Professional Documents
Culture Documents
IEEE Transactions On Microwave Theory and Techniques Vol 65 No 07 2017 07 7th Edition Microwave Theory and Techniques Society
IEEE Transactions On Microwave Theory and Techniques Vol 65 No 07 2017 07 7th Edition Microwave Theory and Techniques Society
https://ebookmeta.com/product/ieee-transactions-on-microwave-
theory-and-techniques-vol-65-no-08-2017-08-8th-edition-
fantomasping/
https://ebookmeta.com/product/ieee-transactions-on-microwave-
theory-and-techniques-v065-i12a-2017-12-fantomasping/
https://ebookmeta.com/product/ieee-
mtt-v065-i11a-2017-11-microwave-theory-and-techniques-ieee-
microwave-theory-and-tecnique-society/
https://ebookmeta.com/product/microwave-theory-and-
techniques-v065-i10-2017-10-10th-edition-ieee/
IEEE MTT V065 I11B 2017 11 Microwave theory and
techniques Fantomasping
https://ebookmeta.com/product/ieee-
mtt-v065-i11b-2017-11-microwave-theory-and-techniques-
fantomasping/
https://ebookmeta.com/product/rf-and-microwave-circuit-design-
theory-and-applications-microwave-and-wireless-technologies-
series-1st-edition-charles-e-free/
https://ebookmeta.com/product/apprehend-me-no-flowers-madison-
night-cozy-mystery-07-diane-vallere-et-el/
https://ebookmeta.com/product/automata-theory-and-formal-
languages-fundamental-notions-theorems-and-techniques-first-
edition-alberto-pettorossi/
JULY 2017 VOLUME 65 NUMBER 7 IETMAB (ISSN 0018-9480)
THIS ISSUE INCLUDES THE JOURNAL WITHIN A JOURNAL ON MICROWAVE SYSTEMS AND
APPLICATIONS
A. A BUNJAILEH M. B OZZI K. G HORBANI R. H ENDERSON P. K HANNA S. KOUL M. M ADIHIAN G. P ONCHAK J. E. R AYAS -S ANCHEZ M. S ALAZAR PALMA D. S CHREURS D. W ILLIAMS
S. BARBIN T. B RAZIL R. G UPTA A. JACOB D. K ISSINGER G. LYONS D. PASQUET S. R AMAN S. R EISING A. S ANADA M. S TEER
IEEE Officers
K AREN BARTLESON, President S. K. R AMESH, Vice President, Educational Activities
JAMES A. J EFFERIES, President-Elect S AMIR M. E L -G HAZALY, Vice President, Publication Services and Products
W ILLIAM P. WALSH, Secretary M ARY E LLEN R ANDALL, Vice President, Member and Geographic Activities
J OHN W. WALZ, Treasurer F ORREST D. W RIGHT, President, Standards Association
BARRY L. S HOOP, Past President M ARINA RUGGIERI, Vice President, Technical Activities
K AREN S. P EDERSEN, President, IEEE-USA
J ENNIFER T. B ERNHARD, Director, Division IV—Electromagnetics and Radiation
IEEE Executive Staff
D R . E. JAMES P RENDERGAST, Executive Director & Chief Operating Officer
T HOMAS S IEGERT, Business Administration C HERIF A MIRAT, Information Technology
J ULIE E VE C OZIN, Corporate Governance K AREN H AWKINS, Marketing
D ONNA H OURICAN, Corporate Strategy C ECELIA JANKOWSKI, Member and Geographic Activities
JAMIE M OESCH, Educational Activities M ICHAEL F ORSTER, Publications
E ILEEN M. L ACH, General Counsel & Chief Compliance Officer KONSTANTINOS K ARACHALIOS, Standards Association
VACANT, Human Resources M ARY WARD -C ALLAN, Technical Activities
C HRIS B RANTLEY, IEEE-USA
IEEE Periodicals
Transactions/Journals Department
Senior Director, Publishing Operations: F RAN Z APPULLA
Director, Editorial Services: DAWN M ELLEY Director, Production Services: P ETER M. T UOHY
Associate Director, Editorial Services: W ILLIAM A. C OLACCHIO Associate Director, Information Conversion and Editorial Support: K EVIN L ISANKIE
Managing Editor: C HRISTOPHER P ERRY Journals Coordinator: C HRISTINA M. R EZES
IEEE T RANSACTIONS ON M ICROWAVE T HEORY AND T ECHNIQUES (ISSN 0018-9480) is published monthly by the Institute of Electrical and Electronics Engineers, Inc. Responsibility for the contents rests upon
the authors and not upon the IEEE, the Society/Council, or its members. IEEE Corporate Office: 3 Park Avenue, 17th Floor, New York, NY 10016-5997. IEEE Operations Center: 445 Hoes Lane, Piscataway,
NJ 08854-4141. NJ Telephone: +1 732 981 0060. Price/Publication Information: Individual copies: IEEE Members $20.00 (first copy only), nonmember $217.00 per copy. (Note: Postage and handling charge
not included.) Member and nonmember subscription prices available upon request. Copyright and Reprint Permissions: Abstracting is permitted with credit to the source. Libraries are permitted to photocopy for
private use of patrons, provided the per-copy fee of $31.00 is paid through the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. For all other copying, reprint, or republication permission,
write to Copyrights and Permissions Department, IEEE Publications Administration, 445 Hoes Lane, Piscataway, NJ 08854-4141. Copyright c 2017 by The Institute of Electrical and Electronics Engineers, Inc.
All rights reserved. Periodicals Postage Paid at New York, NY and at additional mailing offices. Postmaster: Send address changes to IEEE T RANSACTIONS ON M ICROWAVE T HEORY AND T ECHNIQUES, IEEE,
445 Hoes Lane, Piscataway, NJ 08854-4141. GST Registration No. 125634188. CPC Sales Agreement #40013087. Return undeliverable Canada addresses to: Pitney Bowes IMEX, P.O. Box 4332, Stanton Rd.,
Toronto, ON M5W 3J4, Canada. IEEE prohibits discrimination, harassment and bullying. For more information visit http://www.ieee.org/nondiscrimination. Printed in U.S.A.
Abstract— A high-order convergent and robust numerical materials [9] at microwave frequencies, but otherwise WGMs
solver is constructed and used to find complex eigenwavenumbers have been less exploited in microwave technology than in
and electromagnetic eigenfields of dielectric objects with axial optics.
symmetry. The solver is based on Fourier–Nyström discretization
of combined integral equations for the transmission problem and Common numerical methods for the determination of
can be applied to demanding resonance problems at microwave, electromagnetic resonances in dielectric objects include the
terahertz, and optical wavelengths. High achievable accuracy, finite-element method (FEM) [10], the boundary integral
even at very high wavenumbers, makes the solver ideal for equation (BIE) methods [11]–[13], the discrete dipole
benchmarking and for assessing the performance of general- approximation method [14], and the null-field method [15].
purpose commercial software.
The method used in this paper belongs to the category of BIE
Index Terms— Axial symmetry, body of revolution, boundary methods, which comprise a variety of formulations and tech-
integral equation (BIE), dielectric resonator, electric field inte- niques. In [16], the integral equations derived in [17, Sec. 23]
gral equation (EFIE), magnetic field integral equation (MFIE),
Nyström method, whispering-gallery mode (WGM). were applied to scattering from dielectric objects. We use
a related set of integral equations and a modification of the
Fourier–Nyström scheme of [13], which in turn draws on the
I. I NTRODUCTION
progress in [12] and [18]–[21].
III. I NTEGRAL R EPRESENTATIONS AND E QUATIONS The BIEs coming from r ∈ V2 are
The modal counterpart of (24) and (25) becomes The modal representations of the fields in V2 are
⎡ ⎤
I + 2K νn −2iK 25n 2K 26n −2ik S5n 2k S6n 0
⎢ −2iK 12n I − K 1n −iK 2n 2k S3n −2ik S4n 0 ⎥
⎢ ⎥
⎢ 2K 24n −iK 3n I − K 4n 2ik S1n −2k S2n 0 ⎥
Q1n =⎢
⎢
⎥ (43)
⎢ 0 −2k S3n 2ik S4n I − K 1n −iK 2n 2iK 12n ⎥
⎥
⎣ 0 −2ik S1n 2k S2n −iK 3n I − K 4n −2K 24n ⎦
0 −2ik S5n 2k S6n 2iK 25n −2K 26n I + 2K νn
⎡ ⎤
I − 2 K̃ νn 2im K̃ 25n −2m K̃ 26n 2im k S̃5n −2m 2 k S̃6n
2 2 2 0
⎢ 2im −2 K̃ I + K̃ 1n −2k S̃3n ⎥
⎢ 12n i K̃ 2n 2ik S̃4n 0 ⎥
⎢ −2 ⎥
⎢−2m K̃ 24n i K̃ 3n I + K̃ 4n −2ik S̃1n 2k S̃2n 0 ⎥
Q2n =⎢ ⎥ (44)
⎢ 0 2m k S̃3n −2im k S̃4n
2 2 I + K̃ 1n i K̃ 2n −2i K̃ 12n ⎥
⎢ ⎥
⎣ 0 2im 2 k S̃1n −2m 2 k S̃2n i K̃ 3n I + K̃ 4n 2 K̃ 24n ⎦
0 2im 2 k S̃5n −2m 2 k S̃6n −2i K̃ 25n 2 K̃ 26n I − 2 K̃ νn
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
The normalized far-field pattern of a mode with azimuthal of Qn− 1 (χ) as χ → 1+ ; a strategy for when to use forward
2
index n is the φ-dependent function or backward recursion for the evaluation of Qn− 1 (χ); tem-
2
|F n (φ)| porary mesh refinement (upsampling) coupled with temporary
(72) increase of the quadrature order on γ .
max |F n (φ)|
0≤φ≤π The discretization of the modal representation of E and H
where F n (φ) is the Fourier coefficient of F in (11), and by for r ∈ / γ in Section IV-C is done in analogy with the
that discretization of (53).
Fig. 2. Planar field plots of the magnetic field for the fundamental n = 90 mode of a unit sphere with refractive index m = 1.5. The eigenwavenumber
is k = 65.09451518155630 − 1.3 · 10−13 i and 832 discretization points are used on γ . (a), (c), and (e) |Hρ90 (r)|, |Hθ 90 (r)|, and |Hz90 (r)|, respectively.
(b), (d), and (f) log10 of the estimated pointwise absolute error.
[E ρn (r ), E θn (r ), E zn (r )]. In each example, all six coeffi- containing 50% more quadrature panels on γ . The error
cients are evaluated and scaled with a common factor so plots use a logarithmic scale.
that the largest pointwise value of at least one coefficient Our examples cover two modal cases, both with high k:
is unity. The coefficients are evaluated at 5 · 105 points r fundamental modes with large n and a general resonance
on a Cartesian grid in a rectangle of height 2.6 and with a small n. In addition to finding eigenwavenumbers and
width 1.3 and with its left side coinciding with the showing field plots, we also do a convergence study, compute
z-axis in the half-plane depicted in Fig. 1(c). For ease of Q-values, and present far-field patterns. The convergence study
interpretation, we also show mirror images so that a field comprises a comparison between our formulation (53) and a
plot includes 106 points in a square of side length 2.6 homogeneous modal version of the Müller formulation (31).
in the x z plane.
4) The estimated errors in the field plots are taken as the
absolute value of the pointwise difference to a reference A. Fundamental Mode for Large n
solution. In the absence of semianalytic solutions, the 1) Unit Sphere: Our first example is the fundamental
reference solution is obtained with an overresolved mesh n = 90 mode of the unit sphere with m = 1.5 and is
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 3. Same as in Fig. 2, but for the object in Fig. 1 and with m = 1.5 + 5.5 · 10−12 i. The eigenwavenumber is k = 54.72590089140112 − 1.9803 · 10−10 i
and 864 discretization points are used on γ .
intended as a verification of the solver. The reference solu- 2) Lossless Versus Lossy Object Materials: We now look at
tion is evaluated from a semianalytic solution given by Mie the fundamental n = 90 mode of the object in Fig. 1 and
theory [30]. The eigenwavenumber k = 65.09451518155629− compare converged eigenwavenumbers and eigenfields for two
1.3 · 10−13 i, found by the solver, corresponds to a sphere different object materials. The first material is lossless with
diameter of 20.7 vacuum wavelengths and agrees with the m = 1.5. The eigenwavenumber is k = 54.72590089140112−
value k = 65.09451518155630 − 1.3 · 10−13 i, obtained from 1.5 · 10−13 i, corresponding to a generalized object diameter
the semianalytic solution, to almost machine precision. of about 22.8 vacuum wavelengths. The Q-factor (62) is
Fig. 2 shows field plots of [|Hρ90(r )|, |Hθ90(r )|, |Hz90(r )|] Q = 1.8 · 1014 . The second material is lossy with m =
along with estimated absolute pointwise errors, which peak at 1.5 + 5.5 · 10−12 i and has k = 54.72590089140112 − 1.9803 ·
around 100mach . In passing, we mention that our numerical 10−10 i, which corresponds to a skin depth δ ≈ 1010 . The
tests revealed the following relations for the fundamental condition (70) is fulfilled and by that Qrad is independent of
modes of dielectric spheres: m{m} and (71) holds. A comparison of the eigenwavenum-
z bers reveals that e{k} is virtually unaffected by the losses.
E θn (r ) = iE ρn (r ) = i E zn (r ) (79)
ρ Since Qrad Qdiss, it follows from (62), (65), and (71) that
which we then also derived from the semianalytic solution. m{k}/ e{k} ≈ −m{m}/ e{m}. Now Q = 1.382 ·1011 and,
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 4. Same as in Fig. 3, but for n = 450 and with m = 1.5. The eigenwavenumber is k = 258.059066513439 and 352 adaptively spaced discretization
points are used on γ .
since the lossless material has Q = 1.8 · 1014 , the dissipative 3) High Wavenumber WGM: Fig. 4 shows planar plots
Q-factor is Qdiss = 1.383 · 1011 according to (65). This and error estimates of the magnetic field for the fundamental
value agrees well with the approximate expression Qdiss ≈ n = 450 mode of the object in Fig. 1 with m = 1.5. The
1.364 · 1011 from (71). eigenwavenumber of this WGM is k = 258.059066513439,
The losses of the second material are the same as that corresponding to a generalized object diameter of about
of silica at the vacuum wavelength of 1550 nm, which is 107.6 vacuum wavelengths. This is in the regime where
the smallest known loss of any solid material at optical asymptotic methods for WGMs are applicable [31]. The imag-
wavelengths. It indicates that the physical limit for the Q- inary part of k is not identically zero, but it is too small to be
factor is approximately 1011 . To the eye, the field plots and the resolved in double precision arithmetic. The images in Fig. 4
corresponding error plots with the lossless material and with resemble those of the fundamental n = 90 mode in Fig. 3, but
the lossy material are indistinguishable. The images shown in with the fields confined to a smaller region and one digit of
Fig. 3 are thus valid for both object materials. We also tested precision lost. This case has been compared with an evaluation
our scheme for the large loss case with m = 1.5 + 5 · 10−3 i, in COMSOL Multiphysics 5.2, which is an FEM simulation
giving k = 54.72532533461791 − 0.17989547170087i, package. Since the field is confined to a small region, it was
without any problem. possible to reduce the computational domain in COMSOL to
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 5. Planar field plots of the electric field for an n = 1 mode of the object in Fig. 1. The eigenwavenumber is k = 110.041232211051 − 0.404177078290i,
the refractive index is m = 1.5, and 1984 discretization points are used on γ . (a), (c), and (e) |E ρ1 (r)|, |E θ 1 (r)|, and |E z1 (r)|, respectively.
