1 Appendix A 2 Analytical Solutions For Atmospheric Radiative Equilibrium State 3 4 A.1 Infrared (IR) Source Function in Radiative Equilibrium 5 6 The general equation of radiative transfer is: 7 π ππΌ =πΌ−π΅ ππ (π΄. 1) 8 where πΌ is the IR radiation field (intensity), π΅ is the atmospheric source function, π is the 9 optical depth and π is the cosine of the view angle. Under Eddington approximation: 10 πΌ(π) = πΌ0 + πΌ1 π, the moments are: 11 πΌ Μ = ∫ πΌ(π) ππ ππ = πΌ0 12 πΉ = ∫ ππΌ ππ ππ = 4π πΌ 3 1 (π΄. 2π) 13 πΎ = ∫ π 2 πΌ ππ ππ = 4π πΌ 3 0 (π΄. 2π) 14 (π΄. 2π) Substituting into Equation (A.1) and we obtain: 15 4π ππ΅ π2πΉ = 3πΉ − 2 ππ ππ (π΄. 3) 16 The upper boundary condition of thermal source function π΅(π) can be estimated from the 17 two-stream approximation. Suppose we only have one upward stream πΌ + , and one 18 downward stream πΌ − : 19 πΉ = π(πΌ + − πΌ − ) πΌΜ = 20 21 22 23 24 25 πΌ+ + πΌ− 2 (π΄. 4π) (π΄. 4π) So Equation (A. 3) becomes: ππΉ(π) = 2πΉ(π) + 4ππΌ − (π) − 4ππ΅(π) ππ (π΄. 5) If we assume no downward IR flux (πΌ − ) at the upper boundary, we obtain: π΅(0) = πΉ(0) 1 ππΉ − ( ) 2π 4π ππ π=0 The radiative equilibrium state requires: 1 (π΄. 6) 26 ππΉ ππΉβπππ‘ =− = −π(π) ππ ππ 27 where πΉ(π) is the thermal IR flux emitted by the atmosphere and πΉβπππ‘ is the heat flux 28 from outside heat source. π(π) = ππΉβπππ‘ /ππ is called the heating rate. The primary heat 29 flux is from the downward solar flux πΉπ , which is absorbed in the visible or near IR 30 wavelengths so that it does not exchange the photons with the thermal IR wavelengths. 31 On the other hand, there could be an upward heat flux πΉβ¨ , either from (1) the surface that 32 is heated by absorbing the sunlight directly; or (2) the radioactive decay of the heavy 33 elements in the crust of terrestrial plants; or (3) the upward convective heat flux from the 34 deep interior of giant planets. πΉβ¨ is mainly in the thermal IR region and usually treated 35 as the lower boundary condition of the upward flux. (π΄. 7) 36 37 Here we discuss two cases: 38 (a) For a certain incident angle, solar Flux at ππ£ is: 39 πΉπ − (π, ππ£ ) = −ππΉπ π π − π£ (π΄. 8) π 40 where ππ£ is the optical depth at visible (or near IR) wavelength π£ where the solar flux can 41 be absorbed. π is the cosine of the solar incident angle π = πππ π, πΉπ is the solar flux at 42 the top of the atmosphere (TOA). Negative sign means downward flux. The radiative 43 equilibrium thermal flux πΉπΌπ = −πΉβπππ‘ = πΉβ¨ + ππΉπ π 44 depth in the thermal emission wavelengths and πΌ = ππ /π. From Equation (A. 3), we solve 45 the π΅(π): 46 47 π΅(π) = − πΌπ π , where π is the averaged optical 3πΉβ¨ 2 3ππΉπ 2 π πΌ π −πΌπ ( + π) + [ + + ( − )π π ] 4π 3 4π 3 πΌ 3π πΌ (π΄. 