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.

August 5, 2025

(BT) Diversity from (LC) Diversity

Posted by Tom Leinster

Guest post by Mark Meckes

Around 2010, in papers that both appeared in print in 2012, two different mathematical notions were introduced and given the name “diversity”.

One, introduced by Tom Leinster and Christina Cobbold, is already familiar to regular readers of this blog. Say XX is a finite set, and for each x,yXx,y \in X we have a number Z(x,y)=Z(y,x)[0,1]Z(x,y) = Z(y,x) \in [0,1] that specifies how “similar” xx and yy are. (Typically we also assume Z(x,x)=1Z(x,x) = 1.) Fix a parameter q[0,]q \in [0,\infty]. If pp is a probability distribution on XX, then the quantity D q Z(p)=( xsupp(p)( ysupp(p)Z(x,y)p(y)) q1p(x)) 1/(1q) D_q^Z(p) = \left(\sum_{x\in supp(p)} \left( \sum_{y\in supp(p)} Z(x,y) p(y)\right)^{q-1} p(x)\right)^{1/(1-q)} (with the cases q=1,q=1,\infty defined by taking limits) can be interpreted as the “effective number of points” in XX, taking into account both the similarities between points as quantified by ZZ and the weights specified by pp. Its logarithm logD q Z(p)\log D_q^Z(p) is a refinement of the qq-Rényi entropy of pp. The main motivating example is when XX is a set of species of organisms present in an ecosystem, and D q Z(p)D_q^Z(p) quantifies the “effective number of species” in XX, accounting for both similarities between species and their relative abundances. This family of quantities turns out to subsume many of the diversity measures previously introduced in the theoretical ecology literature, and they are now often referred to as Leinster–Cobbold diversities.

The parameter qq determines how much D q Z(p)D_q^Z(p) counts the very “rare” points (those for which p(x)p(x) is very small). An interesting question from an ecological point of view is, given XX and ZZ, which probability distribution pp maximizes the diversity D q Z(p)D_q^Z(p)? It turns out that the answer is independent of qq. Moreover, if XX is a metric space and Z(x,y)=e d(x,y)Z(x,y) = e^{-d(x,y)}, this maximum diversity D(X):=max pD q Z(p) D(X) := \max_p D_q^Z(p) is an isometric invariant closely related to the magnitude of XX. It also extends in a natural way to compact metric spaces.

Independently, David Bryant and Paul Tupper defined a diversity on a set XX to be a [0,)[0,\infty)-valued function δ\delta on the finite subsets of XX which satisfies:

  • δ(A)=0\delta(A) = 0 if AA has at most one element, and

  • δ(AB)δ(AC)+δ(CB)\delta(A\cup B) \le \delta(A \cup C) + \delta(C \cup B) whenever CC \neq \emptyset.

I will refer to a diversity in this sense as a BT diversity. If δ\delta were defined only on sets with at most two elements, this would amount to the definition of a metric. In fact, if dd is a metric on XX, then δ(A)=diam(A):=max a,bAd(a,b) \delta(A) = diam (A) := \max_{a,b \in A} d(a,b) defines a BT diversity on XX, so BT diversities are actually a generalization of metrics.

Here as well, the motivation for the name “diversity” comes from an example in theoretical ecology: suppose XX is a set of species in a phylogenetic tree TT. Define δ(A)\delta(A) to be the length of the smallest subtree of TT containing AA. Then δ\delta is a BT diversity, known in the literature as phylogenetic diversity. However, just as with the maximum diversity discussed above, most of the subsequent work on BT diversities has focused on geometric examples.

So we now have two seemingly quite different geometric notions, introduced about the same time, going by strikingly similar names for conceptually similar reasons. One can’t help wondering, do they have something to do with each other? In particular, could maximum (LC) diversity be an example of a BT diversity?

In a new paper with Gautam Ashwarya, Dongbin Li, and Mokshay Madiman, we show that, after a minor tweak, maximum diversity does give rise to a BT diversity. The minor tweak is necessary to handle the first condition in the definition of BT diversity: if XX is a metric space and xXx \in X, it’s easy to check that D({x})=1D(\{x\}) = 1, whereas a BT diversity must satisfy δ({x})=0\delta(\{x\}) = 0. This can be dealt with in the simplest imaginable way:

Theorem 1 Let XX be a metric space. For each nonempty finite AXA \subseteq X set δ(A)=D(A)1\delta(A) = D(A) - 1, and define also δ()=0\delta(\emptyset) = 0. Then δ\delta is a BT diversity on XX.