(b), (d), and (f) log10 of the estimated pointwise absolute error.
a square 0.25×0.25. By that, the eigenwavenumber and the it can also be understood from a phenomenological description
eigenfield could be evaluated using default meshes. The con- of WGMs in terms of internal reflections. The eigenfields with
vergence of the wavenumber evaluation was of order 1.41. The high k and small n vary rapidly both outside and, in particular,
default mesh referred to as extremely fine used 2557 degrees inside A and the problem is harder to resolve.
of freedom and gave the eigenwavenumber 258.1164 with a Fig. 5 shows an example for the object in Fig. 1 with
relative error of 2.2·10−4 . All of our attempts to analyze other n = 1 and m = 1.5. The converged eigenwavenumber
resonances than WGMs with FEM methods failed. This is in k = 110.041232211051 − 0.404177078290i corresponds to
contrast to our experience from perfectly conducting cavities a generalized object diameter of about 45.9 vacuum wave-
[12], [13], where FEM is an option also for other resonances. lengths. The large value of m{k} makes the exponential
growth of the eigenfields visible already in the object’s imme-
B. High k and Small n diate vicinity.
When m{m} = 0, resonances with high k and small n Fig. 6 confirms that our solver exhibits 16th-order conver-
have much smaller Q-factors than fundamental modes with gence and is stable under uniform overresolution. The average
similar e{k}. We have clearly seen this in numerical tests and pointwise accuracy in the field plots saturates at 12–13 digits,
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 6. Convergence of the electric field plots shown in Fig. 5 and of the Fig. 8. Far-field patterns: curve 1 is for the WGM of the unit sphere of
corresponding magnetic field. The average pointwise accuracy has converged Section VII-A1; curves 2 and 3, which are almost identical, are for the WGMs
to between 12 and 13 digits at 1984 discretization points on γ , corresponding with lossless and lossy object materials of Section VII-A2; and curve 4 is for
to about 26 points per vacuum wavelength along γ . the n = 1 mode of Section VII-B.
C. Far-Field Patterns
Far-field patterns of resonant modes are defined by
(72) and (73). A necessary condition for their meaningful
evaluation when m{m} = 0 is that m{k} is known with
a relative accuracy better than 1%. This means that if Qrad ,
which for m{m} = 0 is equal to Q, is of the order of 10 p , then
k needs to be resolved with at least p+2 digits. This condition,
coupling the magnitude of Qrad to the precision required in k,
seems to hold also when m{m} = 0. The requirement of
p + 2 accurate digits in k is met in all our examples, except
for that of the high wavenumber WGM in Section VII-A3.
Fig. 8 shows that the far-field patterns of the WGMs are
smooth and resemble each other. Their radiated fields peak at
the equator, φ = π/2. The variation in the pattern of the n = 1
mode of Section VII-B is rapid, as expected.
Fig. 7. Same as in Fig. 6, but based on the Müller combination in (31). VIII. C ONCLUSION
The average pointwise accuracy has converged to almost 12 digits at
2640 discretization points on γ , corresponding to about 36 points per vacuum
Our solver, for the determination of resonant modes of
wavelength along γ . axially symmetric dielectric objects, uses integral equations,
related to the Müller formulation, and ChIEs. This, in com-
which compares favorably with the most accurate results we bination with a high-order convergent discretization, allows
have found in the literature for general transmission problems for exceptionally accurate results, and excludes the possibility
involving axially symmetric objects of nontrivial shapes [32]. of finding spurious solutions. Moreover, the solver extends
We have done the same convergence study for our the admissible size of objects for which high accuracy can
Fourier–Nyström scheme applied to (31). The result is given be obtained, based on the full vectorial Maxwell equations,
in Fig. 7. The convergence order is 15 for (31) compared into the regime where asymptotic methods for WGMs are
to 16 for our combination (53). The Müller combination (31) applicable. We stress the following capabilities of our solver
needs 36 discretization points per vacuum wavelength along γ up to such object sizes.
for saturated convergence, compared to 26 for (53). The 1) The evaluation of the entire spectrum and all eigen-
relative error of the largest absolute value of the evaluated field fields with e{k} in a given interval. This includes the
components in Fig. 5 is 1.3 · 10−12 for (31) and 3.7 · 10−13 computationally difficult resonances with small n and
for (53). The construction of the system matrix took 680 s for high k.
(31) and 350 s for (53). It took 45 s to find the solution to (31), 2) The evaluation of eigenfields at any point in space. This
compared to 70 s for (53). It took an average of 0.072 s to includes slowly evanescent and radiated fields.
evaluate the six field components at a point r for (31) and 3) The evaluation of far-field patterns of WGMs with
0.042 s for (53), with (49)–(52). radiative Q-factors up to 1013 .
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
4) Geometric flexibility. While high-order surface infor- [8] T. J. Kippenberg, R. Holzwarth, and S. A. Diddams, “Microresonator-
mation is a prerequisite, nonsmooth boundaries can be based optical frequency combs,” Science, vol. 332, no. 6029,
pp. 555–559, Jun. 2011.
treated. [9] R. Ratheesh et al., “Microwave characterisation of BaCe2 Ti5 O15 and
These capabilities open up for new studies related to the Ba5 Nb4 O15 ceramic dielectric resonators using whispering gallery
coupling of electromagnetic waves into WGMs, to finding mode method,” Mater. Lett., vol. 45, no. 5, pp. 279–285, Sep. 2000.
[10] M. Oxborrow, “Traceable 2-D finite-element simulation of the
new object shapes for WGMs and resonances in objects with whispering-gallery modes of axisymmetric electromagnetic resonators,”
nonlinear as well as active materials. They also make the solver IEEE Trans. Microw. Theory Techn., vol. 55, no. 6, pp. 1209–1218,
ideal for benchmarking. Jun. 2007.
[11] V. S. Bulygin, Y. V. Gandel, A. Vukovic, T. M. Benson, P. Sewell, and
A. I. Nosich, “Nystrom method for the Muller boundary integral
equations on a dielectric body of revolution: Axially symmetric
A PPENDIX problem,” IET Microw., Antennas Propag., vol. 9, no. 11, pp. 1186–
O PERATORS K 24n , K 25n , AND K 26n 1192, Aug. 2015.
[12] J. Helsing and A. Karlsson, “Determination of normalized magnetic
The modal operators K in , i = 24, 25, 26 are most easily eigenfields in microwave cavities,” IEEE Trans. Microw. Theory Techn.,
defined in terms of (41) and (42) and the kernels vol. 63, no. 5, pp. 1457–1467, May 2015.
[13] J. Helsing and A. Karlsson, “Determination of normalized electric
K i (r, r ) = Di (r, r )(1 − ik|r − r |)eik|r−r | , i = 24, 25, 26 eigenfields in microwave cavities with sharp edges,” J. Comput. Phys.,
vol. 304, pp. 465–486, Jan. 2016.
(A.1) [14] V. Amendola, O. M. Bakr, and F. Stellacci, “A study of the surface
plasmon resonance of silver nanoparticles by the discrete dipole
with static factors approximation method: Effect of shape, size, structure, and assembly,”
Plasmonics, vol. 5, no. 1, pp. 85–97, Mar. 2010.
τ · (r − r ) [15] W. Zheng and S. Ström, “The null-field approach to electromagnetic
D24 (r, r ) = (A.2)
4π|r − r |3 resonance of composite objects,” Comput. Phys. Commun., vol. 68,
nos. 1–3, pp. 157–174, Nov. 1991.
(νρ ν · r + νz τ · r ) sin(θ − θ ) [16] P. Ylä-Oijala and M. Taskinen, “Well-conditioned Müller formulation
D25 (r, r ) = i (A.3)
4π|r − r |3 for electromagnetic scattering by dielectric objects,” IEEE Trans.
Antennas Propag., vol. 53, no. 10, pp. 3316–3323, Oct. 2005.
νz ρ − (τ · r + νρ z ) cos(θ − θ ) [17] C. Müller, Foundations of the Mathematical Theory of Electromagnetic
D26 (r, r ) = (A.4)
4π|r − r |3 Waves. Berlin, Germany: Springer-Verlag, 1969.
[18] H. S. Cohl and J. E. Tohline, “A compact cylindrical Green’s function
and, with notation as in [13, Appendix A], corresponding expansion for the solution of potential problems,” Astrophys. J.,
vol. 527, no. 1, pp. 86–101, Dec. 1999.
Fourier coefficients [19] J. Helsing and A. Karlsson, “An explicit kernel-split panel-based
νz Nyström scheme for integral equations on axially symmetric surfaces,”
D24n (r, r ) = −η d(τ )Rn (χ) − Pn (χ) (A.5) J. Comput. Phys., vol. 272, pp. 686–703, Sep. 2014.
ρ [20] J. Helsing and A. Holst, “Variants of an explicit kernel-split panel-
(νρ ν · r + νz τ · r ) based Nyström discretization scheme for Helmholtz boundary value
D25n (r, r ) = η nQn− 1 (χ) (A.6) problems,” Adv. Comput. Math., vol. 41, no. 3, pp. 691–708, Jun. 2015.
ρρ 2 [21] P. Young, S. Hao, and P. G. Martinsson, “A high-order Nyström
(τ · r + νρ z ) discretization scheme for boundary integral equations defined on
D26n (r, r ) = η d(τ )Rn (χ) + Pn (χ) . rotationally symmetric surfaces,” J. Comput. Phys., vol. 231, no. 11,
ρρ pp. 4142–4159, Jun. 2012.
(A.7) [22] M. Taskinen and P. Yla-Oijala, “Current and charge integral equation
formulation,” IEEE Trans. Antennas Propag., vol. 54, no. 1, pp. 58–67,
The reason for including (A.5)–(A.7) in this exposition is that Jan. 2006.
[23] F. Vico, Z. Gimbutas, L. Greengard, and M. Ferrando-Bataller,
these expression are used in the convolutions in our numerical “Overcoming low-frequency breakdown of the magnetic field integral
scheme (see Section VI-A). equation,” IEEE Trans. Antennas Propag., vol. 61, no. 3, pp. 1285–1290,
Mar. 2013.
[24] M. Ganesh, S. C. Hawkins, and D. Volkov, “An all-frequency weakly-
R EFERENCES singular surface integral equation for electromagnetism in dielectric
media: Reformulation and well-posedness analysis,” J. Math. Anal.
[1] A. A. Kishk, Y. Yin, and A. W. Glisson, “Conical dielectric resonator Appl., vol. 412, no. 1, pp. 277–300, Apr. 2014.
antennas for wide-band applications,” IEEE Trans. Antennas Propag., [25] D. Colton and R. Kress, Inverse Acoustic and Electromagnetic Scattering
vol. 50, no. 4, pp. 469–474, Apr. 2002. Theory, 3rd ed. New York, NY, USA: Springer, 2013.
[2] I. M. Reaney and D. Iddles, “Microwave dielectric ceramics for res- [26] J. Wiersig, “Boundary element method for resonances in dielectric
onators and filters in mobile phone networks,” J. Amer. Ceram. Soc., microcavities,” J. Opt. A, Pure Appl. Opt., vol. 5, no. 1, pp. 53–60,
vol. 89, no. 7, pp. 2063–2072, Jul. 2006. Jan. 2003.
[3] S. J. Fiedziuszko et al., “Dielectric materials, devices, and circuits,” [27] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
IEEE Trans. Microw. Theory Techn., vol. 50, no. 3, pp. 706–720, Products, 7th ed. Amsterdam, The Netherlands: Elsevier, 2007.
Mar. 2002. [28] A. Gil, J. Segura, and N. M. Temme, Numerical Methods for Special
[4] A. B. Matsko and V. S. Ilchenko, “Optical resonators with whispering- Functions. Philadelphia, PA, USA: SIAM, 2007, p. 106.
gallery modes—Part I: Basics,” IEEE J. Sel. Topics Quantum Electron., [29] M. T. Heath, Scientific Computing: An Introductory Survey, 2nd ed.
vol. 12, no. 1, pp. 3–14, Jan./Feb. 2006. New York, NY, USA: McGraw-Hill, 2002, p. 241.
[5] L. He, Ş. K. Özdemir, and L. Yang, “Whispering gallery micro- [30] J. A. Stratton, Electromagnetic Theory. New York, NY, USA:
cavity lasers,” Laser Photon. Rev., vol. 7, no. 1, pp. 60–82, McGraw-Hill, 1941, p. 615.
Jan. 2013. [31] I. Breunig, B. Sturman, F. Sedlmeir, H. G. L. Schwefel, and K. Buse,
[6] M. R. Foreman, J. D. Swaim, and F. Vollmer, “Whispering gallery mode “Whispering gallery modes at the rim of an axisymmetric optical
sensors,” Adv. Opt. Photon., vol. 7, no. 2, pp. 168–240, Jun. 2015. resonator: Analytical versus numerical description and comparison with
[7] Y. K. Chembo and N. Yu, “Modal expansion approach to optical- experiment,” Opt. Exp., vol. 21, no. 25, pp. 30683–30692, Dec. 2013.
frequency-comb generation with monolithic whispering-gallery-mode [32] Y. Liu and A. H. Barnett, “Efficient numerical solution of acoustic
resonators,” Phys. Rev. A, Gen. Phys., vol. 82, no. 3, p. 033801, scattering from doubly-periodic arrays of axisymmetric objects,”
Sep. 2010. J. Comput. Phys., vol. 324, pp. 226–245, Nov. 2016.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Johan Helsing was born in Huddinge, Sweden, Anders Karlsson was born in Gothenburg,
in 1961. He received the Ph.D. degree in theoretical Sweden, in 1955. He received the Ph.D. degree in
physics from the KTH Royal Institute of Technol- theoretical physics from the Chalmers University
ogy, Stockholm, Sweden, in 1992. of Technology, Gothenburg, in 1984.
From 1992 to 1996, he held post-doctoral positions During 1985–1987, he was a Post-Doctoral Fellow
with the Courant Institute, New York, NY, USA, with the Applied Mathematical Sciences, Ames,
with The University of Utah, Salt Lake City, UT, IA, USA. Since 2000, he has been a Professor of
USA, and with the Rensselaer Polytechnic Insti- electromagnetic theory with Lund University, Lund,
tute, Troy, NY, USA. From 1997 to 2000, he was Sweden. His current research interests include
with Uppsala University, Uppsala, Sweden, and with direct and inverse scattering, accelerators, and wave
KTH. Since 2001, he has been a Professor with propagation in dispersive materials.
the Centre for Mathematical Sciences, Lund University, Lund, Sweden. His
current research interests include scientific computing, fast algorithms, integral
equations, and potential theory.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Abstract— This paper presents a rigorous scattering model of Containing slits on the upper plate of a PPW is an attractive
an infinite slit on a dielectric-loaded parallel-plate waveguide with feature as this type of deformation has the potential to become
an oblique incident TEM wave. The dyadic Green’s functions metal strip gratings [10] or leaky wave antennas [11], [12].
of an equivalent line source are derived first and then applied
to solve the scattered fields in terms of reflection, transmission, For problems of an incident TEM wave scattered by a slit on
and radiation. Special treatment with PEC/PMC symmetry is the PPW, the field expressions were derived in [13] and [14].
employed for fast and accurate calculation. The proposed model The reflection and transmission of TEM waves with a normal
was carefully validated by the full-wave simulator with excellent incidence were reported in [15], which furthermore provided
agreement. Some interesting scattering phenomena are found and the equivalent circuit parameters. Moreover, the coupling
emphasized, as they are crucial in explaining the special features
of slit-based planar structures. In this paper, we demonstrate problems related to a slitted PPW and a nearby conductor were
the presented model for the analysis and physical explanation of reported in [16] and [17]. In addition, a PPW with periodic slits
unique features found in a slitted substrate integrated waveguide, was also reported for leaky wave antennas’ applications [18].
grounded coplanar waveguide, and grounded coplanar stripline. To the best of our knowledge, all the previous models
The proposed model provides a simple and effective approach reported in the scattering of the slit-based PPWs were limited
for the design of potential slit-based planar circuits and antennas
based on the presented structures in this paper. to the normal incidence. The scattering of oblique incidence
for a slitted PPW involves more complicated Green’s functions
Index Terms— Parallel-plate waveguide (PPW), slitted with an additional dimension (a 2-D problem). However,
structure.
understanding fundamental mechanisms in the scattering of
I. I NTRODUCTION oblique incidence is crucial in the design of many slit-based
planar structures for PCB circuits and antenna applications.