9) This is basically the same as the result in Goody and Yung (1995). 48 49 (b) For a global averaged solar flux (including the diurnally average): π ∫02 πΉπ£− (π, ππ£ ) 2ππ 2 π πππππ 1 − Μ Μ Μ Μ πΉ (ππ£ ) = = − πΉ πΈ (π ) 4ππ 2 2 π 3 π£ 50 ∞ π −π₯π‘ ∞ π₯ π−1 π −π‘ (π΄. 10) 51 where πΈπ (π₯) = ∫1 52 this integral should be understood in terms of the Cauchy principal value, due to the π‘π ππ‘ = ∫π₯ π‘π ππ‘ is the exponential integral, if π₯ is negative, 2 53 singularity in the integrand at zero. When ππ = 0 (TOA), Μ Μ Μ Μ πΉ − (ππ£ ) = πΉπ /4. The radiative 54 equilibrium thermal flux πΉπΌπ = −πΉβπππ‘ = πΉβ¨ + 2 πΉπ πΈ3 (πΌπ). From Equation (A. 2), we 55 solve the π΅(π): 56 1 π΅(π) = 3πΉβ¨ 2 3πΉπ 1 + πΌ πΌ 1 ( + π) + [ + πΈ2 (πΌπ) − πΈ (πΌπ)] 4π 3 4π 6πΌ 6 2πΌ 4 (π΄. 11) 57 58 A. 2. Analytical Solutions For Radiance, Flux, Heating and Cooling Rate 59 60 Based on the radiative equilibrium IR source functions above, we derive the analytical 61 solutions for the TOA radiance, upward and downward fluxes for each layer, and solar 62 heating and thermal cooling rates for each layer. The source function π΅(π) could have 63 one of the following forms, corresponding to the above cases (a) and (b), respectively: 64 1. π΅(π) = π΅0 + π΅1 π + π΅2 π −πΌπ 65 2. π΅(π) = π΅0 + π΅1 π + π΅2 πΈ2 (πΌπ) + π΅3 πΈ4 (πΌπ) 66 We will focus on the form 1 because it is less complicated than the form 2 and therefore 67 the analytical solutions can be obtained through the integrations. Form 1 can describe the 68 typical tropospheric temperature and stratospheric temperature with or without inversion. 69 In the following notation, we will ignore the wavenumber index in the lower subscript 70 but all the formulae should be considered monochromatic and can be used for testing the 71 numerical line-by-line (LBL) radiative transfer models. Please see Appendix D for the 72 details of the integration derivations. 73 74 A.2.1. TOA Radiance 75 76 77 The IR upward radiance at emission angle π at ππ : πΌ + (ππ , π) = π΅(ππ )π π −π − π π π ππ + ∫ π΅(π‘)π ππ − π‘−ππ π ππ‘ π (π΄. 12) 78 where ππ is the total optical depth from the upper to the lower boundary. π = πππ π. 79 IR downward radiance at ππ : 80 πΌ − (π ππ π , π) = ∫ π΅(π‘)π 0 3 π‘−ππ π ππ‘ π (π΄. 13) 81 We are more interested the TOA IR radiance (ππ = 0) that can be measured by the 82 satellite. The TOA IR radiance at emission angle π: 83 πΌ + (0, π) = π΅(ππ π − π )π π ππ + ∫ π΅(π‘)π − 0 84 π‘ π ππ‘ π (π΄. 14) The global averaged TOA IR radiance: π ππ ∫02 πΌ + (0, π) 2ππ 2 π πππππ + Μ Μ Μ πΌ (0, π) = = π΅(ππ )πΈ2 (ππ ) + ∫ π΅(π‘)πΈ1 (π‘)ππ‘ (π΄. 