγ+ψ\dot\ell=e^{\gamma+\psi} on the domain γ,ψ≥0\gamma,\psi\geq0. On a regular branch with ℓ¨>0\ddot\ell>0, the remaining causality condition requires

{w′′(ψ)>w′(ψ),λ>0,w′′(ψ)<w′(ψ),λ<0.\begin{equation} \begin{cases} w''(\psi)>w'(\psi),&\lambda>0,\\ w''(\psi)<w'(\psi),&\lambda<0. \end{cases} \tag*{(13)} \end{equation}

These strict inequalities ensure local invertibility of the Legendre map. Points where w′′=w′w''=w' are degenerate and require separate treatment.

Stress tensor flows

For S2+P2>0S^2+P^2>0, the stress tensor takes the CH form

Tμν=τℓ˙S2+P2TμνMax+(ℓ−τℓ˙)gμν,\begin{equation} T_{\mu\nu} =\frac{\tau\dot\ell}{\sqrt{S^2+P^2}}T^{\mathrm{Max}}_{\mu\nu} +(\ell-\tau\dot\ell)g_{\mu\nu}, \tag*{(14)} \end{equation}

where the Maxwell stress tensor obeys TMax μμ=0,T^{{\mathrm{Max}}\,\mu}{}_\mu=0, and TμνMaxTMaxμν=4(S2+P2).T^{\mathrm{Max}}_{\mu\nu}T_{\mathrm{Max}}^{\mu\nu} =4(S^2+P^2). Using the CH combinations above, one finds

X=4(w′(ψ)2+w(ψ)2)λ2,Y=16w(ψ)2λ2.\begin{equation} X=\frac{4\bigl(w'(\psi)^2+w(\psi)^2\bigr)}{\lambda^2},\qquad Y=\frac{16w(\psi)^2}{\lambda^2}. \tag*{(15)} \end{equation}

In particular,

4X−Y=16w′(ψ)2λ2.\begin{equation} 4X-Y=\frac{16w'(\psi)^2}{\lambda^2}. \tag*{(16)} \end{equation}

Evaluating the coupling derivatives on the auxiliary-field extremum leaves the explicit dependence on λ\lambda and γ\gamma:

∂λL=w(ψ)λ2,∂γL=w′(ψ)λ.\begin{equation} \partial_\lambda\mathcal L=\frac{w(\psi)}{\lambda^2},\qquad \partial_\gamma\mathcal L=\frac{w'(\psi)}{\lambda}. \tag*{(17)} \end{equation}

On the branch where w′(ψ)/λ>0w'(\psi)/\lambda>0, the marginal root-TTˉT\bar T flow becomes

∂γL=12X−14Y.\begin{equation} \partial_\gamma\mathcal L =\frac12\sqrt{X-\frac14Y}. \tag*{(18)} \end{equation}

The positive square-root branch is an additional branch choice; it does not follow from ℓ˙≥1\dot\ell\geq1 alone.

The coupling λ\lambda rescales the potential, whereas γ\gamma shifts its argument. Provided the auxiliary solution is smooth and remains on the same regular branch, the flows commute:

[∂λ,∂γ]L=0,[∂λ,∂γ]H=0.\begin{equation} [\partial_\lambda,\partial_\gamma]\mathcal L=0,\qquad [\partial_\lambda,\partial_\gamma]\mathcal H=0. \tag*{(19)} \end{equation}

The Hamiltonian statement assumes that H\mathcal H is defined on the same smooth branch.

Exact TTˉT\bar T-like flow and reconstruction of the potential

The trace relation alone does not give an autonomous equation for the irrelevant flow, because the flow also depends explicitly on the coupling λ\lambda. We therefore seek to express ∂λL\partial_\lambda\mathcal L solely in terms of the stress-tensor invariants XX and YY. From (15) and (18), the trace and the square root 4X−Y\sqrt{4X-Y} are proportional to w(ψ)w(\psi) and w′(ψ)w'(\psi), respectively. Their ratio is consequently independent of the overall normalization by λ\lambda and of the shift by γ\gamma. This motivates the dimensionless ratio

r=−Tμμ4X−Y=w(ψ)w′(ψ),\begin{equation} r=-\frac{T^\mu{}_\mu}{\sqrt{4X-Y}} =\frac{w(\psi)}{w'(\psi)}, \tag*{(20)} \end{equation}

where the last equality uses the branch on which 4X−Y=4w′(ψ)/λ>0\sqrt{4X-Y}=4w'(\psi)/\lambda>0.

Since rr depends only on the position ψ\psi along the auxiliary potential, and

∂λL=w(ψ)λ2=X w(ψ)4[w′(ψ)2+w(ψ)2],\begin{equation} \partial_\lambda\mathcal L =\frac{w(\psi)}{\lambda^2} =X\,\frac{w(\psi)} {4\bigl[w'(\psi)^2+w(\psi)^2\bigr]}, \tag*{(21)} \end{equation}

the irrelevant exact TTˉT\bar T-like flow takes the autonomous form

∂λL=X F(r),\begin{equation} \partial_\lambda\mathcal L=X\,F(r), \tag*{(22)} \end{equation}

with the exact parametric closed form kernel

F(r(ψ))=w(ψ)4[w′(ψ)2+w(ψ)2].\begin{equation} F\bigl(r(\psi)\bigr) =\frac{w(\psi)} {4\bigl[w'(\psi)^2+w(\psi)^2\bigr]}. \tag*{(23)} \end{equation}

Using w=rw′w=rw', this can equivalently be written as

F(r(ψ))=r(ψ)4w′(ψ)[1+r(ψ)2].\begin{equation} F\bigl(r(\psi)\bigr) =\frac{r(\psi)} {4w'(\psi)\bigl[1+r(\psi)^2\bigr]}. \tag*{(24)} \end{equation}

The kernel is a property of the chosen theory and contains no explicit dependence on λ\lambda or γ\gamma; the coupling dependence enters through the stress-tensor invariant XX and the argument rr.

Reconstructing the auxiliary potential

The kernel also determines the auxiliary potential. Suppose that r(ψ)r(\psi) is locally invertible, so that ψ=ψ(r)\psi=\psi(r). Combining w=rw′w=rw' with (23) gives

w(ψ(r))=r24(1+r2)F(r),w′(ψ(r))=r4(1+r2)F(r).\begin{equation} w\bigl(\psi(r)\bigr) =\frac{r^2}{4(1+r^2)F(r)}, \qquad w'\bigl(\psi(r)\bigr) =\frac{r}{4(1+r^2)F(r)}. \tag*{(25)} \end{equation}

Taking the logarithmic derivative of the first expression yields

ddrlog⁡w=2r−2r1+r2−F′(r)F(r).\begin{equation} \frac{d}{dr}\log w =\frac{2}{r}-\frac{2r}{1+r^2} -\frac{F'(r)}{F(r)}. \tag*{(26)} \end{equation}

Since dw/dψ=w′=w/rdw/d\psi=w'=w/r, it follows that

D(r)≡dψdr=21+r2−rF′(r)F(r).\begin{equation} D(r)\equiv\frac{d\psi}{dr} =\frac{2}{1+r^2} -r\frac{F'(r)}{F(r)}. \tag*{(27)} \end{equation}

Integrating and imposing the vacuum condition ψ(0)=0\psi(0)=0 gives

ψ(r)=2arctan⁡r−∫0rt F′(t)F(t) dt.\begin{equation} \psi(r)=2\arctan r -\int_0^r t\,\frac{F'(t)}{F(t)}\,dt. \tag*{(28)} \end{equation}

Thus, wherever r(ψ)r(\psi) is invertible and FF is regular, the kernel together with the vacuum condition determines ψ(r)\psi(r) and w(r)w(r), and hence reconstructs the potential parametrically.

The vacuum behavior fixes the normalization locally. Assuming that the potential starts at quadratic order, one has

r(ψ)=ψ2+O(ψ2),D(0)=2,\begin{equation} r(\psi)=\frac{\psi}{2}+\mathcal O(\psi^2), \qquad D(0)=2, \tag*{(29)} \end{equation}

and therefore

w′′(0)=18F(0),F(0)=18w′′(0).\begin{equation} w''(0)=\frac{1}{8F(0)}, \qquad F(0)=\frac{1}{8w''(0)}. \tag*{(30)} \end{equation}

The normalization of λ\lambda is conventional. Under a positive rescaling

λ~=cλ,c>0,\begin{equation} \widetilde\lambda=c\lambda,\qquad c>0, \tag*{(31)} \end{equation}

the same physical family is represented by

w~=cw,F~=Fc.\begin{equation} \widetilde w=cw,\qquad \widetilde F=\frac{F}{c}. \tag*{(32)} \end{equation}

Indeed, w~/λ~=w/λ\widetilde w/\widetilde\lambda=w/\lambda, while r~=r\widetilde r=r and the flow XF(r)X F(r) is unchanged when expressed using λ~\widetilde\lambda as the deformation parameter.

Causality constraints on the kernel

We now translate the CH causality condition into a condition on FF. For the regular causal branch with λ>0\lambda>0, equation (13) requires w′′>w′w''>w'. Differentiating r=w/w′r=w/w' gives

drdψ=1−rw′′w′.\begin{equation} \frac{dr}{d\psi}=1-r\frac{w''}{w'}. \tag*{(33)} \end{equation}

Using D=dψ/drD=d\psi/dr, we obtain

w′′−w′w′=(1−r)D−1rD.\begin{equation} \frac{w''-w'}{w'} =\frac{(1-r)D-1}{rD}. \tag*{(34)} \end{equation}

On the increasing branch connected continuously to the vacuum, D>0D>0. For 0<r<10<r<1, the sign of w′′−w′w''-w' is therefore the sign of (1−r)D−1(1-r)D-1. Hence the causality condition is equivalent to

w′′>w′⟺D(r)>11−r.\begin{equation} w''>w' \quad\Longleftrightarrow\quad D(r)>\frac{1}{1-r}. \tag*{(35)} \end{equation}

Substituting (27), we find the differential inequality

21+r2−rF′(r)F(r)>11−r.\begin{equation} \frac{2}{1+r^2} -r\frac{F'(r)}{F(r)} >\frac{1}{1-r}. \tag*{(36)} \end{equation}

On the regular increasing branch with 0<r<10<r<1, this is the necessary and sufficient kernel condition for w′′>w′w''>w' and hence for CH causality when λ>0\lambda>0. For λ<0\lambda<0, the CH condition instead requires w′′<w′w''<w', so the inequality must be reversed with the corresponding branch assumptions.

