Nonlinear Frequency-Momentum Topology and Doubling of Multifold Exceptional Points Tsuneya Yoshida Abstract Even in the linear limit, the topology of multifold (also called higher-order) exceptional points across the Brillouin zone has lacked a general characterization, leaving the doubling theorem essentially limited to two-fold exceptional points. Here, we establish the doubling theorem of n n -fold exceptional points [EP n n s ( n = 2 , 3 , … n=2,3,\ldots )] for systems where nonlinearity enters through eigenvalues. To this end, we introduce new topological invariants, termed frequency-momentum winding numbers, which characterize nonlinear EP n n s in m m -band systems throughout the Brillouin zone for arbitrary n n and m m ( m ≥ n m\geq n ). These invariants enable a unified proof of the doubling theorem in the absence of symmetry and under several symmetry constraints, including parity-time ( P ​ T PT ) and charge-conjugation-parity symmetries. Furthermore, even in the linear limit, the frequency-momentum winding number indicates ℤ \mathbb{Z} topology of P ​ T PT -symmetric EP 2 2 s which is beyond the previously reported ℤ 2 \mathbb{Z}_{2} topology. The frequency-momentum winding numbers can also be extended to a class of coupled resonators in which nonlinearity enters via the eigenvectors, whereas the spectrum is determined by a nonlinear scalar equation for the frequency. Introduction.— Doubling of topological quasiparticles, enforced by the vanishing total topological charge in the Brillouin zone (BZ), plays a fundamental role in topological physics Nielsen and Ninomiya 1981a ; Nielsen and Ninomiya 1981b . A representative example is provided by Weyl semimetals Murakami 2007 ; Wan et al. 2011 ; Burkov and Balents 2011 ; Bevan et al. 1997 ; Armitage et al. 2018 , which host pairs of Weyl nodes, i.e., linear band-touching points with quantized Chern number. The doubling of Weyl nodes underlies characteristic phenomena such as surface Fermi arcs Xu et al. 2015 ; Huang et al. 2015a ; Weng et al. 2015 ; Silaev and Volovik 2012 and anomalous transport responses Nielsen and Ninomiya 1983 ; Yang et al. 2011 ; Burkov 2014 ; Liu et al. 2018 ; Nakatsuji et al. 2015 ; Wang et al. 2018 ; Son and Spivak 2013 ; Huang et al. 2015b ; Zhang et al. 2016 ; Xiong et al. 2015 . Recently, extensions of the doubling theorem to exceptional points in non-Hermitian systems have been explored Yang et al. 2021 ; Yoshida et al. 2025a . At exceptional points, the coalescence of eigenvalues and eigenvectors leads to a dispersion relation with a fractional exponent Berry 2004 ; Heiss 2004 ; Rotter 2009 ; Shen et al. 2018 ; Kawabata et al. 2019a ; Kozii and Fu 2024 ; Bergholtz et al. 2021 ; Ashida et al. 2020 , which is generalized to multifold (or higher-order) exceptional points Heiss 2012 ; Demange and Graefe 2011 ; Delplace et al. 2021 ; Mandal and Bergholtz 2021 ; Sayyad and Kunst 2022 ; Montag and Kunst 2024 ; Yoshida et al. 2025a . The doubling of two-fold exceptional points is proved Yang et al. 2021 and accounts for bulk Fermi arcs Kozii and Fu 2024 ; Doppler et al. 2016 ; Zhou et al. 2018 ; Tang et al. 2021 ; Hofmann et al. 2020 ; Gao et al. 2015 ; Yoshida et al. 2018 ; Yoshida et al. 2019 ; Zyuzin and Zyuzin 2018 . While the above phenomena are described in terms of eigenvalue equations, topological physics has recently been extended to nonlinear systems, including classical metamaterials Leykam and Chong 2016 ; Benzaouia et al. 2022 ; Veenstra et al. 2024 ; Smirnova et al. 2020 ; Xie et al. 2026 ; Sahin et al. 2025 ; Kuzmiak et al. 1994 ; Chern et al. 2006 ; Wang et al. 2008 ; Huang et al. 2009 ; Lee and Wright 2016 ; Ozawa et al. 2019 and interacting quantum matters Gulevich et al. 2017 ; Mostaan et al. 2022 ; Kozii and Fu 2024 ; Yoshida et al. 2018 . In such systems, nonlinearity can enter either through the eigenvectors Sone et al. 2022 ; Sone et al. 2024 ; Sone et al. 2025 ; Sone and Hatsugai 2026 ; Jezequel and Delplace 2022 ; Schindler et al. 2025 or through the eigenvalues Isobe et al. 2024 ; Silveirinha 2015 ; Raman and Fan 2010 ; Feng et al. 2025 ; Xiao et al. 2016 . Nonlinearity of eigenvectors affects the stability of edge states, giving rise to nonlinear excitations such as chiral solitons Ablowitz et al. 2014 ; Gulevich et al. 2017 ; Snee and Ma 2019 ; Ezawa 2022a , while nonlinearity of eigenvalues effectively increases the number of bands Raman and Fan 2010 ; Feng et al. 2025 . Nonlinear topology has been further explored in non-Hermitian systems Bai et al. 2023 ; Bai et al. 2022 ; Fang et al. 2025 ; Wingenbach et al. 2025 ; Ezawa 2022b ; de Castro and Benalcazar 2025 ; Kawabata and Nakamura 2025 ; Hamanaka 2026 ; Yoshida et al. 2025b ; Feng et al. 2025 , particularly through the interplay between nonlinearity and exceptional points. The nonlinearity of eigenvalues can induce exceptional points Yoshida et al. 2025b and even increase their multiplicity 1 1 1 For a Lorentz dispersive medium, Ref. Feng et al. 2025 maps a system with nonlinearity of eigenvalues to an effective linear system. Feng et al. 2025 ; Bai et al. 2023 ; Bai et al. 2022 ; Fang et al. 2025 . Despite the above progress, the doubling theorem of exceptional points in nonlinear systems has yet to be established, especially for n -fold exceptional points (EP n s). The central obstacle is the absence of topological invariants that characterize EP n s in nonlinear systems with an arbitrary number of bands. Furthermore, even in the linear limit, no topological invariant is known to characterize EP n s ( n\geq 3 ) across the BZ in m -band systems Yoshida et al. 2026 . As a consequence, the existing doubling theorem is restricted to EP 2 s in the linear regime Yang et al. 2021 . We hereby establish the doubling theorem of EP n s with arbitrary n ( n\geq 2 ) for systems where nonlinearity enters through eigenvalues (i.e., frequency). To this end, we introduce new topological invariants, termed frequency-momentum (FM) winding numbers, which characterize nonlinear EP n s in m -band systems throughout the BZ for arbitrary n and m . The FM winding numbers enable a proof of the doubling theorem both in the absence and presence of symmetry, including parity-time ( PT ) and charge-conjugation-parity ( CP ) symmetries. In particular, they imply the doubling of EP 3 s in the two-dimensional BZ and EP 4 s in the three-dimensional BZ under PT symmetry, and of EP 5 s and EP 7 s when both PT and CP symmetries are imposed. Furthermore, even in the linear limit, the FM winding numbers reveal previously unrecognized stability of EP 2 s with PT symmetry; our invariant indicates \mathbb{Z} topology refining the existing \mathbb{Z}_{2} topological characterization. The FM winding numbers can also be extended to a class of coupled resonators with nonlinearity of eigenvectors, whose spectrum is determined by a nonlinear scalar equation for frequency. Frequency-momentum winding number.