Skip to the Main Content

Note:These pages make extensive use of the latest XHTML and CSS Standards. They ought to look great in any standards-compliant modern browser. Unfortunately, they will probably look horrible in older browsers, like Netscape 4.x and IE 4.x. Moreover, many posts use MathML, which is, currently only supported in Mozilla. My best suggestion (and you will thank me when surfing an ever-increasing number of sites on the web which have been crafted to use the new standards) is to upgrade to the latest version of your browser. If that's not possible, consider moving to the Standards-compliant and open-source Mozilla browser.

September 9, 2024

The Space of Physical Frameworks (Part 3)

Posted by John Baez

In Part 1, I explained how statistical mechanics is connected to a rig whose operations depend on a real parameter β\beta and approach the ‘tropical rig’, with operations min\min and ++, as β+\beta \to +\infty. I explained my hope that if we take equations from classical statistical mechanics, expressed in terms of this β\beta-dependent rig, and let β+\beta \to +\infty, we get equations in thermodynamics. That’s what I’m slowly trying to show.

As a warmup, last time I explained a bit of thermodynamics. We saw that some crucial formulas involve Legendre transforms, where you take a function f:[,]f \colon \mathbb{R} \to [-\infty,\infty] and define a new one f˜:[,]\tilde{f} \colon \mathbb{R} \to [-\infty,\infty] by

f˜(s)=inf x(sxf(x)) \tilde{f}(s) = \inf_{x \in \mathbb{R}} (s x - f(x))

I’d like the Legendre transform to be something like a limit of the Laplace transform, where you take a function ff and define a new one f^\hat{f} by

f^(s)= e sxf(x)dx \hat{f}(s) = \int_{-\infty}^\infty e^{-s x} f(x) \, d x

Why do I care? As we’ll see later, classical statistical mechanics features a crucial formula that involves a Laplace transform. So it would be great if we could find some parameter β\beta in that formula, take the limit β+\beta \to +\infty, and get a corresponding equation in thermodynamics that involves a Legendre transform!

As a warmup, let’s look at the purely mathematical question of how to get the Legendre transform as a limit of the Laplace transform — or more precisely, something like the Laplace transform. Once we understand that, we can tackle the physics in a later post.

From Legendre to Laplace: the rough idea

In Part 1 we saw that for each β>0\beta \gt 0 we can make (0,](0,\infty] into a rig with addition

x βy=1βln(e βx+e βy) x \oplus_\beta y = -\frac{1}{\beta} \ln(e^{-\beta x} + e^{-\beta y})

As β+\beta \to +\infty, this operation approaches

x y=xminy x \oplus_\infty y = x \min y

and we get a version of the tropical rig. (There’s a more popular version using max\max instead of min\min, but I decided to use min\min.)

Suppose we want to define a notion of integration with β\oplus_\beta replacing ordinary addition. We could call it ‘β\beta-integration’ and denote it by β\int_\beta. Then it’s natural to try this:

βf(x)dx=1βlne βf(x)dx \int_\beta f(x) \, d x = - \frac{1}{\beta} \ln \int e^{-\beta f(x)} d x

If the function ff is nice enough, we could hope that

lim β+ βf(x)dx=inf xf(x) \lim_{\beta \to +\infty} \int_\beta f(x) \, d x \; = \; \inf_x f(x)

Then, we could hope to express the Legendre transform

f˜(s)=inf x(sxf(x)) \tilde{f}(s) = \inf_{x \in \mathbb{R}} (s x - f(x))

as the β+\beta \to + \infty limit of some transform involving β\beta-integration.

Indeed, in Section 6 here, Litvinov claims that the Legendre transform is the β=\beta = \infty analogue of the Laplace transform:

But he doesn’t state any result saying that it’s a limit of the Laplace transform, or something like that.. Touchette states a result along these lines here:

He even applies it to classical statistical mechanics! But he’s operating as a physicist, not a mathematician, so he doesn’t state a precise theorem. I’d like to take a crack at that, just to be sure I’m not fooling myself.

The Legendre transform as a limit

Touchette’s formula gives an Legendre transform involving a sup rather than an inf. I slightly prefer a version with an inf. We’ll get the Legendre transform as a limit of something that is not exactly a Laplace transform, but close enough for our physics application:

Almost Proved Theorem. Suppose that f:[0,)f \colon [0,\infty) \to \mathbb{R} is a concave function with continuous second derivative. Suppose that for some s>0s \gt 0 the function sxf(x)s x - f(x) has a unique minimum at x 0x_0, and f(x 0)<0f''(x_0) \lt 0. Then as β+\beta \to +\infty we have

1βln 0 e β(sxf(x))dxinf x(sxf(x)) \displaystyle{ -\frac{1}{\beta} \ln \int_0^\infty e^{-\beta (s x - f(x))} \, d x \quad \longrightarrow \quad \inf_x \left( s x - f(x)\right) }