15) 2ππ 2 0 85 86 87 Plug in the source function π΅(π) = π΅0 + π΅1 π + π΅2 π −πΌπ , we obtain: πΌ + (0, π) = (π΅0 + π΅1 ππ + π΅2 π −πΌππ π − π )π π ππ + ∫ (π΅0 + π΅1 π‘ + π΅2 π −πΌπ‘ )π − π‘ π 0 88 = π΅0 + π΅1 π(1 − π π − π π π΅ 2 ) + πΌπ+1 (1 + πΌππ 1 π −(πΌ+ )ππ ππ‘ π ) (π΄. 16) 89 and 90 Μ Μ Μ πΌ + (0, π) = (π΅0 + π΅1 ππ + π΅2 π −πΌππ )πΈ2 (ππ ) + ∫ (π΅0 + π΅1 π‘ + π΅2 π −πΌπ‘ )πΈ1 (π‘)ππ‘ ππ 0 91 92 1 π΅2 = π΅0 + π΅1 ( − πΈ3 (ππ )) + [ln(1 + πΌ) − π −πΌππ πΈ1 (ππ ) + πΈ1 ((1 + πΌ)ππ ) 2 πΌ −πΌππ + πΌπ πΈ2 (ππ )] (π΄. 17) 93 94 A.2.2. Upward and Downward Fluxes 95 96 From the downward solar flux in Eqn. (A.8), the global-averaged solar flux at ππ£ is: π ∫02 πΉπ£− (π, ππ£ ) 2ππ 2 π πππππ 1 − Μ Μ Μ Μ πΉ (ππ£ ) = = − πΉπ πΈ3 (ππ£ ) 2 4ππ 2 97 (π΄. 18) 98 From Eqn. (A.12), the IR upward flux at layer ππ is 99 πΉ + (ππ ) = ∫ ππΌ + (ππ , π) ππππ = 2ππ΅(ππ )πΈ3 (ππ − ππ ) + 2π ∫ π΅(π‘)πΈ2 (π‘ − ππ )ππ‘ (π΄. 19) 1 ππ 0 100 ππ From Eqn. (A.13), the IR downward flux at layer ππ : 1 101 πΉ − (π π) = ∫ ππΌ − (π ππ π , π) ππππ 0 102 = 2π ∫ π΅(π‘)πΈ2 (ππ − π‘)ππ‘ 0 Plug in the source function π΅(π) = π΅0 + π΅1 π + π΅2 π −πΌπ , we obtain: 4 (π΄. 20) 103 104 105 πΉ + (ππ ) 1 ππ = ππ΅0 + 2ππ΅1 ( + − πΈ4 (ππ − ππ )) 3 2 −πΌππ 2ππ΅2 π + {π −πΌ(ππ −ππ ) [(πΈ1 (ππ − ππ ) − πΌπΈ2 (ππ − ππ ) + πΌ 2 πΈ3 (ππ − ππ )] πΌ2 106 − πΈ1 ((1 + πΌ)(ππ − ππ )) − ln(1 + πΌ) + πΌ} 107 and 108 πΉ − (ππ ) 109 110 (π΄. 21) 1 1 ππ = 2ππ΅0 ( − πΈ3 (ππ )) + 2ππ΅1 (− + + πΈ4 (ππ )) 2 3 2 −πΌππ 2ππ΅2 π + {π πΌππ [(πΈ1 (ππ ) + πΌπΈ2 (ππ )] − πΈ1 ((1 − πΌ)ππ ) − ln(|πΌ − 1|) − πΌ} 2 πΌ (π΄. 22) 111 112 A.2.3. Solar Heating Rate 113 114 The Solar heating Rate at ππ£ is: 115 116 π»βπππ‘ (ππ£ ) = π ππΉπ − (π, ππ£ ) − π£ = πΉπ π π ππ (π΄. 23) And the global-averaged solar heating rate at ππ£ : − (π ) Μ Μ Μ Μ ππΉ 1 π£ = πΉπ πΈ2 (ππ£ ) ππ 2 117 Μ Μ Μ Μ Μ Μ Μ π»βπππ‘ (ππ£ ) = 118 But the zonally averaged solar heating rate at ππ£ is not integrable. (π΄. 24) 119 120 A.2.4. Thermal Cooling Rate 121 122 123 124 125 The IR Cooling Rate at ππ : π»ππππ = ππΉ(ππ ) = 2π[−π΅(ππ )πΈ2 (ππ ) + (π΅(ππ )−π΅(ππ ))πΈ2 (ππ − ππ ) + πΌ1 + πΌ2 ] ππ where ππ πΌ1 = ∫ (π΅(π‘) − π΅(ππ ))πΈ1 (π‘ − ππ )ππ‘ ππ 126 ππ πΌ2 = ∫ (π΅(π‘) − π΅(ππ ))πΈ1 (ππ − π‘)ππ‘ 0 5 (π΄. 25) 127 Plug in the source function π΅(π) = π΅0 + π΅1 π + π΅2 π −πΌπ , we obtain: 128 π»ππππ = 129 = −2ππ΅0 πΈ2 (ππ ) + 2ππ΅1 (πΈ3 (ππ ) − πΈ3 (ππ − ππ )) 130 + 131 − πΈ1 ((1 − πΌ)ππ )−π −πΌ(ππ −ππ ) πΈ1 (ππ − ππ ) + πΈ1 ((1 + πΌ)(ππ − ππ )) 132 + ln(1 + πΌ) − ln(|πΌ − 1|) − 2πΌ} ππΉ(ππ ) ππ 2ππ΅2 π −πΌππ {πΌπ −πΌ(ππ −ππ ) πΈ2 (ππ − ππ ) + π πΌπ πΈ1 (ππ ) πΌ (π΄. 