— We consider a system described by a nonlinear eigenvalue equation \displaystyle F(\omega_{n},\bm{k})|\psi_{n}(\omega_{n},\bm{k})\rangle \displaystyle=0, (1) where F(\omega_{n},\bm{k}) is an N\times N matrix, |\psi(\omega_{n},\bm{k})\rangle are eigenvectors, and \omega_{n}\in\mathbb{C} are eigenvalues ( n=1,2,\ldots ). The real vector \bm{k} denotes momentum (or wavenumber) in the D -dimensional BZ. The eigenvalues \omega_{n} are obtained by solving \displaystyle\mathrm{det}F(\omega_{n},\bm{k}) \displaystyle=0. (2) We suppose that \mathrm{det}F(\omega,\bm{k}) becomes independent of \bm{k} for large \omega , as exemplified by \mathrm{det}F(\omega,\bm{k})\sim\omega^{r} for |\omega|\to\infty with integer r . Equations ( 1 ) and ( 2 ) describe band structures of metamaterials with dispersive media Kuzmiak et al. 1994 ; Chern et al. 2006 ; Huang et al. 2009 ; Lee and Wright 2016 ; Raman and Fan 2010 ; Feng et al. 2025 and quasiparticles in quantum systems Kozii and Fu 2024 ; Yoshida et al. 2018 . Figure 1: Illustration of the doubling theorem of nonlinear EP n s in 2n -dimensional \omega - \bm{k} space. Panel (a) shows that an EP n with W_{2n-1}=1 (red dot) is accompanied by another EP n with W_{2n-1}=-1 (blue dot) due to the periodicity of the BZ and \bm{k} independence of F(\omega,\bm{k}) for |\omega|\to\infty [see Eqs. ( 5 ) and ( Nonlinear Frequency-Momentum Topology and Doubling of Multifold Exceptional Points )]. In this panel, the axes of \mathrm{Im}\omega and k_{j} ( j=3,\ldots,2n-2 ) are omitted. Panel (b) illustrates the winding structure of \bm{d}(\omega,\bm{k}) [Eq. ( Nonlinear Frequency-Momentum Topology and Doubling of Multifold Exceptional Points )] in the vicinity of the EP n . The green (orange) surface separates the regions of positive and negative d_{2n-1} ( d_{2n} ). The red surface denotes the region where d_{1}=d_{2}=\ldots=d_{2n-2}=0 holds, forming a manifold of EP n^{\prime} with n^{\prime}=n-1 . The vector \bm{d} winds the (2n-1) -sphere in the vicinity of the EP n emerging on the two-dimensional manifold of EP n^{\prime} s with n^{\prime}=n-1 . We begin by considering the codimension of nonlinear EP n s without symmetry in frequency-momentum ( \omega - \bm{k} ) space. An EP n emerges at \omega_{0}\in\mathbb{C} and \bm{k}_{0} if and only if the following conditions are satisfied: \displaystyle\mathrm{det}F(\omega,\bm{k}_{0})|_{\omega=\omega_{0}} \displaystyle=0, \displaystyle\partial_{\omega}\mathrm{det}F(\omega,\bm{k}_{0})|_{\omega=\omega_{0}} \displaystyle=0, \displaystyle\partial^{n-1}_{\omega}\mathrm{det}F(\omega,\bm{k}_{0})|_{\omega=\omega_{0}} \displaystyle=0, (3) and \partial^{n}_{\omega}\mathrm{det}F(\omega,\bm{k}_{0})|_{\omega=\omega_{0}}\neq 0 . Since \mathrm{det}F is a complex function, the emergence of an EP n requires satisfying 2n real-valued constraints. Therefore, the codimension is \delta_{\omega\text{-}\bm{k}}=2n in D+2 -dimensional \omega - \bm{k} space; i.e., EP n s emerge in the D=2(n-1) -dimensional BZ. In passing, eigenvectors coalesce at (\omega,\bm{k})=(\omega_{0},\bm{k}_{0}) unless additional conditions are satisfied. The above codimension argument leads to FM topology of EP n s for systems with an arbitrary number of bands. We rewrite Eq. ( Nonlinear Frequency-Momentum Topology and Doubling of Multifold Exceptional Points ) as \bm{d}(\omega_{0},\bm{k}_{0})=\bm{0} with \displaystyle\bm{d}(\omega,\bm{k}) \displaystyle=\Big(\mathrm{Re}[\mathrm{det}F],\mathrm{Im}[\mathrm{det}F], \displaystyle\quad\quad\ldots,\mathrm{Re}[\partial^{n-1}_{\omega}\mathrm{det}F],\mathrm{Im}[\partial^{n-1}_{\omega}\mathrm{det}F]\Big). (4) The vector \bm{d}(\omega,\bm{k}) is nonzero on the M -sphere S^{M} with M=\delta_{\omega\text{-}\bm{k}}-1 enclosing the EP n in \omega - \bm{k} space. Thus, \bm{d}(\omega,\bm{k}) defines a map from S^{M} in the \omega - \bm{k} space to the unit vector \hat{\bm{d}}(\omega,\bm{k})=\bm{d}(\omega,\bm{k})/\lVert\bm{d}(\omega,\bm{k})\rVert on S^{M} whose topology in \omega - \bm{k} space is classified by the homotopy group \pi_{M}(S^{M})=\mathbb{Z} . This topology is characterized by the FM winding number \displaystyle W_{M} \displaystyle=\frac{1}{A_{M}}\oint d^{M}\bm{p}\epsilon^{\mu_{1}\ldots\mu_{M+1}}f_{\mu_{1}\cdots\mu_{M+1}} (5) with \displaystyle f_{\mu_{1}\cdots\mu_{M+1}} \displaystyle=\hat{d}_{\mu_{1}}\partial_{1}\hat{d}_{\mu_{2}}\partial_{2}\hat{d}_{\mu_{3}}\cdots\partial_{M}\hat{d}_{\mu_{M+1}}, (6) \displaystyle A_{M} \displaystyle=2\pi^{\frac{M+1}{2}}\Big[\Gamma\big(\frac{M+1}{2}\big)\Big]^{-1}. (7) The integral is computed over S^{M} in \omega - \bm{k} space parameterized by \bm{p} . The symbol \epsilon^{\mu_{1}\cdots\mu_{M+1}} denotes an anti-symmetric tensor satisfying \epsilon^{12\cdots M+1}=1 . The derivative with respect to p_{i} is denoted by \partial_{i} . The area of the M -dimensional sphere A_{M} is expressed in terms of the gamma function, which satisfies \Gamma(z+1)=z\Gamma(z) , \Gamma(1)=1 , and \Gamma(1/2)=\sqrt{\pi} ; its explicit form is given in Sec. S1 of the Supplemental Material Sup . Notably, the above FM topological argument reveals the hierarchical structure of nonlinear EP n s. A manifold of EP n s emerges on a manifold of EP n^{\prime} s whose dimension exceeds that of EP n s by 2(n-n^{\prime}) as illustrated in Fig. 1 (b) for n^{\prime}=n-1 in the D -dimensional BZ with D=2n-2 . This structure follows directly from Eqs. ( Nonlinear Frequency-Momentum Topology and Doubling of Multifold Exceptional Points ) and ( 5 ). Doubling of nonlinear EP n s without symmetry.