Eq. (2.6): the optimal hash rate is the unique maximizer of the Hamiltonian
ProvedCryptoMiningMFG.eq_2_6_maximizercryptocurrencymean-field-gamesp2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1point-processesstochastic-control
Let and let , stand for and . Consider the Hamiltonian of the HJB equation (2.5),
and the candidate
Then , and for every with : the formula (2.6) is the unique optimal hash rate.
This is the pointwise optimization behind the feedback (2.6) and, through it, behind the explicit equilibrium.
Formalization Note At the term is , the branch of the intensity (2.2).
Differences from the source:
- generalization: (2.6) is stated with ∂ₓv(t, x; ᾱ) and Δv(t, x; ᾱ) of the value function, both positive by Lemma 2.1 → the maximizer property is stated for arbitrary p > 0, q > 0 in place of ∂ₓv and Δv; reason: the page's case is the instance p = ∂ₓv, q = Δv
- reading: "the optimal hash rate is given by" (2.6) → (2.6) is nonnegative and strictly beats every other α ≥ 0 in the Hamiltonian of (2.5); reason: "optimal" means the maximizer of the supremum in (2.5); uniqueness is part of "the" optimal rate
Preamble
import Mathlib
Formal statement
namespace CryptoMiningMFG
/-- **The optimal hash rate (2.6) maximizes the Hamiltonian of (2.5).** Li, Reppen, Sircar,
*A Mean Field Games Model for Cryptocurrency Mining*, arXiv:1912.01952v2, eq. (2.6), p.11.
Let `𝒦, c, M, ᾱ > 0` and let `p > 0`, `q > 0` play the roles of `∂ₓv` and `Δv`. The function
`H(α) = −c α p + 𝒦 α / (α + M ᾱ) · q` on `α ≥ 0` (the expression inside the supremum of the HJB
equation (2.5), p.10) has the unique maximizer
`α* = −M ᾱ + √(𝒦 M ᾱ q / (c p))` if `ᾱ < 𝒦 q / (M c p)`, and `α* = 0` otherwise:
`α* ≥ 0`, and every other `α ≥ 0` gives a strictly smaller value.
**Formalization Note.** At `α = 0` the intensity `𝒦 α / (α + M ᾱ)` is `0`, which is the `α = 0`
branch of (2.2). -/
theorem eq_2_6_maximizer (K c M abar p q : ℝ) (hK : 0 < K) (hc : 0 < c) (hM : 0 < M)
(habar : 0 < abar) (hp : 0 < p) (hq : 0 < q) :
let H : ℝ → ℝ := fun α => -c * α * p + K * α / (α + M * abar) * q
let αstar : ℝ := if abar < K * q / (M * c * p) then
-M * abar + Real.sqrt (K * M * abar * q / (c * p)) else 0
0 ≤ αstar ∧ ∀ α : ℝ, 0 ≤ α → α ≠ αstar → H α < H αstar := by sorry
end CryptoMiningMFG
Source
Li, Reppen, Sircar, A Mean Field Games Model for Cryptocurrency Mining, arXiv:1912.01952v2 (preprint, 22 Jan 2022), https://arxiv.org/abs/1912.01952v2, eq. (2.6) p.11 (with the Hamiltonian of (2.5), p.10)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.