The interval 0<r<10<r<1 follows from causality and the vacuum normalization w(0)=w′(0)=0w(0)=w'(0)=0. In fact,

ddψ(w′−w)=w′′−w′>0\begin{equation} \frac{d}{d\psi}\bigl(w'-w\bigr)=w''-w'>0 \tag*{(37)} \end{equation}

implies

0<w(ψ)<w′(ψ),ψ>0,\begin{equation} 0<w(\psi)<w'(\psi),\qquad \psi>0, \tag*{(38)} \end{equation}

and therefore 0<r(ψ)<10<r(\psi)<1. A stronger bound follows because

ddψ(e−ψw′(ψ))=e−ψ(w′′−w′)>0.\begin{equation} \frac{d}{d\psi}\bigl(e^{-\psi}w'(\psi)\bigr) =e^{-\psi}\bigl(w''-w'\bigr)>0. \tag*{(39)} \end{equation}

For 0≤t<ψ0\leq t<\psi, this monotonicity gives w′(t)<et−ψw′(ψ)w'(t)<e^{t-\psi}w'(\psi). Integrating from 00 to ψ\psi and using w(0)=0w(0)=0, we obtain

w(ψ)=∫0ψw′(t) dt<w′(ψ)∫0ψet−ψ dt.\begin{equation} w(\psi)=\int_0^\psi w'(t)\,dt <w'(\psi)\int_0^\psi e^{t-\psi}\,dt. \tag*{(40)} \end{equation}

Consequently,

r(ψ)<1−e−ψ,ψ(r)>−log⁡(1−r).\begin{equation} r(\psi)<1-e^{-\psi}, \qquad \psi(r)>-\log(1-r). \tag*{(41)} \end{equation}

This bound constrains the ratio along any causal flow and can also be interpreted as a restriction on the allowed stress-tensor configurations.

Global structure: single-valuedness and principal invariant

The reconstruction of § 3 assumed that r(ψ)r(\psi) is locally invertible. Whether the closed form kernel F(r)F(r) is a single function on the whole physical branch is a global question, to which causality alone does not provide the answer. Causality ensures a unique auxiliary-field solution for each admissible electromagnetic configuration, since it makes Ω′′>0\Omega''>0; it does not ensure that r=w/w′r=w/w' is globally injective.

The local reconstruction in § 3 treats rr as a coordinate on the auxiliary-field branch. For the kernel to be a globally defined function of rr, distinct values of ψ\psi must not map to the same value of rr. This requirement is logically independent of the local CH causality condition.

Differentiating r=w/w′r=w/w' gave dr/dψ=1−rw′′/w′dr/d\psi=1-rw''/w'. A regular global kernel exists on a physical interval provided w′2−ww′′>0w'^2-ww''>0 throughout that interval, i.e. precisely when drdψ>0.\frac{dr}{d\psi}>0. Under causality and this additional condition one has w′′/w′>1w''/w'>1 together with rw′′/w′<1rw''/w'<1, that is

1<w′′w′<1r,\begin{equation} 1<\frac{w''}{w'}<\frac{1}{r}, \tag*{(42)} \end{equation}

and the kernel is globally single-valued on the image of the physical interval. Conversely, let FF be positive on 0<r<r∗0<r<r_* with r∗<1r_*<1, real analytic near r=0r=0 (by which we mean the existence of a Taylor expansion), with F(0)>0F(0)>0, and sufficiently differentiable elsewhere. If D(r)>1/(1−r)D(r)>1/(1-r) for 0<r<r∗0<r<r_*, the reconstruction formulas define a unique causal potential on the corresponding ψ\psi-interval.

Global single-valuedness is a substantive restriction. Since w′′>w′>0w''>w'>0 the function w′w' is strictly increasing, so if two distinct points had the same positive ratio rr their closed form kernel values

F=r4w′(1+r2)\begin{equation} F=\frac{r}{4w'(1+r^2)} \tag*{(43)} \end{equation}

would differ, and this causal theory would have no single globally defined autonomous kernel of the proposed form over its entire physical branch. More precisely, at a point where dr/dψ=0dr/d\psi=0,

dFdψ=−F w′′w′≠0,\begin{equation} \frac{dF}{d\psi}=-F\,\frac{w''}{w'}\neq0, \tag*{(44)} \end{equation}

whereas for a differentiable function F(r)F(r) one would have dF/dψ=F′(r) dr/dψ=0dF/d\psi=F'(r)\,dr/d\psi=0. Hence a differentiable function F(r)F(r) cannot pass regularly through that point. The obstruction is a failure of the stress-tensor ratio to provide a coordinate on the theory.

The ratio rr is useful for reconstructing the potential, but it is not itself written solely in terms of the two polynomial stress-tensor invariants without a square root. We can instead use a dimensionless invariant ratio directly observable from XX and YY. It is convenient to replace rr by the physical principal invariant ρ=Y/X\rho=\sqrt{Y/X}, which depends only on the stress-tensor invariants XX and YY. On the causal branch one finds the relations

ρ=2r1+r2,r=ρ4−ρ2.\begin{equation} \rho=\frac{2r}{\sqrt{1+r^2}}, \qquad r=\frac{\rho}{\sqrt{4-\rho^2}}. \tag*{(45)} \end{equation}

To distinguish the two coordinates we write F(ρ)=F(ρ/4−ρ2)\mathcal{F}(\rho) =F(\rho/\sqrt{4-\rho^2}), and the exact autonomous flow reads

∂λL=X F(ρ).\begin{equation} \partial_\lambda\mathcal{L} =X\,\mathcal{F}(\rho). \tag*{(46)} \end{equation}

The causal energy bound r<1r<1 translates into 0<ρ<20<\rho<\sqrt{2}, away from degenerate zero-invariant configurations; this is the image of the physical branch in the space of stress-tensor invariants. If F\mathcal F is analytic near the vacuum (again in the sense of a Taylor expansion),

F(ρ)=∑m=0∞cmρm,\begin{equation} \mathcal{F}(\rho)=\sum_{m=0}^{\infty}c_m\rho^m, \tag*{(47)} \end{equation}

then

∂λL=∑m=0∞cm Ym/2X1−m/2.\begin{equation} \partial_\lambda\mathcal{L} =\sum_{m=0}^{\infty} c_m\,Y^{m/2}X^{1-m/2}. \tag*{(48)} \end{equation}

Odd powers of ρ\rho generate half-integer powers of YY. Evenness of the kernel removes these half-integer structures, but does not by itself make the operator polynomial in XX and YY: for example, a ρ4\rho^4 term produces Y2/XY^2/X. This shows that analyticity in the electromagnetic invariants and analyticity as a function of independently varied stress-tensor invariants are different properties.

The two notions of “even” that appear above are equivalent at the level of analytic germs at the vacuum:

F(−r)=F(r)⟺w(−ψ)=w(ψ).\begin{equation} F(-r)=F(r) \quad\Longleftrightarrow\quad w(-\psi)=w(\psi). \tag*{(49)} \end{equation}

Indeed, if ww is even then w′w' is odd and r=w/w′r=w/w' is odd, so that the parametric kernel is even. Conversely, if FF is even then D(r)D(r) is even, ψ(r)\psi(r) is odd and the reconstructed expression

w(r)=r24(1+r2)F(r)\begin{equation} w(r)=\frac{r^2}{4(1+r^2)F(r)} \tag*{(50)} \end{equation}

is even. Here negative rr denotes the signed analytic continuation of w/w′w/w'; it is not a negative value of the physical principal root ρ\rho.

Causality gives a unique auxiliary solution but not an injective r(ψ)r(\psi). A globally single-valued autonomous kernel exists exactly when dr/dψ>0dr/d\psi>0, equivalently w′2−ww′′>0w'^2-ww''>0; at a zero of dr/dψdr/d\psi no differentiable kernel exists. In terms of the principal invariant ρ=Y/X\rho=\sqrt{Y/X} the flow is ∂λL=X F(ρ)\partial_\lambda\mathcal L=X\,\mathcal F(\rho) with ρ<2\rho<\sqrt2. For analytic germs, evenness of FF and of ww are equivalent, while odd powers of ρ\rho give half-integer powers of YY.

φ\varphi-Parity and analyticity

The statement “FF is even” is easy to confuse with several different symmetry statements. In this section we separate three of them: reflection about the selected vacuum, full φ\varphi-parity under sign reversal of the square-root invariant, and analyticity of the Lagrangian in the electromagnetic invariants SS and P2P^2. The distinction is governed by the marginal parameter γ\gamma, which shifts the vacuum of the auxiliary field away from φ=0\varphi=0. Let wc(ψ)=w(−ψ)w^{\mathrm c}(\psi)=w(-\psi). Then vacuum-centred reflection acts on the data as

(w,γ)⟼(wc,−γ),rc(ψ)=−r(−ψ).\begin{equation} (w,\gamma) \longmapsto \bigl(w^{\mathrm c},-\gamma\bigr), \qquad r^{\mathrm c}(\psi)=-r(-\psi). \tag*{(51)} \end{equation}

The corresponding analytic kernel satisfies Fc(r)=F(−r)F^{\mathrm c}(r)=F(-r). Thus vacuum-centred reflection reverses all odd Taylor coefficients of the kernel.

