I came across this interesting number in a book I was reading on Mathematics. To quote from pages 97 and 98 of The Story of Numbers by Mallick Asok Kumar:
This, almost certainly, irrational number has been discovered recently by an Indian computer scientist — named Divakar Vishwanath. This number is related to “Randomised Fibonacci Sequence”. Just like the normal Fibonacci sequence (see Section 2.6.1), the randomised version also starts with first two numbers as 1, 1. Thereafter one tosses a fair coin at every stage and use the following formula to generate the sequence:$$F_n=F_{n−2}±F_{n−1} \text{ for }n≥3$$The plus sign is used if the coin shows a head and the minus sign is used if the coin shows a tail. Vishwanath has proved that if you start to generate such avsequence, then, with probability 1, the absolute value of the \(N\)-th number in the randomised sequence will be approximately equal to the \(N\)-th power of 1.13198824 . . . . The bigger the value of \(N\), the closer is the absolute value of the \(N\)-th number of the randomised sequence to the \(N\)-th power of Vishwanath number 1.13198824 . . . . This number is almost certainly an irrational number.
Bear in mind that the recent discovery referred to in the above quote is now about twenty years old. Using SageMathCell, I tested this out for N=1000. Here is the code that I used:
The above box sometimes works but is temperamental. Here is a permalink to the code at SageMathCell. Some results are as follows (note that even though the \(N\)-th term can be very large, the \(N\)-th root of the number always reins it in:
71635048983737736457508495785347136900241837532880818653 1.13724785289546
-3868701952168426630626609949500693480141855962812452339 1.13393344608014
-230368470580308360353351991114643286077160535687 1.11522481150615
-1294461166539244845628908128250544767526170899091979974739663 1.14844999189994
Increasing \(N\) to 100,000 gives the following results:
71635048983737736457508495785347136900241837532880818653 1.13724785289546
-3868701952168426630626609949500693480141855962812452339 1.13393344608014
-230368470580308360353351991114643286077160535687 1.11522481150615
-1294461166539244845628908128250544767526170899091979974739663 1.14844999189994
Increasing \(N\) to 100,000 gives the following results:
- 1.13104709702409
- 1.13203489558527
- 1.12848774540983
- 1.13268434714159
- 1.13181606331044
- 1.13153973067764
- 1.1312382520834
- 1.13181595957100
To give some idea of what this result says, the way the randomized Fibonacci sequence is generated is a bit like the daily weather at a particular location. Today's weather can be assumed to depend on the weather the previous two days, but there is a large element of chance. The analog of the number 1.13198824 . . . for the weather would give a quantitative measure of the unpredictability of weather. It measures the rate at which small disturbances explode exponentially in time. It would tell you for exactly how many days high-speed computers can forecast weather reliably. Unfortunately, nobody knows this number for global weather, and probably never will.
Viswanath's result brings to an end a puzzle that has its origins in 1960. In that year, Hillel Furstenberg (now at the Hebrew University) and Harry Kesten (at Cornell University) showed that for a general class of random-sequence generating processes that includes the random Fibonacci sequence, the absolute value of the \(N\)-th member of the sequence will, with probability 1, get closer to the \(N\)-th power of some fixed number. (The exact formulation of their result is in terms of random matrix products, and is not for the faint-hearted. See Viswanath's paper -- cited below -- for an exact statement, or read the whole story in the book Random Products of Matrices With Applications to Infinite-Dimensional Schrodinger Operators, by P. Bougerol and J. Lacroix, published by Birkhauser, Basel, in 1984.)
Since Furstenberg and Kesten's deep result applied to the randomized Fibonacci process, it followed that, with probability 1, the absolute value of the Nth number in any random Fibonacci sequence will get closer and closer to the Nth power of some fixed number K. But no one knew the value of the number K, or even how to calculate it.
What Viswanath did was find a way to compute \(K\). At least, he computed the first eight decimal places. Almost certainly, \(K\) is irrational, so cannot be computed exactly. Viswanath presented his new result at a colloquium at MSRI last month.
Since there is no known algorithm to compute \(K\), Viswanath had to adopt a circuitous route, showing that \(K\) equals \(e^P\), where \(P\) lies somewhere between 0.1239755980 and 0.1239755995 (and, as usual, \(e\) is the base for natural logarithms). Since those two numbers are equal in their first eight decimal places, that meant he could calculate \(K\) to eight decimal places.
The process involved large doses of mathematics and some heavy duty computing. Since his computation made use of floating point arithmetic -- which is not exact -- Viswanath had to carry out a detailed mathematical analysis to obtain an upper bound on any possible errors in the computation. He describes the key to his new result this way: "The problem was that fractals were coming in the way of an exact analysis. What I did was to guess the fractal and use it to find \(K\). To do this, I made use of some devilishly clever work carried out by Furstenberg in the early 1960s."
There's not been much on the Internet about Vishwanath's Number over the subsequent twenty years. The number did get a brief mention in another blog last year but that's about it. Nonetheless, it's an interesting concept.And with that computation, mathematics has a new constant, a direct descendent of a pair of rabbits in thirteenth century Italy.
