+ Yjȱ^RIt^RIt^RIHt^RIHt]P!]4Pt ]P!]4Pt Rt RtRtRtRtR tR tR tR#) N)qr)get_arrays_tolc V'Ed|\V\P4'dVP^8XgQh\P !V4P V4'gQh\V\P4'dVPVP8XgQh\V\P4'dVPVP8XgQh\V\4'gQh\V\4'gQh\W#4p\P!W'8*4'gQh\P!W7)84'gQh\P!V4'dVR8gQh\P!VR4p\P!VR4pVPp\P !V4p\P !V4p VR8VR8,VR8VR8,,p \P"!V4p \P"!V 4p W ,)W&^p RpRpV \P$!V 48EdW ,pVR \&,V,\)R\P*P-V44,8dEM\/WV4pT)T,RT,8:dEMT!T 4pT T,pT\2\5T4,8d\)T)T, R4pM\P6p\9TT4pT)TRT,T,,,RT,8:dEMT\P6)8T \2)\P4!Y+, 4,8,pT\P68T \2\P4!Y;, 4,8,p\P:!T \P64p\P:!T \P64p\P!TT,T T,, T T,, R4TT&\P!TT,T T,, T T,, R4TT&\P8!T4p\P8!T4p\9TT4p\9TT4pTR8dv\P<!Y,TY,,,Y*,Y:,4Y&TTT,, pTTTRT,T,,,,pT\9TT48dMY ,TT ,,T, pTY,,Y ,, Y&RY(&T ^, p EKTT8dgTT8:d$\P>!T4pTT,T T&M"\P>!T4pTT,T T&RT T&Y ,)Y&RY(&^p EKTT8:d\AT4pTT,T T&RT T&TT8:d\AT4pTT,T T&RT T&RpVPCRR4'EdV'Ed\P !V 4pV V,RV,V!V4,,p \P$!V 4^8EdOW,W,,p!W ,W ,,p"W ,W,,p#\PD!\)V!V",V#R,, R44)pV#W,,V!W ,,, W&RW(&VR V,8gF\PF!V\2)\P4!W,4,84'dEMZW;;,V),uu&\PH!V4p$\PH!V4p%W,R,W,R,,W*,R,, V$V &W,R,W,R,,W:,R,, V%V &\PD!V$V$R8,4V V$R8,, V$V$R8&\PD!V%V%R8,4V V%R8,,V%V%R8&\P!W, R4p&\P!W;, R4p'V$\2V&,8pV%\2V',8p\PJ!V4p(\PJ!V4p)\P!V(V,V&V,V$V,, 4V(V&\P!V)V,V'V,V%V,, 4V)V&\P8!V(4p*\P8!V)4p+\9V*V+4p,V!V 4p-V!V 4pV V-,p.V V,pV V,p/^p0\MV0^, V,,^,4p0\PN!V,V0, V,V04p1RV1,RV1R,,, p2V2V#V1,V, V1V.,, V2V1V/,RVV., ,, ,,,p3\P!V3R8*4'dEM\PP!V34p4RV1V4,R,, RV1V4,R,,, p5V5W,,V2V4,W,,,W&VV5R, V-,V2V4,V,,, pVV3V4,, pV,R8d[V4V0^, 8XdMV*V,8:d\AV(4pVV,V V&RV V&V+V,8:d \AV)4pVV,V V&RV V&EKfEKiW,RV ,V!V 4,,V 8dTp V'dn\P!W+8*4'gQh\P!W8*4'gQh\P*P-V 4R V,8gQhV # \0dEKvi;i) a Minimize approximately a quadratic function subject to bound constraints in a trust region. This function solves approximately .. math:: \min_{s \in \mathbb{R}^n} \quad g^{\mathsf{T}} s + \frac{1}{2} s^{\mathsf{T}} H s \quad \text{s.t.} \quad \left\{ \begin{array}{l} l \le s \le u\\ \lVert s \rVert \le \Delta, \end{array} \right. using an active-set variation of the truncated conjugate gradient method. Parameters ---------- grad : `numpy.ndarray`, shape (n,) Gradient :math:`g` as shown above. hess_prod : callable Product of the Hessian matrix :math:`H` with any vector. ``hess_prod(s) -> `numpy.ndarray`, shape (n,)`` returns the product :math:`H s`. xl : `numpy.ndarray`, shape (n,) Lower bounds :math:`l` as shown above. xu : `numpy.ndarray`, shape (n,) Upper bounds :math:`u` as shown above. delta : float Trust-region radius :math:`\Delta` as shown above. debug : bool Whether