import os import numpy as np import matplotlib.pyplot as plt from scipy.optimize import curve_fit from atlas_fit_function import atlas_invMass_mumu data_path = 'DATA/generated_histograms' histo_name = 'hist_range_110-160_nbin-25_Asimov' labels = ['Background', 'Signal', 'Data'] save_figs = False ############################################################################################################################# # Load data pldict = {} for label in labels: fileName = os.path.join(data_path, histo_name + label + '.npz') with np.load(fileName, 'rb') as data: bin_centers = data['bin_centers'] bin_edges = data['bin_edges'] bin_values = data['bin_values'] bin_errors = data['bin_errors'] pldict[label] = [bin_centers, bin_edges, bin_values, bin_errors] xs = np.linspace(bin_edges[0], bin_edges[-1], 301) width = (np.max(xs) - np.min(xs)) / len(bin_centers) ############################################################################################################################# # Helper fit functions def BackgroundFit(x, a, b, c, d, e): '''Some random function as a weight to the atlas_fit_function''' return atlas_invMass_mumu(np.exp(a * x) + b * x**3 + c * x**2 + d * x + e, x) def CrystalBall(x, A, aL, aR, nL, nR, mCB, sCB): '''Double-Sided Crystal Ball Function, see e.g. arXiv:2009.04363v2''' np.seterr(all="ignore") # Turn off some annoying warnings return np.piecewise(x, [(x - mCB) / sCB <= -aL, (x - mCB) / sCB >= aR], [lambda x: A * (nL / np.abs(aL))**nL * np.exp(-aL**2 / 2) * (nL / np.abs(aL) - np.abs(aL) - (x - mCB) / sCB)**(-nL), lambda x: A * (nR / np.abs(aR))**nR * np.exp(-aR**2 / 2) * (nR / np.abs(aR) - np.abs(aR) + (x - mCB) / sCB)**(-nR), lambda x: A * np.exp(-(x - mCB)**2 / (2 * sCB**2)) ]) ########################################################################################################################### # Fit background from data bin_centers = pldict['Data'][0] bin_edges = pldict['Data'][1] bin_values = pldict['Data'][2] bin_errors = pldict['Data'][3] xerrs = 0.5 * (bin_edges[1:] - bin_edges[:-1]) # Blind signal region signal_region = (117, 135) mask = (bin_centers < signal_region[0]) | (bin_centers > signal_region[1]) bin_centers_masked = bin_centers[mask] bin_values_masked = bin_values[mask] bin_errors_masked = bin_errors[mask] xerrs_masked = xerrs[mask] popt_masked, pcov = curve_fit(BackgroundFit, bin_centers_masked, bin_values_masked, sigma=bin_errors_masked, p0=[0.23, -3.5e10, 1.2e13, -1.5e15, 6e16]) f, (ax1, ax2) = plt.subplots(2, sharex=True, figsize=(16, 9), gridspec_kw={'height_ratios': [3, 1]}) f.suptitle('Fitting background from data', fontsize=22) ax1.errorbar(bin_centers, bin_values, bin_errors, xerrs, marker='o', markersize=5, color='k', ecolor='k', ls='', label='Data') ax1.plot(xs, BackgroundFit(xs, *popt_masked), 'r-', label='Blinded Data fit') ax1.axvspan(signal_region[0], signal_region[1], color='gainsboro', lw=2, ls='--', ec='k') ax1.set_ylabel('Number of events', fontsize=20) ax1.text(123, 2 * 10**5, "Blinded area", size=20) ax1.tick_params(axis='both', which='major', labelsize=20) ax1.legend(fontsize=20) # ax1.set_yscale('log') ax2.bar(bin_centers_masked, bin_values_masked / BackgroundFit(bin_centers_masked, *popt_masked) - 1, width=width, color='k') ax2.axhline(0, color='k', ls='--', alpha=0.7) ax2.set_xlabel(r'$m_{\mu \mu}$', fontsize=20) ax2.set_ylabel('(Data-Pred.)