考虑非负实数域上某可用条件概率刻画的期望
$$E(x) = \int_{0}^{+\infty} xP(t = x|q) \mathrm{d}x
$$
其中 $q$ 为已知
考虑贝叶斯定理
$$E(x) = \int_{0}^{+\infty} x\frac{P(q|t = x)P(t = x)}{P(q)} \mathrm{d}x
$$
观察发现对 $x$ 微分本质上就是在枚举每个 $t = x$ 的情况
$$E(x) = \int_{0}^{+\infty} x\frac{P(q|t = x)}{P(q)} \mathrm{d}x
$$
大多数情况下
$$E(x) = \int_{0}^{+\infty} x\frac{P(q|t = x)}{\int_{0}^{+\infty} P(q|t = y) \mathrm{d}y } \mathrm{d}x
$$
整理可得
$$E(x) = \frac{\int_{0}^{+\infty} xP(q|t = x) \mathrm{d}x}{\int_{0}^{+\infty} P(q|t = x) \mathrm{d}x }
$$
设 $P(q|t = x) = f(x)$
$$E(x) = \frac{\int_{0}^{+\infty} xf(x) \mathrm{d}x}{\int_{0}^{+\infty} f(x) \mathrm{d}x } = \lim_{\varepsilon \to 0^+}^{m \to +\infty}\frac{\delta mf(m) - \delta \varepsilon f(\varepsilon)}{\delta f(m) - \delta f(\varepsilon)}
$$