Skip to main content

Advertisement

Log in

A framework of shape optimisation based on the isogeometric boundary element method toward designing thin-silicon photovoltaic devices

  • Original Article
  • Published:
Engineering with Computers Aims and scope Submit manuscript

Abstract

We propose a gradient-based optimisation framework to design the shape of the interfaces in periodic layered structures, which can model photovoltaic devices, optical gratings, etc., according to a desired electromagnetic property. To this end, we first develop an isogeometric boundary element method (IGBEM) to analyse singly periodic boundary value problems for 2D Helmholtz equation in layered media. By nature, the IGBEM is not only numerically accurate but also suitable for the shape optimisation in comparison with the conventional boundary- and domain-type solvers. Next, we derive the shape derivative (or shape sensitivity) of the objective function in terms of the magnetic field strength or energy absorption rate on the basis of the adjoint variable method. The shape derivative can be computed by solving the primal and adjoint problems with the IGBEM. Subsequently, we map our shape optimisation problems to nonlinear programming problems in order to exploit a general-purpose software for the latter problems. After verifying our optimisation framework rigorously by comparing with the exact solutions, we demonstrate a shape optimisation of the silicon-metal interface in a model of thin-silicon photovoltaic devices: it is crucial to determine the interface so that the energy of the incident light can be confined in the silicon layer as much as possible when the thickness of the layer is unconventionally small; one micrometre in our model. The optimal shape achieved 8.6 times higher absorption rate than the reference (flat) shape. This simulation shows that the proposed framework is capable of making a fundamental contribution to analysing and designing photovoltaic devices as well as photonic and plasmonic devices that consist of singly periodic layers.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8
Fig. 9
Fig. 10
Fig. 11
Fig. 12
Fig. 13
Fig. 14
Fig. 15
Fig. 16
Fig. 17
Fig. 18
Fig. 19

Similar content being viewed by others

Notes

  1. It is worth mentioning that Bai et al. [46] investigated doubly periodic problems for 2D elastostatics with the IGBEM. They utilised the fundamental solution by discretising the periodic boundaries directly. However, the approach based on the fundamental solution does not work (without approximation) for any 2D singly-periodic problem, which is the case in this study, because the periodic boundaries, which are denoted by \(\Gamma ^\mathrm{L}\) and \(\Gamma ^\mathrm{R}\) in this paper, are infinitely long.

  2. Strictly speaking, Kostas et al. [22] use gradient-free algorithms, i.e. the evolutionary algorithms. Meanwhile, Gillebaart et al. [25] perform gradient-based optimisation but compute the sensitivities with the algorithmic (or automatic) differentiation.

  3. This is the case for the BIEs in Eq. 7 and the variation \(\delta \mathcal {G}_i\) in (60), for example.

  4. This is the case for Eq. 66 and the following expressions.

  5. It is possible to consider the TE mode, but it cannot excite surface plasmons on a dielectric-metal surface [4]. This is the reason why we consider the TM mode.

  6. The most LHS element \(I_0^{(a)}\) is singular if we use \(t=1\) as the collocation point, whereas this is not the case in this study. Note that the BIE for the potential collocation point at \(t=1\) is equivalent to that for \(t=0(=t_{p_a}^{(a)})\) due to the periodicity and, thus, they cannot be chosen as the collocation points at the same time.

  7. This is not the case for \(\mu =0\).

  8. It is apparent that \(u^\mathrm{inc}\) and \(u^\mathrm{ref}\) satisfy Eq. 5. On the other hand, the continuity of the magnetic fields must be satisfied at \(y=0\) for any x, i.e. \(u^\mathrm{inc}(x,0)+u^\mathrm{ref}(x,0)=u^\mathrm{tra}(x,0)\) for any x. This results in the relation for the x-component of the wavevectors, i.e. \(k_0\cos \theta =k_1\sin \phi\). Then, \(u^\mathrm{tra}\) satisfies Eq. 5.

  9. Strictly speaking, the control points are indirectly used as the design variables in [22].

  10. In order to maximise the function f, one may consider \(-f\) instead of f.

  11. The x-coordinate of \(\varvec{z}_0^{(0)}\) is irrelevant to this problem theoretically because the magnetic field u is uniform in the x-direction in the present configuration.

  12. This number involves regular iterations and iterations during the restoration phase ([84], Page 60).

  13. The second term in the RHS of Eq. 33 can be rewritten as \(\sum _{\nu =0}^{n_0-1}s_{\nu ,y}^{(0)}\delta p_{{\nu },y}^{(0)}=\left( \sum _{\nu =0}^{n_0-1}s_\nu ^{(0)}\right) \delta l\) in the present case, where \(\delta l:=\delta p_{{0},y}^{(0)}=\cdots =\delta p_{{n_0-1},y}^{(0)}\). Therefore, the summation \(\sum _{\nu =0}^{n_0-1}s_\nu ^{(0)}\) corresponds to the exact shape derivative, i.e. \(\frac{\mathrm {d}\mathcal {M}}{\mathrm {d}l}\).

  14. This wavelength is in the infrared range and available on the earth. At this wavelength, the relative permittivities of silicon and silver are given as \(\varepsilon _1={13.8+0.16}\mathrm {i}\) and \(\varepsilon _2={2.1+34.0}\mathrm {i}\), respectively [86].

  15. The primal-dual interior point method [85], on which the Ipopt is based, uses the information of the past optimisation steps. So, restarting and thus discarding the past information can affect the behaviour of the optimisation, although we cannot clearly explain why our result was improved in the second try (and not improved in the third one) by restarting the optimisation.

  16. This is because \(\nabla \cdot \varvec{v}-\varvec{n}\cdot (\nabla \varvec{v}\varvec{n})\) is identical to \(\partial _y v_y\) when \(\varvec{n}\) and \(\varvec{v}\) are parallel to the x- and y-axis, respectively, and thus common to both \(\Gamma _i^\mathrm{L}\) and \(\Gamma _i^\mathrm{R}\). Meanwhile, \(\lambda ^*q\) on \(\Gamma _i^\mathrm{L}\) and that on \(\Gamma _i^\mathrm{R}\) have the same magnitude and different sign due to the quasi-periodic conditions of \(\lambda ^*\) in Eq. 63 and q in Eq. 65a. Thus, the underlying integrals are cancelled.

  17. The symbol \(\upepsilon _0\) is different from \(\varepsilon _0\), i.e. the relative permittivity in \(\Omega _0\).

References

  1. Polman A, Knight M, Garnett EC , Ehrler B, Sinke WC Photovoltaic materials: present efficiencies and future challenges. Science 352(6283). https://doi.org/10.1126/science.aad4424. URL http://science.sciencemag.org/content/352/6283/aad4424

  2. Atwater HA, Polman A (2010) Plasmonics for improved photovoltaic devices. Nature Mater 9:205–213. https://doi.org/10.1038/nmat2629

    Article  Google Scholar 

  3. Mandal P, Sharma S (2016) Progress in plasmonic solar cell efficiency improvement: a status review. Renew Sustain Energy Rev 65:537–552. 10.1016/j.rser.2016.07.031. URL http://www.sciencedirect.com/science/article/pii/S1364032116303598

  4. Maier SA (2007) Plasmonics-fundamentals and Applications, Springer. URL http://opus.bath.ac.uk/8624/

  5. Ferry VE, Munday JN, Atwater HA (2010) Design considerations for plasmonic photovoltaics. Adv Mater 22(43):4794–4808. 10.1002/adma.201000488. URL http://dx.doi.org/10.1002/adma.201000488

  6. Aydin K, Ferry VE, Briggs RM, Atwater HA (2011) Broadband polarization-independent resonant light absorption using ultrathin plasmonic super absorbers. Nature Commun 2:517. copyright-Copyright Nature Publishing Group Nov 2011; Last updated - 2012-05-25. URL https://search.poquest.com/docview/1002587847?accountid=12653

  7. Spinelli P, Ferry VE , van de Groep J,  van Lare M, Verschuuren MA, Schropp REI, Atwater HA, Polman A (2012) Plasmonic light trapping in thin-film si solar cells. J Opt 14(2):024002. URL http://stacks.iop.org/2040-8986/14/i=2/a=024002

  8. Cao Y-Y, Takeyasu N, Tanaka T, Duan X-M, Kawata S (2009) 3D metallic nanostructure fabrication by surfactant-assisted multiphoton-induced reduction. Small 5(10):1144–1148. https://doi.org/10.1002/smll.200801179

    Google Scholar 

  9. Varghese LT, Fan L, Xuan Y, Tansarawiput C, Kim S, Qi M (2013) Resistless nanoimprinting in metal for plasmonic nanostructures. Small 9(22):3778–3783. https://doi.org/10.1002/smll.201300168

    Article  Google Scholar 

  10. Sugioka K, Cheng Y (2014) Femtosecond laser three-dimensional micro- and nanofabrication. Appl Phys Rev 1(4):041303. https://doi.org/10.1063/1.4904320

    Article  Google Scholar 

  11. Hughes T, Cottrell J, Bazilevs Y (2005) Isogeometric analysis: CAD, finite elements. NURBS, exact geometry and mesh refinement. Comput Methods Appl Mech Eng 194(39–41):4135–4195. 10.1016/j.cma.2004.10.008. URL http://www.sciencedirect.com/science/article/pii/S0045782504005171

  12. Cottrell JA, Hughes TJR, Bazilevs Y (2009) Isogeometric analysis: toward Integration of CAD and FEA, 1st edn. Wiley Publishing, Singapore

    Book  MATH  Google Scholar 

  13. Nguyen VP, Anitescu C, Bordas SP, Rabczuk T (2015) Isogeometric analysis: an overview and computer implementation aspects. Math Comput Simul 117:89–116. 10.1016/j.matcom.2015.05.008. URL http://www.sciencedirect.com/science/article/pii/S0378475415001214

  14. Bonnet M (1999) Boundary integral equation methods for solids and fluids. Wiley, West Sussex

    Google Scholar 

  15. Politis C, Ginnis A, Kaklis P, Belibassakis K, Feurer C (2009) An isogeometric BEM for exterior potential-flow problems in the plane. In: Proceedings of SIAM/ACM joint conference on geometric and physical modeling, pp. 349–354

  16. Takahashi T, Matsumoto T (2012) An application of fast multipole method to isogeometric boundary element method for Laplace equation in two dimensions. Eng Anal Bound Elem 36(12):1766–1775. 10.1016/j.enganabound.2012.06.004. URL http://www.sciencedirect.com/science/article/pii/S0955799712001373

  17. Jeong JW, Oh H-S, Kang S, Kim H (2013) Mapping techniques for isogeometric analysis of elliptic boundary value problems containing singularities. Comput Methods Appl Mech Eng 254:334–352. 10.1016/j.cma.2012.09.009. URL http://www.sciencedirect.com/science/article/pii/S0045782512002915

  18. May S, Kstner M, Mller S, Ulbricht V (2014) A hybrid IGAFEM/IGABEM formulation for two-dimensional stationary magnetic and magneto-mechanical field problems. Comput Methods Appl Mech Eng 273:161–180. 10.1016/j.cma.2014.01.015. URL http://www.sciencedirect.com/science/article/pii/S0045782514000280