— The FM winding number [see Eq. ( 5 )] leads to the doubling theorem of nonlinear EP n s. We consider EP n s in 2n -dimensional \omega - \bm{k} space [i.e., in the D=2(n-1) dimensional BZ] for an m -band system ( m\geq n ). Because of the periodicity of the BZ, the total winding number W_{\mathrm{tot},M} with M=2n-1 is determined by boundary terms at |\omega|\to\infty (see Fig. 1 ). Since \bm{d}(\omega,\bm{k}) is \bm{k} independent for |\omega|\to\infty , the integrant including momentum derivatives vanishes \displaystyle W_{\mathrm{tot},M} \displaystyle\propto\sum_{j}\oint_{S^{M}_{j}}d^{M}\bm{p}\,\epsilon^{\mu_{1}\ldots\mu_{M+1}}f_{\mu_{1}\cdots\mu_{M+1}} \displaystyle\quad=\oint_{|\omega|\to\infty}d^{M}\bm{p}\,\epsilon^{\mu_{1}\ldots\mu_{M+1}}f_{\mu_{1}\cdots\mu_{M+1}}=0. (8) This result implies that any EP n with W_{2n-1}=1 is necessarily paired with a corresponding EP n with W_{2n-1}=-1 , thereby establishing the doubling theorem of EP n for arbitrary m ( m\geq n ), which had not been proved even in the linear case. The FM winding number does not require any information about the band structure and is not affected by fake EP n s reported in a linear system Yoshida et al. 2026 . In addition, our FM winding number reproduces the doubling theorem of EP2s in the linear regime (see Sec. S2 of the Supplemental Material Sup ), which was previously obtained by using the discriminant of the characteristic polynomial Yang et al. 2021 . Cases with symmetry: an overview.— FM winding numbers also imply the doubling of nonlinear EP n s protected by symmetry. Symmetry \omega n Codim. \delta_{\omega\text{-}\bm{k}} \bm{d} none \mathbb{C} e/o 2n Eq. ( Nonlinear Frequency-Momentum Topology and Doubling of Multifold Exceptional Points ) PT \mathbb{R} e/o n Eq. ( 12 )( \ddagger ) CP 0 e n Eq. ( 14 ) o n-1 Eq. ( 14 ) PT and CP 0 e n/2 Eq. ( 15a ) o (n-1)/2 Eq. ( 15b ) chiral \mathrm{i}\mathbb{R} e/o n Eq. ( 12 )(*) PT^{\dagger} \mathbb{C} e/o 2n Eq. ( Nonlinear Frequency-Momentum Topology and Doubling of Multifold Exceptional Points ) CP^{\dagger} \mathrm{i}\mathbb{R} e/o n Eq. ( 12 )(*)( \ddagger ) sublattice 0 e n Eq. ( 14 ) o n-1 Eq. ( 14 ) psH \mathbb{R} e/o n Eq. ( 12 ) Table 1: The codimension \delta_{\omega\text{-}\bm{k}} and the vector \bm{d} for each case of symmetry, which specify the explicit form of the FM winding number in Eq. ( 5 ). The first column denotes symmetry, where pseudo-Hermiticity is denoted by “psH”. The second column indicates the frequency \omega where the symmetries are closed. The third column indicates the parity of n ; “e” (“o”) denotes even (odd) n , and “e/o” denotes even or odd n . We consider symmetry operations that square to 1 . The codimension and degrees of freedom of \omega indicates that nonlinear EP n s emerge in the D -dimensional BZ with D=\delta_{\omega\text{-}\bm{k}}-\mathrm{dim}\omega ( \mathrm{dim}\omega=2 for \omega\in\mathbb{C} , \mathrm{dim}\omega=1 for \omega\in\mathbb{R}\cup\mathrm{i}\mathbb{R} , and \mathrm{dim}\omega=0 for \omega=0 ). For the cases marked with the symbol “(*)”, \bm{d} is defined by replacing F\to\tilde{F}:=-\mathrm{i}F and \omega\to\tilde{\omega}:=-\mathrm{i}\omega . In the cases marked with the symbol “( \ddagger )”, EP2s are characterized by a \mathbb{Z} invariant rather than the previously proposed \mathbb{Z}_{2} invariant in the linear limit, revealing the stability of these EP2s that was previously unrecognized. We consider PT , CP , and chiral symmetries \displaystyle U_{\mathrm{PT}}F^{*}(\omega,\bm{k})U^{-1}_{\mathrm{PT}} \displaystyle=F(\omega^{*},\bm{k}), (9) \displaystyle U_{\mathrm{CP}}F^{T}(\omega,\bm{k})U^{-1}_{\mathrm{CP}} \displaystyle=-F(-\omega,\bm{k}), (10) \displaystyle U_{\Gamma}F^{\dagger}(\omega,\bm{k})U^{-1}_{\Gamma} \displaystyle=-F(-\omega^{*},\bm{k}). (11) We assume that the unitary matrices satisfy U_{\mathrm{PT}}U^{*}_{\mathrm{PT}}=1 , U_{\mathrm{CP}}U^{*}_{\mathrm{CP}}=1 , and U^{2}_{\Gamma}=1 , respectively. These symmetries impose additional conditions on \omega and \mathrm{det}F(\omega,\bm{k}) . Taking these conditions into account, we obtain the codimension \delta_{\omega\text{-}\bm{k}} and \bm{d}(\omega,\bm{k}) (see Table 1 ) as discussed below. Table 1 shows that symmetry-protected EP n s emerge in \delta_{\omega\text{-}\bm{k}} -dimensional \omega - \bm{k} space, i.e., the D -dimensional BZ with D=\delta_{\omega\text{-}\bm{k}}-\mathrm{dim}\omega and \mathrm{dim}\omega which takes \mathrm{dim}\omega=2 , 1 , and 0 for \omega\in\mathbb{C} , \omega\in\mathbb{R}\cup\mathrm{i}\mathbb{R} , and \omega=0 , respectively. As is the case without symmetry, EP n s possess FM topology classified by homotopy group \pi_{M}(S^{M})=\mathbb{Z} with M=\delta_{\omega\text{-}\bm{k}}-1 . The corresponding topological invariant is the winding number in Eq. ( 5 ) with \bm{d}(\omega,\bm{k}) listed in Table 1 . Due to the periodicity of the BZ and the \bm{k} independence of \mathrm{det}F for |\omega|\to\infty , the total winding number of symmetry-protected EP n s vanishes [see Eq. ( Nonlinear Frequency-Momentum Topology and Doubling of Multifold Exceptional Points )]. Thus, a symmetry-protected EP n with W_{M}=1 for M=\delta_{\omega\text{-}\bm{k}}-1 is accompanied by another one with W_{M}=-1 , thereby establishing the doubling theorem. The above argument also extends to cases with PT^{\dagger} symmetry, CP^{\dagger} symmetry, sublattice symmetry, and pseudo-Hermiticity (see Table 1 ), as discussed in Sec. S3 of the Supplemental Material Sup . Codimension under PT symmetry.