Almost Proof. Laplace’s method should give the asymptotic formula

0 e β(sxf(x))dx2πβf(x 0)e β(sx 0f(x 0) \displaystyle{ \int_0^\infty e^{-\beta (s x - f(x))} \, d x \quad \sim \quad \sqrt{\frac{2\pi}{-\beta f''(x_0)}} \; e^{-\beta(s x_0 - f(x_0)} }

Taking the logarithm of both sides and dividing by β-\beta we get

lim β+1βln 0 e β(sxf(x))dx=sx 0f(x 0) \displaystyle{\lim_{\beta \to +\infty} -\frac{1}{\beta} \ln \int_0^\infty e^{-\beta(s x - f(x))} \, d x \; = \; s x_0 - f(x_0) }

since

lim β+1βln2πβf(x 0)=0 \displaystyle{ \lim_{\beta \to +\infty} \frac{1}{\beta} \; \ln \sqrt{\frac{2\pi}{-\beta f''(x_0)}} = 0 }

Since sxf(x)s x - f(x) has a minimum at x 0x_0 we get the desired result:

1βln 0 e β(sxf(x))dxinf x(sxf(x)) \displaystyle{ -\frac{1}{\beta} \ln \int_0^\infty e^{-\beta (s x - f(x))} \, d x \quad \longrightarrow \quad \inf_x \left( s x - f(x)\right) }

The only tricky part is that Laplace’s method as proved here requires finite limits of integration, while we are integrating from 00 all the way up to \infty. However, the function sxf(x)s x - f(x) is concave, with a minimum at x 0x_0, and it has positive second derivative there since f(x 0)<0f''(x_0) \lt 0. Thus, it grows at least linearly for large xx, so as β+\beta \to +\infty the integral

0 e β(sxf(x))dx \displaystyle \int_0^\infty e^{-\beta(s x - f(x))} \, d x

can be arbitrarily well approximated by an integral over some finite range [0,a][0,a].       ∎

Someone must have studied the hell out of this issue somewhere — do you know where?

Now, let’s look at the key quantity in the above result:

1βln 0 e β(sxf(x))dx \displaystyle{ - \frac{1}{\beta} \ln \int_0^\infty e^{-\beta (s x - f(x))} \, d x }

We’ve got a 1/β-1/\beta and then the logarithm of an integral that is not exactly a Laplace transform… but it’s close! In fact it’s almost the Laplace transform of the function

g β(x)=e βf(x) g_\beta(x) = e^{\beta f(x)}

since

0 e β(sxf(x))dx= 0 e βsxg β(x)dx \displaystyle{ \int_0^\infty e^{-\beta (s x - f(x))} \, d x = \int_0^\infty e^{-\beta s x} \, g_\beta(x) \, d x }

The right hand side here would be the Laplace transform of g βg_\beta if it weren’t for that β\beta in the exponential.

So, it seems to be an exaggeration to say the Legendre transform is a limit of the Laplace transform. It seems to be the limit of something that’s related to a Laplace transform, but more complicated in a number of curious ways.

This has made my life difficult (and exciting) for the last few weeks. Right now I believe that all the curious complications do exactly what we want them to in our physics applications. But I should hold off on declaring this until I write up all the details: I keep making computational mistakes and fixing them, with a roller-coaster of emotions as things switch between working and not working.

Posted at September 9, 2024 3:20 PM UTC

TrackBack URL for this Entry:   https://golem.ph.utexas.edu/cgi-bin/MT-3.0/dxy-tb.fcgi/3556

8 Comments & 1 Trackback

Re: The Space of Physical Frameworks (Part 3)

David Corfield may be happy to hear that I’m finally answering his question from April 2014:

Are we going to get to find out the commonality between the Legendre and Laplace transforms (and others)?

Igor Khavkine’s reply to that question was spot on, since Laplace’s method is a cousin to the saddle point or steepest descent method. But there seem to be some subtleties… which I’m tackling now.

Posted by: John Baez on September 11, 2024 4:30 PM | Permalink | Reply to this

Re: The Space of Physical Frameworks (Part 3)

What you seem to want looks like an application of Varadhan’s lemma. The Varadhan’s lemma gives the limit β\beta \to \infty of the “scaled cumulant generating function” 1βln𝔼e βϕ(X)\frac{1}{\beta}\ln \mathbb{E} e^{\beta \phi(X)}, with Xg βX \sim g_\beta and ff being the large deviation rate function of g βg_\beta, as a Legendre transform of ff.

Posted by: Malo Tarpin on September 11, 2024 5:05 PM | Permalink | Reply to this

Re: The Space of Physical Frameworks (Part 3)

https://mathoverflow.net/a/264096 and question above of motivational interest

Posted by: Steve Huntsman on September 12, 2024 1:17 PM | Permalink | Reply to this

Re: The Space of Physical Frameworks (Part 3)

Posted by: Steve Huntsman on September 12, 2024 1:21 PM | Permalink | Reply to this

Re: The Space of Physical Frameworks (Part 3)

Inversion problem, Legendre transform and inviscid Burgers’ equations” by Wenhua Zhao relates compositional inversion to the Legendre transform. See the last paragraph on pg. 301 and note that the Legendre transform f¯(z)\bar{f}(z) there is of the form of a variational principle given by Govind Menon in his publications on the inviscid Burgers’ equation as I present on pg. 3 of my pdf “Compositional Inverse Pairs, the Inviscid Burgers-Hopf Equation, and the Stasheff Associahedra”. (See also Eqn. 14 on pg. 617 of “Making sense of the Legendre transform” by Zia, Redish, and Mckay, a variant of the inverse function theorem.)

