Abstract
We construct and study properties of an infinite dimensional analog of Kahane’s theory of Gaussian multiplicative chaos (Kahane in Ann Sci Math Quebec 9(2):105-150, 1985). Namely, if \(H_T(\omega )\) is a random field defined w.r.t. space-time white noise \(\dot{B}\) and integrated w.r.t. Brownian paths in \(d\ge 3\), we consider the renormalized exponential \(\mu _{\gamma ,T}\), weighted w.r.t. the Wiener measure \(\mathbb {P}_0(\textrm{d}\omega )\). We construct the almost sure limit \(\mu _\gamma = \lim _{T\rightarrow \infty } \mu _{\gamma ,T}\) in the entire weak disorder (subcritical) regime and call it subcritical GMC on the Wiener space. We show that
meaning that \(\mu _\gamma \) is supported almost surely only on \(\gamma \)-thick paths, and consequently, the normalized version is singular w.r.t. the Wiener measure. We then characterize uniquely the limit \(\mu _\gamma \) w.r.t. the mollification scheme \(\phi \) in the sense of Shamov (J Funct Anal 270:3224–3261, 2016) – we show that the law of \(\dot{B}\) under the random rooted measure \(\mathbb Q_{\mu _\gamma }(\textrm{d}\dot{B}\textrm{d}\omega )= \mu _\gamma (\textrm{d}\omega ,\dot{B})P(\textrm{d}\dot{B})\) is the same as the law of the distribution \(f\mapsto \dot{B}(f)+ \gamma \int _0^\infty \int _{\mathbb {R}^d} f(s,y) \phi (\omega _s-y) \textrm{d}s \textrm{d}y\) under \(P \otimes \mathbb {P}_0\). We then determine the fractal properties of the measure around \(\gamma \)-thick paths: \(-C_2 \le \liminf _{\varepsilon \downarrow 0} \varepsilon ^2 \log {\widehat{\mu }}_\gamma (\Vert \omega \Vert< \varepsilon ) \le \limsup _{\varepsilon \downarrow 0}\sup _\eta \varepsilon ^2 \log {\widehat{\mu }}_\gamma (\Vert \omega -\eta \Vert < \varepsilon ) \le -C_1\) w.r.t a weighted norm \(\Vert \cdot \Vert \). Here \(C_1>0\) and \(C_2<\infty \) are the uniform upper (resp. pointwise lower) Hölder exponents which are explicit in the entire weak disorder regime. Moreover, they converge to the scaling exponent of the Wiener measure as the disorder approaches zero. Finally, we establish negative and \(L^p\) (\(p>1\)) moments for the total mass of \(\mu _\gamma \) in the weak disorder regime.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
In this article, we construct and study properties of an infinite dimensional analog of the Gaussian multiplicative chaos (GMC) measures, namely, the measures
Here, \(\mathbb {P}_0\) stands for the Wiener measure corresponding to Brownian paths \(\omega : [0,\infty ) \mapsto \mathbb {R}^d\), and the Gaussian process \(\{H_T(\omega )\}_{\omega \in \Omega }\), indexed by Brownian paths, is driven by a Gaussian space-time white noise \(\dot{B}\) (under the probability measure \(\textbf{P}\)) integrated w.r.t. the Brownian path:
Here, \(\phi \) is a normalized mollifier. Developing this framework for GMC measures is quite natural and important in the field due to the relations to the continuous directed polymers as well as the multiplicative noise stochastic heat equation in \(d\ge 3\) [10, 25]. For this case, it is known that there is a non-trivial constant \(\gamma _c\in (0,\infty )\) such that the weak disorder or sub-critical phase is characterized by the uniform integrability of the martingale \(\{\mu _{\gamma ,T}(\Omega )\}_{T>0}\) when its almost sure limit \(\lim _{T\rightarrow \infty } \mu _{\gamma ,T}(\Omega )\) remains strictly positive. We note that the above definition of \(\gamma _c\) is rather implicit and that makes the weak disorder phase \((0,\gamma _c)\) harder to analyze.Footnote 1 Given this background, the following questions arise naturally:
-
Does the limiting measure \(\lim _{T\rightarrow \infty } \mu _{\gamma ,T}=\mu _\gamma \) exist, almost surely w.r.t. \(\dot{B}\), and in the entire weak disorder regime \((0,\gamma _c)\)?
-
Can we identify the support of the limit \(\mu _\gamma \) on the path space, and characterize its dependence explicitly in terms of the disorder \(\gamma \in (0,\gamma _c)\)?
-
Is there a way to characterize this limit in terms of the mollification scheme \(\phi \)?
-
How does the measure \(\mu _\gamma \) look like in a neighborhood of its support? Phrased differently, can the above construction yield quantitative and tractable (in terms of \(\gamma \in (0,\gamma _c)\) and \(\phi \)) information about the local fractal geometry of \(\mu _\gamma \) on “points" close to its support?
The goal of the present article is to answer the above questions. More precisely, our first main result in Theorem 2.1 shows that, for any \(d\ge 3\) and in the entire weak disorder regime \(\gamma \in (0,\gamma _c)\), the infinite volume GMC measure
exists, is non-trivial and non-atomic. This limit, which is taken w.r.t. the topology of weak convergence, also exists if the Gaussian noise \(\dot{B}\) is replaced by a random environment with finite exponential moment. Next, also in Theorem 2.1, we identify the support of this limit and show that
That is, for almost every realization of \(\dot{B}\) and for any Brownian path \(\omega \) sampled according to infinite-volume limit \(\mu _\gamma (\cdot )\), the value of the underlying field \(H_T(\omega )\) is atypically large – or, for any \(\gamma \in (0,\gamma _c)\), every path \(\omega \) is \(\gamma \)-thick w.r.t. \(\mu _\gamma \). Consequently, the normalized probability measure \({\widehat{\mu }}_\gamma =\mu _\gamma /\mu _\gamma (\Omega )\) is almost surely singular w.r.t. the base measure \(\mathbb {P}_0\).
We next investigate the universality of the limit (1.3) and determine the role of the cut-off \(\phi \). Following Kahane’s construction of log-correlated GMC, one expects that, in the current infinite dimensional setup, a well-defined limiting object should not depend so much on the choice of the mollifier. For this purpose, and to emphasize the role of \(\phi \), let us write \(\mu _{\scriptscriptstyle {\gamma ,H(\phi )}}\) for the infinite-volume limit (which is almost surely a functional of the field \(H(\phi )\)). In this vein, we first show that a strict uniqueness cannot hold in the current infinite dimensional setup – Proposition 3.12 implies that \(\mu _{\scriptscriptstyle {\gamma ,H(\phi )}}(\cdot )\ne \mu _{\scriptscriptstyle {\gamma ,H(\phi ^\prime )}}(\cdot )\) unless \(\phi \) and \(\phi ^\prime \) are identically equal. Thus, at very small scales, the infinite-volume limit still remembers how the field H was regularized (and in infinite dimensions, such a small dependence is conceivable). Then the question naturally arises if one can determine to what degree the limit \(\mu _{\gamma ,H(\phi )}\) depends on the mollifying scheme \(\phi \). In this regard, denote by \(P= \textbf{P}\dot{B}^{-1}\) the law of the white noise (a probability measure on tempered distributions \({\mathcal {S}}^\prime (\mathbb {R}_+\times \mathbb {R}^d)\)). In the third part of Theorem 2.1 we show that \(\mu _{\scriptscriptstyle {\gamma ,H(\phi )}}\) is the unique measure such that the distribution of \(H_T(\phi )\) under \(\mu _\gamma (\textrm{d}\dot{B},\textrm{d}\omega ) P(\textrm{d}\dot{B})\) is the same as the distribution of \(H_T(\phi )+T(\phi \star \phi )(0)\) under \(P\otimes \mathbb {P}_0\). In other words, the only way to perturb linearly the distribution \(\dot{B}\) with the test function
is by using the limiting GMC measure \(\mu _{\gamma ,H(\phi )}\). That is, the limit satisfies a “Cameron-Martin equation"
for all deterministic \(v: \Omega \mapsto \mathbb {R}\) so that the law of \(H(\phi )+v\) is absolutely continuous w.r.t. that of \(H(\phi )\). In other words, this limiting measure can be thought as a family \(\{\mu _{\gamma ,H(\phi )}\}_\phi \), where each member verifies the Cameron-Martin equation (1.5) for a fixed \(\phi \), as the field H (being a function of the noise \(\dot{B}\)) varies. This Cameron-Martin characterization (i.e., validity of (1.5)) is reminiscent of Shamov’s definition [33] of finite-dimensional GMC, see Remark 1. Shamov’s argument shows that, in finite-dimensions and for log-correlated fields, the solution to the Cameron-Martin equation is unique (see also the book by Berestycki and Powell [3, Sec. 3.4] where Shamov’s argument is revisited in a simpler way).
We subsequently deduce fractal properties of \(\mu _\gamma \) and deduce the volume decay or Hölder exponents of the normalized GMC measure \({\widehat{\mu }}_\gamma \) – namely, in Theorem 2.2 we show that, for weak disorder, and \(\textbf{P}\)-almost surely,
Here, \(\Omega _0\) is a subset of the paths carrying a (weighted) norm \(\Vert \cdot \Vert \) that makes \(\Omega _0\) a Banach space with \(\mathbb {P}_0(\Omega _0)=1\), and \(C_1\), \(C_2\) are explicit constants for weak disorder. While for a fixed \(\gamma \in (0,\gamma _c)\), the constants \(C_1\) and \(C_2\) do not match (and, given their nature, they should not match, as will be explained below), the bounds given in (1.6) agree in the limit \(\gamma \rightarrow 0\) and coincide with the scaling exponents of the Wiener measure \(\mathbb {P}_0\) as well. This will also be shown in Theorem 2.2. In Theorem 2.3, we also prove the existence of positive and negative moments of its total mass in the entire weak disorder regime, showing that for all \(\gamma \in (0,\gamma _c)\), there is some \(p>1\) such that \(\mu _\gamma (\Omega )\in L^p(\textbf{P})\) and also \(\mu _\gamma (\Omega ) \in L^{-q}\) for some \(q>0\).Footnote 2 Moreover, for \(\gamma \) even smaller (in the so-called \(L^2\)-regime), it holds that \(\mu _\gamma (\Omega )\) has negative moments of all order. Let us finally mention that, as in the first part of Theorem 2.1, Theorem 2.2 and Theorem 2.3 hold also for continuous directed polymers in random environments with finite exponential moments (i.e., the environment is not required to be Gaussian), while for the second and third parts of Theorem 2.1 we need the Gaussianity of the environment. We also refer to Sect. 2.4 for the main ideas of the proof.
In order to draw analogies, let us briefly recall Kahane’s theory of GMC for log-correlated fields on finite dimensional spaces. Given a domain \(D\subset \mathbb {R}^d\), a GMC is a rigorously defined version of the measures
Here, \(\{h(x)\}_{x\in D}\) is a log-correlated centered Gaussian field with \(\textbf{E}[h(x) h(y)]=-\log |x-y| + O(1)\). The logarithmic divergence along the diagonal prevents to define h pointwise, and a regularization process becomes necessary to define h, and consequently, the measures \(m_{\gamma ,h}\), in a precise sense. Since the work of Kahane [18], there have been very important works in the field by Robert and Vargas [29], Duplantier and Sheffield [15], Shamov [33] and Berestycki [2], who showed that if \((h_\varepsilon )_{\varepsilon \in (0,1)}\) is a suitable approximation of h, then as long as \(\gamma \in (0,\sqrt{2d})\),
This is the so-called subcritical regime. In the critical/supercritical phase, (i.e.,\(\gamma \ge \sqrt{2}\)), the measure cannot be constructed as above, but still one can make sense of it; we refer to the works of Duplantier-Rhodes-Sheffield-Vargas [13, 14], Powell [27, 28], Madaule-Rhodes-Vargas [24] and Biskup-Louidor [7] for the theory of critical and supercritical GMC, which is outside the scope of this work.
In the subcritical phase, \(m_{\gamma ,h}\) is non-trivial, non-atomic and \(\textbf{P}\)-a.s. and for a.e. \(x\in D\) sampled according to \(m_{\gamma ,h}\), it holds that \(h_\varepsilon (x) \sim \gamma \log \big (\frac{1}{\varepsilon }\big )\) – that is, almost every point chosen via \(\mu _{\gamma ,h}\) is \(\gamma \)-thick. Consequently, \(m_{\gamma ,h}\) is singular w.r.t. the Lebesgue measure. We underline the analogies of these statements to (1.3) and (1.4). There have been notable instances where, using the scale-invariance of the logarithmic correlations, studying the positive and negative moments of the total mass \(m_{\gamma ,h}(D)\) have been instrumental (see [16, 31] and [3, Ch. 3.7\(-\)3.9]). A related geometric feature of \(m_{\gamma ,h}\) is captured by the asymptotic volume decay of \(\log m_{\gamma ,h}(B_\varepsilon (x))\) as \(\varepsilon \downarrow 0\), which is the analogue of our bounds (1.6). This is known as the (uniform) Hölder exponents of \(m_{\gamma ,h}\) and it is closely related to its multifractal behavior. Indeed, recall that for \(m_{\gamma ,h}\), the uniform Hölder exponent is given by \(d(1-\frac{\gamma }{\sqrt{2d}})^2\), while its pointwise Hölder exponent is \(d+ {\gamma ^2}/2\) (see [30, Sec. 4.1] for precise definitions). However, none of these exponents fully captures its multifractal spectrum, which roughly says that if a point \(x\in D\) is a-thick, then
Thus, the pointwise (resp. uniform) Hölder scaling exponents are the extremal values of the multifractal spectrum, and these therefore do not match for a fixed \(\gamma \) (but do so as \(\gamma \rightarrow 0\)). We refer to the discussion below (1.6) again to underline the analogy to the Hölder exponents \(C_1, C_2\) in our setup.
In summary, it is for the first time, to the best of our knowledge, that the existence, characterization, thick points and the above fractal properties of \(\mu _\gamma \) in (1.3) have been established in the infinite-dimensional setup and in the entire weak disorder regime. As mentioned in the above paragraph, for log-correlated Gaussian fields on \({\mathbb {R}}^d\), critical parameters such as \(\gamma _c\) and sharp bounds for the positive moments of \(\mu _\gamma (\Omega )\) are well-established. In contrast, in our framework most of these parameters as well as the relevant information are qualitative in the entire weak-disorder regime. We underline that our results on the \(\gamma \)-thick paths, characterization of the measure and the Hölder exponents \(C_1,C_2\) provide tractable and quantitative information for weak disorder \(\gamma \in (0,\gamma _c)\). Our objective is to leverage the GMC approach to derive quantifiable estimates that, until now, have only been validated in a limited number of integrable models. Let us now turn to the precise mathematical layout of the model.