= x̂ ×
2
¯ hm
Ḡ (1) (y|y ) · M(y )d y . (2)
− 2s x=0,y∈[− 2s , 2s ]
s
j kρ k z 2l + 1
π y H1(2)(kρ (yi − y ))d y
2
large original problem into two small equivalent problems. yz
Hil = − sin
That is, the original 2N × 2N inverse matrix issue was 2ωμ0 − 2s s
reduced to two N × N inverse matrix problems. The number ∞ 2l+1 2l+1
kz s π cos s π yi
of required matrix elements was reduced to a quarter of the + 2
dωμ0 n=−∞ γn2 + 2l+1
original number and the CPU time required for solving the s π
inverse matrices was considerably shortened. (−1)l γn cosh(γn yi )
− 2 (27)
γn2 + 2l+1 s π
A. PMC Symmetry s
yy 1 2 2m + 1
In the PMC symmetry, we combine the combined incident Him =− cos πy
fields of the original plane wave and the image plane wave 2ωμ0 − 2s s
and obtain 2 (2) kρ (2)
· k z H0 (kρ |yi − y |)+ H (kρ (yi − y )) d y
s
Convoluting the magnetic currents solved by (24) and the yy 1 2 2mπ
Him = − sin y
Green’s functions (32) and (33), the fundamental mode com- 2ωμ0 − 2s s
ponent of the scattered fields can be determined. By comparing
kρ
the ratio of the backward wave to the incident wave at the · k z2 H0(2)(kρ |yi − y |)+ (2)
H (kρ (yi − y )) d y
yi − y 1
slit edge where the reference plane is defined, we completely ∞
obtain the reflection coefficient of the PMC symmetry as j nπ 2
⎡ − + k z2
dωμ0 n=−∞ d
(−1)l j k y Mzl
cos k y 2s L−1
− j ky s ⎣
H = e 1− 2l+1 2 s yi −(−1) γn s sinh(γn yi )
sin 2mπ m 2mπ
d
l=0 s π − k 2y · 2 .
γn2 + 2mπ
M−1 ⎤ s
(43)
π sin k y 2s 2m + 1 (−1)m k z M ym
− ⎦.
d k y s 2m+1 π 2 − k 2 In the PEC case, the combined incident magnetic fields at
m=0 s y
the i th sampling point are represented as
(34)
2k y
Hzi = cos(k y yi )
ηk
B. PEC Symmetry
2 j kz
For the PEC symmetry, we can modify the combined H yi = sin(k y yi ). (44)
ηk
incident waves and the magnetic currents with the following
expressions: Similarly, once the magnetic currents are solved, we may
convolute with (32) and (33) for the scattered fields and accu-
E x = (e j ky cos θ − e− j ky cos θ )e− j kz sin θ rately obtain the reflection coefficient of the PEC symmetry
= 2 j sin(k y y)e− j kz z (35) expressed by
L−1
2 j kz sin k y 2s (−1)l k y Mzl
Hy = sin(k y y)e− j kz z (36) E = e − j ky s
−1+ 2lπ 2
ηk d
l=0 s − k 2y
2k y s M
Hz = cos(k y y)e− j kz z (37) cos k y 2 2mπ (−1)m j k z M ym
ηk − .
d k y s 2mπ 2 − k 2
L−1
2l m=1 s y
Mz = Mzl cos πy (38) (45)
s
l=0
M−1
2m IV. M ODEL VALIDATION BY F ULL -WAVE S IMULATION
My = M ym sin πy . (39) We validated the presented model with the full-wave simu-
s
m=0
lation tool ANSYS HFSS. The illustrated PCB substrate is
Substituting (36)–(39) into (24), the individual matrix elements 1.6 mm thick with a relative dielectric constant of r =
can be expressed as 4.4, assumed lossless for simplicity. We evaluated various
s parametric effects in terms of frequency dispersion, slit widths,
kρ2 2lπ
y H0(2)(kρ |yi − y |)d y
2
zz
Hil = − cos and incident angles. We conducted the comparison through a
2ωμ0 − 2 s s normal incident case and then an oblique incident case.
∞ Fig. 5 shows the reflection and transmission of a slitted
j k 2y s π yi − (−1) e
cos 2lπ l −γn 2s cosh(γ y )
n i
− 2
PPW in the normal incident case. We set the frequency
dωμ0 n=−∞ γ2 + 2lπ
window from 1 to 10 GHz in view of fundamental TEM
n s
(40) mode of our interest under the PPW dimensions, where the
s cutoff frequency of the next higher mode is 45 GHz. That
j kρ k z 2mπ 2 (2)
y H1 (kρ (yi − y ))d y
zy
Him = − sin is, the reflected and the transmitted waves can be regarded
2ωμ0 s
− 2s as only TEM plane waves. From Fig. 5(a), a wide slit has
∞ a lower transmission and higher reflection than a narrow
k z π 2m cos 2mπ
s yi − (−1) cosh(γn yi )
m
+ 2 slit, as expected. However, the reflection magnitude decreases
dωμ0 n=−∞ s γ 2 + 2mπ and the transmission magnitude increases as the frequency
n s
(41) increases. This phenomenon can be explained by a similar
s behavior of a grounded dielectric slab. The slit region in
j kρ k z 2l 2 (2)
π y H1 (kρ (yi − y ))d y
yz
Hil = − cos this problem can be considered as a grounded dielectric slab
2ωμ0 − 2s s in which the fundamental mode of a TM surface wave has
∞ 2lπ no cutoff frequency. At higher frequencies, the TM surface
k z 2lπ s sin s yi + (−1) γn sinh(γn yi )
l
− 2
waves are relatively more confined in the substrate than in the
dωμ0 n=−∞ γn2 + 2lπ air, leading to its characteristic wave impedance approaching
s
(42) the homogeneous limit of the substrate. Therefore, higher
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Fig. 9. SIW with a finite slit on its upper plate and the illustration of the
equivalent resonant TEM mode for the propagating TE10 mode.
Fig. 16. Grounded coplanar stripline with illustrated equivalent TEM mode
resonance for the propagation mode in the z-direction (w = 5 mm, s =
0.1 mm, and d = 1.6 mm). (a) Geometry. (b) Phase constant.
problem for the dispersive propagating modes and discussed [20] D. C. Chang and E. F. Kuester, “Total and partial reflection from the end
the limitations of the presented model. From an engineering of a parallel-plate waveguide with an extended dielectric slab,” Radio
Sci., vol. 16, no. 1, pp. 1–13, Jan./Feb. 1981.
point of view, the model not only resolves the classic scattering [21] E. F. Kuester, R. T. Johnk, and D. C. Chang, “The thin-substrate
problem of a slitted PPW with an oblique incidence but also approximation for reflection from the end of a slab-loaded parallel-plate
provides practical applications for slit-based planar structures waveguide with application to microstrip patch antennas,” IEEE Trans.
Antennas Propag., vol. AP-30, no. 5, pp. 910–917, Sep. 1982.
with unique features. From a scientific point of view, this [22] M. Bozzi, A. Georgiadis, and K. Wu, “Review of substrate-integrated
paper addresses the physical explanation for the interesting waveguide circuits and antennas,” IET Microw., Antennas Propag.,
phenomena discovered in the slitted PPW that correlate with vol. 5, no. 8, pp. 909–920, Jun. 2011.
[23] H. Shigesawa, M. Tsuji, and A. A. Oliner, “Dominant mode power
the unique features found in slit-based planar structures. leakage from printed-circuit waveguides,” Radio Sci., vol. 26, no. 2,
pp. 559–564, Mar./Apr. 1991.
[24] N. K. Das, “Methods of suppression or avoidance of parallel-plate
R EFERENCES power leakage from conductor-backed transmission lines,” IEEE Trans.
Microw. Theory Techn., vol. 44, no. 2, pp. 169–181, Feb. 1996.
[1] R. F. Harrington, Time-Harmonic Electromagnetic Fields. New York,
NY, USA: Wiley, 2001.
[2] D. M. Pozar, Microwave Engineering, 3rd ed. New York, NY, USA:
Wiley, 2004.
[3] B. Tomasic and A. Hessel, “Electric and magnetic current sources in the
parallel plate waveguide,” IEEE Trans. Antennas Propag., vol. AP-35,
no. 11, pp. 1307–1310, Nov. 1987.
[4] G. K. C. Kwan and N. K. Das, “Excitation of a parallel-plate dielectric
waveguide using a coaxial probe-basic characteristics and experiments,”
IEEE Trans. Microw. Theory Techn., vol. 50, no. 6, pp. 1609–1620,
Jun. 2002.
[5] R. Safavi-Naini and R. H. Macphie, “On solving waveguide junction
scattering problems by the conservation of complex power technique,” Yao-Wen Hsu (S’11) was born in Kaohsiung,
IEEE Trans. Microw. Theory Techn., vol. MTT-29, no. 4, pp. 337–343, Taiwan. He received the B.S. degree in electrical
Apr. 1981. engineering and M.S. degree in communication engi-
[6] C. M. Angulo and W. S. C. Chang, “The launching of surface waves by neering from National Taiwan University, Taipei,
a parallel plate waveguide,” IRE Trans. Antennas Propag., vol. 7, no. 4, Taiwan, in 2009 and 2011, respectively, where he
pp. 359–368, Oct. 1959. is currently pursuing the Ph.D. degree in communi-
[7] E. Arnieri and G. Amendola, “Analysis of substrate integrated waveguide cation engineering.
structures based on the parallel-plate waveguide Green’s function,” His current research interests include the design
IEEE Trans. Microw. Theory Techn., vol. 56, no. 7, pp. 1615–1623, and analysis of leaky wave antennas, especially on
Jul. 2008. the mode analysis and size optimization of low-
[8] Y. Zhang, J. Fan, G. Selli, M. Cocchini, and F. de Paulis, “Analytical profile directive antennas, and also include antennas
evaluation of via-plate capacitance for multilayer printed circuit boards and arrays for satellite and 5G applications, including design, manufacturing,
and packages,” IEEE Trans. Microw. Theory Techn., vol. 56, no. 9, and measurement techniques.
pp. 2118–2128, Sep. 2008.
[9] E. X. Liu, E. P. Li, Z. Z. Oo, X. Wei, Y. Zhang, and R. Vahldieck,
“Novel methods for modeling of multiple vias in multilayered parallel-
plate structures,” IEEE Trans. Microw. Theory Techn., vol. 57, no. 7,
pp. 1724–1733, Jul. 2009.
[10] R. Li and A. Oliner, “Scattering resonances on a fast-wave struc-
ture,” IEEE Trans. Antennas Propag., vol. AP-13, no. 6, pp. 948–959,
Nov. 1965.
[11] R. C. Honey, “A flush-mounted leaky-wave antenna with predictable
patterns,” IRE Trans. Antennas Propag., vol. 7, no. 4, pp. 320–329,
Oct. 1959.
[12] R. E. Collin, “Analytical solution for a leaky-wave antenna,” IRE Trans.
Antennas Propag., vol. AP-10, no. 5, pp. 561–565, Sep. 1962. Yi-Cheng Lin (S’92–M’98–SM’10) received the
[13] C. Butler and E. Yung, “Properties of a slotted parallel-plate waveguide B.S. degree from National Tsing-Hua University,
excited by an incident TEM wave,” in Proc. Antennas Propag. Soc. Int. Hsinchu, Taiwan, the M.S. degree from National
Symp., Oct. 1976, pp. 404–407. Taiwan University, Taipei, Taiwan, and the Ph.D.
[14] T. Keshavamurthy and C. Butler, “Characteristics of a slotted parallel- degree from the University of Michigan, Ann Arbor,
plate waveguide filled with a truncated dielectric,” IEEE Trans. Antennas MI, USA.
Propag., vol. AP-29, no. 1, pp. 112–117, Jan. 1981. From 1997 to 2003, he was with Qualcomm Inc.,
[15] Y. K. Cho, “On the equivalent circuit representation of the slitted San Diego, CA, USA, where he was involved in
parallel-plate waveguide filled with a dielectric,” IEEE Trans. Antennas the design and development of various antennas
Propag., vol. 31, no. 9, pp. 1193–1200, Sep. 1989. for satellites and terrestrial mobile communication
[16] P. D. Mannikko, C. C. Courtney, and C. M. Butler, “Slotted parallel- applications. In 2003, he joined the faculty of the
plate waveguide coupled to a conducting cylinder,” Proc. Inst. Elect. Department of Electrical Engineering and the Graduate Institute of Commu-
Eng.—Microw., Antennas Propag., vol. 139, no. 2, pt. H, pp. 193–201, nication Engineering, National Taiwan University, where he is currently a
Apr. 1992. Professor. From 2014 to 2015, he was a Visiting Scholar with the Center
[17] J.-I. Lee, C.-H. Lee, and Y.-K. Cho, “Electromagnetic coupling mech- of Wireless Communications, University of California at San Diego, La
anism to a conducting strip through a narrow slit in a parallel- Jolla, CA. His current research interests include EBG/metamaterial-embedded
plate waveguide,” IEEE Trans. Antennas Propag., vol. 49, no. 4, antennas, millimeter-wave antennas of system in packaging, ultrawideband
pp. 592–596, Apr. 2001. and multiband antennas, and compact diversity antennas for multi-input multi-
[18] C.-W. Lee and H. Son, “Periodically slotted dielectrically filled parallel- output systems.
plate waveguide as a leaky-wave antenna: E-polarization case,” IEEE Dr. Lin was the recipient of the 2013–2014 Distinguished Lecturer of Taiwan
Trans. Antennas Propag., vol. 47, no. 1, pp. 171–178, Jan. 1999. Electromagnetic Industry–Academia Consortium. His team was the recipient
[19] R. E. Collin, “Green’s functions,” in Field Theory of Guided Waves. of the First Prize of the National Mobile Handset Antenna Design Competition
New York, NY, USA: Wiley, 1991, ch. 2. of Taiwan in 2008.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Abstract— In this paper, the resonant cavity problem with results. Nédélec [4] has proposed the edge element, which is
anisotropic fully conducting media, complex geometrical struc- a curl-conforming element. This edge element can preserve
ture and perfect electric conductor walls is investigated. We solve the tangential continuous property of electric field E on the
this problem based on the finite element method (FEM) with
tangential and linear normal (CT/LN) element and standard interface, and hence, the edge element is widely applied in
linear element. An effective numerical method is proposed by computational electromagnetics. For example, in the first place
us such that it is free of nonphysical modes. After the FEM Lee et al. [5] have made use of tangential vector finite
discretization, we need to solve a quadratic algebraic eigenvalue elements to solve dielectric waveguide problems, and provided
problem with a linear constraint condition. In order to overcome a posteriori error estimates. Early literatures on the edge
this difficulty in the field of numerical algebra, we change
this algebraic eigenvalue problem into a generalized eigenvalue elements claim that the edge elements do not introduce the
problem by introducing an auxiliary zero eigenvector. Moreover, any spurious modes to solve the resonant cavity problem.
when the permittivity and conductivity are two constants, both However, this conclusion is not correct, please see [6, Ch. 9,
the eigenmodes of infinite algebraic multiplicity and all the pp. 306–309], because if we use the edge element to solve
nonphysical modes are also removed by linearization method. the resonant problem and ignore divergence-free condition
Several numerical experiments show that computational method
in this paper can suppress all the spurious modes. in the numerical calculation, then spurious zero modes will
be present in the numerical results [7]. The eigenfunctions
Index Terms— Edge element, nonlinear Maxwell’s eigenvalue associated with these spurious zero modes do not have
problem, nonphysical modes, quadratic algebraic eigenvalue
problem, resonant cavity problem. property of divergence-free condition. How to enforce
divergence-free condition in numerical computation is an
important problem. We must make the eigenfunctions be
I. I NTRODUCTION divergence-free such that numerical solution cannot introduce
any nonphysical modes. For the perfect electric conduc-
F OR a microwave resonator, resonant frequency is a very
important physical parameter. It is very difficult to find
analytical solution of resonant frequency when the resonator
tor (PEC) resonant cavity problem filled with a homogeneous
isotropic medium, in the first place, Kikuchi [8] has introduced
is with complex geometrical structure and/or inhomogeneous a Lagrange multiplier to enforce Gauss’ law, and numerical
media. Therefore, we usually turn to the help of numerical computational method proposed by Kikuchi is free of the
methods to obtain approximate resonant frequencies, for exam- any spurious modes. This numerical method is based on
ple, finite element method (FEM) [1] and finite difference the idea of mixed FEM. About theoretical foundation of
method [2]. mixed FEM (please see the monograph written in [9]). An
According to classic electromagnetic theory [3], it is known efficient numerical method must have ability in removing all
that electric field E is tangential continuous on the interface the nonphysical modes together.
between two different media, while the electric field E is not When the medium in the whole cavity is anisotropic lossless
normal continuous on this interface. If the numerical method and the cavity is with PEC boundary condition, this problem
cannot preserve interface condition of the electric field E, then is a linear vector Maxwell’s eigenvalue problem in 3-D [10].
many spurious nonzero modes will be present in the numerical Venkatarayalu and Lee [11] have proposed a novel computa-
tional method to solve this problem successfully. Their method
Manuscript received September 17, 2016; revised November 7, 2016, needs to solve a generalized matrix eigenvalue problem with
December 19, 2016, and January 14, 2017; accepted January 18, 2017. This a linear constraint condition after the FEM discretization.
work was supported by the National Natural Science Foundation of China
under Grant 11371357, Grant 11101381, and Grant 41390453. (Corresponding In mathematics, this problem is a linear eigenvalue problem
author: Wei Jiang.) about self-adjoint compact operator. On the basis of the spec-
W. Jiang is with the Institute of Geophysics and Geomatics, China Univer- tral theory of self-adjoint compact operator [12], we know that
sity of Geosciences, Wuhan 430074, China (e-mail: jwmathphy@163.com).