— Because PT transformation maps \omega to \omega^{*} , the constraint in Eq. ( 9 ) is closed for \omega\in\mathbb{R} . Thus, we focus on symmetry-protected EP n s that emerge on the real axis of frequency where \mathrm{det}F(\omega,\bm{k}) is real. In this case, the conditions of symmetry-protected EP n s [see Eq. ( Nonlinear Frequency-Momentum Topology and Doubling of Multifold Exceptional Points )] is written as n real constraints \bm{d}(\omega_{0},\bm{k}_{0})=\bm{0} with \displaystyle\bm{d}(\omega,\bm{k}) \displaystyle=\Big(\mathrm{det}F,\partial_{\omega}\mathrm{det}F,\ldots,\partial^{n-1}_{\omega}\mathrm{det}F\Big). (12) Thus, their codimension is \delta_{\omega\text{-}\bm{k}}=n , indicating that symmetry-protected EP n s emerge in the D=(n-1) -dimensional BZ. The above vector \bm{d}(\bm{k}) reveals FM topology of symmetry-protected EP n s characterized by the winding number in Eq. ( 5 ), thereby leading to the doubling theorem. In passing, the condition \bm{d}(\omega_{0},\bm{k})=\bm{0} in FM space reveals the hierarchical structure of EP n s with PT symmetry. A manifold of EP n s emerges on a manifold of EP n^{\prime} s whose dimension exceeds that of the EP n s by (n-n^{\prime}) [see Eq. ( 12 )]. In the linear limit, it is known that a \mathbb{Z}_{2} invariant characterizes EP2s with PT symmetry, whereas our topological invariant takes arbitrary integer values. Our numerical analysis demonstrates the presence of EP2s with PT symmetry beyond the previously discussed \mathbb{Z}_{2} topology (for more details, see Sec. S4 of the Supplemental Material Sup ). Codimension under CP symmetry.— Because CP transformation maps \omega to -\omega , the constraint in Eq. ( 10 ) is closed for \omega=0 , indicating that CP symmetry protects EP n s only at \omega_{0}=0 . In addition, Eq. ( 10 ) leads to \displaystyle\partial^{l}_{\omega}\mathrm{det}F(\omega,\bm{k})\Big|_{\omega=\omega_{0}} \displaystyle=0 (13) for l and N of opposite parity. Here, N denotes the size of the matrix F . Thus, one of the conditions of EP n s, \partial^{n}_{\omega}\mathrm{det}F(\omega,\bm{k}_{0})|_{\omega=\omega_{0}}\neq 0 , is compatible only for n=2s+2 when N is even, and for n=2s+3 when N is odd ( s=0,1,2,\ldots ). The remaining conditions for symmetry-protected EP n s lead to \bm{d}(\omega_{0},\bm{k}_{0})=\bm{0} with \displaystyle\bm{d}(\omega_{0},\bm{k}) \displaystyle=\Big(\mathrm{Re}[\mathrm{det}F],\mathrm{Im}[\mathrm{det}F],\mathrm{Re}[\partial^{2}_{\omega_{0}}\mathrm{det}F], \displaystyle\quad\quad\quad\quad\quad\quad\quad\ldots,\partial^{2s}_{\omega_{0}}\mathrm{Im}[\mathrm{det}F]\Big) (14a) for n=2s+2 (i.e., even N ) and \displaystyle\bm{d}(\omega_{0},\bm{k}) \displaystyle=\Big(\mathrm{Re}[\partial_{\omega_{0}}\mathrm{det}F],\mathrm{Im}[\partial_{\omega_{0}}\mathrm{det}F], \displaystyle\quad\quad\mathrm{Re}[\partial^{3}_{\omega_{0}}\mathrm{det}F],\ldots,\mathrm{Im}[\partial^{2s+1}_{\omega_{0}}\mathrm{det}F]\Big) (14b) for n=2s+3 (i.e., odd N ). Here, we have used a simplified notation, \partial^{l}_{\omega_{0}}\mathrm{det}F:=\partial^{l}_{\omega}\mathrm{det}F(\omega,\bm{k})|_{\omega=\omega_{0}} . We stress that in the above definition, \omega is fixed at \omega_{0}=0 in contrast to that in Eq. ( 12 ). Equation ( 14 ) indicates that the codimension of symmetry-protected EP n s is \delta_{\omega\text{-}\bm{k}}=n [ \delta_{\omega\text{-}\bm{k}}=n-1 ] for even [odd] n . The vector \bm{d}(\bm{k}) reveals FM topology of symmetry-protected EP n s characterized by the winding number in Eq. ( 5 ), thereby establishing the doubling theorem. In passing, the condition \bm{d}(\omega_{0},\bm{k})=\bm{0} in FM space reveals the hierarchical structure of EP n s with CP symmetry. A manifold of EP n s emerges on a manifold of EP n^{\prime} s whose dimension exceeds that of EP n s by n-n^{\prime} , provided that n and n^{\prime} have the same parity [see Eq. ( 14 )]. Codimension under PT and CP symmetries.— Because these symmetries are closed only at \omega=0 , we set \omega_{0}=0 . Due to PT symmetry, \mathrm{det}F(\omega_{0},\bm{k}) is real. In addition, due to CP symmetry, Eq. ( 13 ) holds, indicating that symmetry-protected EP n s with n=2s+2 [ n=2s+3 ] are compatible for even N [odd N ] ( s=0,1,2,\ldots ). This fact leads to the constraints of symmetry-protected EP n s, \bm{d}(\omega_{0},\bm{k})=\bm{0} with \displaystyle\bm{d}(\omega_{0},\bm{k}) \displaystyle=\Big(\mathrm{det}F,\partial^{2}_{\omega_{0}}\mathrm{det}F,\ldots,\partial^{2s}_{\omega_{0}}\mathrm{det}F\Big) (15a) for n=2s+2 and \displaystyle\bm{d}(\omega_{0},\bm{k}) \displaystyle=\Big(\partial_{\omega_{0}}\mathrm{det}F,\partial^{3}_{\omega_{0}}\mathrm{det}F,\ldots,\partial^{2s+1}_{\omega_{0}}\mathrm{det}F\Big) (15b) for n=2s+3 . We recall that \omega is fixed to \omega_{0}=0 and that we have used a simplified notation, \partial^{l}_{\omega_{0}}\mathrm{det}F:=\partial^{l}_{\omega}\mathrm{det}F(\omega,\bm{k})|_{\omega=\omega_{0}} . Equation ( 15 ) indicates that the codimension of symmetry-protected EP n s is \delta_{\omega\text{-}\bm{k}}=n/2 [ \delta_{\omega\text{-}\bm{k}}=(n-1)/2 ] for n=2s+2 ( n=2s+3 ) with s=0,1,2,\ldots . The above vector \bm{d}(\bm{k}) reveals FM topology of symmetry-protected EP n s with n\geq 4 characterized by the winding number in Eq. ( 5 ) which leads to the doubling theorem. For n=2,3 , symmetry-protected EP n s are characterized by the \mathbb{Z}_{2} invariant \nu \displaystyle(-1)^{\nu} \displaystyle=\prod_{k\in\{k_{+0},k_{-0}\}}\mathrm{sgn}\big[d_{1}(\omega_{0},k)\big] (16) with d_{1}(\omega_{0},k)=\mathrm{det}F(\omega_{0},k) [ d_{1}(\omega_{0},k)=\partial_{\omega_{0}}\mathrm{det}F(\omega_{0},k) ] for n=2 [ n=3 ]. The momentum slightly displaced from the EP n is denoted by k_{\pm 0}:=k_{0}\pm\delta k with \delta k>0 . This topological invariant leads to the doubling theorem due to the periodicity of the BZ. In passing, the condition \bm{d}(\omega_{0},\bm{k})=\bm{0} in FM space reveals the hierarchical structure of EP n s with PT and CP symmetries. A manifold of EP n s emerges on a manifold of EP n^{\prime} s whose dimension exceeds that of EP n s by (n-n^{\prime})/2 , provided that n and n^{\prime} have the same parity [see Eq. ( 15 )]. Codimension under chiral symmetry.