Full φ\varphi-parity under sign reversal of the square-root invariant instead acts on the unshifted potential, Wc(φ)=W(−φ)W^{\mathrm c}(\varphi)=W(-\varphi).

An even w(ψ)w(\psi) expresses symmetry about the selected vacuum φ=γ\varphi=\gamma. It expresses invariance of the vacuum-connected theory under full φ\varphi-parity only when γ=0\gamma=0. For γ>0\gamma>0, the full φ\varphi-parity dual has a ModMax vacuum with parameter −γ-\gamma, which violates weak-field causality, whereas for γ=0\gamma=0 both members of a φ\varphi-parity pair can be causal; this does not require that the pair coincide. Evenness of ww, i.e. symmetry of the potential about its vacuum, and invariance of the theory under full φ\varphi-parity are therefore two separate concepts that only coincide at γ=0\gamma=0.

The distinction between vacuum-centred reflection and full φ\varphi-parity becomes especially transparent in the weak-field expansion. The leading term determines whether nonanalytic dependence on the electromagnetic invariants is already present before the irrelevant deformation is switched on.

To see how this reflects in the Lagrangian we expand around the vacuum. The weak-field limit of the Lagrangian is

Lweak=Scosh⁡γ+S2+P2sinh⁡γ.\begin{equation} \mathcal{L}_{\mathrm{weak}} =S\cosh\gamma+\sqrt{S^2+P^2}\sinh\gamma. \tag*{(52)} \end{equation}

Define the combination

Jγ=Ssinh⁡γ+S2+P2cosh⁡γ=eγV+e−γU.\begin{equation} J_\gamma =S\sinh\gamma+\sqrt{S^2+P^2}\cosh\gamma =e^\gamma V+e^{-\gamma}U. \tag*{(53)} \end{equation}

Solving the auxiliary equation near the vacuum gives

ψ=λJγw′′(0)+O(λ2),\begin{equation} \psi=\frac{\lambda J_\gamma}{w''(0)}+O(\lambda^2), \tag*{(54)} \end{equation}

and therefore, expanding the Lagrangian to first order in λ\lambda,

L=Lweak+λ2w′′(0)Jγ2+O(λ2).\begin{equation} \mathcal{L} =\mathcal{L}_{\mathrm{weak}} +\frac{\lambda}{2w''(0)}J_\gamma^2 +O(\lambda^2). \tag*{(55)} \end{equation}

For γ>0\gamma>0, nonanalyticity in S,P2S,P^2 is already present in the leading ModMax term, because the square-root invariant appears linearly there; evenness of the irrelevant kernel cannot remove it. For γ=0\gamma=0, the regular analytic reconstruction gives

L analytic in S,P2⟺w even⟺F even.\begin{equation} \mathcal{L}\text{ analytic in }S,P^2 \quad\Longleftrightarrow\quad w\text{ even} \quad\Longleftrightarrow\quad F\text{ even}. \tag*{(56)} \end{equation}

The leading obstruction in the case γ=0\gamma=0 is explicit; carrying the expansion to second order in λ\lambda,

L=S+λ2w′′(0)(S2+P2)+λ22w′′(0)2S(S2+P2)−w′′′(0)λ26w′′(0)3(S2+P2)3/2+O(λ3).\begin{equation} \begin{aligned} \mathcal{L} ={}&S +\frac{\lambda}{2w''(0)}(S^2+P^2) +\frac{\lambda^2}{2w''(0)^2}S(S^2+P^2)\\ &-\frac{w'''(0)\lambda^2}{6w''(0)^3} (S^2+P^2)^{3/2} +O(\lambda^3). \end{aligned} \tag*{(57)} \end{equation}

The first odd kernel coefficient therefore measures the first possible nonanalytic interaction,

F′(0)=−w′′′(0)6w′′(0)2.\begin{equation} F'(0)=-\frac{w'''(0)}{6w''(0)^2}. \tag*{(58)} \end{equation}

Vanishing of F′(0)F'(0) removes this particular obstruction, and full analyticity requires vanishing of every odd coefficient. The results of this section are collected in the following table.

StatementMaxwell vacuum, γ=0\gamma=0ModMax vacuum, γ>0\gamma>0
Even regular kernelPotential symmetric about the vacuumPotential symmetric about the vacuum
Full φ\varphi-parity invarianceEquivalent to evenness of the potential germChanges the selected vacuum to −γ-\gamma
Analyticity in S,P2S,P^2Equivalent to an even analytic kernel germObstructed by the leading ModMax term
Full φ\varphi-parity dualCan also be causalHas an acausal weak-field limit

Table 1. φ\varphi-Parity, vacuum and analyticity statements for the reconstructed class.

Three notions must be kept apart: evenness of the kernel (symmetry of the potential about its vacuum), invariance under full φ\varphi-parity, and analyticity of the Lagrangian in S,P2S,P^2. For γ=0\gamma=0 they are equivalent for analytic germs; for γ>0\gamma>0 the full φ\varphi-parity dual acquires the vacuum −γ-\gamma and is acausal, and the ModMax term itself obstructs analyticity. The first odd coefficient F′(0)=−w′′′(0)/(6w′′(0)2)F'(0)=-w'''(0)/(6w''(0)^2) controls the leading nonanalytic term (S2+P2)3/2(S^2+P^2)^{3/2} in the Maxwell-vacuum case.

Linear auxiliary and Hamiltonian formulations

The reconstruction of § 3–§ 5 defines the theory through the potential ww and the auxiliary field φ\varphi. We now show that, for the analytic Maxwell-vacuum class γ=0\gamma=0, w(−ψ)=w(ψ)w(-\psi)=w(\psi), the same data admits formulations in which the Lagrangian is linear in SS and PP, and that the Hamiltonian counterpart inherits the flow equation of § 3 with the opposite sign.

For the Maxwell-vacuum case, evenness of ww permits the sign ambiguity generated when the angular auxiliary field is eliminated to be absorbed by reflection of ψ\psi. The resulting formulations are useful because the electromagnetic fields enter linearly, while all nonlinear information is carried by the auxiliary potential.

Introduce an angular auxiliary field ϑ\vartheta in addition to ψ\psi:

L=(cosh⁡ψ+sinh⁡ψcos⁡ϑ)S+sinh⁡ψsin⁡ϑ P−w(ψ)λ.\begin{equation} \mathcal L =\left(\cosh\psi+\sinh\psi\cos\vartheta\right)S +\sinh\psi\sin\vartheta\,P -\frac{w(\psi)}{\lambda}. \tag*{(59)} \end{equation}

Eliminating ϑ\vartheta gives

L=Scosh⁡ψ±S2+P2sinh⁡ψ−w(ψ)λ.\begin{equation} \mathcal L =S\cosh\psi\pm\sqrt{S^2+P^2}\sinh\psi -\frac{w(\psi)}{\lambda}. \tag*{(60)} \end{equation}

The two signs are equivalent because ww is even. This gives a definite Lagrangian linear in S,PS,P. To obtain a formulation in which ww enters analytically, we use the complex auxiliary field and introduce the transformed potential

z=sinh⁡ψ eiϑ,t=zzˉ,wz(t):=w ⁣(arsinh⁡t).\begin{equation} \begin{aligned} z&=\sinh\psi\,e^{i\vartheta}, \qquad t=z\bar z,\\ w_z(t)&:=w\!\left( \operatorname{arsinh}\sqrt{t} \right). \end{aligned} \tag*{(61)} \end{equation}

Since ww is analytic and even, wzw_z is analytic near t=0t=0, and the formulation becomes

L=[1+zzˉ+ℜz]S+(ℑz)P−wz(zzˉ)λ.\begin{equation} \mathcal L =\left[\sqrt{1+z\bar z}+\Re z\right]S +(\Im z)P -\frac{w_z(z\bar z)}{\lambda}. \tag*{(62)} \end{equation}

The kernel reconstructs this potential parametrically:

t(r)=sinh⁡2ψ(r),wz(t(r))=r24(1+r2)F(r).\begin{equation} t(r)=\sinh^2\psi(r), \qquad w_z\bigl(t(r)\bigr) =\frac{r^2}{4(1+r^2)F(r)}. \tag*{(63)} \end{equation}

An equivalent formulation uses two independent real auxiliaries a>0a>0 and bb,

c(a,b)=12[a+1+b2a],U(c)=w ⁣(arcosh⁡c),\begin{equation} c(a,b) =\frac12\left[a+\frac{1+b^2}{a}\right], \qquad \mathcal U(c) =w\!\left(\operatorname{arcosh}c\right), \tag*{(64)} \end{equation}
a=(cosh⁡ψ+sinh⁡ψcos⁡ϑ),b=sinh⁡ψsin⁡ϑ.\begin{equation} a =\left(\cosh\psi+\sinh\psi\cos\vartheta\right),\qquad b=\sinh\psi\sin\vartheta. \tag*{(65)} \end{equation}

so that

L=aS+bP−U(c(a,b))λ.\begin{equation} \mathcal L =aS+bP -\frac{\mathcal U(c(a,b))}{\lambda}. \tag*{(66)} \end{equation}

This is the form in which duality is most transparent, as we see in § 7. Its Hamiltonian counterpart is

H=∣D−bB∣22a+a2∣B∣2+U(c(a,b))λ.\begin{equation} \mathcal H =\frac{\lvert\mathbf D-b\mathbf B\rvert^2}{2a} +\frac a2\lvert\mathbf B\rvert^2 +\frac{\mathcal U(c(a,b))}{\lambda}. \tag*{(67)} \end{equation}

This construction applies without a sign ambiguity to the full φ\varphi-parity-invariant analytic class.

The preceding linear formulation is tailored to the analytic, φ\varphi-parity-invariant class. To cover a general potential, including one that is not even, it is more efficient to introduce the auxiliary variable directly in the Hamiltonian description.