1.1 The model and notation
For a fixed dimension \(d \ge 3\), let \(\Omega := {C}\left( [0,\infty ), \mathbb {R}^d \right) \) be the space of continuous functions from \([0, \infty )\) to \(\mathbb {R}^d\), endowed with the topology of uniform convergence on compact sets. We equip this space with the Wiener measure denoted by \(\mathbb {P}_0\), so a typical path \(\omega =(\omega _s)_{s\in [0,\infty )} \in \Omega \) corresponds to a realization of a \(\mathbb {R}^d\)-valued Brownian motion starting at 0. Similarly, we denote by \(\mathbb {P}_x\) the Wiener measure corresponding to a Brownian motion starting at \(x\in \mathbb {R}^d\). Let \(({\mathcal {E}}, \mathcal {F}, \textbf{P})\) be a complete probability space so that \(\dot{B}\) is a space-time white noise independent of the Wiener measure. More precisely, denote by \({\mathcal {S}}(\mathbb {R}_+ \times \mathbb {R}^d)\) the space of rapidly decaying Schwartz functions on \(\mathbb {R}_+ \times \mathbb {R}^d\). Then \(\dot{B}= {\dot{B}} (f)_{f \in {\mathcal {S}}(\mathbb {R}_+ \times \mathbb {R}^d)}\) is a Gaussian process with mean zero and covariance
Here, we use the notation
We can extend the integral to \(f \in L^2(\mathbb {R}_+ \times \mathbb {R}^d)\) via approximation. Indeed, if \((f_n)_{n \in \mathbb {N}} \subset {\mathcal {S}}(\mathbb {R}_+ \times \mathbb {R}^d)\) approximates f in \( L^2(\mathbb {R}_+ \times \mathbb {R}^d)\) (such a sequence exists since \( {\mathcal {S}}(\mathbb {R}_+ \times \mathbb {R}^d)\) is dense in \( L^2(\mathbb {R}_+ \times \mathbb {R}^d)\)), then the sequence \(({\dot{B}} (f_n))_n\) is Cauchy in \(L^2(\textbf{P})\), since we have
Therefore, for every \(f \in L^2(\mathbb {R}_{+}\times \mathbb {R}^d)\), we can define the \(L^2(\textbf{P})\)-limit
For \(f \in L^2(\mathbb {R}_+\times \mathbb {R}^d)\) and \(T > 0\), we will write
By construction, for \(f \in L^2(\mathbb {R}_+\times \mathbb {R}^d)\), the random variable \(\int _{0}^{\infty } \int _{ \mathbb {R}^d} f(s,y ){\dot{B}} (s,y) \, \textrm{d}y \, \textrm{d}s\) is Gaussian distributed with mean zero and variance \(\Vert f\Vert _{L^2(\mathbb {R}_+\times \mathbb {R}^d)}\) and for \(f,g \in L^2(\mathbb {R}_+\times \mathbb {R}^d)\), the covariance \(\textbf{E}[\dot{B}(f) \dot{B}(g)]\) is also given by (1.7). We also define the family of space-time shifts \(\{\theta _{t,x}:t>0, x\in \mathbb {R}^d\}\) acting on the white-noise environment (i.e., \(\theta _{t,x}(\dot{B} (s, y))=\dot{B}(t+s, x+y)\)), and remind the reader that \(\dot{B}\) is stationary under this action.
Next, let \(\phi \) be a mollifier – that is, a smooth, non-negative, spherically symmetric and compactly supported function \(\phi :\mathbb {R}^d \mapsto \mathbb {R}\) such that \(\int _{\mathbb {R}^d} \phi (x)\textrm{d}x = 1\). We define the Gaussian field \(\left( {H}_T(\omega ) \right) _{\omega \in \Omega }\) as the Itô integral
In particular, \(( {H}_T(\omega ))_{\omega \in \Omega }\) is also a Gaussian process with mean zero and covariance
where \((f\star g)(x) = \int _{ \mathbb {R}^d}f(x-y)g(y)\textrm{d}y\). Note that, for any \(\omega \in \Omega \), \(\textrm{Var}[H_T(\omega )]= T (\phi \star \phi )(0)\) – that is, for \(T\gg 1\) large, the covariance of the field \((H_T(\omega ))_{\omega \in \Omega }\) diverges along the diagonal.
Given any \(\gamma >0\) and \(T>0\), we define the random measure on the path space \(\Omega \)
and in the sequel, we will refer to \(\mu _{\gamma ,T}\) as the finite-volume Gaussian multiplicative chaos on \(\Omega \) or the continuous directed polymer. Also, we will write
for its normalized counterpart. Let \({\mathcal {F}}_T\) be the \(\sigma \)-algebra generated by the noise \(\dot{B}\) until time T. Since \(\dot{B}\) is smoothened only in space, the total mass, or the partition function,
is adapted to the filtration \(({\mathcal {F}}_T)_T\) and is a martingale. Our main results will hold in the uniform integrability phase \(\gamma \in (0,\gamma _c)\) (also called weak disorder) of the martingale \((\mu _{\gamma ,T}(\Omega ))_T\) determined by the critical value \(\gamma _c \in (0,\infty )\) for \(d\ge 3\) defined in (3.3). We are now ready to state our main results.
2 Main results
2.1 Existence, support and characterization
For any fixed realization of \(\dot{B}\) and \(a>0\), we define
to be the (random) set of a-thick paths \(\omega \in \Omega \). Here is our first main result:
Theorem 2.1
Fix \(d\ge 3\) and \(\gamma \in (0, \gamma _c)\). Then the following hold:
-
(Existence) There exists a non-trivial measure \(\mu _{\gamma }\) on \(\Omega \) such that \(\textbf{P}\)-a.s., \(\mu _{\gamma , T}\) converges weakly to \(\mu _{\gamma }\) as \(T \rightarrow \infty \). Moreover, for any fixed Borel set \(A\subset \Omega \) with \(\mathbb {P}_0(A)>0\), \(\mu _\gamma (A)>0\) \(\textbf{P}\)-a.s. Similarly, there exists a probability measure \(\widehat{\mu }_\gamma \) on \(\Omega \) such that \(\textbf{P}\)-a.s., the normalized approximations \(\widehat{\mu }_{\gamma ,T}\) converge weakly to \(\widehat{\mu }_\gamma \). Finally, the same conclusion holds if the noise \(\dot{B}\) is replaced by any other random environment with finite exponential moments.
-
(Identification of thick points and singularity) Let \(\mu _\gamma \) be the limiting measure from above. Then,
$$\begin{aligned} \mu _\gamma \bigg \{\omega \in \Omega : \lim _{T\rightarrow \infty } \frac{H_T(\omega )}{T (\phi \star \phi )(0)} \ne \gamma \bigg \}=0 \qquad \textbf{P}\text{-a.s. } \end{aligned}$$(2.1)That is, almost surely, any path \(\omega \in \Omega \) sampled according to \(\mu _\gamma \) is \(\gamma \)-thick. Consequently, for \(\textbf{P}\)-a.e. realization of \(\dot{B}\), the normalized probability measure \({\widehat{\mu }}_\gamma =\mu _\gamma /\mu _\gamma (\Omega )\) (also defined above) is singular w.r.t. the Wiener measure \(\mathbb {P}_0\). In particular, if \(\phi \) and \(\phi ^\prime \) are two mollifiers, then the limiting measures \(\mu _{\gamma ,\phi }\) and \(\mu _{\gamma ,\phi ^\prime }\) are singular unless \((\phi \star \phi )(0)=(\phi ^\prime \star \phi ^\prime )(0)\).
-
(Characterization) Let \(P=\textbf{P}\circ \dot{B}^{-1}\) be the law of the white noise, and for any measure \(\nu \) on \(\Omega \), which may depend on \(\dot{B}\), we write
$$\begin{aligned} \mathbb {Q}_\nu (\textrm{d}\dot{B}\textrm{d}\omega ):=\nu (\textrm{d}\omega ,\dot{B})P(\textrm{d}\dot{B}). \end{aligned}$$(2.2)Then the (unnormalized) GMC measure \(\mu _\gamma = \mu _{\gamma ,\phi }\) is the unique measure such that the law of \({\dot{B}} \) under \(\mathbb {Q}_{\mu _{\gamma ,\phi }}\) is the same as the law of the (Schwartz) distribution
$$\begin{aligned} {{\dot{B}} }_\phi (f)={\dot{B}} (f)+\gamma \int _{\mathbb {R}_+\times \mathbb {R}^d}f(s,y)\phi (\omega _s-y)\textrm{d}s \textrm{d}y, \qquad f \in {\mathcal {S}}(\mathbb {R}_+\times \mathbb {R}^d) \end{aligned}$$under \(P\otimes \mathbb {P}_0\).Footnote 3 In other words, \(\mu _{\gamma ,\phi }\) is the unique measure satisfying
$$\begin{aligned} \mathbb {E}^P\bigg [\int _\Omega \mu _{\gamma ,\phi }(\textrm{d}\omega ) F(\dot{B}, \omega )\bigg ]= \mathbb {E}^{P\otimes \mathbb {P}_0}\big [F(\dot{B}_\phi , \omega )\big ]. \end{aligned}$$(2.3)for any bounded measurable function \(F: \Omega \times {\mathcal {E}} \mapsto \mathbb {R}\) (cf. Remark 1 below).
Remark 1
We comment on the third part of Theorem 2.1 which determines to what extent the limiting measure \(\mu _\gamma \) depends on the choice of the mollification \(\phi \). Let us first note that, in the infinite-dimensional setup, the measure \(\mu _{\gamma ,\phi }\) does depend on the mollification \(\phi \) in a strict sense, and this dependence is to be expected – indeed, it will be shown in Proposition 3.12 that for two mollifiers \(\phi \) and \(\phi ^\prime \), even if \((\phi \star \phi )(0)=(\phi ^\prime \star \phi ^\prime )(0)\), the two measures \(\mu _{\gamma ,\phi }\) and \(\mu _{\gamma ,\phi ^\prime }\) are different unless \(\phi =\phi ^\prime \). However, this dependence on the approximation procedure is rather small – as shown by the third statement in Theorem 2.1, the only way to perturb linearly the distribution \(\dot{B}\) with the test function
is by using the measure \(\mu _{\gamma ,\phi }\). In particular,
Note that this characterization of \(\mu _\gamma \) is reminiscent of Shamov’s definition of a subcritical GMC [33]. Under this setup, a GMC over the (generalized) Gaussian field H is a random measure \(\mu =\mu _H\) (measurable w.r.t. H) on a measure space \((X,{\mathcal {G}})\) if for all deterministic \(v:X\mapsto \mathbb {R}\) such that the law of \(H+v\) is absolutely continuous w.r.t. that of H, then \(\mu _{H+v}(dx)=\textrm{e} ^{v(x)}\mu _{H}(dx)\). The construction of such measure goes via approximating H by a sequence of fields \((H_n)\) such that \(H_n\rightarrow H\) in a suitable sense. A difference to the present infinite dimensional setup is that the limiting field does depend on the mollifier \(\phi \), since for \(\omega \ne \omega '\in \Omega \),
2.2 Volume decay and Hölder exponents
Recall that by (2.1), \(\mu _\gamma \) is supported only \(\gamma \)-thick paths \(\omega \in \Omega \). Our next goal is to determine the asymptotic decay rate of small balls around each such \(\gamma \)-thick path, that is to determine the behavior of
and compare this decay rate with that of \(\log \mathbb {P}_0(\omega \in \Omega :\Vert \omega \Vert <\varepsilon )\) w.r.t. the Wiener measure \(\mathbb {P}_0\).
Note that \(\Omega =C([0,\infty ); \mathbb {R}^d)\) is not a Banach space.Footnote 4 To define a suitable norm, we will consider a subset \(\Omega _0\subset \Omega \) with a weighted norm \(\Vert \cdot \Vert _w\) such that \(\mathbb {P}_0(\Omega _0)=1\) and that \((\Omega _0, \Vert \cdot \Vert _w)\) is a Banach space. For this purpose, we consider a class of maps \(w:(0,\infty )\mapsto (0,\infty ]\) satisfying
For a fixed w as above, we set
Since \(\mathbb {P}_0\left( \lim _{t\rightarrow \infty }\frac{|\omega _t|}{t}=0\right) =1\), condition (2.5) assures that \(\mathbb {P}_0(\Omega _0)=1\). Moreover, under the norm \(\Vert \cdot \Vert _w\), defined as
\(\Omega _0\) is a separable Banach space. From now on, we consider an arbitrary w satisfying (2.5).
The next theorem consists of two parts. In the first one, we will determine the decay exponents (uniform upper and pointwise lower estimates) for (2.4) valid in the whole weak disorder region \(\gamma \in (0,\gamma _c)\). In the second one, we will show that, as \(\gamma \rightarrow 0\), the upper and lower estimates coincide.
Theorem 2.2
(Volume decay) Fix \(d\ge 3\).
-
Given \(\gamma \in (0,\gamma _c)\) and w satisfying (2.5), there exists \(r_0>0\) such that for all \(r\in (0,r_0)\), there are explicit constants \(0<C_1\le C_2<\infty \) (defined in (4.13) and (4.24)) fulfilling
$$\begin{aligned} -C_2\le & {} \liminf _{\varepsilon \rightarrow 0}\varepsilon ^2\log \widehat{\mu }_\gamma (\Vert \omega \Vert _w<r\varepsilon ) \nonumber \\\le & {} \limsup _{\varepsilon \rightarrow 0}\varepsilon ^2\sup _{\eta \in \Omega _0}\log \widehat{\mu }_\gamma (\Vert \omega -\eta \Vert _w<r \varepsilon )\le -C_1. \end{aligned}$$(2.8) -
Let \(p_0>1\) and \(q_0>0\). Then if \(\gamma \) is small enough, the constants \(C_1,C_2\) can be chosen as
$$\begin{aligned} \begin{aligned} C_1(\gamma ,d,r)&:=\frac{p_0-1}{p_0 }\bigg (\frac{j^2_{\frac{d-2}{2}}}{2r^2}\int _{0}^\infty w^{-2}(t)\textrm{d}t\bigg )-\frac{1}{2p_0}\frac{\gamma ^2}{2}(\phi \star \phi )(0),\\ C_2(\gamma ,d,r)&:=\bigg (\frac{q_0+1}{q_0}\bigg )\bigg (\frac{j^2_{\frac{d-2}{2}}}{2r^2}\int _{0}^\infty w^{-2}(t)\textrm{d}t\bigg )+\frac{\gamma ^2}{2}(\phi \star \phi )(0), \end{aligned} \end{aligned}$$(2.9)where \(j_{\frac{d-2}{2}}\) is the smallest positive root of the Bessel function \(J_{\frac{d-2}{2}}\). \(r>0\),
$$\begin{aligned} \lim _{\gamma \rightarrow 0}C_1(\gamma ,d,r)=\lim _{\gamma \rightarrow 0}C_2(\gamma ,d,r)= & {} \frac{j^2_{\frac{d-2}{2}}}{2r^2}\int _{0}^\infty w^{-2}(t)\textrm{d}t \nonumber \\= & {} \lim _{\varepsilon \rightarrow 0}\varepsilon ^2\log \mathbb {P}_0(\Vert \omega \Vert _w<r\varepsilon ). \end{aligned}$$(2.10)In other words, both exponents \(C_1\) and \(C_2\) converge as \(\gamma \rightarrow 0\) to the volume decay exponent for the Wiener measure.
Remark 2
We note that the upper bound in (2.8) is uniform over shifted balls of a given radius, while the lower bound holds pointwise. In fact, we do not expect a uniform lower bound over \(\eta \in \Omega _0\) in Theorem 2.2. This is reflected already in the behavior of Wiener measure \(\mathbb {P}_0\) (corresponding to the case \(\gamma =0\)): for every function in the Cameron-Martin space, defined as the Hilbert space
with inner product induced by the norm \(\Vert \eta \Vert _{H^1}:=\Vert \dot{\eta }\Vert _{L^2([0,\infty );\mathbb {R}^d)}\), we have (see (4.15) in Lemma 4.2)
That is, we obtain a pointwise lower bound w.r.t. \(\mathbb {P}_0\).