26) 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 6 156 Appendix B 157 Numerical Schemes for Flux and Cooling Rate Calculations 158 159 We consider two numerical schemes for the calculations of upward and downward 160 intensities, fluxes and cooling rate. The first method is the direct integral with Gaussian 161 Quadrature method. But usually it needs larger number of Gaussian points to achieve the 162 satisfactory accuracy. Our second method is the finite difference scheme. In the following 163 notation, we ignore the wavenumber index in the lower subscript but all the formulae 164 should be considered monochromatic and line-by-line (LBL) radiative transfer 165 calculations. Using optical depth as the vertical coordinate, We divide the atmosphere 166 into π layers with π + 1 atmospheric levels as layer boundaries, noted by ππ , where π = 167 1, … , π + 1. The source function π΅π = π΅(ππ ) is defined on the levels. Within each layer 168 [ππ , ππ+1 ], we approximate the source function linearly as function of optical depth: 169 π΅π (π‘) = π΅π + π΅π′ (π‘ − ππ ) , where ππ ≤ π‘ ≤ ππ+1 , where π΅π′ = (π΅(ππ+1 ) − π(ππ ))/βπ and 170 βπ = ππ+1 − ππ . 171 172 The upward intensity π°+ π (ππ , π): π 173 πΌ + (π π , π) = π΅π+1 π −(ππ+1 −ππ )/π ππ+1 + ∑∫ π΅π (π‘)π −(π‘−ππ )/π π=1 ππ ππ‘ π π 174 = π΅π+1 π −(ππ+1 −ππ )/π + ∑ π −(ππ −ππ )/π [(π΅π + π΅π′ π)(1 − π −βπ/π ) − π΅π′ βππ −βπ/π ] (π΅. 1) π=1 175 176 Specifically, the TOA emission is: π 177 πΌ + (0, π) = π΅π π −ππ /π + ∑ π −ππ/π [(π΅π + π΅π′ π)(1 − π −βπ/π ) − π΅π′ βππ −βπ/π ] (π΅. 2) π=1 178 179 The downward intensity π°− π (ππ , π): π−1 180 πΌ − (ππ , π) = ∑ π −(ππ−ππ )/π [π΅π (π βπ/π − 1) + π΅π′ π( π=1 7 βπ βπ/π π − π βπ/π + 1)] π (π΅. 3) 181 182 The upward flux π↑ (ππ ): 183 ↑ (π π πΉ π) = 2ππ΅π+1 πΈ3 (ππ+1 − ππ ) + 2π ∑ πΌππ↑ (π΅. 4) π=π 184 where πΌππ↑ = π΅π [πΈ3 (ππ − ππ ) − πΈ3 (ππ+1 − ππ )] − π΅π′ [(ππ+1 − ππ )πΈ3 (ππ+1 − ππ ) − 185 πΈ4 (ππ − ππ ) + πΈ4 (ππ+1 − ππ )] 186 187 If βπ < 10−4 , we use an approximated formula to minimize the numerical errors: 188 πΌππ↑ = 3π΅π − π΅π+1 πΈ2 (ππ+1 − ππ )βπ 2 189 190 The downward flux π↓ (ππ ): π−1 191 πΉ ↓ (π π) = 2π ∑ πΌππ↓ (π΅. 4) π=1 192 where πΌππ↓ = π΅π [πΈ3 (ππ − ππ+1 ) − πΈ3 (ππ − ππ )] + π΅π′ [(ππ+1 − ππ )πΈ3 (ππ − ππ+1 ) + 193 πΈ4 (ππ − ππ ) − πΈ4 (ππ − ππ+1 )] 194 195 If βπ < 10−4 , we use: 196 πΌππ↓ = π΅π + π΅π+1 πΈ2 (ππ − ππ+1 )βπ 2 197 198 The cooling rate can be calculated in two ways. 199 200 201 (1) The cooling rate defined at level ππ is: π»ππππ π π−1 π=π π=1 ππΉ(ππ ) = = 2π[π΅π+1 πΈ2 (ππ+1 − ππ ) − 2π΅π + ∑ πΌ1 + ∑ πΌ2 ] ππ 202 where πΌ1 = π΅π [πΈ2 (ππ − ππ ) − πΈ2 (ππ+1 − ππ )] − π΅π′ [(ππ+1 − ππ )πΈ2 (ππ+1 − ππ ) − 203 πΈ3 (ππ − ππ ) + πΈ3 (ππ+1 − ππ )] 204 205 If βπ < 10−4 (and π ≠ π − 1), we use: 8 (π΅. 