## Archive for September, 2022

### Is mathematics being driven out by computers?

September 28, 2022

In the past two weeks I attended two conferences. The first was the annual meeting of the Deutsche Mathematikervereinigung (DMV, the German mathematical society) in Berlin. The second was the joint annual meeting of the ESMTB (European Society for Mathematical and Theoretical Biology) and the SMB (Society for Mathematical Biology) in Heidelberg. I had the impression that the participation of the SMB was relatively small compared to previous years. (Was this mainly due to the pandemic or due to other problems in international travel?) There were about 500 participants in total who were present in person and about another 100 online. I was disappointed with the plenary talks at both conferences. The only one which I found reasonably good was that of Benoit Perthame. One reason I did not like them was the dominance of topics like machine learning and artificial intelligence. This brings me to the title of this post. I have the impression that mathematics (at least in applied areas) is becoming ever weaker and being replaced by the procedure of developing computer programmes which could be applied (and sometimes are) to the masses of data which our society produces these days. This was very noticeable in these two conferences. I would prefer if we human beings would continue to learn something and not just leave it to the machines. The idea that some day the work of mathematicians might be replaced by computers is an old one. Perhaps it is now happening, but in a different way from that which I would have expected. Computers are replacing humans but not because they are doing everything better. There is no doubt there are some things they can do better but I think there are many things which they cannot. The plenary talks at the DMV conference on topics of this kind were partly critical. There occurred examples of a type I had not encountered before. A computer is presented with a picture of a pig and recognizes it as a pig. Then the picture is changed in a very specific way. The change is quantitatively small and is hardly noticeable to the human eye. The computer identifies the modified picture as an aeroplane. In another similar example the starting picture is easily recognizable as a somewhat irregular seven and is recognized by the computer as such. After modification the computer recognizes it as an eight. This seems to provide a huge potential for mistakes and wonderful opportunities for criminals. I feel that the trend to machine learning and related topics in mathematics is driven by fashion. It reminds me a little of the ‘successes’ of string theory in physics some years ago. Another aspect of the plenary talks at these conferences I did not like was that the speakers seemed to be showing off with how much they had done instead of presenting something simple and fascinating. At the conference in Heidelberg there were three talks by young prizewinners which were shorter than the plenaries. I found that they were on average of better quality and I know that I was not the only one who was of that opinion.

In the end there were not many talks at these conferences I liked much but let me now mention some that I did. Amber Smith gave a talk on the behaviour of the immune system in situations where bacterial infections of the lung arise during influenza. In that talk I really enjoyed how connections were made all the way from simple mathematical models to insights for clinical practise. This is mathematical biology of the kind I love. In a similar vein Stanca Ciupe gave a talk about aspects of COVID-19 beyond those which are common knowledge. In particular she discussed experiments on hamsters which can be used to study the infectiousness of droplets in the air. A talk of Harsh Chhajer gave me a new perspective on the intracellular machinery for virus production used by hepatitis C, which is of relevance to my research. I saw this as something which is special for HCV and what I learned is that it is a feature of many positive strand RNA viruses. I obtained another useful insight on in-host models for virus dynamics from a talk of James Watmough.

Returning to the issue of mathematics and computers another aspect I want to mention is arXiv. For many years I have put copies of all my papers in preprint form on that online archive and I have monitored the parts of it which are relevant for my research interests for papers by other people. When I was working on gravitational physics it was gr-qc and since I have been working on mathematical biology it has been q-bio (quantitative biology) which I saw as the natural place for papers in that area. q-bio stands for ‘quantitative biology’ and I interpreted the word ‘quantitative’ as relating to mathematics. Now the nature of the papers on that archive has changed and it is also dominated by topics strongly related to computers such as machine learning. I no longer feel at home there. (To be fair I should say there are still quite a lot of papers there which are on stochastic topics which are mathematics in the classical sense, just in a part of mathematics which is not my speciality.) In the past I often cross-listed my papers to dynamical systems and maybe I should exchange the roles of these two in future – post to dynamical systems and cross-list to q-bio. If I succeed in moving further towards biology in my research, which I would like to I might consider sending things to bioRxiv instead of arXiv.