— Chiral symmetry ( 11 ) is closed for \omega\in\mathrm{i}\mathbb{R} . Thus, we focus on symmetry-protected EP n s that emerge on the imaginary axis. In addition, the chiral symmetry is rewritten as \displaystyle U_{\Gamma}\tilde{F}^{\dagger}(\tilde{\omega},\bm{k})U^{\dagger}_{\Gamma} \displaystyle=\tilde{F}(\tilde{\omega}^{*},\bm{k}) (17) with \tilde{F}(\tilde{\omega},\bm{k})=-\mathrm{i}F(\mathrm{i}\tilde{\omega},\bm{k}) and \tilde{\omega}=-\mathrm{i}\omega Kawabata et al. 2019b ; Kawabata et al. 2019c . Because the transpose of an arbitrary matrix does not affect its determinant, the problem is reduced to the case with PT symmetry. Thus, the codimension is \delta_{\omega\text{-}\bm{k}}=n . The vector \bm{d}(\tilde{\omega},\bm{k}) is obtained in the same way as for deriving Eq. ( 12 ) except that F and \omega are replaced by \tilde{F} and \tilde{\omega} , respectively. Application to a toy model with PT symmetry .— Figure 2: (a) Winding number W_{2} for each mesh. The winding number is W_{2}=1 ( W_{2}=-1 ) on the surface of red (blue) cuboids. Here, we take a 10\times 10\times 10 mesh. The winding number is computed based on the method proposed in Ref. Fukui et al. 2005 . For the computation of W_{2} , each cuboid is further subdivided into a 10\times 10\times 10 mesh. (b) Eigenvalues \omega\in\mathbb{R} as functions of \bm{k}=(k_{1},k_{2}) . The data for k_{2}=2.80176 ( k_{1}=2.15984 ) are represented by red (blue) points. The others are represented by gray points. The blue (black) point denotes an EP2 (EP3). Data are obtained for (\delta_{1},\delta_{2})=(0.01,1) . Numerical analysis of a two-dimensional toy model with PT symmetry demonstrates the doubling of nonlinear EP 3 s whose codimension is \delta_{\omega\text{-}\bm{k}}=3 . The model is described by F(\omega,\bm{k})=\omega\mbox{1}\hskip-2.5pt\mbox{l}-H_{0}(\bm{k})-\Sigma(\omega) with a 4\times 4 matrix H_{0}(\bm{k}) exhibiting EP 3 and \Sigma(\omega) including nonlinearity of \omega (for explicit forms of H_{0}(\bm{k}) and \Sigma(\omega) , see Sec. S5 of the Supplemental Material Sup ). The EP 3 s are characterized by the winding number W_{2} with vector \bm{d} [see Eqs. ( 5 ) and ( 12 )]. Figure 2 (a) displays W_{2} for each point in the \omega - \bm{k} space. This figure indicates that the points of W_{2}=1 are accompanied by the ones of W_{2}=-1 . In addition, the point with W_{2}=1 captures a triple root [see Fig. 2 (b)]. These data numerically confirm the doubling theorem of EP 3 s with PT symmetry. Here, we note that the topology of arbitrary \bm{d} is rephrased in terms of a Hermitian Hamiltonian: H_{\bm{d}}(\bm{q})=\sum_{i}d_{i}\gamma_{i} with Hermitian matrices \gamma ’s satisfying \gamma_{i}\gamma_{j}+\gamma_{j}\gamma_{i}=2\delta_{ij} and \bm{q}=(\omega,k_{1},\ldots,k_{D})^{T} (for more details, see Sec. S6 of the Supplemental Material Sup ). This mapping allows us to compute W_{2} by employing the method proposed in Refs. Fukui et al. 2005 . Topology of an EP3 in coupled resonators with nonlinearity of eigenvectors .— FM topology protects an EP3 in a class of coupled resonators with saturable gain Bai et al. 2023 ; Bai et al. 2022 ; Fang et al. 2025 described by H_{|\psi\rangle}|\psi\rangle=\omega|\psi\rangle with \displaystyle H_{|\psi\rangle} \displaystyle=\left(\begin{array}[]{cc}\omega_{\mathrm{A}}+\mathrm{i}g_{\mathrm{A}}(|\psi_{\mathrm{A}}|)&\kappa_{1}\\ \kappa_{1}&\omega_{\mathrm{B}}-\mathrm{i}l_{\mathrm{B}}\end{array}\right). Here, g_{\mathrm{A}}(|\psi_{A}|) denotes a nonlinear saturable gain. The resonant frequencies of each resonator are denoted by \omega_{\mathrm{A}} and \omega_{\mathrm{B}} , respectively. The linear loss is denoted by l_{\mathrm{B}} . After reaching the stable state, the eigenvalues \omega satisfy p(\omega)=0 with p(\omega)=(\omega-\omega_{\mathrm{A}})\big[(\omega-\omega_{\mathrm{B}})^{2}+l^{2}_{\mathrm{B}}\big]-\kappa^{2}_{1}(\omega-\omega_{\mathrm{B}}) . For \omega_{\mathrm{A}}=\omega_{\mathrm{B}}=0 , eigenvalues are written as \omega=0 and \omega=\pm\sqrt{\kappa^{2}_{1}-l^{2}_{\mathrm{B}}} , indicating that the system hosts an EP3 despite the 2\times 2 matrix H_{|\psi\rangle} . Our approach reveals the topological protection of these EP3s. In the vicinity of (\omega,\omega_{\mathrm{A}},l_{\mathrm{B}})=(0,0,\kappa_{1}) with \kappa_{1}>0 , \bm{d}(\delta\omega,\delta\omega_{\mathrm{A}},\delta l_{\mathrm{B}}) for |\delta\omega|,|\delta\omega_{\mathrm{A}}|,|\delta l_{\mathrm{B}}|\ll\kappa_{1} is given by \bm{d}^{T}=\Big(-\kappa^{2}_{1}\delta\omega_{\mathrm{A}},2\kappa_{1}\delta l_{\mathrm{B}},6\delta\omega-2\delta\omega_{\mathrm{A}}\Big) from which we obtain the winding number W_{2}=-1 . A similar analysis applies to an EP 3 in Hermitian nonlinear coupled resonators with Kerr nonlinearity Fang et al. 2025 . Nonlinear Kramers pairs.— While we have focused on the cases of U_{\mathrm{PT}}U^{*}_{\mathrm{PT}}=1 so far, the doubling theorem for cases of U_{\mathrm{PT}}U^{*}_{\mathrm{PT}}=-1 remains an open question due to nonlinear Kramers pairs. If |\psi_{n}\rangle is an eigenstate with eigenvalue \omega_{n} , then U_{\mathrm{PT}}\mathcal{K}|\psi_{n}\rangle is also an eigenstate with eigenvalue \omega^{*}_{n} where \mathcal{K} denotes the complex conjugate operator. This degeneracy prevents a sign change of \mathrm{det}F(\omega,\bm{k}) in the vicinity of EP n s. A similar argument also holds for CP^{\dagger} symmetry since it reduces to PT symmetry. Conclusion.— We have established the doubling theorem of EP n s for m -band systems with nonlinearity of frequency. The introduced FM winding numbers reveal a global topological structure of EP n s over the entire BZ and underlie the doubling theorem both in the absence and presence of symmetry. In the linear limit, our framework suggests \mathbb{Z} topology for PT -symmetric EP2s beyond the previously discussed \mathbb{Z}_{2} topology. The same framework directly covers dispersive metamaterials and quantum quasiparticles with finite lifetime. It can be extended to a class of coupled resonators with nonlinearity of eigenvectors whose spectrum is determined by a nonlinear scalar equation for the frequency. The extension to cases with nonlinear Kramers pairs remains an open question. Acknowledgments.