5) 206 πΌ1 = 3π΅π − π΅π+1 πΈ1 (ππ+1 − ππ )βπ 2 207 208 and πΌ2 = π΅π [πΈ2 (ππ − ππ+1 ) − πΈ2 (ππ − ππ )] + π΅π′ [(ππ+1 − ππ )πΈ2 (ππ − ππ+1 ) + 209 πΈ3 (ππ − ππ ) − πΈ3 (ππ − ππ+1 )] 210 211 If βπ < 10−4 (and π ≠ π − 1), we use: 212 πΌ2 = π΅π + π΅π+1 πΈ1 (ππ − ππ+1 )βπ 2 213 214 215 (2) The averaged cooling rate within layer [ππ , ππ+1 ] is: π»ππππ = ππΉ(ππ ) = [πΉ ↓ (ππ ) + πΉ ↑ (ππ+1 ) − πΉ ↓ (ππ+1 ) − πΉ ↑ (ππ )]/(ππ+1 − ππ ) ππ (π΅. 6) 216 217 Comparison Between Analytical and Numerical Calculations 218 219 In order to test the accuracy of the two numerical schemes, we applied three typical 220 source function profiles to the models: 221 222 (1) π΅(π) = 3 + 0.2 π 223 (2) π΅(π) = 5 + 0.05 π + 5 exp(−0.5 π) 224 (3) π΅(π) = 1 + 0.01 π + 15 exp(−0.9 π) 225 226 For the Gaussian Quadrature method we use 100 Gaussian points. The model has 72 227 levels with optical depth evenly distributed logarithmically from 10-5 to 102. Fig. B1 228 shows the three source function profiles and the corresponding averaged IR cooling rates 229 within each layer, indicated by the middle point of the vertical grid. The comparisons 230 with analytical solutions show that the two integration methods are roughly comparably 231 accurate and the relative errors compared with the analytical solutions are generally less 232 than 1 percent. The differences are larger in the larger optical depth region because (1) 233 the cooling rate is approaching zero, thus limited by the computational precision; (2) the 234 optical grid is coarser, leading to larger numerical errors. For instance, the 720-level 9 235 model agrees much better than the 72-level model. The cooling rate for case (1) has the 236 least errors because the source function is linear in π, in consistent with the assumption in 237 our numerical methods within each grid box. When the non-linear terms are more 238 significant, the relative errors become larger. Fig. B2 shows the averaged IR cooling rates 239 defined at levels and the associated errors. Fig. B3 and B4 are showing the upward and 240 downward thermal fluxes at each level, respectively. The relative difference of the 241 numerical values and analytical solutions are generally within 1 percent. 