  19. Feischl M, Gantner G, Praetorius D (2015) Reliable and efficient a posteriori error estimation for adaptive IGA boundary element methods for weakly-singular integral equations. Comput Methods Appl Mech Eng 290:362–386. 10.1016/j.cma.2015.03.013. URL http://www.sciencedirect.com/science/article/pii/S0045782515001231

  20. Yoon M, Choi M-J, Cho S (2015) Isogeometric configuration design optimization of heat conduction problems using boundary integral equation. Int J Heat Mass Transf 89:937–949. 10.1016/j.ijheatmasstransfer.2015.05.112. URL http://www.sciencedirect.com/science/article/pii/S001793101500616X

  21. Manzoni A, Salmoiraghi F, Heltai L (2015) Reduced basis isogeometric methods (RB-IGA) for the real-time simulation of potential flows about parametrized NACA airfoils. Comput Methods Appl Mech Eng 284:1147–1180 (isogeometric Analysis Special Issue). http://dx.doi.org/10.1016/j.cma.2014.11.037. URL http://www.sciencedirect.com/science/article/pii/S0045782514004708

  22. Kostas K, Ginnis A, Politis C, Kaklis P (2015) Ship-hull shape optimization with a T-spline based BEM-isogeometric solver. Comput Methods Appl Mech Eng 284:611–622 (isogeometric Analysis Special Issue). http://dx.doi.org/10.1016/j.cma.2014.10.030. URL http://www.sciencedirect.com/science/article/pii/S0045782514004009

  23. Feischl M, Gantner G, Haberl A, Praetorius D (2016) Adaptive 2D IGA boundary element methods. Eng Anal Boundary Elem 62:141–153. 10.1016/j.enganabound.2015.10.003. URL http://www.sciencedirect.com/science/article/pii/S0955799715002155

  24. Simpson R, Liu Z (2016) Acceleration of isogeometric boundary element analysis through a black-box fast multipole method. Eng Anal Bound Elem 66:168–182. 10.1016/j.enganabound.2016.03.004. URL http://www.sciencedirect.com/science/article/pii/S0955799716300315

  25. Gillebaart E, Breuker RD (2016) Low-fidelity 2D isogeometric aeroelastic analysis and optimization method with application to a morphing airfoil. Comput Methods Appl Mech Eng 305:512–536. 10.1016/j.cma.2016.03.014. URL http://www.sciencedirect.com/science/article/pii/S0045782516300962

  26. Maestre J, Cuesta I, Pallares J (2016) An unsteady 3D isogeometrical boundary element analysis applied to nonlinear gravity waves. Comput Methods Appl Mech Eng 310:112–133. 10.1016/j.cma.2016.06.031. URL http://www.sciencedirect.com/science/article/pii/S0045782516306569

  27. Feischl M, Gantner G, Haberl A, Praetorius D (2017) Optimal convergence for adaptive IGA boundary element methods for weakly-singular integral equations. Numeri Math 136(1):147–182. 10.1007/s00211-016-0836-8. URL http://dx.doi.org/10.1007/s00211-016-0836-8

  28. Aimi A, Diligenti M, Sampoli M, Sestini A (2017) Non-polynomial spline alternatives in isogeometric symmetric Galerkin BEM. Appl Numer Math 116:10–23 (new Trends in Numerical Analysis: Theory, Methods, Algorithms and Applications (NETNA 2015)). https://doi.org/10.1016/j.apnum.2016.07.004. URL http://www.sciencedirect.com/science/article/pii/S0168927416301118

  29. Kostas K, Ginnis A, Politis C, Kaklis P (2017) Shape-optimization of 2D hydrofoils using an isogeometric BEM solver. Comput Aided Design 82:79–87 (isogeometric Design and Analysis). https://doi.org/10.1016/j.cad.2016.07.002. URL http://www.sciencedirect.com/science/article/pii/S0010448516300653

  30. Gong Y, Dong C, Qin X (2017) An isogeometric boundary element method for three dimensional potential problems. J Comput Appl Math 313:454–468. 10.1016/j.cam.2016.10.003. URL http://www.sciencedirect.com/science/article/pii/S0377042716304824

  31. Gong Y, Dong C, Bai Y (2017) Evaluation of nearly singular integrals in isogeometric boundary element method. Eng Anal Bound Elem 75:21–35. 10.1016/j.enganabound.2016.11.005. URL http://www.sciencedirect.com/science/article/pii/S0955799716302909

  32. Gong Y, Dong C (2017) An isogeometric boundary element method using adaptive integral method for 3D potential problems. J Comput Appl Math 319:141–158. 10.1016/j.cam.2016.12.038. URL http://www.sciencedirect.com/science/article/pii/S037704271730002X

  33. Sapountzakis E, Tsiptsis I (2017) B-splines in the analog equation method for the generalized beam analysis including warping effects. Comput Struct 180:60–73. 10.1016/j.compstruc.2016.03.007. URL http://www.sciencedirect.com/science/article/pii/S0045794916300621

  34. Aimi A, Calabrò F, Diligenti M, Sampoli ML, Sangalli G, Sestini A. New efficient assembly in isogeometric analysis for symmetric Galerkin boundary element method, ArXiv e-prints arXiv:1703.10016

  35. Li K, Qian X (2011) Isogeometric analysis and shape optimization via boundary integral. Comput Aided Design 43(11):1427–1437. 10.1016/j.cad.2011.08.031. URL http://www.sciencedirect.com/science/article/pii/S0010448511002302

  36. Simpson R, Bordas S, Trevelyan J, Rabczuk T (2012) A two-dimensional isogeometric boundary element method for elastostatic analysis. Comput Methods Appl Mech Eng 209:87–100. 10.1016/j.cma.2011.08.008. URL http://www.sciencedirect.com/science/article/pii/S0045782511002635

  37. Simpson R, Bordas S, Lian H, Trevelyan J (2013) An isogeometric boundary element method for elastostatic analysis: 2D implementation aspects. Comput Struct 118:2–12 (special Issue: UK Association for Computational Mechanics in Engineering). 10.1016/j.compstruc.2012.12.021. URL http://www.sciencedirect.com/science/article/pii/S0045794912003355

  38. Scott M, Simpson R, Evans J, Lipton S, Bordas S, Hughes T, Sederberg T (2013) Isogeometric boundary element analysis using unstructured T-splines. Comput Methods Appl Mech Eng 254:197–221. 10.1016/j.cma.2012.11.001. URL http://www.sciencedirect.com/science/article/pii/S0045782512003386

  39. Belibassakis K, Gerostathis T, Kostas K, Politis C, Kaklis P, Ginnis A, Feurer C (2013) A BEM-isogeometric method for the ship wave-resistance problem. Ocean Eng 60:53–67. 10.1016/j.oceaneng.2012.12.030. URL http://www.sciencedirect.com/science/article/pii/S0029801812004490

  40. Mallardo V, Ruocco E (2014) An improved isogeometric boundary element method approach in two dimensional elastostatics. Comput Model Eng Sci 102(5):373–391. https://doi.org/10.3970/cmes.2014.102.373

    MathSciNet  MATH  Google Scholar 

  41. Zechner J, Marussig B, Beer G, Fries T-P. Isogeometric boundary element method with hierarchical matrices, ArXiv e-prints arXiv:1406.2817

  42. Marussig B, Zechner J, Beer G, Fries T-P (2015) Fast isogeometric boundary element method based on independent field approximation. Comput Methods Appl Mech Eng 284:458–488 (isogeometric Analysis Special Issue). http://dx.doi.org/10.1016/j.cma.2014.09.035. URL http://www.sciencedirect.com/science/article/pii/S0045782514003582

  43. Gu J, Zhang J, Chen L, Cai Z (2015) An isogeometric BEM using PB-spline for 3-D linear elasticity problem. Eng Anal Bound Elem 56:154–161. 10.1016/j.enganabound.2015.02.013. URL http://www.sciencedirect.com/science/article/pii/S0955799715000612

  44. Wang Y, Benson D (2015) Multi-patch nonsingular isogeometric boundary element analysis in 3D. Comput Methods Appl Mech Eng 293:71–91. 10.1016/j.cma.2015.03.016. URL http://www.sciencedirect.com/science/article/pii/S0045782515001267

  45. Wang Y, Benson D, Nagy A (2015) A multi-patch nonsingular isogeometric boundary element method using trimmed elements. Comput Mech 56(1):173–191. URL http://search.ebscohost.com/login.aspx?direct=true&db=aph&AN=103364844&lang=ja&site=ehost-live

  46. Bai Y, Dong C, Liu Z (2015) Effective elastic properties and stress states of doubly periodic array of inclusions with complex shapes by isogeometric boundary element method. Compos Struct 128:54–69. 10.1016/j.compstruct.2015.03.061. URL http://www.sciencedirect.com/science/article/pii/S0263822315002445

  47. Beer G, Marussig B, Zechner J, Dnser C, Fries T-P (2016) Isogeometric boundary element analysis with elasto-plastic inclusions. Part 1: plane problems. Comput Methods Appl Mech Eng 308:552–570. 10.1016/j.cma.2016.03.035. URL http://www.sciencedirect.com/science/article/pii/S0045782516301219

  48. Yoon M, Cho S (2016) Isogeometric shape design sensitivity analysis of elasticity problems using boundary integral equations. Eng Anal Bound Elem 66:119–128. 10.1016/j.enganabound.2016.01.010. URL http://www.sciencedirect.com/science/article/pii/S0955799716300017

  49. Mallardo V, Ruocco E (2016) A NURBS boundary-only approach in elasticity. Eur J Comput Mech 25(1-2):71–90. arXiv:http://www.tandfonline.com/doi/pdf/10.1080/17797179.2016.1181034, 10.1080/17797179.2016.1181034. URL http://www.tandfonline.com/doi/abs/10.1080/17797179.2016.1181034

  50. Lian H, Kerfriden P, Bordas SPA (2016) Implementation of regularized isogeometric boundary element methods for gradient-based shape optimization in two-dimensional linear elasticity. Int J Numer Methods Eng 106(12):972–1017. https://doi.org/10.1002/nme.5149

    Article  MathSciNet  MATH  Google Scholar 

  51. Peng X, Atroshchenko E, Kerfriden P, Bordas SPA (2017) Linear elastic fracture simulation directly from CAD: 2D NURBS-based implementation and role of tip enrichment. Int J Fracture 204(1):55–78. https://doi.org/10.1007/s10704-016-0153-3

    Article  Google Scholar 

  52. Peng X, Atroshchenko E, Kerfriden P, Bordas S (2017) Isogeometric boundary element methods for three dimensional static fracture and fatigue crack growth. Comput Methods Appl Mech Eng 316:151–185 (special Issue on Isogeometric Analysis: Progress and Challenges). https://doi.org/10.1016/j.cma.2016.05.038. URL http://www.sciencedirect.com/science/article/pii/S0045782516304832

  53. Lian H, Kerfriden P, Bordas S (2017) Shape optimization directly from CAD: an isogeometric boundary element approach using T-splines. Comput Methods Appl Mech Eng 317:1–41. 10.1016/j.cma.2016.11.012. URL http://www.sciencedirect.com/science/article/pii/S0045782516315365

  54. Vazquez R, Buffa A, Rienzo LD (2012) NURBS-based BEM implementation of high-order surface impedance boundary conditions. IEEE Trans Magn 48(12):4757–4766. https://doi.org/10.1109/TMAG.2012.2204897