— The author thanks Emil J. Bergholtz and Tomáš Bzdušek for the previous collaboration Yoshida et al. 2026 , which provided part of the motivation for this work. This work is supported by JSPS KAKENHI Grant Nos. JP21K13850, JP23KK0247, JP25K07152, and JP25H02136 as well as JSPS Bilateral Program No. JPJSBP120249925. Supplemental Material: Nonlinear Frequency-Momentum Topology and Doubling of Multifold Exceptional Points S1 Area of the M -dimensional sphere For an integer n , the gamma function is written as \displaystyle\Gamma(n) \displaystyle=(n-1)!, (S1) \displaystyle\Gamma\big(n+\frac{1}{2}\big) \displaystyle=\frac{(2n)!}{2^{2n}n!}\sqrt{\pi}. (S2) Thus, the area of the (2M+1) -dimensional sphere is \displaystyle A_{2M+1} \displaystyle=\frac{2\pi^{M+1}}{M!}, (S3) and that of the 2M -dimensional sphere is \displaystyle A_{2M} \displaystyle=\frac{2^{2M+1}\pi^{M}M!}{(2M)!}. (S4) S2 Frequency-momentum winding number in the linear limit In the linear limit, the FM winding number coincides with the discriminant winding number Yang et al. 2021 and thus reproduces the doubling theorem of EP 2 s. We see the coincidence of these two topological invariants. We consider a toy model given by a 2\times 2 matrix \displaystyle H(\bm{k}) \displaystyle=\begin{pmatrix}0&1\\ z_{\eta}(\bm{k})&0\end{pmatrix} (S5) with an integer \eta and z_{\eta}(\bm{k})=(k_{1}+\mathrm{i}k_{2})^{\eta} for \eta>0 , z_{\eta}(\bm{k})=(k_{1}-\mathrm{i}k_{2})^{|\eta|} for \eta<0 and z_{0}(\bm{k})=k^{2}_{1}+k^{2}_{2} for \eta=0 . This model hosts an EP 2 at \omega_{0}=0 and \bm{k}_{0}=\bm{0} characterized by the discriminant winding number W_{\mathrm{D}}=\eta Yang et al. 2021 ; Yoshida et al. 2025a . The FM winding number W_{3} can be computed from \displaystyle W_{3}=\sum_{(\omega,\bm{k})\in\{(0,\bm{k}_{l})\}}\mathrm{sgn}\Big(\mathrm{det}[J(\omega,\bm{k})]\Big) (S6) with \displaystyle J(\omega,\bm{k}) \displaystyle=\frac{\partial(d_{1},d_{2},d_{3},d_{4})}{\partial(\omega_{\mathrm{R}},\omega_{\mathrm{I}},k_{1},k_{2})}, (S7) \displaystyle d_{1}+\mathrm{i}d_{2} \displaystyle=\omega^{2}-z(\bm{k}), (S8) \displaystyle d_{3}+\mathrm{i}d_{4} \displaystyle=2\omega, (S9) and \omega=\omega_{\mathrm{R}}+\mathrm{i}\omega_{\mathrm{I}} ( \omega_{\mathrm{R}},\omega_{\mathrm{I}}\in\mathbb{R} ). For \eta>0 [ \eta<0 ], \bm{k}_{l}=(k_{1l},k_{2l}) satisfies k_{1l}+\mathrm{i}k_{2l}=e^{\frac{2\mathrm{i}\pi l}{\eta}} [ k_{1l}-\mathrm{i}k_{2l}=e^{\frac{2\mathrm{i}\pi l}{|\eta|}} ] with l=0,1,\ldots,\eta-1 . Here, \mathrm{sgn}(x) takes 1 ( -1 ) for x>0 ( x<0 ). At the points (0,\bm{k}_{l}) , the normalized vector satisfies \bm{d}/\lVert\bm{d}\rVert=(-1,0,0,0) . By computing the Jacobian, we have \mathrm{det}J(0,\bm{k}_{l})=4\eta^{2}\mathrm{sgn}(\eta) for \eta\neq 0 which indicates that the FM winding number is W_{3}=\eta . For \eta=0 , the FM winding number vanishes W_{3}=0 because d_{2}=0 for arbitrary (\omega,\bm{k}) . Because these two topological invariants are independent of the details of H(\bm{k}) , this coincidence holds for an arbitrary model. S3 Other cases of symmetry FM topology can also be extended to cases with PT^{\dagger} , CP^{\dagger} , sublattice symmetries, and pseudo-Hermiticity (ps H) \displaystyle U_{\mathrm{PT}^{\dagger}}F^{T}(\omega,\bm{k})U^{-1}_{\mathrm{PT}^{\dagger}} \displaystyle=F(\omega,\bm{k}), (S10) \displaystyle U_{\mathrm{CP}^{\dagger}}F^{*}(\omega,\bm{k})U^{-1}_{\mathrm{CP}^{\dagger}} \displaystyle=-F(-\omega^{*},\bm{k}), (S11) \displaystyle U_{\mathrm{S}}F(\omega,\bm{k})U^{-1}_{\mathrm{S}} \displaystyle=-F(-\omega,\bm{k}), (S12) \displaystyle U_{\mathrm{psH}}F^{\dagger}(\omega,\bm{k})U^{-1}_{\mathrm{psH}} \displaystyle=F(\omega^{*},\bm{k}). (S13) The codimension \delta_{\omega\text{-}\bm{k}} and the vector \bm{d} are summarized in Table 1 . As discussed below, these results are obtained by noting that the transpose of a matrix does not change its determinant \mathrm{det}F^{T}=\mathrm{det}F , which reduces the problem to cases with symmetries defined in Eqs. ( 9 ), ( 10 ), and ( 11 ). S3.1 Codimension under PT^{\dagger} symmetry PT^{\dagger} symmetry is closed for arbitrary \omega\in\mathbb{C} . In addition, the transpose of an arbitrary matrix does not affect its determinant. Thus, the codimension and topology of EP n s are the same as those in the case with no symmetry. S3.2 Codimension under CP^{\dagger} symmetry CP^{\dagger} symmetry is closed for arbitrary \omega\in\mathrm{i}\mathbb{R} . In addition, CP^{\dagger} symmetry is rewritten as \displaystyle U_{CP^{\dagger}}\tilde{F}^{*}(\tilde{\omega},\bm{k})U^{\dagger}_{CP^{\dagger}} \displaystyle=\tilde{F}(\tilde{\omega}^{*},\bm{k}) (S14) with \tilde{F}(\tilde{\omega},\bm{k})=-\mathrm{i}F(\mathrm{i}\tilde{\omega},\bm{k}) and \tilde{\omega}=-\mathrm{i}\omega . The above constraint is nothing but Eq. ( 9 ). Thus, the codimension is \delta_{\omega\text{-}\bm{k}}=n . The vector \bm{d}(\tilde{\omega},\bm{k}) is obtained in the same way as Eq. ( 12 ) with F and \omega replaced by \tilde{F} and \tilde{\omega} . S3.3 Codimension under sublattice symmetry Sublattice symmetry is closed for \omega=0 . In addition, the transpose of an arbitrary matrix does not affect its determinant. Thus, the codimension and topology of EP n s are the same as those in the case with CP symmetry. S3.4 Codimension under pseudo-Hermiticity Pseudo-Hermiticity is closed for \omega\in\mathbb{R} . In addition, the transpose of an arbitrary matrix does not affect its determinant. Thus, the codimension and topology of EP n s are the same as those in the case with PT symmetry. S4 EP2 with PT symmetry In the linear limit, our winding number W_{1} in Eq. ( 5 ) with \bm{d}=(\mathrm{det}F,\partial_{\omega}\mathrm{det}F) reveals the robustness of EP2s with PT symmetry, which was not understood from