\documentclass[11pt,letterpaper,reqno]{amsart} \usepackage[T1]{fontenc} \usepackage{amsmath,amssymb,bm} \usepackage[mathscr]{euscript} \usepackage{microtype} \numberwithin{equation}{section} \newcommand{\R}{\mathbb R} \newcommand{\C}{\mathbb C} \newcommand{\Q}{\mathbb Q} \newcommand{\F}{\mathbb F_2} \newcommand{\Z}{\mathbb Z} \newcommand{\N}{\mathbb N} \newcommand{\T}{\R/\Z} \newcommand{\A}{\mathcal A} \newcommand{\HH}{\mathcal H} \newcommand{\act}{\mathfrak a} \newcommand{\cEH}{c^{\mathrm{EH}}} \newcommand{\FR}{\operatorname{ind}_{\mathrm{FR}}} \newcommand{\EH}{\operatorname{ind}_{\mathrm{EH}}} \newcommand{\FRrho}{\operatorname{ind}_{\Z_2}} \newcommand{\EHrho}{\operatorname{ind}_{\mathrm{EH},\rho}} \newcommand{\Erho}{E_\rho} \newcommand{\Srho}{S_\rho^+} \newcommand{\Gammarho}{\Gamma_\rho} \newcommand{\ip}[2]{\langle #1,#2\rangle} \DeclareMathOperator{\interior}{int} \DeclareMathOperator{\id}{id} \DeclareMathOperator{\Fix}{Fix} \theoremstyle{plain} \newtheorem{mainthm}{Theorem} \renewcommand{\themainthm}{\Alph{mainthm}} \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{corollary}[theorem]{Corollary} \theoremstyle{definition} \newtheorem{remark}[theorem]{Remark} \usepackage{xcolor} \definecolor{citationgreen}{rgb}{0,0.4,0} \usepackage{hyperref} \hypersetup{colorlinks=true,linkcolor=blue,citecolor=citationgreen, urlcolor=blue,linktoc=page} \setcounter{tocdepth}{2} \title[Ekeland--Hofer capacities and convex duality] {Ekeland--Hofer capacities and convex duality:\\ ordinary and real spectral invariants} \date{} \begin{document} \begin{abstract} We identify the rational Ekeland--Hofer capacities of convex bodies with Clarke's homogeneous dual-action selectors by a direct minimax argument. Convex reduction and a common nonresonant quadratic tail give the cohomological real analogue for bodies invariant under a linear anti-symplectic involution. In the real case, the mod-two index of half-period translation requires linking in the presence of fixed Fourier modes. The real selectors are full actions of generalized brake characteristics and converge to twice the least brake action. Convex approximation includes nonsmooth bodies, and Fourier decomposition computes both sequences on diagonal ellipsoids. \end{abstract} \maketitle \tableofcontents \section{Introduction} Fix $n\geq1$, write $\N=\{1,2,\ldots\}$, and identify $\R^{2n}=\C^n$. We use the Euclidean scalar product, $J_0=i$, and \[ \omega_0(v,w)=\ip{J_0v}w,\qquad \lambda_0(z)[v]=\tfrac12\ip{J_0z}v. \] Thus Hamiltonian trajectories satisfy $\dot x=J_0\nabla H(x)$. A convex body is a compact convex set with nonempty interior. Smooth boundaries are $C^\infty$; positive curvature means positive definite second fundamental form. Denote the closed Euclidean ball by $B^{2n}(R)$. For $0\in\interior C$, define \begin{equation}\label{eq:gauge} \begin{aligned} j_C(z)&=\inf\{r>0:z\in rC\},\qquad h_C=j_C^2,\\ h_C^*(p)&=\sup_{z\in\R^{2n}}\bigl(\ip pz-h_C(z)\bigr). \end{aligned} \end{equation} Write $L^p=L^p(\T,\R^{2n})$, with norm $\|\cdot\|_p$, and $H^s=H^s(\T,\R^{2n})$; the subscript $0$ denotes mean-zero loops. All function spaces have their strong topologies. The mean-zero primitive $\Pi:L^2_0\to H^1_0$ is \begin{equation}\label{eq:primitive} (\Pi u)(t)=\int_0^t u(s)\,ds- \int_0^1\int_0^\tau u(s)\,ds\,d\tau. \end{equation} Set \begin{equation}\label{eq:velocity-functionals} \begin{aligned} \A(u)&=\tfrac12\int_0^1\ip{J_0\Pi u}{u}\,dt,\\ \HH_C(u)&=\int_0^1h_C^*(-J_0u)\,dt, \end{aligned} \end{equation} and define the homogeneous dual quotient by \begin{equation}\label{eq:quotient} \begin{aligned} \Lambda&=\{u\in L^2_0:\|u\|_2=1,\ \A(u)>0\},\\ \Psi_C(u)&=\frac{\HH_C(u)}{\A(u)},\qquad u\in\Lambda. \end{aligned} \end{equation} The circle $S^1=\T$ acts by $(\theta\cdot u)(t)=u(t+\theta)$. For its rational Borel index $\FR$, defined in Section~\ref{sec:indices}, Clarke's selectors are \begin{equation}\label{eq:selector} s_k(C)=\inf\{a>0:\FR\{\Psi_Cn. \end{cases} \end{equation} \end{mainthm} \section{Ordinary minimax and dual-action foundations}\label{sec:indices} In the ordinary theory, we use singular cohomology with coefficients in $\Q$. Let $ES^1$ be the unit sphere of $\ell^2(\C)$ with scalar circle action, and put $BS^1=ES^1/S^1$. For an $S^1$-space $X$, let $p_X:X\times_{S^1}ES^1\to BS^1$ be its Borel projection. The universal first Chern class $e\in H^2(BS^1;\Q)$ gives $H^*(BS^1;\Q)=\Q[e]$. Define \begin{equation}\label{eq:FR} \FR(X)=\sup\{j\geq1:p_X^*(e^{j-1})\ne0\},\qquad\FR(\varnothing)=0. \end{equation} An equivariant map $X\to Y$ implies $\FR(X)\leq\FR(Y)$. In particular, no freeness assumption is required. \begin{lemma}\label{lem:weighted-index} If $V=\bigoplus_{j=1}^d\C_{(m_j)}$, where $\theta\cdot z=e^{2\pi i m_j\theta}z$ and $m_j\in\Z\setminus\{0\}$, then its unit sphere satisfies $\FR(S(V))=d$. \end{lemma} \begin{proof}[Proof sketch] The equivariant Euler class is $(\prod_jm_j)e^d$. The sphere-bundle Gysin sequence gives exactly the asserted height of $e$; see \cite[proof of Lemma~A.2]{BBLM23}. The weights are invertible over $\Q$. \end{proof} Let \begin{equation}\label{eq:E} \begin{aligned} E&=H^{1/2}(\T,\C^n)=E^-\oplus E^0\oplus E^+,\\ x(t)&=\sum_{m\in\Z}e^{2\pi imt}x_m,\\ \|x\|_E^2&=|x_0|^2+2\pi\sum_{m\ne0}|m||x_m|^2. \end{aligned} \end{equation} The summands are the negative, constant, and positive Fourier modes; $P^-,P^0,P^+$ denote their projections and $x^\pm=P^\pm x$. These projections also act on $L^2$; $\ip{\cdot}{\cdot}_E$ is the real Hilbert scalar product for the displayed norm. The inclusion $i:E\hookrightarrow L^2$ is compact; write $i^*$ for its Hilbert adjoint. Moreover, $\|x\|_p\leq C_p\|x\|_E$ for every finite $p$. EH minimax sets, neighborhoods, and deformations use the $E$ topology; $H^1$ records primitive and critical-loop regularity. The map \[ \Pi:L^2_0\longrightarrow H^1_0\hookrightarrow E \] is bounded into $H^1$ and compact into $E$; differentiation recovers $u$. Define the free action on all of $E$ by \begin{equation}\label{eq:free-action} \begin{aligned} \act(x)&=\pi\sum_{m\in\Z}m|x_m|^2\\ &=\tfrac12(\|x^+\|_E^2-\|x^-\|_E^2),\qquad x\in E. \end{aligned} \end{equation} For $x\in H^1$, \[ \act(x)=\int_x\lambda_0=\tfrac12\int_0^1\ip{J_0x}{\dot x}\,dt. \] In particular, $\A(u)=\act(\Pi u)$ for $u\in L^2_0$. Put $S^+=\{x\in E^+:\|x\|_E=1\}$. The original EH group $\Gamma$ consists of equivariant homeomorphisms \begin{equation}\label{eq:deformation} \Theta(x)=e^{\theta_+(x)}x^++x^0+e^{\theta_-(x)}x^-+K(x), \end{equation} where $\theta_\pm$ are continuous invariant functions bounded on bounded sets and $K$ is continuous, equivariant, and compact on bounded sets. For some $\rho>0$, all three vanish when $\act(x)\leq0$ or $\|x\|_E\geq\rho$. Use the neighborhood pseudoindex \begin{equation}\label{eq:pseudoindex} \EH(V)=\inf_{\Theta\in\Gamma} \inf_{\substack{U\supset V\cap\Theta(S^+)\\U\text{ open, invariant in }E}} \FR(U). \end{equation} These are the original deformations \cite[Section~II]{EH90}, with the singular-cohomology convention