The paper "On the Instability of Bitcoin Without the Block Reward" by Carsten et al, suggests that as the Bitcoin block subsidy decreases relative to the transaction fees, there is a possibility that Bitcoin will experience "Mining gaps". This occurs when a substantial fraction of miners turn off their mining equipment for a short period immediately after a block is mined (when the mempool is smaller and when there is accordingly less of a reward for mining blocks), in order to save on electricity costs.
This question resolves to "YES" if the Bitcoin blockchain shows clear evidence of a at least 10% of the network hashrate using a Mining gap strategy before the first 365-day period where transaction fees make up at least 50% of total block rewards. Note that the argument for the presence of mining must consist entirely of analysis of chain data. An example of "clear evidence" might be a statistical analysis of inter-block times, showing that these times are not exponentially distributed (as we would expect if hashrate was constant) or showing a difference in inter-block times depending on the state of the mempool as extrapolated from fees in previous blocks.
Dec 20, 6:29pm: Will Bitcoin experience "mining gaps" before the first 365-day period where transaction fees have >1/2 of total block rewards? → Will Bitcoin experience "mining gaps" before the first 365-day period where transaction fees are >1/2 of total block rewards?
10% of miners dropping out when the queue is empty and the block reward is <0.25btc is a plausible number, but it would have to be 80% to even become a nuisance. It's extremely unlikely that the cost of electricity is homogenous enough to cause mining gaps to be a serious problem. Staying always on is easiest because it's the default, and defaults have a lot of staying power. Sometimes they have a long term deal with the power company so they're contractually obligated to stay on. Sometimes they're generating their own solar power, or using natural gas that would otherwise be flared, so staying on is essentially free.
Also the demand for bitcoin transactions is pretty elastic so mempool zero is a very rare occurrence. It's much more likely that the expected value of transaction fees varies smoothly as miners work through a backlog of low fee transactions hundreds of times larger than the block size.
Bitcoin difficulty adjustment is every 2016 blocks = 2 weeks and there are currently 650 blocks worth of transactions in the mempool = 4.5 days. If necessary they can soft fork to increase the mempool size relative to the block size to stabilize things.
@JonathanRay I have been thinking more about this.
If you look in the appendix A of the paper, you find their derivation of the ratio of time that miners spend mining as opposed to not when block rewards are zero and everyone mines only when it is profitable. Their algebra basically shows that this is equal to the ratio between the rate at which fees accrue to the cost rate of running the whole network, which makes sense, because it's essentially a zero-profit theorem.
USD-denominated Bitcoin fees have historically fluctuated by factors of 10 or more on the time scale of just a few months. If we expect that ASICs don't get produced or break down on that timescale, it seems like we could indeed get usage rates of less than 10% at times, no?