Partially Active Automated Market Makers
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.
Ruth: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We're looking at the paper: "Partially Active Automated Market Makers".
Warren: Here's the gist of the paper. We introduce a new class of automated market maker (AMM), the partially active automated market maker (PA-AMM), which divides its reserves into active and passive parts to reduce adverse selection costs while balancing tracking error.
Ruth: The PA-AMM mechanism partitions the pool’s total reserves into two distinct components, active reserves (Ractive) and passive reserves (Rpassive), and uses only the active portion for trading.
Warren: This division is done exactly once at the beginning of each block, ensuring that only a fraction λ ∈ (0, 1] of the total liquidity is exposed to arbitrageurs at the top of each block.
Ruth: The special case where λ = 1 recovers the standard fully-active CFMM with invariant function φ.
Paper summary: Iris: To wrap up, the paper introduces the partially active automated market maker or PA-AMM, which divides reserves into active and passive parts to trade only with the active part at every block.
Ruth: The main point is that this mechanism reduces loss versus rebalancing measured by LVR and thereby improves the wealth of liquidity providers relative to plain constant function market makers.
Warren: But they also found that this comes at a cost, specifically an increase in asset weights within the pool which might deviate from their target allocation.
Iris: The optimization problem they solve finds the optimal activeness level lambda star that balances that tracking error against the reduction of LVR.
Ruth: In simple terms, PA-AMM limits how much liquidity is active in a single block by gradually executing rebalancing, which cuts down on losses at the expense of some tracking error.
Warren: The paper suggests that PA-AMM might be a better option for liquidity provision in prediction markets where the costs from adverse selection are significantly higher than in other markets.
Iris: They also note that the work abstracts away gas costs and sets fees to zero, but they point out that future work should look at how to find an optimal combination of fee rates and activeness levels.
Ruth: The paper concludes by suggesting that PA-AMM might incentivize liquidity providers to provide more liquidity because the loss function is continuous with respect to latency.
Conclusion: Ruth: So we’ve been looking at this paper on Partially Active Automated Market Makers, and now we need to talk about what this actually means for liquidity providers and markets as a whole.
Warren: Yeah, I mean the title itself—Partially Active AMM—it sounds like they're introducing a way to manage risk by not being fully active all the time.
Iris: Exactly. The authors are focusing on how you can split your reserves into active and passive parts to reduce those nasty costs from adverse selection.
Ruth: Adverse selection costs, right? That’s when someone knows you’re about to trade and they profit off that information advantage before you can react.
Warren: So the mechanism they propose here, dividing liquidity into Ractive and Rpassive, is basically a way to cut down on those immediate losses at the top of each block.
Iris: They model this as an optimization problem balancing two things: tracking error and this loss versus rebalancing measure we call LVR.
Ruth: It’s a trade-off, isn't it? You reduce the risk of getting hit by bad trades, but you might end up with a bigger gap between your pool price and the actual market price.
Warren: Right. The math they use suggests there's an optimal level for this activeness, called lambda star, that balances that tracking error against minimizing LVR.
Iris: In the limit of very small time steps for block execution, this optimal activeness level settles down to a specific constant value based on some parameters from the model.
Ruth: That’s interesting because it suggests there isn't just one perfect setting for everyone; it depends on how much you care about minimizing loss versus how much you care about keeping your prices tight.
Warren: It definitely implies that for certain markets, like prediction markets where those adverse selection costs are huge, this PA-AMM approach could be a better way to set up liquidity provision.
Iris: It does. And what it means for the world is that we might need to think about dynamic ways of managing liquidity exposure in decentralized finance environments.
KAIST
q-fin.MF
Submitted: 2026-02-10
Updated: 2026-11-03
Comments: 32 pages; minor metadata correction
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 99/100
The gist: We introduce a new class of automated market maker (AMM), the partially active automated market maker (PA-AMM)…
Key concepts
- Active Reserves (Ractive)
- These are the portion of the pool's total reserves that are actually used for trading. In PA-AMM, only this active fraction is exposed to arbitrageurs at the top of each block. This division happens once per block.
- Passive Reserves (Rpassive)
- This is the reserve portion that remains unused for immediate trading purposes. By keeping a passive part, the mechanism reduces exposure to per-block adverse selection, which is when bad trades occur at the end of a trading period.