    Article  Google Scholar 

  55. Peake M, Trevelyan J, Coates G (2013) Extended isogeometric boundary element method (XIBEM) for two-dimensional Helmholtz problems. Comput Methods Appl Mech Eng 259:93–102. 10.1016/j.cma.2013.03.016. URL http://www.sciencedirect.com/science/article/pii/S0045782513000674

  56. Simpson R, Scott M, Taus M, Thomas D, Lian H (2014) Acoustic isogeometric boundary element analysis. Computer Methods in Applied Mechanics and Engineering 269:265–290. 10.1016/j.cma.2013.10.026. URL http://www.sciencedirect.com/science/article/pii/S0045782513002788

  57. Peake M, Trevelyan J, Coates G (2015) Extended isogeometric boundary element method (XIBEM) for three-dimensional medium-wave acoustic scattering problems. Comput Methods Appl Mech Eng 284:762–780 (isogeometric Analysis Special Issue). http://dx.doi.org/10.1016/j.cma.2014.10.039. URL http://www.sciencedirect.com/science/article/pii/S0045782514004095

  58. Coox L, Atak O, Vandepitte D, Desmet W (2017) An isogeometric indirect boundary element method for solving acoustic problems in open-boundary domains. Comput Methods Appl Mech Eng 316:186–208 (special Issue on Isogeometric Analysis: Progress and Challenges). https://doi.org/10.1016/j.cma.2016.05.039. URL http://www.sciencedirect.com/science/article/pii/S0045782516304844

  59. Liu Z, Majeed M, Cirak F, Simpson RN. Isogeometric FEM-BEM coupled structural-acoustic analysis of shells using subdivision surfaces, ArXiv e-prints arXiv:1704.08491

  60. Heltai L, Arroyo M, DeSimone A (2014) Nonsingular isogeometric boundary element method for Stokes flows in 3D. Comput Methods Appl Mech Eng 268:514–539. 10.1016/j.cma.2013.09.017. URL http://www.sciencedirect.com/science/article/pii/S0045782513002442

  61. van Opstal T, van Brummelen E, van Zwieten G (2015) A finite-element/boundary-element method for three-dimensional, large-displacement fluidstructure-interaction. Comput Methods Appl Mech Eng 284:637–663 (isogeometric Analysis Special Issue). http://doi.org/10.1016/j.cma.2014.09.037. URL http://www.sciencedirect.com/science/article/pii/S0045782514003600

  62. Joneidi A, Verhoosel C, Anderson P (2015) Isogeometric boundary integral analysis of drops and inextensible membranes in isoviscous flow. Comput Fluids 109:49–66. 10.1016/j.compfluid.2014.12.011. URL http://www.sciencedirect.com/science/article/pii/S0045793014004745

  63. Heltai L, Kiendl J, DeSimone A, Reali A (2017) A natural framework for isogeometric fluid-structure interaction based on BEM-shell coupling, Comput Methods Appl Mech Eng 316:522–546 (special Issue on Isogeometric Analysis: Progress and Challenges). https://doi.org/10.1016/j.cma.2016.08.008. URL http://www.sciencedirect.com/science/article/pii/S0045782516308854

  64. Boedec G, Leonetti M, Jaeger M (2017) Isogeometric FEM-BEM simulations of drop, capsule and vesicle dynamics in Stokes flow. J Comput Phys 342:117–138. 10.1016/j.jcp.2017.04.024. URL http://www.sciencedirect.com/science/article/pii/S0021999117302954

  65. Li J, Dault D, Liu B, Tong Y, Shanker B (2016) Subdivision based isogeometric analysis technique for electric field integral equations for simply connected structures. J Comput Phys 319:145–162. 10.1016/j.jcp.2016.04.008. URL http://www.sciencedirect.com/science/article/pii/S0021999116300407

  66. Simpson RN, Liu Z, Vazquez R, Evans JA An isogeometric boundary element method for electromagnetic scattering with compatible B-spline discretizations, ArXiv e-prints arXiv:1704.07128

  67. Greengard L, Rokhlin V (1987) A fast algorithm for particle simulations. J Comput Phys 73(2):325–348. https://doi.org/10.1016/0021-9991(87)90140-9

    Article  MathSciNet  MATH  Google Scholar 

  68. Nishimura N (2002) Fast multipole accelerated boundary integral equation methods. Appl Mech Rev 55(4):299–324. 10.1115/1.1482087. URL http://link.aip.org/link/?AMR/55/299/1

  69. Liu Y (2009) Fast multipole boundary element method: theory and applications in engineering. Cambridge University Press, Cambridge. URL http://opac.inria.fr/record=b1132939

  70. Bebendorf M (2008) Hierarchical matrices: a means to efficiently solve elliptic boundary value problems, lecture notes in computational science and engineering. Springer Berlin, Heidelberg. URL https://books.google.co.jp/books?id=hnIJOwyz9Z4C

  71. Hirai T, Takahashi T, Isakari H, Matsumoto T (2017) A development of isogeometric Galerkin boundary element method for 2D periodic problem. In: The Proceedings of Conference of the Japan Society of Mechanical Engineers, Tokai Branch 2017.66 (2017) 511. 10.1299/jsmetokai.2017.66.511

  72. Wall WA, Frenzel MA, Cyron C (2008) Isogeometric structural shape optimization. Comput Methods Appl Mech Eng 197(33):2976–2988. 10.1016/j.cma.2008.01.025. URL http://www.sciencedirect.com/science/article/pii/S0045782508000509

  73. Komkov V, Choi KK, Haug EJ (1986) Design sensitivity analysis of structural systems. Academic Press, The United States of America

    MATH  Google Scholar 

  74. Sokolowski J, Zolesio J-P (1992) Introduction to shape optimization: shape sensitivity analysis. Springer, Berlin

    Book  MATH  Google Scholar 

  75. Udawalpola R, Wadbro E, Berggren M (2011) Optimization of a variable mouth acoustic horn. Int J Numer Methods Eng 85(5):591–606

    Article  MathSciNet  MATH  Google Scholar 

  76. List of optimization software. URL https://en.wikipedia.org/wiki/List_of_optimization_software

  77. R. Petit (Ed.), Electromagnetic theory of gratings, topics in current physics, Springer, 1980. URL https://books.google.co.jp/books?id=mrnvAAAAMAAJ

  78. Otani Y, Nishimura N (2008) An FMM for periodic boundary value problems for cracks for Helmholtz’ equation in 2D. Int J Numer Methods Eng 73(3):381–406. https://doi.org/10.1002/nme.2077

    Article  MathSciNet  MATH  Google Scholar 

  79. Barnett A, Greengard L (2010) A new integral representation for quasi-periodic fields and its application to two-dimensional band structure calculations. J Comput Phys 229(19):6898–6914. 10.1016/j.jcp.2010.05.029. URL http://www.sciencedirect.com/science/article/pii/S0021999110002901

  80. Cazeaux P, Zahm O (2015) A fast boundary element method for the solution of periodic many-inclusion problems via hierarchical matrix techniques. ESAIM Proc 48:156–168. https://doi.org/10.1051/proc/201448006

    Article  MathSciNet  MATH  Google Scholar 

  81. Arens T, Sandfort K, Schmitt S (2013) Analysing Ewald’s method for the evaluation of Green’s functions for periodic media. IMA J Appl Math 78:405–431. https://doi.org/10.1093/imamat/hxr057

    Article  MathSciNet  MATH  Google Scholar 

  82. Telles JCF (1987) A self-adaptive co-ordinate transformation for efficient numerical evaluation of general boundary element integrals. Int J Numer Methods Eng 24(5):959–973. https://doi.org/10.1002/nme.1620240509

    Article  MATH  Google Scholar 

  83. Feijóo GR , Oberai AA, Pinsky PM (2004) An application of shape optimization in the solution of inverse acoustic scattering problems. Inverse Prob 20(1):199–228. URL http://stacks.iop.org/0266-5611/20/i=1/a=012

  84. Kawajir Y, Laird C, Vigerske S, Wächter A (2015) A tutorial for downloading, installing, and using Ipopt (revision: 2538). URL https://projects.coin-or.org/Ipopt

  85. Wächter A, Biegler LT (2006) On the implementation of an interior-point filter line-search algorithm for large-scale nonlinear programming. Math Progr 106(1):25–57. https://doi.org/10.1007/s10107-004-0559-y

    Article  MathSciNet  MATH  Google Scholar 

  86. Palik E (1985) Handbook of Optical Constants of Solids, no. 1 in Academic Press handbook series, Academic Press, 1985. URL https://books.google.co.jp/books?id=opDvAAAAMAAJ

  87. Isakari H, Kondo T, Takahashi T, Matsumoto T (2017) A level-set-based topology optimisation for acousticelastic coupled problems with a fast BEM-FEM solver. Comput Methods Appl Mech Eng 315:501–521. 10.1016/j.cma.2016.11.006. URL http://www.sciencedirect.com/science/article/pii/S0045782516305187

  88. Linton C (1998) The Green’s function for the two-dimensional Helmholtz equation in periodic domains. J Eng Math 33(4):377–401. https://doi.org/10.1023/A:1004377501747

    Article  MathSciNet  MATH  Google Scholar 

Download references

Acknowledgements

This research was supported by JSPS KAKENHI (No. 15K06683). We would like to thank Mr. Tetsuro Hirai at Nagoya University for his helpful discussions on the proposed IGBEM, especially for the singular integrals. We also would like to thank Mr. Shinichi Hayashi of former graduate student for his contributions to the present research project at the early stage.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Toru Takahashi.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendices

Appendix A: Evaluation of the quasi-periodic Green’s function by Kummer’s transformation

In this appendix, we will denote the quasi-periodic Green’s function \(G_i\) and the wavenumber \(k_i\) in Eq. 8 as \(G\) and k, respectively, for concise. Also, we denote \(\varvec{x}\) and \(\varvec{y}\in \mathbb {R}^2\) as \((x_1,x_2)^T\) and \((y_1,y_2)^T\), respectively. In addition, we consider the case that k is real, because \(G\) converges rapidly if k is not real.

Applying Kummer’s transformation [88] to \(G\) leads to the two series \(G^{1}\) and \(G^{2}\) that can converge more rapidly than \(G\), i.e.

$$\begin{aligned} G^{1}(\varvec{x}-\varvec{y}):= & {} \frac{1}{2\pi }\sum _{m=-\infty }^{\infty }K_0(k\left| \varvec{x}-(\varvec{y}+ mL\varvec{e}_x)\right| )e^{\mathrm {i}\beta L m}, \end{aligned}$$
(42a)
$$\begin{aligned} G^{2}(\varvec{x}-\varvec{y}):= & {} \frac{1}{2L}\sum _{m=-\infty }^{\infty }\left( \frac{e^{-\sqrt{\xi _m^2-k^2}|x_2-y_2|}}{\sqrt{\xi _m^2-k^2}} - \frac{e^{-\sqrt{\xi _m^2+k^2}|x_2-y_2|}}{\sqrt{\xi _m^2+k^2}} \right) e^{\mathrm {i}\xi _m(x_1-y_1)}, \end{aligned}$$
(42b)

where \(G\equiv G^{1}+G^{2}\) holds, \(K_0\) denotes the modified Bessel function of the second kind of order zero and \(\xi _m:=\beta +\frac{2\pi m}{L}\). Note that, regarding any real wavenumber k, we evaluate \(\sqrt{\xi _m^2-k^2}\) as follows:

$$\begin{aligned} \sqrt{\xi _m^2-k^2}= {\left\{ \begin{array}{ll} \sqrt{\xi _m^2-k^2} &{} ~\mathrm{for}~ \xi _m^2> k^2, \\ -\mathrm {i}\sqrt{k^2-\xi _m^2} &{} ~\mathrm{for}~ \xi _m^2 < k^2. \end{array}\right. } \end{aligned}$$
(43)

Note also that the term of \(m=0\) and \(-1\) in Eq. 42a can be singular under the configuration of the present IGBEM, as mentioned in Subsubsect. 2.3.5.

