本論文利用蒙地卡羅方法模擬日本高能加速器B介子工廠(KEKB)中之Belle偵測器所收集的771百萬個B介子對與背景來尋找B0介子衰變至四個輕子的事件,其中包含三種衰變模式B0→μ+μ-μ+μ-、B0→e+e-e+e-和B0→e+e-μ+μ-。我們使用NeuroBayes演算法來分離訊號與背景以提升訊噪比,接著再以B0介子衰變至一個J/ψ介子、一個K介子和一個π介子,J/ψ介子再衰變至一個輕子對的事件作為對照組來確認分析步驟是否正確並得到校正因子,然後利用校正因子作為B0→l+l-l+l-之中模擬與數據的校正。最後我們估計在90%的信賴區間之下的衰變機率上限為: B(B0→μ+μ-μ+μ-)<1.28×10E-7,B(B0→e+e-e+e-)<8.34×10E-8和B(B0→e+e-μ+μ-)<5.09×10E-8。
This thesis uses Monte Carlo method to simulate 771 million BB pairs and background collected in Belle detector at KEKB to search for B0→l+l-l+l-, it includes B0→μ+μ-μ+μ-, B0→e+e-e+e- and B0→e+e-μ+μ-. We use NeuroBayes algorithm to separate signal from background so that it enhances the signal-to-noise ratio, then we do B0→J/ψ(J/ψ→l+l-)K+π- as the control sample to check whether the analysis is correct or not and get the calibration factor, and then we use calibration factor to be the correction between simulation and data in B0→l+l-l+l-. Finally, we estimate the upper limit of branching fraction at 90% confidence interval: B(B0→μ+μ-μ+μ-)<1.28×10E-7, B(B0→e+e-e+e-)<8.34×10E-8 and B(B0→e+e-μ+μ-)<5.09×10E-8.