of \cite[Section~3.1, Remark~A.1]{BBLM23}. For invariant open $V$, the inner infimum equals \begin{equation}\label{eq:open-index} \FR(V\cap\Theta(S^+)). \end{equation} Indeed, for $Y=\Theta^{-1}V\cap S^+$, the open set $\{x:x^+\ne0,\ x^+/\|x^+\|_E\in Y\}$ retracts equivariantly to $Y$. The Borel spaces of these open Hilbert-manifold slices are metrizable and locally contractible: their bundles over $BS^1$ are locally products. Thus singular and Alexander--Spanier cohomology agree naturally there, preserving $p_X^*e^j$. Let $\mathcal F(C)$ consist of smooth $H:\R^{2n}\to[0,\infty)$, zero near $C$, with $H(z)=b|z|^2$ outside a compact set, where $b>\pi$ and $b\notin\pi\N$. Set \begin{equation}\label{eq:Phi} \Phi_H(x)=\act(x)-\int_0^1H(x(t))\,dt. \end{equation} \begin{equation}\label{eq:EH-body} \begin{aligned} \cEH_k(H)&=\inf\{a\in\R:\EH\{\Phi_H0,\ p\in\R^{2n}). \end{gathered} \end{equation} If $H\in\mathcal F(C)$ and $0\in\interior C$, then $\cEH_k(H)>0$; it is finite if its outer slope satisfies $b>\pi\lceil k/n\rceil$. Every finite positive $\cEH_k(H)$ is a critical value of $\Phi_H$. For admissible $H,G$, \[ H\leq G\quad\Longrightarrow\quad\cEH_k(H)\geq\cEH_k(G). \] If their difference is bounded and their selectors are finite, then \begin{equation}\label{eq:eh-bounded-perturbation} |\cEH_k(H)-\cEH_k(G)|\leq\|H-G\|_\infty. \end{equation} Thus a family with a common quadratic tail, continuous in the uniform norm of differences, cannot have its finite $k$th selector cross a positive level which is never critical. \end{proposition} \begin{proof}[Proof sketch] These are the EH minimax properties \cite[Section~II, pp.~558--559]{EH90}. We recall the estimates relevant to our conventions. If $H=0$ on $B^{2n}(r)$, quadratic growth gives $K>0$ such that \[ \begin{gathered} H(z)\leq K|z|^4/r^2,\\ \Phi_H\big|_{\rho S^+}\geq\rho^2/2-KC_4^4\rho^4/r^2>0 \end{gathered} \] for small $\rho>0$. For any $a>0$, choose smooth cutoffs $\beta:\R\to[0,1]$ and $\gamma:[0,\infty)\to[0,1]$ with \[ \begin{gathered} \beta=1\text{ on }[\tfrac12\min\{1,a^2\},\tfrac12\max\{1,a^2\}], \qquad \beta=0\text{ on }(-\infty,0],\\ \gamma=1\text{ on }[0,\max\{1,a^2\}],\qquad \gamma\text{ of compact support}. \end{gathered} \] The time-$\log a$ flow of $\dot x=\beta(\act(x))\gamma(\|x\|_E^2)x^+$ belongs to $\Gamma$ and maps $S^+$ to $aS^+$. Indeed, its trajectories from $S^+$ are $x(t)=e^tv$ between times $0$ and $\log a$; only the positive component is rescaled, and the cutoff fixes the required regions. The EH linking estimate \cite[Proposition~1]{EH90} gives \begin{equation}\label{eq:eh-finite-cone-linking-input} \EH(E^-\oplus E^0\oplus V_N)\geq nN. \end{equation} Its finite-dimensional proof uses the weighted Euler class and piercing; the singular argument uses \cite[Proposition~A.4]{BBLM23}. Since $H\geq b|z|^2-A_0$, the functional is bounded above by $A_0$ on this cone when $b>\pi N$, proving finiteness. For compactness, put $L=P^+-P^-$ and $q(z)=\nabla H(z)-2bz$. Then \[ \nabla\Phi_H=(L-2bi^*i)x-i^*q(x). \] The last term has relatively compact image since $q$ is bounded. The linear term has Fourier eigenvalues \[ \begin{cases} -2b,&m=0,\\ 1-b/(\pi m),&m>0,\\ -1-b/(\pi|m|),&m<0. \end{cases} \] Nonresonance bounds their absolute values away from zero. Thus every sequence with gradient tending to zero has a convergent subsequence. For a critical point $x\in E$, the Fourier equations give $2\pi m x_m=(\nabla H(x))_m$. Since $\nabla H(x)\in L^2$, they imply $x\in H^1$ and $\dot x=J_0\nabla H(x)$, hence $x$ is smooth. This regularity identifies the $E$-critical points with the periodic orbits used below. Fix a positive noncritical value $c$ and choose $\varepsilon>0$ so that $[c-2\varepsilon,c+2\varepsilon]\subset(0,\infty)$ contains no critical value. Put $Y_\pm=\{\Phi_H\sup_{v\in S^+}\|\Theta(v)\|_E+3\varepsilon\|Z\|_\infty \] and a smooth $\zeta:[0,\infty)\to[0,1]$, equal to one on $[0,R^2]$ and zero outside a bounded interval. Let $\varphi_t$ be the flow of \[ \begin{aligned} \dot x&=a_H(x)\bigl(Lx-i^*\nabla H(x)\bigr),\\ a_H(x)&=\frac{\chi(\Phi_H(x))\zeta(\|x\|_E^2)} {\|\nabla\Phi_H(x)\|_E^2}, \end{aligned} \] where $a_H=0$ outside the band. With $\beta_t(x)=\int_0^t a_H(\varphi_s(x))\,ds$, variation of constants gives \[ \varphi_t(x)=e^{\beta_t(x)}x^++x^0+e^{-\beta_t(x)}x^-+K_t(x). \] Here $\beta_t$ is continuous and invariant, and $K_t$ is compact on bounded sets: its integral expression is a uniform limit, on bounded sets, of Riemann sums with relatively compact image. The norm cutoff fixes all sufficiently large points. Since $H\geq0$, the flow also fixes $\{\act\leq0\}$. Thus $g=\varphi_{3\varepsilon}\in\Gamma$ and \[ Y_+\cap g\Theta(S^+)\subset g\bigl(Y_-\cap\Theta(S^+)\bigr). \] Transporting an open index witness for $Y_-$ by $g$ excludes an index threshold at $c$. Sublevel inclusions prove Hamiltonian monotonicity and \eqref{eq:eh-bounded-perturbation}, and continuity then gives no crossing. Monotonicity is immediate from the admissible families; translation and dilation are the EH invariances \cite[Section~II, p.~559]{EH90}. For the exact tail convention, allowing $b|z-q|^2$ gives the same body infimum for every center $q$: any such $H$ admits a majorant with tail $b'|z-q'|^2$, $b'>b$, by interpolation outside a sufficiently large ball. The reverse comparison interchanges $q,q'$. Translation is therefore compatible with the centered definition; $r^2H(z/r)$ proves scaling. \end{proof} \begin{lemma}\label{lem:quotient-finite} For centered convex bodies, the quotient $\Psi_C$ is continuous, and $s_k$ is monotone and homogeneous of degree two. These assertions include nonsmooth bodies. If $rB^{2n}(1)\subset C\subset RB^{2n}(1)$, then \begin{equation}\label{eq:finite} \pi r^2\lceil k/n\rceil\leq s_k(C) \leq\pi R^2\lceil k/n\rceil. \end{equation} \end{lemma} \begin{proof} For the support function $\sigma_C(p)=\max_{z\in C}\ip pz$, \begin{equation}\label{eq:support} h_C^*(p)=\sup_{t\geq0}(t\sigma_C(p)-t^2)=\tfrac14\sigma_C(p)^2. \end{equation} Boundedness of $\Pi$ gives continuity of $\A$. Also \[ \tfrac14r^2\|u\|_2^2\leq\HH_C(u)\leq\tfrac14R^2\|u\|_2^2, \] and $|\sigma_C(p)-\sigma_C(q)|\leq R|p-q|$ gives \[ |\HH_C(u)-\HH_C(v)|\leq\tfrac14R^2(\|u\|_2+\|v\|_2)\|u-v\|_2. \] Moreover $C\subset D$ implies $\Psi_C\leq\Psi_D$, and $\Psi_{rC}=r^2\Psi_C$. Writing $u_m$ for the Fourier coefficients of $u$ gives \[ \A(u)=\frac1{4\pi}\sum_{m\ne0}\frac{|u_m|^2}{m}. \] On the $L^2$-unit sphere of $V_N$, $\Psi_C\leq\pi R^2N$ and the index is $nN$. Taking strict levels above $\pi R^2N$ proves the upper bound. If $P_{V_N}u=0$ and $\|u\|_2=1$, then $\A(u)\leq1/(4\pi(N+1))$ and $\HH_C(u)\geq r^2/4$. Hence $L^2$-normalized projection to $V_N$ is defined on $\{\Psi_C0$ such that \begin{equation}\label{eq:h-uniform-convexity} \begin{aligned} \ip{\nabla h(z)-\nabla h(z')}{z-z'}&\geq\mu|z-z'|^2, \\ |\nabla h(z)-\nabla h(z')|&\leq L|z-z'|. \end{aligned} \end{equation} \begin{equation}\label{eq:h-quadratic-bounds} m_0|z|^2\leq h(z)\leq M_0|z|^2. \end{equation} Write $q_\pm(v)=\frac12\|v\|_E^2$ on $E^\pm$. For $\tau>0$, put $F_\tau(x)=\tau\int_0^1h(x(t))\,dt$ for $x\in L^2$, and define, for $v\in E^+$, \begin{equation}\label{eq:reduced-K} K_\tau(v)=\min_{w\in E^-,\ c\in E^0} \bigl(F_\tau(v+w+c)+q_-(w)\bigr). \end{equation} The minimized expression is strictly convex and weakly lower semicontinuous, with the coercive bound \[ F_\tau(v+w+c)+q_-(w) \geq\tau m_0(\|v\|_2^2+\|w\|_2^2+|c|^2)+q_-(w). \] The direct method gives a unique minimizer $(w_\tau(v),c_\tau(v))$. Define \begin{equation}\label{eq:reduced-f} \begin{aligned} x_\tau(v)&=v+w_\tau(v)+c_\tau(v), \\ f_\tau(v)&=q_+(v)-K_\tau(v). \end{aligned} \end{equation} The minimizing pair is equivariant and positively homogeneous, and \begin{equation}\label{eq:reduced-direct-maximum} f_\tau(v)=\max_{w\in E^-,\ c\in E^0} \bigl(\act(v+w+c)-F_\tau(v+w+c)\bigr). \end{equation} \begin{lemma}\label{lem:negative-graph-regularity} The map $W_\tau:E^+\to H^1\cap(E^-\oplus E^0)$, $W_\tau=w_\tau+c_\tau$, is Lipschitz from the $L^2$ norm on $E^+$ to the $H^1$ norm and compact into $E$ on $E$-bounded sets. The full graph point $x_\tau(v)$ is only required to lie in $E$. \end{lemma} \begin{proof} Fix $v_1,v_2\in E^+$ and put $p_i=\tau\nabla h(x_\tau(v_i))$ for $i=1,2$. Let $\Delta$ denote the difference between the corresponding quantities at $v_1$ and $v_2$. The Fourier Euler equations are \begin{equation}\label{eq:negative-euler} \begin{gathered} (p_i)_0=0,\\ (p_i)_m=2\pi m\,(w_\tau(v_i))_m\qquad(m<0). \end{gathered} \end{equation} Pairing with $\Delta w$ and using $(\Delta p)_0=0$ gives \[ \ip{\Delta p}{\Delta x}_{L^2} =\ip{\Delta p}{\Delta v}_{L^2}-\|\Delta w\|_E^2. \] Strong monotonicity and the Lipschitz bound therefore yield \[ \|\Delta w\|_E^2+\tau\mu\|\Delta x\|_2^2 \leq\tau L\|\Delta x\|_2\|\Delta v\|_2, \] hence \begin{equation}\label{eq:negative-graph-estimate} \|\Delta w\|_E^2+\frac{\tau\mu}{2}\|\Delta x\|_2^2 \leq\frac{\tau L^2}{2\mu}\|\Delta v\|_2^2. \end{equation} The Fourier equations also give \begin{equation}\label{eq:negative-graph-H1} \begin{aligned} |\Delta c|&\leq\|\Delta x\|_2,\\ \|\partial_t\Delta w\|_2 &=\|P^-\Delta p\|_2\leq\tau L\|\Delta x\|_2. \end{aligned} \end{equation} Together with \eqref{eq:negative-graph-estimate}, these prove the $H^1$ estimate. Compactness follows from $E\hookrightarrow L^2$. \end{proof} \begin{lemma}\label{lem:homogeneous-cone-index} For every $\tau>0$, \begin{equation}\label{eq:homogeneous-cone-index} \begin{aligned} &\FR(\{v\in E^+:f_\tau(v)<0\})\\ &\qquad=\FR(\{u\in\Lambda:\Psi_C(u)<\tau\}). \end{aligned} \end{equation} Intersecting the left-hand set with $S^+$ preserves its index. \end{lemma} \begin{proof} The conjugate of $F_\tau$ on $L^2$ is $F_\tau^*(p)=\tau^{-1}\int h^*(p)$. For $v\in E^+$, define \begin{equation}\label{eq:positive-dual-map} \begin{aligned} p(v)&=\tau\nabla h(x_\tau(v)),\\ U(v)&=J_0p(v),\qquad r(v)=P^+\Pi U(v). \end{aligned} \end{equation} Equation~\eqref{eq:negative-euler} gives $U(v)\in L^2_0$ and \[ \Pi U(v)=r(v)+w_\tau(v). \] The maps $U:E^+\to L^2_0$ and $r:E^+\to E^+$ are continuous, by Lemma~\ref{lem:negative-graph-regularity} and the Lipschitz continuity of $\nabla h$. Abbreviate $r=r(v)$ and $w=w_\tau(v)$. Fenchel equality and Euler's identity give \[ \begin{aligned} F_\tau^*(p(v))&=F_\tau(x_\tau(v)),\\ \ip{r}{v}_E&=\ip{p(v)}{x_\tau(v)}_{L^2}+2q_-(w)\\ &=2K_\tau(v). \end{aligned} \] Therefore \begin{equation}\label{eq:cone-forward-identity} \begin{aligned} \tau^{-1}\HH_C(U(v))-\A(U(v))&=K_\tau(v)-q_+(r)\\ &=f_\tau(v)-q_+(r-v)\leq f_\tau(v). \end{aligned} \end{equation} Conversely, for $u\in L^2_0$, put $z=\Pi u$ and $y=P^+z$. Fenchel's inequality at $(x_\tau(y),-J_0u)$ gives \begin{equation}\label{eq:cone-reverse-inequality} \begin{aligned} \tau^{-1}\HH_C(u)-\A(u) &\geq\ip{-J_0u}{x_\tau(y)}_{L^2}-F_\tau(x_\tau(y))\\ &\qquad-q_+(y)+q_-(z^-)\\ &=f_\tau(y)+q_-(z^--w_\tau(y))\geq f_\tau(y). \end{aligned} \end{equation} The pairing uses only $x_\tau(y)\in L^2$ and $z\in H^1$ and does not differentiate the graph point. The first inequality gives $\A(U(v))>0$ when $f_\tau(v)<0$; the second gives $f_\tau(P^+\Pi u)<0$ when $\Psi_C(u)<\tau$. Hence the continuous equivariant maps \[ \begin{aligned} \{v\in E^+:f_\tau(v)<0\} &\longrightarrow\{u\in\Lambda:\Psi_C(u)<\tau\},\\ v&\longmapsto\frac{U(v)}{\|U(v)\|_2},\\[3pt] \{u\in\Lambda:\Psi_C(u)<\tau\} &\longrightarrow\{v\in E^+:f_\tau(v)<0\},\\ u&\longmapsto P^+\Pi u \end{aligned} \] are well-defined. Index monotonicity in both directions proves the equality; positive radial normalization proves the sphere assertion. \end{proof} \begin{lemma}\label{lem:homogeneous-uniform-gap} For $0<\tau<\tau'$, put $X_{\tau'}=\{v\in S^+:f_{\tau'}(v)<0\}$ and \[ \alpha=\frac{(\tau'-\tau)m_0} {2(\tau M_0+(\tau'-\tau)m_0)}>0. \] Then $f_\tau\geq\alpha$ on $S^+\setminus X_{\tau'}$. \end{lemma} \begin{proof} Using the $K_{\tau'}$ minimizer as a competitor for $K_\tau$ gives \[ \begin{aligned} K_{\tau'}(v)-K_\tau(v) &\geq(\tau'-\tau)\int_0^1 h(x_{\tau'}(v)(t))\,dt\\ &\geq(\tau'-\tau)m_0\|v\|_2^2. \end{aligned} \] On the stated complement $K_{\tau'}\leq1/2$, while the competitor $(w,c)=(0,0)$ gives $K_\tau(v)\leq\tau M_0\|v\|_2^2$. Consequently, with $t=\|v\|_2^2$, \[ \begin{aligned} f_\tau(v)&\geq\max\{(\tau'-\tau)m_0t,\tfrac12-\tau M_0t\}\\ &\geq\alpha. \end{aligned}\qedhere \] \end{proof} \begin{proposition}\label{prop:graph-pseudoindex-bound} Fix $0<\tau<\tau'$ and $a>0$. There are $\Theta\in\Gamma$ and $B_0<\infty$ with $\|\Theta(v)\|_E\leq B_0$ on $S^+$ such that, for every $H\in\mathcal F(C)$, \begin{equation}\label{eq:graph-tail-error} \int_0^1(H(x(t))-\tau h(x(t)))\,dt\leq1 \quad(\|x\|_E\leq B_0) \end{equation} implies \begin{equation}\label{eq:graph-pseudoindex-bound} \EH(\{\Phi_Ha+1$, put $W=W_\tau$, and take $\|W(v)\|_E\leq C_0\|v\|_E$. Set $B_0=R(1+C_0)$. For $v\in S^+\setminus X_{\tau'}$ and $0\leq t\leq1$, \begin{align} \act(Rv+tRW(v)) &=R^2(q_+(v)-t^2q_-(w_\tau(v)))\notag\\ &\geq R^2f_\tau(v)\geq R^2\alpha.