Remark 3
We also remark that instead of considering weighted norms on \(\Omega _0\), it is also possible to obtain analogous estimates as in Theorem 2.2 on \(\Omega \) by considering the sets \(\{\omega \in \Omega : \sup _{0\le t\le r}|\omega _t|<\varepsilon \}\) (i.e., by considering balls in the uniform metric in [0, r]) as \(\varepsilon \rightarrow 0\).
2.3 Moments of \(\mu _\gamma (\Omega )\) in weak disorder
Finally we turn to the positive and negative moments of the total mass \(\mu _\gamma (\Omega )\). By the martingale convergence theorem, we always have \(\textbf{E}[\mu _\gamma (\Omega )]=1\) for all \(\gamma \in (0,\gamma _c)\). Moreover, we also have
Theorem 2.3
(Positive and negative moment) Fix \(d\ge 3\) and \(\gamma \in (0,\gamma _c)\). Then there is some \(p>1\) and \(q>0\) such that
Moreover, if \(\gamma \in (0,\gamma _c)\) is chosen so that the martingale \((\mu _{\gamma ,T}(\Omega ))_T\) is bounded in \(L^2(\textbf{P})\), then
Remark 4
We can show (2.11) and (2.12) exactly in the same manner if \(\dot{B}\) is replaced by a random environment with finite exponential moments. In [11], it was shown (using Talagrand’s concentration inequality and for Gaussian random environment) that in the “\(L^2\) region" (i.e., when \(\gamma \in (0,\gamma _c)\) is restricted so that the martingale \((\mu _{\gamma ,T}(\Omega ))_T\) remains \(L^2(\textbf{P})\) bounded) and for all \(q\in (-\infty ,0)\), \((\mu _{\gamma ,T}(\Omega ))_T\) is \(L^q(\textbf{P})\) bounded. It remains an open problem to extend (2.13) in the full weak disorder regime \(\gamma \in (0,\gamma _c)\). For negative moments of the total mass (or more generally, the tail distribution of the inverse of the partition function) of finite dimensional log-correlated GMC we refer to [16, 29].
2.4 Ideas of the proofs and comparison with literature
As a guidance to the reader, we will now outline the key technical constituents of the proofs of the above results. In this description, we will also underline similarities and differences from existing techniques in the literature.
Step 1 (Weak convergence): For the existence part of the proof of Theorem 2.1, we have drawn inspiration from the simple and very elegant approach of Berestycki [2] for the construction of GMC for log-correlated fields on finite dimensional spaces. This approach goes via employing a second moment method to prove convergence in the \(L^2\) region, then using Girsanov’s formula to determine thick points and showing that every Liouville point \(x\in D\) is \(\gamma \)-thick, and finally covering the entire uniform integrability phase by removing points which are thicker than this prescribed value. In the current setup, we approach the uniform integrability phase directly by using martingale properties. The starting point is that, for any Borel set \(A\subset \Omega \), \((\mu _{\gamma ,T}(A))_{T\ge 0}\) is a martingale w.r.t. the white noise filtration (see Lemma 3.1 and Lemma 3.2). This guarantees convergence of \(\mu _{\gamma ,T}(A)\) \(\textbf{P}\)-a.s. and in \(L^1(\textbf{P})\) to some random variable \(\mu _\gamma (A)\) in the uniform integrability phase (the same statement holds for the normalized version \({\widehat{\mu }}_{\gamma ,T}(A)\rightarrow \mu _\gamma (A)/\mu _\gamma (\Omega )\)). However, it is not clear that the limiting random variable defines a probability measure, so we cannot derive weak convergence just from this fact. Thus, we first prove that the sequence \(({\widehat{\mu }}_{\gamma ,T})_{T\ge 0}\) is tight, in order to obtain a weakly convergent subsequence. For this purpose, first it is necessary to find a countable collection \({\mathcal {X}}\) of Borel sets in \(\Omega \) where the almost sure convergence \({\widehat{\mu }}_{\gamma ,T}(A)\rightarrow \mu _\gamma (A)/\mu _\gamma (\Omega )\) takes places simultaneously for all \(A\in {\mathcal {X}}\). Using this construction and a continuity property of the limiting random variables \(\mu _\gamma (\cdot )\), we then show that \(\textbf{P}\)-a.s., the family \(\{{\widehat{\mu }}_{\gamma ,T}\}_T\) is uniformly tight, and moreover, the weak limit of \({\widehat{\mu }}_{\gamma ,T}\) is uniquely determined. We refer to Proposition 3.8 for details. It might be worth pointing out that this approach is robust in the sense that it does not rely on a \(L^2\) computation, and it does not need Gaussianity of the environment. These arguments constitute Sects. 3.1, 3.2, 3.3. See also Remark 5 below for a comparison with existing literature.
Step 2 (Rooted random measures): In the second step for the proof of Theorem 2.1, we work with the measures
Using Theorem 2.1, we show that \(\mathbb {Q}_{\mu _{\gamma ,T}}(\textrm{d}\omega \textrm{d}\dot{B})\) converges weakly towards \(\mathbb {Q}_{\mu _\gamma }(\textrm{d}\omega \textrm{d}\dot{B})\), and apply Girsanov’s theorem to determine, under the conditional measure \(\mathbb {Q}_{\mu _\gamma }(\textrm{d}\omega |\dot{B})\), the mean and the variance the Gaussian process \(\{\dot{B}(f)\}_{f\in {\mathcal {S}}(\mathbb {R}_+\times \mathbb {R}^d)}\). Combining these two facts, it can be shown that the law of \(\dot{B}\) under \(\mathbb {Q}_{\mu _\gamma }\) is the same as the law of \(\dot{B}(f)+ \int _{\mathbb {R}_+\times \mathbb {R}^d} f(s,y) \phi (\omega _s- y) \textrm{d}s \textrm{d}y\) under \(\mathbb {P}_0\otimes P\). Combined with a law of large numbers of the martingale
it can then be shown that
These arguments constitute the second and third parts of the proof of Theorem 2.1 in Sect. 3.4.
Step 3 (Volume estimates): In this step we will show decay rate of \({\widehat{\mu }}_\gamma (\Vert \omega - \eta \Vert _w < \varepsilon )\) of microscopic balls in the weighted norm \(\Vert \cdot \Vert _w\) for any \(\gamma \in (0,\gamma _c)\). Note that because of uniform integrability, the total mass \(\log (\mu _\gamma (\Omega ))>-\infty \) a.s., so it suffices to estimate the unnormalized volume \(\mu _\gamma (\Vert \omega - \eta \Vert _w < \varepsilon )\). For the upper bound, we approximate this quantity by \(\log \mu _{\gamma , S+ \varepsilon ^{-2}}(\Vert \omega - \eta \Vert _{w} < \varepsilon )\) for any fixed \(\varepsilon >0\) and large \(S\gg 1\), and use Markov property and Hölder’s inequality to split the latter term into three contributions, namely, the uniform small ball probabilities \(\sup _\eta \log \mathbb {P}_0[\Vert \omega - \eta \Vert _{w,\varepsilon } < r \varepsilon ]\) (under a truncated weighted norm \(\Vert \cdot \Vert _{w,\varepsilon }\)), the uniform space-time translations
and then the contribution (up to time \(\varepsilon ^{-2}\)) from \(\mu _{\varphi (\gamma ), \varepsilon ^{-2}}(\Omega )\) for a disorder \(\varphi (\gamma )\) depending on the exponents coming from Hölder’s inequality. The first term \(\sup _\eta \log \mathbb {P}_0[\Vert \omega - \eta \Vert _{w,\varepsilon } < r \varepsilon ]\) is handled by log-concavity of Gaussian measures and an asymptotic small ball probability in the weighted norm; this provides a rate involving Bessel functions. To handle the second term, we crucially use the space-time stationarity of the white noise \(\dot{B}\) – this fact, together with Doob’s inequality and the \(L^p(\textbf{P})\) moment (2.11) for any \(\gamma \in (0,\gamma _c)\), dictate that the ergodic theorem for stationary processes is applicable, implying that the LHS of (2.14) is dominated by \(C(\dot{B}) \varepsilon ^{-2}\) almost surely w.r.t. \(\textbf{P}\), thereby removing the influence of this term on a logarithmic scale.
For the third term, the asymptotic behavior of \(\varepsilon ^2 \log \mu _{\gamma ,\varepsilon ^{-2}}(\Omega )\), for any \(\gamma >0\), is provided by a tractable variational formula over the translation-invariant compactification \(\widetilde{{\mathcal {X}}}\) of probability measures introduced in [26] – briefly, this space consists of elements \(\xi =\{{\widetilde{\alpha }}_i\}_i\) of orbits \({\widetilde{\alpha }}_i\) of sub-probability measures \(\alpha _i\) with \(\sum _i \alpha _i(\mathbb {R}^d)\le 1\), see Sect. 4.2. For our purposes, it might be worthwhile explaining the structure of this formula. First, we can decompose \(\log \mu _{\gamma ,\varepsilon ^{-2}}(\Omega )\) in terms of a martingale and an additive (and shift-invariant) energy functional of the probability distributions \({\widehat{\mu }}_{\gamma ,\varepsilon ^{-2}}[\omega _{\varepsilon ^{-2}}\in \cdot ]\), but embedded into the aforementioned compactification \(\widetilde{{\mathcal {X}}}\), and these distributions also define a dynamics defined by transition probabilities. This dynamics was introduced in [1] as a fixed point approach related to the cavity method from spin glasses for discrete time random walks on \(\mathbb {Z}^d\) over the aforementioned compactified space (this used the countability of \(\mathbb {Z}^d\) by employing a different, but equivalent, metric to the one in [26]). In the present context, we will follow the approach developed in [6] which directly used the precise metric structure from [26] in the continuum showing that the above Markovian transition kernel is continuous w.r.t. the metric on \(\widetilde{{\mathcal {X}}}\), implying existence of invariant measures (over probability measures \({{\mathcal {M}}}_1(\widetilde{{\mathcal {X}}})\) on \(\widetilde{{\mathcal {X}}}\)). The key aspect here hinges upon the notion of total disintegration of mass as well as a decoupling phenomenon of two independent GMC chains at large distances, both traits being compatible with the topology of \(\widetilde{{\mathcal {X}}}\). Then the aforementioned representation of \(\varepsilon ^{2} \log \mu _{\gamma , \varepsilon ^{-2}}(\Omega )\) and the continuity of the above map provide, for any temperature \(\gamma >0\), a tractable quenched variational formula \(\varepsilon ^{2} \log \mu _{\gamma , \varepsilon ^{-2}}(\Omega )= - \sup _{\vartheta } \int _{\widetilde{{\mathcal {X}}}} \Phi _{\gamma }(\xi ) \vartheta (\textrm{d}\xi )\) almost surely w.r.t. \(\textbf{P}\). Here
The supremum in the variational formula is taken (and attained) over the invariant measures of the dynamic in \({{\mathcal {M}}}_1(\widetilde{{\mathcal {X}}})\). The maximizer(s) depend on the temperature \(\gamma \) and undergoes a qualitative change as \(\gamma \) varies, see Sect. 4.2 for details. Combining these estimates will provide the required upper bound in Theorem 2.2. Note that since the \(L^p\) bound in (2.11) holds for some (possibly implicit) \(p>1\), the subsequent application of the Hölder’s inequality is valid for this (pre-determined) \(p>1\). In this vein, the structure of the formula (in terms of \(\gamma \) and p) is instrumental in providing a non-trivial uniform upper bound \(\exp (-C_1/\varepsilon ^2)\) with an explicit \(C_1>0\), for any \(\gamma \in (0,\gamma _c)\) and \(r>0\) sufficiently small. We refer to Proposition 4.1 for details.
We also remark that an orthogonal approach for handling asymptotic behavior of the logarithmic partition function goes via exploiting its sub-additivity [9, 12], which, unlike the present approach, does not immediately yield an explicit limiting object.
Now, for the proof of the lower bound in Theorem 2.2 we will apply reverse Hölder’s inequality to split the probabilities also into three factors similar to the one above. The applicative of this inequality then requires a negative moment from (2.12), and then the space-time translation invariance of the white noise allows the use of the ergodic theorem again. However, now a concentration bound coming from stochastic calculus allows us to replace the quenched probabilities \(\varepsilon ^{2} \log \mu _{\gamma , \varepsilon ^{-2}}(\Vert \omega \Vert _{w,\varepsilon }< r\varepsilon )\) by its annealed counterpart \(\varepsilon ^{2} \textbf{E}[\log \mu _{\gamma , \varepsilon ^{-2}}(\Vert \omega \Vert _{w,\varepsilon }< r\varepsilon )]\), and subsequently an application of Jensen’s inequality removes any contribution coming from this term. This leads to a lower bound \(\exp (-C_2/\varepsilon ^2)\) for any \(\gamma \in (0,\gamma _c)\) with \(C_2 <\infty \), see Proposition 4.3. Similar to \(C_1\), the lower bound constant \(C_2\) is also explicit, but unlike the upper bound, \(C_2\) carries only two contributions coming from the normalization constant and from the exit time estimate for small ball probability under the Wiener measure. As remarked earlier, the upper and lower bounds \(C_1\) and \(C_2\) differ for a fixed \(\gamma \in (0,\gamma _c)\), but as \(\gamma \rightarrow 0\), these exponents coincide with an explicit constant involving the Bessel function (the scaling exponent of the Wiener measure \(\mathbb {P}_0\)), see Corollary 4.4.
Finally, let us mention that it is shown in [5] that, for any \(\gamma >0\) and \(d\ge 1\), the finite-volume measure \({\widehat{\mu }}_{\gamma ,T}\) decays exponentially on small balls (in the uniform metric on paths in [0, T]). Even though the infinite-volume limit of \({\widehat{\mu }}_{\gamma ,T}\) exists for \(\gamma \in (0,\gamma _c)\), it is not clear how a similar property for \(\varepsilon ^2 \log {\widehat{\mu }}_\gamma (\Vert \omega \Vert < \varepsilon )\) can be deduced from the statements there directly.
Step 4 (Moments for \(\gamma \in (0,\gamma _c)\)): It remains to show the estimates in Theorem 2.3. For this purpose, we will adapt an argument which was developed recently in [17] to show that (2.11) and (2.12) hold for discrete polymers in a bounded environment in weak disorder. This argument involved exploiting properties of a non-negative martingale \((M_n,{\mathcal {F}}_n)_n\) satisfying three assumptions: (i) First, for any \(k,\ell \in \mathbb {N}\), \(\textbf{E}[f(M_{k+\ell }/M_k)| {\mathcal {F}}_k]\le \textbf{E}[f(M_\ell )]\) for any convex function f (as mentioned there, this condition is originally known to be satisfied by well-known branching processes, e.g. Galton-Watson processes); (ii) If \(\textbf{P}[\lim _n M_n>0]>0\) and condition (i) holds, then \(\textbf{E}[M_\infty ^\star ]< \infty \), where \(M_n^\star =\sup _{0\le k \le n} M_k\) is the running maximum; (iii) If \(\textbf{P}[\lim _n M_n>0]>0\) and condition (i) holds, and moreover if for some \(K<\infty \), we have \(M_{n+1}/M_n \le K\) for all n, then \(\sup _n \Vert M_n\Vert _p < \infty \) for some \(p>1\). These three properties are satisfied by the partition function of the discrete directed polymer in a bounded environment. For the continuous directed polymer (in an unbounded environment)Footnote 5 we will modify the first step and show that the martingale \((\mu _{\gamma ,T}(\Omega ))_T\), satisfies, for a suitable passage time \(\tau =\tau _u:= \inf \{t: \mu _{\gamma ,t}(\Omega )=u\}\) and for any \(\gamma >0\), \(\textbf{E}\big [f\big (\mu _{\gamma ,T}(\Omega )/\mu _{\gamma ,\tau }(\Omega )\big ); \tau \le T\big ] \le \textbf{P}[\tau \le T] \textbf{E}[f(\mu _{\gamma ,T}(\Omega ))]\) (see Lemma 5.2). This in turn will imply that for \(\gamma \in (0,\gamma _c)\), we have \(\textbf{E}[M_\infty ] < \infty \), where \(M_\infty =\lim _{T\rightarrow \infty } \, \sup _{0\le S \le T} \mu _{\gamma ,S}(\Omega )\) (see Lemma 5.1). These two modifications will subsequently suffice to show (2.11) for some \(p>1\), without requiring the third assumption above.
