ScalingStacks

5.4. Maslov’s dequantization [04T8]

Original official author HTML, exact retained edition. Historical TeX conversion verdicts remain unchanged. Cited-edition alignment and mathematical self-containment are not assessed.

Complete original source context · Original author HTML

5.4. Maslov’s dequantization

Consider the following family of binary operations on ℝ∋x,y\mathbb{R}\ni x,y:

x⊕ty=logt(tx+ty),x\oplus_{t}y=\log_{t}(t^{x}+t^{y}),

for t>1t>1 and

x⊕∞y=limt→0x⊕ty=max{x,y}.x\oplus_{\infty}y=\lim\limits_{t\to 0}x\oplus_{t}y=\max\{x,y\}.

This is a commutative semigroup operation (no inverse elements and no zero) for each tt. The set ℝ\mathbb{R} equipped with this operation for addition and with x⊙y=x+yx\odot y=x+y for multiplication is a semiring ℝt\mathbb{R}_{t}. Indeed, for any x,y,z∈ℝx,y,z\in\mathbb{R} we have x⊙(y⊕tz)=(x⊙z)⊕t(y⊙tz)x\odot(y\oplus_{t}z)=(x\odot z)\oplus_{t}(y\odot_{t}z).

Passing from a finite tt to infinity in this family of semirings is called Maslov’s dequantization, cf. [10]. Note that for all finite values of tt the semiring is isomorphic to the semiring of real positive numbers equipped with the usual addition and multiplication. But the behavior at t=∞t=\infty is qualitatively different, the addition becomes idempotent, x⊕∞x=xx\oplus_{\infty}x=x. The prefix “de” reflects the fact that in this deformation the classical calculus operations appear on the quantum side.

There is a universal bound for the convergence of the operations ⊕t\oplus_{t} to ⊕∞=max\oplus_{\infty}=\max. Namely, we have

(4) max{x1,…,xN}≤x1⊕t⋯⊕txN≤max{x1,…,xN}+logtN\max\{x_{1},\dots,x_{N}\}\leq x_{1}\oplus_{t}\dots\oplus_{t}x_{N}\leq\max\{x_{1},\dots,x_{N}\}+\log_{t}N

The dequantization point of view can be used to reinterpret Viro’s patchworking, see [16]. Instead of deforming the coefficients of the polynomial we may keep them constant, but deform the addition operation instead. This point of view yields some useful estimates on the zero set of the patchworking polynomial as shown below.

One way to think of a polynomial is to think of it as a collection of coefficients at its monomials. Fix a polynomial p⁡(x)=∑jcj​xjp(x)={\sum\limits_{j}}c_{j}x^{j} in n+1n+1 variables, where the arithmetic operations are taken from the semiring ℝt\mathbb{R}_{t}. Depending on tt this polynomial defines different functions pt:ℝn+1→ℝp_{t}:\mathbb{R}^{n+1}\to\mathbb{R}. Note that the function

ft​(z)=tpt​(Logt⁡(z))f_{t}(z)=t^{p_{t}(\operatorname{Log}_{t}(z))}

coincides with the patchworking polynomials where all aj=1a_{j}=1 and v⁡(j)=cjv(j)=c_{j}. Here Logt⁡(z1,…,zn+1)=(logt⁡(z1),…,log⁡(zn+1)).\operatorname{Log}_{t}(z_{1},\dots,z_{n+1})=(\log_{t}(z_{1}),\dots,\log(z_{n+1})).

Lemma 5.3.

If a point x∈ℝn+1x\in\mathbb{R}^{n+1} belongs to the amoeba

Logt⁡({z∈(ℂ∗)n+1|ft​(z)=0})\operatorname{Log}_{t}(\{z\in(\mathbb{C}^{*})^{n+1}\ |\ f_{t}(z)=0\})

then the monomials cj​xjc_{j}x^{j} from ptp_{t} satisfy to the generalized triangle inequality in ℝt\mathbb{R}_{t}, i.e. for each index kk we have

ck⊙xk≤⨁j≠kcj⊙xj.c_{k}\odot x^{k}\leq{\bigoplus\limits_{j\neq k}}c_{j}\odot x^{j}.
Proof.