previous \mathbb{Z}_{2} invariants. For concreteness, we consider toy models \displaystyle H_{4\times 4} \displaystyle=\left(\begin{array}[]{cccc}0&1&0&0\\ k&0&V&0\\ 0&V&0&1\\ 0&0&k-\kappa&0\end{array}\right) and \displaystyle H_{2\times 2} \displaystyle=\left(\begin{array}[]{cc}0&1\\ m_{0}-\sin k&0\\ \end{array}\right) with real values V , \kappa , and m_{0} . Figure S1: (a) [(c)]: Color plot of the argument \mathrm{Arg}[d_{1}+\mathrm{i}d_{2}]/\pi as a function of k and \omega ( \omega\in\mathbb{R} ) of the Hamiltonian H_{4\times 4} for \kappa=0.5 [ \kappa=0 ]. In these panels, the FM winding number takes the value W_{1}=-1 along the path enclosing points denoted by black arrows. (b) [(d)]: Eigenvalues as functions of k of the Hamiltonian H_{4\times 4} for \kappa=0.5 [ \kappa=0 ]. The real and imaginary parts of eigenvalues are represented by red and blue lines. The data are obtained for V=0.3 . Figure S2: (a) [(c)]: Color plot of the argument \mathrm{Arg}[d_{1}+\mathrm{i}d_{2}]/\pi as a function of k and \omega ( \omega\in\mathbb{R} ) of the Hamiltonian H_{2\times 2} for m_{0}=0.5 [ m_{0}=1.1 ]. In these panels, the FM winding number takes the value W_{1}=-1 ( W_{1}=1 ) along the path enclosing points denoted by the black (white) arrow. (b) [(d)]: Eigenvalues as functions of k of the Hamiltonian H_{2\times 2} for m_{0}=0.5 [ m_{0}=1.1 ]. The real and imaginary parts of eigenvalues are represented by red and blue lines. The winding number and the band structure of H_{4\times 4} are displayed in Fig. S1 . Figures S1 (a) and S1 (b) indicate the presence of two EP2s characterized by W_{1}=-1 for \kappa=0.5 . As \kappa decreases, these two EP2s approach each other. Notably, however, they do not annihilate each other because they have the same winding number. This behavior is in sharp contrast to the EP2s of H_{2\times 2} displayed in Fig. S2 , where the pair annihilation is observed. S5 Details of the toy model Data displayed in Fig. 2 are obtained for the following toy model \displaystyle F(\omega,\bm{k})=\omega\mbox{1}\hskip-2.5pt\mbox{l}-H_{0}(\bm{k})-\Sigma(\omega) (S21) with \displaystyle H_{0}(\bm{k}) \displaystyle=\left(\begin{array}[]{cccc}0&1&0&0\\ 0&0&1&0\\ 0&0&0&\delta_{1}\\ 3\sin k_{2}&\cos k_{1}+\cos k_{2}-\frac{3}{2}&\delta_{1}&\delta_{2}\end{array}\right), \displaystyle\Sigma(\omega) \displaystyle=0.1\omega\tanh(\omega)\mathrm{diag}(1,-1,1,-1). (S26) For \delta_{1}=0 , H_{0}(\bm{k}) is decoupled into 3\times 3 and 1\times 1 blocks. The former reduces to a 3\times 3 Jordan block at \bm{k}=\bm{0} and exhibits an EP 3 . In order to compute the FM winding number W_{2} , we map the vector \bm{d} to a Hermitian Hamiltonian H_{\bm{d}}=d_{1}\sigma_{1}+d_{2}\sigma_{2}+d_{3}\sigma_{3} with Pauli matrices \sigma_{i} . By employing the method proposed in Ref. Fukui et al. 2005 , we numerically obtain the Chern number of the Hermitian Hamiltonian H_{\bm{d}} and thereby obtain W_{2} (see also Appendix S6 ). S6 Doubling of nonlinear EP n s in terms of Hermitian Hamiltonians The FM winding numbers can be reinterpreted as topological invariants of Hermitian Hamiltonians, implying that the doubling of nonlinear EP n s can be understood in terms of the doubling of topological nodes of the corresponding Hermitian Hamiltonian. With a given vector \bm{d} that vanishes on an EP n [e.g., see Eqs. ( Nonlinear Frequency-Momentum Topology and Doubling of Multifold Exceptional Points ) and ( 12 )], we consider a Hermitian Hamiltonian \displaystyle H_{\bm{d}}(\bm{q}) \displaystyle=\sum_{i=1,2,\ldots,\mathrm{dim}\bm{d}}d_{i}\gamma_{i} (S27) with Hermitian matrices \gamma_{i} satisfying \gamma_{i}\gamma_{j}+\gamma_{j}\gamma_{i}=2\delta_{ij} and \bm{q} begin the vector in \omega - \bm{k} space. Here, the dimension of the \omega - \bm{\bm{k}} space is equal to \dim\bm{d} because EP n s emerge as zero-dimensional manifolds. Sets of \gamma ’s are obtained by taking the direct product of Pauli matrices \sigma_{i} ( i=1,2,3 ) and the identity matrix \sigma_{0} . For instance, a set of \gamma_{i} is written as \displaystyle\{\sigma_{1},\sigma_{2},\sigma_{3}\}, (S28) for 2\times 2 matrices and \displaystyle\{\sigma_{1}\otimes\sigma_{0},\sigma_{2}\otimes\sigma_{0},\sigma_{3}\otimes\sigma_{1},\sigma_{3}\otimes\sigma_{2},\sigma_{3}\otimes\sigma_{3}\}, (S29) for 4\times 4 matrices. For m_{\bm{d}}\times m_{\bm{d}} matrices, the number of possible \gamma matrices is 2m_{\bm{d}}+1 . The \gamma matrices satisfy \displaystyle\mathrm{tr}\big(\gamma_{1}\gamma_{2}\cdots\gamma_{2m_{\mathbb{d}}+1}\big) \displaystyle=(2\mathrm{i})^{m_{\mathbb{d}}}. (S30) The vanishing of the vector \bm{d} at the EP n can be detected as the gap-closing of the Hermitian Hamiltonian H_{\bm{d}}(\bm{q}) . This is because the eigenvalues of H_{\bm{d}}(\bm{q}) are given by E_{\bm{d}}=\pm\sqrt{\lVert\bm{d}\rVert} . When \mathrm{dim}\bm{d} is odd, the symmetry class of H_{\bm{d}} is class A because all possible \gamma ’s are exhausted. In contrast, when \mathrm{dim}\bm{d} is even, the symmetry class of H_{\bm{d}} is class AIII due to the presence of one additional \gamma matrix. The above fact leads to the correspondence between the FM winding number and the Chern number (winding number) of the Hermitian Hamiltonian H_{\bm{d}} when \mathrm{dim}\bm{d} is odd (even). Specifically, when \mathrm{dim}\bm{d} is odd, the Chern number is written as \displaystyle\mathcal{C}_{n_{\mathrm{C}}} \displaystyle=-\frac{1}{2^{2n_{\mathrm{C}}+1}}\frac{1}{n_{\mathrm{C}}!