## Abstract

We develop the theory of all-dielectric absorbers based on temporal coupled mode theory (TCMT), with parameters extracted from eigenfrequency simulations. An infinite square array of cylindrical resonators embedded in air is investigated, and we find that it supports two eigenmodes of opposite symmetry that are each responsible for half of the total absorption. The even and odd eigenmodes are found to be the hybrid electric (EH_{111}) and hybrid magnetic (HE_{111}) waveguide modes of a dielectric wire of circular cross section, respectively. The geometry of the cylindrical array is shown to be useful for individual tuning of the radiative loss rates of the eigenmodes, thus permitting frequency degeneracy. Further, by specifying the resonators’ loss tangent, the material loss rate can be made to equal the radiative loss rate, thus achieving a state of degenerate critical coupling and perfect absorption. Our results are supported by S-parameter simulations, and agree well with waveguide theory.

© 2017 Optical Society of America

## 1. Introduction

Electromagnetic wave absorbers based on metamaterials have attracted much interest over the past decade. The most extensively explored design is the metal-dielectric-metal (MDM) three layered structure due to its ease of fabrication, frequency scalability, wide-angle absorption, tunability, and application in diverse areas as: thermal emitters, sensors, detectors, and spatial light modulators [1–7]. Although there is continued interest in the basic physics and applications of metal-based absorbers, they posses various shortcomings, which may ultimately limit their usefulness. For example, the high operational temperatures (T>1000 °C) required for energy harvesting using thermal emitters in thermophotovoltaics (TPVs) is above the melting point of commonly used metals [8]. Further, the performance of metallic metamaterials is inextricably interwoven with both the electrical conductivity and thermal conductivity, as prescribed by the Wiedmann-Franz law [9]. Thus high performance metal-based metamaterials are constrained to also be good thermal conductors.

An alternative approach using all-dielectric materials to form metasurface absorbers has recently been proposed and demonstrated [9–11]. Experimental verification of high absorption at both 1 THz and 600 GHz was demonstrated, and a test system consisting of an uncooled terahertz imager was also shown. It has further been proposed that dielectric metasurfaces may be useful as high temperature emitters for energy harvesting applications [9]. Metasurface absorbers are fashioned from subwavelength dielectric particles [12,13], which can be achieved utilizing arrays of various geometrical shapes with a specific amount of material loss. In [10], Shadrivov et. al. show that the high absorption occurs due to the overlap of electric dipole (ED) and magnetic dipole (MD) resonances supported by a square array of dielectric cylinders. In [9] and [11], Padilla et.al. experimentally verified and confirmed that the high absorptive state is due to dipole resonances – in particular the ED and MD resonances were shown to be the hybrid electric (EH_{111}) and hybrid magnetic (HE_{111}) modes, respectively, of a dielectric cylindrical waveguide.

Here we investigate the mechanism underlying high absorption in all-dielectric absorbers using temporal coupled mode theory (TCMT) [14–17]. We find that all-dielectric metamaterial absorbers achieve degenerate coupling of the EH_{111} and HE_{111} modes [16], with each providing half of the absorption. Further, the two hybrid modes possess opposite symmetry and each achieves a state where the radiation loss rate (*γ*) is equal to the material loss rate (*δ*), i.e. critical coupling *γ* = *δ*. For simplicity, we consider an array of sub-wavelength free-standing lossy dielectric disks in air at normal incidence. The all-dielectric metasurface absorber is thus a mirror-symmetric 2-port resonator with the symmetry plane lying in the center of the disk perpendicular to the cylindrical axis. It is important to note that the all-dielectric absorber may be coupled to with one or two inputs [16]. Thus it is distinct from MDM absorbers where often a continuous ground plane only permits coupling with a single port [18, 19], and different from coherent perfect absorbers [20] where two ports must be simultaneously excited with specific inputs. In what follows, we will separate contributions of the even EH_{111} and odd HE_{111} modes to absorption, and demonstrate that they are degenerate, uncoupled to each other, with each independently achieving nearly critical coupling and thus high absorption.