to make debugging tests during the execution. Returns ------- `numpy.ndarray`, shape (n,) Approximate solution :math:`s`. Other Parameters ---------------- improve_tcg : bool, optional If True, a solution generated by the truncated conjugate gradient method that is on the boundary of the trust region is improved by moving around the trust-region boundary on the two-dimensional space spanned by the solution and the gradient of the quadratic function at the solution (default is True). Notes ----- This function implements Algorithm 6.2 of [1]_. It is assumed that the origin is feasible with respect to the bound constraints and that `delta` is finite and positive. References ---------- .. [1] T. M. Ragonneau. *Model-Based Derivative-Free Optimization Methods and Software*. PhD thesis, Department of Applied Mathematics, The Hong Kong Polytechnic University, Hong Kong, China, 2022. URL: https://theses.lib.polyu.edu.hk/handle/200/12294. F?:0yE>?T improve_tcg@皙?$:0yE)) isinstancenpndarrayndiminspect signaturebindshapefloatboolrallisfiniteminimummaximumsizecopy zeros_like count_nonzeroEPSmaxlinalgnorm _alpha_trZeroDivisionErrorTINYabsinfmin full_likeclipargmin_argmingetsqrtanyzerosonesintlinspaceargmax)6grad hess_prodxlxudeltadebugkwargstoln grad_origfree_bdstepsdkreductboundary_reachedgrad_sdalpha_trhess_sdcurv_sd alpha_quadalphai_xli_xu all_alpha_xl all_alpha_xualpha_xlalpha_xualpha_bdbetai_new step_basestep_comparatorstep_sqgrad_sq grad_steptemp_xltemp_xudist_xldist_xuall_t_xlall_t_xut_xlt_xut_bd hess_step curv_step curv_step_sd n_samples t_samples sin_values all_reducti_max cos_values6&&&&&&, h/home/luca/screen-checking/.venv/lib/python3.14/site-packages/scipy/_external/cobyqa/subsolvers/optim.pytangential_byrd_omojokunrn sd @ u$ ++ Q>>  +006666"bjj))bhh$**.DDD"bjj))bhh$**.DDD%''''%&&&&R$vvbi    vvbDj!!!!{{5!!eck11 B B B B A 774=D ISTCZ(b3h4#:-FGG == D t B=.BK A F bw' ') eckAoC1E(FF F   51H 9w $- / B-w, TCL( (gX/5JJHj) 6WsU{W44 5 F bffW tebffRY.?&?!?@RVV TBFF29,=%= =>||D"&&1 ||D"&&1 ZZ XT "bh .  T ZZ XT "bh .  