The convergence rate of the series \(G^1\) is \(\mathcal {O}(m^{-\frac{1}{2}}e^{-m})\), while that of \(G^2\) is \(\mathcal {O}(m^{-1}e^{-m})\) if \(|x_2-y_2|>0\) and \(\mathcal {O}(m^{-3})\) otherwise. These rates are genuinely better than that of \(G\), i.e. \(\mathcal {O}(m^{-1})\).

We need to truncate the infinite series of Eq. 42a and b. To this end, we compare the value of \(G^1\) or \(G^2\) evaluated for \(\left| m\right| \le p\) for a certain number p with the previous value for \(\left| m\right| \le p-1\). If the change of the former value relative to the latter one is smaller than \(10^{-10}\), we employ the former value. Otherwise, we increase the number p by one and repeat the evaluation and comparison.

Regarding the double layer kernel, i.e. \(\frac{\partial G(\varvec{x}-\varvec{y})}{\partial n}\), the normal derivative of \(G^1\) and \(G^2\) is computed as follows:

$$\begin{aligned}&\frac{\partial G^1(\varvec{x}-\varvec{y})}{\partial n(\varvec{y})} = \frac{k}{2\pi }\sum _{m=-\infty }^{\infty }K_1(k|\varvec{x}-(\varvec{y}+ mL\varvec{e}_x)|)\frac{(\varvec{x}-(\varvec{y}+ mL\varvec{e}_x))\cdot \varvec{n}(\varvec{y})}{|\varvec{x}-(\varvec{y}+ mL\varvec{e}_x)|}e^{\mathrm {i}\beta L m},\end{aligned}$$
(44a)
$$\begin{aligned}&\frac{\partial G^2(\varvec{x}-\varvec{y})}{\partial n(\varvec{y})} = \frac{1}{2L}\sum _{m=-\infty }^{\infty }\left( \left( -\mathrm {i}\xi _m n_1(\varvec{y})+\mathrm {sgn}(x_2-y_2)\sqrt{\xi _m^2-k^2}n_2(\varvec{y})\right) \frac{e^{-\sqrt{\xi _m^2-k^2}|x_2-y_2|}}{\sqrt{\xi _m^2-k^2}}\right. \nonumber \\&\left. -\left( -\mathrm {i}\xi _m n_1(\varvec{y})+\mathrm {sgn}(x_2-y_2)\sqrt{\xi _m^2+k^2}n_2(\varvec{y})\right) \frac{e^{-\sqrt{\xi _m^2+k^2}|x_2-y_2|}}{\sqrt{\xi _m^2+k^2}} \right) e^{\mathrm {i}\xi _m(x_1-y_1)}, \end{aligned}$$
(44b)

where \(K_1\) denotes the modified Bessel function of the second kind of order one and \(\mathrm {sgn}(f)\) denotes the sign of f. The way to truncate the infinite series in Eq. 44 is the same as \(G^1\) and \(G^2\).

Appendix B: B-spline basis function of degree 2

Letting \(\{t_0,\ldots ,t_{n+2}\}\) be the knots, the B-spline basis function of degree 2, denoted by \(N_\nu ^2\), and its derivative are given as follows:

$$\begin{aligned} N^2_\nu (t) = {\left\{ \begin{array}{ll} 0 &{} t< t_\nu \\ \frac{(t-t_\nu )^2}{(t_{\nu +2}-t_\nu )(t_{\nu +1}-t_\nu )} &{} t_\nu \le t< t_{\nu +1} \\ \frac{(t-t_\nu )(t_{\nu +2}-t)}{(t_{\nu +2}-t_\nu )(t_{\nu +2}-t_{\nu +1})} + \frac{(t_{\nu +3} - t)(t - t_{\nu +1})}{(t_{\nu +3} - t_{\nu +1})(t_{\nu +2} - t_{\nu +1})} &{} t_{\nu +1} \le t< t_{\nu +2}\\ \frac{(t_{\nu +3} - t)^2}{(t_{\nu +3} - t_{\nu +1})(t_{\nu +3} - t_{\nu +2})} &{} t_{\nu +2} \le t < t_{\nu +3} \\ 0 &{} t \ge t_{\nu +3},\\ \end{array}\right. } \end{aligned}$$
$$\begin{aligned} \frac{\mathrm {d}N^2_\nu (t)}{\mathrm {d}t} = {\left\{ \begin{array}{ll} 0 &{} t< t_\nu \\ \frac{2(t-t_\nu )}{(t_{\nu +2}-t_\nu )(t_{\nu +1}-t_\nu )} &{} t_\nu \le t< t_{\nu +1} \\ \frac{-2t+(t_\nu +t_{\nu +2})}{(t_{\nu +2}-t_\nu )(t_{\nu +2}-t_{\nu +1})} + \frac{-2t+(t_{\nu +1}+t_{\nu +3})}{(t_{\nu +3} - t_{\nu +1})(t_{\nu +2} - t_{\nu +1})} &{} t_{\nu +1} \le t< t_{\nu +2}\\ \frac{-2(t_{\nu +3} - t)}{(t_{\nu +3} - t_{\nu +1})(t_{\nu +3} - t_{\nu +2})} &{} t_{\nu +2} \le t < t_{\nu +3} \\ 0 &{} t \ge t_{\nu +3}. \end{array}\right. } \end{aligned}$$

Note that, when the denominator of a particular term in the RHSs vanishes, the term can be considered as zero.

Appendix C: Derivation of shape derivatives in Eqs. (28) and (31)

We derive shape derivative, i.e. the sensitivity of an objective function \(\mathcal {J}\) with respect to the shape of \(\Gamma\), in the framework of the adjoint variable method. To this end, we consider the following Lagrangian \(\mathcal {L}\):

$$\begin{aligned} \mathcal {L}(\Gamma ;u,\lambda ):=\mathcal {J}(\Gamma ;u) + \sum _{i=0}^{d-1} \mathop {\mathrm {Re}}\left[ \int _{\Omega _i} \frac{\lambda ^*}{\varepsilon _i}( {{{\Delta }}}u + k_i^2 u) \, \mathrm {d}\Omega \right] , \end{aligned}$$
(45)

where the Lagrange multiplier \(\lambda\) is an arbitrary function and the operator \({}^*\) denotes the complex conjugate. As long as u satisfies Eq. 2, \(\mathcal {L}\) is equivalent to \(\mathcal {J}\). Hence, we may maximise or minimise \(\mathcal {L}\) instead of \(\mathcal {J}\). Note that dividing \(\lambda\) by \(\varepsilon _i({}\ne 0)\) is not essential but necessary to obtain the adjoint problem in the same fashion as the primal problem. In what follows, we denote the integral over \(\Omega _i\) in the RHS of Eq. 45 by \(\mathcal {G}\), i.e.

$$\begin{aligned} \mathcal {G}(\Omega _i;u,\lambda ):=\int _{\Omega _i} \frac{\lambda ^*}{\varepsilon _i}( {{{\Delta }}}u + k_i^2 u) \, \mathrm {d}\Omega . \end{aligned}$$

This can be rewritten by the Green’s first identity as follows:

$$\begin{aligned} \mathcal {G}(\Omega _i;u,\lambda ) = \int _{\Omega _i} \frac{1}{\varepsilon _i} \left( \nabla \lambda ^* \cdot \nabla u - k_i^2 \lambda ^* u \right) \mathrm {d}\Omega - \int _{\partial \Omega _i} \lambda ^* q \, \mathrm {d}\Gamma . \end{aligned}$$
(46)

Next, we slightly move (oneperturbate) every point \(\varvec{x}\) in \(\Omega\) by \(\epsilon \varvec{v}(\varvec{x})\). Here, a real number \(\epsilon\) is infinitesimal and represents the degree of the onedesign perturbation (variation) and the vector \(\epsilon \varvec{v}(\varvec{x})\) represents the onedesign velocity at \(\varvec{x}\). Then, the domain \(\Omega _i\) after onedesign perturbation can be represented by

$$\begin{aligned} \tilde{\Omega }_i := \{\tilde{\varvec{x}} | \tilde{\varvec{x}} := \varvec{x} + \epsilon \varvec{v}(\varvec{x}), \,\varvec{x} \in \Omega _i\}. \end{aligned}$$
(47)

Here and hereafter, \(\tilde{f}\) stands for the quantity f after onedesign perturbation. We now give some assumptions for \(\varvec{v}\) on \(\Gamma _{-1}\), \(\Gamma _{d}\), \(\Gamma _i^\mathrm{L}\) and \(\Gamma _i^\mathrm{R}\). Regarding the infinite boundaries \(\Gamma _{-1}\) and \(\Gamma _{d-1}\), we may suppose that they are fixed, i.e.

$$\begin{aligned} \varvec{v}\equiv \varvec{0}\quad \text {on } \Gamma _{-1} \,\,\text {and}\,\, {\Gamma _{d-1}}. \end{aligned}$$
(48)

Meanwhile, we assume that the periodic boundaries \(\Gamma _i^\mathrm{L}\) and \(\Gamma _i^\mathrm{R}\) can move only in the \(+y\)-direction so that the period L is maintained, i.e.

$$\begin{aligned} \varvec{v}\cdot \varvec{n}=0\quad \text {on } \Gamma _i^\mathrm{L}\cup \Gamma _i^\mathrm{R}, \end{aligned}$$
(49)

where the normal vector \(\varvec{n}\) is identical to \((-1,0)^T\) and \((1,0)^T\) on \(\Gamma _i^\mathrm{L}\) and \(\Gamma _i^\mathrm{R}\), respectively. Moreover, the periodicity of the geometry in the x-direction imposes that the onedesign velocity at \(\varvec{x}\in \Gamma _i^\mathrm{L}\) and \(\varvec{x}+L\varvec{e}_x\in \Gamma _i^\mathrm{R}\) must be same, i.e.

$$\begin{aligned} \varvec{v}(\varvec{x}+L\varvec{e}_x) = \varvec{v}(\varvec{x})\quad ~\mathrm{for}~ \varvec{x}\in \Gamma _i^\mathrm{L}. \end{aligned}$$
(50)

