This assertion (whether thermodynamic entropy and information theoretic entropy are equivalent) is an ongoing controversy open to debate - it shouldn't be stated as fact. See wikipedia for (a summary of) both sides of the story: https://en.wikipedia.org/wiki/Entropy_in_thermodynamics_and_... (specifically the #Theoretical_relationship and #Criticism sections)
As an analogy, if I make a huge number of random files on my computer and execute them, the probability of getting a self replicating computer virus is really low. And one that successfully spreads itself to other computers over the internet should never happen.
If the first successful computer virus is 100 bits long, then probability of finding it in a single random search is 1/2^100. Which is just incredibly small and should never ever happen.
There could be 10 different ways to make a virus N bits long. There could be 1000 ways to do it in N+1 bits, and a billion ways in N+2.
The relationship between how many possible successful viruses exist to number of bits doesn't have to scale by any obvious pattern.
There could be a magic threshold where if you're shorter than that you don't work.
For ex, maybe a valid file has to be at least 100 bytes, or the OS can't treat it as a file. Any random file 100bytes or longer could have a 10% chance of being a virus (the actual number depends on how the os works).
This would mean a really high chance of randomness creating a virus, even though N is 100.
The probability of vitality depends on the specifics of the operating system, and we can't infer it by looking at the size of the smallest possible virus.
There is a reasonable theory that I remember discussing/reading once that proteins may have started out as long sequences and were then selected for shorter more stable versions. Presumably the early proteins were partly disordered, and only optimised for having stable cores and domain structures later on.
In any case, it is not as simple as the number of bits in a program, as macromolecules are not 'just' software, but actual physical devices that have to (self)-assemble and carry out functions at various temperatures, pressures, pH, etc.
Saying that "the odds of abiogenesis are ~1/2^n" isn't an assertion of fact, it's merely defining n = -log2(probability of abiogenesis). (Of course, the probability of abiogenesis occurring in any particular body of water is not a constant, it depends on conditions; but all the oceans on Earth are only 80 binary orders of magnitude bigger than a single mL [1], so n for a lab experiment and n for the entire Earth can't be more than a couple decimal orders of magnitude different, so it's immaterial.)
In other words, your argument about convection rolls is an excellent argument that the n for abiogenesis is much, much lower than -log2(probability of organic molecules in solution spontaneously assembling themselves into bacteria), which grandparent claimed is ~1 million. I have no reason to believe ~1 million is accurate, but I don't care, because grandparent has convinced me that the upper bound is many orders of magnitude greater than 73. Your argument that n could be much less than that upper bound is convincing, but unless you're arguing it's so much lower that it isn't much greater than 73, grandparent's counter to the paper still holds.
[1]: http://www.wolframalpha.com/input/?i=log2((volume+of+ocean+i...)
Of course it can be easy to over-do the analogies between things like convection rolls and the emergence of lipid bilayers (or whatever). However, the origin of life necessarily came about through purely physical processes and only once that bootstrap phase was over did evolution come into play as biological systems took over acting upon themselves.
As I understand it the idea is that a closed system that is driven far from equilibrium by a constant influx of energy can undergo 'symmetry breaking' where the symmetries of the system are changed. So the symmetric situation of the boiling pot is broken such that convection rolls are formed - which is new structure.
Similarly, a uniform soup of primitive chemical compounds could have been broken into less-uniform patches by some energy gradient (black or white smokers, for example). This diversity of chemical conditions then helped to form the initial replicators that could then further exploit various temperature, chemical, or proton gradients to create primitive life.
Statistical mechanical entropy and information theoretical entropy ('complexity in bits') are not the same quantities! But even if they were, your expression would still be incorrect because living system are definitionally not at equilibrium - so equilibrium statistical mechanics (whose predictions you are using to get that expression) does not apply.
No one is surprised when convection rolls spontaneously form as they boil a pot of water. But equilibrium statistical mechanics predicts that this should be extremely unlikely as convection roll states have very low entropy. That is, of all of the ways that the molecules in that pot of water could be oriented, vanishingly few of them correspond to convection roll states, where several Avogrado's numbers worth of particles conspire to all swirl around in a coordinated fashion to dissipate the heat of the burner faster than would occur just by diffusion.
The origin of life is very much analogous to the formation of convection rolls in the pot. And I believe is far more common than classical predictions imagine it to be - nearly everywhere there is an energy gradient in the universe, you will probably find life.
Citation: I have a degree in physics