}\Big(\frac{\mathrm{i}}{2\pi}\Big)^{n_{\mathrm{C}}}\int\mathrm{tr}\Big[Q(dQ)^{2n_{\mathrm{C}}}\Big] (S31) with n_{\mathrm{C}}=(\mathrm{dim}\bm{d}-1)/2 , Q=\sum_{i}\hat{d}_{i}\gamma_{i} and \hat{d}=\bm{d}/\lVert\bm{d}\rVert . Here, Q is the flattened Hamiltonian associated with H_{\bm{d}} whose eigenvalues are \pm 1 and dQ denotes the exterior derivative of Q . The integrant is rewritten as \displaystyle\mathrm{tr}\Big[Q(dQ)^{2n_{\mathrm{C}}}\Big] \displaystyle=\mathrm{sgn}\left(\begin{array}[]{ccc}1&\cdots&2n_{\mathrm{C}}\\ \mu_{1}&\cdots&\mu_{2n_{\mathrm{C}}}\end{array}\right)\mathrm{tr}[\gamma_{i_{1}}\cdots\gamma_{i_{2n_{\mathrm{C}}+1}}]\big(\hat{d}_{i_{1}}\partial_{\mu_{1}}\hat{d}_{i_{2}}\cdots\partial_{\mu_{2n_{\mathrm{C}}}}\hat{d}_{i_{2n_{\mathrm{C}}+1}}\big)d^{2n_{\mathrm{C}}}\bm{q} \displaystyle=\mathrm{sgn}\left(\begin{array}[]{ccc}1&\cdots&2n_{\mathrm{C}}\\ \mu_{1}&\cdots&\mu_{2n_{\mathrm{C}}}\end{array}\right)\mathrm{sgn}\left(\begin{array}[]{ccc}1&\cdots&2n_{\mathrm{C}}+1\\ i_{1}&\cdots&i_{2n_{\mathrm{C}}+1}\end{array}\right)\mathrm{tr}[\gamma_{1}\cdots\gamma_{2n_{\mathrm{C}}+1}]\big(\hat{d}_{i_{1}}\partial_{\mu_{1}}\hat{d}_{i_{2}}\cdots\partial_{\mu_{2n_{\mathrm{C}}}}\hat{d}_{i_{2n_{\mathrm{C}}+1}}\big)d^{2n_{\mathrm{C}}}\bm{q} \displaystyle=(2\mathrm{i})^{n_{\mathrm{C}}}\mathrm{sgn}\left(\begin{array}[]{ccc}i_{2}&\cdots&i_{2n_{\mathrm{C}}+1}\\ \sigma(i_{2})&\cdots&\sigma(i_{2n_{\mathrm{C}}+1})\end{array}\right)\mathrm{sgn}\left(\begin{array}[]{ccc}1&\cdots&2n_{\mathrm{C}}+1\\ i_{1}&\cdots&i_{2n_{\mathrm{C}}+1}\end{array}\right)\big(\hat{d}_{i_{1}}\partial_{1}\hat{d}_{\sigma(i_{2})}\cdots\partial_{2n_{\mathrm{C}}}\hat{d}_{\sigma(i_{2n_{\mathrm{C}}+1})}\big)d^{2n_{\mathrm{C}}}\bm{q} \displaystyle=(2\mathrm{i})^{n_{\mathrm{C}}}\mathrm{sgn}\left(\begin{array}[]{cccc}1&2&\cdots&2n_{\mathrm{C}}+1\\ i_{1}&\sigma(i_{2})&\cdots&\sigma(i_{2n_{\mathrm{C}}+1})\end{array}\right)\big(\hat{d}_{i_{1}}\partial_{1}\hat{d}_{\sigma(i_{2})}\cdots\partial_{2n_{\mathrm{C}}}\hat{d}_{\sigma(i_{2n_{\mathrm{C}}+1})}\big)d^{2n_{\mathrm{C}}}\bm{q}, where the summation is taken over repeated indices. The symbol “ \mathrm{sgn}(\cdots) ” denotes 1 ( -1 ) when the second row is an even (odd) permutation of the first row. The symbol “ d^{2n_{\mathrm{C}}}\bm{q} ” is defined as d^{2n_{\mathrm{C}}}\bm{q}=dq_{1}dq_{2}\ldots dq_{2n_{\mathrm{C}}} . Here \sigma(i) in the fourth line denotes the permutation of indices so that \displaystyle\mathrm{sgn}\left(\begin{array}[]{ccc}1&\cdots&2n_{\mathrm{C}}\\ \mu_{1}&\cdots&\mu_{2n_{\mathrm{C}}}\end{array}\right)\big(\hat{d}_{i_{1}}\partial_{\mu_{1}}\hat{d}_{i_{2}}\cdots\partial_{\mu_{2n_{\mathrm{C}}}}\hat{d}_{i_{2n_{\mathrm{C}}+1}}\big) \displaystyle=\sum_{\sigma}\mathrm{sgn}\left(\begin{array}[]{ccc}i_{2}&\cdots&i_{2n_{\mathrm{C}}+1}\\ \sigma(i_{2})&\cdots&\sigma(i_{2n_{\mathrm{C}}+1})\end{array}\right)\big(\hat{d}_{i_{1}}\partial_{1}\hat{d}_{\sigma(i_{2})}\cdots\partial_{2n_{\mathrm{C}}}\hat{d}_{\sigma(i_{2n_{\mathrm{C}}+1})}\big) holds. The symbol “ \sum_{\sigma} ” denotes the summation of all possible permutations. From the third to the fourth line of Eq. ( S6 ), we have used Eq. ( S30 ). Substituting Eq. ( S6 ) into Eq. ( S31 ), we obtain \displaystyle(-1)^{n_{\mathrm{C}}+1}\mathcal{C}_{n_{\mathrm{C}}} (S44) \displaystyle=\frac{(2n_{\mathrm{C}})!}{\pi^{n_{\mathrm{C}}}n_{\mathrm{C}}!2^{2n_{\mathrm{C}}+1}}\int d^{2n_{\mathrm{C}}}\bm{q}\epsilon^{i_{1}\ldots i_{2n_{\mathrm{C}}}}f_{i_{1}\ldots i_{2n_{\mathrm{C}}}} \displaystyle=W_{2n_{\mathrm{C}}}. (S45) From the second to the third lines, we have used Eqs. ( 5 ) and ( S4 ). This relation indicates the correspondence between FM winding numbers W_{2n_{\mathrm{C}}} and Chern numbers \mathcal{C}_{n_{\mathrm{C}}} of Hermitian Hamiltonians. When \mathrm{dim}\bm{d} is even, the winding number of H_{\bm{d}} is written as \displaystyle\mathcal{W}_{2n_{\mathrm{W}}+1} \displaystyle=\frac{(-1)^{n_{\mathrm{W}}}n_{\mathrm{W}}!}{2(2\pi\mathrm{i})^{n_{\mathrm{W}}+1}(2n_{\mathrm{W}}+1)!}\int\mathrm{tr}\Big[\gamma_{2n_{\mathrm{W}}+3}Q(dQ)^{2n_{\mathrm{W}}+1}\Big] (S46) with n_{\mathrm{W}}=\mathrm{dim}\bm{d}/2-1 , Q=\sum_{i=1,\ldots,2n_{\mathrm{W}}+2}\hat{d}_{i}\gamma_{i} and \hat{d}=\bm{d}/\lVert\bm{d}\rVert . The integrant is rewritten as \displaystyle\mathrm{tr}\Big[\gamma_{2n_{\mathrm{W}}+3}Q(dQ)^{2n_{\mathrm{W}}+1}\Big] \displaystyle=\mathrm{sgn}\left(\begin{array}[]{ccc}1&\cdots&2n_{\mathrm{W}}+1\\ \mu_{1}&\cdots&\mu_{2n_{\mathrm{W}}+1}\end{array}\right)\mathrm{tr}[\gamma_{2n_{\mathrm{W}}+3}\gamma_{i_{1}}\cdots\gamma_{i_{2n_{\mathrm{W}}+2}}]\big(\hat{d}_{i_{1}}\partial_{\mu_{1}}\hat{d}_{i_{2}}\cdots\partial_{\mu_{2n_{\mathrm{W}}+1}}\hat{d}_{i_{2n_{\mathrm{W}}+2}}\big)d^{2n_{\mathrm{W}}+1}\bm{q} \displaystyle=\mathrm{sgn}\left(\begin{array}[]{ccc}1&\cdots&2n_{\mathrm{W}}+1\\ \mu_{1}&\cdots&\mu_{2n_{\mathrm{W}}+1}\end{array}\right)\mathrm{sgn}\left(\begin{array}[]{ccc}1&\cdots&2n_{\mathrm{W}}+2\\ i_{1}&\cdots&i_{2n_{\mathrm{W}}+2}\end{array}\right)\mathrm{tr}[\gamma_{1}\cdots\gamma_{2n_{\mathrm{W}}+3}]\big(\hat{d}_{i_{1}}\partial_{\mu_{1}}\hat{d}_{i_{2}}\cdots\partial_{\mu_{2n_{\mathrm{W}}+1}}\hat{d}_{i_{2n_{\mathrm{W}}+2}}\big)d^{2n_{\mathrm{W}}+1}\bm{q} \displaystyle=(2\mathrm{i})^{n_{\mathrm{W}}+1}\mathrm{sgn}\left(\begin{array}[]{ccc}i_{2}&\cdots&i_{2n_{\mathrm{W}}+2}\\ \sigma(i_{2})&\cdots&\sigma(i_{2n_{\mathrm{W}}+2})\end{array}\right)\mathrm{sgn}\left(\begin{array}[]{ccc}1&\cdots&2n_{\mathrm{W}}+2\\ i_{1}&\cdots&i_{2n_{\mathrm{W}}+2}\end{array}\right)\big(\hat{d}_{i_{1}}\partial_{1}\hat{d}_{\sigma(i_{2})}\cdots\partial_{2n_{\mathrm{W}}+1}\hat{d}_{\sigma(i_{2n_{\mathrm{W}}+2})}\big)d^{2n_{\mathrm{W}}+1}\bm{q} \displaystyle=(2\mathrm{i})^{n_{\mathrm{W}}+1}(2n_{\mathrm{W}}+1)!\mathrm{sgn}\left(\begin{array}[]{cccc}1&2&\cdots&2n_{\mathrm{W}}+2\\ i_{1}&i_{2}&\cdots&i_{2n_{\mathrm{W}}+2}\end{array}\right)\big(\hat{d}_{i_{1}}\partial_{1}\hat{d}_{i_{2}}\cdots\partial_{2n_{\mathrm{W}}+1}\hat{d}_{i_{2n_{\mathrm{W}}+2}}\big)d^{2n_{\mathrm{W}}+1}\bm{q}, in a similar way as Eq. ( S6 ). Substituting Eq. ( S6 ) into Eq. ( S6 ), we have \displaystyle(-1)^{n_{\mathrm{W}}}\mathcal{W}_{2n_{\mathrm{W}}+1} \displaystyle=\frac{n_{\mathrm{W}}!}{2\pi^{n_{\mathrm{W}}+1}}\int d^{2n_{\mathrm{W}}+1}\bm{q}\epsilon^{i_{1}\ldots i_{2n_{\mathrm{W}}+1}}f_{i_{1}\ldots i_{2n_{\mathrm{W}}+1}} \displaystyle=W_{2n_{\mathrm{W}}+1}. (S59) From the second to the third lines, we have used Eqs. ( 5 ) and ( S3 ). This relation indicates the correspondence between FM winding numbers W_{2n_{\mathrm{W}}+1} and winding numbers \mathcal{W}_{2n_{\mathrm{W}}+1} of Hermitian Hamiltonians.