In this post I have written a lot which is negative. I feel the danger of falling into the role of a ‘grumpy old man’. Nevertheless I think it is good that I have done so. Talking openly about what you are unsatisfied with is a good starting point for going out and starting in new positive directions.

In my research I have often used centre manifolds but I have not thoroughly studied the proof of their existence. The standard reference I have quoted for this topic is the book of Carr. The basic statement is the following. Let $\dot x=f(x)$ be a dynamical system on $R^n$ and $x_0$ a point with $f(x_0)=0$. Let $A=Df(x_0)$. Then $R^n$ can be written as a direct sum of invariant subspaces of $A$, $V_-\oplus V_c\oplus V_+$, such that the real parts of the eigenvalues of the restrictions of $A$ to these subspaces are negative, zero and positive, respectively. $V_c$ is the centre subspace. The centre manifold theorem states that there exists an invariant manifold of the system passing through $x_0$ whose tangent space at $x_0$ is equal to $V_c$. This manifold, which is in general not unique, is called a centre manifold for the system at $x_0$. Theorem 1 on p. 16 of the book of Carr is a statement of this type. I want to make two comments on this theorem. The first is that Carr states and proves the theorem only in the case that the subspace $V_+$ is trivial although he states vaguely that this restriction is not necessary. The other concerns the issue of regularity. Carr assumes that the system is $C^2$ and states that the centre manifold obtained is also $C^2$. In the book of Perko on dynamical systems the result is stated in the general case with the regularity $C^r$ for any $r\ge 1$. No proof is given there. Perko cites a book of Guckenheimer and Holmes and one of Ruelle for this but as far as I can see neither of them contains a proof of this statement. Looking through the literature the situation of what order of differentiability is required to get a result and whether the regularity which comes out is equal to that which goes in or whether it is a bit less seems quite chaotic. Having been frustrated by this situation a trip to the library finally allowed me to find what I now see as the best source. This is a book called ‘Normal forms and bifurcations of planar vector fields’ by Chow, Li and Wang. Despite the title it offers an extensive treatment of the existence theory in any (finite) dimension and proves, among other things, the result stated by Perko. I feel grateful to those authors for their effort.
A general approach to proving the existence of a local centre manifold, which is what I am interested in here, is first to do a cut-off of the system and prove the existence of a global centre manifold for the cut-off system. It is unique and can be obtained as a fixed point of a contraction mapping. A suitable restriction of it is then the desired local centre manifold for the original system. Due to the arbitrariness involved in the cut-off the uniqueness gets lost in this process. A mapping whose fixed points correspond to (global) centre manifolds is described by Carr and is defined as follows. We look for the centre manifold as the graph of a function $y=h(x)$. The cut-off is done only in the $x$ variables. If a suitable function $h$ is chosen then setting $y=h(x)$ gives a system of ODE for $x$ which we can solve with a prescribed initial value $x_0$ at $t=0$. Substituting the solution into the nonlinearity in the evolution equation for $y$ defines a function of time. If this function were given we could solve the equation for $y$ by variation of constants. A special solution is singled out by requiring that it vanishes sufficiently fast as $t\to -\infty$. This leads to an integral equation of the general form $y=I(h)$. If $y=h$, i.e. $h$ is a fixed point of the integral operator then the graph of $h$ is a centre manifold. It is shown that when certain parameters in the problem are chosen correctly (small enough) this mapping is a contraction in a suitable space of Lipschitz functions. Proving higher regularity of the manifold defined by the fixed point requires more work and this is not presented by Carr. As far as I can see the arguments he does present in the existence proof nowhere use that the system is $C^2$ and it would be enough to assume $C^1$ for them to work. It is only necessary to replace $O(|z|^2)$ by $o(|z|)$ in some places.