- Loss-versus-Rebalancing (LVR)
- LVR measures how much value is lost due to an arbitrage opportunity. It compares the total asset value before and after an arbitrage event. The goal of PA-AMM is to minimize this loss while controlling tracking error.
- Activeness Level ($\lambda$)
- This parameter determines what fraction of immediate rebalancing occurs at the top of each block. A value of $\lambda=1$ is a standard fully-active model, while $\lambda < 1$ means only a fraction of rebalancing happens, trading off loss reduction against tracking error.
Terminology
Summary
We introduce a new class of automated market maker (AMM), the partially active automated market maker (PA-AMM), which divides its reserves into active and passive parts to reduce adverse selection costs while balancing tracking error. p. 0 Partially Active Automated Market Makers
Ko Sunghun‗
February 11, 2026
Abstract
We introduce a new class of automated market maker (AMM), the partially active automated
market maker (PA-AMM). PA-AMM divides its reserves into two parts, the active and the passive
parts, and uses only the active part for trading.Where the paper says it
Introduction to PA-AMM
The PA-AMM mechanism partitions the pool’s total reserves into two distinct components, active reserves (Ractive) and passive reserves (Rpassive), and uses only the active portion for trading. This division is done exactly once at the beginning of each block, ensuring that only a fraction λ ∈ (0, 1] of the total liquidity is exposed to arbitrageurs at the top of each block. p. 4 The underlying invariant function φ enforces trading
constraints solely on the active portion, thereby limiting the liquidity available for immediate
arbitrage within any single block. The mechanism ensures that the active-passive division is done
exactly once at the beginning of each block, triggered by the first interaction (swap, addition, or
removal of liquidity) of that block. This partition is refreshed at the top of each
block, so that a λ ∈ (0, 1] fraction of the total liquidity remains active at all times. The special case
of λ = 1 recovers the standard fully-active CFMM with invariant function φ. The special case
of λ = 1 recovers the standard fully-active CFMM with invariant function φ. When λ < 1, the pool executes only a λ-fraction of the immediate rebalancing that would occur
under a CFMM, thereby reducing the exposure to per-block adverse-selection.Where the paper says it
Where the paper says it
Where the paper says it
Modeling Loss and Dynamics
Loss-versus-rebalancing (LVR) is measured by the difference in the total value of assets within the pool before and after an arbitrage, defined as V (P, S) − V (S). The instantaneous LVR is defined by LVR(P):= −σ2P2/2V''(P) ≥ 0. For a Geometric Mean Market Maker (G3M), the instantaneous rate of LVR becomes LVR(P) = σ2/2θ(1 − θ)V (P). p. 4 For instance, when the invariant curve is given by weighted geometric mean, i. , the
CFMM is geometric mean market maker (G3M) with weights (θ, 1 − θ), the instantaneous rate of
LVR becomes
LVR(P) = σ
2
2
θ(1 − θ)V (P). The following proposition tells us that there exists
a unique stationary distribution π∆t of the gap process {gn}n∈N with the estimation on its second
moment up to leading order. 5
In the sense that the trader should divide the order into many sub-orders and gradually submit them over multiple
blocks for optimal execution, one may consider PA-AMM as a form of implementation of a fully continuous exchange
proposed in [KL17], given that the block arrival interval ∆t is short enough.Where the paper says it
Where the paper says it
Optimization of Activeness
The paper formalizes the trade-off between tracking error and LVR reduction as a discounted infinite-horizon stochastic control problem. The objective functional is defined as J (λ) ≈ E[e−ϱn∆t ((1 − λ)2 + γλ)g2]. p. 9 To leading order, we have
TEn ≈ θ
2
(1 − θ)
2
(1 − λn)
2
g
2
n
, LVR ]n ≈
λn
2
θ(1 − θ)g
2
n
,
and for the portfolio manager who weights the LVR cost γ
′
times more than the tracking error, the
one-stage cost becomes
TEn + γ
′LVR ]n ∝ ((1 − λn)
2 + γλn)g
2
n
,
where γ :=
γ
′
2θ(1−θ)
is the constant that absorbs the fixed factors for notational convenience. We
define the objective functional as a weighted sum of one-stage costs discounted at a rate ϱ > 0,
J ({λn}) := E
"X∞
n=1
e
−ϱn∆t
TEn + γ
′ LVR ]n
#
. Such tracking error becomes more severe as the activeness parameter λ decreases. Throughout the paper, we formalize this trade-off, define and solve the optimization problem of