To pass to the general (not necessarily even) case we work directly with the Hamiltonian variables. Define

s=12(∣D∣2+∣B∣2),p=∣D×B∣.\begin{equation} s=\frac12\left(|\mathbf D|^2+|\mathbf B|^2\right), \qquad p=|\mathbf D\times\mathbf B|. \tag*{(68)} \end{equation}

and

u=s−s2−p22,v=s+s2−p22.\begin{equation} u=\frac{s-\sqrt{s^2-p^2}}{2}, \qquad v=\frac{s+\sqrt{s^2-p^2}}{2}. \tag*{(69)} \end{equation}

The corresponding auxiliary Hamiltonian is

Haux=eφu+e−φv+w(ψ)λ.\begin{equation} \mathcal H_{\mathrm{aux}} =e^\varphi u+e^{-\varphi}v+\frac{w(\psi)}{\lambda}. \tag*{(70)} \end{equation}

Its field equation is

w′λ=e−φv−eφu.\begin{equation} \frac{w'}{\lambda} =e^{-\varphi}v-e^\varphi u. \tag*{(71)} \end{equation}

The equivalent equation

Ω′(y)=−u+vy2\begin{equation} \Omega'(y)=-u+\frac{v}{y^2} \tag*{(72)} \end{equation}

has at most one physical solution. At a physical solution,

v≥e2φu,\begin{equation} v\geq e^{2\varphi}u, \tag*{(73)} \end{equation}

or equivalently,

s≥pcosh⁡φ.\begin{equation} s\geq p\cosh\varphi. \tag*{(74)} \end{equation}

The Hamiltonian Courant–Hilbert boundary function is reconstructed by

σ(ψ)=eγ+ψw′(ψ)λ,h(σ(ψ))=w′(ψ)+w(ψ)λ.\begin{equation} \sigma(\psi) =\frac{e^{\gamma+\psi}w'(\psi)}{\lambda}, \qquad h\bigl(\sigma(\psi)\bigr) =\frac{w'(\psi)+w(\psi)}{\lambda}. \tag*{(75)} \end{equation}

The exact kernel gives both boundary functions of the CH and Hamiltonian descriptions explicitly:

τ(r)=e−γ−ψ(r)r4λ(1+r2)F(r),σ(r)=eγ+ψ(r)r4λ(1+r2)F(r),\begin{equation} \tau(r) =\frac{e^{-\gamma-\psi(r)}r} {4\lambda(1+r^2)F(r)}, \qquad \sigma(r) =\frac{e^{\gamma+\psi(r)}r} {4\lambda(1+r^2)F(r)}, \tag*{(76)} \end{equation}

and

ℓ(r)=r(1−r)4λ(1+r2)F(r),h(r)=r(1+r)4λ(1+r2)F(r).\begin{equation} \ell(r) =\frac{r(1-r)}{4\lambda(1+r^2)F(r)}, \qquad h(r) =\frac{r(1+r)}{4\lambda(1+r^2)F(r)}. \tag*{(77)} \end{equation}

Their derivatives satisfy

h′=e−φ,h′′=−λe−2φw′′+w′,\begin{equation} h'=e^{-\varphi}, \qquad h''=-\frac{\lambda e^{-2\varphi}}{w''+w'}, \tag*{(78)} \end{equation}

and

h′+2σh′′=e−φw′′−w′w′′+w′.\begin{equation} h'+2\sigma h'' =e^{-\varphi}\frac{w''-w'}{w''+w'}. \tag*{(79)} \end{equation}

Hence the Hamiltonian causality conditions follow directly from the same reconstructed potential: for w′′>w′>0w''>w'>0 one has w′′+w′>0w''+w'>0, so h′+2σh′′h'+2\sigma h'' is positive precisely when w′′>w′w''>w'.

Finally, the coupling derivatives of the Hamiltonian at fixed D,B\mathbf D,\mathbf B are

∂λH=−wλ2=−XF(r),∂γH=−w′λ=−144X−Y\begin{equation} \partial_\lambda\mathcal H =-\frac{w}{\lambda^2} =-X F(r), \qquad \partial_\gamma\mathcal H =-\frac{w'}{\lambda} =-\frac14\sqrt{4X-Y} \tag*{(80)} \end{equation}

The Lagrangian and Hamiltonian coupling derivatives therefore have opposite signs, while using the same stress-tensor invariants and the same kernel.

For γ=0\gamma=0 and even ww the theory has linear formulations with an angle ϑ\vartheta, a complex field zz, or two real auxiliaries (a,b)(a,b), with an analytic potential wz(zzˉ)=U(c(a,b))w_z(z\bar z)=\mathcal U(c(a,b)) reconstructed from FF. The auxiliary Hamiltonian Haux\mathcal H_{\rm aux} has a unique physical solution for every configuration, its CH data (σ,h)(\sigma,h) are given explicitly by the kernel, and the Hamiltonian flow is ∂λH=−X F(r)\partial_\lambda\mathcal H=-X\,F(r), ∂γH=−144X−Y\partial_\gamma\mathcal H=-\tfrac14\sqrt{4X-Y}, i.e. the Lagrangian flow with reversed sign.

Duality, characteristic cones and energy conditions

We now draw the physical consequences of the reconstructed potential. We first show that duality is automatic, then compute the propagation of perturbations on a static background, which is controlled by A±A_\pm, and finally relate the kernel variable rr to the energy density and pressures.

The auxiliary formulation also makes it possible to separate properties that are sometimes conflated. Electromagnetic self-duality is a continuous duality property of the equations, Born–Legendre self-duality is a property of the Hamiltonian under a discrete Legendre transformation, and full φ\varphi-parity is an additional reflection symmetry. The first two need not imply the third.

Every reconstructed theory remains electromagnetically self-dual because the Lagrangian satisfies

LULV=−1,\begin{equation} \mathcal{L}_U \mathcal{L}_V = -1, \tag*{(81)} \end{equation}

with U,V=(S2+P2∓S)/2U,V=(\sqrt{S^2+P^2}\mp S)/2, a property that is inherited from the structure of L\mathcal L as a linear function of SS and S2+P2\sqrt{S^2+P^2} with a duality-invariant potential. Born’s Legendre self-duality follows from the discrete duality rotation

H(D,B)=H(B,−D).\begin{equation} \mathcal{H}(\mathbf{D},\mathbf{B}) = \mathcal{H}(\mathbf{B},-\mathbf{D}). \tag*{(82)} \end{equation}

It therefore holds for the reconstructed theory irrespective of whether FF is even. Within the linear auxiliary formulation, the result can also be seen directly. Under

a~=aa2+b2,b~=−ba2+b2,\begin{equation} \widetilde{a} = \frac{a}{a^2+b^2}, \qquad \widetilde{b} = -\frac{b}{a^2+b^2}, \tag*{(83)} \end{equation}

one has

c(a~,b~)=c(a,b).\begin{equation} c(\widetilde{a},\widetilde{b}) = c(a,b). \tag*{(84)} \end{equation}

The reconstructed interaction potential is therefore invariant under the auxiliary redefinition implementing the covariant Legendre transformation. Electromagnetic self-duality, Born’s Legendre self-duality and full φ\varphi-parity under φ↦−φ\varphi\mapsto-\varphi are distinct statements: the first two hold throughout the reconstructed class, while the third imposes the additional φ\varphi-parity condition of § 5.

Causality can be tested not only through convexity of the CH function, but also through the propagation of small disturbances on a background electromagnetic field. In a frame where the background fields are parallel, the two polarizations have dispersion relations controlled by two coefficients, which can be expressed directly in terms of the reconstructed potential.

For plane-wave perturbations about a static background, the two dispersion relations are

ω2=k∥2+A±k⊥2.\begin{equation} \omega^2 = k_\parallel^2+A_\pm k_\perp^2. \tag*{(85)} \end{equation}

and the auxiliary-field expressions of the two birefringence coefficients are

A−=e−2φ,A+=w′′−w′w′′+w′.\begin{equation} A_-=e^{-2\varphi}, \qquad A_+=\frac{w''-w'}{w''+w'}. \tag*{(86)} \end{equation}

Using w′′/w′=(D−1)/(rD)w''/w'=(D-1)/(rD) the exact kernel reconstructs them as

A−(r)=e−2γ−2ψ(r),\begin{equation} A_-(r)=e^{-2\gamma-2\psi(r)}, \tag*{(87)} \end{equation}

and

A+(r)=(1−r)D(r)−1(1+r)D(r)−1.\begin{equation} A_+(r) = \frac{(1-r)D(r)-1}{(1+r)D(r)-1}. \tag*{(88)} \end{equation}

The kernel causality inequality implies

0<A+≤1,0<A−≤1.\begin{equation} 0<A_+\leq 1, \qquad 0<A_-\leq 1. \tag*{(89)} \end{equation}

At the weak-field limit,

A+(0)=1,A−(0)=e−2γ.\begin{equation} A_+(0)=1, \qquad A_-(0)=e^{-2\gamma}. \tag*{(90)} \end{equation}

Thus FF determines one characteristic coefficient directly and the other after reconstruction of ψ\psi, with γ\gamma specifying its vacuum value. The ratio of the two coefficients is

A−A+=e−2γ−2ψ(r)(1+r)D(r)−1(1−r)D(r)−1.\begin{equation} \frac{A_-}{A_+} = e^{-2\gamma-2\psi(r)} \frac{(1+r)D(r)-1}{(1-r)D(r)-1}. \tag*{(91)} \end{equation}

It is natural to ask when this ratio stays at its vacuum value along the whole flow, i.e. when the two cones are related by a constant factor. Requiring

A−A+=e−2γ.\begin{equation} \frac{A_-}{A_+}=e^{-2\gamma}. \tag*{(92)} \end{equation}

for all ψ\psi is equivalent to

w′′w′=coth⁡ψ.\begin{equation} \frac{w''}{w'}=\coth\psi. \tag*{(93)} \end{equation}

The solution is

w(ψ)=w′′(0)(cosh⁡ψ−1).\begin{equation} w(\psi)=w''(0)(\cosh\psi-1). \tag*{(94)} \end{equation}

Consequently,

F(r)=1−r28w′′(0)(1+r2),\begin{equation} F(r)=\frac{1-r^2}{8w''(0)(1+r^2)}, \tag*{(95)} \end{equation}

or

F(ρ)=18w′′(0)−ρ216w′′(0).\begin{equation} \mathcal{F}(\rho) = \frac{1}{8w''(0)}-\frac{\rho^2}{16w''(0)}. \tag*{(96)} \end{equation}

which is generalized Born–Infeld electrodynamics (§ 8): constant ratio A−/A+A_-/A_+ singles out the generalized Born–Infeld family.