If x=Logt⁡(z)x=\operatorname{Log}_{t}(z) with ft​(z)=0f_{t}(z)=0 then zero is the sum of the monomials tcj​zjt^{c_{j}}z^{j} is zero and thus their norms must satisfy to the triangle inequality. ∎

Let ft=∑j∈Δ∩ℤn+1aj​t−v⁡(j)​zjf_{t}=\sum\limits_{j\in\Delta\cap\mathbb{Z}^{n+1}}a_{j}t^{-v(j)}z^{j} now be a general patchworking polynomial. Denote Vt∘={ft=0}⊂(ℂ∗)n+1{V}^{\circ}_{t}=\{f_{t}=0\}\subset(\mathbb{C}^{*})^{n+1}. The family ftf_{t} can be treated as a single polynomial in (K∗)n+1(K^{*})^{n+1} (see 5.2). It defines a hypersurface VK∘⊂(K∗)n+1{V}^{\circ}_{K}\subset(K^{*})^{n+1}. Consider the Hausdorff metric on the subsets of ℝn+1\mathbb{R}^{n+1} induced by the Euclidean metric on ℝn+1\mathbb{R}^{n+1}. Denote 𝒜t=Logt⁡(Vt∘)\mathcal{A}_{t}=\operatorname{Log}_{t}({V}^{\circ}_{t}) and 𝒜K=LogK⁡(VK∘)\mathcal{A}_{K}=\operatorname{Log}_{K}({V}^{\circ}_{K}).

Corollary 5.4.

The amoebas 𝒜t\mathcal{A}_{t} converge in the Hausdorff metric to the non-Archimedian amoeba 𝒜K\mathcal{A}_{K} when t→∞t\to\infty.

Proof.

Lemma 5.3 and the inequality (4) imply that 𝒜t\mathcal{A}_{t} converge to a subset of 𝒜K\mathcal{A}_{K}. Indeed, for each tt we can rewrite |aj​tv⁡(j)​zj||a_{j}t^{v(j)}z^{j}| as |tcj​zj||t^{c_{j}}z^{j}|, cj=v⁡(j)+logt⁡|aj|c_{j}=v(j)+\log_{t}|a_{j}|. Such a monomial induces a linear function cj+j​xc_{j}+jx in ℝn+1\mathbb{R}^{n+1}. The inequalities

(5) ck+k​x≤maxj≠k⁡(cj+j​x)+logt⁡(N),c_{k}+kx\leq\max\limits_{j\neq k}(c_{j}+jx)+\log_{t}(N),

where N+1N+1 is the number of monomials in ftf_{t}, cut out a uniformly bounded neighborhood of 𝒜K\mathcal{A}_{K} which contains 𝒜K\mathcal{A}_{K}.

The limit of 𝒜t\mathcal{A}_{t} cannot be any smaller than 𝒜K\mathcal{A}_{K} by the following topological reason. A component of the complement of the set described by the inequalities (5) is given by the inequality ck+k​x>maxj≠k⁡(cj+j​x)+logt⁡(N)c_{k}+kx>\max\limits_{j\neq k}(c_{j}+jx)+\log_{t}(N). By [3] this component is contained in the component of ℝn+1∖𝒜t\mathbb{R}^{n+1}\smallsetminus\mathcal{A}_{t} corresponding to the index kk. Thus, different components of the set described by (5) must be contained in different components of ℝn+1∖𝒜t\mathbb{R}^{n+1}\smallsetminus\mathcal{A}_{t}. ∎

This corollary can be strengthened to describe the limits of the varieties Vt∘⊂(ℂ∗)n+1{V}^{\circ}_{t}\subset(\mathbb{C}^{*})^{n+1} under the corresponding renormalization of the norms of their points. The description is in terms of the lifts of non-Archimedian amoebas, see 5.3. Let Ht:(ℂ∗)n+1→(ℂ∗)n+1H_{t}:(\mathbb{C}^{*})^{n+1}\to(\mathbb{C}^{*})^{n+1} be the transformation defined by

Ht​(z1,…,zn+1)=(t−|z1|​z1|z1|,…,t−|zn+1|​zn+1|zn+1|).H_{t}(z_{1},\dots,z_{n+1})=(t^{-|z_{1}|}\frac{z_{1}}{|z_{1}|},\dots,t^{-|z_{n+1}|}\frac{z_{n+1}}{|z_{n+1}|}).

We have Logt=Log∘Ht\operatorname{Log}_{t}=\operatorname{Log}\circ H_{t}.

Theorem 5.

The sets Ht​(Vt∘)H_{t}({V}^{\circ}_{t}) converge in the Hausdorff metric to W⁡(VK∘)W({V}^{\circ}_{K}) when t→∞t\to\infty.

The proof is the same as the proof of Corollary 5.4. The only difference we have to make is to incorporate the arguments of the monomials to the inequalities (5).

Original mathematics by the credited authors. Source-backed reader collection; mathematical self-containment is not assessed.