## 2. Degenerate critical coupling of EH_{111} and HE_{111} modes

In Fig. 1(a) we show a schematic of one unit cell of the all-dielectric absorber disk array embedded in air, with *r* the radius, *h* the height, and *p* the periodicity of the square array. The mirror plane is shown in the *z*=0 plane in Fig. 1. One typically excites metamaterial absorbers with electromagnetic radiation incident from a single side, i.e. a 1-port excitation. However, due to mirror symmetry, the all-dielectric absorber possesses two identical input facets. Thus we may decompose the 1-port excitation into a combination of even and odd eigenexcitations from the two opposing identical ports [16], each with half of the total power as depicted in Fig. 1. The two excitations – one from each port – are equal in electric field amplitude (E_{0}/2), but possess opposite symmetry with respect to the mirror plane. The net result of the above procedure is to decompose a single excited two port resonator into two 2-port resonators, each excited with half of the original power. Further, the decomposition specifies a boundary condition for the mirror symmetry plane, i.e. a perfect magnetic conductor (PMC) for even eigenexcitation, and perfect electric conductor (PEC) for odd eigenexcitation. Therefore the even and odd eigenexcitations only couple to either the EH_{111} (even) or HE_{111} (odd) modes of the same symmetry, respectively [16, 21].

If the EH_{111} and HE_{111} modes possess resonant frequencies that are close to each other – but far from other higher order modes – their behavior can be described with TCMT [14–17]. We may thus express the total absorption as a linear sum of absorption due to even and odd eigenexcitations. If the incident field values are not too large, such that the system responds linearly, the loss from each mode remains independent and thus the total absorption can be expressed as (see Appendix A) a sum of two Lorentz terms,

_{111}) and odd (HE

_{111}) hybrid waveguide modes, respectively. In Eq. (1),

*ω*

_{0}is the center frequency of each mode,

*γ*and

*δ*are the radiation loss rate and material loss rate, respectively, for each mode. We note that the Lorentz parameters given in Eq. (1) are governed by the geometry of the resonator, filling fraction of the array, and the resonator’s complex dielectric function [18, 19]. As can be seen from Eq. (1) both terms can achieve a maximum absorption of 50% at

*ω*=

*ω*

_{0}when the radiation loss is equal to the dissipation loss, i.e. the so-called critically damped state

*γ*=

*δ*. Moreover, if we additionally have the condition

*ω*

_{0,1}=

*ω*

_{0,2}, we achieve a state of degeneracy, and thus perfect absorption at

*ω*=

*ω*

_{0}. Relaxing the condition for critical coupling, we note that it is still possible to achieve significant absorption in each mode if the radiative rate and dissipation rate are close, since Eq. (1) for each mode gives,

*ω*

_{0}and achieve the criteria specified in Eq. 2, the total absorption is at least

*A*(

*ω*

_{0}) =

*A*+

_{even}*A*= 98%.

_{odd}The structure investigated here is similar to that in [11], where the disk material consists of boron doped silicon with relative permittivity described by the Drude model [22] as ${\mathit{\u03f5}}_{r}={\mathit{\u03f5}}_{1r}\left(1+i\text{tan}\delta \right)={\mathit{\u03f5}}_{\infty}-{\omega}_{p}^{2}/\left({\omega}^{2}+i{\omega}_{s}\omega \right)$, *ϵ*_{1}* _{r}* and

*ϵ*

_{2}

*are the real and imaginary parts, respectively, of the relative permittivity*

_{r}*ϵ*, and tan

_{r}*δ*=

*ϵ*

_{2}

*/*

_{r}*ϵ*

_{1}

*is the loss tangent. Here we use*

_{r}*ϵ*

_{∞}= 11.9,

*ω*= 2

_{p}*π*× 1.27 × 10

^{12}Hz and

*ω*= 2

_{s}*π*× 0.64 × 10

^{12}Hz. In order to reduce the existence of – and potential coupling to – higher order modes (evidenced by the occurrence of Fano lineshapes [23]) – the height

*h*and radius

*r*of disks are chosen to be close to the cut-off values of the EH

_{111}mode [9, 11, 21], determined by $h={\lambda}_{0}/2{\mathit{\u03f5}}_{1r}^{1/2}$ and

*r*=

*J*

_{1,1}

*λ*

_{0}/2

*π*(

*ϵ*

_{1}

*−1)*

_{r}^{1/2}, where

*J*

_{1,1}is the first positive solution of the Bessel function of the first kind, and

*λ*

_{0}= 2

*πc/ω*

_{0}, where

*c*is the speed of light. We further require that the period

*p*of the array should be smaller than

*λ*

_{0}in order to avoid any diffractive effects. For a target frequency of

*ω*

_{0}= 2

*π*× 1.0 THz, we find

*h*=45.8µm,

*r*=58.6µm from the cut-off conditions of the EH

_{111}mode, and use a periodicity of

*p*<300µm. In our study, we investigate the absorptive properties of the all-dielectric metamaterial using the optimized parameters of

*h*=50µm,

*r*=60µm, and

*p*=210µm, unless specified otherwise.