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 10 266 Appendix C 267 Non-LTE Formulism 268 269 In this study we consider a simple two-level non-LTE radiative transfer between a 270 vibrational level and the ground level. We ignore the interaction between the vibrational 271 levels, especially the excitation of the low-lying fundamentals (e.g., in the thermal IR 272 wavelength) by the higher vibrational levels (i.e., the vibrational-vibration transfer, or 273 “V-V transfer”). In the absence of the vibrational chemistry, collisions will be the only 274 cause of the departure of the source function from LTE through the vibrational- 275 translational transfer (“V-T transfer”). The non-LTE source function, π½ν at wavenumber 276 ν, can be formulated as (see Appleby, 1990; Yelle et al., 1991; Goody and Yung, 1995; 277 López-Puertas and Taylor, 2001): 278 π½ν (π) = πΌ (Μ π) + π(π)π΅ν (π) 1 + π(π) (πΆ. 1) 279 where ε ≡ C10 /A10 , C10 is the de-activation rates due to thermal collisions and A10 is the 280 Einstein coefficient of the vibrational band. π΅ν is the LTE source function (Planck 281 function). πΌ Μ is the mean radiation field averaged in the vibrational band: πΌ (Μ π) = 282 ∫ πΌνΜ (π)πν πν ∫ πν πν (πΆ. 2) 283 where πΌνΜ (π) is the mean intensity averaged for all the emission angles (Goody and Yung, 284 2.103): 285 πΌνΜ (π) = 1 1 ππ π΅ν (ππ )πΈ2 (ππ − π) + ∫ π½ν (π‘)πΈ1 (|π‘ − π|)ππ‘ 2 2 0 (πΆ. 3) 286 where π is the optical depth. πΈ1 (π₯) and πΈ2 (π₯) are exponential integral functions. For gas 287 giant planets, we need to include the continuum effect from H2-H2 and H2-He CIA that is 288 assumed to be always in LTE. The source function is modified to include this 289 partitioning: 290 π½ν (π) = 1 − π(π) [πΌ (Μ π) + ε(π)π΅ν (π)] + π(π)π΅ν (π) 1 + ε(π) (πΆ. 4) 291 where π(π) = ππ»2 −π»2 /(ππππ + ππ»2 −π»2 ) is the ratio of the continuum optical depth to the 292 total optical depth. 11 293 Substituted in Eqn. (C.3) and we obtain the equation for the mean radiation field: 294 πΌνΜ (π) 295 = 296 1 ππ 1 − π(π‘) ∫ πΌνΜ (π‘)πν πν + ∫ { [ + ε(t)π΅ν (π‘)] + π(π‘)π΅ν (π‘)} πΈ1 (|π‘ 2 0 1 + ε(t) ∫ πν πν 297 − π|)ππ‘ 298 Once we solve πΌνΜ (π), and the source function π½ν (π) can be derived. We now introduce 299 two numerical schemes to solve πΌνΜ (π). 1 π΅ (π )πΈ (π − π) 2 ν π 2 π (πΆ. 5) 300 301 C.1 Iteration Method 302 303 In the iteration method, first we write the equations in the discretized format: π½ν (π) = 304 1 − π(π) ∑ πΌνΜ (π)πν Δν [ + ε(π)π΅ν (π)] + π(π)π΅ν (π) ∑ πν Δν 1 + ε(π) (πΆ. 6π) π π½ν (ππ ) + π½ν (ππ+1 ) 1 1 πΌΜ ν (π) = π΅ν (ππ )πΈ2 (ππ − π) + ∑ |πΈ2 (|π − ππ |) − πΈ2 (|π − ππ+1 |)| 2 2 2 { π=1 305 Start from πΌΜ ν (π) = ∫ π΅ν (π)πν πν / ∫ πν πν and iterate the above two following equations. 306 Usually πΌΜ ν (π) converges after four or five iterations. 