T66,'66,'x*E8$ 3;GG  33  DM EGO #D ewuw)>>? ?F 3x* *MGG$44?D,t}L IY]d2Q67I D9$4dIFIyC)S.,@AJ#I%i'(|+cWy5H.IIKKJvvjC'((IIj)Ey/366i&#--IDM)Ju,= ,KK M Y_ 1Ju4E4OO OD j' 'F czey1}44<#H-E"$U)DK%*GEN4<#H-E"$U)DK%*GEN   cDj9T?: :_ LD vvbj!!!!vvdj!!!!yy~~d#cEk111 KY!   s9 s++ s;:s;c  V$V'Ed\V\P4'dVP^8XgQh\P !V4P V4'gQh\V\P4'dVPVP8XgQh\V\P4'dVPVP8XgQh\V\P4'd4VP^8Xd#VP^,VP8XgQh\V\P4'd4VP^8Xd#VPVP^,8XgQh\V\P4'd4VP^8Xd#VP^,VP8XgQh\V\4'gQh\V\4'gQh\W#4p \P!W*8*4'gQh\P!W:)84'gQh\P!WZ)84'gQh\P!V4'dVR8gQh\P!VR4p\P!VR4p\P!VR4pVPp \P !V4p\P !V4p VR8VR8,p VR8VR8,pVR8W@,R8,p\#WFWV4wpp\P$!V4pVRVR13,)VRVR13,P&V,,p\P !V4p^pRpRpVV V, 8EdgVV,pVR \(,V ,\+R\P,P/V44,8dEM\1VVV4pT)T,RT,8:dEMT!T4pTT,pT\4\7T4,8d\+T)T, R4pM\P8p\;TT4pT)TRT,T,,,RT,8:dEMdY\P8)8,T\4)\P6!TT, 4,8,pY\P88,T\4\P6!TT, 4,8,p\P<!T\P84p \P<!T\P84p!\P!TT,TT,, TT,, R4T T&\P!TT,TT,, TT,, R4T!T&\P:!T 4p"\P:!T!4p#\;T"T#4p$TT,p%TT%\4\P6!T4,8,p&\P<!T\P84p'TT&,T%T&,, T'T&&\P:!T'\P8R7p(\;TT$T(4pTR8d\P>!TTT,,Y#4pTTT,, p\P!RTTT%,, 4pTTTRT,T,,,,pT\;TT$T(48d]TRTR13,TRTR13,P&T,,p)T)T,T, p*T*T,T), pT^, pEK;TT8dT"T8:d)\P@!T 4p+TT+,TT+&RT T+&MJT#T8:d)\P@!T!4p+TT+,TT+&RTT+&M\P@!T'4p+RTT+&\#TTT TT4wppTRTR13,)TRTR13,P&T,,p^pEKT"T8:d\CT 4p+TT+,TT+&RT T+&T#T8:d\CT!4p+TT+,TT+&RTT+&T(T8:d\CT'4p+RTT+&\#TTT TT4wppR pV PER R 4'EdV'EdVV 8Ed\P !V4p,VV 8EdYVRVR13,VRVR13,P&V,,p-VRVR13,VRVR13,P&V,,p)V-V-,p.V)V),p/V)V-,p0\PF!\+V.V/,V0R ,, R44)pVRVR13,VRVR13,P&V0V,V.V,, ,,pVRV,8g@\PH!V\4)\P6!V4,84'dEMVV),p\PJ!V 4p1\PJ!V 4p2\P!VV, R4p3\P!VV, R4p4VV ,R ,V3V ,V3V ,R V-V ,,, ,, V1V &VV,R ,V4V,V4V,R V-V,,,,, V2V&\PF!V1V1R8,4VV1R8,, V1V1R8&\PF!V2V2R8,4VV2R8,,V2V2R8&V1\4V3,8pV2\4V4,8p\PL!V 4p5\PL!V 4p6\P!V5V,V3V,V1V,, 4V5V&\P!V6V,V4V,V2V,, 4V6V&\P:!V54p7\P:!V64p8\;V7V84p9\P$!V4p:VV-,p;VV,p%V%V,R ,VV,VV,R V;V,,,,, V:V&\PF!V:V:R8,4V%V:R8,,V:V:R8&V:\4V,8p&\PN!V4p<\P!VV!V-4p?V!V4pV-V?,p@VV,pV-V,pA^pB\QVB^, V>,^,4pB\PR!V>VB, V>VB4pCR VC,RVCR ,,, pDVDV0VC,V, VDRVCR ,,V@,R VC,VA,, RV,,,, ,pE\P!VER8*4'dEM\PT!XE4pFRXCVF,R ,, RVCVF,R ,,, pG\P>!VVGR, V-,,XDVF,V,,VV4pVVGR, V?,VDVF,V,,, p\P!RVVGR, V;,, VDVF,V%,, 4pVVEVF,, pV>R8dXFXB^, 8XdsV7V>8:d\CV54p+VV+,VV+&RV V+&V8V>8:d\CV64p+VV+,VV+&RV V+&V=V>8:d\CV<4p+RVV+&\#VVV VV4wppEK_V V,RV,V!V4,,V V,,RV,,V!V,4,,8dT,pV'd\W#4p \P!VV8*4'gQh\P!VV8*4'gQh\P!VV,WZ,8*4'gQh\P!\P6!VV,4V 8*4'gQh\P,P/V4R V,8gQhV# \2dEKi;i)az Minimize approximately a quadratic function subject to bound and linear constraints in a trust region. This function solves approximately .. math:: \min_{s \in \mathbb{R}^n} \quad g^{\mathsf{T}} s + \frac{1}{2} s^{\mathsf{T}} H s \quad \text{s.t.