## 3. Computational simulations

#### 3.1. Scattering parameter

We perform S-parameter simulations of a single unit cell of the cylindrical resonator, with periodic boundary conditions at the perimeter of the unit cell (±*xz* −plane and *yz* − plane), and port boundaries (with perfectly matched layers (PMLs)) on the ±*xy* −planes placed equidistant from the mirror symmetry plane – see Fig. 1. Thus we may characterize even-eigenexcitations (Fig. 1(b)) or odd-eigenexcitations (Fig. 1(c)) of the metamaterial by driving the ports in-phase or out-of-phase, respectively. Absorptivity due to even or odd eigenexcitations we denote as *A _{even}* or

*A*, respectively. The power supplied to each port – for both the even and odd excitations – is half of the power supplied in the single port simulation shown in Fig. 1(a). The S-parameter configuration described above, and depicted in Fig. 1, further allows us to investigate the sum of even and odd absorption, i.e.

_{odd}*A*

_{Σ}=

*A*+

_{even}*A*, as well as the single port absorption

_{odd}*A*. It is important to note that the 1-port excitation (Fig. 1(a)) drives both even and odd modes simultaneously, in contrast to Fig. 1(b) and Fig. 1(c), where only the even-eigenexcitations or odd-eigenexcitations are driven.

The absorption spectra for both 1-port excitation (red curves) and 2-port excitation of the even (black curve) and odd (gray curve) eigenmodes are shown in Fig. 2. The absorption spectra is calculated as $A=1-{\displaystyle {\sum}_{i,j=1}^{2}}|{S}_{ij}{|}^{2}$ from S-parameter simulations for cylindrical radii of *r*=45*µ*m, 60*µ*m and 70*µ*m, all for a constant height of *h* =50*µ*m and period *p* =210*µ*m. For the parameters studied here, a radius of *r* =60*µ*m is optimal and gives us a maximum absorption of *A* = 99.4% at *ω*_{0} = 2*π* × 1.048THz. For radii smaller than optimal we observe that both even and odd modes shift to higher frequencies, and are no longer degenerate. In contrast, for a larger radius of *r* =70*µ*m, the odd eigenexcitation is nearly unshifted from the optimal case, but the even eigenexcitation shifts to lower frequencies. For the optimal case shown in Fig. 2(b) both the even and odd modes achieve values close to *A* =50%. The open blue circles in Fig. 2 are the sum of even and odd absorption, i.e. *A*_{Σ}, and we find excellent agreement between *A* and *A*_{Σ}.

In order to gain insight into the symmetry of each eigenmode, we plot the electric and magnetic fields of the even and odd modes in Fig. 3, on the vertical cut-plane of the disk (y=0 plane), parallel to the incident electric field. As can be observed, the x-component of the electric field (black arrows) is symmetric to the mirror plane for even eigenexcitation (Fig. 3(a)), while for the odd eigenexcitation (Fig. 3(b)), it is anti-symmetric. Also plotted in Fig. 3 is the y-component of the magnetic field (*H _{y}*), shown as the colormap. We note that the fields are mainly confined in the disk and their spatial dependence is consistent with the EH

_{111}and HE

_{111}waveguide modes [21,23,24]. In Fig. 3(c) we show a plot of the summed electric and magnetic fields from Fig. 3(a) and Fig. 3(b), as well as the fields resulting from a 1-port excitation in Fig. 3(d). As can be observed, the resulting electric field is asymmetric for both for the summed and 1-port simulations. We also find asymmetry in the magnetic field plotted in Fig. 3(c) and Fig. 3(d), which is not obvious since its amplitude in the HE

_{111}mode (Fig. 3(b)) dominates over that in the EH

_{111}mode (Fig. 3(a)) [10,11,13].

