Optimal-Growth Liquidity Providers

• Optimal-growth liquidity providers are agents in AMMs and LOBs that maximize long-run logarithmic wealth using Kelly-based strategies, balancing asset volatility and risk.
• They employ mathematical models, such as geometric mean market makers and convex optimization, to determine optimal fee schedules, allocation ranges, and liquidity adjustments.
• Empirical tests on platforms like Uniswap v3 demonstrate that these strategies can enhance fee revenues by up to 25% while mitigating adverse selection and market impact.

Optimal-growth liquidity providers are agents in automated market makers (AMMs) or limit order book (LOB) protocols who structure their quoting, allocation, and fee/range selection to maximize the long-run (Kelly-style) logarithmic growth rate of their on-chain wealth. This paradigm integrates adversarial market microstructure dynamics, price volatility, arbitrage, fee schedules, and strategic range selection, producing protocols and algorithmic recipes that outperform naive liquidity provision in both theoretical models and empirical deployments.

1. Mathematical Foundations of Growth-Optimal Provision

The foundational objective for optimal-growth LPs is maximization of log-wealth under stochastic asset price dynamics, subject to inventory, risk aversion, capital charge, and rebalancing costs. For a geometric mean market maker (G3M)—generalizing Uniswap v2 and Balancer—the pool invariant is $V(x) = \prod_{i=1}^n x_i^{\theta_i}$ for reserves $x(t)=(x_1,\ldots,x_n)$ and weight vector $\theta=(\theta_1,\ldots,\theta_n)$, $\sum_i \theta_i=1$ (Lee et al., 2024).

Mark-to-market LP wealth in reference units is $W(t) = \sum_{i=1}^n S^i(t)x_i(t)$, with external prices modeled by correlated Brownian motions: $d\ln S^i(t) = \mu^i dt + \sigma^i dB^i(t)$ Under Kelly log-utility, the long-run growth rate is: $$g = \lim_{T\to\infty} \frac{1}{T}\mathbb{E}\left[\ln W(T) - \ln W(0)\right]$$ This rate decomposes into linear drift ($\sum_i \theta_i \mu^i$) and a nonlinear excess-growth term controlled by fee rate $f$, weights $\theta$, fee parameter $\gamma=f$, and volatility $\sigma$. The optimal $(f^*,\theta^*)$ is found by maximizing $g$ subject to weight constraints, leveraging closed-form or numerical ergodic averages over the no-arbitrage corridor $Z(t)\in[\ln \gamma, -\ln \gamma]$ (Lee et al., 2024).

Growth-optimal impact laws generalize this principle to quoted liquidity in AMMs and LOBs. Under random walk price dynamics, the optimal impact curve for absorbing trade size $Q$ is the square-root law: $\Delta P^* \propto \sigma^2 \sqrt{Q} / k$ where $k$ is the LP's capital buffer, derived from balancing expected edge and variance extraction per Kelly’s criterion (Meister, 16 Jan 2026).

2. Pool Structure, Fee Optimization, and Concentration

Optimal-growth LP strategies structurally depend on both the functional form of liquidity curves and dynamic fee scheduling. In G3Ms, fee tier $f$ tightens or widens the arbitrage corridor, directly shaping the tradeoff between accumulated fees and impermanent loss. Maximizing $g$ with respect to $f$ yields a unique interior maximizer, with nonzero drift or asymmetric weights leading to phase transitions and possibly skewed optimums (Lee et al., 2024).

In AMMs operating parallel to CEXs, optimal fee selection is characterized by a threshold-type schedule. For moderate volatility: $\eta^{1,*}(\sigma,V) \in (0, \eta^0)$ with

$$\frac{\partial \eta^{1,*}}{\partial \sigma} > 0, \quad \frac{\partial \eta^{1,*}}{\partial V} < 0$$

Optimal base fees typically undercut CEX execution cost by $\sim$30\% in normal markets; in high volatility, "protection" fees can spike or liquidity can be withdrawn entirely to mitigate adverse-selection losses (Campbell et al., 11 Aug 2025).

Liquidity concentration intervals are determined via self-financing stochastic programs that trade off fee income, predictable loss (PL), and concentration risk. Closed-form solutions for optimal half-width $\delta^*$ and skew are available for log-utility LPs: $\delta^* = \frac{4\gamma}{8\pi - \sigma^2}$ where $\pi$ is fee rate, $\gamma$ penalizes tight ranges, and $\sigma$ is volatility. Asymmetric drift $\mu_t\ne0$ produces optimal skew $\rho^* = 1/2 + \mu_t/\delta^*_t$ (Cartea et al., 2023).

3. Strategic Range, Fragmentation, and Dynamic Allocation

Optimal-growth protocols exploit dynamic range adjustment, pool fragmentation, and agent heterogeneity. In concentrated liquidity AMMs (e.g., Uniswap v3), strategic LPs calibrate their range intervals either statically or through τ-reset policies. A τ-reset strategy dynamically reallocates all liquidity into the price band of width $2\tau+1$ ticks centered at the market price, resetting exactly when price leaves this interval. ML models can be trained to select optimal allocations from market features, ensuring predictive fee uplift of 15–25% over uniform baselines (Urusov et al., 21 May 2025).