determining the optimal activeness level λ
∗
that balances the trade-off.Where the paper says it
Where the paper says it
Key Findings and Implications
The analysis reveals a fundamental trade-off where smaller activeness λ reduces loss to arbitrageurs (LVR) but increases the typical price gap between the pool’s marginal price and the true price. The optimal policy is of feedback form, λopt(g) = clip 1 − γ2(1 + βv2) + β(2v2µ∆t + v1)2/g, λ, 1. p. 1 In the optimal index-tracking and optimal
execution literature, this is rarely optimal. An optimal policy often comprises two components:
a no-trading-zone and an execution plan, which it follows when a weight discrepancy escapes
that no-trading-zone [BFP18, DY03, HH13, GP13, GP16]. In this section, we formalize this trade-off as a discounted infinite-horizon stochastic control
problem, where {λn}n∈N is the control and the (linearized) gap process is the state. We then derive
a closed-form expression for the small-∆t optimal activeness λ
∗
. We show that this simple mechanism reduces adverse selection
costs, measured by loss-versus-rebalancing (LVR), and thereby improves the wealth of liquidity
providers (LPs) relative to plain constant-function market makers (CFMMs). As a trade-off,
the asset weights within a PA-AMM pool may deviate from their target weights implied by its
invariant curve.Where the paper says it
Where the paper says it
Where the paper says it
Mathematical Derivations
The proof of Proposition 1 relies on the Lipschitz property of the map Ψ, which ensures that the Markov chain admits a unique invariant distribution and converges to it geometrically. The derivation of E[g2] = σ2∆tλ(2 − λ) + O(∆t3/2) involves bounding the second derivative of the gap function Ψ, yielding a global bound Ψ''(g) ≤ 1/4. The solution to the Bellman equation yields the Riccati equation for v2, which determines the optimal control. p. 7 Notice that the coefficient λ(1 − λ) in Equation (3) is maximized at λ =
1
2
, i. , for a fixed small
gap gn, this choice maximizes the leading-order per-block increase of log-liquidity. For a more detailed explanation,
we refer the reader to check [Eva20]. 1 Constant Function Market Makers
Let invariant function φ : R
2
+ → R+ be C
3
, strictly concave, and 1-homogeneous.Where the paper says it
Where the paper says it
Conclusion and Future Work
In summary, PA-AMM limits the available liquidity within a single block by gradually executing rebalancing, which reduces LVR at the cost of tracking error. The work abstracts away gas costs and sets fees to zero, suggesting future work should adjust the optimization problem to find an optimal combination of fee rate and activeness. p. 2 In Section 5 we conclude with implications, limitations
(e. , fees and gas costs), and directions for future work. 1 Background and Related Works
Automated market makers (AMMs) have gained popularity and become a dominant type of
decentralized exchange (DEX) on blockchains due to their simplicity and low computational and
operational costs. However, because of their passive nature, AMM quotes cannot be updated in
real time, and liquidity providers (LPs) suffer from adverse selection.Where the paper says it
Where the paper says it
References
[AC01] Robert Almgren and Neil Chriss. Optimal execution of portfolio transactions. p. 1 Within such
a framework, CFMMs and the aforementioned improved mechanisms operate suboptimally in
the sense that they immediately remove any deviation beyond a certain threshold from the target
weights, that is, if the arbitrage happens in current block and the true price of the asset remains
the same in the next block, no trade from arbitrageurs will happen even with zero friction since
the opportunity has already gone in the last block. In the optimal index-tracking and optimal
execution literature, this is rarely optimal.Where the paper says it
[AD21] Andreas A. Aigner and Gurvinder Dhaliwal. Uniswap: Impermanent loss and risk profile of a liquidity provider. p. 0 There has been extensive work on the structure and magnitude of such losses, both theoretically
and empirically. In their early attempts practitioners used the notion of “impermanent loss”
(sometimes called “divergence loss”) as a cost of liquidity provision [Pin19a, Pin19b, AD21, Der20].Where the paper says it
[AF24] Abe Alexander and Lars Fritz. Fees in amms: A quantitative study. p. 1 Optimal execution is usually not a single