Remark 5
To compare our Step 1 (sketched above) with existing literature, let us mention that, for discrete (in time and in space \(\mathbb {Z}^d\)) directed polymers, Comets and Yoshida [12, Sec. 4] observed that the normalized polymer measure \(\mu _n\) at time n defines a time-inhomogeneous Markov chain with transition probabilities
Here, P is the law of a simple random walk on \(\mathbb {Z}^d\), and \(W_n\) the partition function that also converges a.s to \(W_\infty >0\). Thus, by using the existence of the limit \(\lim _{n\rightarrow \infty }\mu _n(A)\) for \(A\in \sigma \left( \bigcup _n {\mathcal {F}}_n\right) \), it can be easily shown that there is a limiting measure \(\mu \) on \(\sigma \)-algebra generated by finite paths, and this limiting measure is a Markov chain with
Note that, \(\mu _n\) defined in (2.15) is time-inhomogeneous and also inconsistent, as the transition probabilities depend on the time horizon n. Unless \(\beta =0\), there exists no Markov chain on the set of paths with infinite length such that, for any \(n\ge 1\), the marginal on the set of paths of length n is \(\mu _n\), see the discussion in Comets [8, p.15, Sec. 2.1.2]. We note that in the current setup, we are working from the beginning on the Borel \(\sigma \)-algebra of \(\Omega = C(([0,\infty );\mathbb {R}^d)\) of paths of infinite length and the topology of uniform convergence on compact sets. This is different to the one generated by \(\bigcup _T {\mathcal {F}}_T\), where \({\mathcal {F}}_T\) is the Borel \(\sigma \)-algebra generated by continuous paths up to time T under the supremum norm on [0, T]. Also, even if we were to adapt the Markovian approach from discrete polymers in the current continuum framework (and chose to work with \(\sigma \)-algebras on finite-time horizons), it would still require showing tightness in the continuum set up (which is not needed in the discrete setup, as mentioned). Therefore, in the current framework, we followed an alternate (and perhaps a direct) route of showing weak convergence of the measures \(\mu _{\gamma ,T}\) in the space \(\Omega = C(([0,\infty );\mathbb {R}^d)\), as described above in Step 1. \(\square \)
Organization of the article: Sects. 3.1, 3.2, 3.3, 3.4 constitute the proof of Theorem 2.1; here Sect. 3.1, 3.2, 3.3 are devoted to the existence of the measure, while Sect. 3.4 will provide the identification of thick points and characterization of the measure. Proofs of Theorem 2.2 and Theorem 2.3 can be found Sect. 4 and Sect. 5, respectively.
3 Proof of theorem 2.1
3.1 Martingale arguments
Lemma 3.1
Let \((\mathcal {F}_T)\) be the \(\sigma \)-algebra generated by the noise up to time T. Then the following statements hold with respect to the filtration \((\mathcal {F}_T)_{T>0}\):
-
(a)
For every \(\omega \in \Omega \), the process \(H(\omega )= ({H}_T(\omega ))_{T>0}\) is a martingale. Its quadratic variation is given by
$$\begin{aligned} \langle H(\omega )\rangle _T = \textbf{E}[{H}_T(\omega )^2] = \int _{0}^{T} (\phi \star \phi )(0)\textrm{d}s = T(\phi \star \phi )(0). \end{aligned}$$ -
(b)
For every \(\omega \in \Omega \) and \(\gamma >0\), the process \(\left( \exp \left\{ \gamma {H}_T(\omega ) - \frac{\gamma ^2}{2}T (\phi \star \phi )(0) \right\} \right) _{T>0}\) is a martingale.
-
(c)
For every Borel set \(A \subset \Omega \) and \(\gamma >0\),
$$\begin{aligned} \big \{\mu _{\gamma ,T}(A)\big \}_{T>0}= \bigg \{ \int _A \exp \left\{ \gamma {H}_T(\omega ) - \frac{\gamma ^2}{2}T (\phi \star \phi )(0) \right\} \textrm{d}\mathbb {P}_0(\omega ) \bigg \}_{T>0} \end{aligned}$$is a martingale.
Proof
Parts (a) and (b) are well-known. The statement in (c) follows by the lemma below together with part (b). \(\square \)
Lemma 3.2
Let \(X \in L^1(\textbf{P} \otimes \mathbb {P}_0)\) and \({\mathcal {G}}\subset \mathcal {F}\) be a \(\sigma \)-algebra. Then
Proof
The proof is standard and a direct consequence of Fubini’s theorem for conditional expectation and is therefore skipped. \(\square \)
3.2 Uniform integrability
Lemma 3.3
If for some \(\gamma >0\), \((\mu _{\gamma ,T}(\Omega ))_{T>0}\) is uniformly integrable, then the same holds for all \(\gamma '<\gamma \).
Proof
Fix \(\gamma >0\) such that \((\mu _{\gamma ,T}(\Omega ))_{T\ge 0}\) is uniformly integrable. Let \(\dot{B},\dot{B}'\) be independent copies of the noise, and let \(\gamma '<\gamma \), so that \(\gamma '=c\gamma \) for some \(0<c<1\). To avoid ambiguities, we write \(\mu _{\gamma ,T}(\Omega )=\mu _{\gamma ,T}(\Omega ,\dot{B})\). Note that
Since \((\mu _{\gamma ,T}(\Omega ,\dot{B}))_{T\ge 0}\) is uniformly integrable, then by de La Vallée-Poussin’s theorem, there exists a function \(f:[0,\infty )\mapsto [0,\infty )\) convex and increasing such that
By (3.2) and Jensen’s inequality,
It follows that \(\sup _{T}\textbf{E}[f(\mu _{\gamma ,T}(\Omega ,\dot{B}))]<\infty \), which in turn implies the uniform integrability of \((\mu _{\gamma ',T}(\Omega ,\dot{B}))_{T\ge 0}\). \(\square \)
The last lemma allows us to define the critical parameter
In particular, if \(\gamma <\gamma _c\), then \((\mu _{\gamma ,T}(\Omega ))_{T\ge 0}\) is uniformly integrable, while for \(\gamma >\gamma _c\), \((\mu _{\gamma ,T}(\Omega ))_{T\ge 0}\) is not uniformly integrable. Since we will be dealing also with the normalized probability measure \(\widehat{\mu }_{\gamma , T}\) from (1.10), we need to know whether the denominator vanishes or not at the limit. The next lemma tell us that in the uniform integrable phase, the limit \(\lim _{T\rightarrow \infty }\mu _{\gamma ,T}(\Omega )\) is positive \(\textbf{P}\)-a.s:
Lemma 3.4
If \((\mu _{\gamma ,T}(\Omega ))_{T \ge 0}\) is uniformly integrable, then \(\lim _{T\rightarrow \infty }\mu _{\gamma ,T}(\Omega )>0\) \(\textbf{P}\)-a.s.
Proof
First we note that \(\textbf{E}[\mu _{\gamma ,T}(\Omega )]=1\), so that \(\textbf{P}(\lim _{T\rightarrow \infty }\mu _{\gamma ,T}(\Omega )>0)>0\). Moreover, the event
has probability 0 or 1 in virtue of Kolmogorov’s 0-1 law. \(\square \)
The non-triviality of \(\gamma _c\), i.e., that \(\gamma _c>0\) for \(d\ge 3\) is implied by the existence of the so-called \(L^2\)-phase:
Lemma 3.5
For \(d\ge 3\) and \(\gamma >0\) small enough, the martingale \((\mu _{\gamma ,T}(\Omega ))_{T\ge 0}\) is bounded in \(L^2(\textbf{P})\). In particular, \(\gamma _c>0\).
Proof
Let \(\omega '\) be an independent copy of \(\omega \) under \(\mathbb {P}_0\). Denote by \(\mathbb {P}_0^{\otimes 2}\) the product measure of \(\mathbb {P}_0\). Then by definition and Fubini’s theorem,
Recall that for \(\omega ,\omega '\in \Omega \), \(H_T(\omega )+H_T(\omega ')\) is a Gaussian random variable with mean 0 and variance
and so
Thus,
Since \(d\ge 3\) and \(\phi \star \phi \) is bounded with compact support,
so that that for \(\gamma >0\) small enough,
By Khas’minski’s lemma [20] we deduce that
\(\square \)
Remark 6
By an application of Kahane’s inequality [19], in [25] it is proved additionally that \(\gamma _c<\infty \), and
strengthening Lemma 3.4 to an “if and only if" statement. These results are not required in the sequel. We also refer to [32] where the weak and strong disorder for the partition function as well as concentration inequalities were obtained for the continuous directed polymer in a Gaussian random environment.
3.3 Weak convergence of the approximating measures \({\widehat{\mu }}_{\gamma ,T}\)
The proof of the first part of Theorem 2.1 is split into two parts. At first, we will show that for a fixed Borel set \(A \subset \Omega \), the sequence \((\mu _{\gamma , T}(A))_{T\ge 0}\) converges to some random variable. The second step will be to deduce that \(\mu _{\gamma , T}\) and its normalized version converge weakly as a measure.
Lemma 3.6
Fix a Borel set \(A\subset \Omega \). For \(\gamma < \gamma _c\), the sequence \((\mu _{\gamma , T}(A))_{T\ge 0}\) converges in \(L^1(\textbf{P})\) and almost surely to a random variable \(\mu _{\gamma }(A)\).
Proof
By Lemma 3.1, for any Borel set \(A \subset \Omega \),
is a martingale with respect to the filtration \((\mathcal {F}_T)_T\). By our choice of \(\gamma \), the sequence \((\mu _{\gamma , T}(\Omega ))_T \) is uniformly integrable and therefore the same is true for \((\mu _{\gamma , T}(A))_T \). By the martingale convergence theorem, the latter converges in \(L^1(\textbf{P})\) and almost surely to a random variable. We denote the limit by \(\mu _{\gamma }(A)\).
\(\square \)
Remark 7
By Lemma 3.4, we know that \(\mu _\gamma (\Omega )>0\) \(\textbf{P}\)-a.s. Hence, we can replace \(\mu _{\gamma ,T}\) by its normalized version while preserving the almost sure convergence. Indeed, for any \(A \in {\mathcal {B}}\), we have
Lemma 3.7
Let \(\mu _\gamma \) be defined as in Lemma 3.6. If \((A_n)_n \subset {\mathcal {B}}\) is a sequence of subsets of \(\Omega \) that is increasing in the sense that \(A_n \subset A_{n+1}\) for all \(n \ge 1\), then \(\textbf{P}\)-a.s. it is true that
An analogous statement holds for decreasing events.
Proof
First note that the limit on the right hand side in (3.6) exists since the sequence \((\mu _\gamma (A_n))_n\) is non-decreasing (by monotonicity of the limit and of \(\mu _{\gamma , T}\) as a measure) and it is bounded by
To prove almost sure equality, it suffices to show that their expectations coincide. By \(L^1(\textbf{P})\)-convergence, the definition of the measure \(\mu _{\gamma , T}\) and Fubini’s theorem, we obtain
On the other hand, by monotone convergence, \(L^1(\textbf{P})\)-convergence and the same argument as above, we obtain
This concludes the proof. \(\square \)
Remark 8
If we normalize \(\mu _{\gamma }(A)\), i.e. if we consider the random variables given by \(\frac{\mu _{\gamma }(A)}{\mu _{\gamma }(\Omega )}\) for \(A \in {\mathcal {B}}\), then clearly the statement in Lemma 3.7 remains true.
The key argument for the proof of Theorem 2.1 is provided by
Proposition 3.8
Suppose that for each Borel set \(A\subset \Omega \), the sequence \(\mu _{\gamma , T}(A)\) converges in \(L^1(\textbf{P})\) and almost surely to some random variable \(\mu _{\gamma }(A)\) as \(T \rightarrow \infty \). Then the sequence \((\mu _{\gamma ,T})_{T\ge 0}\) converges weakly \(\textbf{P}\)-a.s. to a random measure \(\mu _\gamma \). Similarly, the sequence of normalized measures \((\widehat{\mu }_{\gamma , T})_{T\ge 0}\) converges weakly \(\textbf{P}\)-a.s. to a random probability measure \(\widehat{\mu }_{\gamma }\).
Proof
We will carry out the proof in few steps.
Step 1: In the first step, we will show that, there is a collection \({\mathcal {X}}\) of countably many Borel measurable subsets of \(\Omega \) such that
Indeed, we define the sets
and
Analogously, we define \({\mathcal {X}}_c^1\) (by replacing \({\mathcal {A}}_o\) by \({\mathcal {A}}_c\) in the above definition) and \({\mathcal {X}}_s^1\) (by replacing \({\mathcal {A}}_o\) by \({\mathcal {A}}_s\)), and set
Furthermore, for any \(\delta >0\), let
![](http://media.springernature.com/lw226/springer-static/image/art%3A10.1007%2Fs00440-024-01271-7/MediaObjects/440_2024_1271_Equ36_HTML.png)
and set
Thus, \({\mathcal {X}}\) contains countably many Borel measurable subsets of \(\Omega \). By Lemma 3.6 (actually, by Remark 7), for any Borel subset A of \(\Omega \),
\(\textbf{P}\)-almost surely. But since \({\mathcal {X}}\) is countable, the above almost sure convergence happens for all \(A \in {\mathcal {X}}\) simultaneously, proving (3.7).
Step 2: We will now show that, \(\textbf{P}\) almost surely, for any \(\ell \in \mathbb {Q}_+\) and \(\eta \in \mathbb {Q}_+\), there is \(a\in \mathbb {Q}_+\) such that
Note that for a fixed \(\ell \in \mathbb {Q}^+\), the events
are decreasing. Hence, by Lemma 3.7 (and the subsequent remark), we deduce that
In particular, for fixed \(\ell , \eta \in \mathbb {Q}^+\), one can find \(a \in \mathbb {Q}^+\) such that \(\textbf{P}\)-a.s. we have
Now by (3.7) in Step 1, we have
for any \(\ell \in \mathbb {Q}_+\) and corresponding \(a\in \mathbb {Q}_+\) simultaneously. Therefore, by (3.13), \(\textbf{P}\)-a.s., for any \(\ell \in \mathbb {Q}_+\) and \(\eta \in \mathbb {Q}_+\), there is \(a\in \mathbb {Q}_+\) such that
for some \(T>0\) large enough. Hence, we can choose \(a\in \mathbb {Q}_+\) large enough which guarantees that the above inequality holds for all \(T >0\). This shows (3.12).
