(\Rightarrow) Let a random process X(\cdot) be an intrinsic random function with order d on the real line, and define
\lambda_{\Delta_{\iota,t}^{d}} := \sum_{k=0}^{d} (-1)^k \binom{d}{k} \delta_{t - k\iota}, \quad t,\iota \in \mathbb{R}.
Then, X(\lambda_{\Delta_{\iota,t}^{d}}) = \Delta_{\iota}^{d} X(t) and by Lemma 1, \lambda_{\Delta_{\iota,t}^{d}} is allowable with its degree d on \mathbb{R}.
By Definition 2.3 of an intrinsic random function, for an allowable measure \lambda_{\Delta_{\iota,t}^{d}}, \lambda_{\Delta_{\iota,s}^{d}} \in \Lambda_{d} and any t,s,x \in \mathbb{R},
\begin{aligned}
&E\bigl(X(\lambda_{\Delta_{\iota,t}^{d}}) \bigl) = E\bigl(X(\tau_x\lambda_{\Delta_{\iota,t}^{d}}) \bigl) \\
&Cov\bigl(X(\lambda_{\Delta_{\iota,t}^{d}}), X(\lambda_{\Delta_{\iota,s}^{d}}) \bigl) = Cov\bigl(X(\tau_x \lambda_{\Delta_{\iota,t}^{d}}), X(\tau_x \lambda_{\Delta_{\iota,s}^{d}}) \bigl)
\end{aligned}
Since X(\lambda_{\Delta_{\iota,t}^{d}}) = \Delta_{\iota}^{d} X(t) and X(\lambda_{\Delta_{\iota,s}^{d}}) = \Delta_{\iota}^{d} X(s), Equation (?) and (?) can also be expressed as
\begin{aligned}
&E\bigl( \Delta_{\iota}^{d} X(t) \bigl) = E\bigl( \Delta_{\iota}^{d} X(\tau_x t) \bigl)\\
&Cov\bigl( \Delta_{\iota}^{d} X(t), \Delta_{\iota}^{d} X(s) \bigl) = Cov\bigl( \Delta_{\iota}^{d} X(\tau_x t), \Delta_{\iota}^{d} X(\tau_x s) \bigl)
\end{aligned}
Hence, \Delta_{\iota}^{d} X(t) is stationary, which implies that X(t) is a random process with stationary increments of order d.
(\Leftarrow) Suppose X_d(t) is a random process with stationary increments of order d. Then, by its spectral representation given in Remark 1 and the condition in Remark 2,
X(t) = \int_{-\infty}^{\infty} \bigl( e^{it\omega} -1 - it\omega - \cdots - \frac{(i t \omega)^{d-1}}{(d-1)!} \bigl) dZ(\omega)
Then, for any allowable measure \lambda \in \Lambda_d with compact support,
\begin{aligned}
X(\lambda) &= \int_\mathbb{R} \int_{-\infty}^{\infty} \bigl( e^{it\omega} -1 - it\omega - \cdots - \frac{(i t \omega)^{d-1}}{(d-1)!} \bigl) dZ(\omega) \lambda(dt)\\
&= \int_{-\infty}^{\infty} \int_\mathbb{R} \bigl( e^{i t \omega} -1 - it\omega - \cdots - \frac{(i t \omega)^{d-1}}{(d-1)!} \bigl) \lambda(dt) dZ(\omega) \quad \text{(by Fubini's theorem).}\\
\end{aligned}
Since \lambda has compact support, the inner integral is finite. Furthermore, because X(t) is a random process (assumed L^2) and \lambda \in \Lambda_d, X(\lambda) is guaranteed to be in L^2(\Omega), where \mathbf{\Omega} denotes the sample space. This justifies the use of the Fubini’s theorem to interchange the order of stochastic and Lebesgue-Stieltjes integration. By the definition of an allowable measure, any \lambda \in \Lambda_{d} annihilates polynomials of its order d-1. Hence,
X(\lambda)=\int_{-\infty}^{\infty} \int_\mathbb{R} e^{i t \omega} \lambda(dt) dZ(\omega) = \int_\mathbb{R} \int_{-\infty}^{\infty} e^{i t \omega} dZ(\omega) \lambda(dt) \quad \text{(by Fubini's theorem again).}
Therefore, by defining Y^*(t) := \int_{-\infty}^{\infty} e^{it\omega} dZ(\omega), which is a stationary process for t \in \mathbb{R},
X(\lambda) = \int_\mathbb{R} Y^*(t) \lambda(t) = Y^*(\lambda)
Since Y^*(t) is stationary with respect to t, Y^*(\lambda) is stationary with respect to \lambda by Lemma 2. Therefore, X(\lambda) is stationary for \lambda as well. In conclusion, X(t) is an intrinsic random function of order d on the real line.