Setup
  circuit: Fig. 4's: couplers k1 to k9 and two pairs of rings between MZI arms of phases φ1 to φ8, lossless, solved as 2 × 2 transfer matrices and by photonoxide's circuit solve
  k1 to k9: 0.29, 0.097747, 1, 0.097747, 0.7617, 0.7715, 0.9142, 0.7715, 0.7617
  φ1 to φ8: 0π, 1π, 0π, 0π, 0π, 1π, 0π, 0π
  designs: elliptic low-passes of order 4, passband edge 0.5π, 2 dB ripple and 40 dB attenuation; 1 dB ripple and 60 dB attenuation
  frequencies: ω the rings' round-trip phase: 501 over 0 to π for the optimizer's target, 20000 for a response's ripple and attenuation
  grid: none: closed forms and the circuit solve, exact
  solver: Nelder–Mead on the paper's Eq. 12 (weights 1 and 100) from Fig. 4's values, 20000 evaluations a target
  tolerance: each design at its edges to 1e-8 of A_p and A_s; the 2 × 2 model against the circuit to 1e-12

Wang et al. 2022's double-ring-loaded MZI filter: the fourth-order elliptic low-passes, derived here, against its circuit, exact

Elliptic, order 4, 2 dB ripple, 40 dB attenuation, passband edge 0.5π: stopband edge 0.606941π
  b: 0.0925, 0.2221, 0.3002, 0.2221, 0.0925
  a: 1.0000, -0.7992, 1.2899, -0.6645, 0.3438
  response at ω_p, dB (40 dB)           -2.000000   −A_p -2.000000  ± 2e-8       ok
  response at ω_s, dB (40 dB)          -40.000000   −A_s -40.000000 ± 4e-7       ok

Elliptic, order 4, 1 dB ripple, 60 dB attenuation, passband edge 0.5π: stopband edge 0.754271π
  b: 0.0702, 0.2385, 0.3403, 0.2385, 0.0702
  a: 1.0000, -0.7055, 1.0843, -0.5410, 0.2367
  response at ω_p, dB (60 dB)           -1.000000   −A_p -1.000000  ± 1e-8       ok
  response at ω_s, dB (60 dB)          -60.000000   −A_s -60.000000 ± 6e-7       ok

Eq. 10 as printed: 1.02 dB of ripple, 20.00 dB of attenuation from the 40 dB design's stopband edge; 20.00 dB at π
  Eq. 11's sections multiplied: b 0.1170, 0.1705, 0.2674, 0.1705, 0.1170, a 1.0000, -0.3396, 0.9775, -0.3142, 0.0708; their peak -4.38 dB. Printed, not checked

The circuit (Fig. 4's values) against photonoxide's circuit solve of it as a netlist, 2 × 2 at 3 frequencies: up to 3e-16
  largest |ΔT|                           0.000000   round-off 0.000000   ± 1e-12 ok
  Fig. 4's circuit: 1.38 dB of ripple, 19.26 dB of attenuation from the 40 dB design's stopband edge (paper, Fig. 5: "around 34 dB extinction ratio"). Printed, not checked

Nelder–Mead from Fig. 4's values on Eq. 12, weights 1 and 100, 20000 evaluations. Printed, not checked
  towards 40 dB and 2 dB of ripple: 6.94 dB of ripple, 39.62 dB of attenuation, Eq. 12 at 68.649 (Fig. 6: the 40 dB filter)
    k1–k9 0.2553, 0.1888, 0.9919, 0.2944, 0.4702, 0.6848, 0.7228, 0.7068, 0.1995; φ1–φ8 -0.5809, 2.2721, -0.7609, -0.4831, -0.4936, 2.0512, -0.2186, -0.9211
  towards 60 dB and 1 dB of ripple: 1.64 dB of ripple, 59.44 dB of attenuation, Eq. 12 at 39.937 (Fig. 7: 60 dB, 1 dB of ripple)
    k1–k9 0.1108, 0.2309, 1.0000, 0.2010, 0.5064, 0.7583, 0.5158, 0.7466, 0.3327; φ1–φ8 0.1630, 3.3376, 0.2289, 0.1361, 0.3816, 3.4437, 0.3530, 0.3147
all 5 within tolerance of the paper