large market order that is filled immediately, but rather a series of sub-orders executed over time
[AC01, OW13]. Except [FSM+24], to our knowledge, there is no study on the mechanisms for
AMMs that perform such gradual rebalancing to align the price at AMMs to the true price of assets. xyz, Matroos Labs and Department of Mathematical Sciences, KAIST. 1
arXiv:2602.Where the paper says it
Where the paper says it
[AMRR24] Austin Adams, Ciamac C Moallemi, Sara Reynolds, and Dan Robinson. am-amm: An auction-managed automated market maker. p. 0 Partially Active Automated Market Makers
Ko Sunghun‗
February 11, 2026
Abstract
We introduce a new class of automated market maker (AMM), the partially active automated
market maker (PA-AMM). PA-AMM divides its reserves into two parts, the active and the passive
parts, and uses only the active part for trading. xyz, Matroos Labs and Department of Mathematical Sciences, KAIST. 1
arXiv:2602.Where the paper says it
Where the paper says it
[AZS+21] Hayden Adams, Noah Zinsmeister, Moody Salem, River Keefer, and Dan Robinson. Uniswap v3 core. p. 1 1Note that the target weight of each asset need not be kept fixed. For instance, the weight of assets in a Uniswap
V3 LP position may vary as prices move within the liquidity range if the range is finite [AZS+21]. 1Note that the target weight of each asset need not be kept fixed. For instance, the weight of assets in a Uniswap
V3 LP position may vary as prices move within the liquidity range if the range is finite [AZS+21].Where the paper says it
Where the paper says it
[Bal] Balancer. Liquidity bootstrapping pool. p. 1 For instance, the weight of assets in a Uniswap
V3 LP position may vary as prices move within the liquidity range if the range is finite [AZS+21]. Balancer’s liquidity
bootstrapping pool (LBP) is an example of CFMM with time-varying weights [Bal]. For instance, the weight of assets in a Uniswap
V3 LP position may vary as prices move within the liquidity range if the range is finite [AZS+21]. Balancer’s liquidity
bootstrapping pool (LBP) is an example of CFMM with time-varying weights [Bal].Where the paper says it
Where the paper says it
[BFP18] Konstantinos Benidis, Yiyong Feng, and Daniel P. Palomar. Optimization methods for financial index tracking: From theory to practice. p. 0 As a trade-off,
the asset weights within a PA-AMM pool may deviate from their target weights implied by its
invariant curve. Motivated by the optimal index-tracking problem literature, we also propose
and solve an optimization problem that balances such deviation and the reduction of LVR.Where the paper says it
[BHSB25] Leonardo Baggiani, Martin Herdegen, and Leandro Sánchez-Betancourt. Optimal dynamic fees in automated market makers. p. 0 Partially Active Automated Market Makers
Ko Sunghun‗
February 11, 2026
Abstract
We introduce a new class of automated market maker (AMM), the partially active automated
market maker (PA-AMM). PA-AMM divides its reserves into two parts, the active and the passive
parts, and uses only the active part for trading. xyz, Matroos Labs and Department of Mathematical Sciences, KAIST. 1
arXiv:2602.Where the paper says it
Where the paper says it
[BM20] Anton Bukov and Mikhail Melnik. Mooniswap. https://blockchainwelt.de/wp-content/uploads/2023/05/1inch-whitepaper. p. 0 More specifically, since asset
prices are often discovered on centralized exchanges (CEXs) and AMMs lack access to off-chain
information about the true value of assets, arbitrageurs can exploit stale AMM quotes and take
profit. Such profit is estimated to reach over 230 million USD between August 2023 and March
2025 [WSTP25].Where the paper says it
[CBMN25] Steven Campbell, Philippe Bergault, Jason Milionis, and Marcel Nutz. Optimal fees for liquidity provision in automated market makers. p. 0 1 Background and Related Works
Automated market makers (AMMs) have gained popularity and become a dominant type of
decentralized exchange (DEX) on blockchains due to their simplicity and low computational and
operational costs. However, because of their passive nature, AMM quotes cannot be updated in
real time, and liquidity providers (LPs) suffer from adverse selection. xyz, Matroos Labs and Department of Mathematical Sciences, KAIST. 1
arXiv:2602.Where the paper says it
Where the paper says it
[CDM23] Álvaro Cartea, Fayçal Drissi, and Marcello Monga. Predictable losses of liquidity provision in constant function markets and concentrated liquidity markets. p. 1 With additional assumptions such
