import sys, json sys.path.insert(0,'.') import numpy as np from math import factorial import matrices as m import cases as cs exec(open("basisglow.py").read().split("out={}")[0].split("import cases as cs")[1]) allc=dict(extra); allc.update(cs.C) J=json.load(open('basisglow.json')) def S(M,k): K=M**k; N=M.shape[0]; return np.linalg.norm(K@K.conj().T-N*np.eye(N))**2 rows=[] for name,mk in allc.items(): H=np.array(mk(),complex); N=H.shape[0]; d=J[name]["dev"] D=[d[p-1]*factorial(p)*N**(2*p) for p in range(1,len(d)+1)] if len(D)<4: continue rel=abs(D[3]-18*S(H,2))/max(1,abs(D[3])) rows.append((name,N,D,S(H,2),S(H,3))) print(f"{name:26s} N={N:2d} relerr(D4 vs 18 S2)={rel:.1e} S2={S(H,2):10.3f} S3={S(H,3):10.3f}") # p=6 at N=6: fit D6 = a S2 + b S3 for N in [6]: R=[r for r in rows if r[1]==N and len(r[2])>=6] A=np.array([[r[3],r[4]] for r in R]); y=np.array([r[2][5] for r in R]) c,res,_,_=np.linalg.lstsq(A,y,rcond=None) print("N=%d p=6: D6 = %.4f S2 + %.4f S3, max residual %.2e"%(N,*c,np.abs(A@c-y).max())) y=np.array([r[2][6] for r in R]); c,res,_,_=np.linalg.lstsq(A,y,rcond=None) print("N=%d p=7: D7 = %.4f S2 + %.4f S3, max residual %.2e"%(N,*c,np.abs(A@c-y).max())) for r in R: print(" ",r[0],round(r[2][5],3),round(r[3],3),round(r[4],3))