} \quad \left\{ \begin{array}{l} l \le s \le u,\\ A_{\scriptscriptstyle I} s \le b_{\scriptscriptstyle I},\\ A_{\scriptscriptstyle E} s = 0,\\ \lVert s \rVert \le \Delta, \end{array} \right. using an active-set variation of the truncated conjugate gradient method. Parameters ---------- grad : `numpy.ndarray`, shape (n,) Gradient :math:`g` as shown above. hess_prod : callable Product of the Hessian matrix :math:`H` with any vector. ``hess_prod(s) -> `numpy.ndarray`, shape (n,)`` returns the product :math:`H s`. xl : `numpy.ndarray`, shape (n,) Lower bounds :math:`l` as shown above. xu : `numpy.ndarray`, shape (n,) Upper bounds :math:`u` as shown above. aub : `numpy.ndarray`, shape (m_linear_ub, n) Coefficient matrix :math:`A_{\scriptscriptstyle I}` as shown above. bub : `numpy.ndarray`, shape (m_linear_ub,) Right-hand side :math:`b_{\scriptscriptstyle I}` as shown above. aeq : `numpy.ndarray`, shape (m_linear_eq, n) Coefficient matrix :math:`A_{\scriptscriptstyle E}` as shown above. delta : float Trust-region radius :math:`\Delta` as shown above. debug : bool Whether to make debugging tests during the execution. Returns ------- `numpy.ndarray`, shape (n,) Approximate solution :math:`s`. Other Parameters ---------------- improve_tcg : bool, optional If True, a solution generated by the truncated conjugate gradient method that is on the boundary of the trust region is improved by moving around the trust-region boundary on the two-dimensional space spanned by the solution and the gradient of the quadratic function at the solution (default is True). Notes ----- This function implements Algorithm 6.3 of [1]_. It is assumed that the origin is feasible with respect to the bound and linear constraints, and that `delta` is finite and positive. References ---------- .. [1] T. M. Ragonneau. *Model-Based Derivative-Free Optimization Methods and Software*. PhD thesis, Department of Applied Mathematics, The Hong Kong Polytechnic University, Hong Kong, China, 2022. URL: https://theses.lib.polyu.edu.hk/handle/200/12294. rNNNNFrrr initialTr r r r r)+rrrrrrrrrrrrrrrrrqr_tangential_byrd_omojokunrTr!r"r#r$r%r&r'r(r)r*r+r,r-r.r/r0r1r2r3 ones_liker4r5r6)Hr7r8r9r:aubbubaeqr;r<r=r>r?r@free_xlfree_xufree_ubn_actqrBrCresidrDrErFrGrHrIrJrKrLrMrNrOrPrQrRrSaub_sdi_ub all_alpha_ubalpha_ub grad_projrTrUrV step_projrXrYrZr[r\r]r^r_r`rarbrctemp_ubaub_stepall_t_ubt_ubt_minrdrerfrgrhrirjrkrlsH&&&&&&&&&, rm$constrained_tangential_byrd_omojokunrCsf u$ ++ Q>>  +006666"bjj))bhh$**.DDD"bjj))bhh$**.DDD sBJJ ' 'A  !  )  * sBJJ ' 'A CIIaL(  ) sBJJ ' 'A  !  )  *%''''%&&&&R$vvbi    vvbDj!!!!vvcTk""""{{5!!eck11 B B B B **S# C A 774=D ICxD3J'GCxD3J'GSySZ#-.G*3WwOHE1 == D AuvI,!AuvI,..4/ 0B GGCLE A F a%i-) eckAoC1E(FF F   r51H 9w $- / B-w, TCL( (gX/5JJHj) 6WsU{W44 5 F w,'2rDy8I0I+IJrvv+&"tbffR$Y6G/G*GH||D"&&1 ||D"&&1 ZZ XT "bh .  T ZZ XT "bh .  T66,'66,'x*r&4"&&-#778||C0 "4[6$<7 T66,7E8X. 3;774%"*,b5D EGO #DJJsEEFN$:;E ewuw)>>? ?F 3x84 4!UV) !UV) (=>I'72DY&B FA X 5  ,/ iU !&U" ,/ iU !& ,/!&2 HE1AuvI,!AuvI,..4"78BA 5  - iU !&5  - iU !&5  -!&2 HE1 $  zz-&&+;+; GGDM ai!UV) !UV) (=>I!UV) !UV) (=>I)+G)+G!I-Iwws7W#4y#~#EsKLLG1ef9!UV) )d"2Wt^"CDB%&.(BFFD5266":---- 7(NB hhqkGhhqkGjjC0GjjdC0G!'{c1GG4D 37);#;;5 GG  "'{c1GG4D 37);#;;5 GG # ./"Ws]2CC GcM "# ./"Ws]2CC GcM "TG^+DTG^+DwwqzHwwqzHZZ  -HTN ZZ  -HTN66(#D66(#DtT?DmmE*GYH2XF%g#5ggx'8!889 GG # ./&32GG GcM "TE\)D||E*HZZd gdm+HTN66(C0DdOE"),ImG!I-I7lG$w.L IY]e3a78I EI$5uiHIyC)S.,@AJ#I%)S.(94Io 45Gm$ JvvjC'((IIj)Ey/366i&#--I77 C944z%7H27MMD Y_ 1Ju4E4OO ODJJs?h./U#f,-E j' 'F s{u A 55=#H-E"$U)DK%*GEN5=#H-E"$U)DK%*GEN5=#H-E%*GEN6 q t cDj9 ,    !C)Oi 6J$J J KD R$vvbDj!!!!vvdbj!!!!vvcDjCI-....vvbffS4Z(C/0000yy~~d#cEk111 Ke!   s* AHH AH(H'AH(c %V'Ede\V\P4'dVP^8XgQh\V\P4'd4VP^8Xd#VPVP ^,8XgQh\V\P4'd;VP^8Xd*VP ^,VP ^,8XgQh\V\P4'd4VP^8Xd#VPVP ^,8XgQh\V\P4'd$VP VP ^,38XgQh\V\P4'd$VP VP ^,38XgQh\V\ 4'gQh\V\4'gQh\WE4p \P!WI8*4'gQh\P!WY)84'gQh\P!V4'dVR8gQh\P!VR4p\P!VR4pVP wr\PVPV),\P!RV)43,p VR8V RV R8,p VR8V RV R8,pVR8pVR8W RV ,WR, R8,p\VV VVV4wpp\P !W3,WRWR,,4p\P"!V 4pVRVR13,)VRVR13,PV ,,pWV R,p^pRpRpVW,V, 8Ed|V V,pVR \$,V ,\'R\P(P+V 44,8dEM)\-VVRV V4p\3V\-WRVV RV44pV)V,RV,8:dEM\PVPVVRV ,,VV R3,pVV,pV\4\7V4,8d\'V)V, R4pM\P0p\3VV4pV)VRV,V,,,RV,8:dEM*W\P0)8,VRV \4)\P6!VV, 4,8,p W\P08,VRV \4\P6!VV, 4,8,p!VVV R\4)\P6!WR4,8,p"\P8!V\P04p#\P8!V\P04p$\P8!V\P04p%\P!VV ,VV ,, VRV V ,, R4V#V &\P!VV!,VV!,, VRV V!,, R4V$V!&\P!WRV",)VV RV",, R4V%V"&\P2!V#4p&\P2!V$4p'\P2!V%\P0R7p(\3V&V'V(4p)VVRV ,VV R, p*VV*\4\P6!V4,8,p+\P8!V\P04p,VV+,V*V+,, V,V+&\P2!V,\P0R7p-\3VV)V-4pVR8d\P:!VVVRV ,,WE4pV VV,, p \P!RVVV*,, 4pVVVRV,V,,,,pV\3VV)V-48d]VRVR13,VRVR13,PV ,,p.V.V,V, p/V/V,V., pV^, pEK^VV8dV&V8:d)\P<!V#4p0VV0,VV0&RV V0&MmV'V8:d)\P<!V$4p0VV0,VV0&RVV0&M>V(V8:d\P<!V%4p0RVV0&M\P<!V,4p0RVV0&\W WV4wppVRVR13,)VRVR13,PV ,,p^pEKGV&V8:d\?V#4p0VV0,VV0&RV V0&V'V8:d\?V$4p0VV0,VV0&RVV0&R pVPAR R 4'EdV'Ed\PB!V4p1W,p2VP\P!VV,V, R4,VPVV,V, ,,p \P"!V 4p\PD!V24^8EduVV2,VV2,,p3V V2,V V2,,p4V V2,VV2,,p5\P !