as zero opportunity cost and the asset price following geometric Brownian motion with constant
volatility, the PL degenerates to the LVR. Moreover, studies on optimal liquidity provision, which concern the selection of the fee rate and
the liquidity provision range that balances fee revenue from noise traders and losses to arbitrageurs
[EAC21, CDM24, Pow24, AF24, MC24, VPT24, CBMN25], have emerged.Where the paper says it
[CDM24] Álvaro Cartea, Fayçal Drissi, and Marcello Monga. Decentralized finance and automated market making: Predictable loss and optimal liquidity provision. p. 0 1 Background and Related Works
Automated market makers (AMMs) have gained popularity and become a dominant type of
decentralized exchange (DEX) on blockchains due to their simplicity and low computational and
operational costs. However, because of their passive nature, AMM quotes cannot be updated in
real time, and liquidity providers (LPs) suffer from adverse selection.Where the paper says it
[CDSB+24] Álvaro Cartea, Fayçal Drissi, Leandro Sánchez-Betancourt, Dusan Siska, and Lukasz Szpruch. Strategic bonding curves in automated market makers. p. 0 Partially Active Automated Market Makers
Ko Sunghun‗
February 11, 2026
Abstract
We introduce a new class of automated market maker (AMM), the partially active automated
market maker (PA-AMM). PA-AMM divides its reserves into two parts, the active and the passive
parts, and uses only the active part for trading.Where the paper says it
[CF23] Andrea Canidio and Robin Fritsch. Arbitrageurs’ profits, lvr, and sandwich attacks: batch trading as an amm design response. p. 1 Many mechanisms that mitigate LVR and protect LPs have also been proposed. These proposed
mechanisms fall largely into three categories: batch execution of trades [RGGM24, CF23], dynamic
fee rates [Cro22, CDSB+24, Ego21, BHSB25], and discriminatory quotes based on (inferred) toxicity of orders [BM20, AMRR24, sol23]. xyz, Matroos Labs and Department of Mathematical Sciences, KAIST. 1
arXiv:2602.Where the paper says it
Where the paper says it
[Cro22] CrocSwap. Designing a dynamic fee policy that outperforms all uniswap eth/usdc pools. p. 8 For the middle and right panels, we used the historical ETH price from May
2025 to October 2025. All of the pools were initialized with 1000 ETH and an equivalent amount
of USDT.Where the paper says it
[Der20] Deribit Insights. Impermanent loss and bancor v2. p. 0 There has been extensive work on the structure and magnitude of such losses, both theoretically
and empirically. In their early attempts practitioners used the notion of “impermanent loss”
(sometimes called “divergence loss”) as a cost of liquidity provision [Pin19a, Pin19b, AD21, Der20].Where the paper says it
[DY03] Christopher Donohue and Kenneth Yip. Optimal portfolio rebalancing with transaction costs. p. 8 4 Finding Optimal λ
As we saw in Section 3, PA-AMM introduces a fundamental trade-off: smaller activeness λ reduces
loss to arbitrageurs (loss-versus-rebalancing, LVR), but increases the typical price gap between
the pool’s marginal price and the true price. For G3Ms, this trade-off admits a natural portfolio
interpretation.Where the paper says it
[EAC21] Alex Evans, Guillermo Angeris, and Tarun Chitra. Optimal fees for geometric mean market makers. p. 3 We further assume that φ is sufficiently smooth so that the aforementioned notions are all
well-defined. A geometric mean market maker (G3Ms) is a CFMM with invariant function φ defined
as
φ(x1, x2, · · · , xn) :=
Yn
i=1
x
wi
i
,
where (wi)i∈[n]
is weight of i-th asset satisfying Pn
i=1 wi = 1. We note that Corollaries 2 and 3 together reveal a fundamental trade-off between the price gap
and LVR reduction. Such a trade-off persists even when the true price is not a Brownian motion;
Figure 1 illustrates the results of the simulation using historical ETH price data.Where the paper says it
Where the paper says it
[Ego21] Michael Egorov. Automatic market-making with dynamic peg. p. 1 Many mechanisms that mitigate LVR and protect LPs have also been proposed. These proposed
mechanisms fall largely into three categories: batch execution of trades [RGGM24, CF23], dynamic
fee rates [Cro22, CDSB+24, Ego21, BHSB25], and discriminatory quotes based on (inferred) toxicity of orders [BM20, AMRR24, sol23].Where the paper says it
[Eva20] Alex Evans. Liquidity provider returns in geometric mean markets. p. 3 A geometric mean market maker (G3Ms) is a CFMM with invariant function φ defined
as
φ(x1, x2, · · · , xn) :=