Pool fragmentation yields endogenous LP sorting equilibria: small LPs prefer high-fee pools, minimizing adverse selection and gas costs via wider ranges and lower execution frequencies; large LPs concentrate on low-fee pools with tight ranges and active rebalancing. The equilibrium split and participation thresholds are given by explicit formulas dependent on fee differential, gas, and market informativeness. Empirical calibration shows large LPs (top 25%) dominate low-fee pools, small LPs are passive in high-fee venues (Lehar et al., 2023).

4. Solution Algorithms and Empirical Validation

Growth-maximizing LP strategies are solved via convex optimization, infinite-dimensional KKT/Lagrangian conditions, or numerically (sample-average-approximation, branch-and-bound, ML regression, etc.). For general beliefs about future prices, (Goyal et al., 2022) provides a convex program minimizing expected trade failure subject to liquidity, capital, and reserve constraints. The optimal trading curve $\phi$ is reconstructed from liquidity density $L(p)$ uniquely determined by the KKT multipliers.

Validation across real Uniswap v3 pools (ETH/USDC, WBTC/ETH, USDC/USDT) confirms the outperformance of optimally concentrated and dynamically allocated LPs over historical or uniform strategies (Cartea et al., 2023, Urusov et al., 21 May 2025). ML-augmented τ-reset plus Gaussian liquidity calibration achieves fee uplifts ranging from 7–23% depending on pool and tick granularity (Urusov et al., 21 May 2025).

JIT LPs in concentrated AMMs solve a per-trade nonlinear maximization problem (utility = fees – price impact – inclusion cost) over liquidity size and tick-range, with explicit 1D-plus-enumeration search algorithms provably attaining the optimal solution. Empirical study shows that typical JIT LPs leave up to 69% of profits on the table by ignoring price impact, and that full optimization would erode passive LP fee share by up to 44% per trade (while improving trader slippage) (Trotti et al., 19 Sep 2025).

5. Market Impact, Adverse Selection, and Microstructure

Optimal-growth LP protocols produce market impact profiles sharply divergent from classical linearized CFMM models. The canonical optimal-impact is the square-root law ($\beta=1/2$); generalized fractional OU processes yield $P_{\rm post}-P_{\rm pre} \propto Q^{2H-1/2}$, with sublinear/convexity depending on the Hurst exponent $H$ (Meister, 16 Jan 2026). Linear CFMMs always undercompensate LPs for adverse selection and order toxicity; in contrast, dynamically adjusting the curve exponent and scale restores Kelly-optimal risk utility and fee extraction.

Market microstructure models (Glosten–Milgrom) with adaptive bonding curves for AMMs guarantee zero expected-loss versus the "true" external price. These are implemented via real-time Kalman filtering of trade histories, with the curve parameter $\theta$ calibrated to current volatility and noise. Adversarial performance is robust and Bayesian-efficient, with on-chain/off-chain architecture (Uniswap v4 hooks plus Axiom coprocessors) (Nadkarni et al., 2024).

6. Limit Order Book Analogs and Classical Results

Outside DeFi, the small-spread, high-frequency optimal LP policy in limit order books is the reflect-at-boundaries strategy: always post at best bid/ask, choosing sizes to keep inventory between

$$\left[ -\frac{2\varepsilon_t \alpha_t^{(1)}}{\mathrm{ARA}\,\sigma_t^2},\, \frac{2\varepsilon_t \alpha_t^{(2)}}{\mathrm{ARA}\,\sigma_t^2} \right]$$

for spread $\varepsilon_t$, intensities $\alpha_t^{(i)}$, volatility $\sigma_t$, and absolute risk aversion $\mathrm{ARA}$ (Kühn et al., 2013). The leading-order safe rate grows as $\varepsilon^2 \alpha^2 / \mathrm{ARA}\,\sigma^2$ for the symmetric case.

7. Practical Considerations, Assumptions, and Limitations

Growth-optimal LP analysis assumes frictionless reference markets, instantaneous arbitrage, continuous-time or high-frequency tick setting, and ergodic log-price dynamics. Real-world performance is affected by gas costs, pool depth, discrete-trade price impact, MEV-sandwich effects, latency, and deviations from lognormality or Brownian motion. The theoretical frameworks provide first-order guidance: match pool weights to relative drift-to-volatility ($\mu/\sigma^2$), set fee schedule adaptively, concentrate where anticipated price ranges maximize expected log-growth, and leverage ML or convex programs for live recalibration (Lee et al., 2024, Nadkarni et al., 2024, Campbell et al., 11 Aug 2025, Urusov et al., 21 May 2025).

—

In summary, optimal-growth liquidity provision synthesizes advanced stochastic models, microstructure theory, convex optimization, adaptive algorithmics, and empirically-grounded range and fee selection to maximize long-run log-wealth for LPs. The literature establishes explicit recipes, validates their performance, and supplies dynamic adjustment and risk-mitigation mechanisms suitable for the full spectrum of DeFi and classical markets.