(In the paper itself, we adopt the term complexity when referring to the quantities logD q Z(p)\log D_q^Z(p) and logD(X)\log D(X), and state most of the results in terms of complexity instead of maximum diversity; we further deduce from Theorem 1 that the complexity logD(X)log D(X) is also a BT diversity. This terminology is used partly to cut down on the potential confusion created by using “diversity” in multiple ways. It also alludes to the relationship between logD q Z(p)\log D_q^Z(p) and Rényi entropy, which is widely understood as a measure of “complexity”. Further connections between LC complexity and Rényi entropy are the subject of forthcoming work that I hope to be able to tell you more about soon! But for the remainder of this blog post I’ll stick to the maximum diversity formulation.)

Interestingly, maximum diversity has some properties that are quite nice and natural, but turn out to make it intriguingly different from the heretofore most thoroughly studied BT diversities. For example, D=1+δD = 1 + \delta has the following subadditivity property, which is not shared by the functional 1+diam1 + diam:

Theorem 2 Let XX be a metric space, and let A 1,,A nXA_1, \ldots, A_n \subseteq X be compact subsets. Then D( i=1 nA i) i=1 nD(A i). D\left(\bigcup_{i=1}^n A_i \right) \le \sum_{i=1}^n D(A_i).

Maximum diversity actually satisfies a much stronger property called fractional subadditivity, which arises naturally in inequalities for entropy. Another special case of fractional subadditivity is the following.

Theorem 3 Let X={x 1,,x n}X = \{x_1, \ldots, x_n\} be a finite metric space. Then D(X)n1n i=1 nD(X{x i})n1. \frac{D(X)}{n} \le \frac{1}{n} \sum_{i=1}^n \frac{D(X \setminus \{x_i\})}{n-1}.

Theorem 3 can be interpreted as saying that the “complexity per element” of XX is at most the average complexity per element of a randomly chosen subset of cardinality n1n-1. This captures the natural intuition that as the size of a metric space increases, its complexity per element decreases.

In the setting of n\mathbb{R}^n, many examples of BT diversities are homogeneous, in the sense that δ(λA)=λδ(A)\delta(\lambda A) = \lambda \delta(A) for all λ0\lambda \ge 0 and nonempty finite A nA \subseteq \mathbb{R}^n, and either sublinear, meaning homogeneous and also satisfying δ(A+B)δ(A)+δ(B), \delta(A + B) \le \delta(A) + \delta(B), or else linear, where we have equality in the condition above. For example, the diameter is a sublinear diversity. (Diversities with these properties are the focus of a recent work by Bryant and Tupper.)

By contrast, maximum diversity has no simple homogeneity property; in fact its complex behavior with respect to scaling is part of what gives it such rich geometric interest. And at least in one dimension, the diversity δ=logD\delta = \log D satisfies the following superlinearity properties.

Theorem 4 Let δ\delta be the diversity δ=logD\delta = \log D defined on compact subsets of \mathbb{R}. Then δ(A+B)δ(A)+δ(B) \delta(A + B) \ge \delta(A) + \delta(B) and δ(λA+(1λ)B)λδ(A)+(1λ)δ(B) \delta(\lambda A + (1-\lambda)B) \ge \lambda \delta(A) + (1-\lambda) \delta(B) for every 0λ10 \le \lambda \le 1 and nonempty compact A,BA,B \subseteq \mathbb{R}.

The first inequality in Theorem 4 can be regarded as a generalization of the Cauchy–Davenport inequality in \mathbb{R}, and the second as a version of the Brunn–Minkowski inequality in \mathbb{R}. (In fact, since Lebesgue measure can be recovered from maximum diversity, it implies the Brunn–Minkowski inequality in \mathbb{R}.) It is an open question, for which we know some partial results, whether Theorem 4 can be extended to higher dimensions.

In conclusion, our results make (at least) the following points:

  • The seemingly independent mathematical notions of diversity introduced by Leinster and Cobbold on the one hand, and Bryant and Tupper on the other hand, are actually closely connected.