307 308 C.2 Matrix Inversion Method 309 310 We can also directly solve πΌΜ ν (π) by the Matrix Inversion Method. Discretize the equation: 311 πΌνΜ (ππ ) 312 1 1 = π΅ν (ππ )πΈ2 (ππ − ππ ) + ∑[(1 − ππ )π½ππ + ππ π½ππ ]|πΈ2 (|ππ − ππ |) − πΈ2 (|ππ − ππ+1 |)| 2 2 π π=1 313 314 315 316 (πΆ. 7) where π½ππ and π½ππ are the continuum and gas band contribution: π½ππ = (π΅π + π΅π+1 ) 1 πΌπΜ + εj π΅π πΌΜ Μ Μ Μ Μ π+1 + εj+1 π΅π+1 & π½ππ = [ + ] 2 2 1 + εj 1 + εj+1 Therefore: 12 (πΆ. 8) (πΆ. 6π) 317 πΌνΜ (ππ ) 318 ππ + εj ππ + εj+1 1 1 = π΅π πΈ2 (ππ − ππ ) + ∑[ π΅π + π΅ ] |πΈ2 (|ππ − ππ |) − πΈ2 (|ππ − ππ+1 )|| 2 4 1 + εj 1 + εj+1 π+1 π π=1 319 π (1 − ππ )πΌΜ π (1 − ππ )πΌΜ Μ Μ Μ Μ 1 π+1 + ∑(|πΈ2 (|ππ − ππ | − πΈ2 (|ππ − ππ+1 |)|)[ + ] 4 1 + εj 1 + εj+1 (πΆ. 9) π=1 320 Using Einstein notation, πΌνΜ (ππ ) ≡ πΌπ + π½ππ πΌπΜ , where: 321 πΌπ 322 = 323 ππ + εj ππ + εj+1 1 + ∑( π΅π + π΅ )|πΈ (|π − ππ |) − πΈ2 (|ππ − ππ+1 |)| 4 1 + εj 1 + εj+1 π+1 2 π 1 π΅ (π )πΈ (π − ππ ) 2 ν π 2 π π π=1 324 1 1 − π1 π½π,1 = (|πΈ2 (|ππ − π1 |) − πΈ2 (|ππ − π2 |)|)( ) 4 1 + ε1 325 π½π,π (π = 2 … π) 326 = 327 1 1 − ππ + |πΈ2 (|ππ − ππ |) − πΈ2 (|πΈπ − ππ+1 |)|( ) 4 1 + εk 328 1 1 − ππ π½π,π+1 = (|πΈ2 (|ππ − ππ |) − πΈ2 (|ππ − ππ+1 |)|)( ) 4 1 + εN+1 329 Take the band average (“~” stands for the matrix form): 330 1 1 − ππ−1 |πΈ2 (|ππ − ππ−1 |) − πΈ2 (|ππ − ππ |)|( ) 4 1 + εk ∫ πΌΜ πν πν ∫ π½Μ πν πν Μ + π½ΜΜ πΌ Μ πΌΜΜ = + πΌ Μ = πΌΜ ∫ πν πν ∫ πν πν −1 (πΆ. 10) 331 Therefore the final solution is πΌΜΜ = (π°Μ − π½ΜΜ ) 332 Μ. LTE situation (ε → ∞), π½ΜΜ → 0Μ, πΌΜΜ → πΌΜ 333 Our calculations show that the matrix inversion method is less accurate than the iteration 334 method. Therefore in this study we adopt the latter. 335 336 337 13 πΌΜ Μ , where π°Μ is the identity matrix. For the 338 Appendix D 339 Derivations for the Integral Involving Exponential Integral Functions 340 341 The general properties of the exponential integral functions are: 1 (π ≥ 2) π−1 342 πΈπ (0) = 343 ∫ πΈπ (π₯)ππ₯ = −πΈπ+1 (π₯) 344 Therefore, we obtain: π 345 π ∫ π‘πΈπ (π‘)ππ‘ = −ππΈπ+1 (π) + ∫ πΈπ+1 (π‘)ππ‘ = 0 0 1 − πΈπ+2 (π) − ππΈπ+1 (π) π+1 346 π 347 D.1. Derivation of ππ = ∫π π π−πΆπ π¬π (π − π)π π : 348 πΉπ = π −πΌπ ∫ ππ −π π −πΌπ‘ πΈπ (π‘)ππ‘ 0 π −πΌπ 349 =− 350 Therefore, [π −πΌ(ππ −π) πΈπ (ππ − π) − πΌ 1 + π πΌπ πΉπ−1 ] π−1 π 1 π−1 1 π−π+1 −πΌπ π −πΌπ πΉπ = (− ) πΉ1 + ∑ (− ) [π π πΈπ (ππ − π) − ] πΌ πΌ π−1 351 (π·. 1) π=2 352 where, 353 πΉ1 = ∫ π −πΌπ‘ πΈ1 (π‘ − π)ππ‘ = π −πΌπ ∫ ππ π ππ −π ∞ π −πΌπ‘ ππ‘ ∫ 0 1 π −π‘π§ ππ§ π§ ∞ 1 − π −(ππ −π)π₯ π −πΌπ ∞ 1 1 ππ₯ = ∫ ( − ) (1 − π −(ππ −π)π₯ ) ππ₯ πΌ 1+πΌ π₯ − πΌ π₯ 1+πΌ π₯(π₯ − πΌ) 354 = π −πΌπ ∫ 355 = 356 π −πΌπ = [−π −πΌ(ππ −π) πΈ1 (ππ − π) + πΈ1 ((1 + πΌ)(ππ − π)) + ln(1 + πΌ)] πΌ π −πΌπ π −πΌπ ∞ π −(ππ −π)π₯ π −(ππ −π)π₯ ln(1 + πΌ) + ∫ ( − )ππ₯ πΌ πΌ 1+πΌ π₯ π₯−πΌ 357 358 when n=2, 14 (π·. 