\label{eq:graph-positive-path} \end{align} These segments lie in the ball of radius $B_0$. Choose smooth $[0,1]$-valued cutoffs $\chi=0$ on $(-\infty,R^2\alpha/2]$, $\chi=1$ on $[R^2\alpha,\infty)$, and $\zeta=1$ on $[0,B_0^2]$ with bounded support. The equivariant field \begin{equation}\label{eq:graph-vector-field} \mathcal V(x)=\chi(\act(x))\zeta(\|x\|_E^2)W(x^+) \end{equation} is globally Lipschitz since $W$ is Lipschitz and the norm cutoff bounds its support. Its flow $\varphi_t$ has constant positive component, so its time-one map is \[ \begin{aligned} T(x)&=x+\eta(x)W(x^+),\\ \eta(x)&=\int_0^1\chi(\act(\varphi_t(x))) \zeta(\|\varphi_t(x)\|_E^2)\,dt\in[0,1]. \end{aligned} \] Here $\eta$ is continuous and invariant. The added term is compact on bounded sets, and $T$ fixes $\{\act\leq0\}$ and all sufficiently large points. Thus $T\in\Gamma$, and uniqueness of the flow along the displayed segments gives \begin{equation}\label{eq:graph-flow-exact} T(Rv)=Rx_\tau(v)\qquad(v\in S^+\setminus X_{\tau'}). \end{equation} Let $D_R\in\Gamma$ be the radial deformation constructed in the proof of Proposition~\ref{prop:eh-facts}, so $D_R(v)=Rv$ on $S^+$. Put $\Theta=T\circ D_R$. Then $\|\Theta(v)\|_E\leq B_0$ on $S^+$ and, outside $X_{\tau'}$, \[ \Phi_H(\Theta(v))\geq R^2f_\tau(v)-1>a. \] Hence $\Theta^{-1}(\{\Phi_H0,\\ [c/(1+2\delta),c]\cap\mathcal S(C)=\varnothing. \end{gathered} \end{equation} Choose a smooth convex profile $f:\mathbb R\to[0,\infty)$ with \begin{equation}\label{eq:cap-profile} \begin{gathered} f=0\ \text{on }(-\infty,1+\delta],\\ 00$ and \[ b>\max\{\pi,M\max_{|z|=1}h(z)\},\qquad b\notin\pi\N, \] there is a family $\widehat H_s\in\mathcal F(C)$, $s\in[\tau,M]$, equal to $b|z|^2$ outside one fixed ball and continuous in the uniform norm of differences. Each member satisfies $\widehat H_s\geq A_s$ and $\Phi_{\widehat H_s}(x)\ne c$ for every one-periodic orbit $x$. Uniformly for $s\in[\tau,M]$, \begin{equation}\label{eq:cap-integrated-error} 0\leq\int_0^1(\widehat H_s-A_s)(x(t))\,dt<\zeta \qquad(\|x\|_E\leq B). \end{equation} \end{lemma} \begin{proof} Choose $\alpha,\beta>0$ with $\alpha|z|^2\leq h(z)\leq\beta|z|^2$ and $b>M\beta$, and $\varepsilon>0$ with $\tau\kappa-\varepsilon>c$. An even mollification $\vartheta$ of $|\cdot|$ may be chosen so that \[ \begin{gathered} \vartheta(t)=|t|\quad(|t|\geq\varepsilon),\qquad |\vartheta'|\leq1,\\ 0\leq\vartheta-t\vartheta'\leq\varepsilon,\\ 0\leq\vartheta-|t|\leq\varepsilon. \end{gathered} \] Put $\operatorname{mx}(a,d)=(a+d+\vartheta(a-d))/2$. Choose $R\geq1$ so large that \begin{equation}\label{eq:cap-energy-separation} \begin{gathered} \alpha R^2>1+2\delta,\\ \tau(\alpha R^2-\kappa)>M f(1+2\delta),\\ \frac{(b+\varepsilon/2)C_4^4B^4}{R^2}<\zeta, \end{gathered} \end{equation} where $\|x\|_4\leq C_4\|x\|_E$. For $L>\max\{bR^2+\varepsilon,c+\varepsilon\}$, define \[ F_s(z)=\operatorname{mx}(A_s(z),b|z|^2-L). \] Then $F_s=A_s$ for $|z|\leq R$, while $b>M\beta$ gives a common $R_2>R$ with \begin{equation}\label{eq:cap-radial-tail} F_s(z)=b|z|^2-L\qquad(|z|\geq R_2). \end{equation} For $\mathcal DF=\langle\nabla F,z\rangle/2-F$, on $\{h>1+2\delta\}$ the chain rule gives \begin{equation}\label{eq:cap-Euler-bound} \begin{aligned} \mathcal DF_s &=w_1s\kappa+w_2L+\tfrac12(t\vartheta'(t)-\vartheta(t))\\ &\geq\min\{s\kappa,L\}-\varepsilon/2>c, \end{aligned} \end{equation} where $t=A_s-b|z|^2+L$ and $w_1,w_2$ are the nonnegative partial derivatives of $\operatorname{mx}$, with $w_1+w_2=1$. To restore the exact quadratic tail, take a smooth nonincreasing $\chi:[0,\infty)\to[0,1]$, equal to $1$ on $[0,1]$ and $0$ on $[2,\infty)$. Choose \[ T>R_2^2,\qquad (L/T)\|\chi'\|_\infty<\tfrac12\operatorname{dist}(b,\pi\Z). \] Set \begin{equation}\label{eq:cap-final-H} \widehat H_s(z)=F_s(z)+L\bigl(1-\chi(|z|^2/T)\bigr). \end{equation} For $|z|\geq R_2$ this equals $q(|z|^2)$, where \[ \begin{aligned} q(t)&=bt-L\chi(t/T),\\ q'(t)&=b-(L/T)\chi'(t/T)\notin\pi\Z. \end{aligned} \] A trajectory meeting this region stays on its radial sphere by uniqueness of the Hamiltonian flow, and hence cannot be one-periodic. Thus every nonconstant one-periodic orbit lies in $\{|z|A_s$, its energy exceeds $M f(1+2\delta)$ by \eqref{eq:cap-energy-separation}. It cannot enter $\{h\leq1+2\delta\}$, where $F_s=A_s\leq M f(1+2\delta)$; \eqref{eq:cap-Euler-bound} therefore makes its action $\int_0^1\mathcal DF_s(x(t))\,dt>c$. Otherwise it is an $A_s$-orbit (the nonnegative difference $F_s-A_s$ has zero gradient at its zeros). Its level $h=r$ is constant. For $r\geq1+2\delta$ its action is $s\kappa>c$; for $1+\deltaR\}},\\ \int_0^1(\widehat H_s-A_s)(x(t))\,dt \leq\frac{(b+\varepsilon/2)C_4^4B^4}{R^2}<\zeta. \end{gathered} \] The pointwise bound uses $\operatorname{mx}\leq\max+\varepsilon/2$ before the radial correction and $\widehat H_s\leq b|z|^2$ during it. This proves \eqref{eq:cap-integrated-error}. \end{proof} \section{Comparison and convex approximation}\label{sec:comparison} \begin{lemma}\label{lem:positive-ray-energy} Let $0\in\interior C$ and let $v\in H^1$ satisfy $\act(v)>0$ and $\HH_C(\dot v)\leq b\act(v)$ for some $b>0$. For $r>0$ and $y\in E^-\oplus E^0$, \begin{equation}\label{eq:cone-energy-bound} \act(y+rv)\leq b\int_0^1h_C(y(t)+rv(t))\,dt. \end{equation} \end{lemma} \begin{proof} Put $x=y+rv$. The assertion is immediate if $\act(x)\leq0$. Otherwise set $B(v,x)=\int_0^1\ip{-J_0\dot v}{x}\,dt$. Since $\act(x-rv)=\act(y)\leq0$, \[ \begin{aligned} rB(v,x)&=\act(x)-\act(y)+r^2\act(v)\\ &\geq\act(x)+r^2\act(v) \geq2r\sqrt{\act(x)\act(v)}>0. \end{aligned} \] Fenchel's inequality for $t>0$ gives \[ tB(v,x)-t^2\HH_C(\dot v)\leq\int_0^1h_C(x(t))\,dt. \] Its maximum over positive $t$ is attained at $t=B(v,x)/(2\HH_C(\dot v))$. Hence \[ \int_0^1h_C(x(t))\,dt \geq\frac{B(v,x)^2}{4\HH_C(\dot v)} \geq\frac{\act(x)}b. \] Only positive scalings of $\dot v$ were used. \end{proof} \begin{proposition}\label{thm:smooth} If $D$ is smooth and positively curved, with $0\in\interior D$, then $\cEH_k(D)=s_k(D)$ for every $k\geq1$. \end{proposition} \begin{proof} Write $h=h_D$ and $s=s_k(D)\in(0,\infty)$. The upper inequality is proved in \cite[equation~(3.3), Proposition~3.1]{BBLM23}. For its proof, fix $b>s\geq s_1(D)=\min\mathcal S(D)$. By \cite[Lemma~3.5]{BBLM23}, choose a compact invariant set $V_b\subset E$ of mean-zero $H^1$ primitives and put $W_b=\{rv:r>0,\ v\in V_b\}$. These sets satisfy \[ \begin{gathered} \act(v)>0,\qquad \HH_D(\dot v)\leq b\act(v)\quad(v\in V_b),\\ \FR(W_b)\geq\FR\{\Psi_D\leq b\}\geq k. \end{gathered} \] Put $Z_b=E^-+E^0+W_b$. Lemma~\ref{lem:positive-ray-energy} gives $\act(x)\leq b\int_0^1h(x(t))\,dt$ on $Z_b$. For every $\varepsilon>0$, the majorant in \cite[Lemma~3.3]{BBLM23} satisfies $H_b\in\mathcal F(D)$ and $H_b\geq bh-b-\varepsilon$. Together with the linking statement for this carrier \cite[Lemma~3.6]{BBLM23}, this yields \[ \sup_{Z_b}\Phi_{H_b}\leq b+\varepsilon,\qquad \EH(Z_b)\geq k. \] Every strict level above $b+\varepsilon$ therefore qualifies in the minimax, so $\cEH_k(D)\leq\cEH_k(H_b)\leq b+\varepsilon$. Let $\varepsilon\downarrow0$ and $b\downarrow s$. For the lower inequality, fix $H\in\mathcal F(D)$ and $c\in(0,s)\setminus\mathcal S(D)$. Choose \[ c0$ so that \[ \begin{gathered} H=0\text{ on }\{h\leq1+2\delta\},\\ [c/(1+2\delta),c]\cap\mathcal S(D)=\varnothing. \end{gathered} \] For the profile $f$ in \eqref{eq:cap-profile}, the quotient $H/f(h)$ is bounded where defined: $H$ vanishes near its denominator's zero set, and both functions have quadratic growth. Choose $M>\tau$ with $H\leq Mf(h)$. Apply Lemma~\ref{lem:fixed-cap} with error $\zeta=1$, radius $B_0$, and common slope \[ \begin{gathered} b_\infty>\max\{\pi\lceil k/n\rceil,M\max_{|z|=1}h(z)\},\\ b_\infty\notin\pi\N. \end{gathered} \] For $\|x\|_E\leq B_0$, \eqref{eq:cap-integrated-error} and $f(h)\leq h$ give \[ \int_0^1(\widehat H_\tau-\tau h)(x(t))\,dt \leq\int_0^1(\widehat H_\tau-A_\tau)(x(t))\,dt<1. \] Proposition~\ref{prop:graph-pseudoindex-bound} therefore gives \[ \begin{gathered} \EH\{\Phi_{\widehat