Now compositional inversion can be related to the Laplace transform with the simplest examples given for the classic Fuss-Catalan sequences of numbers as demonstrated in my pdf “Discriminating Deltas, Depressed Equations, and Generalized Catalan Numbers”. The last section of these notes shows how to achieve the compositional inversions that define these sequences via the Legendre-Fenchel transform, analytically and graphically. (I have written a few later sets of notes on connections between the Laplace transform and compositional inversion. Compositional inversion and the Legendre transform are also related to quadratic operads as noted in my MO-Q “Inversion, Koszul duality, combinatorics and geometry”.)

Note also eqn. 4 on pg. 6 of “Legendre Transform, Hessian Conjecture and Tree Formula” by Meng for an equation of the logarithmic type that you discuss.

Posted by: Tom Copelandt on September 12, 2024 4:26 PM | Permalink | Reply to this

Re: The Space of Physical Frameworks (Part 3)

There’s an interesting account of a Legendre transform in terms of symmetric functions in

\S 7.14 p 30 of E Getzler, M Kapranov, Modular operads

https://arxiv.org/abs/dg-ga/9408003

Posted by: jack morava on September 13, 2024 2:01 PM | Permalink | Reply to this

Re: The Space of Physical Frameworks (Part 3)

The log of a Bargmann-Segal (Gaussian) transform, similar to your logarithmic formula, is presented by Getzler in “The semi-classical approximation for modular operads”. He relates this to the Legendre transform LL of a function ff, which he characterizes at the top of pg. 4 (and on pg. 2) as (Lf)f=x.(Lf)'\circ f' = x. (Never have truly understood the math behind this paper although the table beneath Theorem (1.3) contains the Lagrange inversion polynomials of OEIS A134685. If anyone can point me to another more pedestrian explanatory set of notes, I’d appreciate it.)

A general perspective to incorporate into any comprehensive theory of entropy and its relation to the Legendre transform is presented in “The concept of duality in convex analysis, and the characterization of the Legendre transform” by Artstein-Avidan and Milman.

Posted by: Tom Copeland on September 13, 2024 8:50 PM | Permalink | Reply to this

Re: The Space of Physical Frameworks (Part 3)

I explored the Borel-Laplace transform

0 1zexp(utF(u)z)du \int_0^{\infty} \frac{1}{z}\exp\left(-\frac{u-t\;F(u)}{z}\right) du

in relation to compositional inversion in “The Lagrange Reversion Theorem and Inversion Formula Revisited”.

And, here is a simple geometric interpretation of the Legendre transform, aside from the one involving a tangent envelope, that illustrates the connection to compositional inversion.

Given a function f(x)f(x) with f(0)=0f(0)=0 and f(0)>0f'(0) \gt 0. Draw the curve y=f(x)y=f(x) from the origin to the point (x e,y e=f(x e))(x_e,y_e=f(x_e)) a small distance from the origin. The compositional inverse function is associated with the curve x=f(y)x = f(y), or y=f (1)(x)y = f^{(-1)}(x). The rectangular box with diagonal from (0,0)(0,0) to (x e,y e)=(f (1)(y e),f(x e))(x_e,y_e)= (f^{(-1)}(y_e),f(x_e)) has area x ey ex_e \cdot y_e, and, within the box, this is equal to the area between the curve y=f(x)y=f(x) and the xx-axis plus the area between the same curve and the yy-axis; that is,

x eye= 0 x ef(x)dx+ 0 y ef (1)(y)dy=F(x e)+G(y e).x_ey_e = \int_0^{x_e} f(x) dx + \int_0^{y_e} f^{(-1)}(y)dy = F(x_e) + G(y_e).

Then we have the Legendre transform

G(y e)=x ey eF(x e)G(y_e) = x_e y_e - F(x_e)

with

F(x)=f(x)F'(x) = f(x) and G(x)=f (1)(x),G'(x) = f^{(-1)}(x),

giving a relation between the Legendre transform and compositional inversion.

Posted by: Tom Copeland on September 15, 2024 3:24 AM | Permalink | Reply to this
Read the post The Space of Physical Frameworks (Part 4)
Weblog: The n-Category Café
Excerpt: I'll explain exactly what I mean by 'classical statistical mechanics', and how free entropy is defined in this subject. Then I'll show that as Boltzmann's constant approaches zero, this free entropy approaches the free entropy we've already seen in t...
Tracked: September 30, 2024 9:18 PM

Post a New Comment