\'V3V4,V5R ,, R44)pV5VV2,,V3V V2,,, VV2&RVV2(&VRV,8gG\PF!V\4)\P6!VV2,4,84'dEMuVV2;;,V),uu&\P"!V 4p6\P"!V 4p7VV2,R ,VV2,R ,,VV2,R ,, V6V2&VV2,R ,VV2,R ,,VV2,R ,, V7V2&\P !V6V6R8,4VV6R8,, V6V6R8&\P !V7V7R8,4VV7R8,,V7V7R8&\P!VV, R4p8\P!VV, R4p9V6\4V8,8p V7\4V9,8p!\PH!V 4p:\PH!V 4p;\P!V:V ,V8V ,V6V ,, 4V:V &\P!V;V!,V9V!,V7V!,, 4V;V!&\P2!V:4p<\P2!V;4p=\3V^p?\KV?^, V>,^,4p?\PL!V>V?, V>V?4p@\P!VV,V, R4pAVV,V, pB\PB!V4pCRVCV2(&\PN!V?4pD\QV?4FpER X@VE,,RV@VE,R ,,, pF\P:!VVFVV@VE,XC,, ,,VV4pG\P!VVG,V, R4pHVVG,V, pIRXAVA,XBVB,,VHVH,, VIVI,, ,XDVE&K \P!XDR8*4'dEMk\PR!XD4pJRX@VJ,R ,, RV@VJ,R ,,, pKR V@VJ,,RV@VJ,R ,,, pFVKVV2,,VFVV2,,,VV2&VP\P!VV,V, R4,VPVV,V, ,,p VVDVJ,, pV>R8d[XJV?^, 8XdMV8:d\?V:4p0VV0,VV0&RV2V0&V=V>8:d \?V;4p0VV0,VV0&RV2V0&EKEK\P!VV,V, R4pA\P!VV1,V, R4pLVV,V, pBVV1,V, pMVAVA,VBVB,,VLVL,VMVM,,8dT1pV'dp\P!VV8*4'gQh\P!VV8*4'gQh\P(P+V4R V,8gQhV# \.d\P0pELi;i \.dELi;i)a Minimize approximately a linear constraint violation subject to bound constraints in a trust region. This function solves approximately .. math:: \min_{s \in \mathbb{R}^n} \quad \frac{1}{2} \big( \lVert \max \{ A_{\scriptscriptstyle I} s - b_{\scriptscriptstyle I}, 0 \} \rVert^2 + \lVert A_{\scriptscriptstyle E} s - b_{\scriptscriptstyle E} \rVert^2 \big) \quad \text{s.t.} \quad \left\{ \begin{array}{l} l \le s \le u,\\ \lVert s \rVert \le \Delta, \end{array} \right. using a variation of the truncated conjugate gradient method. Parameters ---------- aub : `numpy.ndarray`, shape (m_linear_ub, n) Matrix :math:`A_{\scriptscriptstyle I}` as shown above. bub : `numpy.ndarray`, shape (m_linear_ub,) Vector :math:`b_{\scriptscriptstyle I}` as shown above. aeq : `numpy.ndarray`, shape (m_linear_eq, n) Matrix :math:`A_{\scriptscriptstyle E}` as shown above. beq : `numpy.ndarray`, shape (m_linear_eq,) Vector :math:`b_{\scriptscriptstyle E}` as shown above. xl : `numpy.ndarray`, shape (n,) Lower bounds :math:`l` as shown above. xu : `numpy.ndarray`, shape (n,) Upper bounds :math:`u` as shown above. delta : float Trust-region radius :math:`\Delta` as shown above. debug : bool Whether to make debugging tests during the execution. Returns ------- `numpy.ndarray`, shape (n,) Approximate solution :math:`s`. Other Parameters ---------------- improve_tcg : bool, optional If True, a solution generated by the truncated conjugate gradient method that is on the boundary of the trust region is improved by