Step 3: Recall the definition of \(m^T(\omega ,\delta )\) from (3.10). We will now show that, \(\textbf{P}\) almost surely, for any \(\ell , \eta , \varepsilon \in \mathbb {Q}_+\), there is \(\delta \in \mathbb {Q}_+\) such that
Indeed, note that for fixed \(\ell , \eta , \varepsilon \in \mathbb {Q}^+ \), the events
are decreasing as \(\delta \searrow 0\). Again by Lemma 3.7, it holds that
Thus, again using (3.7) from Step 1, \(\textbf{P}\)-a.s., for any \(\ell , \eta , \varepsilon \in \mathbb {Q}_+\), there exists a \(\delta \in \mathbb {Q}^+\) such that
Hence, \(\textbf{P}\)-a.s.,
if \(T>0\) is sufficiently large. If we now choose \(\delta \) sufficiently small, we can assure that the inequality holds for all \(T >0\), proving (3.14).
Step 4: We will now conclude that
Fix \(\eta \in \mathbb {Q}_+\) and also \(\ell , m \in \mathbb {N}\). Then by (3.12) and (3.14), \(\textbf{P}\)-a.s., there is \(a_\ell , \delta _{m,\ell }\in \mathbb {Q}_+\) such that for all \(T>0\),
Note that for each \(\ell ,m \in \mathbb {N}\), \(A_\ell \) and \(B_{m,\ell }\) are closed. Therefore,
By (3.12), the family of functions K is uniformly bounded and by (3.14), K is equicontinuous, and hence, by Arzela-Ascoli theorem, K is relatively compact. Together with (3.17), we conclude that \(K\subset \Omega \) is compact. Finally, by (3.16), \(\textbf{P}\)-a.s., for any \(\eta \in \mathbb {Q}_+\) and all \(T>0\), \({\widehat{\mu }}_{\gamma ,T}(K) \ge 1- \sum _{\ell \in \mathbb {N}} \eta /2^{\ell +1} - \sum _{\ell ,m\in \mathbb {N}} \eta /2^{\ell +1+m}=1- \eta \). This proves (3.15).
Step 5: We will now conclude the proof of the proposition. By tightness from Step 4, for each subsequence of \((\widehat{\mu }_{\gamma , T})_{T\ge 0}\) there is a further sub-subsequence converging weakly to some random measure \(\widehat{\mu }_{\gamma }\). It remains to show that the limiting measure \(\widehat{\mu }_{\gamma }\) is uniquely determined that is, this limit is independent of the subsequence. We will use Portmanteau’s Theorem to show that for all Borel sets \(A \subset \Omega \), it is true that any weak limit satisfies
More precisely, it suffices to show that this is true for all sets \(A \in {\mathcal {X}}_s^1\) (recall (3.9)). Hence, let \(A \in {\mathcal {X}}_s^1\), i.e. A is of the form \(A= \{\omega : \omega (t_1)\in A_1, \ldots , \omega (t_n) \in A_n \}\) for some \(n \in \mathbb {N}, t_1< \ldots < t_n \in \mathbb {Q}^+\) and \(A_1, \ldots , A_n \in {\mathcal {A}}_s\). Then by Lemma 3.7,
where the supremum is over \(A_1', \ldots , A_n' \in {\mathcal {A}}_c\) (recall (3.8)) such that \(A_i' \subset A_i\) for all \(1 \le i \le n\). Similarly, we have
where the infimum is over all \(A_1', \ldots , A_n' \in {\mathcal {A}}_o\) such that \(\bar{A_i} \subset A_i'\) for all \(1\le i\le n\). By Lemma 3.7 and Portmanteau’s Theorem, for any \(A \in {\mathcal {X}}_s^1\) and any \(A_1', \ldots , A_n' \in {\mathcal {A}}_o\) with \(\bar{A_i}\subset A_i'\) for all \(1\le i\le n\), \(\textbf{P}\)-a.s., we have
Since \(A_1', \ldots , A_n'\) are arbitrary, by taking the infimum and using (3.19), we can deduce that \(\textbf{P}\)-a.s., \(\widehat{\mu }_{\gamma }(A)\le \nicefrac {\mu _\gamma (A)}{\mu _\gamma (\Omega )}\). Proceeding analogously with sets \(A_1', \ldots , A_n' \in {\mathcal {A}}_c\) such that \(A_i' \subset A_i\) for all \(1 \le i \le n\) and using (3.18), we obtain \(\textbf{P}\)-a.s., \(\widehat{\mu }_\gamma (A)\ge \nicefrac {\mu _\gamma (A)}{\mu _\gamma (\Omega )}\). Therefore, \(\widehat{\mu }_{\gamma }(A)= \nicefrac {\mu _\gamma (A)}{\mu _\gamma (\Omega )}\) \(\textbf{P}\)-a.s. as desired. The proof of the weak convergence for the unnormalized measures can be performed by the same methods, by noting that \(\mu _\gamma (\Omega )>0\) \(\textbf{P}\)-a.s. \(\square \)
We can now complete the
Proof of existence Theorem 2.1
By Lemma 3.6, for each Borel set \(A\subset \Omega \), the sequence \((\widehat{\mu }_{\gamma , T}(A))\) converges \(\textbf{P}\)-a.s. to \(\nicefrac {\mu _\gamma (A)}{\mu _\gamma (\Omega )}\) as \(T \rightarrow \infty \). Hence, by Proposition 3.8, \(\widehat{\mu }_{\gamma , T}\) converges weakly \(\textbf{P}\)-a.s. to the probability measure \(\widehat{\mu }_{\gamma }\). This completes the existence part of the proof of Theorem 2.1. \(\square \)
3.4 Thick points, support and characterization
We turn to the proofs of the second and the third parts of Theorem 2.1. We start with the third part, which will be shown in
Proposition 3.9
(Characterization of \(\mu _\gamma )\) Fix \(d\ge 3\), \(\gamma \in (0,\gamma _c)\) and a mollifier \(\phi \), Then the (unnormalized) GMC measure \(\mu _\gamma = \mu _{\gamma ,\phi }\) is the unique measure such that the law of \({\dot{B}} \) under \(\mathbb {Q}_{\mu _{\gamma ,\phi }}\) (recall the notation (2.2)) is the same as the law of the Schwartz distribution
under \(P\otimes \mathbb {P}_0\). In other words, \(\mu _{\gamma ,\phi }\) is the unique measure satisfying
for any bounded measurable function \(F: \Omega \times {\mathcal {E}} \mapsto \mathbb {R}\).
Proof
Recall that \({H}_T(\omega ) = {H}_T(\omega , {\dot{B}} )\) where \({\dot{B}} \) is the space-time white noise. Similarly, we write \(\mu _{\gamma }(\textrm{d}\omega ) = \mu _{\gamma }(\textrm{d}\omega , {\dot{B}} )\). Given \(T>0\), set (recall the notation from (2.2))
This is a probability measure since \(E \left[ \exp \left\{ \gamma {H}_T(\omega , {\dot{B}} ) - \frac{\gamma ^2}{2}T(\phi \star \phi )(0) \right\} \right] = 1\). Similarly define
Before continuing, we show that
Lemma 3.10
\(\mathbb {Q}_{\mu _{\gamma ,T}} \rightarrow \mathbb {Q}_{\mu _{\gamma }}\) weakly as \(T \rightarrow \infty \).
Proof
We first prove first that \((\mathbb {Q}_{\mu _{\gamma ,T}})_{T\ge 0}\) is tight. It is enough to verify that the marginals
Clearly \(\mathbb {Q}_{\mu _{\gamma ,T}}^{1}\) is tight since it is a single probability measure. On the other hand, since \({\mathcal {S}}'\) is \(\sigma \)-compact (by the Banach-Alaoglu theorem) and \(\mathbb {Q}_{\mu _{\gamma ,T}}(\Omega \times {\mathcal {S}}')\) is uniformly bounded in T by the uniform integrability of \((\mu _{\gamma ,T}(\Omega ))_{T\ge 0}\), then \(\mathbb {Q}_{\mu _{\gamma ,T}}^{2}\) is also tight. Therefore, \(\mathbb {Q}_{\mu _{\gamma ,T}}\) converges weakly to some measure \(\mathbb {Q}\) as \(T\rightarrow \infty \). We claim that \(\mathbb {Q}=\mathbb {Q}_{\mu _{\gamma }}\). Indeed, for Borel subsets \(A_1\subset \Omega , A_2\subset {\mathcal {S}}'\), it holds that \((\mu _{\gamma ,T}(A_1))_{T\ge 0}\) is uniformly integrable and \(\mu _{\gamma ,T}(A_1)\rightarrow \mu _{\gamma }(A_1)\) P-a.s. and in \(L^1\). Therefore,
as \(T\rightarrow \infty \). Thus, for any Borel sets \(A_1\subset \Omega \) and \(A_2\subset {\mathcal {S}}'\), \(\lim _{T\rightarrow \infty }\mathbb {Q}_{\mu _{\gamma ,T}}(A_1\times A_2)=\mathbb {Q}_{\mu _{\gamma }}(A_1\times A_2)\). By Portmanteau’s theorem, we deduce that \(\mathbb {Q}=\mathbb {Q}_{\mu _{\gamma }}\). \(\square \)
We now continue with the proof of Proposition 3.9. By the previous lemma, if \(n \in \mathbb {N}, f_1, \ldots , f_n \in {\mathcal {S}}(\mathbb {R}_{+}\times \mathbb {R}^d)\) and \(g: \Omega \mapsto \mathbb {R}\) is bounded and continuous, i.e the map \((\omega , {\dot{B}} ) \mapsto g(\omega ){\dot{B}} (f_1)\cdot \ldots \cdot {\dot{B}} (f_n)\) is a bounded continuous function on \(\Omega \times {\mathcal {S}}'\), then
Now, conditioning \(\mathbb {Q}_{\mu _{\gamma ,T}}(\textrm{d}{\dot{B}}, \textrm{d}\omega )\) on \(\omega \in \Omega \), we obtain
By the Cameron-Martin-Girsanov Theorem, we know that under the measure \(\mathbb {Q}_{\mu _{\gamma ,T}}(\textrm{d}{\dot{B}} \, | \, \omega )\), \(({\dot{B}} (f))_f\) is a Gaussian process with the same covariance structure as in (1.7) and mean given by
Hence, the expression in (3.21) is equal to
Writing \(\mathbb {Q}_{\mathbb {P}_0}(\textrm{d}{\dot{B}}, \textrm{d}\omega ) {=} \mathbb {P}_0(\textrm{d}\omega ) P(\textrm{d}{\dot{B}} )\), and using (3.21), we deduce that
Since \(g, f_1, \ldots , f_n\) are arbitrary, we conclude that the law of \({\dot{B}} \) under \(\mathbb {Q}_{\mu _\gamma }\) is the same as the law of \(\widetilde{{\dot{B}} }\) defined by \(\widetilde{{\dot{B}} }(f) {:=} {\dot{B}} (f) + \gamma \int _{R_{+}\times \mathbb {R}^d} f(s,y) \phi (\omega _s - y) \textrm{d}s \textrm{d}y\) under \(\mathbb {Q}_{\mathbb {P}_0}\). The uniqueness is immediate from this construction. This completes the proof of Proposition 3.9. \(\square \)
Finally, the second part of Theorem 2.1 will be shown in
Proposition 3.11
Fix \(d\ge 3\) and \(\gamma \in (0,\gamma _c)\) and let \(\mu _\gamma \) be the infinite volume measure. Then,
Proof
First recall that, for every \(\omega \in \Omega \), the stochastic integral \(H_T(\omega )= \int _0^T \int _{\mathbb {R}^d} \phi (\omega _s-y) \dot{B}(s,y) \textrm{d}s \textrm{d}y\) is a continuous martingale with quadratic variation
It follows that, for every \(\omega \in \Omega \),Footnote 6
Next, by Theorem 3.9, we know that the law of \({H}_T\) under the measure \(\mathbb {Q}_{\mu _\gamma }(\textrm{d}{\dot{B}} ,\textrm{d}\omega )=\mu _\gamma (\textrm{d}\omega ,{\dot{B}} )P(d{\dot{B}} )\) is the same as the law of \({H}_T + \gamma T (\phi \star \phi )(0)\) under \(P\otimes \mathbb {P}_0\). Combining these two facts, we have
so that \(\mu _{\gamma }\left( \lim _{T \rightarrow \infty } \frac{{H}_T}{T} \ne \gamma (\phi \star \phi )(0) \right) =0\) \(\textbf{P}\)-a.s. This completes the proof of Proposition 3.11, and therefore, that of Theorem 2.1. \(\square \)
We will end this section with
Proposition 3.12
Let \(\phi \) and \(\phi ^\prime \) be two mollifiers. If \(\phi (\cdot )\not \equiv \phi ^\prime (\cdot )\), then \(\mu _{\gamma ,\phi }\ne \mu _{\gamma ,\phi ^\prime }\).
Proof
Since for any \(A\subset \Omega \) Borel measurable, \((\mu _{\gamma ,T,\phi }(A))_{T\ge 0}\) and \((\mu _{\gamma ,T,\phi '}(A))_{T\ge 0}\) are uniformly integrable martingales converging respectively to \(\mu _{\gamma ,\phi }(A)\) and \(\mu _{\gamma ,\phi '}(A)\), we have \(\textbf{P}\)-a.s. the following identities:
Thus, if \(\mu _{\gamma ,\phi }(A)=\mu _{\gamma ,\phi '}(A)\) \(\textbf{P}\)-a.s., then we deduce that for all \(T\ge 0\), \(\mu _{\gamma ,T,\phi }(A)=\mu _{\gamma ,T,\phi '}(A)\) \(\textbf{P}\)-a.s. By choosing an appropriate countable collection of Borel sets \(A\in \Omega \) (such as the sets \({\mathcal {X}}^1_s\) that appear in the proof of Proposition 3.8, one can deduce that \(\textbf{P}\otimes \mathbb {P}_0\)-a.s. for all \(T\in \mathbb {Q}^+\) and assuming that \(\phi \star \phi (0)=\phi '\star \phi '(0)\),
concluding that \(\textbf{P}\otimes \mathbb {P}_0\)-a.s., \(H_{T,\phi }=H_{T,\phi '}\). Note that that for \(\omega \in \Omega \), the quadratic variation of \(H_{T,\phi }(\omega )-H_{T,\phi '}(\omega )\) is equal to
By Cauchy-Schwarz inequality, we know that the last display is equal to zero if and only if \(\phi =\lambda \phi '\) for some \(\lambda >0\), but using that \(\int _{\mathbb {R}^d}\phi (y)\textrm{d}y=\int _{\mathbb {R}^d}\phi '(y)\textrm{d}y=1\), we deduce that \(\lambda =1\), so \(\phi =\phi '\). \(\square \)
Remark 9
Looking at (2.1) in Theorem 2.1, one may wonder if for a fixed \(\phi \), \(\mu _{\gamma ,\phi }\) is the unique measure such that
However, this is false: by perturbing \(\mathbb {P}_0\) by \(\textrm{e} ^{\gamma (H_T+H_1)-\frac{\gamma ^2}{2}\phi \star \phi (0)(T+1+2(T\wedge 1))}\) instead of by \(\textrm{e} ^{\gamma H_T-\frac{\gamma ^2}{2}T\phi \star \phi (0)} \) leads to a measure \(\tilde{\mu }_{\gamma ,\phi }\) (at least for \(\gamma \) small enough so that the limiting measure also exists) such that (3.23) still holds. Indeed, one can follow the proof of Theorem 3.9 and notice that the law of \(\dot{B}\) under \(\mathbb {Q}_{\tilde{\mu }_{\gamma ,\phi }}\) is the same as the law of
under \(P\otimes \mathbb {P}_0\). In particular, \(\mu _{\gamma ,\phi }\ne \tilde{\mu }_{\gamma ,\phi }\). Nevertheless, the distribution of \((H_T)_{T\ge 1}\) under \(\mathbb {Q}_{\tilde{\mu }_{\gamma ,\phi }}\) is the same as the distribution of \((H_T+\gamma (T+1)\phi \star \phi (0))_{T\ge 1}\) under under \(P\otimes \mathbb {P}_0\). Thus, \(\tilde{\mu }_{\gamma ,\phi }\) also satisfies (3.23).