Yn
i=1
x
wi
i
,
where (wi)i∈[n]
is weight of i-th asset satisfying Pn
i=1 wi = 1. For two-assets G3M with reserves
(x, y) and weights (θ, 1 − θ), the marginal price of asset X in terms of asset Y is then:
Pmargin(x, y) = θ
1 − θ
·
y
x
,
and for given log marginal price p and liquidity L, the reserves (x, y) are:
x(L, p) = L
θ
1 − θ
· e
−p
1−θ
, y(L, p) = L
1 − θ
θ
· e
p
θ
. xyz, Matroos Labs and Department of Mathematical Sciences, KAIST. 1
arXiv:2602.Where the paper says it
Where the paper says it
[FSM+24] Robin Fritsch, Maria Inês Silva, Akaki Mamageishvili, Benjamin Livshits, and Edward W Felten. Mev capture through time-advantaged arbitrage. p. 3 and ∆t is the block time. We further assume that there are infinitely many arbitrageurs competing with one another to
arbitrage against the AMMs, and that there is no retail investor for simplicity. xyz, Matroos Labs and Department of Mathematical Sciences, KAIST. 1
arXiv:2602.Where the paper says it
Where the paper says it
[GP13] Nicolae Gârleanu and Lasse Heje Pedersen. Dynamic trading with predictable returns and transaction costs. p. 1 Many mechanisms that mitigate LVR and protect LPs have also been proposed. These proposed
mechanisms fall largely into three categories: batch execution of trades [RGGM24, CF23], dynamic
fee rates [Cro22, CDSB+24, Ego21, BHSB25], and discriminatory quotes based on (inferred) toxicity of orders [BM20, AMRR24, sol23].Where the paper says it
[GP16] Nicolae Gârleanu and Lasse Heje Pedersen. Dynamic portfolio choice with frictions. p. 1 Many mechanisms that mitigate LVR and protect LPs have also been proposed. These proposed
mechanisms fall largely into three categories: batch execution of trades [RGGM24, CF23], dynamic
fee rates [Cro22, CDSB+24, Ego21, BHSB25], and discriminatory quotes based on (inferred) toxicity of orders [BM20, AMRR24, sol23].Where the paper says it
[HH13] Helge Holden and Lars Holden. Optimal rebalancing of portfolios with transaction costs. p. 1 Moreover, studies on optimal liquidity provision, which concern the selection of the fee rate and
the liquidity provision range that balances fee revenue from noise traders and losses to arbitrageurs
[EAC21, CDM24, Pow24, AF24, MC24, VPT24, CBMN25], have emerged. Especially, these studies
collectively indicate that the optimal choice is determined by various factors, such as the arrival
rate and distribution of order size of noise traders, asset volatility, competition between liquidity
providers, and the parameters determined by underlying blockchains, such as gas fee costs and
the time interval between two consecutive blocks.Where the paper says it
[KL17] Albert S Kyle and Jeongmin Lee. Toward a fully continuous exchange. p. 5 The following proposition tells us that there exists
a unique stationary distribution π∆t of the gap process {gn}n∈N with the estimation on its second
moment up to leading order. 5
In the sense that the trader should divide the order into many sub-orders and gradually submit them over multiple
blocks for optimal execution, one may consider PA-AMM as a form of implementation of a fully continuous exchange
proposed in [KL17], given that the block arrival interval ∆t is short enough.Where the paper says it
[MC24] Julian Ma and Davide Crapis. The cost of permissionless liquidity provision in automated market makers. p. 0 1 Background and Related Works
Automated market makers (AMMs) have gained popularity and become a dominant type of
decentralized exchange (DEX) on blockchains due to their simplicity and low computational and
operational costs. However, because of their passive nature, AMM quotes cannot be updated in
real time, and liquidity providers (LPs) suffer from adverse selection. Initially, its closed-form formula was based on assumptions of an
ideal situation (Poisson block arrivals, no friction in arbitrageurs’ trades, no gas fee), but such
‗
kosunghun317@matroos. xyz, Matroos Labs and Department of Mathematical Sciences, KAIST.Where the paper says it
Where the paper says it
[MMR24] Jason Milionis, Ciamac C Moallemi, and Tim Roughgarden. Automated market making and arbitrage profits in the presence of fees. p. 0 Partially Active Automated Market Makers
Ko Sunghun‗
February 11, 2026
Abstract
We introduce a new class of automated market maker (AMM), the partially active automated
market maker (PA-AMM). PA-AMM divides its reserves into two parts, the active and the passive
parts, and uses only the active part for trading. We note that Corollaries 2 and 3 together reveal a fundamental trade-off between the price gap