2) 359 ππ πΉ2 = ∫ π −πΌπ‘ πΈ2 (π‘ − π)ππ‘ = π −πΌπ 360 − π {π −πΌ(ππ −π) (πΌπΈ2 (ππ − π) − πΈ1 (ππ − π)) + πΈ1 ((1 + πΌ)(ππ − π)) + ln(1 + πΌ) − πΌ} πΌ2 361 (π·. 3) 362 363 π D.2 Derivation of π²π = ∫π π−πΆπ π¬π (π − π)π π : π 364 1 π −πΌπ [πΈπ (π) − + πΎπ−1 ] πΌ π−1 πΎπ = π −πΌπ ∫ π πΌπ‘ πΈπ (π‘)ππ‘ = 0 365 366 Therefore, π πΎ1 1 π−π+1 π −πΌπ πΎπ = π−1 + ∑ ( ) [πΈπ (π) − ] πΌ πΌ π−1 367 (π·. 4) π=2 368 where, 369 πΎ1 = ∫ π −πΌπ‘ πΈ1 (π − π‘)ππ‘ = π −πΌπ ∫ π πΌπ‘ ππ‘ ∫ π π 0 ∞ 0 ∞ −ππ₯ −πΌπ 1 π −π‘π§ ππ§ π§ ∞ 1−π π 1 1 ππ₯ = ∫ ( − ) (1 − π −ππ₯ ) ππ₯ π₯(π₯ + πΌ) πΌ π₯ π₯ + πΌ 1−πΌ 1−πΌ 370 = π −πΌπ ∫ (π·. 5) 371 372 Note that if πΌ > 1, there is a singularity at π₯ = 1. But the Cauchy principal value still 373 exists. For example, if πΌ > 1, ∞ 374 −π ∞ 1 1 1 ∫ ππ₯ = lim(∫ ππ₯ + ∫ ππ₯) π→0 1−πΌ π₯(π₯ + πΌ) 1−π π₯(π₯ + πΌ) π π₯(π₯ + πΌ) 375 = lim 376 =− πΌ+π ln (πΌ − π ) − ln(πΌ − 1) π→0 πΌ ln(πΌ − 1) πΌ 377 378 if πΌ < 1, 379 ∫ ∞ 1 ln(1 − πΌ) ππ₯ = − πΌ 1−πΌ π₯(π₯ + πΌ) 380 15 381 Therefore, 382 −π −πΌπ π −πΌπ ∞ π −ππ₯ π −ππ₯ πΎ1 = ln(|πΌ − 1|) + ∫ ( − ) ππ₯ πΌ πΌ 1−πΌ π₯ + πΌ π₯ 383 = π −πΌπ πΌπ [π πΈ1 (π) − πΈ1 ((1 − πΌ)π) − ln(|πΌ − 1|)] πΌ (π·. 6) 384 385 when n=2, 386 πΎ2 = ∫ π −πΌπ‘ πΈ2 (π − π‘)ππ‘ π 0 387 π −πΌπ πΌπ = 2 {π [(πΈ1 (π) + πΌπΈ2 (π)] − πΈ1 ((1 − πΌ)π) − ln(|πΌ − 1|) − πΌ} πΌ (π·. 7) 388 389 390 391 Note that there is no singularity at πΌ = 1, since: −πΈ1 ((1 − πΌ)π) − ln(|πΌ − 1|) = πΎ + ln(|πΌ − 1|π) − ln(|πΌ − 1|) = πΎ + ln(π) where πΎ is the Euler–Mascheroni constant. 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 16 409 Figure Captions 410 411 Fig. B.1. Test cases (1: black; 2: orange; 3: blue) for averaged cooling rates within each 412 layer from different numerical schemes. Three symbols indicate three solutions. Squares 413 are analytical solutions; triangles are from the Gaussian Quadrature method; Crosses are 414 from the finite difference scheme. The relative difference is defined as the absolute value 415 of (numerical-analytical)/analytical. 416 417 Fig. B.2. Same as Fig. B.1, but for the cooling rates defined at levels. 418 419 Fig. B.3. Test cases of upward IR fluxes πΉ ↑ (π) from different numerical schemes and 420 compared with the analytical solutions. The symbols and colors follow Fig. B.1. 421 422 Fig. B.4. Test cases of downward IR fluxes πΉ ↓ (π) from different numerical schemes and 423 compared with the analytical solutions. The symbols and colors follow Fig. B.1. 17 424 425 Fig. B.1. 18 426 427 Fig. B.2. 428 429 430 431 432 19 433 434 Fig. B.3. 435 436 437 438 20 439 440 Fig. B.4. 441 442 21