J. Liu and X. Xiong are with the Department of Electronic Science, Institute the eigenvalues of self-adjoint compact operator are a series of
of Electromagnetics and Acoustics, Xiamen University, Xiamen 361005, real numbers, which is in accordance with the conclusion that
China (e-mail: liujie190484@163.com; iema@xmu.edu.cn). this cavity problem has real resonant frequencies. By virtue of
Q. H. Liu is with the Department of Electrical and Computer Engineering,
Duke University, Durham, NC 27708 USA (e-mail: qhliu@duke.edu). mixed FEM, Jiang et al. [13] have solved this resonant cavity
Digital Object Identifier 10.1109/TMTT.2017.2661740 problem, and it is shown that mixed FEM cannot introduce
0018-9480 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
the nonphysical modes. Based on mixed spectral element vector on the boundary ∂. Let E and H be the electric field
method (SEM), Liu et al. [14] have also solved this problem, phasor and magnetic field phasor, respectively. ω is angular
and mixed SEM cannot also introduce the nonphysical modes. frequency of time-harmonic electromagnetic wave.
In fact, the idea of mixed SEM and mixed FEM is almost the Permittivity and permeability in vacuum is usually denoted
same, because these two numerical methods all employ the by 0 and μ0 , respectively. Let r be relative permittivity
same mixed variational form. The unique difference between tensor in the medium. Relative permeability tensor in the
mixed SEM and mixed FEM is that the constructions of medium is usually denoted by μr . Suppose that r and μr
basis functions in these two methods are different. When are two piecewise constant tensors, and they satisfy
the eigenfunction for the linear vector Maxwell’s eigenvalue † †
problem about curl–curl operator is very smooth, the numerical r = r μr = μr (1)
eigenvalue obtained by mixed SEM has spectral accuracy, but and
the numerical eigenvalue obtained by mixed FEM of low order
does not have spectral accuracy. But for practical resonant ξ † r ξ > 0 ∀ ξ ∈ C3 and ξ = 0 (2)
cavity problem with complex geometry, the corresponding ξ μr ξ > 0 ∀ ξ ∈ C
† 3
and ξ = 0 (3)
eigenfunction is usually not smooth, which will lead to the lack
of spectral accuracy of numerical eigenvalue if we use mixed where the superscript † denotes conjugate transpose of a
SEM to solve this resonant cavity. In this case, superiority tensor. From (1)–(3), we know that r and μr are two
of mixed FEM is much better than the one of mixed SEM, second-order positive definite Hermitian tensors, which are
because mixed FEM can adopt an unstructured grid to deal the conditions for an anisotropic lossless medium according to
with a very complex geometrical structure. electromagnetic theory [3], [15]. Let σ e be conductivity tensor
To our best knowledge, the research about the computational in the medium. Moreover, suppose that the medium is fully
method without any nonphysical mode for the resonant cavity conducting in the whole cavity. In this case, the conductivity
problem with complex geometrical structure and anisotropic σ e in the whole cavity is also a second-order positive
fully conducting media has not appeared yet in existing liter- definite Hermitian tensor, that is,
†
ature until now. This problem is a nonlinear vector Maxwell’s σe = σe
eigenvalue problem, which is a much more difficult problem
than the liner vector Maxwell’s eigenvalue problem. This and
paper proposes an efficient numerical method to solve it. It is ξ † σ e ξ > 0 ∀ ξ ∈ C3 and ξ = 0.
known that CT/LN edge element space can approximate curl-
conforming space very well, and standard linear element space Note that σ e may not be a constant tensor in the cavity .
can approximate grad-conforming space very well. Therefore, According to the Maxwell’s equations with source-free we
CT/LN edge element space and standard linear element space arrive at
⎧
are adopted in this paper such that Gauss’ law and charge ⎪
⎪ ∇ × E = − j ωμ0 μr H, in (4a)
⎪
⎪ ρec
conservation law are satisfied in the weak sense together. ⎪
⎪ ∇ · ( r E) = , in
⎪
⎪ (4b)
Moreover, numerical experiments show that the computational ⎪
⎪ 0
method in this paper does not introduce the any nonphysical ⎪
⎪
⎨ ∇ · (σ e E) = − j ωρec , in (4c)
modes.
The outline of this paper is as follows. Variational formula- ⎪ ∇ × H = j ω0 r E + σ e E, in
⎪
(4d)
⎪
⎪
⎪
⎪ ∇ · (μr H) = 0, in (4e)
tions for the resonant cavity problem filled with anisotropic ⎪
⎪
⎪
⎪ n̂ × E = 0, on ∂
fully conducting media are given in Section II. The FEM ⎪
⎪ (4f)
discretizations corresponding to the variational forms of con- ⎩
n̂ · (μr H) = 0, on ∂ (4g)
tinuous problem are given in Section III. In Section IV, we
recall how to solve a quadratic algebraic eigenvalue prob- where ρec is the conduction electric charge density in the
lem, and give an effective algorithm about how to solve a cavity . We want to seek a quaternion (ω, E, H, ρec ) with
quadratic algebraic eigenvalue problem with a linear constraint (ω, E, H, ρec ) = (0, 0, 0, 0), such that (4) is valid.
condition. Finally, several numerical experiments verify that From Maxwell’s system (4), eliminating the scalar field ρec
numerical algorithm in this paper can solve the resonant and the vector magnetic field H, we achieve the following
cavity with anisotropic fully conducting media and PEC walls, partial differential equations (PDEs): seek k0 ∈ C and E = 0
such that
and our proposed numerical algorithm can remove all the ⎧ −1
nonphysical modes. ⎪
⎨ ∇ × (μr ∇ × E) = k0 (k0 r − j η0 σ e )E, in (5a)
⎪ ∇ · ((k0 r − j η0 σ e )E) = 0, in (5b)
II. G OVERNING E QUATIONS ⎩
n̂ × E = 0, on ∂ (5c)
A. Preliminaries where k0 = ω(0 μ0 )1/2 is the wavenumber in vacuum, and
Assume that ⊂ R3 is a bounded connected domain. ∂ is η0 = (μ0 /0 )1/2 is the wave impedance in vacuum. It is clear
the boundary of the cavity . Note that the boundary ∂ can that when k0 = 0, (5b) is the conclusion of (5a). But (5b) is
be disconnected. In this case, the cavity has multiple perfect not omitted, otherwise nonphysical zero modes (0, ∇ϕ) will
electric conductors. Suppose that n̂ is the outward normal unit be introduced in the eigenmodes. In order to suppress these
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
nonphysical zero modes, we must consider divergence-free associates with a linear eigenvalue problem about self-adjoint
condition (5b) in numerical simulation. compact operator. PDEs (7) can be solved by mixed FEM
Now we prove that the nonzero eigenmodes between successfully [13].
PDEs (5) and PDEs (4) are the same. Obviously, a nonzero However, PDEs (5) is a nonlinear vector Maxwell’s eigen-
eigenmode to PDEs (4) is a nonzero eigenmode to PDEs (5). value problem in 3-D, which is a much more difficult problem
Conversely, we prove that a nonzero eigenmode to PDEs (5) than linear vector Maxwell’s eigenvalue problem (7) in 3-D.
is also a nonzero eigenmode to PDEs (4). When k0 = 0, then PDEs (5) is a nonlinear eigenvalue problem about a noncom-
ω = 0, set pact operator, because from numerical experiments we find
−1 that the spectrum of PDEs (5) has a nonzero accumulation
μ ∇×E point or a continuous spectrum. This will lead to the difficulty
H= r . (6)
− j ωμ0 of the numerical simulation. In mathematics, it is known that
It is clear that (4a) holds. Substituting (6) into (5a), we can research about the spectral theory of compact operator in
achieve functional analysis is very thorough [12], but the research
achievements about the spectral theory of noncompact operator
k02 r E − j η0 k0 σ e E
∇×H= = j ω0 r E + σ e E in functional analysis are very limited. Therefore the spectrum
− j ωμ0 properties of the operator associated with PDEs (5) are almost
which is just (4d) unknown in theory.
1 −1
∇ · (μr H) = ∇ · (μr μr ∇ × E) = 0 B. Linearized Method
− j ωμ0
which is just (4e) Specially when σ e = σe I and r = r I , where σe and r
are two positive constants in , and I is the identical tensor,
1 then ( j σe r−1 0−1 , −∇ϕ, 0, −0 ∇ ·(r ∇ϕ)) is an eigenmode to
n̂ · (μr H) = ∇ · (n̂ × E) = 0, on ∂
j ωμ0 Maxwell’s system (4), where ϕ ∈ H01(). In this case, ω =
which is just (4g). Finally, we verify Gauss’ law for electric j σe r−1 0−1 is an eigenvalue of infinite algebraic multiplicity,
field E (4b) and charge conservation equation (4c), which can and E = −∇ϕ belongs to function space ∇ H01(), which is
be deduced from (5b). In fact, from (5b), we obtain a function space with infinite dimension. In mathematics, the
√ eigenvalue j σe r−1 0−1 is called an essential spectral point. But
ω 0 μ0
∇ · (σ e E) = ∇ · ( r E) = − j ω0 ∇ · ( r E). these eigenmodes do not contain magnetic energy, therefore
j η0
we are not interested in these eigenmodes. This paper can
Set ρec = 0 ∇ · ( r E), then we have suppress them by virtue of the method based on solving a
quadratic algebraic equation.
∇ · (σ e E) = − j ωρec
First, using mixed FEM to solve a linear vector Maxwell’s
which are just (4b) and (4c). eigenvalue problem: seek ∈ R and E = 0 such that
⎧ −1
When the cavity is simply connected, it is easy to prove ⎪
⎨ ∇ × μr ∇ × E = E, in
2
(8a)
that ω = 0 is not an eigenvalue for Maxwell’s system (4).
In fact, if ω = 0 is an eigenvalue for Maxwell’s system (4), ⎪ ∇ · E = 0, in (8b)
⎩
then Maxwell’s system (4) will have a trivial solution. By (4a), n̂ × E = 0, on ∂. (8c)
we get ∇ × E = 0 in , since is simply connected, then Next, solve the roots to a quadratic algebraic equation: seek
E = −∇ϕ can be deduced. Substituting this into (4c), we can k0 ∈ C, such that
get −∇ · (σ e ∇ϕ) = 0. Since σ e is a second-order positive
k02 r − j η0 σe k0 − 2
= 0. (9)
definite Hermitian tensor in the whole cavity and (4f), then
it is very easy to prove that ∇ϕ = 0, thus E = 0 in . In It is clear that the solutions to (9) are
terms of (4b), we get ρec = 0 in . Similarly, we can prove
H = 0 in . ± 4r 2 − η02 σe2 + j η0 σe
k0 = .
When σ e = 0 in the whole cavity , we have ρec = 0 2r
in . In this case, charge conversation law (4c) holds spon- If we employ mixed FEM to solve PDEs (8), next
taneously. Therefore we do not need to consider charge using (9) to obtain approximate wavenumber, then this lin-
conversation law (4c) in the numerical simulation. In this earized method can not only suppress all the spurious zero
case, the governing equations about the electric field E are and nonzero modes successfully, but it can also suppress the
the following: find k0 ∈ R and E = 0 such that physical modes ( j σe r−1 0−1 , −∇ϕ, 0, −0 ∇·(r ∇ϕ)) in which
⎧ −1
⎪ we are not interested. Notice that numerical simulation about
⎨ ∇ × μr ∇ × E = k0 r E, in
2
(7a)
the modes ( j σe r−1 0−1 , −∇ϕ, 0, −0 ∇ · (r ∇ϕ)) will spend
⎪ ∇ · ( r E) = 0, in (7b) a lot of CPU time. Note that PDEs (8) and (9) can also
⎩
n̂ × E = 0, on ∂. (7c) simulate the resonant cavity problem with σ e = σe I and
Obviously, when k0 = 0, (7b) is the conclusion of (7a). r = r I , where σe and r are two positive constants in the
PDEs (7) is investigated in [13]. PDEs (7) is a linear PDE whole domain , but μr may be inhomogeneous anisotropic
eigenvalue problem about curl–curl operator in 3-D, which in .
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
III. F INITE E LEMENT D ISCRETIZATION edge in mesh Th . We usually employ CT/LN edge element
Because it is very difficult to establish the mixed variational space Wh to approximate curl-conforming space H(curl, ),
form for PDEs (5), we plan to give the variational form for and standard linear element space S h to approximate the
PDEs (5) according to the Green’s formulas directly. Hilbert space H 1(). From [17], it is known that Wh is
First, we introduce some important function spaces over C a finite element subspace of curl-conforming function space
in FEM associated with PDEs (5) H(curl, ); from [16], it is known that S h is a finite element
subspace of the Hilbert space H 1(). Set W0h = Wh V
L 2 () = f : | f (x, y, z)|2 dxdydz < +∞ and S0h = S h H01(). For the definitions of both CT/LN
edge element space W0h and linear element space S0h (please
H 1() = { f ∈ L () : ∇ f ∈ (L 2 ())3 }
2 see [13]). Suppose that
H01() = { f ∈ H 1() : f |∂ = 0} W0h = span{N1 , N2 , N3 , . . . , Nn }
H(curl, ) = {M ∈ (L 2 ())3 : ∇ × M ∈ (L 2 ())3 } S0h = span{φ1 , φ2 , φ3 , . . . , φm }
H0 (curl, ) = {M ∈ H(curl, ) : n̂ × M|∂ = 0}
where Ni , i = 1, 2, 3, . . . , n (n = dim W0h ) are the global basis
Set V = H0 (curl, ). functions in W0h , and φi , i = 1, 2, 3, . . . , m (m = dim S0h ) are
Second, denote the continuous sesquilinear forms the global basis functions in S0h .
Now, we give discrete variational forms corresponding
a: V×V → C to variational forms (10): seek kh ∈ C, Eh ∈ W0h with
(E, M) →
−1
μr ∇ × E · ∇ × M∗ d x d ydz Eh b1 = 1, such that
kh2 b1 (Eh , M) − j η0 kh b2 (Eh , M) − a(Eh , M) = 0 ∀M ∈ W0h
b1 : (L 2 ())3 × (L 2 ())3 → C
(11a)
(E, M) → r E · M∗ d x d ydz kh c1 (Eh , q) − j η0 c2 (Eh , q) = 0 ∀q ∈ S0h
b2 : (L 2 ())3 × (L 2 ())3 → C (11b)
(E, M) → σ e E · M∗ d x d ydz where kh in (11) is the approximate wavenumber in vacuum.
The notation kh is in order to distinguish the exact wavenum-
c1 : V × H01() → C ber k0 . Write
(M, q) → r M · ∇q ∗ d x d ydz n
Eh = ξi Ni = [N]ξ ξ T = [ξ1 , ξ2 , . . . , ξn ].
c2 : V × H01() →C i=1
Taking M = Ni , q = φi in (11), then we obtain the algebraic
(M, q) → σ e M · ∇q ∗ d x d ydz
eigenvalue problem
where ∗ stands for complex conjugate. kh2 B1 − j η0 kh B2 − A ξ = 0 (12a)
Define M b1 = (b1 (M, M))1/2 . It is easy to prove that
kh C1 − j η0 C2 ξ = 0 (12b)
· b1 is a norm in (L 2 ())3 . This norm is equivalent to the
usual norm · 0 in (L 2 ())3 . In our numerical experiments, where
we normalize the numerical eigenfunction according to the
A = (ai j ) ∈ Cn×n ai j = a(N j , Ni )
norm · b1 . (1) (1)
Finally, we give variational form of PDEs (5) by using B1 = bi j ∈ Cn×n bi j = b1 (N j , Ni )
Green’s formulas: seek k0 ∈ C and E ∈ V with E b1 = 1,
B2 = bi(2)
j ∈ Cn×n bi(2)
j = b2 (N j , Ni )
such that (1) (1)
C1 = ci j ∈ Cm×n ci j = c1 (N j , φi )
k02 b1 (E, M) − j η0 k0 b2 (E, M) − a(E, M) = 0 ∀M ∈ V
C2 = ci(2)
j ∈ Cm×n ci(2)
j = c2 (N j , φi ).