4 Volume decay and Hölder exponents
4.1 Proof of theorem 2.2
Recall the definition of \((\Omega _0, \Vert \cdot \Vert _w)\) from Sect. 2.2. The proof of Theorem 2.2 is split in four parts. First, we will show the (uniform on \(\eta \in \Omega _0\)) upper bound, stated as
Proposition 4.1
Given \(\gamma \in (0,\gamma _c)\) and w satisfying (2.5), there exists \(r_0>0\) such that for all \(r\in (0,r_0)\), there is an explicit constant \(C_1\in (0,\infty )\) (defined in (4.13)) such that
Proof
We will prove this result in four steps.
Step 1: Since
and since \(\gamma <\gamma _c\), we know that \(\textbf{P}\)-a.s., when letting \(\varepsilon \rightarrow 0\), the second term vanishes. So we consider from now on only the asymptotic behavior of the first term \(\varepsilon ^2\log \mu _{\gamma }(\Vert \omega -\eta \Vert <r\varepsilon )\). Let us set
Using that for every \(\varepsilon >0\), \(\mu _\gamma = \lim _{s\rightarrow \infty } \mu _{\gamma ,s+ \varepsilon ^{-2}}(\cdot )\) and the monotonicity
we obtain
where in the penultimate step we used the Markov property (upon conditioning on the \(\sigma \)-algebra generated by the Brownian path until time \(\varepsilon ^{-2}\); we remind the reader that \(\theta _{t,x}\) denotes the time-space shift on the noise \(\dot{B}\)).
Step 2: Now, since \(\gamma <\gamma _c\), by Part (i) of Theorem 2.3 and Doob’s maximal inequality (applied to the Martingale \((\mu _{\gamma ,S}(\Omega ))_{S\ge 0}\)),
For any \(q,\ell >1\) such that \(\frac{1}{p_0}+\frac{1}{q}+\frac{1}{\ell }=1\), we apply Hölder’s inequality to (4.5), so that
The first term on the right hand side above is controlled by an explicit constant \(\lambda (\cdot )\) such that for any \(\gamma >0\) and \(\textbf{P}\)-a.s.
In order to not ebb the flow of the proof, we defer the proof of the above fact to Theorem 4.8 in Sect. 4.2. We therefore bound the second term in the last display of the right hand side of (4.7) as follows. By Anderson’s inequality (see (4.14) below in Lemma 4.2), for any \(\eta \in \Omega _0\),
To estimate the right hand side, we will need the following estimate valid for the Wiener measure \(\mathbb {P}_0\) on \(\Omega _0\) (see [21, Theorem 1.4]): Let \(j_{\frac{d-2}{2}}\) be the smallest positive root of the Bessel function \(J_{\frac{d-2}{2}}\). If w satisfies (2.5), then
Thus, applying (4.9) leads to
Step 3: Thus, we concentrate on the third term of the last display of the (r.h.s.) of (4.7). Set
Since \(\dot{B}\) is stationary with respect to space-time shifts \((\theta _{t,x})_{t>0,x\in \mathbb {R}^d}\), then \(f_{\varepsilon }\) is stationary and by (4.6), \(\textbf{E}[f_{\varepsilon }]=\textbf{E}[\sup _{S\ge 0}\mu _{\gamma ,S}(\Omega )^p]< \infty \). By the ergodic theorem for stationary processes,Footnote 7 there is a (possibly random) \(C=C(\dot{B})\) such that for all \(\varepsilon >0\) sufficiently small,
and consequently,
Combining (4.7–4.12) and optimizing over \(q,\ell \), we obtain
and observe that \(C_1>0\) if we choose r small enough. This completes the proof of the proposition.
Step 4: In Step 2 above, we have used the following general result for Gaussian measures:
Lemma 4.2
Let E be a separable Banach space, with X being a E-valued centered Gaussian with law \(\mu \). Then the following hold:
-
(i)
For any symmetric convex subset \(A\subset E\) and \(x\in E\),
$$\begin{aligned} \mu (A+ x) \le \mu (A). \end{aligned}$$(4.14) -
(ii)
If \(H_\mu \) is the Cameron-Martin space of \(\mu \),Footnote 8 then for any \(r>0\), and \(\eta \in H_\mu \),
$$\begin{aligned} \textrm{e} ^{-\frac{1}{2}\Vert \eta \Vert _\mu ^2} \mu (\omega \in E: \Vert \omega \Vert \le r) \le \mu (\omega \in E:\Vert \omega - \eta \Vert \le r) \le \mu (\omega \in E: \Vert \omega \Vert \le r). \end{aligned}$$(4.15)
Proof
(4.14) is Anderson’s inequality, which follows from log-concavity of Gaussian measures (see [4, Thm. 2.8.10], or [22, Thm. 2.13]). The upper bound of (4.15) follows from (4.14), while the lower bound follows from the Cameron-Martin formula
as well as Hölder’s inequality and the symmetry of \(\langle \omega , \eta \rangle _\mu \) on \(\{\omega \in E: \Vert \omega \Vert \le r\}\), see [22, Thm. 3.1]. \(\square \)
We now turn to the lower bound in Theorem 2.2, stated as
Proposition 4.3
Given \(\gamma \in (0,\gamma _c)\) and w satisfying (2.5), there exists \(r_0>0\) such that for all \(r\in (0,r_0)\), there is an explicit constant \(C_2 \in (0,\infty )\) (defined in (4.24)) such that
Proof
This will also be shown in four steps.
Step 1: Recall the norm \(\Vert \cdot \Vert _{w,\varepsilon }\) from (4.3). We will first prove a lower bound on \(\liminf _{\varepsilon \rightarrow 0}\varepsilon ^2\log \widehat{\mu }_\gamma (\Vert \omega \Vert _{w,\varepsilon }<r\varepsilon )\). As remarked below (4.2), for \(\gamma \in (0,\gamma _c)\), it is sufficient to handle
For a fixed \(\gamma <\gamma _c\), we apply part (ii) of Theorem 2.3 to find some \(q_0>0\) such that for all \(0<q\le q_0\),
Then
In the second equality above we used the Markov property (again by conditioning on the \(\sigma \)-algebra generated by the Brownian path until time \(\varepsilon ^{-2}\)), in the subsequent lower bound we used Fatou’s lemma and in the last lower bound we invoked reverse Hölder’s inequality. To justify this step, recall that Hölder’s inequality implies that if \(\theta >1\) and \(f(\omega ),g(\omega )\) are measurable functions satisfying \(\Vert fg\Vert _1 < \infty \) and \(\Vert g\Vert _{-\frac{1}{\theta -1}} <\infty \), then
To deduce the last lower bound in (4.18), we apply the above inequality for
Thus,
Step 2: In this step, we will show that for all \(\gamma >0\),
This in turn will imply that
To show (4.20), we first note that, for any event \(A_\varepsilon \subset \Omega \) with \(\mathbb {P}_0(A_\varepsilon )>0\) it holds \(\textbf{P}\)-almost surely that
We refer to Lemma 4.9 for a proof of this result. Therefore,
We note that in the lower bound above we applied Jensen’s inequality and in the following display we used that \(H_{\varepsilon ^{-2}}\) is an Itô integral with \(\textbf{E}[H_{\varepsilon ^{-2}}]=0\). We now choose \(A_\varepsilon =\{\omega :\Vert \omega \Vert _{w,\varepsilon } < r\varepsilon \}\) and note that (4.4) and (4.9) imply in particular that for this choice, \(\mathbb {P}_0(A_\varepsilon )>0\) for \(\varepsilon >0\) sufficiently small. Thus these two facts, combined with the above estimate imply (4.20).
Step 3: Now using (4.17), the stationarity of the white noise, combined with the ergodic theorem and invoking (4.17) (similar to the argument for deducing (5.7)), we obtain that
Finally, combining (4.19)-(4.23), and optimizing over \(0<q<q_0\), it holds \(\textbf{P}\)-a.s.
Clearly, \(C_2<\infty \).
Step 4: Note that we have shown Proposition 4.3 for the restricted weighted norm \(\Vert \cdot \Vert _{w,\varepsilon }\). To extend the argument to the weighted norm \(\Vert \cdot \Vert _w\), we can apply again the reverse Hölder’s inequality to deduce, for any \(p>1\),
As \(p>1\) is arbitrary, we conclude the proof of (2.8). \(\square \)
To finish the proof of Theorem 2.2, we show
Corollary 4.4
Let \(p_0>1\) and \(q_0>0\). Then for any \(\gamma \) small enough, the constants \(0<C_1\le C_2<\infty \) can be chosen as
where \(j_{\frac{d-2}{2}}\) is the smallest positive root of the Bessel function \(J_{\frac{d-2}{2}}\). In particular, for any \(r>0\),
In other words, both exponents converge as \(\gamma \rightarrow 0\) to the volume decay exponent for the Wiener measure.
Proof
First we check (4.25). Since the constant \(C_2\) is the same from (4.24), we only show the corresponding estimate for \(C_1\). We note that by Theorem 4.8, it holds that \(\lambda (\gamma )\le 0\) for all \(\gamma >0\). Therefore, if \(\gamma \) is small enough so that \(\mu _\gamma \in L^{2 p_0}(\textbf{P})\), the formula from (4.13) can be bounded by below by (choosing \(\ell =\frac{p_0}{p_0-1}\) and \(q=2p_0\)),
Next, we prove (4.26). By the previous part, let \(\varepsilon >0\) and \(p_0>1\), \(q_0>0\) such that \(\frac{p_0-1}{p_0}>1-\varepsilon \) and \(\frac{q_0+1}{q_0}<1+\varepsilon \). Then if \(\gamma \) is small enough so that (4.25) holds and \(\frac{\gamma ^2}{2}(\phi \star \phi )(0)<\varepsilon \), then
and
Since \(\varepsilon >0\) is arbitrary, this completes the proof of Corollary 4.4. \(\square \)
4.2 A variational formula on a group-invariant compactification
In Step 2 of the proof of Proposition 4.1 we have used (4.8) which gives an explicit variational formula for the free energy \(\lim _{T\rightarrow \infty }\frac{1}{T}\log \mu _{\gamma ,T}(\Omega )\) as well its properties for all \(\gamma >0\). Description of this formula and its properties need some setting up and further notation.
We denote by \({{\mathcal {M}}}_1= {{{\mathcal {M}}}_1}(\mathbb {R}^d)\) (resp., \({{\mathcal {M}}}_{\le 1}\)) the space of probability (resp., subprobability) distributions on \(\mathbb {R}^d\) and by \({\widetilde{{{\mathcal {M}}}}}_1= {{\mathcal {M}}}_1 \big / \sim \) the quotient space of \({{\mathcal {M}}}_1\) under the action of \(\mathbb {R}^d\) (as an additive group on \({{\mathcal {M}}}_1\)), that is, for any \(\mu \in {{\mathcal {M}}}_1\), its orbit is defined by \(\widetilde{\mu }=\{\mu \star \delta _x:\, x\in \mathbb {R}^d\}\in {\widetilde{{{\mathcal {M}}}}}_1\). Then we define
to be the space of all empty, finite or countable collections of orbits of subprobability measures with total masses bounded by 1. Note that the quotient space \({\widetilde{{{\mathcal {M}}}}}_1(\mathbb {R}^d)\) is embedded in \(\widetilde{{\mathcal {X}}}\) – that is, for any \(\mu \in {{\mathcal {M}}}_1(\mathbb {R}^d)\), \({\widetilde{\mu }}\in {\widetilde{{{\mathcal {M}}}}}_1(\mathbb {R}^d)\) and the single orbit element \(\{{\widetilde{\mu }}\}\in \widetilde{{\mathcal {X}}}\) belongs to \(\widetilde{{\mathcal {X}}}\) (in this context, sometimes we will write \({\widetilde{\mu }}\in \widetilde{{\mathcal {X}}}\) for \(\{{\widetilde{\mu }}\}\in \widetilde{{\mathcal {X}}}\)).
The space \(\widetilde{{\mathcal {X}}}\) also comes with a metric structure. If for any \(k\ge 2\), \({\mathcal {H}}_k\) is the space of functions \(h:\left( \mathbb {R}^d\right) ^k\rightarrow \mathbb {R}\) which are invariant under rigid translations and which vanish at infinity, we define, for any \(h\in {\mathcal {H}}=\bigcup _{k\ge 2}{\mathcal {H}}_k\), the functionals
A sequence \(\xi _n\) is said to converge to \(\xi \) in the space \(\widetilde{{\mathcal {X}}}\) if
This leads to the following definition of the metric \(\textbf{D}\) on \(\widetilde{{\mathcal {X}}}\). For any \(\xi _1,\xi _2\in \widetilde{{\mathcal {X}}}\), set
The following result was proved in [26, Theorem 3.1\(-\)3.2].
Theorem 4.5
We have the following properties of the space \(\widetilde{{\mathcal {X}}}\).
-
\(\textbf{D}\) is a metric on \(\widetilde{{\mathcal {X}}}\) and the space \({\widetilde{{{\mathcal {M}}}}}_1(\mathbb {R}^d)\) is dense in \((\widetilde{{\mathcal {X}}},\textbf{D})\).
-
Any sequence in \({\widetilde{{{\mathcal {M}}}}}_1(\mathbb {R}^d)\) has a convergent subsequence with a limit point in \(\widetilde{{\mathcal {X}}}\). Thus, \(\widetilde{{\mathcal {X}}}\) is the completion and the compactification of the totally bounded metric space \({\widetilde{{{\mathcal {M}}}}}_1(\mathbb {R}^d)\) under \(\textbf{D}\).
-
Let a sequence \((\xi _n)_n\) in \(\widetilde{{\mathcal {X}}}\) consist of a single orbit \({\widetilde{\gamma }}_n\) and \(\textbf{D}(\xi _n,\xi )\rightarrow 0\) where \(\xi =({\widetilde{\alpha }}_i)_i\in \widetilde{{\mathcal {X}}}\) such that \(\alpha _1(\mathbb {R}^d)\ge \alpha _2(\mathbb {R}^d) \ge \dots \). Then given any \(\varepsilon >0\), we can find \(k\in \mathbb {N}\) such that \(\sum _{i>k} \alpha _i(\mathbb {R}^d) <\varepsilon \) and we can write \(\gamma _n= \sum _{i=1}^k\alpha _{n,i}+ \beta _n\), such that
-
for any \(i=1,\dots ,k\), there is a sequence \((a_{n,i})_n\subset \mathbb {R}^d\) satisfying
$$\begin{aligned} \begin{aligned} \alpha _{n,i}\star \delta _{a_{n,i}} \Rightarrow \alpha _i \quad \text{ with }\quad \lim _{n\rightarrow \infty } \, \inf _{i\ne j}\,\, |a_{n,i}- a_{n,j}| =\infty . \end{aligned} \end{aligned}$$ -
The sequence \(\beta _n\) totally disintegrates, meaning that for any \(r>0\), \(\sup _{x\in \mathbb {R}^d} \beta _n\big (B_r(x)\big )\rightarrow 0\).