and LVR reduction. Such a trade-off persists even when the true price is not a Brownian motion;
Figure 1 illustrates the results of the simulation using historical ETH price data.Where the paper says it
Where the paper says it
[MMRZ22] Jason Milionis, Ciamac C Moallemi, Tim Roughgarden, and Anthony Lee Zhang. Automated market making and loss-versus-rebalancing. p. 0 At the top of every block, such a division is
done again to keep the active reserves always being λ-portion of total reserves, where λ ∈ (0, 1]
is an activeness parameter. We show that this simple mechanism reduces adverse selection
costs, measured by loss-versus-rebalancing (LVR), and thereby improves the wealth of liquidity
providers (LPs) relative to plain constant-function market makers (CFMMs). xyz, Matroos Labs and Department of Mathematical Sciences, KAIST. 1
arXiv:2602.Where the paper says it
Where the paper says it
[NT25] Alex Nezlobin and Martin Tassy. Loss-versus-rebalancing under deterministic and generalized block-times. p. 0 At the top of every block, such a division is
done again to keep the active reserves always being λ-portion of total reserves, where λ ∈ (0, 1]
is an activeness parameter. We show that this simple mechanism reduces adverse selection
costs, measured by loss-versus-rebalancing (LVR), and thereby improves the wealth of liquidity
providers (LPs) relative to plain constant-function market makers (CFMMs). xyz, Matroos Labs and Department of Mathematical Sciences, KAIST. 1
arXiv:2602.Where the paper says it
Where the paper says it
[OW13] Anna A Obizhaeva and Jiang Wang. Optimal trading strategy and supply/demand dynamics. p. 2 Throughout the paper, we formalize this trade-off, define and solve the optimization problem of
determining the optimal activeness level λ
∗
that balances the trade-off. We also observe an increase
in liquidity per unit of LP supply, due to the concavity of φ.Where the paper says it
[Pin19a] Pintail. Understanding uniswap returns. p. 1 1Note that the target weight of each asset need not be kept fixed. For instance, the weight of assets in a Uniswap
V3 LP position may vary as prices move within the liquidity range if the range is finite [AZS+21].Where the paper says it
[Pin19b] Pintail. Uniswap: A good deal for liquidity providers? p. 1 1Note that the target weight of each asset need not be kept fixed. For instance, the weight of assets in a Uniswap
V3 LP position may vary as prices move within the liquidity range if the range is finite [AZS+21].Where the paper says it
[Pow24] Corinne Powers. A tick-by-tick solution for concentrated liquidity provisioning. p. 9 (10)
Since multiplicative constants do not affect the optimizer of {λn}, we work with the leading-order
approximate objective
J ({λn}) ≈ E
"X∞
n=1
e
−ϱn∆t
(1 − λn)
2 + γλn
g
2
n
#
. (11)
We further assume that there exists a strictly positive lower bound λ on possible λn to ensure
the finiteness of the moments of gn and the validity of its linearized update rule, which justifies
the solution introduced below under a small ∆t regime. xyz, Matroos Labs and Department of Mathematical Sciences, KAIST. 1
arXiv:2602.Where the paper says it
Where the paper says it
[RGGM24] Geoffrey Ramseyer, Mohak Goyal, Ashish Goel, and David Mazières. Augmenting batch exchanges with constant function market makers. p. 0 At the top of every block, such a division is
done again to keep the active reserves always being λ-portion of total reserves, where λ ∈ (0, 1]
is an activeness parameter. We show that this simple mechanism reduces adverse selection
costs, measured by loss-versus-rebalancing (LVR), and thereby improves the wealth of liquidity
providers (LPs) relative to plain constant-function market makers (CFMMs).Where the paper says it
[SK25] Christoph Schlegel and Quintus Kilbourn. Arbitrage with bounded liquidity. p. 0 , the cost) of liquidity
providers (LPs) of AMMs from adverse selection. It has become a go-to metric for measuring
loss from CEX-DEX arbitrage. xyz, Matroos Labs and Department of Mathematical Sciences, KAIST. 1
arXiv:2602. xyz, Matroos Labs and Department of Mathematical Sciences, KAIST. 1
arXiv:2602.Where the paper says it
Where the paper says it
Where the paper says it
[sol23] solarcurve. [bip-295] fee discount for cowswap solvers. p. 0 It has become a go-to metric for measuring
loss from CEX-DEX arbitrage. Initially, its closed-form formula was based on assumptions of an
