import mpmath as mp, json, numpy as np
mp.mp.dps=60
def qm(a,b):
    a0,a1,a2,a3=a; b0,b1,b2,b3=b
    return (a0*b0-a1*b1-a2*b2-a3*b3, a0*b1+a1*b0+a2*b3-a3*b2, a0*b2-a1*b3+a2*b0+a3*b1, a0*b3+a1*b2-a2*b1+a3*b0)
def eq(a,b): return all(abs(x-y)<mp.mpf(10)**-40 for x,y in zip(a,b))
def close(gens):
    els=[(mp.mpf(1),mp.mpf(0),mp.mpf(0),mp.mpf(0))]; fr=list(els)
    while fr:
        new=[]
        for x in fr:
            for g in gens:
                y=qm(x,g)
                if not any(eq(y,e) for e in els): els.append(y); new.append(y)
        fr=new
    return els
h=mp.mpf(1)/2; r2=1/mp.sqrt(2); ph=(1+mp.sqrt(5))/2
G={'2T':close([(0,1,0,0),(h,h,h,h)]),'2O':close([(0,1,0,0),(h,h,h,h),(r2,r2,0,0)]),'2I':close([(h,h,h,h),(h,ph/2,1/(2*ph),0)])}
def classes(g):
    cl=[];used=set()
    for i,x in enumerate(g):
        if i in used: continue
        orb=set()
        for y in g:
            c=qm(qm(y,x),(y[0],-y[1],-y[2],-y[3]))
            for j,e in enumerate(g):
                if eq(c,e): orb.add(j)
        used|=orb; cl.append([g[j] for j in orb])
    return cl
ref=json.load(open('refine.json')); out={}
for name,g in G.items():
    cl=classes(g); N=len(g); I0=mp.matrix(ref[name]['argmin'])
    data=[]
    for C in cl:
        s2=1-C[0][0]**2
        if s2<mp.mpf(10)**-30: continue
        axes=[mp.matrix(c[1:])/mp.norm(mp.matrix(c[1:])) for c in C]
        data.append((mp.mpf(len(C))/N,s2,axes))
    def beta(I):
        I=I/mp.norm(I); return sum(d*s2*(1-max((a.T*I)[0]**2 for a in ax)) for d,s2,ax in data)
    # active axes and tie normals
    normals=[]; active=[]
    for d,s2,ax in data:
        dots=[((a.T*I0)[0]) for a in ax]; m=max(abs(x) for x in dots)
        tied=[a*mp.sign(x) for a,x in zip(ax,dots) if abs(abs(x)-m)<1e-6]
        uniq=[]
        for t in tied:
            if not any(mp.norm(t-u)<1e-20 for u in uniq): uniq.append(t)
        active.append((d,s2,uniq))
        for k in range(1,len(uniq)): normals.append(uniq[k]-uniq[0])
    if normals:
        nrm=normals[0]/mp.norm(normals[0])
        par=max(mp.norm(mp.matrix(np.cross([float(x) for x in nrm],[float(x) for x in v/mp.norm(v)]).tolist())) for v in normals)
        # orthonormal basis of plane perp nrm
        u=I0-(nrm.T*I0)[0]*nrm; u=u/mp.norm(u)
        v=mp.matrix(3,1); v[0]=nrm[1]*u[2]-nrm[2]*u[1]; v[1]=nrm[2]*u[0]-nrm[0]*u[2]; v[2]=nrm[0]*u[1]-nrm[1]*u[0]
        f=lambda th: beta(mp.cos(th)*u+mp.sin(th)*v)
        # 1D min near 0, golden refinement with findroot on derivative using fixed active axes
        A=[(d,s2,uq[0]) for d,s2,uq in active]
        c=sum(d*s2 for d,s2,_ in A)
        Muu=sum(d*s2*(a.T*u)[0]**2 for d,s2,a in A); Mvv=sum(d*s2*(a.T*v)[0]**2 for d,s2,a in A); Muv=sum(d*s2*(a.T*u)[0]*(a.T*v)[0] for d,s2,a in A)
        lam=(Muu+Mvv)/2+mp.sqrt(((Muu-Mvv)/2)**2+Muv**2)
        bstar=c-lam
        th=mp.atan2(2*Muv,Muu-Mvv)/2; Istar=mp.cos(th)*u+mp.sin(th)*v
        # check: active set consistent at Istar, and beta(Istar)==bstar; and in-plane perturbations
        chk=beta(Istar); 
        # off-plane perturbation check
        eps=mp.mpf('1e-6'); offs=[beta(Istar+eps*nrm)-chk, beta(Istar-eps*nrm)-chk]
        poly=mp.findpoly(bstar,4,maxcoeff=10**8)
    else:
        bstar=beta(I0); poly=None; chk=bstar; offs=[]; par=0
    print(name, mp.nstr(bstar,30), mp.nstr(chk-bstar,5), 'normals parallel resid',mp.nstr(par,3), 'off-plane',[mp.nstr(o,5) for o in offs], 'poly',poly)
    out[name]=dict(beta_min=mp.nstr(bstar,40),beta_check_diff=mp.nstr(chk-bstar,5),minpoly_coeffs_high_to_low=poly,offplane_increase=[mp.nstr(o,8) for o in offs],tie_normals_parallel_resid=mp.nstr(par,3))
json.dump(out,open('exact.json','w'),indent=1)