The same auxiliary ratio also has a direct interpretation in terms of the energy density and principal pressures. This interpretation makes the range 0<r<10<r<1 physically transparent and allows the energy conditions to be checked without choosing a specific model.

We now turn to the physical meaning of the kernel variable rr. Causality implies

w′λ>wλ>0(ψ>0).\begin{equation} \frac{w'}{\lambda}>\frac{w}{\lambda}>0 \qquad (\psi>0). \tag*{(97)} \end{equation}

These inequalities yield the dominant and strong energy conditions. To make the statement explicit, take a non-null field and choose a frame in which the electric and magnetic fields are parallel,E⃗=(0,0,E),\vec{E}=(0,0,E), and B⃗=(0,0,B)\vec{B}=(0,0,B). The energy density and the pressures are

ε=w′+wλ,\begin{equation} \varepsilon=\frac{w'+w}{\lambda}, \tag*{(98)} \end{equation}
p⊥=w′−wλ,p∥=−w′+wλ.\begin{equation} p_{\perp}=\frac{w'-w}{\lambda}, \qquad p_{\parallel}=-\frac{w'+w}{\lambda}. \tag*{(99)} \end{equation}

Therefore, in terms of A=w′/λA=w'/\lambda,

ε=A(1+r),p⊥=A(1−r),p∥=−ε.\begin{equation} \varepsilon=A(1+r), \qquad p_{\perp}=A(1-r), \qquad p_{\parallel}=-\varepsilon. \tag*{(100)} \end{equation}

The ratio rr has the invariant physical interpretation

r=ε−p⊥ε+p⊥.\begin{equation} r=\frac{\varepsilon-p_{\perp}} {\varepsilon+p_{\perp}}. \tag*{(101)} \end{equation}

Since 0<r<10<r<1,

ε>0,0<p⊥<ε,p∥<0.\begin{equation} \varepsilon>0, \qquad 0< p_{\perp}<\varepsilon, \qquad p_{\parallel}<0. \tag*{(102)} \end{equation}

Moreover,

ε+p⊥=2A>0,ε+p∥=0,\begin{equation} \varepsilon+p_{\perp}=2A>0, \qquad \varepsilon+p_{\parallel}=0, \tag*{(103)} \end{equation}

and

ε+2p⊥+p∥=2A(1−r)>0.\begin{equation} \varepsilon+2p_{\perp}+p_{\parallel} =2A(1-r)>0. \tag*{(104)} \end{equation}

The null, weak, dominant and strong energy conditions follow. The kernel reconstructs these quantities:

A(r)=r4λ(1+r2)F(r),\begin{equation} A(r)=\frac{r}{4\lambda(1+r^2)F(r)}, \tag*{(105)} \end{equation}
ε(r)=r(1+r)4λ(1+r2)F(r),p⊥(r)=r(1−r)4λ(1+r2)F(r).\begin{equation} \varepsilon(r) =\frac{r(1+r)}{4\lambda(1+r^2)F(r)}, \qquad p_{\perp}(r) =\frac{r(1-r)}{4\lambda(1+r^2)F(r)}. \tag*{(106)} \end{equation}

Finally, the traceless part of the mixed stress tensor, T^μν=Tμν−Tμμ4δμν\hat T^\mu{}_\nu=T^\mu{}_\nu-\tfrac{T^\mu{}_\mu}{4}\delta^\mu{}_\nu, is proportional to the Maxwell one and therefore squares (as a matrix product over the contracted index) to a multiple of the identity,

T^μρ T^ρν=4X−Y16 δμν,\begin{equation} \hat T^\mu{}_\rho\, \hat T^\rho{}_\nu =\frac{4X-Y}{16}\,\delta^\mu{}_\nu, \tag*{(107)} \end{equation}

Hence every algebraic scalar constructed from TμνT^\mu{}_\nu reduces to a function of TμμT^\mu{}_\mu and XX; this is the reason why the flow can be closed on these two invariants.

All reconstructed theories are electromagnetically self-dual and Born–Legendre self-dual, independently of the φ\varphi-parity of FF. The two characteristic coefficients are A−=e−2φA_-=e^{-2\varphi} and A+=(w′′−w′)/(w′′+w′)A_+=(w''-w')/(w''+w'), both reconstructed from the kernel; constancy of A−/A+A_-/A_+ selects generalized Born–Infeld theory, F=(1−r2)/[8w′′(0)(1+r2)]F=(1-r^2)/[8w''(0)(1+r^2)]. Causality gives 0<r<10<r<1, the dominant and strong energy conditions, and r=(ε−p⊥)/(ε+p⊥)r=(\varepsilon-p_\perp)/(\varepsilon+p_\perp).

Exact models

In this section we apply the construction to the standard families of causal self-dual electrodynamics. For each theory we start from the CH boundary function ℓ(τ)\ell(\tau), read off the auxiliary potential by w=λ(τℓ˙−ℓ)w=\lambda(\tau\dot\ell-\ell), check causality through w′′−w′w''-w', invert r=w/w′r=w/w' and insert the result into the kernel formula F=r/[4w′(1+r2)]F=r/[4w'(1+r^2)]. The last step is the only one that requires solving an algebraic equation, and it is where the global single-valuedness criterion of § 4 is needed.

We now illustrate the reconstruction for representative potentials. In each case, the auxiliary potential determines r(ψ)=w/w′r(\psi)=w/w', and inversion gives the kernel F(r)F(r). The invariant kernel F(ρ)\mathcal F(\rho) then follows by the universal change of variable in (46).

Generalized Born–Infeld

For generalized Born–Infeld, start directly from

ℓGBI(τ)=1−1−2eγλτλ.\begin{equation} \ell_{\mathrm{GBI}}(\tau) =\frac{1-\sqrt{1-2e^\gamma\lambda\tau}}{\lambda}. \tag*{(108)} \end{equation}

Its auxiliary potential iswGBI(ψ)=cosh⁡ψ−1w_{\mathrm{GBI}}(\psi)=\cosh\psi-1. Thus

r(ψ)=cosh⁡ψ−1sinh⁡ψ=tanh⁡ψ2,ψ(r)=2arctanh⁡r.\begin{equation} r(\psi)=\frac{\cosh\psi-1}{\sinh\psi} =\tanh\frac{\psi}{2}, \qquad \psi(r)=2\operatorname{arctanh}r. \tag*{(109)} \end{equation}

The exact kernel is

FGBI(r)=1−r28(1+r2),\begin{equation} F_{\mathrm{GBI}}(r) =\frac{1-r^2}{8(1+r^2)}, \tag*{(110)} \end{equation}

and therefore

FGBI(ρ)=18−ρ216.\begin{equation} \mathcal F_{\mathrm{GBI}}(\rho) =\frac18-\frac{\rho^2}{16}. \tag*{(111)} \end{equation}

The complete flow is

∂λLGBI=X FGBI(YX).\begin{equation} \partial_\lambda\mathcal L_{\mathrm{GBI}} =X\,\mathcal F_{\mathrm{GBI}} \left(\sqrt{\frac{Y}{X}}\right). \tag*{(112)} \end{equation}

This is the unique case with constant A−/A+A_-/A_+ found in § 7.

Logarithmic theory

For the logarithmic theory,

ℓlog⁡(τ)=−1λlog⁡(1−eγλτ),\begin{equation} \nonumber \ell_{\log}(\tau) = -\frac{1}{\lambda} \log\left(1-e^\gamma\lambda\tau\right), \end{equation}

the auxiliary potential is

wlog⁡(ψ)=eψ−1−ψ.\begin{equation} \nonumber w_{\log}(\psi)=e^\psi-1-\psi. \end{equation}

Setting s=e−ψs=e^{-\psi},

w′=1−ss,w=1s−1+log⁡s,\begin{equation} \nonumber w'=\frac{1-s}{s}, \qquad w=\frac{1}{s}-1+\log s, \end{equation}

and

r(s)=1+slog⁡s1−s.\begin{equation} \nonumber r(s)=1+\frac{s\log s}{1-s}. \end{equation}

The physical inverse is

s(r)=r−1W−1((r−1)e(r−1)),0<r<1.\begin{equation} \nonumber s(r) =\frac{r-1}{W_{-1}((r-1)e^{(r-1)})}, \qquad 0<r<1. \end{equation}

The lower real Lambert branch supplies the nontrivial physical solution; the vacuum value is obtained by continuity. The exact kernel is

Flog⁡(r)=r s(r)4[1−s(r)](1+r2).\begin{equation} \nonumber F_{\log}(r) =\frac{r\,s(r)} {4[1-s(r)](1+r^2)}. \end{equation}

Since w′′−w′=1w''-w'=1, causality holds throughout the physical auxiliary interval, while the non-even potential accounts for the odd kernel coefficients.

The qq-deformed family.