moving around the trust-region boundary on the two-dimensional space spanned by the solution and the gradient of the quadratic function at the solution (default is True). Notes ----- This function implements Algorithm 6.4 of [1]_. It is assumed that the origin is feasible with respect to the bound constraints and that `delta` is finite and positive. References ---------- .. [1] T. M. Ragonneau. *Model-Based Derivative-Free Optimization Methods and Software*. PhD thesis, Department of Applied Mathematics, The Hong Kong Polytechnic University, Hong Kong, China, 2022. URL: https://theses.lib.polyu.edu.hk/handle/200/12294. rNrpFrrr rqTr r r r r)*rrrrrrrrrrrrrr_rtqr_normal_byrd_omojokunr0r2r!r"r#r$r%r&r)r*r'r(r+r,r-r.r/rr r1r3r4r5emptyranger6)Nrvrwrxbeqr9r:r;r<r=r> m_linear_ubr?r7ryrz free_slackr{r|r} delta_slackrBrCr~rDrErFrGrHrIrJrKrLrMrNi_slackrOrPall_alpha_slackrQrR alpha_slackrSrrrrrrTrUrVrArXrYrZr[r\r]r^r_r`rarbrcrgrhresid_ubresid_eqrrji sin_valuestep_alt resid_ub_alt resid_eq_altrkrl resid_ub_base resid_eq_basesN&&&&&&&&, rmnormal_byrd_omojokunrs>F u#rzz**sxx1}<< sBJJ ' 'A CIIaL(  ) sBJJ ' 'A  !  ! ,  - sBJJ ' 'A CIIaL(  )"bjj))bhh399Q>? ?F 3x84 4!UV) !UV) (=>I'72DY&B FA X 5  ,/ iU !&U" ,/ iU !&% /2$) 5! ,/!&.g7HE1AuvI,!AuvI,..4"78BA 5  - iU !&5  - iU !&#  zz-&&+;+;GGDM #uurzz#*s"2C88355 $J <  XXa[w'!+7md7m3G7md7m3GW W 5Iwws7W#4y#~#EsKLLG#d7m3gW 6MMBwKBxL%&.(BFFD5266"W+#666-- wKG8 #K hhqkGhhqkGW $r'{c'99BwK3D "X~3 #x')*"\12#\12! 1 &vvjC'((IIj)Ey/366i&#--Iy//)E"2c"99;I%W 5 BwK8OODM552::cDj3&6< :#i#- x (X"5 5m+mm.KK LD vvbDj!!!!vvdbj!!!!yy~~d#cEk111 Km! vvH !   s.5AJ5AKJ5AKKAKK AK&K%AK&c tVPp\P!V4p\\P!V.W(R3,.Wb(R3,).Wc(R3,..4P RR7wrxp \P !\P!\P!V44R\,V,\PPVR\P!VP41R\P!VP413,^R7,84p W3#rpT)pivotingg$@N)axis)rreyerblockrtr r(diagr!r#r$r*r) rvrxryrzr{r?identityr}r_r|s &&&&& rmrsrscs AvvayH Xq[!"8Q;''((A+&'    ! GA!    rwwqz      ))..,RVVAGG_,.?qww.??@q. I J J E 8OcVPwrV\P!V4p\P!V4p\\P!W(R3,W(R3,).\P !V\P !V4, V34W(R3,).Wq(R3,)\P !V\P !V4, V34.Wr(R3,\P !V\P !V4, V34..4PRR7wrp \P !\P!\P!V 44R\,We,,\PPV R\P!V P41R\P!V P413,^R7,84p W3#r)rrrrrr2r rtr(rr!r#r$r*) rvryrzrr{rr? identity_n identity_mr}rrr|s &&&&& rmrr{sYYNKJ $J ! $! ,, HHkB,<, ?D#~%$T^)<"<<%/5 O T^ tj0 0zG^4c: O rc\\P!V\P!V484#N)r flatnonzeror"xs&rm_argmaxr >>!rvvay. ))rc\\P!V\P!V48*4#r)rrr*rs&rmr.r.rr)rnumpyr scipy.linalgrutilsrfinfortinyr'epsr!rnrrrsrr%rr.rrmrsh" xxhhuosl `F zz 0"J **r