12 Comments
User's avatar
Kevin O'Bryant's avatar

One way forward is to replace "tweetable" with something we can define, like: strings of length at most 280 on the alphabet {0,1,2,3,4,5,6,7,8,9,+,-,*,/} that are well-formed expressions with the usual meanings attached to the symbols. I think I could work out the largest natural number that can be so expressed, but finding the smallest number that cannot be so expressed feels rather difficult.

Kevin O'Bryant's avatar

Wait, both questions are pretty easy! Better throw "^" into the alphabet!

Kevin O'Bryant's avatar

The language with ^ can definitely represent everything up to 10^280+10^273+10^266+ ... = 10000001000000100000010000001000000100000010000001000000100000010000001000000100000010000001000000100000010000001000000100000010000001000000100000010000001000000100000010000001000000100000100000100000100000100000100000100000100000100000100000100000100000100000100000100000100000999. But, I'd be surprised if that's the end of it.

Can you get f(10), where f(x) = x^3+\frac{x^{98}-x^{287}}{1-x^7}+\frac{x^8-x^{98}}{1-x^6}?

Joel David Hamkins's avatar

What is the smallest number not expressible in your first language?

Kevin O'Bryant's avatar

10^{280}. We can get every smaller natural by just listing the digits. Now consider an expression in the alphabet {0,...,9,+,*} (with division and subtraction removed) that isn't just a number. The expression's tree has a top level operation + or *, and both are monotone in both of their arguments. So consider a bottom-level operation a+b or a*b. Define a new operation anb which is just a string of 9's with the same length as "a+b". Since a+b<anb and a*b < anb, we can simplify the tree by replacing some bottom-level operation with n, while increasing the number represented.

The operations / and - are annoying because they aren't monotone and might leave the natural numbers. But, they can't possibly lead to getting something bigger --- a fact that deserves a proof that I will leave to the reader.

Kevin O'Bryant's avatar

Let f(n) be the smallest natural number that can NOT be represented by a string of length n on the alphabet {"0","1",...,"9", "*", "+", "^"}. Note that 0^0 = 1. We have f(1)=10, f(2)=100, f(3)=1000, f(4)=10001, f(5)=100001. Not in the OEIS. Computing f(6) naively doesn't work, as some of the representable numbers are too large. Seems a bit light for a MathOverflow question...By hand, I get f(6) = (10^6+9)+1. Conceivably, f(n)-1 is a sum of different powers of 10, but I fear some cosmic coincidence that lands some other perfect high power slightly less than f(n), allowing one to go a bit further.

Joel David Hamkins's avatar

Great! I'd encourage a MathOverflow question. Can you give insightful asymptotic bounds? also: f(0)=0. Also, it seems likely that you can add subtraction, division without changing the answer.

Valentin's avatar

Your conjectured pattern breaks holds up to n=13, but I am quite sure that f(14)=100000100001025 .

Note that f(14)-1 = 10000005^2+999.

All 15 digit numbers below 10000005^2 can be represented with 14 symbols using expressions involving a plain decimal number, 10^14, and up to two additional terms from the set {10^4, 10^5, 10^6, 10^7, 10^8, 22^3, 26^3, 6^5, 7^5, 3^9}.

This is off topic, I am happy to share more results via DM or if you open a question on math overflow.

Benson Margulies's avatar

It strikes me that there is some point of contact between this paradox and the finitists and/or some intuitionists and/or Reeb. On the one hand, a googolplex - arguably we can all hold the concept of that - the program that emits it - in our heads. So perhaps everyone in the tent agrees that it 'exists'. On the other hand, some arbitrary number 'a bit smaller' for which we lack any concise description: we can't exhibit a program that produces it, and we arguably can't be said to hold a concept of it. So perhaps it doesn't 'exist'?

Joel David Hamkins's avatar

I think that is just how the ultrafinitists would view it.