(10a)
k0 c1 (E, q) − j η0 c2 (E, q) = 0 ∀q ∈ H01(). If we do not consider constraint (5b), finally one will only
(10b) obtain a quadratic algebraic eigenvalue problem (12a).
According to the continuous sesquilinear forms above, it is
Now we already know that edge element space can approx- known that A is a positive semidefinite Hermitian matrix, and
imate curl-conforming space V very well, and the function B1 and B2 are two positive definite Hermitian matrices. It is
space based on nodal-basis functions can approximate the clear that kh = 0 is an eigenvalue with finite algebraic multi-
Hilbert space H 1() very well. Moreover, edge element space plicity to problem (12a) because A is a positive semidefinition
has ability in dealing with linear electromagnetic eigenvalue Hermitian matrix. But most of these eigenvalues kh = 0 are
problem with discontinuous media, and it will not introduce nonphysical zero eigenvalues for original PDEs (4). In order
nonphysical nonzero modes. to eliminate these nonphysical zero eigenvalues, we have to
Let Th be a regular tetrahedral partition [16] of with solve problem (12), which is a difficult numerical algebraic
mesh parameter h, where h is the length of the longest eigenvalue problem in the field of numerical algebra. In the
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
next section, we will deal with this troublesome numerical Since φ ∈ H01(), then
algebraic problem.
M = ∇φ ∈ H0 (curl, ). (16)
IV. A LGEBRAIC E IGENVALUE P ROBLEMS Substituting (16) into the continuous sesquilinear form
Problem (12a) is a quadratic algebraic eigenvalue prob- about c1 , and taking q = φ in the continuous sesquilin-
lem, which is the easiest nonlinear algebraic eigenvalue prob- ear form about c1 , from (15) we achieve c1 (∇φ, φ) = 0.
lem, while problem (12) is a quadratic algebraic eigenvalue According to the positive definiteness of relative permittivity
problem with a linear constraint condition. About review paper tensor r , we can infer that ∇φ = 0 in . Thanks to
for how to solve the quadratic algebraic eigenvalue problem, the Poincaré inequality [16] in Hilbert space H01(), i.e.,
we are glad to recommend [18]. The most common way of φ 1 ≤ C ∇φ 0 = 0 in , then φ = 0 in . Since the basis
dealing with the quadratic algebraic eigenvalue problem is to functions φi (i = 1, 2, 3, . . . , m) are linear independent, thus
transform it into a linear generalized eigenvalue problem [19]. di∗ = 0 (i = 1, 2, 3, . . . , m), i.e., di = 0 (i = 1, 2, 3, . . . , m),
But this linearized method will double the matrix size. This this implies that ζi (i = 1, 2, 3, . . . , m) are linear independent.
paper adopts this linearized method to solve the quadratic alge- Therefore Rank(C1 ) = m, this also implies Rank(C1† ) = m.
braic eigenvalue problem. The algebraic eigenvalue problem Similarly, we can also prove that Rank(C2† ) = m. Note that in
(12a) can be changed into a generalized eigenvalue problem the proof we need to use positive definiteness of conductivity
tensor σ e , this is why we assume that the medium in the cavity
j η0 B2 A kh ξ B1 O kh ξ
= kh (13) is fully conducting. By virtue of linear algebra theory [20], the
In O ξ O In ξ
solution spaces to linear algebraic equations (14a) and (14b)
where In is the identity matrix of order n. are two trivial spaces. Hence D = {0}.
Obviously, problem (13) has dim (Null(A)) zero eigenval- Next we can change the algebraic eigenvalue problem (12)
ues, where Null stands for taking kernel space of a matrix. into the following generalized eigenvalue problem:
Most of these zero eigenvalues are nonphysical zero eigenval- ⎡ ⎤⎡ ⎤
ues. We want to suppress these nonphysical zero eigenvalues j η0 B2 A C1† kh ξ
in the numerical computation. ⎣ In O j η0 C2† ⎦ ⎣ ξ ⎦
Commercial softwares COMSOL Multiphysics 4.2a and C1 − j η0 C2 O α
HFSS (High Frequency Structure Simulator) 10.0 usually ⎡ ⎤⎡ ⎤
B1 O O kh ξ
employ eigenvalue problem (13) to solve (5). It is clear that = k h ⎣ O In O ⎦ ⎣ ξ ⎦ . (17)
this method will introduce many nonphysical zero modes. O O O α
Now let us consider how to solve the algebraic eigenvalue
problem (12). First, we consider the following linear algebraic From the proof above, we know that α = 0 in the general-
equations: ized eigenvalue problem (17). The numerical algebraic prob-
lem (17) can suppress all the nonphysical modes, including
C1† α = 0 (14a) nonphysical zero modes, because we have already considered
C2† α = 0. (14b) Gauss’ law (4b) and charge conversation law (4c) for the
electric field E in the numerical computation.
It is well-known that the solutions to linear algebraic equations
(14a) and (14b) are two linear subspaces in Cm , respectively.
V. N UMERICAL E XPERIMENTS
Assume that solution spaces to (14a) and (14b) are D1 and D2 ,
respectively. Define the sum spaces D = Null(C1† )+Null(C2† ). In this section, five numerical experiments are carried out
In fact, D1 = Null(C1† ) and D2 = Null(C2† ). Now, we prove in order to examine the effectiveness of the proposed com-
that Rank(C1† ) = Rank(C2† ) = m. We can divide into row putational method. Our aim is to confirm that the proposed
(1) (1) (1) algorithm cannot introduce the any spurious modes. In these
groups to the matrix C1 , i.e., set ζi = [ci1 , ci2 , . . . , cin ],
five numerical experiments, we shall take the numerical
i = 1, 2, . . . , m. Suppose that
wavenumber kh in vacuum as the eigenvalue. We do not list the
m
continuous spectrum in our numerical results, because the task
d i ζi = 0 of listing the continuous spectrum is very difficult. We plan to
i=1
list the first eight numerical eigenvalues in all the examples. By
then we have the way, if we adopt the generalized eigenvalue problem (17)
m to solve PDEs (5), we will validate the auxiliary eigenvector
di ci(1)
j = 0, j = 1, 2, . . . , n α = 0 corresponding to the numerical eigenvalue kh =
i=1 a +bj (a > 0), and record the size of the matrix in generalized
that is, eigenvalue problem (17) and the number of nonzero matrix
m entries in problem (17). Recall that α = [α1 , α2 , . . . , αm ]T ,
c1 (N j , di∗ φi ) = 0, j = 1, 2, . . . , n. where αi = ai + j bi , ai ∈ R and bi ∈ R (i = 1, 2, . . . , m).
i=1 Define the norm: α = max1≤i≤m (ai2 + bi2 )1/2 . Let #1 be the
m ∗ size of the matrix in generalized eigenvalue problem (17). Let
Set φ = i=1 di φi , then
#2 and #3 be the number of nonzero entries in the left and
c1 (N j , φ) = 0, j = 1, 2, . . . , n. (15) right matrix in problem (17), respectively.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
TABLE I
D ISCRETE E IGHT E IGENVALUES F ROM THE R ECTANGULAR C AVITY F ILLED W ITH A H OMOGENEOUS M EDIUM
TABLE II
D ISCRETE E IGHT E IGENVALUES F ROM THE R ECTANGULAR C AVITY F ILLED W ITH AN I NHOMOGENEOUS M EDIUM
TABLE III
D ISCRETE E IGHT E IGENVALUES F ROM THE T ORIC C AVITY
and
3 1−2j 1− j
σe = 1+2j 4 2+ j × 10−3 S/m.
1+ j 2− j 4
Fig. 1. Toric cavity.
Here, we also adopt the generalized eigenvalue problem (17)
to solve PDEs (5), and show the computational results in
Table IV.
Table IV implies that our method is free of nonphysical
modes. In this example, there does not exist a physical zero
mode. However, for the case in anisotropic lossless media,
there exists a physical zero mode. This is because this cavity
has two disconnected PEC boundaries [13]. From Table IV,
one can observe that numerical eigenvalues are convergent
to the corresponding exact eigenvalues as mesh parameter h
decreases, and α (i) ≈ 0 (i = 1, 2, 3, 4) in , which is in
agreement with the theory in Section IV.
TABLE IV
D ISCRETE E IGHT E IGENVALUES F ROM THE C AVITY W ITH O NE H OLE
TABLE V
D ISCRETE E IGHT E IGENVALUES F ROM THE C AVITY W ITH T WO H OLES
⎡ ⎤
3 2+ j j
(1) ⎢ 0⎥ −3
σe = ⎣2− j 2 ⎦ × 10 S/m,
−j 0 4
⎡ ⎤
4 2+3j 0
(2) ⎢ 0⎥ −4
σe = ⎣2− 3j 4 ⎦ × 10 S/m.
0 0 1
a quadratic algebraic eigenvalue problem with a linear con- Wei Jiang received the B.S. degree in mathematics
straint condition. This difficult numerical algebraic eigenvalue and applied mathematics from Hubei Normal Uni-
versity, Huangshi, China, in 2008, the M.S. degree
problem can be reduced to a generalized eigenvalue problem in computational mathematics from Guizhou Normal
by introducing an auxiliary zero eigenvector. The proposed University, Guiyang, China, in 2011, and the Ph.D.
method can eliminate nonphysical zero and nonzero modes degree in radio physics from Xiamen University,
Xiamen, China, in 2016.
together. Moreover, when the permittivity and conductivity are He is currently a Post-Doctoral Researcher with
two constants, the linearization method can remove both the the Institute of Geophysics and Geomatics, China
nonphysical modes and the eigenmodes of infinite algebraic University of Geosciences, Wuhan, China. His
current research interests include the finite ele-
multiplicity. The proposed method only employs the easiest ment method for PDE eigenvalue problems, applied numerical algebra,
CT/LN edge element space and standard linear element space, and computational electromagnetics, especially for eigenvalue problems in
but it is very easy to extend our proposed method to higher electromagnetics.
order FEM. Unfortunately, the proposed method in this paper Jie Liu received the B.S. degree in mathematics and
cannot be extended to solve the resonant cavity problem with applied mathematics and M.S. degree in computa-
tional mathematics from Guizhou Normal Univer-
anisotropic partially conducting media and PEC walls. In the sity, Guiyang, China, in 2008 and 2012, respectively.
future, we would like to develop a new numerical algorithm He is currently pursuing the Ph.D. degree at the
to solve this difficult problem successfully. Institute of Electromagnetics and Acoustics and the
Department of Electronic Science, Xiamen Univer-
The PEC walled resonant cavity problem with complex sity, Xiamen, China.
geometry structure and anisotropic fully conducting media From 2012 to 2015, he was a Lecturer with
may have a continuous spectrum. Continuous spectrum is the School of Mathematics and Statistics, Guizhou
√ University of Finance and Economics, Guiyang. His
(a, b) j , where 0 < a < b and j = −1, but it is very current research interests include the numerical solution of partial differential
difficult to find sup{a} and inf{b} in theory. equations, numerical algebra, and computational electromagnetics.
R EFERENCES Xiaoping Xiong received the B.S. degree in
mechanical design, manufacturing, and automation
[1] G. Strang and G. J. Fix, An Analysis of the Finite Element Method. from the Hefei University of Technology, Hefei,
Englewood Cliffs, NJ, USA: Prentice-Hall, 1973. China, in 2010, and the M.S. degree in electronics
[2] G. D. Smith, Numerical Solution of Partial Differential Equations: Finite and communication engineering from Xiamen Uni-
Difference Methods. New York, NY, USA: Oxford Univ. Press, 1985. versity, Xiamen, China, in 2015.
[3] C. A. Balanis, Advanced Engineering Electromagnetics. New York, NY, Since 2015, she has been with the Department
USA: Wiley, 1989. of Electronic Science, Xiamen University, as an
[4] J. C. Nédélec, “Mixed finite elements in R3 ,” Numer. Math., vol. 35,
Assistant Engineer. Her current research interests
no. 3, pp. 315–341, 1980.
[5] J.-F. Lee, D.-K. Sun, and Z. J. Cendes, “Full-wave analysis of dielec- include planar antenna design.
tric waveguides using tangential vector finite elements,” IEEE Trans.
Microw. Theory Techn., vol. 39, no. 8, pp. 1262–1271, Aug. 1991. Qing Huo Liu (S’88–M’89–SM’94–F’05) received
[6] D. B. Davidson, Computational Electromagnetics for RF and Microwave the B.S. and M.S. degrees in physics from
Engineering. Edinburgh, U.K.: Cambridge Univ. Press, 2010. Xiamen University, Xiamen, China, in 1983 and
[7] S. G. Perepelitsa, R. Dyczij-Edlinger, and J.-F. Lee, “Finite-element 1986, respectively, and the Ph.D. degree in elec-
analysis of arbitrarily shaped cavity resonators using H1 (curl) elements,” trical engineering from the University of Illinois at
IEEE Trans. Magn., vol. 33, no. 2, pp. 1776–1779, Mar. 1997. Urbana–Champaign, Champaign, IL, USA, in 1989.
[8] K. Fumio, “Mixed and penalty formulations for finite element analysis He was with the Electromagnetics Laboratory,
of an eigenvalue problem in electromagnetism,” Comput. Methods Appl. University of Illinois at Urbana–Champaign, as a
Mech. Eng., vol. 64, nos. 1–3, pp. 509–521, Oct. 1987. Research Assistant from 1986 to 1988, and as a Post-
[9] F. Brezzi and M. Fortin, Mixed and Hybrid Finite Element Methods. Doctoral Research Associate from 1989 to 1990.
New York, NY, USA: Springer-Verlag, 1991. He was a Research Scientist and Program Leader
[10] J. Jin, The Finite Element Method in Electromagnetics. New York, NY, with Schlumberger-Doll Research, Ridgefield, CT, USA, from 1990 to 1995.
USA: Wiley, 1993. From 1996 to 1999, he was an Associate Professor with New Mexico State
[11] N. V. Venkatarayalu and J.-F. Lee, “Removal of spurious DC modes University, Las Cruces, NM, USA. Since 1999, he has been with Duke
in edge element solutions for modeling three-dimensional resonators,” University, Durham, NC, USA, where he is currently a Professor of electrical
IEEE Trans. Microw. Theory Techn., vol. 54, no. 7, pp. 3019–3025, and computer engineering. He has authored over 300 papers in refereed
Jul. 2006. journals and 400 papers in conference proceedings. His current research inter-
[12] F. Chatelin, Spectral Approximation of Linear Operators. New York, ests include computational electromagnetics and acoustics, inverse problems,
NY, USA: Academic, 1983. and their application in nanophotonics, geophysics, biomedical imaging, and
[13] W. Jiang, N. Liu, Y. Yue, and Q. H. Liu, “Mixed finite-element method
electronic packaging.
for resonant cavity problem with complex geometric topology and
Dr. Liu is a Fellow of the Acoustical Society of America, a member of Phi
anisotropic lossless media,” IEEE Trans. Magn., vol. 52, no. 2, pp. 1–8,
Kappa Phi and Tau Beta Pi, and a full member of the U.S. National Committee
Feb. 2016.
of the URSI Commissions B and F. Currently, he serves as the Deputy Editor-
[14] N. Liu, L. Tobón, Y. Zhao, Y. Tang, and Q. H. Liu, “Mixed spectral-
in-Chief of Progress in Electromagnetics Research, an Associate Editor of
element method for 3-D Maxwell’s eigenvalue problem,” IEEE Trans.
the IEEE T RANSACTIONS ON G EOSCIENCE AND R EMOTE S ENSING, and
Microw. Theory Techn., vol. 64, no. 2, pp. 317–325, Feb. 2015.