#### 3.2. Eigenanalysis

The S-parameter study presented in Fig. 2 and Fig. 3 verify that the all-dielectric resonator supports two modes of opposite symmetry, but is unable to independently determine the value and significance of the radiative and material loss rates. Thus we next turn toward an eigenanalysis in order to calculate the Lorentz parameters in Eq. (1), i.e. *ω*_{0}, *γ* and *δ*. Here, only the top half of the resonator is simulated and we use a boundary condition in place of the mirror symmetry plane, i.e. a perfect magnetic conductor (PMC) for the EH_{111} mode and perfect electric conductor (PEC) for HE_{111} mode [21]. We describe the resulting complex eigenfrequency as $\tilde{\omega}={\omega}_{0}-i{\omega}_{2}$, where *ω*_{2} = *γ* + *δ*. We may also repeat the analysis with material loss removed, i.e. *δ* = 0, thereby determining the individual radiative and material contributions to resonator loss [16, 18]. Before carrying out eigenfrequency numerical modeling, we first determine the approximate resonant frequencies for the EH_{111} and HE_{111} modes from waveguide theory (Eqs. (9)–(12) in Appendix B) and find values of 1.086THz and 1.053THz, respectively [21,24,25]. The Lorentz parameters determined from the waveguide equations and eigenfrequency simulations are shown in Table 1.

#### 3.3. Comparison of TCMT with numerical results

We next compare the absorption calculated from Eq. (1) using the eigenfrequencies shown in Table 1, to that of the absorption computed from a 1-port S-parameter simulation. In Fig. 4 we plot the S-parameter *A*(*ω*) (open blue circles) and *A* good (*ω*) from Eq. (1) as the red curve. We find good agreement between the calculated eigenfrequency *A*(*ω*) and the S-parameter simulated absorption near *ω*_{0}, which gradually worsens away from the resonance frequency. The poor agreement at higher frequencies is due to the occurrence of higher order modes [15, 26, 27], not accounted for in our analysis. Also shown in Fig. 4 are the simulated even (black curve) and odd (gray curve) eigenmode absorptivities. Notably, we find our reflectivity is relatively low across the frequency range investigated (not shown – see [9]), which may be understood by noting that we achieve a critically coupled state *γ* = *δ* for each of the two modes and, importantly, all loss rates are close to each other in value. Thus the all-dielectric absorber realizes conditions similar to that of the all-pass filter [17].

## 4. Dependence of loss rates on metasurface geometry

We next detail the dependence of absorptivity *A*(*ω*) and resonant frequency *ω*_{0} on the metasurface geometrical parameters of height, radius, and periodicity. In Fig. 5 we show the resonant frequency of both the even (open black circles) and odd (open gray triangles) modes as determined from eigenfrequency simulation. It can be observed that *ω*_{0} follows the absorptivity peaks determined by S-parameter simulation, shown as the colormap which is similar to Fig. 1 in [11], for both the even and odd eigenexcitations. The solid curves shown in Fig. 5 are the resonant frequencies for the even (black) and odd (gray) modes, calculated from Eqs. 9–12. We note that the analytical solutions yield good approximate estimates of the geometrical values for design and, in particular, the HE_{111} mode (solid gray curve) agrees quite well with simulation (gray triangles) for all radii values shown, and we find relatively good agreement with the height. In contrast, the analytically calculated EH_{111} mode (solid black curve) resonant frequency deviates from the eigenfrequency (black circles) for both *r* and *h*, and we believe this is due to the PMC boundary condition assumption in Eq. (9) (see Appendix B).

In Fig. 6 we show the dependence of the radiative loss rate (solid curves) and material loss rate (dashed curves) on the geometry of the metasurface for both the even EH_{111} (red) and odd HE_{111} (blue) eigenfrequencies. The optimal geometrical parameters of *h* =50*µ*m, *r* =60*µ*m, and *p* =210*µ*m for peak absorptivity are shown as the dashed vertical black lines. The waveguide cutoff condition for EH_{111} is shown as the gray shaded area, and we only consider heights and radii outside of this region. Generally we observe that the material loss rates *δ* for both modes are roughly independent of the geometrical parameters, whereas *γ* values vary widely – especially as a function of periodicity where we use an expanded vertical scale.