  • Maximum diversity, in the sense of LC diversities, leads to a geometrically interesting example of a BT diversity whose behavior is quite different from many of the previously studied examples of BT diversities.

  • Maximum diversity, at least in certain contexts, satisfies a number of inequalities which extend important classical inequalities, and it would be especially interesting to push this line of inquiry further.

Please read the paper itself for more detail and other remarks (it’s short!).

Posted at August 5, 2025 4:34 PM UTC

TrackBack URL for this Entry:   https://golem.ph.utexas.edu/cgi-bin/MT-3.0/dxy-tb.fcgi/3612

3 Comments & 0 Trackbacks

Re: (BT) Diversity from (LC) Diversity

That’s super nice!

In the remarks after Theorem 1 of your post, and also in Theorem 1.7 of your paper, you point out that LC maximum diversity D(X)D(X) gives rise to a BT diversity in two ways. You can either use D(X)1D(X) - 1 or logD(X)\log D(X).

First question: is ϕ(D(X))\phi(D(X)) a BT diversity for any function ϕ\phi satisfying ϕ(1)=0\phi(1) = 0 and ϕ(1)=1\phi'(1) = 1? And maybe we also want ϕ\phi to be increasing.

I’m asking not only because that’s a common generalization, and not only with the idea that xx1x \mapsto x - 1 is the linear approximation to any such function ϕ\phi, but also because of stuff about “Tsallis” vs Rényi entropies. Maybe I’ll elaborate on that if the answer to the first question is yes :-)

Second, vaguer, question: do you have a clear sense of whether we should view D1D - 1 or logD\log D as the primary player here?

Posted by: Tom Leinster on August 5, 2025 5:22 PM | Permalink | Reply to this

Re: (BT) Diversity from (LC) Diversity

Regarding your first question, what we do in the paper is first show that D1D-1 is a BT diversity (using a reformulation of the definition contained in Lemma 2.2 in the paper). Then we apply Lemma 2.1, which states that if δ\delta is a BT diversity, then so is log(δ+1)log (\delta + 1). What the proof of Lemma 2.1 needs is that the function ψ:[0,)[0,)\psi:[0,\infty) \to [0,\infty) given by ψ(x)=log(x+1)\psi(x) = log (x+1) satisfies

  • ψ(0)=0\psi(0) = 0,

  • ψ\psi is (weakly) increasing, and

  • ψ\psi is subadditive (ψ(x+y)ψ(x)+ψ(y))\psi(x+y) \le \psi(x) + \psi(y)).

Those hypotheses could perhaps be weakened, but I think the hypotheses you suggest for ϕ\phi don’t give enough control for large arguments to substitute for the subadditivity property.

As for the second question, I think the answer is very context-dependent.

I don’t think it’s possible to deduce that D1D-1 is a BT diversity from the fact that logDlog D is. So from the point of view of Theorem 1 in the post and Theorem 1.7 of the paper, I’d say D1D-1 is the primary player.

On the other hand, I just spotted a typo in Theorem 4 of the post and the line above it (maybe you could fix this, Tom!): in that result, I should have said δ=logD\delta = \log D. One could argue that the result amounts to a similar result for DD itself with products instead of sums on the right hand side, but given the interest in linearity properties for BT diversities, logDlog D is the version that has a more directly comparable (in the sense of almost opposite!) behavior to more classical examples. So it perhaps fits into the existing world of BT diversities more naturally.

And in the forthcoming work that I alluded to, focusing more on relationships between diversity and entropy, logDlog D will definitely be the primary player.

Posted by: Mark Meckes on August 5, 2025 6:32 PM | Permalink | Reply to this

Re: (BT) Diversity from (LC) Diversity

Thanks for the information on what’s needed of ψ\psi. Evidently the situation is different from the one I had in mind, so it’s probably not worth elaborating on what I was thinking.

I just spotted a typo in Theorem 4 of the post

That should be fixed now.

And in the forthcoming work that I alluded to, focusing more on relationships between diversity and entropy, logD\log D will definitely be the primary player.

Looking forward to it!

For those not following closely, “logD\log D” is a metric-sensitive maximum Rényi entropy.

Posted by: Tom Leinster on August 5, 2025 6:39 PM | Permalink | Reply to this

Post a New Comment