-
Recall the definition of \(\Phi _\gamma :\widetilde{{\mathcal {X}}}\rightarrow \mathbb {R}\) from (4.35):
Because of shift-invariance of the integrand in (4.29), \(\Phi _\gamma \) is well-defined on \(\widetilde{{\mathcal {X}}}\). Moreover, we have
Lemma 4.6
\(\Phi _\gamma \) is continuous and non-negative on \(\widetilde{{\mathcal {X}}}\), and \(\Phi _\gamma (\cdot )\le \frac{\gamma ^2}{2}V(0)\).
Proof
For the continuity of \(\Phi _\gamma \), we refer to [26, Corollary 3.3]. Recall that \(V=\phi \star \phi \) and \(\phi \) is rotationally symmetric. Hence, for any \(\alpha \in {{\mathcal {M}}}_{\le 1}(\mathbb {R}^d)\), by Cauchy-Schwarz inequality,
Thus, \(\Phi _\gamma (\xi )\le \frac{\gamma ^2 (\phi \star \phi )(0)}{2} \sum _{i\in I} (\alpha _i(\mathbb {R}^d))^2 \le \frac{\gamma ^2 (\phi \star \phi )(0)}{2}\) since, for \(\xi =({\widetilde{\alpha }}_i)_{i\in I} \in \widetilde{{\mathcal {X}}}\), we have \(\sum _{i\in I} \alpha _i(\mathbb {R}^d)\le 1\). Moreover, since \(V=\phi \star \phi \) is non-negative, also \({\Phi _\gamma }(\cdot )\ge 0\).
\(\square \)
To define the required variational formula, we also need a Markovian dynamics on the space \({{\mathcal {M}}}_1(\widetilde{{\mathcal {X}}})\) of probability measures on the compactification \({\mathcal {X}}\). To this end, we recall the basic notations and properties of this dynamics.
For any \(\alpha \in {{\mathcal {M}}}_\le (\mathbb {R}^d)\), let
and note that for any \(a\in \mathbb {R}^d\) and \(t>0\), \({\mathscr {F}}_t(\alpha _i)\overset{{\scriptscriptstyle {({d}})}}{=}{\mathscr {F}}_t(\alpha _i\star \delta _a)\). Hence, we may define \({\mathscr {F}}_t, \overline{{\mathscr {F}}}_t: \widetilde{{\mathcal {X}}}\mapsto \mathbb {R}\) as
Next, for any \(t>0\), and for \(\xi =({\widetilde{\alpha }}_i)_i \in \widetilde{{\mathcal {X}}}\), we set
Recall that \({\mathscr {F}}_t(\alpha _i)\overset{{\scriptscriptstyle {({d}})}}{=}{\mathscr {F}}_t(\alpha _i\star \delta _a)\) and likewise, \((\alpha _i\star \delta _a)^{{\scriptscriptstyle {({t}})}}(\textrm{d}x)\overset{(d)}{=}(\alpha _i^{{\scriptscriptstyle {({t}})}}\star \delta _a)(\textrm{d}x)\). For any \(\vartheta \in {{\mathcal {M}}}_1(\widetilde{{\mathcal {X}}})\), then (4.32) further defines a transition kernel
For any \(\gamma >0\), the set of fixed points of this dynamics is non-empty, as shown by
Lemma 4.7
The set
of fixed points of \(\Pi _t\) is a non-empty, compact subset of \({{\mathcal {M}}}_1(\widetilde{{\mathcal {X}}})\).
Proof
Note that \({\mathfrak m_\gamma }\ne \emptyset \), because \(\delta _{\widetilde{0}}\in {\mathfrak m_\gamma }\). Moreover, using the definition of the metric \(\textbf{D}\) on \(\widetilde{{\mathcal {X}}}\) and by the resulting convergence criterion determined by Theorem 4.5, it was shown in [6, Theorem 3.1] that the map
This property, together with the compactness of \(\widetilde{{\mathcal {X}}}\) (and therefore also that of \({{\mathcal {M}}}_1(\widetilde{{\mathcal {X}}})\)), we have that \({{\mathcal {M}}}_1(\widetilde{{\mathcal {X}}})\ni \vartheta \mapsto \Pi _t(\vartheta ,\cdot )\) is continuous too for any \(t>0\). It follows that \({\mathfrak m_\gamma }\) is a closed subset of the compact metric space \({\mathcal {M}}_1(\widetilde{{\mathcal {X}}})\), implying the compactness of \({\mathfrak m_\gamma }\). \(\square \)
The following result provides the required variational formula (4.8) and additional information about the maximizers of the continuous map \({\mathcal {E}}_{\Phi _{\gamma }}\), defined in (4.35) in the compact set \(\mathfrak {m}_\gamma \).
Theorem 4.8
Fix \(d\in \mathbb {N}\) and \(\gamma >0\), recall the set \(\mathfrak m_\gamma \) from (4.34) and the functional
Then we have the following implications:
-
\(\textbf{P}\)-almost surely,
$$\begin{aligned} \begin{aligned} \lim _{\varepsilon \downarrow 0}\varepsilon ^2\log \mu _{\gamma ,\varepsilon ^{-2}}(\Omega )&= \lim _{\varepsilon \downarrow 0}\varepsilon ^2 \textbf{E}\big [\log \mu _{\gamma ,\varepsilon ^{-2}}(\Omega )\big ] \\&= \lambda (\gamma ):=-\sup _{\vartheta \in {{\mathfrak m_\gamma }}}{{\mathscr {I}}}_{\Phi _{\gamma }}(\vartheta ) =- \sup _{\vartheta \in {\mathfrak m_\gamma }} \,\,\int _{\widetilde{{\mathcal {X}}}} \Phi _{\gamma }(\xi )\vartheta (\textrm{d}\xi ) \end{aligned} \end{aligned}$$(4.36)The above supremum in (4.36) is attained, and we always have \(\lambda (\gamma ) \in [-\gamma ^2(\phi \star \phi )(0)/2, 0]\).
-
Moreover, there exists \(\gamma _1=\gamma _1(d)\) such that \(\gamma _1>0\) if \(d\ge 3\) and if \(\gamma \in (0,\gamma _1]\), then \({\mathfrak m_\gamma }=\{\delta _{\widetilde{0}}\}\) is a singleton consisting of the Dirac measure at \(\widetilde{0} \in \widetilde{{\mathcal {X}}}\). Consequently, in this regime, \(\sup _{{\mathfrak m_\gamma }}{{\mathcal {E}}}_{\Phi _{\gamma }}(\cdot ) =0\) and also,
$$\begin{aligned} \lambda (\gamma )=0 \qquad d\ge 3 \text{ and } \gamma \in (0,\gamma _c)\text{. } \end{aligned}$$(4.37)Moreover, if \(\gamma >\gamma _1\), then \(\sup _{{\mathfrak m_\gamma }}{{\mathscr {I}}}_{\Phi _{\gamma }}>0\).
-
Finally, if \(\vartheta \in {\mathfrak m_\gamma }\) is a maximizer of \({{\mathscr {I}}}_{\Phi _{\gamma }}(\cdot )\) and \(\vartheta (\xi )>0\) for \(\xi =({\widetilde{\alpha }}_i)_{i\in I}\in \widetilde{{\mathcal {X}}}\), then \(\sum _{i\in I}\alpha _i(\mathbb {R}^d)=1\). In other words, any maximizer of (4.36) assigns positive mass only to those elements of \(\widetilde{{\mathcal {X}}}\) whose total mass add up to one.
Proof
We will briefly sketch the argument and refer to [6] for details. To derive (4.36), note that \(\log \mu _{\gamma ,T}(\Omega )\) can be decomposed
in terms of a square-integrable martingale \(M_T=\gamma \int _0^T \int _{\mathbb {R}^d} \mathbb {E}^{{\widehat{\mu }}_{\gamma ,t}}\big [\phi (y- \omega _t)\big ] \dot{B}(t,y) \textrm{d}y \textrm{d}t\) and an additive functional \(\int _0^T \Phi _\gamma (\widetilde{\mathbb Q}_t) \textrm{d}t\), where \(\mathbb {Q}_t={{\widehat{\mu }}_{\gamma ,t}[\omega _t \in \cdot ]}\in {{\mathcal {M}}}_1(\mathbb {R}^d)\) and \({\widetilde{\mathbb {Q}}}_t\in \widetilde{{\mathcal {X}}}\). As a consequence,
Here \({\mathscr {I}}_{\Phi _\gamma }\) is defined in (4.36). Therefore
Thus, studying the behavior of the left hand side reduces to studying the asymptotic behavior of \(\nu _T\) in the space \({{\mathcal {M}}}_1(\widetilde{{\mathcal {X}}})\). Because of the properties of the dynamics \(\Pi _t\) mentioned in Lemma 4.7 and its proof, it can be shown that \(d(\nu _T, m_\gamma )\rightarrow 0\) almost surely, for any \(\gamma >0\) (where d is a metric which induces the weak topology on probability measures of the compactification \(\widetilde{{\mathcal {X}}}\)). This can be used to derive the variational formula for the (quenched) free energy \(\lim _{\varepsilon \rightarrow 0}\varepsilon ^2 \log \mu _{\gamma ,\varepsilon ^{-2}}(\Omega )= - \sup _{\vartheta \in \mathfrak m_\gamma } \int \Phi _\gamma (\xi ) \vartheta (\textrm{d}\xi )\), where the infimum is taken (and given the continuity of the map \(\Pi _t\)), attained over the compact set \(\mathfrak m=\{\vartheta \in {{\mathcal {M}}}_1(\widetilde{{\mathcal {X}}}):\Pi _t(\vartheta ,\cdot )=\vartheta \,\forall t\ge 0\}\) of fixed points of \(\Pi _t(\vartheta ,\cdot )=\int \pi _t(\xi ,\cdot )\vartheta (\textrm{d}\xi )\) for \(\vartheta \in {{\mathcal {M}}}_1(\widetilde{{\mathcal {X}}})\). We refer to the proof of [6, Theorem 3.7] for details.
Note that the second part of the first item follows from Lemma 4.6 (which implies continuity of \(\widetilde{{\mathcal {X}}} \ni \xi \mapsto \Phi _\gamma (\xi )\), and therefore that of \({{\mathcal {M}}}_1(\widetilde{{\mathcal {X}}}) \ni \vartheta \rightarrow {\mathscr {I}}_{\Phi _\gamma }(\vartheta )\)) and the compactness of \(\mathfrak m_\gamma \) shown in Lemma 4.7. The bounds on \(\sup _{{\mathfrak m_\gamma }}{{\mathscr {I}}}_{\Phi _{\gamma }}\) also follows from the bounds on \(\Phi _\gamma \) shown in Lemma 4.6.
For the second item, we recall the known fact that for any \(d\ge 1\), \(\gamma \mapsto \lambda (\gamma )\) is non-decreasing and continuous in \((0,\infty )\). Thus, (4.37) follows since \(\lambda (\gamma )>0\) clearly implies that \(\lim _{T\rightarrow \infty }\mu _{\gamma ,T}(\Omega )=0\). Thus, with \(\gamma _1:=\inf \{\gamma>0:\lambda (\gamma )>0\}\), it follows that \(\gamma _1\ge \gamma _c\) with \(\gamma _c\) being defined by uniform integrability as in Remark 6 (in fact, it is conjectured that \(\gamma _1=\gamma _c\)). This shows (4.37). For the third item we refer to [6, Lemma 4.4]. \(\square \)
Lemma 4.9
For any sequence of events \((A_\varepsilon )_{\varepsilon } \subset \Omega \) with \(\mathbb {P}_0(A_\varepsilon )>0\) for all \(\varepsilon >0\), it holds \(\textbf{P}\)-almost surely that
Proof
We fix an event \(A_\varepsilon \) such that \(\mathbb {P}_0(A_\varepsilon )>0\). Then, for any \(\varepsilon , t>0\) and \(x\in \mathbb {R}^d\),
where \(D_{t,x}\) denotes the Malliavin derivative. Moreover, \(\textbf{P}\)-a.s. it holds that
In the second line of the last display, \(\omega ^{{\scriptscriptstyle {({1}})}}\) and \(\omega ^{{\scriptscriptstyle {({2}})}}\) denote two independent Brownian paths, \({\widehat{\mu }}_{\gamma ,t}^{\otimes 2}\) stands for the corresponding normalized probability measureFootnote 9 and the conditioning event \(A_\varepsilon ^{\otimes 2}\) denotes two independent copies of the same event (corresponding to \(\omega ^{{\scriptscriptstyle {({1}})}}\) and \(\omega ^{{\scriptscriptstyle {({2}})}}\)). In the second line, we have used Fubini’s theorem and symmetry of \(\phi (\cdot )\), and in the third line we have estimated \((\phi \star \phi )(\cdot ) \le (\phi \star \phi )(0)\). Hence, for any \(\varepsilon , \gamma >0\), we have that \(\Vert D \log \mu _{\gamma ,\cdot }(A_\varepsilon )\Vert _{L^2([0,\varepsilon ^{-2}]\otimes \mathbb {R}^d)}^2 \le \gamma ^2 \varepsilon ^{-2} (\phi \star \phi )(0)\). Thus, by the Gaussian concentration inequality [34, Theorem B.8.1], for any \(u>0\),
The result is now a consequence of the Borel-Cantelli lemma. Indeed, if we define \(t_1:= 1\) and
By the above upper bound and Borel-Cantelli lemma, we have that \(\textbf{P}\)-a.s.
(For notational simplicity, we write \(A_{t_n}\) instead of \(A_{\frac{1}{\sqrt{t_n}}}\)). To complete the proof, we again apply Itô’s formula to decompose \(\log \mu _{\gamma ,t}(A_t)= M_t - \frac{1}{2} \langle M_t\rangle + \frac{\gamma ^2}{2} t (\phi \star \phi )(0)\) for a continuous martingale \(M_t:= \gamma \int _0^t \int _{\mathbb {R}^d} \mathbb {E}^{{\widehat{\mu }}_{\gamma ,t}}[ \phi (\omega _t - y) \mathbb {1}_{A_t}] \dot{B}(t,y) \textrm{d}y \textrm{d}t\) which satisfies
Let us fix a sequence
Then
by Doob’s inequality. Since
we have that the probability on the left hand side of (4.44) is bounded above by \(\frac{1}{\delta _n t_{n+1}}\). If we now choose \(\delta \in (\frac{1}{2}, 1)\) sufficiently large, then these probabilities are summable and by Borel-Cantelli lemma, combined with (4.41), we have \(\textbf{P}\)-a.s.,
To conclude (4.39), we use that for any sequence \(X_t\) and \(Y_t\) satisfying \(\limsup _{t\rightarrow \infty } |X_t - Y_t|=0\), we have \(\liminf _{t\rightarrow \infty } X_t= \liminf _{t\rightarrow \infty } (Y_t + (X_t - Y_t)) \ge \liminf _{t\rightarrow \infty } Y_t\) and by reverting the argument we also have \(\liminf _{t\rightarrow \infty } Y_t \ge \liminf _{t\rightarrow \infty } X_t\). Hence, \(\liminf _{t\rightarrow \infty } X_t= \liminf _{t\rightarrow \infty } Y_t\). Invoking (4.46) and the latter fact for \(X_t= \frac{1}{t} \log \mu _{\gamma ,t}(A_{t})\) and \(Y_t= \frac{1}{t} \textbf{E}[\log \mu _{\gamma ,t}(A_{t})]\) we have (4.39). \(\square \)
5 Moments: proof of theorem 2.3
The proof of Theorem 2.3 will require the following result:
Lemma 5.1
Given \(T>0\) and \(\gamma <\gamma _c\), let \(M_T:=\sup _{0\le s\le T}\mu _{\gamma ,s}(\Omega )\) and \(M_\infty :=\lim _{T\rightarrow \infty }M_T\). Then \(\textbf{E}[M_\infty ]<\infty \).