We next explore the impact of the material loss tangent (tan *δ* = *ϵ*_{2}_{r}/ϵ_{1}* _{r}*) on the reflectivity (R), transmissivity (T), absorptivity, and loss rates. Figure 7(a) shows R, T, and A as a function of tan

*δ*. Here we use a constant

*ϵ*

_{1}

*= 10.83, thus tan*

_{r}*δ*∝

*ϵ*

_{2}

*. For zero loss we find that the metasurface achieves a good transmissivity (*

_{r}*T*≈ 50%), with only 1×10

^{−3}% of the power absorbed. However as loss increases,

*T*and

*R*drop rapidly while

*A*peaks at tan

*δ*= 0.06. In Fig. 7(a) we also plot

*A*(

*ω*=

*ω*

_{0,1}=

*ω*

_{0,2}) as the solid red curve from TCMT, i.e. Eq. (2), as a function of the loss ratio, defined as

*δ*/

*γ*, on a separate horizontal scale (top axis). In Fig. 7(b) we show

*γ*and

*δ*for both eigenmodes. We find that

*γ*for both modes does not depend on loss tangent, and that

*δ*is proportional to the materials dielectric loss, i.e.

*δ*∝ tan

*δ*.

## 5. Discussion

We find that the relative difference between A and A_{Σ} is within numerical precision, thus indicating each mode is orthogonal and provides independent absorption [16]. Further, when we excite only the even (or odd) eigenexcitation – as shown in Fig. 2(a) and Fig. 2(c) – we do not find any characteristic absorption features from the opposite symmetry mode. We note that indeed the sum of the fields for the degenerate critically coupled case, shown in Fig. 3(c), are identical to that shown in Fig. 3(d), and explain the overall asymmetry of the 1-port excited case. That is, when light is incident on the metasurface absorber from one side, the electric fields from the even and odd modes point in opposite directions on the incident side of the mirror plane, and in the same direction on the exit side of the mirror plane. This is the underlying reason for the asymmetric power loss density plots shown in [9] and [11].

The resonate frequencies and loss rates of both eigenmodes are determined from eigenfrequency simulations, which are then used in TCMT to calculate *A*(*ω*), and we find excellent agreement with the *A*(*ω*) calculated from S-parameter simulations. We also find good agreement between our analytical solution of the waveguide equations for the HE_{111} mode – given by Eqs. (11) and (12) – and moderate agreement for the EH_{111} mode determined by Eqs. (9) and (10). Although we have not explored the reason for the discrepancy in the even mode, we believe this is due to the form of Eqs. (9) and (10), which are only approximate.

Although the absorptivity plotted in Fig. 5(c) appears to be a weak function of periodicity, we note that our eigenfrequency analysis shows that the even and odd modes have a strong dependence on *p*, as shown in Fig. 6(c). Thus neighbor interactions are significant and the relatively weak dependence of the absorptivity on periodicity may be understood through examination of Eq. (2), where it can be seen that the conditions for strong absorptivity from each mode are relatively robust against mismatch of the radiative and material loss rates. More generally, we note that the radiative loss rates for both the EH_{111} and HE_{111} modes are relatively strong functions of all geometrical parameters of a cylindrical square array, in comparison to the material loss rates, which realize only weak dependence, as can be observed from Fig. 6. Not surprisingly, Fig. 7(b) shows that, in contrast, radiative loss rates are independent of loss tangent, whereas the material loss rate is linearly proportional, i.e. *δ* ∝ tan *δ* ∝ *ϵ*_{2}* _{r}* (for

*ϵ*

_{1}

*constant).*

_{r}As a general design rule for the construction of all-dielectric absorbers, we first use Eqs. (7) and (8) from [9] to make our cylinder just slightly larger than cutoff for the EH mode. Next we select a periodicity that is large enough to minimize neighbor interaction, but still smaller than the operational wavelength. It’s important to note that we have demonstrated here that – for a square array of cylinders – the geometrical parameters largely determine the radiative loss rate *γ*. Thus with our geometry now set, we select a loss tangent that give us a material loss rate approximately equal to our radiative loss rate, i.e. *δ* = *γ*, as shown in Fig. 7. The later requirement is straightforward when using a semiconductor as the base waveguide material, as one can typically choose a doping in order to provide a prescribed amount of loss.