[15] W. C. Chew, Waves and Fields in Inhomogeneous Media. New York, the Editor of the Journal of Computational Acoustics. He also served as a
NY, USA: Van Nostrand, 1990. Guest Editor of the P ROCEEDINGS of the IEEE. Since 2015, he has been the
[16] P. G. Ciarlet, The Finite Element Method for Elliptic Problems. Editor-in-Chief of the IEEE J OURNAL ON M ULTISCALE AND M ULTIPHYSICS
Amsterdam, The Netherland: North Holland, 1978. C OMPUTATIONAL T ECHNIQUES . He was a recipient of the 1996 Presidential
[17] R. Hiptmair, “Finite elements in computational electromagnetism,” Acta Early Career Award for Scientists and Engineers from the White House,
Numer., vol. 11, pp. 237–339, Jan. 2002. the 1996 Early Career Research Award from the Environmental Protection
[18] F. Tisseur and K. Meerbergen, “The quadratic eigenvalue problem,” Agency, and the 1997 CAREER Award from the National Science Foundation.
SIAM Rev., vol. 43, no. 2, pp. 235–286, 2001. He was an elected Fellow of the IEEE for his contributions to computational
[19] Y. Saad, Numerical Methods for Large Eigenvalue Problems. electromagnetics and to subsurface sensing applications, and a Fellow of
Philadelphia, PA, USA: SIAM, 2011. the Acoustical Society of America for his contributions to computational
[20] G. Strang, Linear Algebra and Its Applications, 4th ed. Pacific Grove, acoustics and elasticity. He served as an IEEE Antennas and Propagation
CA, USA: Brooks/Cole, 2006. Society Distinguished Lecturer during 2014–2016.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Abstract— This paper presents a frequency-sweep technique propagating and slowly decaying WG modes at the operating
based on model-order reduction and finite elements, for the frequency [1]–[4], and utilize the mode patterns both as
broadband analysis of structures fed by waveguides (WGs) excitations of the FE model of the device and for expanding
possessing frequency-dependent modal field patterns. Stan-
dard order reduction requires the matrices and right-hand the resulting fields at the ports into WG modes. Two widely
sides (RHSs) to exhibit affine frequency parameterization. This used FE implementations differ in the kind of boundary
precondition is violated when the transverse fields of the condition (BC) imposed at the WG ports and, in consequence,
WG modes vary with frequency. The proposed solution involves the resulting network matrix: the impedance (Z ) formula-
two steps. First, a reduced-order model (ROM) for the WG is tion uses perfect magnetic conductors (PMCs) for nonexcited
constructed. It enables the accurate yet inexpensive computation
of propagation characteristics. Second, order reduction is applied modes and leads to a modal impedance matrix Z, whereas
to the driven problem, wherein the reduced WG model is utilized the scattering (S) formulation, realized by the transfinite
to construct affine approximations to the matrices and RHSs. element (TFE) approach, employs mode-specific transparent
Since this process requires operations on reduced-order matrices BCs [5], [6] and leads to a modal scattering matrix S. The
only, it is computationally cheap and enables offline/online Z formulation is easier to implement but bears the risk of
decomposition. Both impedance and scattering formulations are
considered. For the latter, an alternative to the transfinite element artificial interior resonances [7] (see Section II). In contrast,
method is proposed, which does not employ modal field patterns the S formulation is free of interior resonances and provides a
as shape functions. It avoids interior resonances and computes more efficient way of computing S parameters when structures
scattering parameters more efficiently when only a limited set with large numbers of ports are considered, but only a limited
of excitations is of interest. The resulting algebraic system is of set of inputs is of interest, e.g., clock trees in integrated
somewhat larger dimension but easier to assemble. Its simple
structure greatly facilitates the construction of the ROM. circuits or EM compatibility analysis of high-current paths on
printed circuit boards. In such applications, the responses to
Index Terms— Finite-element (FE) analysis, reduced-basis the relevant excitations suffice, whereas using the Z approach,
methods, reduced-order systems.
the FE solutions for all excitations are needed to obtain S
from Z.
I. I NTRODUCTION Broadband analysis generally requires that the system
response be evaluated at a large number of frequency samples;
T HE subject of this paper is the numerical characterization
of linear time-invariant microwave devices fed by electro-
magnetic (EM) waveguides (WGs). The considered structures
this insures that sharp resonances do not get missed. The
traditional approach is to set up and solve the frequency-
may exhibit complicated shape and inhomogeneous, possibly dependent FE system at each sampling point. Since FE matri-
frequency-dependent, material properties. The analysis is to ces are typically of high dimension, this procedure becomes
be performed at the level of EM fields, by the method of computationally expensive, or even impracticable.
finite elements (FEs) in the frequency ( f ) domain. The general Model-order reduction (MOR) provides highly effective
procedure is to define port cross sections by cutting the WGs tools for reducing the compute times of frequency sweeps.
some distance away from all discontinuities, compute the In the FE context, projection-based MOR [8]–[14] is partic-
ularly suitable because of its capabilities of handling large-
Manuscript received October 18, 2016; revised January 30, 2017; accepted scale systems and preserving the system structure. One critical
February 16, 2017. This work was supported by the Deutsche Forschungsge-
meinschaft under Grant DY 112/1-2. limitation of projection-based MOR is that the parameteriza-
R. Baltes and R. Dyczij-Edlinger are with the Chair of Electromag- tion of the FE matrices must be affine. This holds true when
netic Theory, Saarland University, D-66123 Saarbrücken, Germany (e-mail: the transverse modal field patterns of the feeding WGs are
r.baltes@lte.uni-saarland.de).
A. Schultschik and O. Farle were with the Chair of Electromagnetic Theory, frequency-independent, i.e., for WGs of homogeneous mate-
Saarland University, D-66123 Saarbrücken, Germany. They are now with rial, which support transverse electric, transverse magnetic,
CST-Computer Simulation Technology AG, D-64289 Darmstadt, Germany. and—in case of multiple electrodes—transverse EM (TEM)
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. modes. Moreover, the propagation coefficients of these types
Digital Object Identifier 10.1109/TMTT.2017.2679181 of modes, which are required as scaling factors [10], are easily
0018-9480 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
A. FE Calculation of Modal Impedance Matrix Z The use of the modal basis on p reduces the number
The entry Z ik is computed by exciting the structure by a of DOFs in (24) but greatly complicates MOR when the
single modal current i k of unit amplitude, terminating all other WG mode patterns become frequency-dependent. Then, all
modes by open circuits (OCs), and measuring the resulting submatrices in (24) that involve P will exhibit nonaffine
modal voltages u i via (7). Thus, the BC (12) takes the form parameter dependence.
2) Proposed Formulation: To facilitate MOR, the goal is to
(νr curl E) × n̂ = − j k0 η0 ĥk (k0 ) on P (19) isolate the effects of frequency-dependent WG mode patterns
and the weak form (14) together with the output Z ik (7) yield within the system matrix and simplify matrix assembly when
k0 is varying. The key idea is to employ conventional FE basis
a(E, w) = j k0 η0 b(ĥk (k0 ), w) ∀w ∈ Hcurl
E functions and solve for the total field on P . For this purpose,
(20)
Z ik = b(ĥi (k0 ), E). the BC (22) is reformulated as
⎛ ⎞
The discrete formulation is obtained by considering a finite
number of modes M and restricting (20) to an N-dimensional (νr curl E) × n̂ = − j k0η0 ⎝2ĥk − (δ j k + b j )ĥ j ⎠
FE space V ⊂ HcurlE . The resulting FE system reads j
⎛ ⎞
Sν + j k0η0 Tσ − k02 Tε x E = j k0η0 bk (k0 )
(21) = − j k0η0 ⎝2ĥk − u j ĥ j ⎠
Z ik = biT (k0 )x E j
⎛ ⎞
with x E , bi , bk ∈ C N and Sν , Tσ , Tν ∈ C N×N ; the
entries of the system matrices and vectors are given by (1) = − j k0η0 ⎝2ĥk − b(ĥ j , E)ĥ j ⎠ . (26)
.
and (16)–(18). FE formulations for immittance matrices have j
been in use for a long time [22]–[24]. Their advantages The corresponding weak form and output are given by
include simplicity and that the system matrix is real valued ⎧
in the lossless case. A potential disadvantage is that Z will ⎪
⎪ a(E, w) + j k η b b( ĥ , E) ĥ , w
⎨ 0 0 j j
become unbounded when the structure behaves as an OC.
Moreover, the formulation may give rise to artificial interior ⎪ = 2 j k0η0 b(ĥk , w) ∀w ∈ Hcurl (27)
⎪
⎩ S = b(ĥ , E) − δ
E
resonances [7]. ik i ik
Moreover, assembling (28) or (32) is easier, because the origi- A = AT and c = b, the preferred choice is one-sided
nal FE matrix R remains unchanged. Since FE basis functions projection [8], with
are used everywhere, standard FE postprocessing tools are
immediately applicable. For the same reason, the number V = U. (41)
of DOFs is somewhat larger than in (24). Another, subtle,
By (34), also the EVP (9) exhibits affine k0 -dependence
difference emerges from the way the BC (22) is implemented.
and is thus suitable for MOR [25]–[27]. In [27], even the
For higher order WG modes that are truncated from the series
projection matrix W was affinely parameterized, to incorporate
expansion, i.e., for j > M, the port cross section P presents
a constraint. Thus, the general framework is as follows. Let
a perfect electric conductor (PEC) in (24), whereas it behaves
Ñ E N E denote the ROM dimension. Given scalar functions
as a PMC in the present approach, as can be seen from (26).
A numerical comparison of the formulations (21), (24), (28), wq (k0 ) and constant matrices Wq ∈ C N E × Ñ E such that
and (32) is given in Section VIII.
W(k0 ) = wq (k0 )Wq (42)
v(k0 ) ≈ W(k0 )ṽ(k0 ) (43)
VI. P ROJECTION -BASED MOR
The parameter dependence of vectors t(k0 ) ∈ C N or the ROM resulting from the EVP (9) reads
matrices M(k0 ) ∈ C N×N is said to be affine, if they decom-
pose into finite sums nl (k0 )Ãl ṽ = λ̃ m l (k0 )C̃l ṽ (44)
Nt A
wherein the ROM matrices M̃ ∈ {Ãl , C̃l } are given by
t(k0 ) = τi (k0 )ti , Nt A N (33)
i=1 M̃(k0 ) = W∗ (k0 )MW(k0 )
N
wq∗ (k0 )wr (k0 )Wq∗ MWr .
MA
M(k0 ) = μ j (k0 )M j , NM A N (34) = (45)
j =1
Various approaches to constructing the projection matrices
wherein τ j (k0 ) and μi (k0 ) are scalar functions of k0 , whereas are available from the literature [8], [9], [11], [12]. Adaptive
the vectors ti and matrices M j do not depend on k0 . methods employ error indicators that are commonly based on
Given the affinely parameterized linear system the norm of the ROM residual r(k0 ), defined by
f (k )A x = g j (k0 )b j u AUx̃ − b, for driven problems
i0 i (35) r= (46)
y= h k (k0 )ckT x + d(k0 )u (A − λ̃C)Wṽ, for eigenvalue problems.
with input u, output y, scalar functions f i , g j , h k , and constant As will be exemplified in Section VII-B3, computing r 2 is
matrices Ai ∈ C N×N and vectors b j , ck ∈ C N . The general inexpensive, because it requires operations on ROM quantities
approach of projection-based MOR is to construct suitable trial only. Implementation issues are discussed in [28] and [29].
and testing matrices U, V ∈ R N× Ñ , with Ñ N, which are
taken to be orthogonal for numerical stability, and to seek an
VII. FAST F REQUENCY S WEEPS FOR D RIVEN S YSTEMS
approximate solution in the column space of U
A. Frequency-Independent WG Mode Patterns
x ≈ Ux̃, with x̃ ∈ C Ñ (36)
The modes supported by WGs of homogeneous material
by projecting the first equation of (35) on the column space properties possess k0 -independent transverse fields, and their
of V. The ROM reads dispersion characteristics are fully characterized by the respec-
tive cutoff wavenumber kc,i [19, p. 171]. It thus suffices to
f (k )Ã x̃ = g j (k0 )b̃ j u
i0 i (37) compute the WG modes at an arbitrary reference freespace
ỹ = h k (k0 )c̃kT x̃ + d(k0 )u wavenumber k R = kc,i . As suggested in [10], normalization
in (2) may be preserved by scaling factors di (k0 ) of the
wherein reduced-order quantities are denoted by tilde. Thus
form
Ãi = VT Ai U (38) ⎧ 2 2
⎪
⎪ k R kc,i −k0
b̃ j = V b j T
(39) ⎪
⎪ , for TE modes
⎪
⎪
⎪
⎪ k k 2 −k 2
c̃k = UT ck . (40) ⎨ 0 c,i R
di (k0 ) =
k0 kc,i
2 −k 2 (47)
The method comprises an offline stage for compu- ⎪
⎪ R
⎪
⎪
, for TM modes
ting (38)–(40) once, and a fast online stage for solving the ⎪
⎪
2 −k 2
k R kc,i
⎪
⎪
0
ROM (37) for arbitrary values of k0 . Thanks to the affine ⎩
1, for TEM modes.
structure of (35), the procedure is online efficient in the
sense that assembling and solving (37) for given k0 requires The corresponding scaling matrix D(k0 ) ∈ C M×M is given by
operations on quantities of reduced order only. Since the
systems (21), (24), and (28) are all complex symmetric, with D(k0 ) = diag(d1 (k0 ), . . . , d M (k0 )). (48)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
In the following, subscript (·) R denotes values at k R . One FE discretization of and solve the EVP (9). Depending on
obtains the WG formulation chosen in (9), the linear mapping from
vi to (êi , ĥi ) may exhibit affine dependence on k0 and/or λi .
ĥi (k0 ) = di (k0 )ĥi (k R ) (49)
Thus, by (1), (20), and (21), the excitation vector bi takes the
bi (k0 ) = di (k0 )bi (k R ) = di (k0 )b R,i (50) general form
B(k0 ) = B(k R )D(k0 ) = B R D(k0 ) (51)
bi (k0 ) = α p (k0 , λi (k0 ))Q p vi (k0 ) (58)
P(k0 ) = P(k R ) D−1 (k0 ) = P R D−1 (k0 ). (52) p
1) Impedance Formulation: Thanks to (50), the Z for- with scalar functions α p and constant matrices Q p ∈ C N×N E ,
mulation (21) exhibits affine k0 -dependence and is therefore the specifics of which depend on the WG formulation
accessible to projection-based MOR. The resulting ROM reads in (9). Similar equations hold for the basis functions pi
of Section V-B1.
S̃ν + j k0η0 T̃σ − k02 T̃ε x̃ E = j k0 η0 dk (k0 )b̃ R,k Assembling and solving the FE system (9) at each sampling
(53)
Z̃ ik = di (k0 )b̃TR,i x̃ E point are too slow for fast frequency sweeps. We suggest to
replace the eigenpair (λi , vi ) in (58) by the solution (λ̃i , ṽi )
wherein the vectors and matrices are defined by (38), (40), of the ROM (44), which is cheap to compute. Plugging (43)
and (41). into (58) yields an affine approximation b̂i ≈ bi
2) TFE Method [5]: Applying (52) to (24) leads to the
rescaled TFE formulation b̂i (k0 ) = α p (k0 , λ̃i (k0 ))wq (k0 )Q p Wq ṽi (k0 ). (59)
p q
R I,I R I, p P R 0 0 xI
+ j k η
P TR R p,I P TR R p, p P R 0 0
0 D2 D−1 b 1) Impedance Formulation [18]: Since the k0 -dependent
WG mode patterns do not affect the system matrix of (21),
R I, p P R 0
=− − j k0η0 2 D−1 ak . (54) it suffices to build a WG-ROM after (44), substitute (59) for
P TR R p, p P R D
the RHS vectors of (21), and apply the MOR of (37) to (21),
Since R is affinely parameterized by (25), and D is of small using a projection matrix U and (41). The resulting ROM reads
dimension independent of N, (54) is suitable for projection-
based MOR, and the resulting ROM is online efficient. Since S̃ν + j k0η0 T̃σ − k02 T̃ε x̃ E = j k0η0 b̃i (k0 )
(60)
its structure is the same as (54), the formula is omitted. Z̃ ik = b̃kT x̃ E
3) Proposed S Formulation: Plugging (51) into (32) yields
⎧ with S̃ν , T̃σ , T̃ε ∈ C Ñ × Ñ according to (38) and (41). By (39)
⎪
⎨ R k0 η0 B R D x E 2 j k0η0 dk b R,k and (59), the reduced vectors b̃ = UT b̂ take the form
=
k0 η0 DBTR j k0η0 I x̂ 0 (55)
⎪
⎩ b̃i (k0 ) = α p (k0 , λ̃i (k0 ))wq (k0 )Q̃ p,q ṽi (k0 ) (61)
Sik = di b R,i x E − δik .