We will define the shape derivative of \(\mathcal {L}\) from its variation due to the infinitesimal displacement \(\epsilon \varvec{v}\). We suppose that \(\tilde{u}\) is the solution of the primal problem with respect to \(\tilde{\Omega }\). Then, the Lagrangian after onedesign perturbation can be obtained from Eq. 45 as

$$\begin{aligned} \mathcal {L}(\tilde{\Gamma }; \tilde{u}, \lambda ) = \mathcal {J}(\tilde{\Gamma };\tilde{u}) + \sum _{i=0}^{d-1} \mathop {\mathrm {Re}}\left[ \mathcal {G}(\tilde{\Omega }_i; \tilde{u}, \lambda ) \right] , \end{aligned}$$
(51)

where

$$\begin{aligned} \mathcal {G}(\tilde{\Omega }_i; \tilde{u}, \lambda ) = \int _{\tilde{\Omega }_i} \frac{1}{\varepsilon _i} \left( \tilde{\nabla } \lambda ^* \cdot \tilde{\nabla } \tilde{u} - k_i^2 \lambda ^* \tilde{u} \right) \mathrm {d}\tilde{\Omega } -\int _{\partial \tilde{\Omega }_i} \lambda ^* \tilde{q} \, \mathrm {d}\tilde{\Gamma } \end{aligned}$$
(52)

follows from Eq. 46. In Eq. 51, it is allowed to use the same \(\lambda\) as that in Eq. 45 because the variation of \(\mathcal {J}\), which we actually desire to evaluate, is irrelevant to the choice of \(\lambda\) in Eq. 51 as long as \(\tilde{u}\) satisfies Eq. 2. Subtracting Eq. 45 from Eq. 51, we can obtain the variation of \(\mathcal {L}\), i.e.

$$\begin{aligned} \delta \mathcal {L}:= \mathcal {L}(\tilde{\Gamma }; \tilde{u}, \lambda ) - \mathcal {L}(\Gamma ; u, \lambda ) = \delta \mathcal {J}+ \sum _{i=0}^{d-1} \mathop {\mathrm {Re}}\left[ \delta \mathcal {G}_i \right] , \end{aligned}$$
(53)

where \(\delta \mathcal {J}:=\mathcal {J}(\tilde{\Gamma };\tilde{u})-\mathcal {J}(\Gamma ;u)\) and \(\delta \mathcal {G}_i:=\mathcal {G}(\tilde{\Omega }_i;\tilde{u},\lambda )-\mathcal {G}(\Omega _i;u,\lambda )\). Here, since \(\delta \mathcal {J}\rightarrow 0\) as \(\epsilon \rightarrow 0\), the variation \(\delta \mathcal {J}\) can be expanded as the polynomial of \(\epsilon\) without the term of \(\epsilon ^0\), i.e.

$$\begin{aligned} \delta \mathcal {J}= \epsilon \mathcal {S}(\Gamma ) + \mathcal {O}(\epsilon ^2) \Longleftrightarrow \mathcal {J}(\tilde{\Gamma }) = \mathcal {J}(\Gamma ) + \epsilon \mathcal {S}(\Gamma ) + \mathcal {O}(\epsilon ^2), \end{aligned}$$
(54)

where \(\mathcal {S}\) denotes the coefficient of \(\epsilon\) and is defined as the (first-order) shape derivative of \(\mathcal {J}\), i.e.

$$\begin{aligned} \mathcal {S}(\Gamma ):=\lim _{\epsilon \rightarrow 0}\frac{\delta \mathcal {J}}{\epsilon }=\lim _{\epsilon \rightarrow 0}\frac{\delta \mathcal {L}}{\epsilon }=\lim _{\epsilon \rightarrow 0}\frac{\delta \mathcal {J}+ \sum _{i=0}^{d-1} \mathop {\mathrm {Re}}\left[ \delta \mathcal {G}_i \right] }{\epsilon }. \end{aligned}$$
(55)

To express \(\mathcal {S}\) explicitly, we need to expand \(\mathcal {J}(\tilde{\Gamma };\tilde{u})\) and \(\mathcal {G}(\tilde{\Omega }_i; \tilde{u}, \lambda )\) in the RHS of Eq. 51 with respect to the infinitesimal number \(\epsilon\). First of all, \(\tilde{u}\) and \(\tilde{q}\) can be expanded in terms of \(\epsilon\) as follows:

$$\begin{aligned} \tilde{u}(\tilde{\varvec{x}}) = u(\varvec{x}) + \epsilon \dot{u}(\varvec{x}) + \mathcal {O}( \epsilon ^2),\quad \tilde{q}(\tilde{\varvec{x}}) = q(\varvec{x}) + \epsilon \dot{q}(\varvec{x}) + \mathcal {O}( \epsilon ^2), \end{aligned}$$
(56)

where

$$\begin{aligned} \dot{u}(\varvec{x}):=\left. \frac{\mathrm {d}u(\tilde{\varvec{x}})}{\mathrm {d}\epsilon } \right| _{\epsilon =0},\quad \dot{q}(\varvec{x}):=\left. \frac{\mathrm {d}q(\tilde{\varvec{x}})}{\mathrm {d}\epsilon } \right| _{\epsilon =0}. \end{aligned}$$

Accordingly, applying Eq. 56 to \(\tilde{u}_j(\tilde{\varvec{x}})=\tilde{u}_{j+1}(\tilde{\varvec{x}})\) and \(\tilde{q}_j(\tilde{\varvec{x}})=-\tilde{q}_{j+1}(\tilde{\varvec{x}})\) obtained from the boundary conditions in Eq. 3 and then gathering the terms of \(\varepsilon\), we can obtain the following boundary conditions in terms of \(\dot{u}\) and \(\dot{q}\):

$$\begin{aligned} \dot{u}_j = \dot{u}_{j+1}(=:\dot{u}),\quad \dot{q}_j = -\dot{q}_{j+1}(=:\dot{q})\quad \text { on }\Gamma _j. \end{aligned}$$
(57)

On the other hand, substituting Eq. 56 into Eq. 5 where \(u(\varvec{x})\) is replaced with \(\tilde{u}(\tilde{\varvec{x}})\) yields the following quasi-periodic condition for \(\dot{u}\):

$$\begin{aligned} \partial _x^n \dot{u}(\varvec{x}+ L\varvec{e}_x) = e^{\mathrm {i}\beta L} \partial _x^n \dot{u}(\varvec{x}),\quad \partial _y^n \dot{u}(\varvec{x}+ L\varvec{e}_x) = e^{\mathrm {i}\beta L} \partial _y^n \dot{u}(\varvec{x}). \end{aligned}$$
(58)

Next, we will write down \(\delta \mathcal {G}_i\), using the quantities without tilde. To this end, we utilise the following formulae [83]:

$$\begin{aligned}&\mathrm {d}\tilde{\Omega } = \det \varvec{F}\,\mathrm {d}\Omega = (1 + \epsilon \nabla \cdot \varvec{v})\,\mathrm {d}\Omega + \mathcal {O}(\epsilon ^2), \end{aligned}$$
(59a)
$$\begin{aligned}&\mathrm {d}\tilde{\Gamma } = \det \varvec{F}\left| \varvec{F}^{-T} \varvec{n}\right| \,\mathrm {d}\Gamma = \left[ 1 + \epsilon \bigl ( \nabla \cdot \varvec{v} - \varvec{n} \cdot (\nabla \varvec{v}) \varvec{n} \bigr )\right] \, \mathrm {d}\Gamma + \mathcal {O}(\epsilon ^2), \end{aligned}$$
(59b)

where \(\varvec{F}\in \mathbb {R}^{2\times 2}\) represents the deformation gradient defined by \(F_{ij}:=\frac{\partial \tilde{x}_i}{\partial x_j}\) and \(\varvec{n}\) denotes the outward normal vector of \(\partial \Omega _i\) in this context. Now, by substituting Eq. 56 and Eq. 59a into Eq. 52, we can express \(\delta \mathcal {G}_i\) as follows:

$$\begin{aligned} \delta \mathcal {G}_i= & {} \epsilon \, \Bigg [ \int _{\partial \Omega _i} \left( \varphi ^*\dot{u} - \lambda ^* \dot{q}\right) \, \mathrm {d}\Gamma \nonumber \\&- \int _{\Omega _i} \frac{1}{\varepsilon _i} \left( {{{\Delta }}}\lambda ^* + k_i^2 \lambda ^* \right) \dot{u} \, \mathrm {d}\Omega \nonumber \\&+ \int _{\partial \Omega _i} \frac{1}{\varepsilon _i} \left( \nabla \lambda ^* \cdot \nabla u - k_i^2 \lambda ^* u \right) \varvec{v} \cdot \varvec{n} \, \mathrm {d}\Gamma \nonumber \\&- \int _{\partial \Omega _i} \left( q (\varvec{v} \cdot \nabla \lambda ^*) + \varphi ^* (\varvec{v} \cdot \nabla u) \right) \, \mathrm {d}\Gamma \nonumber \\&- \int _{\partial \Omega _i} \lambda ^* q \left( \nabla \cdot \varvec{v} - \varvec{n} \cdot (\nabla \varvec{v}) \varvec{n} \right) \, \mathrm {d}\Gamma \nonumber \\&+ \int _{\Omega _i} \frac{1}{\varepsilon _i} \left( {{{\Delta }}}\lambda ^* + k_i^2 \lambda ^*\right) (\varvec{v} \cdot \nabla u) \, \mathrm {d}\Omega \Bigg ] + \mathcal {O}(\epsilon ^2), \end{aligned}$$
(60)

where the Gauss’s divergence theorem was used and \(\varphi ^*\) represents the flux of \(\lambda ^*\), i.e. \(\varphi ^*:=\frac{1}{\varepsilon _i}\frac{\partial \lambda ^*}{\partial n_i}\); note that the relative permittivity \(\varepsilon _i\) is not conjugated.

Meanwhile, the expression of \(\delta \mathcal {J}\) (thus, \(\mathcal {S}\)) depends on the choice of \(\mathcal {J}\). In this paper, we consider two cases of \(\mathcal {J}\) in the subsequent sections.

1.1 Case I: magnetic field strength

Let us consider the objective function \(\mathcal {J}^\mathrm{I}\) in Eq. 27. From Eq. 56, we can obtain

$$\begin{aligned} \frac{\left| \tilde{u}(\tilde{\varvec{x}})\right| ^2}{2} - \frac{\left| u(\varvec{x})\right| ^2}{2}= & {} \frac{(u^*(\varvec{x}) + \epsilon \dot{u}^*(\varvec{x}))(u(\varvec{x}) + \epsilon \dot{u}(\varvec{x}))}{2} - \frac{u^*(\varvec{x}) u(\varvec{x})}{2} + \mathcal {O}(\epsilon ^2)\\= & {} \epsilon \mathop {\mathrm {Re}}[ u^*(\varvec{x}) \dot{u}(\varvec{x})] + \mathcal {O}(\epsilon ^2). \end{aligned}$$

Thus, we can compute the variation of \(\mathcal {J}^\mathrm{I}\) as follows:

$$\begin{aligned} \delta \mathcal {J}^\mathrm{I}:= & {} \mathcal {J}^\mathrm{I}(\tilde{\Gamma }; \tilde{u}) - \mathcal {J}^\mathrm{I}(\Gamma ; u) \nonumber \\= & {} \epsilon \sum _{i=0}^{d-1} \sum _{m=0}^{M_i-1} \mathop {\mathrm {Re}}\left[ \int _{\Omega _i} u^*(\varvec{x}) \dot{u}(\varvec{x}) \updelta (\varvec{x} - \varvec{z}_m^{(i)}) \, \mathrm {d}\Omega \right] + \mathcal {O}(\epsilon ^2), \end{aligned}$$
(61)

where we assume that each observation point belongs to the same domain before and after deformation.