Lastly we clarify that we have only considered a free-standing disk array. Use of a substrate for structural support [11] introduces asymmetry which has not been addressed. However, if the refractive index of the supporting substrate is low loss – with real values near that of free space [11] – it may be treated as a symmetric structure. Thus the theoretical treatment presented here may be used for initial analysis.

## 6. Conclusion

We find that an all-dielectric metamaterial fashioned from a square array of cylinders embedded in air can support two hybrid waveguide modes of opposite symmetry. The geometrical parameters of height, radius, and periodicity may be used to overlap these eigenmodes in frequency, thereby achieving degeneracy. Equally important, we find that the geometry specifies a particular radiative loss rate. Thus by controlling the imaginary portion of the cylinder’s dielectric constant, the material loss rate may be made equal to radiative losses, thus achieving critical damping and perfect absorption. Temporal coupled mode theory and eigenfrequency simulations accurately describe the frequency dependent absorptivity of each eigenmode, as well as their sum – equal to the 1-port excitation. Eigenmode simulations elucidate the radiative and material loss rates, and their dependence on geometry and loss tangent. S-parameter simulations detail the relation of R, T, and A, on the loss tangent, and match well the form of (*A* tan *δ*) predicted by TCMT. Other geometrical shapes and alternative systems may be studied with the methods we show here, including bound states in the continuum and coherent perfect absorbers.

## Appendix A: Derivation of absorption from TCMT

The behavior of the EH_{111} and HE_{111} modes of the mirror-symmetric cylindrical resonator can be described with TCMT [14–17] as,

*a*is a vector with an assumed time dependence of exp (−

*iωt*) which describes the mode amplitude, with the stored energy for each mode given by |

*a*|

_{j}^{2}. The center frequencies of each modes are given by the real diagonal matrix

*Ω*

_{0};

*Γ*and Δ are two real diagonal matrices describing the radiation loss rate and dissipation loss rate of the modes, respectively. Coupling between the two modes is given by an anti-diagonal matrix

*K*; s

*is a vector that represents inputs from each port with |*

_{in}*s*,

_{in}*|*

_{j}^{2}equal to the input power, with a similar term for the outputs s

*; the matrix*

_{out}*D*describes the coupling between modes and inputs, where

*D*

^{∗}

*D*= 2Γ; the matrix

*C*represents the background scattering between ports.

We consider the case with input from only a single port and thus write the input *s _{in}* = [

*s*

_{0}, 0]

*as a decomposition of even and odd eigenexcitations, as*

^{T}*s*

_{in}_{,}

*= [*

_{even}*s*

_{0}/2,

*s*

_{0}/2]

*and*

^{T}*s*= [

_{in,odd}*s*

_{0}/2, −

*s*

_{0}/2]

*. Due to the different symmetry of even and odd modes [15,28], they are orthogonal and uncoupled, i.e.*

^{T}*K*= 0. Thus for an even eigenexcitation of our mirror symmetric resonator, the radiation rates at the two ports is given by

*γ*

_{1,1}=

*γ*

_{2,1}=

*γ*

_{1}/2 due to symmetry, and ${D}_{even}=\left[\sqrt{{\gamma}_{1}},0;\sqrt{{\gamma}_{1}}0\right]$, the complex mode amplitude for the EH

_{111}mode is

Similar process is applicable to the absorption resulting from odd eigenexicitation, and total absorption due to independent absorption of the two modes is thus

## Appendix B: Analytical resonant frequencies of EH_{111} and HE_{111} modes

The resonant frequencies of EH_{111} and HE_{111} modes can be analytically estimated from those of single lossless cylinder dielectric resonator antennas in air with some approximation of the boundary conditions from waveguide theory [21,24,25].

For EH_{111} mode, perfect magnetic wall boundary is set for the side-wall of the cylinder while non-perfect magnetic walls for the top and bottom flat walls, which yield

*k*is the radial wave vector component in the cylinder,

_{r}*k*is the z component of wave vector in the cylinder, while

_{z}*k*

_{z}_{0}describes the z component of wave vector in air, they satisfy ${k}_{z}^{2}={k}_{0}^{2}{\mathit{\u03f5}}_{1r}-{k}_{r}^{2}$ and ${k}_{z0}^{2}={k}_{r}^{2}-{k}_{0}^{2}$, where

*k*

_{0}=

*ω*

_{0}/

*c*is the wavenumber in air and

*ϵ*

_{1}

*is the real part of permittivity of the cylinder. In Eq. (9) the first non-trivial solution is used.*

_{r}Meanwhile, for the HE_{111} mode, non-perfect magnetic wall boundary is set for the side-wall of the cylinder while perfect magnetic wall boundary for the top and bottom flat walls, which yield,

*u*=

*k*,

_{r}r*v*=

*k*

_{r}_{0}

*,*

_{r}*k*

_{r}_{0}is the radial wave vector in air,

*J*

_{1}(

*u*) is the first order Bessel function of the first kind, and

*K*

_{1}(

*u*) is the first order modified Hankel function.