T p q
For the same reasons as in Section VII-A2, MOR is applicable. with constant matrices Q̃ p,q ∈ C Ñ × Ñ E given by
Since ROM matrices according to (38) are fully populated,
the simpler structure of (28) is preferable over (32). Starting Q̃ p,q = UT Q p Wq . (62)
from a projection matrix U for x E , the augmented matrices The WG-ROM matrices Ãl and C̃l of (44), the driven ROM
I matrices S̃ν , T̃σ , and T̃ε of (60), as well as excitation
Va = Ua = U (56) matrices Q̃ p,q of (62) are computed upfront, during the offline
j DBTR
stage of the method. The online stage serves to evaluate
map (55) to the sought ROM Z̃ ik (k0 ). For every k0 , this involves assembling and solving the
⎧
⎪ WG-ROM (44), building b̃i from the resulting (λ̃i , ṽi ) and,
⎨ S̃ν + j k0η0 T̃σ − k0 T̃ε
2
finally, assembling and solving the driven ROM (60). Thanks
+ j k0η0 B̃ R D2 (k0 )B̃TR x̃ E = 2 j k0 η0 dk (k0 )b̃ R,k (57)
⎪
⎩ to the affine parameterizaton present in (44) and (61),
S̃ik = di (k0 )b̃TR,i x̃ E − δik the online stage requires operations on ROM matrices only
wherein the vectors and the matrices are defined by (38), (39), and is therefore computationally cheap.
and (41), using U. Since B̃ R = UT B R ∈ C Ñ ×M is constant 2) Scattering Formulation: The main challenge with the
and of small dimension independent of N, the critical matrix S approach is that, in contrast to immittance formulations (60),
B̃ R D2 (k0 )B̃TR is computed effiently during the online phase. frequency-dependent WG mode patterns enter not only the
RHS but also the system matrices.
As for the original TFE method [5], Section V-B1, it is
B. Frequency-Dependent WG Mode Patterns certainly possible to construct affine approximations p̃i ≈ pi
If the WG feeding contains inhomogeneous materi- to the columns of the mode matrix P(k0 ) of (23) from the
als, its transverse modal field patterns will generally exhibit WG-ROM, similar to (59). Let the linear mapping vi → pi ,
nonaffine frequency dependence, and analytical formulas for reflecting changes in numbering and formulation between the
the propagation coefficients will not be available. To obtain WG and driven case, be given by
numerical approximations to the eigenpairs (vi , γi ), the usual
procedure is to extract the surface mesh on P from the pi (k0 ) = ζ p (k0 , λi )N p vi (k0 ) (63)
p
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
with scalar functions n p and constant matrices N p ∈ C N×N E . matrices S̃ν , T̃σ and T̃ε in (69) are fully populated already, the
Then, by (42) and (43), p̃i (k0 ) takes the form occurence of B̃B̃T does not lead to a significant increase of the
numerical costs for solving (69). The derivation of (69) shows
p̃i (k0 ) = ζ p (k0 , λi )wq (k0 )N p Wq ṽi (k0 ). (64) that the ROM approximation error in the excitations b̃ and
p q
driven solutions x̃ E may be controlled individually, by adjust-
However, (24) reveals that the resulting approximation P̃ ≈ P ing the column spaces of W and U, respectively.
needs to be plugged into system matrices of the form RP 3) Error Indicator: The MOR framework presented in this
and P T RP, wherein R contains terms that are constant, linear, paper does not prescribe a specific way of constructing the
and quadratic in k0 [see (25)]. In consequence, the number of projection matrix U. If an iterative method for ROM generation
coefficient matrices in the approximate system will become is used, some error indicator is required for use as termination
very large. This will not only affect the offline runtime for criterion. In adaptive multipoint methods, such error indicator
construcing the corresponding reduced-order matrices, but also is also utilized for guiding the placement of expansion fre-
the online runtime for scaling all of them by their respec- quencies. One popular choice is to evaluate the residual of
tive k0 -dependences and assembling them to a single system the ROM solution with respect to the full-order system on a
matrix, for each considered value of k0 . Since the approach of dense sampling of the frequency interval under investigation.
Section V-B2 will lead to a more efficient solution, no further The key point is that the 2-norm of the residual is computable
details are given. at low cost from ROM quantities only, without referring to the
In the S formulation proposed in Section V-B2, as real- underlying full-order system.
ized by (28) and (32), only the vectors bi and thus the In the following, an error indicator for the S formulation
matrix B = [b1 , . . . , b N ] are affected by frequency-dependent with frequency-dependent WG mode patterns will be devel-
WG mode patterns. This allows for an MOR technique similar oped. It does not account for the approximation error in B̂,
to Section VII-A3. According to (61), the WG-ROM yields i.e., the WG-ROM is assumed to be exact. Error indicators for
affine approximations b̂m to the excitation vectors and, in con- the WG-ROM are found in [30]; the Z formulation is treated
sequence, an approximate excitation matrix B̂ ≈ B, with in [18]. For clarity, the calculation of the relative residual
B̂(k0 ) = [b̂1 (k0 ), . . . , b̂ M (k0 )]. (65) will be presented in a way that is known to saturate [31] at
approximately the square root of the machine epsilon, which
Substituting b̂ for b and B̂ for B leads to affine representations the authors deem good enough for typical applications. Still,
for the FE system (28) the following equation (72) allow for more robust numerics,
along the lines of [29].
(R + j k0 η0 B̂B̂T )x E = 2 j k0η0 b̂k
(66) By (46) and (66), the residual r of x̃ E is given by
Sik = b̂iT x E − δik
r(k0 ) = Sν + j k0 η0 Tσ − k02 Tε + j k0η0 B̂(k0 )B̂T (k0 ) Ux̃ E
and for the augmented FE system (32), respectively,
⎧ − j 2k0η0 b̂k (k0 ). (71)
⎪
⎨ R k0 η0 B̂ xE 2 j k0η0 b̂k
= Let ¯ denote the complex conjugate and ∗ the complex conju-
k0 η0 B̂T j k0η0 I x̂ 0 (67) gate transpose. Then, the relative residual norm ρ of r reads
⎪
⎩
Sik = b̂i x E − δik .
T
1 ||r(k0 )||2 1 r∗ r(k0 )
As in Section VII-A3, the projection matrix U is constructed ρ(k0 ) = = . (72)
2k0 η0 ||b̂(k0 )||2 2k0 η0 b̂∗ b̂(k0 )
from solutions x E (k0 ) of the augmented FE system (32). The
augmented projection matrices Va and Ua defined by By (59), we have
Va = Ua =
I
U (68) b̂∗ b̂ = α m wn αk wq ṽ∗ Wn∗ Q∗m Qk Wq ṽ. (73)
j B̂T m n k q
map (67) to an ROM of (66) which takes the form The square matrices (Wn∗ Q∗m Qk Wq ) are of the dimension of
⎧ the WG-ROM Ñ E . Similarly, r∗ r may be expanded term by
⎪
⎨ S̃ν + j k0 η0 T̃σ − k0 T̃ε
2
term via (59), (65), and (71). This lengthy, but elementary,
+ j k0 η0 B̃(k0 )B̃T (k0 ) x̃ E = j 2k0η0 b̃k (k0 ) (69) procedure reveals that, again, all the matrices involved are
⎪
⎩
S̃ik = b̃iT (k0 )x̃ E − δik . of reduced order. For paper length constraints, the results are
omitted.
Herein, S̃ν , T̃σ , T̃ε , and x̃ E are defined by (38) and (41), 4) Computer Implementation: We employ multipoint
using U, the vectors b̃i and b̃k are given by (61), and MOR methods [28], [30] with adaptive expansion point place-
B̃(k0 ) = [b̃1 (k0 ), . . . , b̃m (k0 ), . . . , b̃ M (k0 )]. (70) ment for both the WG modes [32] and the driven case. The
WG eigenmode problem is stated in terms of a magnetic
A few comments are in order. As in the case of the Z vector potential and an electric scalar potential (see [4] for
formulation, the structure of (69) allows for online/offline details). The corresponding WG-ROM is constructed by the
decomposition. The fact that the b̃m values are in terms MOR method of [30]. The FE system for the driven case
of ROM quantities only, and M N, enables the effi- is based on the augmented form (32) of the proposed S
cient computation of B̃ during the online phase. Since the formulation.
Another random document with
no related content on Scribd:
Long Island Sound Line, 40
Longitudinal system of ship construction, 268, 348
Louis Philippe of France, escape of, 113
Louvre Museum, Kirkaldy’s designs in, 246
Lund, Mr. W., 297
Lungley, Mr., ship built by, 264
Lyttleton, Wm., 207
McDougall, Capt., 55
McGregor, Mr. John, and early Chinese paddle-wheels, 4
McGregor, Laird & Co., 196
MacIver, Mr. David, and Mr. S. Cunard, 150
McKean, McLarty, and Lament, 254
Mackenzie, William, master of the Comet, 63
McKinnon & Co., Glasgow, 181
MacLachlan, Archibald, 66
McQueen, Robert, 39, 52
Mahmoudieh Canal, 179
Mails, officer in charge of, to West Indies, 190
Mails to America, 149; to India, 176, 177; to Ireland, 102
Makaroff, Vice-Admiral, 367
Malcomson’s London and Dublin Line, 99
Malta floating dock, 363
Man, Isle of, first steamers at, 88; first built there, 89; history of the Manx
service, 87-94; Barrow service, 96
Manby, Mr., 195
Maples and Morris, Messrs., 106, 107
Mare’s Shipyard, Blackwall. See Ditchburn and Mare
Marine engines. See Engines
Marinsky Canal, 364
Maryland Steel Co., Baltimore, floating docks, 358, 362
Maschinenbau-Gesellschaft, 303
Mason, Mr., 262
Masts on steam-ships, 41; used as funnels, 212, 218; tripod, 332; on
warships, 338
Mastless steamers, 346
Maudslay, Sons & Field, founder of the firm, 70; engines by, 148, 201, 202,
233, 253, 319, 372; connection with the Royal Navy, 311; yards of, 234
Maury, Lieut., 241
Mediterranean ports, 111
Mediterranean service, 267
Merchants’ Shipping Co., 47
Mersey ferries, 366
Messageries Maritimes de France, 267
Mexican Government and iron frigate, 316
Mexican War, 174
Middleton Yard, Hartlepool, 305
Midland Railway Co.’s steamers, 121
Milford-Rosslare service, 116
Millard and Kirby, Messrs., and Fulton’s Clermont, 50
Miller and Ravenhill, engines by, 187, 372
Miller, Patrick, 57, 58, 388
Mills, Mr. Edward, 154
Mississippi River steamers, 53; Fitch’s steamboat, 23; Fulton’s steamboat,
32; Moselle and Oroonoko blown up, 53; intentional collisions, 53
Mitchell, Mr. Charles, 212
Monaco, Prince of, yacht of, 373
“Monitors,” 334
Monroe, President, 123
Moore, Admiral Sir Grayham, 217
Moray, John, on James Rumsay as inventor of steamboats, 19
Morey, Samuel, invents a steamboat, 24
Morgan Combine, 228
Morisot’s “Orbis Maritimi,” 6
Morland, Sir S., 10
Motor-boats (hydroplane), 385; (hydrocurve), 385
Napier, Admiral Sir Charles, 195
Napier, David, and the boiler of the Comet, 63; and the shape of bows of
steamers, 71; provides engines, 72
Napier, Robert & Sons, engines by, 72, 88, 89, 147, 151, 157; and Mr. S.