For the qq-deformed theory

ℓq(τ)=1−(1−eγλτq)qλ,q>0.\begin{equation} \ell_q(\tau) = \frac{ 1-\left(1-\frac{e^\gamma\lambda\tau}{q}\right)^q }{\lambda}, \qquad q>0. \tag*{(113)} \end{equation}

For q≠1q\neq 1,

ℓq′(τ)=eγ(1−eγλτq)q−1.\begin{equation} \ell_q'(\tau) = e^\gamma \left(1-\frac{e^\gamma\lambda\tau}{q}\right)^{q-1}. \tag*{(114)} \end{equation}

Since ℓq′=eγ+ψ\ell_q'=e^{\gamma+\psi},

1−eγλτq=eψ/(q−1).\begin{equation} 1-\frac{e^\gamma\lambda\tau}{q} = e^{\psi/(q-1)}. \tag*{(115)} \end{equation}

Using w=λ(τℓq′−ℓq)w=\lambda(\tau\ell_q'-\ell_q), the auxiliary potential is therefore

wq(ψ)=qeψ+(1−q)eqψ/(q−1)−1.\begin{equation} w_q(\psi)=q e^\psi+(1-q)e^{q\psi/(q-1)}-1. \tag*{(116)} \end{equation}

Its relevant derivatives are

wq′=q(eψ−eqψ/(q−1)),\begin{equation} w_q' = q\left(e^\psi-e^{q\psi/(q-1)}\right), \tag*{(117)} \end{equation}
wq′′−wq′=q1−qeqψ/(q−1),wq′′(0)=q1−q.\begin{equation} w_q''-w_q' = \frac{q}{1-q}e^{q\psi/(q-1)}, \qquad w_q''(0)=\frac{q}{1-q}. \tag*{(118)} \end{equation}

Thus, for γ≥0\gamma \geq 0, the causal class has

0<q<1λ>0,q>1λ<0.\begin{equation} 0 < q < 1\qquad\lambda>0,\qquad q>1 \qquad \lambda<0. \tag*{(119)} \end{equation}

The endpoint q=1q = 1 gives ModMax and a vanishing irrelevant flow. For 0<q<10<q<1, set

s=e−ψ,b=11−q.\begin{equation} s=e^{-\psi},\qquad b=\frac{1}{1-q}. \tag*{(120)} \end{equation}

Then

wq=qs+(1−q)sb−1−1,wq′=q(1−sb)s,\begin{equation} w_q=\frac{q}{s}+(1-q)s^{b-1}-1, \qquad w_q'=\frac{q(1-s^b)}{s}, \tag*{(121)} \end{equation}

and

r=q+(1−q)sb−sq(1−sb).\begin{equation} r=\frac{q+(1-q)s^b-s}{q(1-s^b)}. \tag*{(122)} \end{equation}

Consequently, the physical inverse sq(r)s_q(r) is the solution in 0<s<10<s<1 of

(1−q+qr)s1/(1−q)−s+q(1−r)=0,0<r<1.\begin{equation} \bigl(1-q+qr\bigr)s^{1/(1-q)} -s+q(1-r)=0, \qquad 0<r<1. \tag*{(123)} \end{equation}

The root s=1s=1, which also satisfies this rearranged equation, is excluded for r>0r>0; the vacuum is recovered by continuity. The exact kernel is

Fq(r)=r sq(r)4q[1−sq(r)1/(1−q)](1+r2),Fq(0)=1−q8q.\begin{equation} F_q(r) =\frac{r\,s_q(r)} {4q\left[1-s_q(r)^{1/(1-q)}\right](1+r^2)}, \qquad F_q(0)=\frac{1-q}{8q}. \tag*{(124)} \end{equation}

Accordingly,

∂λLq=X Fq(ρ),Fq(ρ)=Fq(ρ4−ρ2).\begin{equation} \partial_\lambda \mathcal{L}_q =X\,\mathcal{F}_q(\rho), \qquad \mathcal{F}_q(\rho) =F_q\left(\frac{\rho}{\sqrt{4-\rho^2}}\right). \tag*{(125)} \end{equation}

This family has a globally increasing ratio. Indeed, with a=q/(1−q)>0a=q/(1-q)>0,

(wq′)2−wqwq′′=qe(1−a)ψ[eaψ+ae−ψ−(a+1)]>0\begin{equation} (w_q')^2-w_q w_q'' =q e^{(1-a)\psi} \left[e^{a\psi}+a e^{-\psi}-(a+1)\right]>0 \tag*{(126)} \end{equation}

for ψ>0\psi>0. The bracket vanishes at the vacuum and has positive derivative. Hence FqF_q is single-valued throughout 0<r<10< r<1. For q=12q=\tfrac{1}{2}, this reproduces GBI result exactly:

w1/2=cosh⁡ψ−1,F1/2(r)=1−r28(1+r2),F1/2(ρ)=18−ρ216.\begin{equation} w_{1/2}=\cosh\psi-1, \qquad F_{1/2}(r)=\frac{1-r^2}{8(1+r^2)}, \qquad \mathcal{F}_{1/2}(\rho)=\frac{1}{8}-\frac{\rho^2}{16}. \tag*{(127)} \end{equation}

For q=34q=\tfrac34,

w3/4(ψ)=34eψ+14e−3ψ−1,w3/4′′−w3/4′=3e−3ψ.\begin{equation} w_{3/4}(\psi) =\frac34 e^\psi+\frac14 e^{-3\psi}-1, \qquad w_{3/4}''-w_{3/4}'=3e^{-3\psi}. \tag*{(128)} \end{equation}

The inverse equation reduces, after removing the vacuum root, to

s3+s2+s=3(1−r)1+3r.\begin{equation} s^3+s^2+s=\frac{3(1-r)}{1+3r}. \tag*{(129)} \end{equation}

It has a unique physical root s(r)∈(0,1)s(r)\in(0,1).

s(r)=−13−22337+811+3r−81r1+3r+366−8r+3r2(1+3r)23+7+811+3r−81r1+3r+366−8r+3r2(1+3r)23323.\begin{equation} \begin{aligned} s(r)={}&-\frac{1}{3} -\frac{2\sqrt[3]{2}}{ 3\sqrt[3]{ 7+\frac{81}{1+3r}-\frac{81r}{1+3r} +36\sqrt{\frac{6-8r+3r^2}{(1+3r)^2}} } } \\[6pt] &+\frac{ \sqrt[3]{ 7+\frac{81}{1+3r}-\frac{81r}{1+3r} +36\sqrt{\frac{6-8r+3r^2}{(1+3r)^2}} } }{ 3\sqrt[3]{2} }. \end{aligned} \tag*{(130)} \end{equation}

Thus

F3/4(r)=r s(r)3[1−s(r)4](1+r2).\begin{equation} F_{3/4}(r) =\frac{r\,s(r)} {3\bigl[1-s(r)^4\bigr](1+r^2)}. \tag*{(131)} \end{equation}

For q=32q=\tfrac{3}{2} (causal for λ<0\lambda<0),

ℓ3/2(τ)=1−(1−23eγλτ)3/2λ,\begin{equation} \ell_{3/2}(\tau) =\frac{1-\left(1-\frac{2}{3}e^\gamma\lambda\tau\right)^{3/2}} {\lambda}, \tag*{(132)} \end{equation}

whose second derivative is positive for λ<0\lambda<0. Its auxiliary potential is

w3/2(ψ)=32eψ−12e3ψ−1,w3/2′′−w3/2′=−3e3ψ<0.\begin{equation} w_{3/2}(\psi) =\frac{3}{2}e^\psi-\frac{1}{2}e^{3\psi}-1, \qquad w_{3/2}''-w_{3/2}'=-3e^{3\psi}<0. \tag*{(133)} \end{equation}

Writing u=eψ>1u=e^\psi>1,

r=u2+u−23(u2+u)>0,\begin{equation} r=\frac{u^2+u-2}{3(u^2+u)}>0, \tag*{(134)} \end{equation}

with inverse

u(r)=12[3(3−r)1−3r−1].\begin{equation} u(r)=\frac{1}{2} \left[ \sqrt{\frac{3(3-r)}{1-3r}}-1 \right]. \tag*{(135)} \end{equation}

The exact kernel is

F3/2(r)=−(1−3r)2144(1+r2)[3+3(3−r)1−3r],r>0.\begin{equation} F_{3/2}(r) =-\frac{(1-3r)^2}{144(1+r^2)} \left[ 3+\sqrt{\frac{3(3-r)}{1-3r}} \right], \qquad r>0. \tag*{(136)} \end{equation}

No-τ\tau-maximum theory

For the no-τ\tau-maximum theory (−23λNM=λ32-\frac{2}{3}\lambda_{NM}=\lambda_{\frac{3}{2}}),

ℓNM(τ)=23λ[(1+eγλτ)3/2−1],\begin{equation} \ell_{\mathrm{NM}}(\tau) = \frac{2}{3\lambda} \left[ \left(1 + e^{\gamma}\lambda\tau\right)^{3/2} - 1 \right], \tag*{(137)} \end{equation}

the auxiliary potential is (−32wNM=w32-\frac{3}{2}w_{NM}=w_{\frac{3}{2}})

wNM(ψ)=13e3ψ−eψ+23.\begin{equation} w_{\mathrm{NM}}(\psi) = \frac{1}{3}e^{3\psi} - e^{\psi} + \frac{2}{3}. \tag*{(138)} \end{equation}

Let u=eψ>1u=e^\psi> 1. Then (rNM=r32r_{NM}=r_{\frac{3}{2}})

w′=u3−u,r=u2+u−23(u2+u).\begin{equation} w' = u^3-u, \qquad r = \frac{u^2+u-2}{3(u^2+u)}. \tag*{(139)} \end{equation}

The physical inverse is

u(r)=12[3(3−r)1−3r−1],0<r<13.\begin{equation} u(r) = \frac12\left[ \sqrt{\frac{3(3-r)}{1-3r}}-1 \right], \qquad 0< r<\frac13. \tag*{(140)} \end{equation}

The exact kernel becomes (−23FNM=F32-\frac{2}{3}F_{NM}=F_{\frac{3}{2}})

FNM(r)=(1−3r)296(1+r2)[3+3(3−r)1−3r].\begin{equation} F_{\mathrm{NM}}(r) = \frac{(1-3r)^2}{96(1+r^2)} \left[ 3+\sqrt{\frac{3(3-r)}{1-3r}} \right]. \tag*{(141)} \end{equation}

Since

w′′−w′=2e3ψ>0,\begin{equation} w''-w' = 2e^{3\psi}>0, \tag*{(142)} \end{equation}

this theory is causal throughout its physical auxiliary interval.