*k*,

_{r}*k*

_{r}_{0}and

*k*satisfy ${k}_{r}^{2}={k}_{0}^{2}{\mathit{\u03f5}}_{1r}-{k}_{z}^{2}$ and ${k}_{r0}^{2}={k}_{z}^{2}-{k}_{0}^{2}$.

_{z}By numerically solving Eqs. (9)–(12) with a graphical method, we obtain an analytical estimate of the resonant frequencies of EH_{111} and HE_{111} modes. Solutions to Eqs. (9)–(12) are plotted as the solid lines in Fig. 5(a) and Fig. 5(b). Because the approximations for HE_{111} mode are closer to the physical situation than those of EH_{111} mode, they yield analytical resonant frequencies closer to the accurate ones.

## Funding

Department of Energy (DOE) (DE-SC0014372); National Key Foundation for Exploring Scientific Instrument of China (2012YQ0901670602); China Scholarship Council (CSC) (201606210317).

## Acknowledgments

WJP and XL acknowledge support from the Department of Energy (DOE) (DE-SC0014372). LS and XM acknowledge support from the National Key Foundation for Exploring Scientific Instrument of China (2012YQ0901670602). XM is partly supported by China Scholarship Council (CSC) (201606210317). We acknowledge Kebin Fan, Ilya Shadrivov, Andrew Cardin, and David Powell for useful discussions.

## References and links

**1. **C. M. Watts, X. Liu, and W. J. Padilla, “Metamaterial electromagnetic wave absorbers,” Adv. Mater. **24**(23), OP98–OP120 (2012). [PubMed]

**2. **Y. P. Lee, J. Y. Rhee, Y. J. Yoo, and K. W. Kim, *Metamaterials for Perfect Absorption* (Springer, 2016). [CrossRef]

**3. **X. Liu and W. J. Padilla, “Reconfigurable room temperature metamaterial infrared emitter,” Optica **4**(4), 430–433 (2017). [CrossRef]

**4. **K. Fan, J. Suen, X. Wu, and W. J. Padilla, “Graphene metamaterial modulator for free-space thermal radiation,” Opt. Express **24** (22), 25189–25201 (2016). [CrossRef] [PubMed]

**5. **N. Liu, M. Mesch, T. Weiss, M. Hentschel, and H. Giessen, “Infrared perfect absorber and its application as plasmonic sensor,” Nano Lett. **10** (7), 2342–2348 (2010). [CrossRef] [PubMed]

**6. **T. Maier and H. Brückl, “Wavelength-tunable microbolometers with metamaterial absorbers,” Opt. Lett. **34** (19), 3012–3014 (2009). [CrossRef] [PubMed]

**7. **C. M. Watts, D. Shrekenhamer, J. Montoya, G. Lipworth, J. Hunt, T. Sleasman, S. Krishna, D. R. Smith, and W. J. Padilla, “Terahertz compressive imaging with metamaterial spatial light modulators,” Nat. Photonics **8**(8), 605–609 (2014). [CrossRef]

**8. **C. Shemelya, D. DeMeo, N. P. Latham, X. Wu, C. Bingham, W. Padilla, and T. E. Vandervelde, “Stable high temperature metamaterial emitters for thermophotovoltaic applications,” Appl. Phys. Lett. **104**, 201113 (2014). [CrossRef]

**9. **K. Fan, J. Y. Suen, X. Liu, and W. J. Padilla, “All-dielectric metasurface absorbers for uncooled terahertz imaging,” Optica **4**(6), 601–604 (2017). [CrossRef]

**10. **M. A. Cole, D. A. Powell, and I. V. Shadrivov, “Strong terahertz absorption in all-dielectric Huygens’ metasurfaces,” Nanotechnology **27**, 424003 (2016). [CrossRef]