Before proving the lemma, we introduce, for \(u>0\), the stopping time
Lemma 5.2
For every convex function \(f:[0,\infty )\mapsto {\mathbb {R}}\) and \(\gamma ,T>0\),
Proof
Let \((\tau _n)_n\) be a discrete approximation of \(\tau \) such that \(\tau _n\searrow \tau \). If (5.2) holds for \(\tau _n\), for each n, then by Fatou’s lemma we can deduce (5.2) for \(\tau \). Thus, we can assume that \(\tau \) takes values in a discrete, countable set \(\{t_i\}_{n\in \mathbb {N}}\) (which we may assume to be ordered in increasing order).
By the Markov property, if \(s\le T\),
where we remind the reader that \(\theta _{t,x}\) is the space-time shift in the environment.
If \([0,T]\cap \{t_i\}_{n\in \mathbb {N}}=\{t_1,\ldots ,t_n\} \), then using the convexity of f and Jensen’s inequality,
By Lemma 3.2, the last expression is equal to
Note that \(f(\mu _{\gamma ,T-t_i}(\Omega )\circ \theta _{t_i,\omega _{t_i}})\) is independent on \({\mathcal {F}}_{t_i}\), and so the last sum reduces to
where we used that f is convex, so that \((f(\mu _{\gamma ,T}(\Omega )))_{T\ge 0}\) is a submartingale. \(\square \)
Proof of Lemma 5.1
Given \(\varepsilon >0\), let
We note that \(f_{\varepsilon }\) is concave and for all \(x\ge 0\),
The proof is complete once we can find \(\varepsilon >0\) such that for all \(T>0\) and \(u>1\),
Indeed,
For a fixed \(u>1\), recall the definition of \(\tau =\tau _u\) from (5.1), so that \(\mu _{\gamma ,\tau }(\Omega )=u\). Hence, by Lemma 5.2 and Eq. (5.3)
We use again (5.3) to deduce
To see the last equality, note that
and
Letting \(\varepsilon \rightarrow 0\) in the last display and noting that \(\textbf{P}(\inf _{T\ge 0}\mu _{\gamma ,T}(\Omega )=0)=\textbf{P}(\mu _{\gamma }(\Omega )=0)=0\), we conclude that
Thus, for \(\varepsilon >0\) small enough, and recalling the computations from (5.5), we conclude that
which is (5.4). \(\square \)
Now we are ready to give the
Proof of Theorem 2.3
Set \(\gamma <\gamma _c\). For some fixed \(u>1\), we recall the stopping time \(\tau =\tau _u\) defined in (5.1). For any \(p>1\) and \(T>0\),
where in the last line we used Lemma 5.2. By Lemma 5.1,
so that there exists some \(u>1\) satisfying \(\textbf{P}(M_\infty >u)\le \frac{1}{2u}\). Since
we deduce from (5.6) the upper bound
If we choose \(p>1\) satisfying \(u^{p-1}<2\), we conclude that for all \(T>0\),
We turn to the proof of (2.12) and (2.13). To show (2.12), we use the stopping time \(\tau =\tau _{1/u}\) for \(u>1\) (recall (5.1)). Proceeding as in (5.6), noting that \(x\mapsto x^{-q}\) is convex on \((0,\infty )\) and using Lemma 5.2, we have
Since \(\textbf{P}(\tau \le T)\le \textbf{P}(\inf _{T\ge 0} \mu _{\gamma ,T}\le u^{-1})\rightarrow 0\) as \(u\rightarrow \infty \), we infer
for \(u>1\) large enough. If \(q\in (0,1)\) is chosen so that \(u^q\le 2\), from (5.7) we conclude that
and hence
This finishes the proof of (2.12). To show (2.13), we appeal to [11, Theorem 1.3], where it was shown that in the “\(L^2\)-region" (i.e., when the martingale \((\mu _{\gamma ,T}(\Omega ))_T\) is bounded in \(L^2(\textbf{P})\)), for all \(q\in (0,\infty )\),
Since Theorem 2.1 holds in the entire weak disorder regime (and therefore, in particular in the \(L^2\) region), the above estimate also implies (2.13) for all negative q for \(L^2\) disorder. \(\square \)
Notes
Also, in \(d\ge 3\), \(\gamma _c\) is different and strictly larger than the \(L^2\)-threshold \(\gamma _{L^2}:=\sup \{\gamma >0: \sup _T \mu _{\gamma ,T}(\Omega ) \in L^2(\textbf{P})\}\) defined by the \(L^2\)-boundedness of the martingale \(\{\mu _{\gamma ,T}(\Omega )\}_T\).
Here and through the sequel, we say for \(q>0\) that \(X\in L^{-q}\) if and only if \(X^{-q} \in L^1\).
To emphasize the dependence on the mollification, here we write \(\mu _{\gamma ,\phi ,T}= \mu _{\gamma ,T}\) for the approximating measure defined in (1.10) and \(\mu _{\gamma ,\phi }=\lim _{T\rightarrow \infty } \mu _{\gamma ,\phi ,T}\) for its limit, defined in the first part.
The metric \(d(\omega ,\eta )=\sum _n \frac{1}{2^n} \sup _{0\le t \le n}[|\omega (t)- \eta (t)| \wedge 1]\) makes \(\Omega \) a complete separable metric space, compatible with the topology of uniform convergence on compacts. However, there is a no canonical way to choose a norm on \(\Omega \) that makes it a separable Banach space.
As remarked before, we do not need the Gaussianity of the environment here – only finite exponential moment of the noise is sufficient.
Here we are using the fact that for a continuous martingale \((M_T)_T\) vanishing at 0, if \(\langle M\rangle _T \rightarrow \infty \) as \(T\rightarrow \infty \) a.s., then \(\frac{M_T}{\langle M\rangle _T} \rightarrow 0\) a.s. as \(T\rightarrow \infty \), see [23, Cor. 1, p. 144].
Here we invoke the ergodic theorem for general stationary processes which implies that the time averages of a stationary process evaluated w.r.t. a \(L^1\) function f converge almost surely to the conditional expectation of f w.r.t. the invariant \(\sigma \)-algebra. In the present context, the function (4.11) is in \(L^1(\textbf{P})\), therefore the conditional expectation \(C(\dot{B}) < +\infty \) almost surely. Moreover, for \(\gamma \in (0,\gamma _c)\), \(\mu _\gamma (\Omega )>0\) almost surely w.r.t. \(\textbf{P}\), therefore \(C(\dot{B})>0\) almost surely for s large enough, hence \(\log C(\dot{B}) < \infty \), implying (4.12).
Let \(E^\star \) denote the dual of E, with \(S: E^\star \mapsto E\) being the map defined via the integral \(S\xi = \int _E x \xi (x) \mu (d x)\) for \(\xi \in E^\star \). Then \(H_\mu \subset E\) is the Hilbert space which is the completion of the range of S under the inner product \(\langle \xi ,\xi ^\prime \rangle _\mu := \int \xi (x)\xi ^\prime (x) \mu (\textrm{d}x)\). This inner product induces a norm \(\Vert \cdot ||_\mu \) on \(H_\mu \).
Concretely, we have \({\widehat{\mu }}_{\beta ,t}^{\otimes 2}(\textrm{d}\omega ^{{\scriptscriptstyle {({1}})}}\otimes \textrm{d}\omega ^{{\scriptscriptstyle {({2}})}}) = \frac{1}{\mu _{\gamma ,t}(\Omega )^2} \exp \big [ \sum _{i=1}^2 \big (\gamma H_t(\omega ^{{\scriptscriptstyle {({i}})}}) - \frac{\gamma ^2}{2} t (\phi \star \phi )(0)\big )\big ] \mathbb {P}_0(\textrm{d}\omega ^{{\scriptscriptstyle {({1}})}}) \mathbb {P}_0(\textrm{d}\omega ^{{\scriptscriptstyle {({2}})}})\).
References
Bates, E., Chatterjee, S.: Endpoint distribution of directed polymers. Ann. Probab. 48, 817–871 (2020)
Berestycki, N.: An elementary approach to Gaussian multiplicative chaos. Electron. Commun. Probab. 22, 1–12 (2017)
Berestycki, N., Powell, E.: Gaussian free field. Liouville quantum gravity and Gaussian multiplicative chaos, Lecture notes (2021)
Bogachev, V.I.: Gaussian Measures. American Mathematical Soc, Providence (1998)
Bazaes, R., Bröker, Y., Mukherjee, C.: Geometry of the Gaussian Multiplicative Chaos in the Wiener Space. Preprint, arXiv:2008.04290 (2022)
Bröker, Y., Mukherjee, C.: Localization of the Gaussian multiplicative chaos in the Wiener space and the stochastic heat equation in strong disorder. Ann. Appl. Probab. 29, 3745–3785 (2019)
Biskup, M., Louidor, O.: Full extremal process, cluster law and freezing for the two-dimensional discrete Gaussian Free Field. Adv. Math. 330, 589–687 (2018)
Comets, F.: Directed polymers in random environments. École d” Et’ e de Probabilités de Saint- Flour XLVI - 2016 Online link
Comets, F., Cranston, M.: Overlaps and pathwise localization in the Anderson polymer model. Stoch. Process. Appl. 123, 2446–2471 (2013)
Comets, F., Cosco, C., Mukherjee, C.: Space-time fluctuation of the Kardar-Parisi-Zhang equation in \(d\ge 3\) and the Gaussian free field. Annales de l’Institut Henri Poincaré - Probabilités et Statistiques 60(1), 82–112 (2024). https://doi.org/10.1214/22-AIHP1272
Comets, F., Cosco, C., Mukherjee, C.: Renormalizing the Kardar-Parisi-Zhang equation in \(d\ge 3\) in weak disorder. J. Stat. Phys. 179, 713–728 (2020)
Comets, F., Yoshida, N.: Directed polymers in random environment are diffusive in weak disorder. Ann. Probab. 34, 1746–1770 (2006)
Duplantier, B., Rhodes, R., Sheffield, S., Vargas, V.: Critical Gaussian multiplicative chaos: convergence of the derivative martingale. Ann. Probab. 42, 1769–1808 (2014)
Duplantier, B., Rhodes, R., Sheffield, S., Vargas, V.: Renormalization of critical Gaussian multiplicative chaos and KPZ relation. Comm. Math. Phys. 330, 283–330 (2014)
Duplantier, B., Sheffield, S.: Liouville quantum gravity and KPZ. Inventiones mathematicae 185, 333–393 (2011)
Garban, C., Holden, N., Sepúlveda, A., Sun, X.: Negative moments for Gaussian multiplicative chaos on fractal sets. Electron. Commun. Probab. 23, 1–10 (2018)
Junk, S.: New characterization of the weak disorder phase of directed polymers in bounded random environments. Comm. Math. Phys. 389(2), 1087–1097 (2022)
Kahane, J.-P.: Sur le chaos multiplicatif. Ann. SC. Math. Quebec. 9(2), 105–150 (1985)
Kahane, J.-P.: Une inégalité du type de Slepian et Gordon sur les processus Gaussiens. Israel J. Math. 55(1), 109–110 (1986)
Khas’minski, R.Z.: On a positive solution of the equation \(Au+ Vu=0\). Theoret. Probab. Appl. 4, 309–318 (1959)
Li, W.V.: A Gaussian correlation inequality and its applications to small ball probabilities electron. Commun. Probab. 4, 111–118 (1999)
Li, W.V., Shao, Q.-M.: Gaussian processes: inequalities, small ball probabilities and applications. Handbook Stat. 19, 533–597 (2001)
Liptser, R.S.: Theory of Martingales. Springer, Dordrecht (1989)
Madaule, T., Rhodes, R., Vargas, V.: Glassy phase and freezing of log-correlated Gaussian potentials. Ann. Appl. Probab. 26, 643–690 (2016)
Mukherjee, C., Shamov, A., Zeitouni, O.: Weak and strong disorder for the stochastic heat equation and continuous directed polymers in \(d\ge 3\). Electron. Commun. Probab. 21, 1–12 (2016)
Mukherjee, C., Varadhan, S.R.S.: Brownian occupation measures, compactness and large deviations. Ann. Probab. 44, 3934–3964 (2016)
Powell, E.: Critical Gaussian multiplicative chaos: a review. Markov Process. Relat. Fields. 27, 557–606 (2021)
Powell, E.: Critical Gaussian chaos: convergence and uniqueness in the derivative normalisation. Electron. J. Prob. 23(31), 1–26 (2018)
Robert, R., Vargas, V.: Gaussian multiplicative chaos revisited. Ann. Probab. 38(2), 605–631 (2010)
Rhodes, R., Vargas, V.: Gaussian multiplicative chaos and applications: a review. Prob. Surv. 11, 315–392 (2014)
Rhodes, R., Vargas, V.: The tail expansion of Gaussian multiplicative chaos and the Liouville reflection coefficient. Ann. Probab. 47, 3082–3107 (2019)
Rovira, C., Tindel, S.: On the Brownian-directed polymer in a Gaussian random environment. J. Funct. Anal. 222, 178–201 (2005)
Shamov, A.: Gaussian multiplicative chaos. J. Funct. Anal. 270, 3224–3261 (2016)
Üstünel, S., Zakai, M.: Transformation of measure on Wiener space. Springer- Verlag, Berlin (2000)
Acknowledgements
We would like to thank Nathanael Berestycki for explaining his approach [2], sharing many insights and comments on previous versions, and Vincent Vargas for very inspiring communications and comments on the first draft, in particular about characterization of GMC and Hölder exponents. We would also like to thank an anonymous referee for a very careful reading and valuable suggestions for improvement. The research of all three authors is funded by the Deutsche Forschungsgemeinschaft (DFG) under Germany’s Excellence Strategy EXC 2044-390685587, Mathematics Münster: Dynamics-Geometry-Structure.
Funding
Open Access funding enabled and organized by Projekt DEAL.
Author information
Authors and Affiliations
Contributions
All authors contributed equally.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Bazaes, R., Lammers, I. & Mukherjee, C. Subcritical Gaussian multiplicative chaos in the Wiener space: construction, moments and volume decay. Probab. Theory Relat. Fields (2024). https://doi.org/10.1007/s00440-024-01271-7
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00440-024-01271-7
Keywords
- Continuous directed polymer
- Weak disorder
- Gaussian multiplicative chaos
- Liouville quantum gravity
- Thick points
- Girsanov transformation
- Moments
- Hölder exponents
- Small ball probability
- Multifractal spectrum