Cunard, 149; present engine of the Comet to South Kensington Museum,
64; and David Kirkaldy, 243; and high-pressure boilers of steel, 279
Napoleon III., yacht of, 373
National Line, 254
Naval Construction Co., Barrow, 99
Navy, Royal, steam-power and the, 311-340; last wooden battleship, 319;
first twin-screw boat, 328; ironclads without masts, 333; torpedo boats,
336; destroyers, 336; development, 336
Neilson, Walter N., 229
New England Ocean Steamship Co., 155
New York celebrates the arrival of early steamers, 141
New York-Aspinwall mails, 188
New York-Bremen service, 154
New York-Chagres line, 188
New York-Havana service, 189
New York-Liverpool, lines in 1850, 155
New York and Havre Steam Navigation Co., 154
New York Shipbuilding Co., Camden, N.J., 51
New Zealand Government subsidy, 185; service to, 298
New Zealand Shipping Co., 310
Newcomen and Savery, 11
Newfoundland Government and mails, 162
Newhaven-Dieppe service, 106
Newport News Shipbuilding, &c., Co., 340
Niger exploration, 280
Norddeutsche Werft, 303
Norddeutscher Lloyd, 267, 302, 303-305
Normand, A, Havre, 373
North Lancashire Railways, 102, 103
North Sea, 84
Northumberland Straits passenger service, 370
Oak, scarcity of, and use of iron for ships, 195
Ocean liner, express, modern type of, 252
Ocean Steam Navigation Co., 154
Oceanic Steam Navigation Co., 252
Ogden, Mr., American Consul, 219
Oil-tank steamers, 348, 351
Oldham’s revolving bars, 195
Orient Line, 264, 291, 294-296
Orient-Pacific Line, 295
Orient Royal Line, 295
Original Steam Packet Co., 72
Ostend-Dover service, 309
Oude, Rajah of, generosity of, 165
Ouseburn engine works, 306
Overcrowding passenger steamers, 79
“Overland Route” to India. See Suez
Pacific coast of S. America trade, 187
Pacific and Australasian Co., 239
Pacific Mail Line, 188
Pacific Steam Navigation Co., 186, 187, 189, 191, 229, 263, 291, 294, 295
Paddle-wheels, evolution of, 1; motive-power, 1; animal-driven, 2; early
forms, 2, 4; early experiments, 10, 12; Jouffroy’s invention, 17; Morey’s
inventions, 24; Roosevelt’s invention, 25; Patrick Miller’s invention, 58;
vertical, 25; disconnecting, 33; Seward’s invention, 110; development in
construction, 197-199; duck-foot paddles, 207; elliptical, 208; horizontal
centrifugal, 208; superseded by screw, 191
Paddle v. screw races, 259; tests, 312
Paddle-boxes as lifeboats, 78
Palmer, Sir Charles, 214
Palmer Bros. & Co., ships built by, 114, 213, 247, 248; and rolled armour
plates, 385
Panama-Astoria service, 189
Panama-San Francisco mails, 188, 189
Panama, New Zealand, and Australian Royal Mail Co., 185
Panama Railway, 174, 187, 191, 262
Panama route, 187
Panciroli’s “Rerum memorabilium,” 6
Papal yacht, 372
Papin, Dr. Dennis, inventions of, 11
Paris Exhibition, 1878, traffic, 109
Parsee custom at launch, 202
Parsons, Hon. A. C., on turbines, 307
Parsons turbines, 118, 307, 338
Passengers carried by Sirius across Atlantic, 141; first steamer for
passengers and cargo, 72
Patersen, Capt. Robert, 86
Paterson of Bristol, 141, 221
Paul, Capt. Fred, R.N., 113
Paulding, James Kirke, 339
Peacock, Capt. George, and mechanical swan yacht, 383
Pearse & Co., Stockton-on-Tees, 205
Penarth floating dock, 359
Peninsular and Oriental (P. & O.) Co., incorporated, 178; first steamer to
India, 179; transport over Suez isthmus, 179; services to India and China,
180; subsidy for Indian mails, 180; Australian service, 180; difficulties on
opening of Suez Canal, 182; overland route through Egypt closed, 182;
ships, 260-261; increase of size of ships, 291, 293; and Australian trade,
294; acquires Blue Anchor Line, 297
Peninsular Steam Navigation Co., 176-178; becomes the P. & O. Co., 178
Penn, John, and Son, engines by, 226, 233, 260; oscillating engines, 201,
314; number of engines fitted by, 315; for the Crimean War, 319; and screw
bearings, 219
Périer’s fire pump, 16
Perkins’ tri-compound engines, 306
Peru, 189
Petroleum steamers, 351
Philadelphia Line, 43
Philippines, floating dock for, 362
Phillips, Sir Richard, 69
Pirrie, Lord, 298
Porter’s patent anchor, 223
Portsmouth-Ryde, 232
Potomac, early steamboats on the, 20
Powell (H. & Co.) Line, 99
Propeller, screw. See Screws
Propelling vessels by recoil from cannon, 8; by animals, 2; by steam, early
experiments, 10-11; by pumping water, 12; by screws, 29. See also
Paddle-wheels
Propelling without paddles, reward for, 210
Pyroscaphe, the, 15
Quebec and Halifax Steam Navigation Co., 134
Racing, Ocean, 247; steamboat, 53; paddle v. screw, 259
Railway companies and their steamships, 102-121
Railway trains, ferrying of, 363-366
Ramage and Ferguson, Ltd., Leith, 375
Rams, 329
Ramsay’s (David) patent boats (1618), 6
Ramus, Rev. C. M., and hydroplane, 386
Randolph, Charles, 229
Randolph, Elder & Co., 229
Rangoon wooden dock, 354
Rate wars, 74, 80, 94
Rateau turbines, 307
Red Cross Line, 231
Red Sea steamer service, 166; to the Mediterranean transport, 179
Red Star Line, 256
Refrigerators, 298
Registration of steamers, 77
Reid, Mr. E. J., designs Koenig Wilhelm, 333
Reid’s U bow, 332
Reiherstieg yard, Hamburg, 302, 303
Rennie, Capt. George, 183
Rennie’s “Aberdeen” Line, 183
Rennie, G. & J., engines by, 233, 313; and Ship Propeller Co., 216; floating
docks, 355, 363
Renwick, Dr. James, 29
Repairs to steam-ships, 300
Reversing machinery, 70
Richardson Bros. & Co., 238, 239
Rivalry between steam-ship companies, 73
Roberts, Lieut., R.N., 138, 145
Robertson, John, 62
Robertson, Robert, engineer, 63
Robinson and Russell, 232, 319
Roebuck, Dr., 86
Rogers, Capt., of the Savannah, 125
Rogers, Moses, pioneer steam navigator, 30, 123
Roosevelt, Nicholas J., invents paddle-boat, 25; associated with Fulton, 42;
experiments in steam propulsion, 208
Ropner & Sons, Ltd., 348
Rostock “Neptun” yard, 302
Rotterdam, railway round, 117
Rouss, Mr. W. P., yacht of, 374
Royal Academy, steam-ship designs exhibited at, 245
Royal Mail Steam Packet Co., 185, 189-191, 262-263, 291, 295, 299-300
Royal Netherlands Steamship Co., 91
Royal yachts, 371-374
Rubic and Blaker, Northam, 110
Rudders, bow, 106; balanced for turbine vessels, 105; submerged, 290
Rumsay, James, as the inventor of the steamboat, 19
Rumsay Society, 21
Rupert, Prince Palatine, and boat propeller, 11
Rushen, Mr. P. C., on Jonathan Hulls’ invention, 14
Russell & Co., clipper built by, 173
Russell, Robinson & Co., 107
Russell, Mr. Scott, and the Wave Queen, 107; and wave-line construction,
236, 316, 320; shipbuilding on the Thames, 204, 234; designs Victoria,
263; and the Great Eastern, 268, 278
Russian Government ice-breaker, 367; Navy floating dock, 363; royal yachts,
371, 373
Ruthven’s hydraulic propulsion, 208, 321-325
“Sag,” 46, 194, 268
Sail power on liners, 158
Sailing clippers, American, fast passages of, 153
Sailing vessels, engines put into, 135, 136; vessel with steam as auxiliary
crosses Atlantic, 122; steam auxiliary to, 164-192
St. George Steam Packet Co., 72, 94, 97, 100, 101
St. Lawrence River ice-breaker, 369
Saloons above deck first fitted, 206; oscillating, 253
Samuda Bros., 204, 234
San Francisco Union Iron Works, 340
Saône, paddle-steamer on the (1783), 17
Sassnitz-Trelleborg railway ferry, 365
Sault Ste. Marie Canal, 52
Savery, Thomas, invention of, 11
Scarborough and Isaacs, Messrs., 122
Schlick balancing of engines, 120
Schultz turbines, 388
Scott, Capt., of Rising Star, 131
Scott, Mr. John, figure-head of, 318
Scott, Russell & Co., Millwall, 204
Scott, Sinclair & Co., Greenock, 318
Screw propellers, invention of 29; first Manx steamer to use, 92; for sea-
going steamers, 97; supersede paddle-wheels, 191; tried in 1802, 192;
earliest attempts to apply, 206, 207; movement of vessels with single
screw, 209; twin-screws, 210; first ocean steamer with twin-screws, 265;
fantastic forms, 215; first sea-going vessel with screw, 216; definitely
adopted, 219; lifting propeller, 253; for long voyages, 256; adopted for mail
boats, 262; multiple screws, 310; first vessel in the Royal Navy with, 313;
removable screws, 318; twin screws, 325; tests of twin screws, 326
Sea-sickness, steamers to prevent, 253, 377-379
Sea voyage, first British steamer to make a, 64
Seamen, pay of, in 1821, 132
Seaward and Capel, Limehouse, 169
Seaward, J., & Co., Millwall, 373
Seaward’s vibrating paddles, 110
Seine, first iron steamer on the, 195
Sewall & Co., 194
Sewell and Faron, 158
Shaw, Savill & Albion Co., 297
Shelter deck, 344
Ship Propeller Co., 216
Shipbuilding, German competition, 302. See also Thames
Ships named:
Aaron Manby, 195
Aberdeen, 296, 307
Achilles, 315
Aconcagua, 264
Active, 311
Ada, 116
Adelaide, 269
Adirondack, 48, 170
Admiral Moorsom, 119
Adriatic, 161, 163, 253, 289
Aetna, 35
Africa, 153, 155
African, 176
Agamemnon, 315
Aguila, 112
Ajax, 315
Alabama, 175
Alaska, 172, 250
Alberta, 116
Alecto, 312
Alexandra (L. & S.W.R.), 116
Alexandra (L. & N.W.R.), 119
Alexandra (Royal Yacht), 371
Alice, 115
Alida, 49
Alliance, 113, 114
Alma, 114, 116
Amazon, 300
America (Cunard Co.), 152, 245, 286
America (National Line), 254
America (Yacht), 158
American Turtle, 376
Amerika, 305
Amethyst, 309, 335
Anglia, 104, 120
Anglo-Saxon, 255
Annette, 173
Antarctic, 157
Antelope, 235
Antrim, 121
Apollo, 110
Aquila, 107
Arabia, 153
Arago, 154
Aragon, 300
Araguaya, 300
Arcadia, 151
Archimedes, 216, 222
Arctic, 157-160
Argyle, 66
Ariadne, 110, 316
Arizona, 249
Arkansas, 340
Armenia, 51
Arrogant, H.M.S., 314
Arundel, 109
Asia, 153, 157
Assiniboia, 301
Assyrian, 316
Astarte, 255
Asturias, 300
Atalanta, 116, 110, 111, 166
Athenia, 255
Athole, 206
Atlantic, 156, 158
Atrato, 271
Augusta, 99
Aurania, 281
Aurora, 327
Austral, 295
Australasian, 296
Avoca, 99
Avon, 222, 300
Ayrshire Lassie, 106
Balmoral Castle, 292
Baltic, 157, 158, 181, 287, 288
Bann, 319
Banshee, 119
Barbarossa, 304
Baron Osy, 269
Barracouta, 175
Basilisk, 313
Bay State, 47
Belfast, 72
Belgic, 253
Bélier, 334
Bellerophon, 80, 315, 334
Ben-my-Chree, 89, 92, 93
Berenice, 166
Bertha, 116
Bessemer, 253, 379
Birkenhead, 317
Black Eagle, 314
Black Prince, 315
Bogota, 229
Borussia, 267, 305
Bremen, 267
Brighton, 109, 112
Bristol, 47, 337
Britannia, 151, 154
Britannic, 253
British Queen, 138, 145-147, 169, 216
Brittany, 109, 114, 115, 116
Brune, 319
Buenos Ayrean, 281
Buffalo, 35
C. Vanderbilt, 49
C. W. Morse, 48
Calais, 105
Calais-Douvres, 378
Caledonia, 64, 151
California, 188
Callao, 229
Caloric, 384
Calvados, 109
Cambria, 104
Cambria (Cunard Co.), 151
Cambria (L. & N.W.R.), 120
Camden, 35
Camilla, 110
Campania, 282, 287
Canada, 152, 245
Canadian, 254
Cape of Good Hope, 181
Captain, 334
Car of Neptune, 35, 36, 38, 44
Carbon, 235
Carmania, 282, 285, 309
Caronia, 282
Carpathia, 283
Carron, 79, 86, 176
Cassandra, 255
Castalia, 377
Cedric, 288
Celtic, 253, 288
Cerberus, 335, 376
Chancellor Livingston, 35, 42, 43
Charles Wetmore, 55
Charleston, 340
Charlotte Dundas, 28, 59, 135, 199
Cherbourg, 116
Chicago, 248
Chili, 187
Chimborazo, 264, 295
China, 246, 247, 293
Cincinnati, 305
City of Baltimore, 239
City of Belfast, 121
City of Berlin, 241, 242
City of Bristol, 242
City of Brussels, 241
City of Chicago, 243
City of Cleveland, 54
City of Dublin, 72
City of Edinburgh, 81
City of Glasgow, 96, 237
City of Limerick, 97
City of Manchester, 238
City of New York, 240, 256, 290
City of Paris, 241, 246, 256, 290
City of Philadelphia, 239
City of Pittsburg, 239
City of Rome, 242
City of Washington, 239
Claremont, 307
Clermont, 20, 29 et seq., 49, 135
Cleveland, 305
Clyde, 64
Cobra, 308
Coffee Mill, 66
Collier, 107
Collingwood, 55
Colombia, 176
Colombo, 181
Colorado, 248
Columbia, 114, 116, 151
Columbus, 288
Comet (Bell’s), 62, 135
Comet (Dawson’s), 70
Comet (French Co.), 112
Commerce, 73
Commonwealth, 48
Conde de Patmella, 122
Confiance, 176
Connector, 379
Connemara, 120
Coogee, 96
Copenhagen, 118
Cotopaxi, 295
Countess of Dublin, 98
Countess of Erne, 119
Countess of Strathmore, 214
Courier, 113
Craster Hall, 348
Crœsus, 233
Cuckoo, H.M.S., 110
Culloden, 106
Curaçoa, 133
Cuzco, 264, 295
Cyclops, 316
Cygnus, 112
Cymba, 99
Dakota, 248
Damascus, 296
Dane, 183
Daniel Drew, 51
Dantzig, 319
Dasher, H.M.S., 110
De Witt Clinton, 45, 46
Dee, 262
Defiance, 69
Delaware, 340
Delcomyn, 297
Delta, 260
Demologos, 35
Destroyer, 339
Deutschland, 305
Devastation, 333
Devonshire, 97
Diana, 115
Dieppe, 107, 109
Dispatch, 113
Dominion, H.M.S., 358
Doncaster, 207
Donegal, 121
Dora, 93
Douglas, 90, 93
Douro, 263
Dover, 105, 317
Dreadnought, 309, 315, 335
Drottning Victoria, 365
Duchess of Albany, 116
Duchess of Buccleuch, 96
Duchess of Connaught, 116
Duchess of Devonshire, 96, 121
Duchess of Edinburgh, 116
Duchess of Fife, 116
Duchess of Kent, 116
Duchess of Sutherland, 119
Duchess of York, 106
Duke of Cornwall, 97
Duke of Sutherland, 119
Duke of Wellington, 272
Dumbarton Castle, 70
Dumfries, 113
Duncannon, 75
Dundee, 87
Dwarf, 313
Eagle, 376
Earl Grey, 370
Earl of Hardwicke, 167
Earl of Liverpool, 82
Earl Spencer, 120
Echo, 176
Echunga, 347
Eclipse, 54
Eden, 309, 335
Edinburgh, 240
Edith, 119, 172, 328
Egypt, 293
Eleanor, 120
Elizabeth, 64
Ella, 115, 116
Ellan Vannin, 91
Empire, 48
Empire of Troy, 48
Empress, 105
Empress of Russia, 35
Empress Queen, 93
Encounter, H.M.S., 314
Endeavour, 201
Enterprise, 43, 165-166, 306
Ericsson, 384
Erin, 254
Ermack, 367, 368, 369
Esk, 262
Etna, 45
Etruria, 281, 282
Europa, 152, 245
Excellent, 325
Experiment, 328
Express, 113
F. P. Smith, 216
Faid Gihaad, 372
Fairy, 371
Falcon, 165
Falken, 373
Fannie, 115
Far East, 265
Fenella, 92
Firebrand, 176
Firefly, 35, 41, 44, 45
Flora, 325
Florida, 288
Forth, 191
Foyle, 98
Francis B. Ogden, 218
Franklin, 43, 154
Frederica, 116
Frolic, 96
Fulton, 35, 41, 154
Gaelic, 253
Galtee-More, 120
Garonne, 264, 295
Garry Owen, 196, 221
Gascony, 349
Gemini, 376
General Admiral Apraxine, 368
George Canning, 74
George Washington, 304
Georgia, 188
Germanic, 253
Geyser, 313
Glasgow, 86, 240
Glatton, 334
Glen Cove, 50
Gloire, 320
Gorgon, 316
Grace, 106
Grand Turk, 111
Great Britain, 217, 221, 256, 271
Great Eastern, 193, 230, 241, 268-278, 284, 288
Great Liverpool, 178
Great Western, 138, 141-144, 145, 147-148, 150, 169, 220, 238, 271
Greenock, 88, 318
Griffin, 115
Guadeloupe, 316
Guernsey, 116
Hansa, 301
Havre, 113, 114
Hazard, 331
Hebe, 315, 327
Helvetia, 254
Hendrick Hudson, 48, 49
Henry Bell, 73, 88, 100
Henry Clay, 170
Her Majesty, 232
Herald, 96
Hercules, 315, 332
Hermann, 154
Hermes, 176
Hibernia, 71, 102, 104, 120, 151
Hilda, 115, 116
Himalaya, 180, 260, 271
Hindostan, 179
Hohenzollern, 371
Ho-Nam, 206
Honfleur, 108, 116
Hope, 36, 38, 39, 45, 69, 107
Howe, 315
Hudson, 300
Hugh Lindsay, 166
Humber, 64
Humboldt, 154
Iberia, 177
Idaho, 248
Immacolata Concezione, 372
Immingham, 118
Inconstant, 315, 333
Independencia, 315
Indian, 254
Indian Empire, 162
Industry, 66
Invincible, 336
Iolanda, 375
Irishman, 100, 101
Iroquois, 351
Isa, 306
Isabella, 120
Italy, 109, 254
Ivernia, 283
James Joicey, 307
James Watt, 81, 100
Jerome Napoleon, 373
John Bowes, 211, 213
John Elder, 264
John W. Richmond, 46
Jumna, 202
Kaiser Wilhelm II., 287, 305
Kaiser Wilhelm der Grosse, 304, 305
Kaiserin Auguste Victoria, 305
Kangaroo, 239
Karamea, 298
Kate, 327
Kearsarge, 176, 340
Kentucky, 340
King Edward, 309
King Orry, 89, 92
Kingfisher, 83
Kite, 124
Koenig Wilhelm, 333
Koenig Wilhelm II., 305
Kronprinz Wilhelm, 305
Kronprinzessin Cecilie, 305
La France, 259
La Plata, 262
Lady de Saumarez, 110
Lady Derby, 233
Lady Eglinton, 98, 245
Lady Grey, 369
Lady Hudson-Kinahan, 99
Lady Martin, 99
Lady Olive, 99
Lady Roberts, 99
Lady Wodehouse, 98
Lady Wolseley, 99
Ladybird, 107
Lancashire Witch, 96
Larriston, 245
Laura, 116
Laurentic, 289
Le Nord, 105
Leinster, 204
Leven, 72
Leviathan, 270, 276
Lewis, 155
Lexington, 45, 46
Liffey, 73
Lightning, 264, 311, 336
Lily, 119
Lima, 229
Livadia, 373
Liverpool, 145