**11. **X. Liu, K. Fan, I. V. Shadrivov, and W. J. Padilla, “Experimental realization of a terahertz all-dielectric metasurface absorber,” Opt. Express **25**(1), 191–201 (2017). [CrossRef] [PubMed]

**12. **I. Staude and J. Schilling, “Metamaterial-inspired silicon nanophotonics,” Nat. Photonics **11**(5), 274–284 (2017). [CrossRef]

**13. **M. Decker, I. Staude, M. Falkner, J. Dominguez, D. N. Neshev, I. Brener, T. Pertsch, and Y. S. Kivshar, “High-efficiency dielectric Huygens’ surfaces,” Adv. Opt. Mater. **3**(6), 813–820 (2015). [CrossRef]

**14. **H. A. Haus, *Waves and Fields in Optoelectronics* (Prentice-Hall, 1984), Chap. 7.

**15. **W. Suh, Z. Wang, and S. Fan, “Temporal coupled-mode theory and the presence of non-orthogonal modes in lossless multimode cavities,” IEEE J. Quantum Electron. **40**(10), 1511–1518 (2004). [CrossRef]

**16. **J. R. Piper, V. Liu, and S. Fan, “Total absorption by degenerate critical coupling,” Appl. Phys. Lett. **104**, 251110 (2014). [CrossRef]

**17. **W. Suh and S. Fan, “All-pass transmission or flattop reflection filters using a single photonic crystal slab,” Appl. Phys. Lett. **84** (24), 4905–4907 (2004). [CrossRef]

**18. **C. Wu, B. Neuner III, G. Shvets, J. John, A. Milder, B. Zollars, and S. Savoy, “Large-area wide-angle spectrally selective plasmonic absorber,” Phys. Rev. B **84**, 075102 (2011). [CrossRef]

**19. **C. Qu, S. Ma, J. Hao, M. Qiu, X. Li, S. Xiao, Z. Miao, N. Dai, Q. He, S. Sun, and L. Zhou, “Tailor the functionalities of metasurfaces based on a complete phase diagram,” Phys. Rev. Lett. **115**, 235503 (2015). [CrossRef] [PubMed]

**20. **W. Wan, Y. Chong, L. Ge, H. Noh, A. D. Stone, and H. Cao, “Time-reversed lasing and interferometric control of absorption,” Science **331**(6019), 889–892 (2011). [CrossRef] [PubMed]

**21. **R. K. Mongia and P. Bhartia, “Dielectric resonator antennas – a review and general design relations for resonant frequency and bandwidth,” Int. J. RF Microw. Comput-Aid. Eng. **4**(3), 230–247 (1994).

**22. **S. Nashima, O. Morikawa, K. Takata, and M. Hangyo, “Measurement of optical properties of highly doped silicon by terahertz time domain reflection spectroscopy,” Appl. Phys. Lett. **79**(24), 3923–3925 (2001). [CrossRef]

**23. **D. Cai, Y. Huang, W. Wang, W. Ji, J. Chen, Z. Chen, and S. Liu, “Fano resonances generated in a single dielectric homogeneous nanoparticle with high structural symmetry,” J. Phys. Chem. C **119**(8), 4252–4260 (2015). [CrossRef]

**24. **E. Snitzer, “Cylindrical dielectric waveguide modes,” J. Opt. Soc. Am. **51**(5), 491–498 (1961). [CrossRef]

**25. **P. Guillon and Y. Garault, “Accurate resonant frequencies of dielectric resonators,” IEEE Trans. Microw. Theory Tech. **25**(11), 916–922 (1977). [CrossRef]

**26. **P. T. Bowen and D. R. Smith, “Coupled-mode theory for film-coupled plasmonic nanocubes,” Phys. Rev. B **90**, 195402 (2014). [CrossRef]

**27. **B. Hopkins, A. N. Poddubny, A. E. Miroshnichenko, and Y. S. Kivshar, “Revisiting the physics of Fano resonances for nanoparticle oligomers,” Phys. Rev. A **88**, 053819 (2013). [CrossRef]

**28. **D. A. Powell, “Interference between the modes of an all-Dielectric meta-atom,” Phys. Rev. Appl. **7**, 034006 (2017). [CrossRef]