H_\tau}c. \end{gathered} \] The choice of $b_\infty$ and Proposition~\ref{prop:eh-facts} make $\nu\mapsto\cEH_k(\widehat H_\nu)$ finite-valued and continuous on $[\tau,M]$. Spectrality and the exclusion of $c$ in Lemma~\ref{lem:fixed-cap} give $\cEH_k(\widehat H_\nu)>c$ throughout this interval. At the endpoint, Hamiltonian monotonicity gives \[ \begin{gathered} H\leq Mf(h)\leq\widehat H_M,\\ \cEH_k(H)\geq\cEH_k(\widehat H_M)>c. \end{gathered} \] Take the infimum over $H$, then let $c\uparrow s$ through the complement of $\mathcal S(D)$, using Lemma~\ref{lem:spectrum-gap}. \end{proof} \begin{lemma}\label{lem:convex-approximation} Every convex body $C$ with $0\in\interior C$ admits smooth positively curved approximants $D_\varepsilon$ and $\eta_\varepsilon\to0$, $0<\eta_\varepsilon<1$, such that \begin{equation}\label{eq:sandwich} \sqrt{1-\eta_\varepsilon}\,C\subset D_\varepsilon \subset\sqrt{1+\eta_\varepsilon}\,C. \end{equation} The approximants may preserve every orthogonal symmetry of $C$. \end{lemma} \begin{proof} Choose $R>0$ with $C\subset B^{2n}(R)$ and a radial nonnegative smooth unit-mass mollifier $\rho_\varepsilon$ supported in $B^{2n}(\varepsilon)$. Put \begin{equation}\label{eq:convex-regularization} g_\varepsilon=h_C*\rho_\varepsilon+\varepsilon|z|^2, \qquad D_\varepsilon=\{g_\varepsilon\leq1\}. \end{equation} Then \[ \begin{gathered} D^2g_\varepsilon\geq2\varepsilon I,\\ g_\varepsilon(z)\geq\frac{(\max\{|z|-\varepsilon,0\})^2}{R^2}, \end{gathered} \] Thus $D_\varepsilon\subset B^{2n}(R+\varepsilon)$, and $g_\varepsilon\to h_C$ locally uniformly. For small $\varepsilon$, $g_\varepsilon(0)<1$ and on its unit level \[ \ip{\nabla g_\varepsilon(z)}z\geq1-g_\varepsilon(0)>0. \] Thus $D_\varepsilon$ is smooth and positively curved. For small $\varepsilon$, both $C$ and $D_\varepsilon$ lie in $B^{2n}(2R)$. Set \[ \eta_\varepsilon=\varepsilon+ \sup_{B^{2n}(2R)}|g_\varepsilon-h_C|\longrightarrow0. \] Then $\{h_C\leq1-\eta_\varepsilon\}\subset D_\varepsilon \subset\{h_C\leq1+\eta_\varepsilon\}$, giving \eqref{eq:sandwich}. Radial convolution and $|z|^2$ preserve every orthogonal symmetry of $h_C$. \end{proof} \begin{proof}[Proof of Theorem~\ref{thm:main}] First suppose $0\in\interior C$. Both values on $C$ are finite and positive: for $s_k$ use \eqref{eq:finite}, and for $\cEH_k$ compare with inner and outer smooth balls and apply Proposition~\ref{thm:smooth}. Take the approximants from Lemma~\ref{lem:convex-approximation}. For $I_k=s_k$ and $I_k=\cEH_k$, monotonicity and scaling give \[ (1-\eta_\varepsilon)I_k(C)\leq I_k(D_\varepsilon) \leq(1+\eta_\varepsilon)I_k(C). \] Since $s_k(D_\varepsilon)=\cEH_k(D_\varepsilon)$ by Proposition~\ref{thm:smooth}, letting $\varepsilon\downarrow0$ proves $s_k(C)=\cEH_k(C)$. For arbitrary $p\in\interior C$, translation invariance gives $s_k(C-p)=\cEH_k(C-p)=\cEH_k(C)$. \end{proof} \begin{corollary}\label{cor:continuity} If $rB^{2n}(1)\subset C$ and the Hausdorff distance between convex bodies $C,D$ is $\delta0$, define \begin{equation}\label{eq:ellipsoid-quadratic} q_a(u)=\sum_{j,m\ne0}(a_j-a/m)|u_{jm}|^2. \end{equation} Since $q_a=4\pi(\HH_{E(\bm a)}-a\A)$ and $\HH_{E(\bm a)}(u)>0$ for $u\ne0$, the strict sublevel is \[ \{\Psi_{E(\bm a)}0$ with $ma_j0$ admitting an absolutely continuous $y:\T\to\partial C$ with \begin{equation}\label{r:eq:rc-generalized-brake} \begin{aligned} y(-t)&=\rho y(t),\\ \dot y(t)&=TJp(t),\qquad p(t)\in\partial h(y(t))\quad\text{a.e.} \end{aligned} \end{equation} Its full action is $T$. For smooth $C$, these are Hamiltonian trajectories. \end{proposition} \begin{proof} Jin--Lu's dual formula \cite[Proposition~7.1, $p=2$]{JL20} applies to the mean-zero $H^1$ brake loops $z=\Pi u$ with $\act(z)=1$; their $H_C=j_C^2$ and action~(1.8) agree with ours. It gives attainment and the brake minimum without boundary regularity. For the normalization, their proof, Steps~2--4, yields, at a minimizer of value $s$, a constant $c\in L$ such that \begin{equation}\label{r:eq:rc-minimizer-duality} q=s\Pi u+c\in\partial h^*(-Ju),\qquad h(q)\equiv s. \end{equation} Thus $y=q/\sqrt s$ satisfies \eqref{r:eq:rc-generalized-brake} with $T=s$. Conversely, \eqref{r:eq:rc-generalized-brake} gives $\act(y)=T$ and $\HH_C(\dot y/\sqrt T)=T$ by Euler and Fenchel identities. Positive homogeneity identifies the two infima in \eqref{r:eq:rc-beta}. The bounds follow from the dual estimates in Lemma~\ref{lem:quotient-finite} and the first positive brake mode. \end{proof} \begin{remark}\label{r:rem:rc-first-existing} This minimum equals Jin--Lu's real EH capacity \cite[Theorems~1.3, 1.11 and Remark~1.12]{JL20}. \end{remark} \begin{proposition}\label{r:prop:rc-covariance} For symplectic $A$ and $\widehat\rho=A\rho A^{-1}$, the map \[ \Lambda_\rho\longrightarrow\Lambda_{\widehat\rho},\qquad u\longmapsto\frac{Au}{\|Au\|_2} \] is an equivariant homeomorphism preserving the quotient for $C$ and $AC$. Moreover, $\beta_\rho(C-p)$ is independent of $p\in L\cap\interior C$. \end{proposition} \begin{proof} In the original Euclidean coordinates, \[ \begin{gathered} h_{AC}^*(q)=h_C^*(A^Tq),\qquad -J_0A=A^{-T}(-J_0),\\ \Pi(Au)=A\Pi u,\\ \HH_{AC}(Au)=\HH_C(u),\qquad \A(Au)=\A(u). \end{gathered} \] Homogeneity proves the first assertion; the inverse uses $A^{-1}$. The second follows from the first-capacity identification above and its fixed-translation invariance \cite[Propositions~1.7(i), 1.10]{JL20}. \end{proof} \section{The mod-two index and linking}\label{r:sec:index} All spaces are metrizable and all actions continuous. We use singular cohomology over $\F$. Take $E\Z_2=S(\ell^2(\R))$ with the antipodal action and put $B\Z_2=E\Z_2/\Z_2$. For a $\Z_2$-space $X$, write $X_{\Z_2}=E\Z_2\times_{\Z_2}X$ and $H^*_{\Z_2}(X)=H^*(X_{\Z_2};\F)$. The canonical projection is \[ p_X:X_{\Z_2}\longrightarrow B\Z_2,\qquad [e,x]\longmapsto[e]. \] Let $w\in H^1(B\Z_2;\F)$ be the nonzero class, so that $H^*(B\Z_2;\F)=\F[w]$. The cohomological index \cite{FR78} is \[ \FRrho(X)=\sup\{j\geq1:p_X^*(w^{j-1})\ne0\},\qquad \FRrho(\varnothing)=0. \] Below, $w$ also denotes its pullback to each Borel space. In relative cohomology, multiplication by $w$ means cup product with this absolute class. An equivariant map $f:X\to Y$ induces $f_{\Z_2}:X_{\Z_2}\to Y_{\Z_2}$ with $p_Y\circ f_{\Z_2}=p_X$. Hence $\FRrho(X)\leq\FRrho(Y)$, and equivariant homotopy equivalences preserve the index. A fixed point gives a section of $p_X$, so $X^{\Z_2}\ne\varnothing$ implies $\FRrho(X)=\infty$. For a centered invariant convex body define \begin{equation}\label{r:eq:selector} s_k(C,\rho)=\inf\{a>0:\FRrho(\{u\in\Lambda_\rho: \Psi_C(u)1$, the induced map $E\Z_2\times Y_*\to E\Z_2\times X$ has zero pullback in degree $k-1$. Naturality of transfer therefore forces $w^{k-1}=0$ on $(Y_*)_{\Z_2}$. For $k=1$, $X$ is path connected: the two sign rays connect through any nonzero point of $Y$. The degree-zero transfer of its connected two-sheeted covering is multiplication by $2=0$ in $\F$, so exactness forces $w\ne0$ on $X_{\Z_2}$. Both cases contradict $\FRrho(X)=k$. Induction proves the formula. \end{proof} \begin{lemma}[Relative cone class]\label{r:lem:two-relative-cone} Let $Y$ be a cone and $T=\R^d$ a trivial representation, $d\geq1$. Choose a generator $\mu\in H^{d-1}(T_*;\F)$ for $d>1$, and a nonconstant $\mu\in H^0(T_*;\F)$ for $d=1$. Under $(T_*)_{\Z_2}=B\Z_2\times T_*$, we also denote by $\mu$ its pullback from $T_*$. Let \[ \delta:H^*_{\Z_2}(T_*)\longrightarrow H^{*+1}_{\Z_2}((Y\oplus T)_*,T_*) \] be the connecting map. Then, for every $j\geq0$, \[ w^j\delta\mu\ne0 \quad\Longleftrightarrow\quad w^j\ne0\text{ on }(Y_*)_{\Z_2}. \] \end{lemma} \begin{proof} Apply Mayer--Vietoris to \[ (Y\oplus T)_*=(Y_*\oplus T)\cup(Y\oplus T_*). \] Since $Y$ contracts equivariantly to zero and $T_*\simeq S^{d-1}$, the K\"unneth formula identifies the restriction-difference map as \[ \begin{gathered} H^*_{\Z_2}(Y_*)\oplus H^*_{\Z_2}(T_*) \longrightarrow H^*_{\Z_2}(Y_*)\otimes H^*(T_*),\\ (\eta,w^j\mu)\longmapsto \eta\otimes1+w^j|_{Y_*}\otimes\mu. \end{gathered} \] The classes $1,\mu$ are independent, including when $d=1$. Thus $w^j\mu$ extends from $T_*$ to $(Y\oplus T)_*$ exactly when $w^j|_{Y_*}=0$. Exactness for the pair proves the assertion; $\delta$ commutes with multiplication by $w$. \end{proof} \begin{lemma}[Piercing]\label{r:lem:two-piercing} Let $Y$ be a cone in a finite-dimensional representation, $T=\R^d$ trivial with $d\geq1$, and $B$ an invariant compact neighborhood of zero in the ambient space of $Y\oplus T$. Suppose \[ \begin{gathered} \psi:Y\oplus T\longrightarrow\R_{\mathrm{sgn}}^q\oplus T \quad\text{is continuous and equivariant},\\ \psi(0,t)=(0,t). \end{gathered} \] Every invariant relatively open neighborhood $U\subset(Y\oplus T)_*$ of $\psi^{-1}(0)\cap\partial B$ satisfies \[ \FRrho(U)+q\geq\FRrho(Y_*). \] \end{lemma} \begin{proof} Put $X=(Y\oplus T)_*$ and suppose $l=\FRrho(U)<\infty$, $l+q<\FRrho(Y_*)$. Set \[ \begin{gathered} V=X\setminus\psi^{-1}(0),\qquad W=U\cup V,\\ X'=(X\cap B)\cup W,\qquad X''=(X\setminus B)\cup W. \end{gathered} \] Since $\psi|_{T_*}=\id$, we have $T_*\subset V\subset W$. As $W$ contains $X\cap\partial B$, the sets $X',X''$ form an open cover of $X$ with intersection $W$. Choose invariant balls about zero, the smaller contained in $\interior B$ and the larger containing $B$. Radial shrinking into the smaller ball and expansion beyond the larger give homotopy right inverses to the inclusions, as pairs relative to $T_*$. Both restrictions from $H^*_{\Z_2}(X,T_*)$ are therefore injective. Mayer--Vietoris also gives an injection \[ H^*_{\Z_2}(X,T_*)\longrightarrow H^*_{\Z_2}(W,W\cap T_*). \] Indeed, if $\alpha|_W=0$, the matching pair $(\alpha|_{X'},0)$ extends to $\beta$ on $(X,T_*)$; injectivity on $X''$ gives $\beta=0$, and injectivity on $X'$ gives $\alpha=0$. For $\eta=\delta\mu$, Lemma~\ref{r:lem:two-relative-cone} yields \[ w^{l+q}\eta\ne0\quad\text{in }H^*_{\Z_2}(W,W\cap T_*). \] On the target pair $((\R_{\mathrm{sgn}}^q\oplus T)_*,T_*)$, $w^q\delta\mu=0$ by sign stabilization and the relative cone lemma (also for $q=0$, when the pair is identical). Pullback by $\psi$ therefore gives $w^q\eta=0$ on $(V,V\cap T_*)$. The triple $W\cap T_*\subset V\subset W$ and the pair $(W,U)$ give the respective lifts \[ w^q\eta\in H^*_{\Z_2}(W,V),\qquad w^l\in H^*_{\Z_2}(W,U). \] Their relative product lies in $H^*_{\Z_2}(W,U\cup V)=0$, contradicting the displayed nonvanishing. \end{proof} On the brake space put \[ \Srho=\{v\in\Erho^+:\|v\|_E=1\}. \] Let $\Gammarho$ consist of the halfshift-equivariant homeomorphisms \begin{equation}\label{r:eq:gamma} \Theta(x)=e^{\theta_+(x)}x^++x^0+e^{\theta_-(x)}x^-+K(x) \qquad(x\in\Erho), \end{equation} with precisely the boundedness, compactness, and vanishing conditions of \eqref{eq:deformation}, now on $\Erho$. The coefficient bounds, compactness of $K$, and cutoff outside a ball ensure that these maps and their inverses preserve bounded sets. They preserve $\{\act>0\}$ because they fix its complement. Solving the formula for the inverse and composing maps preserve its form and conditions. Hence $\Gammarho$ is a group. Define \begin{equation}\label{r:eq:pseudoindex} \EHrho(A)=\inf_{\Theta\in\Gammarho} \inf_{\substack{O\supset A\cap\Theta(\Srho)\\ O\text{ open, invariant in }\Erho}}\FRrho(O). \end{equation} Take the Fourier truncations $F_N\subset\Erho$ with $|m|\leq N$ and orthogonal projections $P_N$. \begin{proposition}[Compact-cone linking]\label{r:prop:two-cone-linking} Let $V\subset\{\act>0\}$ be compact and invariant, and put $\mathcal W=\{rv:r>0,\ v\in V\}$. Then \[ \EHrho(\Erho^-+\Erho^0+\mathcal W)\geq\FRrho(\mathcal W). \] Positive fixed directions are allowed. \end{proposition} \begin{proof} The empty case is immediate. By inversion in $\Gammarho$, it suffices to prove $\FRrho(O)\geq\FRrho(\mathcal W)$ whenever \[ O\supset\Theta(\Erho^-+\Erho^0+\mathcal W)\cap \Srho, \qquad O\text{ open and invariant}. \] Write $F_N^\pm=F_N\cap \Erho^\pm$, $P_N^\pm=P_NP^\pm$. Compactness and positive action give $\min_{v\in V}\|P^+v\|>0$. Uniform convergence of $P_N^+$ on $V$ gives \[ \inf_{v\in V}\|P_N^+v\|\geq\varepsilon>0 \quad\text{for all sufficiently large }N. \] Decompose the negative truncation and define \[ \begin{aligned} F_N^-&=N_N\oplus T_N^-,\\ T_N&=T_N^-\oplus \Erho^0,\\ W_N&=P_N^+(\mathcal W)\cup\{0\},\\ Y_N&=N_N\oplus W_N. \end{aligned} \] Here $N_N$ consists of negative odd modes, and $T_N$ is the trivial representation of negative even modes and constants. Positive even modes remain in $W_N$. The full negative space has infinite index, but $q_N=\dim N_N$ is finite. Monotonicity and sign stabilization give \[ \begin{aligned} \FRrho((Y_N)_*)&=q_N+\FRrho((W_N)_*)\\ &\geq q_N+\FRrho(\mathcal W). \end{aligned} \] We claim, for large $N$, \[ P_N\Theta(Y_N\oplus T_N)\cap \Srho\subset O. \] Otherwise choose $x_N\in Y_N\oplus T_N$ with $P_N\Theta(x_N)\in \Srho\setminus O$. The identity of $\Theta$ outside a ball bounds $x_N$, and \[ x_N^-+x_N^0 =-e^{-\theta_-(x_N)}P_N^-K(x_N)-P^0K(x_N) \] is precompact. Write $x_N^+=r_NP_N^+v_N$, with $v_N\in V$, $r_N\geq0$. Since $\|x_N^+\|=r_N\|P_N^+v_N\|\geq\varepsilon r_N$, the sequence $(r_N)$ is bounded. Passing to a subsequence, we obtain \[ \begin{gathered} r_N\to r,\qquad v_N\to v,\qquad x_N\to x,\\ x^+=rP^+v,\qquad \Theta(x)\in \Srho\setminus O. \end{gathered} \] If $r=0$, then $x\in \Erho^-\oplus \Erho^0$ and $\Theta(x)=x\notin \Srho$. If $r>0$, then $x\in \Erho^-+\Erho^0+\mathcal W$. Both cases contradict the choice of $O$. Fix such an $N$ and put \[ \begin{gathered} \phi=P_N\Theta|_{F_N},\qquad B=\phi^{-1}\{z\in F_N:\|z\|\leq1\},\\ \psi=(P_N^-+P^0)\phi|_{Y_N\oplus T_N}: Y_N\oplus T_N\longrightarrow N_N\oplus T_N. \end{gathered} \] Since $\phi$ is the identity outside a ball and on $F_N^-\oplus \Erho^0$, $B$ is a compact invariant neighborhood of zero and $\psi|_{T_N}=\id$. For $Z=\psi^{-1}(0)\cap\partial B$, \[ \phi(Z)\subset \Srho\cap O. \] Choose an invariant relatively open neighborhood $U\subset(Y_N\oplus T_N)_*$ of $Z$ with $\phi(U)\subset