Theoryw′′−w′w''-w'Physical rr-range
of A+A_+
GBIe−ψe^{-\psi}0<r<10< r<100
Logarithmic110<r<10< r<100
No-τ\tau-maximum2e3ψ2e^{3\psi}0<r<1/30< r<1/31/21/2
q-deformedq1−qe−qψ/(1−q)\dfrac{q}{1-q}e^{-q\psi/(1-q)}
0<r<q−1q0<r<\frac{q-1}{q} (q>1)(q>1)
12q−1\frac{1}{2q-1} (q>1)(q>1)

Table 2. Causal models: w′′−w′w''-w', physical range of rr and strong-field limit of A+A_+.

Generalized Born–Infeld (w=cosh⁡ψ−1w=\cosh\psi-1, F=(1−r2)/[8(1+r2)]F=(1-r^2)/[8(1+r^2)]) is the only member of the list with constant A−/A+A_-/A_+. The logarithmic theory has w′′−w′=1w''-w'=1 and a kernel in terms of the Lambert function; the qq-deformed family (0<q<10<q<1, λ>0\lambda>0, or q>1q>1, λ<0\lambda<0) has a globally increasing ratio r(ψ)r(\psi), so that its kernel is single-valued on 0<r<10<r<1 (for q<1q<1); q=1/2q=1/2 is GBI, q=3/4q=3/4 is solved in radicals, and q=3/2q=3/2 together with its no-τ\tau-maximum partner has the finite range 0<r<1/30<r<1/3. Table 2 summarises w′′−w′w''-w', the rr-range and the strong-field limit of A+A_+.

Discussion

In this paper we have developed the non-perturbative TTˉT\bar T-like flows of causal self-dual electrodynamics in the auxiliary-field formalism. The central structural result is that a pair of commuting flows, an irrelevant one with coupling λ\lambda and a marginal ModMax one with parameter γ\gamma, acts on the single-variable potential W(φ)=λ−1w(φ−γ)W(\varphi)=\lambda^{-1}w(\varphi-\gamma) by rescaling and translation, and that the corresponding flow of the Lagrangian takes the autonomous form ∂λL=X F(r)\partial_\lambda\mathcal L=X\,F(r) and ∂γL=144X−Y\partial_\gamma\mathcal L=\tfrac14\sqrt{4X-Y} in terms of the stress-tensor invariants. The kernel F(r)F(r) and the potential w(ψ)w(\psi) are two descriptions of the same theory, connected by the parametric formulas (23) and (25)–(28); this makes it possible to construct the flow from the potential and, conversely, to build the potential from a prescribed flow.

In the auxiliary formulation of Russo-Townsend [15, 18], weak-field causality implies strong-field causality, and causality is equivalent to a condition on the second derivative of the potential. Our analysis adds that, at the level of the kernel, causality is the first-order inequality D(r)>1/(1−r)D(r)>1/(1-r) and implies 0<r<1−e−ψ0<r<1-e^{-\psi}. The range 0<r<10<r<1 implies the dominant and strong energy conditions, with the physical meaning r=(ε−p⊥)/(ε+p⊥)r=(\varepsilon-p_\perp)/(\varepsilon+p_\perp). The condition for an autonomous, globally single-valued flow, w′2−ww′′>0w'^2-ww''>0, is additional to causality. Theories that violate it are causal but do not have a closed flow equation in the ratio rr alone, since the stress-tensor ratio fails to be a coordinate on the theory. Characterising these theories, and deciding whether the flow can be closed by including further invariants, is a natural open problem.

A subtle but important result is the separation between evenness of the potential about its vacuum, invariance under full φ\varphi-parity, and analyticity in the Lorentz invariants. For γ=0\gamma=0 all three are equivalent for analytic germs, but for γ>0\gamma>0 the ModMax term itself is nonanalytic, and the full φ\varphi-parity dual of a causal theory acquires an acausal vacuum. The first odd kernel coefficient F′(0)F'(0) gives the first nonanalytic interaction, (S2+P2)3/2(S^2+P^2)^{3/2}. Since odd powers of ρ=Y/X\rho=\sqrt{Y/X} produce half-integer powers of YY, while even powers can still produce non-polynomial terms such as Y2/XY^2/X, analyticity of the Lagrangian in S,P2S,P^2 and polynomial dependence of the flow on X,YX,Y are separate properties.

All reconstructed theories are self-dual, the Hamiltonian flow is the Lagrangian flow with a reversed sign, and the characteristic cones A±A_\pm are given explicitly by D(r)D(r) and ψ(r)\psi(r). In particular the two cones coincide up to a constant factor only for generalized Born–Infeld theory, which provides a kernel-level characterisation of this theory.

The exact kernels of generalized Born–Infeld, logarithmic, qq-deformed and no-τ\tau-maximum electrodynamics are summarised in Table 2 and § 8. For q<1q<1 the ratio rr is monotonic on the whole physical branch, while for q>1q>1 and for the no-τ\tau-maximum theory the physical rr-range is 0<r<(q−1)/q0<r<(q-1)/q (resp. 1/31/3 for q=3/2q=3/2) and the strong-field limit of A+A_+ is 1/(2q−1)1/(2q-1) rather than zero.

Several extensions are suggested by these results. Because the auxiliary-field formalism is also the language in which integrable sigma models and root-TTˉT\bar T flows are related to four-dimensional duality-invariant electrodynamics [28, 32, 33]. Interestingly, the reconstruction and single-valued measures obtained here are directly equivalent to those in two dimensions [55]. The same questions can be asked for causal chiral two-form electrodynamics [14, 56] and for higher-derivative extensions of ModMax [20]. The characteristic-curve viewpoint on TTˉT\bar T flows and self-dual electrodynamics [40, 39] and the applications of the CH construction to black-hole thermodynamics [41] offer further settings in which the single-valuedness criterion could be tested.

Acknowledgments

We are grateful to Song He, Jorge G. Russo, Dmitri Sorokin, and Roberto Tateo for their interest in this work and the fruitful discussions that followed. We would like to express our special thanks to Dmitri Sorokin for his valuable insights, generous feedback, and encouraging discussions throughout this project. The work of H.B.-A. was conducted as part of the PostDoc Program on Exploring TT-bar Deformations: Quantum Field Theory and Applications, sponsored by Ningbo University.