By substituting Eqs. 60 and 61 into 55, the shape derivative of \(\mathcal {J}^\mathrm{I}\), denoted by \(\mathcal {S}^\mathrm{I}\), can be written as follows:

$$\begin{aligned} \mathcal {S}^\mathrm{I}(\Gamma )= & {} \mathop {\mathrm {Re}}\sum _{i=0}^{d-1} \Bigg [ \int _{\partial \Omega _i} \left( \varphi ^* \dot{u} - \lambda ^* \dot{q} \right) \, \mathrm {d}\Gamma \nonumber \\&+ \int _{\partial \Omega _i} \frac{1}{\varepsilon _i}\left( \nabla \lambda ^* \cdot \nabla u - k_i^2 \lambda ^* u\right) \varvec{v} \cdot \varvec{n} \, \mathrm {d}\Gamma \nonumber \\&- \int _{\partial \Omega _i} \frac{1}{\varepsilon _i} \left( q (\varvec{v} \cdot \nabla \lambda ^*) + \varphi ^* (\varvec{v} \cdot \nabla u) \right) \, \mathrm {d}\Gamma \nonumber \\&- \int _{\partial \Omega _i} \lambda ^* q \left( \nabla \cdot \varvec{v} - \varvec{n} \cdot (\nabla \varvec{v} \varvec{n}) \right) \, \mathrm {d}\Gamma \nonumber \\&+ \int _{\Omega _i} \frac{1}{\varepsilon _i} \left( {{{\Delta }}}\lambda ^* + k_i^2 \lambda ^*\right) (\varvec{v} \cdot \nabla u) \, \mathrm {d}\Omega \nonumber \\&- \int _{\Omega _i} \frac{1}{\varepsilon _i} \left( {{{\Delta }}}\lambda ^* + k_i^2 \lambda ^* - \varepsilon _i \sum _{m=0}^{M_i-1} u^* \updelta (\varvec{x} - \varvec{z}_m^{(i)}) \right) \dot{u} \, \mathrm {d}\Omega \Bigg ]. \end{aligned}$$
(62)

Now, we will re-express the boundary integrals in terms of \(\partial \Omega _i\) in Eq. 62 with those in terms of \(\Gamma _j\). To this end, we will show that \(\Gamma _i^\mathrm{R}\), \(\Gamma _i^\mathrm{L}\), \(\Gamma _{-1}\) and \(\Gamma _{d-1}\) do not actually contribute to Eq. 62 by giving the quasi-periodic condition, i.e.

$$\begin{aligned} \partial _x^n \lambda ^*(\varvec{x}+ L\varvec{e}_x) = e^{-\mathrm {i}\beta L} \partial _x^n \lambda ^*(\varvec{x}),\quad \partial _y^n \lambda ^*(\varvec{x}+ L\varvec{e}_x) = e^{-\mathrm {i}\beta L} \partial _y^n \lambda ^*(\varvec{x}), \end{aligned}$$
(63)

and the radiation condition, i.e.

$$\begin{aligned} \lambda ^*(x,y)= {\left\{ \begin{array}{ll} \sum _n C_n e^{\mathrm {i}\phi _n x} e^{\mathrm {i}\psi _n (y-y_0)} &{} ~\mathrm{for}~ y>y_0,\\ \sum _n D_n e^{\mathrm {i}\phi _n x} e^{\mathrm {i}\psi _n (-y-y_0)} &{} ~\mathrm{for}~ y<-y_0, \end{array}\right. } \end{aligned}$$
(64)

where \(C_n\) and \(D_n\) are complex numbers, \(\phi _n:=-\beta +\frac{2\pi n}{L}\) and \(\psi _n:=\sqrt{k^2-\phi _n^2}\). Similarly to the radiation condition for \(u^\mathrm{sca}\) in Eq. 6, we consider \(n\in \mathbb {Z}\) such as \(k^2\ge \phi _n^2\). Note that, unlikely to \(\beta\) for u, the phase difference is \(-\beta\) for \(\lambda ^*\) in both Eqs. 63 and 64. On the other hand, the quasi-periodic conditions of u in Eq. 5, \(\dot{u}\) in Eq. 58 and \(\lambda ^*\) in Eq. 63 yield those of q, \(\dot{q}\) and \(\varphi ^*\), respectively, i.e.

$$\begin{aligned}&q(\varvec{x}+L\varvec{e}_x)=-e^{\mathrm {i}\beta L}q(\varvec{x})\quad ~\mathrm{for}~ \varvec{x}\in \Gamma _i^\mathrm{L},\end{aligned}$$
(65a)
$$\begin{aligned}&\dot{q}(\varvec{x}+L\varvec{e}_x)=-e^{\mathrm {i}\beta L}\dot{q}(\varvec{x})\quad ~\mathrm{for}~ \varvec{x}\in \Gamma _i^\mathrm{L},\end{aligned}$$
(65b)
$$\begin{aligned}&\varphi ^*(\varvec{x}+L\varvec{e}_x)=-e^{-\mathrm {i}\beta L}\varphi ^*(\varvec{x})\quad ~\mathrm{for}~ \varvec{x}\in \Gamma _i^\mathrm{L}, \end{aligned}$$
(65c)

where the negative sign in the RHSs contemplates the fact that the normal vector at \(\varvec{x}\in \Gamma _i^\mathrm{L}\) is opposite to that at the corresponding point \(\varvec{x}+L\varvec{e}_x\in \Gamma _i^\mathrm{R}\). First, from Eqs. 58, 63, 65b and 65c, the integral over \(\Gamma _i^\mathrm{R}\) is cancelled with that over \(\Gamma _i^\mathrm{L}\) in the first boundary term of Eq. 62. Second, the integrals over \(\Gamma _i^\mathrm{L}\) and \(\Gamma _i^\mathrm{R}\) in the second boundary term vanish because of Eq. 49. Third, the integrals over \(\Gamma _i^\mathrm{L}\) and \(\Gamma _i^\mathrm{R}\) in the third boundary term are cancelled because of the identity of \(\varvec{v}\) in Eq. 50 and the quasi-periodic conditions of \(\nabla u\) from Eq. 5, \(\nabla \lambda ^*\) from Eq. 63, q in Eq. 65a and \(\dot{q}\) in Eq. 65b. Fourth, the integrals over \(\Gamma _i^\mathrm{L}\) and \(\Gamma _i^\mathrm{R}\) in the fourth boundary term are cancelled.Footnote 16 Fifth, because of the radiation conditions for u in Eq. 6 and \(\lambda ^*\) in Eq. 64, the integrals over the infinite boundaries \(\Gamma _{-1}\) and \(\Gamma _{d-1}\) in the first boundary term vanish (see the proof in 8.1). Sixth and last, the integrals over \(\Gamma _{-1}\) and \(\Gamma _{d-1}\) in the second, third and fourth boundary terms vanish because of the assumption in Eq. 48. Therefore, we can regard \(\partial \Omega _i\) in Eq. 62 as follows:

$$\begin{aligned} \partial \Omega _i={\left\{ \begin{array}{ll} \Gamma _0 &{} ~\mathrm{for}~ i=0,\\ \Gamma _{d-2} &{} ~\mathrm{for}~ i=d-1,\\ \Gamma _{i-1}\cup \Gamma _i &{} \mathrm{otherwise}. \end{array}\right. } \end{aligned}$$

Hence, by considering the both sides of each boundary, we can rewrite Eq. 62 in the following boundary-centric expression:

$$\begin{aligned}\mathcal {S}^\mathrm{I}(\Gamma )= & {} \mathop {\mathrm {Re}}\sum _{j=0}^{d-2} \Bigg [ \int _{\Gamma _j} \left\{ \left( \varphi _j^* + \varphi _{j+1}^* \right) \dot{u} - \left( \lambda _j^* - \lambda _{j+1}^* \right) \dot{q} \right\} \, \mathrm {d}\Gamma \nonumber \\&+ \int _{\Gamma _j} \left\{ \left( \frac{1}{\varepsilon _j} \nabla \lambda _j^* \cdot \nabla u_j - \frac{1}{\varepsilon _{j+1}} \nabla \lambda _{j+1}^* \cdot \nabla u_{j+1} \right) - \left( \frac{k_j^2}{\varepsilon _j} \lambda _j^* - \frac{k_{j+1}^2}{\varepsilon _{j+1}} \lambda _{j+1}^* \right) u \right\} \varvec{v} \cdot \varvec{n} \, \mathrm {d}\Gamma \nonumber \\&- \int _{\Gamma _j} \left\{ q \varvec{v} \cdot ( \nabla \lambda _j^* - \nabla \lambda _{j+1}^* ) + \left( \varphi _j^* (\varvec{v} \cdot \nabla u_j) + \varphi _{j+1}^* (\varvec{v} \cdot \nabla u_{j+1}) \right) \right\} \mathrm {d}\Gamma \nonumber \\&- \int _{\Gamma _j} (\lambda _j^* - \lambda _{j+1}^* ) q \left( \nabla \cdot \varvec{v} - \varvec{n} \cdot (\nabla \varvec{v} \varvec{n}) \right) \, \mathrm {d}\Gamma \Bigg ] \nonumber \\&+\mathop {\mathrm {Re}}\sum _{i=0}^{d-1} \Bigg [\int _{\Omega _i} \frac{1}{\varepsilon _i} ({{{\Delta }}}\lambda ^* + k_i^2 \lambda ^*) ( \varvec{v} \cdot \nabla u) \, \mathrm {d}\Omega \nonumber \\&- \int _{\Omega _i} \frac{1}{\varepsilon _i} \left( {{{\Delta }}}\lambda ^* + k_i^2 \lambda ^* - \varepsilon _i \sum _{m=0}^{M_i-1} u^* \updelta (\varvec{x} - \varvec{z}_m^{(i)}) \right) \dot{u} \, \mathrm {d}\Omega \Bigg ] \end{aligned}$$
(66)

where the boundary conditions Eqs. 3 and 57 were considered and thus the symbols u, q, \(\dot{u}\) and \(\dot{q}\) (without subscript) have appeared on \(\Gamma _j\). Regarding the symbol \(\varvec{n}\) on \(\Gamma _j\), recall Fig. 3 (right). Note that we define \(\nabla u_j(\varvec{x}):=\displaystyle \lim _{\varvec{x'}\in \Omega _{j}\rightarrow \varvec{x}}\nabla u(\varvec{x'})\) and \(\nabla u_{j+1}(\varvec{x}):=\displaystyle \lim _{\varvec{x'}\in \Omega _{j+1}\rightarrow \varvec{x}}\nabla u(\varvec{x'})\) for \(\varvec{x}\in \Gamma _j\) and these gradients are different from each other in general. Similarly, \(\nabla \lambda _j\), \(\nabla \lambda _{j+1}\), \(\varphi _j^*\) and \(\varphi _{j+1}^*\) are defined as the limits to \(\Gamma _j\) from \(\Omega _j\) or \(\Omega _{j+1}\).

In accordance with the concept of the adjoint variable method, we give some constraints to \(\lambda ^*\) so that we can avoid calculating the derivatives with respect to \(\epsilon\), i.e. \(\dot{u}\) and \(\dot{q}\), in the first and last terms of Eq. 66. First, we request that \(\lambda\) should be the solution of the following inhomogeneous Helmholtz equation:

$$\begin{aligned} {{{\Delta }}}\lambda ^*(\varvec{x}) + k_i^2 \lambda ^*(\varvec{x}) = \varepsilon _i \sum _{m=0}^{M_i-1} u^*(\varvec{x}) \updelta (\varvec{x} - \varvec{z}_m^{(i)})\quad ~\mathrm{for}~ \varvec{x}\in \Omega _i. \end{aligned}$$
(67)

Second, we request that \(\lambda\) should satisfy the following boundary conditions:

$$\begin{aligned} \lambda _j^* = \lambda _{j+1}^*(=:\lambda ^*),\quad \varphi _j^*=-\varphi _{j+1}^*(=:\varphi ^*)\quad \text {on }\Gamma _j. \end{aligned}$$
(68)

Then, by substituting Eq. 67 into the two domain integrals and Eq. 68 into the first and fourth boundary integrals, we can write Eq. 66 as follows:

$$\begin{aligned} \mathcal {S}^\mathrm{I}(\Gamma )= & {} \mathop {\mathrm {Re}}\sum _{j=0}^{d-2} \Bigg [ \int _{\Gamma _j} \left( \frac{1}{\varepsilon _j} \nabla \lambda _j^* \cdot \nabla u_j - \frac{1}{\varepsilon _{j+1}} \nabla \lambda _{j+1}^* \cdot \nabla u_{j+1} \right) \varvec{v} \cdot \varvec{n} \, \mathrm {d}\Gamma \nonumber \\&- \int _{\Gamma _j} q \varvec{v} \cdot ( \nabla \lambda _j^* - \nabla \lambda _{j+1}^* ) \, \mathrm {d}\Gamma - \int _{\Gamma _j} \varphi ^* \varvec{v} \cdot (\nabla u_j - \nabla u_{j+1}) \, \mathrm {d}\Gamma \Bigg ], \end{aligned}$$
(69)

where the first domain integral in Eq. 66, which is reduced to \(u^*(\varvec{z}_m^{(i)})\varvec{v}(\varvec{z}_m^{(i)})\cdot \nabla u(\varvec{z}_m^{(i)})\) due to Eq. 67 and the property of the Dirac’s delta function, vanished because of the assumption that the observation points are fixed, i.e. \(\varvec{v}(\varvec{z}_m^{(i)})\equiv \varvec{0}\). In addition, \(\frac{k_j^2}{\varepsilon _j}-\frac{k_{j+1}^2}{\varepsilon _{j+1}}\equiv 0\) was used.

We can show that any deformation in tangential direction does not contribute to the shape derivative. Namely, if we set \(\varvec{v}=\sigma \varvec{s}\), where \(\varvec{s}\) is the unit tangential vector to \(\Gamma _j\) and \(\sigma\) is any real number, we can derive \(\mathcal {S}^\mathrm{I}\equiv 0\). As a matter of fact, since \(\varvec{v}\cdot \varvec{n}=0\) holds, the first integral in Eq. 69 vanishes. In addition, the boundary conditions \(u_j=u_{j+1}\) in Eq. 3 and \(\lambda _j^*=\lambda _{j+1}^*\) in Eq. 68 yield the continuities of the tangential derivatives, i.e. \(\varvec{s}\cdot \nabla u_j=\varvec{s}\cdot \nabla u_{j+1}\) and \(\varvec{s}\cdot \nabla \lambda _j^*=\varvec{s}\cdot \nabla \lambda _{j+1}^*\) on \(\Gamma _j\), respectively, from which the second and third integrals vanish. Consequently, we can write \(\varvec{v}\) as \((\varvec{v}\cdot \varvec{n})\varvec{n}\) in Eq. 69.

Now, we represent \(\nabla u_j\) and \(\nabla \lambda _j^*\) in Eq. 69 with their normal and tangential components, i.e.

$$\begin{aligned} \nabla u_j = \frac{\partial u_j}{\partial n} \varvec{n} + \frac{\partial u_j}{\partial s} \varvec{s},\quad \nabla \lambda _j^* = \frac{\partial \lambda _j^*}{\partial n} \varvec{n} + \frac{\partial \lambda _j^*}{\partial s} \varvec{s}. \end{aligned}$$

Then, the first term in Eq. 69 can be reduced as:

$$\begin{aligned} \int _{\Gamma _j} \left( (\varepsilon _{j} - \varepsilon _{j+1}) \varphi ^* q + \left( \frac{1}{\varepsilon _{j}} - \frac{1}{\varepsilon _{j+1}}\right) \frac{\partial \lambda ^*}{\partial s} \frac{\partial u}{\partial s} \right) \varvec{v} \cdot \varvec{n} \, \mathrm {d}\Gamma , \end{aligned}$$

while the second and third integrals can be expressed as:

$$\begin{aligned}&- \int _{\Gamma _j} q (\varvec{v}\cdot \varvec{n})\varvec{n}\cdot \left( \frac{\partial \lambda _j^*}{\partial n} \varvec{n} + \frac{\partial \lambda _j^*}{\partial s} \varvec{s} - \frac{\partial \lambda _{j+1}^*}{\partial n} \varvec{n} - \frac{\partial \lambda _{j+1}^*}{\partial s} \varvec{s} \right) \, \mathrm {d}\Gamma = - \int _{\Gamma _j} \left( \varepsilon _j - \varepsilon _{j+1} \right) \varphi ^* q \varvec{v}\cdot \varvec{n}\, \mathrm {d}\Gamma ,\\&- \int _{\Gamma _j} \varphi ^* (\varvec{v}\cdot \varvec{n})\varvec{n}\cdot \left( \frac{\partial u_j}{\partial n} \varvec{n} + \frac{\partial u_j}{\partial s} \varvec{s} - \frac{\partial u_{j+1}}{\partial n} \varvec{n} - \frac{\partial u_{j+1}}{\partial s} \varvec{s} \right) \, \mathrm {d}\Gamma = - \int _{\Gamma _j} \left( \varepsilon _j - \varepsilon _{j+1} \right) \varphi ^* q \varvec{v}\cdot \varvec{n}\, \mathrm {d}\Gamma . \end{aligned}$$

In these expressions, the boundary conditions in Eqs. 3 and 68 were used. Consequently, we can obtain the shape derivative \(\mathcal {S}^\mathrm{I}\) in Eq. 28.

1.2 Appendix C.2: Case II: energy absorption rate

We consider the objective function \(\mathcal {J}^\mathrm{II}\) in Eq. 30. Using Eqs. 56 and 59b, we can obtain the variation of \(\mathcal {J}^\mathrm{II}\) as follows:

$$\begin{aligned} \delta \mathcal {J}^\mathrm{II}:= & {} \mathcal {J}^\mathrm{II}(\tilde{\Gamma };\tilde{u},\tilde{q})-\mathcal {J}^\mathrm{II}(\Gamma ;u,q)\nonumber \\= & {} \kappa \epsilon \mathop {\mathrm {Re}}\Bigg [ \int _{\partial \Omega _\alpha } \mathrm {i}(\dot{u} q^* - u^* \dot{q}) \, \mathrm {d}\Gamma - \int _{\partial \Omega _\alpha } \mathrm {i}u^* q \big (\nabla \cdot \varvec{v} - \varvec{n} \cdot (\nabla \varvec{v} \varvec{n}) \big ) \, \mathrm {d}\Gamma \Bigg ] + \mathcal {O}(\epsilon ^2). \end{aligned}$$
(70)

Successively, we can compute the shape derivative, denoted by \(\mathcal {S}^\mathrm{II}\), by substituting \(\delta \mathcal {J}^\mathrm{II}\) in Eq. 70 and \(\delta \mathcal {G}_i\) in Eq. 60 into the definition of the shape derivative in Eq. 55. The way of the derivation is similar to that of the previous case. In this case, it is requested that the (conjugated) Lagrange multiplier \(\lambda ^*\) should be the solution of the following adjoint problem:

$$\begin{aligned}&{{{\Delta }}}\lambda ^* + k_i^2 \lambda ^* = 0 \quad \text {in} \Omega _i, \end{aligned}$$
(71a)
$$\begin{aligned}&\lambda _{\alpha -1}^* = \lambda _\alpha ^* + \mathrm {i}\kappa u_\alpha ^* \, (=: \lambda ^*), \quad \varphi _{\alpha -1}^* = - \varphi _\alpha ^* - \mathrm {i}\kappa q_\alpha ^* \, (=: \varphi ^*) \quad \text {on } \Gamma _{\alpha -1}, \end{aligned}$$
(71b)
$$\begin{aligned}&\lambda _\alpha ^* = \lambda _{\alpha +1}^* - \mathrm {i}\kappa u_\alpha ^*\, (=: \lambda ^*), \quad \varphi _\alpha ^* = - \varphi _{\alpha +1}^* - \mathrm {i}\kappa q_\alpha ^*\, (=: \varphi ^*) \quad \text {on } \Gamma _{\alpha }, \end{aligned}$$
(71c)
$$\begin{aligned}&\lambda _j^* = \lambda _{j+1}^* \, (=: \lambda ^*),\quad \varphi _j^* = - \varphi _{j+1}^* \, (=: \varphi ^*) \quad \text {on } \Gamma _{j} for j\ne \alpha -1,\alpha , \end{aligned}$$
(71d)
$$\begin{aligned}&\lambda ^*(\varvec{x} + L \varvec{e}_x) = e^{-\mathrm {i}\beta L} \lambda ^*(\varvec{x}),\quad \varphi ^*(\varvec{x} + L \varvec{e}_x) = e^{-\mathrm {i}\beta L} \varphi ^*(\varvec{x})\quad \text {on }\Gamma _j^\mathrm{L}, \end{aligned}$$
(71e)
$$\begin{aligned}&\lambda ^*(x,y)= {\left\{ \begin{array}{ll} \sum _n C_n e^{\mathrm {i}\phi _n x} e^{\mathrm {i}\psi _n (y-y_0)} &{} ~\mathrm{for}~ y>y_0,\\ \sum _n D_n e^{\mathrm {i}\phi _n x} e^{\mathrm {i}\psi _n (-y-y_0)} &{} ~\mathrm{for}~ y<-y_0, \end{array}\right. } \end{aligned}$$
(71f)

where \(u_\alpha\) and \(q_\alpha\) [which are defined in Eq. 3] in the boundary conditions Eqs. 71b and 71c are the solutions obtained from the primal problem. Finally, we can obtain \(\mathcal {S}^\mathrm{II}\) in Eq. 31.

Appendix D: Evaluation of the integral of \(\varphi ^*\dot{u}-\lambda ^*\dot{q}\) over the infinite boundaries \(\Gamma _{-1}\) and \(\Gamma _{d-1}\) in Eq. (62)

Let us consider the case of \(y>y_0\). Similarly to Eq. 56, the scattered field \(\tilde{u}^\mathrm{sca}\) for the deformed domain \(\tilde{\Omega }_0\) can be expanded in terms of \(\epsilon\) as follows:

$$\begin{aligned} \tilde{u}^\mathrm{sca}(\tilde{\varvec{x}}) = u^\mathrm{sca}(\varvec{x}) + \epsilon \dot{u}^\mathrm{sca}(\varvec{x}) + \mathcal {O}( \epsilon ^2). \end{aligned}$$

Here, \(\dot{u}^\mathrm{sca}(\varvec{x}):=\left. \frac{\mathrm {d}u^\mathrm{sca}(\tilde{\varvec{x}})}{\mathrm {d}\epsilon }\right| _{\epsilon =0}\) can be rewritten from the radiation condition of \(u^\mathrm{sca}\) in Eq. 6 as follows:

$$\begin{aligned} \dot{u}^\mathrm{sca}(x,y)=\left. \frac{\mathrm {d}}{\mathrm {d}\epsilon }\sum _m A_m e^{\mathrm {i}\xi _m (x+\epsilon v_x)} e^{\mathrm {i}\eta _m (y+\epsilon v_y-y_0)}\right| _{\epsilon =0} =\sum _m A_m \mathrm {i}(\xi _m v_x + \eta _m v_y) e^{\mathrm {i}\xi _m x}e^{\mathrm {i}\eta _m (y-y_0)}, \end{aligned}$$

where \(\epsilon \varvec{v}=(\epsilon v_x,\epsilon v_y)^T\) is the infinitesimal vector defined in Eq. 47. By comparing with the radiation condition of \(u^\mathrm{sca}\) in Eq. 6, this equation means that \(\dot{u}^\mathrm{sca}\) also satisfies the radiation condition. Here, \(\dot{u}^\mathrm{sca}\) is equivalent to \(\dot{u}(\equiv \dot{u}^\mathrm{inc}+\dot{u}^\mathrm{sca})\) because the incident field is irrelevant to the deformation, i.e. \(\tilde{u}^\mathrm{inc}(\tilde{\varvec{x}})\equiv u^\mathrm{inc}(\varvec{x})\Leftrightarrow \dot{u}^\mathrm{inc}\equiv 0\). Therefore, \(\dot{u}\) satisfies the following radiation condition:

$$\begin{aligned} \dot{u}(x,y)=\sum _m E_m e^{\mathrm {i}\xi _m x}e^{\mathrm {i}\eta _m (y-y_0)}, \end{aligned}$$
(72)

where \(E_m:=A_m \mathrm {i}(\xi _m v_x + \eta _m v_y)\).

With using the radiation condition for \(\dot{u}\) in Eq. 72 as well as \(\varphi ^*\) in Eq. 64, we can write down the integrand of the first boundary term in Eq. 62 regarding the upper infinite boundary \(\Gamma _{-1}\), where the outward normal is \((0,1)^T\), as follows:

$$\begin{aligned} \varphi ^*\dot{u}-\lambda ^*\dot{q} =\frac{1}{\varepsilon _0}\frac{\partial \lambda ^*}{\partial y}\dot{u}-\lambda ^*\frac{1}{\varepsilon _0}\frac{\partial \dot{u}}{\partial y}=\frac{1}{\varepsilon _0}\sum _n \sum _m C_n E_m e^{\mathrm {i}\frac{2\pi (m+n)}{L}x} e^{\mathrm {i}(\psi _n + \eta _m)(y-y_0)} \mathrm {i}(\psi _n-\eta _m). \end{aligned}$$

Then, the terms such as \(n+m\ne 0\) vanish when they are integrated over \(\Gamma _{-1}\) (i.e. \(x\in [0,L]\) for any \(y>y_0\)). Meanwhile, the terms such as \(n+m=0\) also vanish because \(n+m=0\) is followed by \(\phi _n=\phi _{-m}=-\beta -\frac{2\pi m}{L}=-\xi _m\) and thus \(\psi _n=\sqrt{k^2-(-\xi _m)^2}=\eta _m\). Therefore, the integral of \(\varphi ^*\dot{u}-\lambda ^*\dot{q}\) over \(\Gamma _{-1}\) vanishes and does not contribute to \(\mathcal {S}^\mathrm{I}\) in Eq. 62.

Similarly, the integral of \(\varphi ^*\dot{u}-\lambda ^*\dot{q}\) over the lower infinite boundary \(\Gamma _{d-1}\) results in zero by starting with \(y<-y_0\) instead of \(y>y_0\).

Appendix E: Energy absorption rate \(\mathcal {A}\)

The time-averaged energy per unit time that absorbed in a closed domain \(\Omega _\alpha\) (where \(\alpha \ne 0\) and \(n-1\)) is defined by

$$\begin{aligned} e_\mathrm{absorp} := - \int _{\Omega _\alpha } \mathrm {div} \langle {\varvec{S}}\rangle \, \mathrm {d}\Omega = - \int _{\partial \Omega _\alpha } \langle {\varvec{S}}\rangle \cdot \varvec{n} \, \mathrm {d}\Gamma , \end{aligned}$$
(73)

where the Gauss’s divergence theorem was used, \(\varvec{n}\) denotes the unit outward normal to the boundary of \(\Omega _\alpha\), \(\varvec{S}(\varvec{x}, t):=\varvec{E}(\varvec{x}, t)\times \varvec{H}(\varvec{x}, t)\) denotes the (real-valued) Poynting vector, and \(\langle {\varvec{S}}\rangle\) denotes the time-average of \(\varvec{S}\) over the period \(T:=\frac{2\pi }{\omega }\), i.e. \(\langle {\varvec{S}(\varvec{x})}\rangle := \frac{1}{T} \int _0^T \varvec{S}(\varvec{x},t) \mathrm {d}t\). Since we suppose the TM-mode, where the magnetic field \(\varvec{H}\) is given by Eq. 1, the corresponding electric field \(\varvec{E}\) in \(\Omega _\alpha\) is represented as:

$$\begin{aligned} \varvec{E}(\varvec{x},t)=\left( \mathop {\mathrm {Re}}\left[ \frac{-1}{\mathrm {i}\omega \upepsilon _0\varepsilon _\alpha }\frac{\partial u(\varvec{x})}{\partial x_2}e^{-\mathrm {i}\omega t}\right] ,\mathop {\mathrm {Re}}\left[ \frac{1}{\mathrm {i}\omega \upepsilon _0\varepsilon _\alpha }\frac{\partial u(\varvec{x})}{\partial x_1}e^{-\mathrm {i}\omega t}\right] ,0\right) ^T \equiv \mathop {\mathrm {Re}}\left[ \hat{\varvec{E}}(\varvec{x})e^{-\mathrm {i}\omega t}\right] , \end{aligned}$$

where \(\upepsilon _0\) denotes the permittivity in vacuum (\(\approx\)air)Footnote 17 and \(\hat{\varvec{E}}(\varvec{x}):=\left( \frac{-1}{\mathrm {i}\omega \upepsilon _0\varepsilon }\frac{\partial u(\varvec{x})}{\partial x_2},\frac{1}{\mathrm {i}\omega \upepsilon _0\epsilon }\frac{\partial u(\varvec{x})}{\partial x_1},0\right) ^T\in \mathbb {C}^{3}\). Similarly to \(\hat{\varvec{E}}\), we define \(\hat{\varvec{H}}(\varvec{x}):=\left( 0,0,u(\varvec{x})\right) ^T\in \mathbb {C}^{3}\) so that \(\varvec{H}(\varvec{x},t)\equiv \mathop {\mathrm {Re}}[\hat{\varvec{H}}(\varvec{x})e^{-\mathrm {i}\omega t}]\) holds. Subsequently, we obtain \(\langle {\varvec{S}}\rangle = \frac{1}{2}\left( \mathop {\mathrm {Re}}[\hat{\varvec{E}}]\times \mathop {\mathrm {Re}}[\hat{\varvec{H}}]+\mathfrak {I}[\hat{\varvec{E}}]\times \mathfrak {I}[\hat{\varvec{H}}]\right)\). By substituting this into Eq. 73 and doing some calculus by means of the Maxwell equations, \(e_\mathrm{absorp}\) can be expressed as follows:

$$\begin{aligned} e_\mathrm{absorp} = - \frac{\mathrm {i}}{4 \omega \upepsilon _0} \int _{\partial \Omega _\alpha } (u q^* - u^* q) \mathrm {d}\Gamma , \end{aligned}$$
(74)

where \(q:=\frac{1}{\varepsilon _\alpha }\frac{\partial u}{\partial n}\), which was defined in Eq. 7. Note that, since u satisfies the quasi-periodicity in Eq. 5 and, thus, \(uq^*-u^*q(\equiv u\frac{\mathrm {d}u^*}{\mathrm {d}x}-u^*\frac{\mathrm {d}u}{\mathrm {d}x})\) on \(\Gamma _\alpha ^\mathrm{R}\) is identical to that on \(\Gamma _\alpha ^\mathrm{L}\) except the sign, the integral over \(\Gamma _\alpha ^\mathrm{L}\cup \Gamma _\alpha ^\mathrm{R}\) vanishes. Therefore, the integral over \(\partial \Omega _\alpha\) is equivalent to that over \(\Gamma _{\alpha -1}\cup \Gamma _\alpha\) in net.

To normalise \(e_\mathrm{absorp}\) in Eq. 74, we consider the time-averaged energy of the incident planewave \(u^\mathrm{inc}\) (see Eq. 4) through the effective length of \(L\sin \theta\) per unit time, i.e.

$$\begin{aligned} e_\mathrm{inc}:=\frac{1}{2}\sqrt{\frac{{\mu }_0}{\upepsilon _0}}\left| a^\mathrm{inc}\right| ^2 L\sin \theta , \end{aligned}$$

where \({\mu }_0\) denotes the permeability in vacuum.

Consequently, we define the (normalised) energy absorption rate \(\mathcal {A}\) as:

$$\begin{aligned} \mathcal {A}:=\frac{e_\mathrm{absorp}}{e_\mathrm{inc}}=\frac{\mathrm {i}\kappa }{2} \int _{\partial \Omega _\alpha } (u q^* - u^* q)\mathrm {d}\Gamma = \kappa \mathfrak {I}\left[ \int _{\partial \Omega _\alpha } u q^* \mathrm {d}\Gamma \right] \end{aligned}$$
(75)

where \(\kappa :=-\frac{1}{kL\sin \theta \left| a^\mathrm{inc}\right| ^2}\). Note that \(k:=\frac{\omega }{c}\) denotes the wavenumber in vacuum, which corresponds to \(k_0\), i.e. the wavenumber in \(\Omega _0\) (to which the incident wave is given), in the main text. Note also that \(0\le \mathcal {A}\le 1\) holds if \(\mathfrak {I}\varepsilon \ge 0\) because of the conservation of the energy.

Appendix F: Lists of optimal control points

Table 1 shows the control points that can maximise the objective function \(\mathcal {A}\) in the first and second tries of the demonstration described in Subsect. 4.3.

Table 1 Design variables (i.e. coordinates of the sixteen control points) with side constraints \((x_\nu ,y_\nu )^T(\equiv \varvec{p}_\nu ^{(1)})\) obtained by the first try (left) and second try (right)

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Takahashi, T., Yamamoto, T., Shimba, Y. et al. A framework of shape optimisation based on the isogeometric boundary element method toward designing thin-silicon photovoltaic devices. Engineering with Computers 35, 423–449 (2019). https://doi.org/10.1007/s00366-018-0606-6

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00366-018-0606-6

Keywords

Navigation