O$. Piercing gives \[ \begin{aligned} \FRrho(O)+q_N&\geq\FRrho(U)+q_N\\ &\geq\FRrho((Y_N)_*)\\ &\geq\FRrho(\mathcal W)+q_N. \end{aligned} \] Since $q_N<\infty$, cancellation is valid, including when $\FRrho(\mathcal W)=\infty$. \end{proof} \begin{lemma}\label{r:lem:rc-fixed-minimum} The halfshift fixed sector has attained minimum \begin{equation}\label{r:eq:rc-fixed-minimum} \min_{\Lambda_\rho\cap\Fix(\sigma)}\Psi_C=2\beta_\rho(C). \end{equation} Thus $\{\Psi_C2\beta_\rho(C)$. \end{lemma} \begin{proof} Define \[ D_2:V_\rho\longrightarrow V_\rho\cap\Fix(\sigma),\qquad (D_2v)(t)=v(2t). \] Fixed velocities are $1/2$-periodic, so the inverse is $e\mapsto(t\mapsto e(t/2))$. Moreover, \[ \begin{aligned} \|D_2v\|_2&=\|v\|_2,\\ \HH_C(D_2v)&=\HH_C(v),\\ \A(D_2v)&=\tfrac12\A(v). \end{aligned} \] Apply Proposition~\ref{r:prop:rc-dual-minimum}. A fixed point gives a section of the Borel projection, so all powers of its universal class survive. \end{proof} \section{Real Hamiltonian minimax and comparison}\label{r:sec:eh} For an invariant convex body $C$, put \[ \mathcal F_\rho(C)=\{H\in\mathcal F(C):H\circ\rho=H\}. \] Its quadratic tails use the adapted norm of Section~\ref{r:sec:rc-dual}. Restrict $\Phi_H$ from \eqref{eq:Phi} to $\Erho$ and define \begin{equation}\label{r:eq:eh-definition} \begin{aligned} \cEH_k(H,\rho) &=\inf\{a\in\R:\EHrho(\{x\in\Erho:\Phi_H(x)0$; all these values are finite if the quadratic tail has slope $b>2\pi$. Every finite value is a critical value of $\Phi_H|_{\Erho}$, attained at a one-periodic brake orbit of $H$. For admissible $\rho$-invariant Hamiltonians $H,G$ with finite $k$th selectors and bounded difference, \[ |\cEH_k(H,\rho)-\cEH_k(G,\rho)|\leq\|H-G\|_\infty. \] The Hamiltonian selectors decrease under increasing $H$. The body selectors are monotone and homogeneous of degree two. \end{proposition} \begin{proof} The reflection $\mathcal R x(t)=\rho x(-t)$ is an isometric involution of $E$ preserving $\Phi_H$ and each Fourier summand. Hence \[ \nabla(\Phi_H|_{\Erho})(x)=\nabla\Phi_H(x)\in\Erho \qquad(x\in\Erho). \] The small-sphere estimate, compact nonlinear term, nonresonant linear inverse, and slice deformation of Proposition~\ref{prop:eh-facts} therefore restrict to $\Erho$. Its Fourier regularity argument gives $x\in H^1$ and $\dot x=J_0\nabla H(x)$ at every critical point. For finiteness let $\mathcal V_{\rho,2}=\{t\mapsto e^{4\pi J_0t}v:v\in L\}\subset\Erho^+$ be the second Fourier mode. Its sphere is fixed, so Proposition~\ref{r:prop:two-cone-linking} gives \[ \EHrho(\Erho^-\oplus L\oplus\mathcal V_{\rho,2})=\infty. \] Since $H(z)\geq b|z|^2-M_0$, \[ \Phi_H(x)\leq M_0+\sum_{m\leq0}(\pi m-b)|x_m|^2 +(2\pi-b)|x_2|^2\leq M_0 \] on this cone when $b>2\pi$. All cutoffs and normalized gradient flows in Proposition~\ref{prop:eh-facts} preserve $\Erho$ and commute with $\sigma$, hence belong to $\Gammarho$; the slice proof uses only index monotonicity and is valid for $\FRrho$. Sublevel inclusions give the remaining Hamiltonian claims. Dilation conjugates $\Gammarho$ to itself. The radial flow in Proposition~\ref{prop:eh-facts} restricts to $\Erho$ and maps $\Srho$ to $r\Srho$ at time $\log r$. Thus $H_r(z)=r^2H(z/r)$ and $\Phi_{H_r}(rx)=r^2\Phi_H(x)$ give homogeneity; inclusion of Hamiltonian classes gives body monotonicity. \end{proof} \begin{lemma}\label{r:lem:translation} For $p\in L$, $\cEH_k(C+p,\rho)=\cEH_k(C,\rho)$. \end{lemma} \begin{proof} The equivariant translation $T_px=x+p$ preserves $\act$ and conjugates $\Gammarho$ to itself. For $v\in\Srho$ and $0\leq t\leq1$, \[ \act(v+tp)=\tfrac12,\qquad \|v+tp\|_E^2=1+t^2|p|^2. \] Choose smooth $\chi:\R\to[0,1]$ and $\zeta:[0,\infty)\to[0,1]$ with $\chi=0$ on $(-\infty,0]$, $\chi=1$ near $1/2$, and $\zeta=1$ on $[0,1+|p|^2]$ with compact support. The time-one map of $\dot x=\chi(\act(x))\zeta(\|x\|_E^2)p$ agrees with $T_p$ on $\Srho$. It belongs to $\Gammarho$: its bounded displacement lies in $L$, and it fixes $\{\act\leq0\}$ and all sufficiently large points. Hence $T_p$ preserves the pseudoindex. The tail-center comparison in Proposition~\ref{prop:eh-facts} uses invariant radial annuli when both centers lie in $L$: for a tail $b|z-p|^2$ and $q\in L$, choose nonresonant $b'>b$; then $b'|z-q|^2$ dominates outside a large ball. Interpolation gives an invariant majorant with the new tail. Interchanging the centers proves cofinality. Apply this to $\Phi_{H(\cdot-p)}(T_px)=\Phi_H(x)$. \end{proof} \subsection{Restriction of the convex reduction}\label{r:sec:rc-convex-reduction} Throughout this subsection, assume $C=\rho C$, $0\in\interior C$, and that $\partial C$ is smooth with positive curvature. For $v\in\Erho^+$ define \begin{equation}\label{r:eq:rc-reduction} \begin{aligned} K_{\tau,\rho}(v) &=\min_{w\in\Erho^-,\ c\in L} \{F_\tau(v+w+c)+q_-(w)\},\\ W_{\tau,\rho}(v)&=w_{\tau,\rho}(v)+c_{\tau,\rho}(v),\\ x_{\tau,\rho}(v)&=v+W_{\tau,\rho}(v),\\ f_{\tau,\rho}(v)&=q_+(v)-K_{\tau,\rho}(v). \end{aligned} \end{equation} Here $F_\tau$ and $q_\pm$ are those of Section~\ref{sec:homogeneous-graph}. The involution $\mathcal R$ preserves the Fourier splitting and the minimized functional. Uniqueness of its ordinary minimizer therefore gives \[ K_{\tau,\rho}=K_\tau|_{\Erho^+},\qquad W_{\tau,\rho}=W_\tau|_{\Erho^+},\qquad f_{\tau,\rho}=f_\tau|_{\Erho^+}. \] In particular Lemma~\ref{lem:negative-graph-regularity}, including \eqref{eq:negative-graph-H1}, applies: the correction is Lipschitz from the $L^2$ norm into $H^1\cap\Erho$ and compact into $\Erho$. The full graph point is only required to lie in $H^{1/2}$. \begin{proposition}\label{r:prop:homogeneous-cone-index} For every $\tau>0$, the sublevels $\{f_{\tau,\rho}<0\}\subset\Erho^+$ and $\{\Psi_C<\tau\}\subset\Lambda_\rho$ admit continuous equivariant maps in both directions. Their indices satisfy, including infinite values, \[ \begin{aligned} \FRrho\{v\in\Erho^+:f_{\tau,\rho}(v)<0\} &=\FRrho\{v\in\Srho:f_{\tau,\rho}(v)<0\}\\ &=\FRrho\{u\in\Lambda_\rho:\Psi_C(u)<\tau\}. \end{aligned} \] \end{proposition} \begin{proof} For $p(v)=\tau\nabla h(x_{\tau,\rho}(v))$, the maps of Lemma~\ref{lem:homogeneous-cone-index} restrict to \[ v\longmapsto\frac{J_0p(v)}{\|J_0p(v)\|_2},\qquad u\longmapsto P^+\Pi u. \] Indeed $p(v)(-t)=\rho p(v)(t)$, hence $J_0p(v)$ is anti-brake. Its constant coefficient vanishes: $p(v)_0\in L$, and variation in the constant directions $L$ gives $p(v)_0=0$. The negative Euler equations \eqref{eq:negative-euler} remain valid. Thus \eqref{eq:cone-forward-identity} and \eqref{eq:cone-reverse-inequality} restrict without change and prove nonvanishing, positivity of action, and the strict sublevel inclusions. Neither argument differentiates a graph point. Index monotonicity in both directions and positive radial normalization prove the claim. \end{proof} \begin{proposition}\label{r:prop:graph-pseudoindex-bound} For $0<\tau<\tau'$ and $a>0$, there are $\Theta\in\Gammarho$ and $B_0<\infty$ with $\sup_{v\in\Srho}\|\Theta(v)\|_E\leq B_0$ such that, for every $H\in\mathcal F_\rho(C)$, \[ \int_0^1(H(x(t))-\tau h(x(t)))\,dt\leq1 \quad(x\in\Erho,\ \|x\|_E\leq B_0) \] implies \begin{equation}\label{r:eq:graph-pseudoindex-bound} \EHrho\{\Phi_Hs_k(C,\rho)$. Compact detection supplies invariant compact $K\subset\{\Psi_C0,\ v\in V\}$. Then \[ \FRrho(\mathcal W)\geq k,\qquad \HH_C(\dot v)0$, choose $b\max\{2\pi,M\max_{|z|=1}h(z)\}$, an invariant annular cutoff from $A$ to its majorant $B|z|^2$ gives $H\in\mathcal F_\rho(C)$ with $H\geq A$. Consequently \[ \sup_{\Erho^-+L+\mathcal W}\Phi_H\leq b+\varepsilon. \] Proposition~\ref{r:prop:two-cone-linking} and the minimax give $\cEH_k(C,\rho)\leq b+\varepsilon$. Let $\varepsilon\downarrow0$ and $b\downarrow s_k(C,\rho)$. For the lower comparison put $s=s_k(C,\rho)$ and follow the lower proof of Proposition~\ref{thm:smooth}, with $H\in\mathcal F_\rho(C)$ and the graph from Proposition~\ref{r:prop:graph-pseudoindex-bound}. Thus choose $c\notin\mathcal S(C)$ and $0\tau$, with $H\leq Mf(h)$ and the spectrum gap \eqref{eq:cap-data}. These choices use only the same vanishing and quadratic-growth hypotheses. Apply Lemma~\ref{lem:fixed-cap} with radius $B_0$, error $1$, and \[ b_\infty>\max\{2\pi,M\max_{|z|=1}h(z)\},\qquad b_\infty\notin\pi\N. \] Its smooth maximum and radial correction preserve $\rho$, so $\widehat H_\nu\in\mathcal F_\rho(C)$ for every $\nu\in[\tau,M]$. The lemma excludes $c$ as a critical value of $\Phi_{\widehat H_\nu}$ on $E$, hence on $\Erho$ by Proposition~\ref{r:prop:ham-properties}. Since $f(h)\leq h$, the initial cap satisfies \[ \begin{gathered} \EHrho\{\Phi_{\widehat H_\tau}c. \end{gathered} \] Proposition~\ref{r:prop:ham-properties} makes the selectors finite and continuous along this common-tail family. None can cross $c$. At the other endpoint $H\leq Mf(h)\leq\widehat H_M$, hence \[ \cEH_k(H,\rho)\geq\cEH_k(\widehat H_M,\rho)>c. \] Take the infimum over $H$ and let $c\uparrow s_k(C,\rho)$ through the complement of the closed null spectrum from Lemma~\ref{lem:spectrum-gap}. Now let $C$ be any centered invariant convex body. Proposition~\ref{r:prop:rc-dual-minimum} and Lemma~\ref{r:lem:rc-fixed-minimum} give $00$ with $B^{2n}(r)\subset C\subset B^{2n}(R)$. $s_1(C,\rho)=\inf\Psi_C=\beta$, and Lemma~\ref{r:lem:rc-fixed-minimum} gives infinite index above $2\beta$, hence $s_k(C,\rho)\leq2\beta$. Fix $00$, the compression $v(t)=e(t/2)$ satisfies $\HH_C(v)=\HH_C(e)$ and $\A(v)=2\A(e)$. Using $\HH_C(e)\geq0$ when $\A(e)\leq0$, we obtain \[ \HH_C(u)\geq\HH_C(e)\geq2\beta\A(e). \] Let $Q_N$ project onto the positive odd frequencies $m\leq N$. If $Q_Nu=0$, the Fourier formula and $\|u\|_2=1$ give \[ \begin{aligned} \A(o)&\leq\frac1{4\pi(N+1)},\\ \A(u)&=\A(e)+\A(o)\\ &\leq\frac{\HH_C(u)}{2\beta}+\frac1{4\pi(N+1)}. \end{aligned} \] For $u\in\{\Psi_C0$ is noncritical, Lemma~\ref{r:lem:dual-ps} gives $\delta,\eta>0$, $2\delta0$, the inclusions $B^{2n}(r)\subset C_j\subset B^{2n}(R)$ imply \[ 2=\ip{p_j}{y_j}=\sigma_{C_j}(p_j)\geq r|p_j|. \] Hence $(y_j)$ is bounded and equi-Lipschitz. A subsequence satisfies \[ y_j\to y\text{ uniformly},\qquad p_j\rightharpoonup^*p\text{ in }L^\infty. \] The limit $y:\T\to\partial C$ satisfies $y(-t)=\rho y(t)$ and $\dot y=sJp$ almost everywhere. For each $z\in\R^{2n}$, the subgradient inequality gives \[ h_{C_j}(z)\geq1+\ip{p_j(t)}{z-y_j(t)}. \] Integrate against nonnegative $L^1$ test functions and use locally uniform gauge convergence. A countable dense set of $z$, followed by continuity, gives $p(t)\in\partial h_C(y(t))$ almost everywhere. Finally, \[ \int_y\lambda_0=\frac s2\int_0^1\ip{y}{p}\,dt=s. \] \end{proof} \section{Real ellipsoids and symmetry examples}\label{r:sec:examples} \begin{proposition}\label{r:prop:real-ellipsoid} For $E(\bm a)$ with $02a_1. \end{cases} \end{equation} Its selectors therefore satisfy \eqref{r:eq:ellipsoid-intro}. \end{proposition} \begin{proof} The Fourier formulas \eqref{eq:ellipsoid-Fourier} restrict to the anti-brake coefficients $u_{jm}\in i\R$. The negative eigenspace of \eqref{eq:ellipsoid-quadratic} is thus the real span $V_b$ of the positive modes with $ma_j2a_1$, the second mode in the first coordinate is fixed. Sign stabilization and the fixed-point index therefore give \eqref{r:eq:ellipsoid-index}. \end{proof} For the ball, $a_j=\pi R^2$ and the sequence is \[ \underbrace{\pi R^2,\ldots,\pi R^2}_{n\text{ entries}}, 2\pi R^2,2\pi R^2,\ldots. \] More generally, the first $N=\#\{j:a_j<2a_1\}$ entries have their axis multiplicities; all later entries equal $2a_1$. For arbitrary convex bodies, only convergence to $2\beta$ is asserted. \begin{proof}[Proof of Theorem~\ref{r:thm:plateau}] Apply Proposition~\ref{r:prop:dual-terminal}, Proposition~\ref{r:prop:dual-spectrality}, Corollary~\ref{r:cor:dual-nonsmooth-spectrality}, and Proposition~\ref{r:prop:real-ellipsoid}. \end{proof} \begin{remark} The triangle with vertices $(-1,-1)$, $(-1,1)$, $(2,0)$ is $\rho_0$-invariant and contains zero, but is not centrally symmetric. Half-period translation still preserves its dual action. The symplectic shear $A_s(q,p)=(q+sp,p)$ gives \[ A_s\rho_0A_s^{-1}(q,p)=(q-2sp,-p). \] This involution is nonorthogonal for $s\ne0$; the selectors of $A_sE(\bm a)$ equal those of $E(\bm a)$ by Theorem~\ref{r:thm:main}. \end{remark} \begin{thebibliography}{BBLM23} \bibitem[BBLM23]{BBLM23} L.~Baracco, O.~Bernardi, C.~Lange, and M.~Mazzucchelli, \emph{On the local maximizers of higher capacity ratios}, \href{https://arxiv.org/abs/2303.13348v1}{arXiv:2303.13348v1}, 2023. Ann. Sc. Norm. Super. Pisa Cl. Sci. (5) \textbf{27} (2026), no.~2, 1057--1108. Result numbers refer to the arXiv version. \bibitem[EH90]{EH90} I.~Ekeland and H.~Hofer, \emph{Symplectic topology and Hamiltonian dynamics. II}, Math. Z. \textbf{203} (1990), 553--567. \href{https://www.ceremade.dauphine.fr/~ekeland/Articles/Sympectic1.pdf}{Full text}. \bibitem[FR78]{FR78} E.~R.~Fadell and P.~H.~Rabinowitz, \emph{Generalized cohomological index theories for Lie group actions with an application to bifurcation questions for Hamiltonian systems}, Invent. Math. \textbf{45} (1978), 139--174. \bibitem[GR24]{GR24} J.~Gutt and V.~G.~B.~Ramos, \emph{The equivalence of Ekeland--Hofer and equivariant symplectic homology capacities}, \href{https://arxiv.org/abs/2412.09555v1}{arXiv:2412.09555v1}, 2024. \bibitem[JL20]{JL20} R.~Jin and G.~Lu, \emph{Representation formula for symmetrical symplectic capacity and applications}, Discrete Contin. Dyn. Syst. \textbf{40} (2020), 4705--4765, \href{https://arxiv.org/abs/1903.00678v3}{arXiv:1903.00678v3}. Result numbers refer to the arXiv version. \bibitem[Mat25]{Mat25} S.~Matijevi\'c, \emph{Positive ($S^1$-equivariant) symplectic homology of convex domains, higher capacities, and Clarke's duality}, \href{https://arxiv.org/abs/2410.13673v2}{arXiv:2410.13673v2}, 2025. \bibitem[SL22]{SL22} K.~Shi and G.~Lu, \emph{Higher $P$-symmetric Ekeland--Hofer capacities}, Commun. Pure Appl. Anal. \textbf{21} (2022), 1049--1070, \href{https://arxiv.org/abs/2102.00600v1}{arXiv:2102.00600v1}. \end{thebibliography} \end{document}