References

  1. [1]Bialynicki-Birula, I. (1984). NONLINEAR ELECTRODYNAMICS: VARIATIONS ON A THEME BY BORN AND INFELD.
  2. [2]Gibbons, G. W., & Rasheed, D. A. (1995). Electric–magnetic duality rotations in nonlinear electrodynamics. Nucl. Phys. B, 454, 185–206. https://arxiv.org/abs/hep-th/9506035
  3. [3]Gaillard, M. K., & Zumino, B. (1981). Duality rotations for interacting fields. Nucl. Phys. B, 193, 221–244.
  4. [4]Babaei-Aghbolagh, H., Velni, K. B., Yekta, D. M., & Mohammadzadeh, H. (2022). Emergence of non-linear electrodynamic theories from TT -like deformations. Phys. Lett. B, 829, 137079. https://doi.org/10.1016/j.physletb.2022.137079
  5. [5]Borsato, R., Ferko, C., & Sfondrini, A. (2023). Classical integrability of root-TT flows. Phys. Rev. D, 107(8), 086011. https://doi.org/10.1103/PhysRevD.107.086011
  6. [6]Ferko, C., Sfondrini, A., Smith, L., & Tartaglino-Mazzucchelli, G. (2022). Root-$T T$ Deformations in Two-Dimensional Quantum Field Theories. Phys. Rev. Lett., 129(20), 201604. https://doi.org/10.1103/PhysRevLett.129.201604
  7. [7]Conti, R., Romano, J., & Tateo, R. (2022). Metric approach to a $ T T $-like deformation in arbitrary dimensions. JHEP, 09, 085. https://doi.org/10.1007/JHEP09(2022)085
  8. [8]Ferko, C., & Smith, L. (2024). Infinite Family of Integrable Sigma Models Using Auxiliary Fields. Phys. Rev. Lett., 133(13), 131602. https://doi.org/10.1103/PhysRevLett.133.131602
  9. [9]Bielli, D., Ferko, C., Smith, L., & Tartaglino-Mazzucchelli, G. (2025). Integrable higher-spin deformations of sigma models from auxiliary fields. Phys. Rev. D, 111(6), 066010. https://doi.org/10.1103/PhysRevD.111.066010
  10. [10]Bielli, D., Ferko, C., Galli, M., & Tartaglino-Mazzucchelli, G. (2025). Higher-Spin Currents and Flows in Auxiliary Field Sigma Models. https://arxiv.org/abs/2504.17294
  11. [11]Smirnov, F. A., & Zamolodchikov, A. B. (2017). On space of integrable quantum field theories. Nucl. Phys. B, 915, 363–383. https://arxiv.org/abs/1608.05499
  12. [12]Cavaglià, A., Negro, S., Szécsényi, I. M., & Tateo, R. (2016). $T T$-deformed 2D quantum field theories. JHEP, 10, 112. https://arxiv.org/abs/1608.05534
  13. [13]Russo, J. G., & Townsend, P. K. (2024). Causality and energy conditions in nonlinear electrodynamics. JHEP, 06, 191. https://doi.org/10.1007/JHEP06(2024)191
  14. [14]Russo, J. G., & Townsend, P. K. (2025). Causal chiral 2-form electrodynamics. https://arxiv.org/abs/2504.01467
  15. [15]Russo, J. G., & Townsend, P. K. (2025). Simplified self-dual electrodynamics. https://arxiv.org/abs/2505.08869
  16. [16]Bandos, I., Lechner, K., Sorokin, D., & Townsend, P. K. (2020). A non-linear duality-invariant conformal extension of Maxwell's equations. Phys. Rev. D, 102, 121703. https://doi.org/10.1103/PhysRevD.102.121703
  17. [17]Russo, J. G., & Townsend, P. K. (2024). Causal Self-Dual Electrodynamics. Phys. Rev. D, 109, 105023. https://doi.org/10.1103/PhysRevD.109.105023
  18. [18]Russo, J. G., & Townsend, P. K. (2024). Dualities of Self-Dual Nonlinear Electrodynamics. https://arxiv.org/abs/2407.02577
  19. [19]Ferko, C., Kuzenko, S. M., Smith, L., & Tartaglino-Mazzucchelli, G. (2023). Duality-Invariant Non-linear Electrodynamics and Stress Tensor Flows. Phys. Rev. D, 108, 106021. https://doi.org/10.1103/PhysRevD.108.106021
  20. [20]Kuzenko, S. M., & Raptakis, E. S. N. (2024). Higher-derivative deformations of the ModMax theory. https://arxiv.org/abs/2404.09108
  21. [21]Courant, R., & Hilbert, D. (1962). Methods of Mathematical Physics, Vol. II. Wiley‐Interscience.
  22. [22]Babaei-Aghbolagh, H., He, S., & Ouyang, H. (2024). Generalized $ T T $-like deformations in duality-invariant nonlinear electrodynamic theories. JHEP, 09, 137. https://doi.org/10.1007/JHEP09(2024)137
  23. [23]Jiang, Y. (2021). A pedagogical review on solvable irrelevant deformations of 2D quantum field theory. Commun. Theor. Phys., 73(5), 057201. https://doi.org/10.1088/1572-9494/abe4c9
  24. [24]He, S., Li, Y., Ouyang, H., & Sun, Y. (2025). $T T$ Deformation: Introduction and Some Recent Advances. https://arxiv.org/abs/2503.09997
  25. [25]Ivanov, E. A., & Zupnik, B. M. (2004). New approach to nonlinear electrodynamics: Dualities as symmetries of interaction. Phys. Atom. Nucl., 67, 2188–2199. https://doi.org/10.1134/1.1842299
  26. [26]Ferko, C., Luke Martin, C., & Sharma, P. (2026). Auxiliary Field Deformations of the Lambda Model. https://arxiv.org/abs/2609.18831
  27. [27]Bielli, D., Ferko, C., Galli, M., & Tartaglino-Mazzucchelli, G. (2026). The classical Yangian symmetry of Auxiliary Field Sigma Models. https://arxiv.org/abs/2605.18213
  28. [28]Baglioni, N., Bielli, D., Galli, M., & Tartaglino-Mazzucchelli, G. (2026). Relating auxiliary field formulations of 4d duality-invariant and 2d integrable field theories. JHEP, 07, 039. https://doi.org/10.1007/JHEP07(2026)039
  29. [29]Fukushima, O., & Yoshida, K. (2025). 4D Chern-Simons theory with auxiliary fields. JHEP, 09, 001. https://doi.org/10.1007/JHEP09(2025)001
  30. [30]Fukushima, O., Matsumoto, T., & Yoshida, K. (2026). The Courant-Hilbert construction in 4D Chern-Simons theory. JHEP, 01, 122. https://doi.org/10.1007/JHEP01(2026)122
  31. [31]Fukushima, O., Matsumoto, T., & Yoshida, K. (2026). Courant-Hilbert deformations of Yang-Baxter sigma models. JHEP, 06, 084. https://doi.org/10.1007/JHEP06(2026)084
  32. [32]Babaei-Aghbolagh, H., Chen, B., & He, S. (2025). Root-TT flows unify 4D duality-invariant electrodynamics and 2D integrable sigma models. Phys. Rev. D, 112(10), L101702. https://doi.org/10.1103/1r4p-3r5q
  33. [33]Babaei-Aghbolagh, H., Chen, B., & He, S. (2026). Integrable sigma models and universal root $T T $ deformation via Courant-Hilbert approach. JHEP, 01, 108. https://doi.org/10.1007/JHEP01(2026)108
  34. [34]Babaei-Aghbolagh, H., Babaei Velni, K., He, S., & Pezhman, Z. (2025). Root-$ T T $ deformations on causal self-dual electrodynamic theories. JHEP, 07, 227. https://doi.org/10.1007/JHEP07(2025)227
  35. [35]Russo, J. G., & Townsend, P. K. (2024). Born again. SciPost Phys., 16(5), 124. https://doi.org/10.21468/SciPostPhys.16.5.124
  36. [36]Babaei-Aghbolagh, H., Babaei Velni, K., He, S., & Pezhman, Z. (2026). Classifying Causal Nonlinear Electrodynamics via $ $-Parity and Irrelevant Deformations. https://arxiv.org/abs/2602.03426
  37. [37]Kuzenko, S. M., & Ruhl, J. (2026). Generalisations of the Russo-Townsend formulation. Phys. Lett. B, 874, 140258. https://doi.org/10.1016/j.physletb.2026.140258
  38. [38]Kuzenko, S. M., & Ruhl, J. (2026). Causal self-dual nonlinear electrodynamics from the Born-Infeld theory. https://arxiv.org/abs/2605.06193
  39. [39]Chen, B., He, S., & Hou, J. (2026). Self-dual electrodynamics via the characteristic method: relativistic and Carrollian perspectives. JHEP, 02, 055. https://doi.org/10.1007/JHEP02(2026)055
  40. [40]Hou, J. (2023). $ T T $ flow as characteristic flows. JHEP, 03, 243. https://doi.org/10.1007/JHEP03(2023)243
  41. [41]Babaei-Aghbolagh, H., Babaei Velni, K., He, S., & Isapour, F. (2026). Unified causal framework for a nonlinear electrodynamics black hole from the Courant-Hilbert approach: Thermodynamics and singularity. Phys. Rev. D, 113(8), 084041. https://doi.org/10.1103/zfc2-52md
  42. [42]Babaei-Aghbolagh, H., Babaei Velni, K., Yekta, D. M., & Mohammadzadeh, H. (2021). $ T T $-like flows in non-linear electrodynamic theories and S-duality. JHEP, 04, 187. https://doi.org/10.1007/JHEP04(2021)187
  43. [43]Ferko, C., Smith, L., & Tartaglino-Mazzucchelli, G. (2022). On Current-Squared Flows and ModMax Theories. SciPost Phys., 13(2), 012. https://doi.org/10.21468/SciPostPhys.13.2.012
  44. [44]Ferko, C., Smith, L., & Tartaglino-Mazzucchelli, G. (2023). Stress tensor flows, birefringence in non-linear electrodynamics, and supersymmetry. https://arxiv.org/abs/2301.10411
  45. [45]Babaei-Aghbolagh, H., He, S., Morone, T., Ouyang, H., & Tateo, R. (2024). Geometric Formulation of Generalized Root-TT Deformations. Phys. Rev. Lett., 133(11), 111602. https://doi.org/10.1103/PhysRevLett.133.111602
  46. [46]Born, M., & Infeld, L. (1934). Foundations of the new field theory. Proc. Roy. Soc. Lond. A, 144(852), 425–451. https://doi.org/10.1098/rspa.1934.0059
  47. [47]Sorokin, D. P. (2022). Introductory Notes on Non-linear Electrodynamics and its Applications. Fortsch. Phys., 70(7-8), 2200092. https://doi.org/10.1002/prop.202200092
  48. [48]Zamolodchikov, A. B. (2004). Expectation value of composite field T anti-T in two-dimensional quantum field theory. https://arxiv.org/abs/hep-th/0401146
  49. [49]Taylor, M. (2018). $T T$ deformations in general dimensions. https://arxiv.org/abs/1805.10287
  50. [50]Brizio, N., Kade, M., Sfondrini, A., & Sorokin, D. P. (2026). More on $ T T $-like deformations in higher dimensions. JHEP, 06, 262. https://doi.org/10.1007/JHEP06(2026)262
  51. [51]Pezhman, Z., Babaei-Aghbolagh, H., & Babaei Velni, K. (2026). Non-perturbative $T T$-like deformations on causal electrodynamics: Courant-Hilbert approach.
  52. [52]Baglioni, N., Bielli, D., Shah, P., & Tartaglino-Mazzucchelli, G. (2026). Integrable auxiliary field sigma model and the metric approach to $T T$-like deformations. https://arxiv.org/abs/2610.13501
  53. [53]Fukushima, O., Kurihara, M., Matsumoto, T., & Yoshida, K. (2026). A gravitational realization of Courant--Hilbert deformations. https://arxiv.org/abs/2609.35452
  54. [54]Pezhman, Z., Babaei-Aghbolagh, H., & Babaei Velni, K. (2026). Exact $T T$-Like Flows in Nonlinear Electrodynamics. https://arxiv.org/abs/2610.23281
  55. [55]Babaei-Aghbolagh, H., He, S., & Pezhman, Z. (2026). Exact stress-tensor generators distinguish integrable continuations. https://arxiv.org/abs/2610.31297
  56. [56]Ferko, C., Kuzenko, S. M., Lechner, K., Sorokin, D. P., & Tartaglino-Mazzucchelli, G. (2024). Interacting chiral form field theories and $ T T $-like flows in six and higher dimensions. JHEP, 05, 320. https://doi.org/10.1007/JHEP05(2024)320

Paper details

vlnc:W_x34kbwe9
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents