Direct Measurement of the Top Quark Mass

Abstract

We measure the top quark mass m_t using t tbar pairs produced in the D0 detector by root(s) = 1.8 TeV p pbar collisions in a 125 pb^-1 exposure at the Fermilab Tevatron. We make a two constraint fit to m_t in t tbar --> b W^+ bbar W^- final states with one W decaying to q qbar and the other to e nu or mu nu. Events are binned in fit mass versus a measure of probability for events to be signal rather than background. Likelihood fits to the data yield m_t = 173.3 +- 5.6 (stat) +- 6.2 (syst) GeV/c^2.

Figures (3)

• Figure 1: Events per bin vs. event selection variables defined in the text, plotted for (a--b, g--h) top quark mass analysis samples, and (c--f) $W$+3 jet control samples. Histograms are data, filled circles are expected top + background mixture, and open triangles are expected background only. Solid arrows in (a--b) show cuts applied to all events; the open arrow in (g) illustrates the LB cut. The nonuniform bin widths in (g--h) are chosen to yield uniform bin populations.
• Figure 2: Events per bin ($\propto$ areas of boxes) vs.$D_{\rm NN}$ (ordinate) and $m_{\rm fit}$ (abscissa) for (a) expected 172 GeV/c$^2$ top signal, (b) expected background, and (c) data. $D_{\rm NN}$ is binned as in Fig. \ref{['fig1']}(h).
• Figure 3: (a--b) Events per bin vs.$m_{\rm fit}$ for events (a) passing or (b) failing the LB cut. Histograms are data, filled circles are the predicted mixture of top and background, and open triangles are predicted background only. The circles and triangles are the average of the LB and NN fit predictions, which differ by $<$10%. (c) Log of arbitrarily normalized likelihood $L$vs. true top quark mass $m_t$ for the LB (filled triangles) and NN (open squares) fits, with errors due to finite top MC statistics. The curves are quadratic fits to the lowest point and its 8 nearest neighbors. In MC studies, 7% (27%) of simulated experiments yield a smaller LB (NN) maximum likelihood.