/Pred.', fontsize=20) ax2.set_xticks(bin_edges[::2]) ax2.tick_params(axis='both', which='major', labelsize=20) ax2.grid(True) f.tight_layout() if save_figs: plt.savefig('DataBkgFit.pdf') ############################################################################################################################# # Fit simiulated signal with the crystal ball function bin_centers = pldict['Signal'][0] bin_edges = pldict['Signal'][1] bin_values = pldict['Signal'][2] bin_errors = pldict['Signal'][3] popt_CB, pcov = curve_fit(CrystalBall, bin_centers, bin_values, sigma=bin_errors, p0=[2.6 * 10**4., 1.5, 1.5, 5., 5., 124.6, 2.5]) f, (ax1, ax2) = plt.subplots(2, sharex=True, figsize=(16, 12), gridspec_kw={'height_ratios': [3, 1]}) f.suptitle('Fitting inflated simulated signal', fontsize=22) ax1.scatter(bin_centers, bin_values, color='k', label='Simulated 100x signal') ax1.plot(xs, CrystalBall(xs, *popt_CB), color='r', label='Fitted signal') ax1.axvline(popt_CB[-2], color='k', ls='--', alpha=0.3) ax1.annotate(r'$m_{{Higgs}}^{{fit}}= {:.2f}$ GeV'.format(popt_CB[-2]) + '\n' + r'$N_{{Higgs}}^{{exp}} = {:d}$'.format(int(popt_CB[0])), (popt_CB[-2], 10000), xytext=(148, 2000), fontsize=20, arrowprops=dict(facecolor='black', shrink=0.05), size=20, bbox=dict(facecolor='w', edgecolor='gray', boxstyle='round,pad=0.5')) string = 'CB parameters after fit:\n' for par, pop in zip(['A', r'$\alpha_L$', r'$\alpha_R$', r'$n_L$', r'$n_R$', r'$\overline{m}_{CB}$', r'$\sigma_{CB}$'], popt_CB): string += par + f': {pop:.3f}\n' ax1.text(148, 10**4, string[:-1], size=20, bbox=dict(facecolor='w', edgecolor='gray', boxstyle='round,pad=0.5')) ax1.legend(loc='upper right', fontsize=20) ax1.set_ylabel('Number of events', fontsize=20) ax1.set_xticks(bin_edges[::2]) ax1.tick_params(axis='both', which='major', labelsize=20) ax1.grid(True) ax2.bar(bin_centers, bin_values / CrystalBall(bin_centers, *popt_CB) - 1, width=width, color='k') ax2.axhline(0, color='k', ls='--', alpha=0.7) ax2.set_xlabel(r'$m_{\mu \mu}$', fontsize=20) ax2.set_ylabel('(Data-Pred.)/Pred.', fontsize=20) ax2.set_xticks(bin_edges[::2]) ax2.tick_params(axis='both', which='major', labelsize=20) ax2.grid(True) f.tight_layout() if save_figs: plt.savefig('AsimovSimSignalFit.pdf') ############################################################################################################################# # Extract and fit our signal # Signal = Data - Fitted Background bin_values = pldict['Data'][2] extracted_signal = bin_values - BackgroundFit(bin_centers, *popt_masked) bin_centers = pldict['Signal'][0] bin_edges = pldict['Signal'][1] bin_values = pldict['Signal'][2] bin_errors = pldict['Signal'][3] def scale_signal(x, scale, popt_CB=popt_CB): return scale * CrystalBall(x, *popt_CB) # Fit extracted signal with the crystal ball function popt_final, pcov = curve_fit(scale_signal, bin_centers, extracted_signal, sigma=bin_errors, p0=[1.]) NHiggs = int(popt_CB[0] * popt_final[0]) f, (ax1, ax2) = plt.subplots(2, sharex=True, figsize=(16, 12), gridspec_kw={'height_ratios': [3, 1]}) f.suptitle('Fitting extracted signal', fontsize=22) ax1.plot(xs, scale_signal(xs, *popt_final), color='r', label='Fitted signal') ax1.scatter(bin_centers, extracted_signal, color='k', label='Extracted signal') ax1.axvline(popt_CB[-2], color='k', ls='--', alpha=0.3) ax1.text(110, 15000, r'$\alpha_{{scale}} = {:.3f}$'.format(*popt_final) + '\n' + r'$N_{{Higgs}} = {:d}$'.format(NHiggs), size=25, bbox=dict(facecolor='w', edgecolor='gray', boxstyle='round,pad=0.5')) ax1.legend(loc='upper right', fontsize=20) ax1.set_ylabel('Number of events', fontsize=20) ax1.set_xticks(bin_edges[::2]) ax1.tick_params(axis='both', which='major', labelsize=20) ax1.set_xlim((108, 140)) # Plot only relevant part of the signal ax1.grid(True) ax2.bar(bin_centers, extracted_signal / scale_signal(bin_centers, *popt_final) - 1, width=width, color='k') ax2.axhline(0, color='k', ls='--', alpha=0.7) ax2.set_xlabel(r'$m_{\mu \mu}$', fontsize=20) ax2.set_ylabel('(Data-Pred.)/Pred.', fontsize=20) ax2.set_xticks(bin_edges[::2]) ax2.tick_params(axis='both', which='major', labelsize=20) ax2.set_xlim((108, 140)) # Plot only relevant part of the signal ax2.set_ylim((-2, 2)) ax2.grid(True) f.tight_layout() if save_figs: plt.savefig('AsimovSignalFit.pdf')