ideal situation (Poisson block arrivals, no friction in arbitrageurs’ trades, no gas fee), but such
‗
kosunghun317@matroos.Where the paper says it
[VPT24] Vladislav Volosnikov, Alexander Pimenov, and Vladimir Tikhomirov. The impact of market conditions and fee algorithms on the design of a competitive amm. p. 7 Reduced λ allows the LPs to set a lower swap
fee rate and attract more order flow from retail while bearing the same adverse selection cost as
its CFMM equivalent. However, as liquidity available per block declines, the price impact as a
function of swap size will become steeper and eventually overwhelm the benefit of the lower fee
rate.Where the paper says it
[WSTP25] Fei Wu, Danning Sui, Thomas Thiery, and Mallesh Pai. Measuring cex-dex extracted value and searcher profitability: The darkest of the mev dark forest. p. 0 , the cost) of liquidity
providers (LPs) of AMMs from adverse selection. It has become a go-to metric for measuring
loss from CEX-DEX arbitrage.Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Trading strategy optimization using PA-AMM allows for dynamic allocation of liquidity based on block arrival rates and volatility, enabling AI systems to choose an optimal activeness parameter lambda star (lambda star) to balance adverse selection costs and tracking error. This enables the AI system to set a trading policy that minimizes the weighted sum of tracking error and loss versus rebalancing.
-
Enhanced risk management for decentralized finance applications by using PA-AMM can significantly reduce losses to arbitrageurs, as
PA-AMM may be a better option for liquidity provision to prediction markets, where LPs face much larger adverse selection costs.
This means the AI system can provide more reliable capital allocation in high-risk environments. -
Improved liquidity bootstrapping and value estimation by utilizing the
liquidity per unit of LP supply grows over time even with a zero swap fee rate
property of PA-AMM. The AI system can better forecast the long-term growth potential of liquidity pools, leading to more informed decisions on capital deployment over extended periods. -
Automated execution planning for complex trading strategies by leveraging the
feedback form lambda opt(g)
solution. The AI system can dynamically adjust the fraction of liquidity available in each block based on the current price gap (g), optimizing execution speed and minimizing tracking error in real-time.
Abstract
We introduce a new class of automated market maker (AMM), the partially active automated market maker (PA-AMM). PA-AMM divides its reserves into two parts, the active and the passive parts, and uses only the active part for trading. At the top of every block, such a division is done again to keep the active reserves always being λ-portion of total reserves, where λ in (0, 1] is an activeness parameter. We show that this simple mechanism reduces adverse selection costs, measured by loss-versus-rebalancing (LVR), and thereby improves the wealth of liquidity providers (LPs) relative to plain constant-function market makers (CFMMs). As a trade-off, the asset weights within a PA-AMM pool may deviate from their target weights implied by its invariant curve. Motivated by the optimal index-tracking problem literature, we also propose and solve an optimization problem that balances such deviation and the reduction of LVR.
Sources
- Fees in AMMs: A quantitative study
- am-AMM: An Auction-Managed Automated Market Maker
- Optimal Dynamic Fees in Automated Market Makers
- Optimal Fees for Liquidity Provision in Automated Market Makers
- Arbitrageurs' profits, LVR, and sandwich attacks: batch trading as an AMM design response
- Liquidity Provider Returns in Geometric Mean Markets
- MEV Capture Through Time-Advantaged Arbitrage
- Automated Market Making and Loss-Versus-Rebalancing
- Loss-Versus-Rebalancing under Deterministic and Generalized block-times
- A Tick-by-Tick Solution for Concentrated Liquidity Provisioning
- Arbitrage with bounded Liquidity
- Measuring CEX-DEX Extracted Value and Searcher Profitability: The Darkest of the MEV Dark Forest
Related papers
- Valuation Reveals Uncertainty
- GROMM: Generative Risk-neutral Option Market Models
- Sublinear Markov Semigroups Beyond Uniqueness
- Price-impact Martingale optimal transport
- Low-rank and graphon limits for dynamic threshold distress contagion in heterogeneous financial networks
- Financial Resilience Evaluation: From Conditional Expectations to Dynamic Convex Risk Measures