The upper and lower Maslov dequantizations are respectively the limits $h \downarrow 0$ and $h \uparrow 0$ of deformations of the semifield $(\mathbb{R}_+,+,\cdot)$ defined for $0 \ne h \in \mathbb{R}$ by the map \begin{equation} \Phi_h(x) := h \log x \end{equation} which sends $\mathbb{R}_+$ to $\mathbb{R} \cup \{-\sigma(h) \cdot \infty\}$. Since the map $\Phi_h$ is invertible, it defines deformed addition and multiplication operations: i.e., \begin{equation} \label{eq:MaslovArithmetic} u +_h v := \Phi_h (\Phi_h^{-1}(u) + \Phi_h^{-1}(v)); \quad u \cdot_h v := \Phi_h (\Phi_h^{-1}(u) \cdot \Phi_h^{-1}(v)). \end{equation} Explicitly, we have \begin{equation} u +_h v := h \log \left ( e^{u/h} + e^{v/h} \right ); \quad u \cdot_h v := h \log \left ( e^{u/h} \cdot e^{v/h} \right ) \equiv u+v. \end{equation} Now writing $\lozenge$ for $\vee$ when $h > 0$ and $\wedge$ when $h < 0$, it is easy to see that $(\sum_{(h)})_{j \in [n]} u_j := u_1 +_h \dots +_h u_n = h \log \sum_{j \in [n]} e^{u_j/h} = \lozenge_j u_j + h \log \sum_j e^{(u_j-\lozenge_k u_k)/h}$. Clearly, as $h \rightarrow 0$, this tends to $\lozenge_j u_j$. This is manifested in the figure below:

enter image description here

There is a well-established theory of analysis in the dequantization limit $h \rightarrow 0$: this is essentially the Ansatz \begin{equation} ^\lozenge : \sum \mapsto \lozenge; \quad ^\lozenge : \prod \mapsto \sum \end{equation} where again $\lozenge$ is either $\vee$ or $\wedge$, as appropriate. The first of the correspondences above naturally leads to \begin{equation} ^\lozenge : \int \mapsto \text{ext}^\lozenge, \end{equation} where here the extremum is given by $\text{ext}^\vee := \sup$ and $\text{ext}^\wedge := \inf$. This turns out to have a rigorous interpretation, so that, e.g., \begin{equation} ^\lozenge : \int K(t,u) f(u) du \mapsto \text{ext}^\lozenge_u \left ( K(t,u) + f(u) \right ) \end{equation} is not only suggestive but correct, and in fact such integral transforms exhaust the class of linear operators. There is also a correspondence between the Fourier and Legendre transformations that is a bit trickier.


Is there a reasonable notion of deformed integral for the case where $h$ is small but nonzero?

The Ansatz suggests something which diverges as $\Delta x \downarrow 0$ unless simultaneously $h \rightarrow 0$ in a controlled way, in which event the extremum results. For my purposes this is degenerate.

  • $\begingroup$ I do not quite understand your question. You can prove that the integral of a tropical function $f$ over an interval gives you tropically the maximal value of the function. What else do you want? $\endgroup$ Mar 7 '17 at 9:15
  • $\begingroup$ @NikitaKalinin Tropical is the limit $h \rightarrow 0$. I want $h$ small but nonzero. $\endgroup$ Mar 7 '17 at 10:24

So my original thinking was that a Riemann sum over $[a,b]$ with $x_0 := a$, $x_n := b$ and $x_j := (n-j)x_0/n + jx_1/n$ would deform to $(\sum_{(h)})_{j = 1}^n (f(x_j) + \Delta x)$, where $\Delta x = (a-b)/n = x_j-x_{j-1}$. This deformed sum equals $\Delta x + h \log \sum_j \exp(f(x_j)/h)$, which as I suggested in my question diverges for fixed $h$ as $\Delta x \rightarrow 0$. If instead we take $h \rightarrow 0$, then this just yields the dequantization limit, i.e., the extremum of $f$.

A better approach seems to be to perform a "Euclidean Wick rotation" and set $\beta := -1/h$: then the deformed sum is just the "free energy", which yields a heuristic for the deformed integral, viz. $\left ( \int_\beta \right ) f \equiv -\frac{1}{\beta} \log \int e^{-\beta f(x)} dx$. Of course it is generally very difficult to get good behavior out of an integral of this form.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.