# Freeloading Games in Reward Symmetry (This formulation is a next iteration following a [previous article](https://hackmd.io/@onur/risk_sharing_in_pos)) Let us assume that each validator has staked some number of tokens, which determines how much reward they will receive. Let $w_i$ be the normalized weight of validator $i$. Validators have to have nonzero weights. Each validator is supposed to send exactly one message, and the strategies are `SEND` and `DON'T SEND` respectively. Let $x=(x_1,\dots,x_N)$ be the tuple of boolean values that hold whether a validator has sent a message. Then participating weight is calculated as $$ W_p(x) = \sum_{i\in V}x_iw_i $$ The total base reward is defined as R, and the actual amount to be distributed after correction is calculated as $W_p(x)R$. Reward symmetry dictates that each validator receives the same reward per weight, regardless of their sending a message or not. Thus, a validator $i$ receives $w_iW_p(x)R$ for a given participation profile $x$. Moreover, we dictate that a minimum participating weight $W_\min$ needs to be observed in order for everyone to receive any reward. If participating weight is greater than or equal to this value, we call the outcome *successful*, and *unsuccessful* otherwise. In unsuccessful rounds, no validator receives any reward. $$ s(x) = \begin{cases} 1 & \text{if } W_p(x)\geq W_\min \\ 0 & \text{otherwise} \end{cases}$$ Let us assume that the cost of sending a message for a given validator depends on weight, given by $c(w)$. **We also assume that $c(w)>0$ for $w>0$ and $c$ is monotonically increasing in $w$.** A validator $i$ incurs the cost only if they send a message, i.e. $x_i=1$. The main point of this game is that validators may "freeload" by not sending any message, because then they would not incur the cost, but still receive some rewards Finally, payoff of a validator $i$ is defined as $$ \pi_i(x) = \begin{cases} w_iW_p(x)R-c(w_i) & \text{if } s(x) = 1 \text{ and } x_i=1 \\ w_iW_p(x)R & \text{if } s(x) = 1 \text { and } x_i = 0 \\ -c(w_i) & \text{if } s(x) = 0 \text{ and } x_i=1 \\ 0 & \text{if } s(x) = 0 \text{ and } x_i=0 \end{cases} $$ This formulation is supposed to aid in choosing a base reward value such that there exists a Nash equilibrium $x$ where $s(x)=1$. It should be chosen so that being a validator is still profitable with a relatively small stake, e.g. $wR - c(w)> 0$ for $w=1/1000$. In fact, the maximum economically viable number of validators can be calculated as $N_\text{viable}=1/w$ where $w$ is obtained by solving $wR = c(w)$. ## Nash equilibria Ideally, one Nash equilibrium should be $x_1=x_2=\dots=x_n=1$. However, if the base reward is too low or the costs are too high, that equilibrium could be located at a lower value of participating weight. **Corollary:** If no validator $i$ exists such that $w_i\geq W_\min$, then $x_1=x_2=\dots=x_n=0$ will be a Nash equilibrium. **Theorem:** Let $\bar{V}\subset V$ be a totally ordered subset of validators such that - $w_i \leq w_j$ for all consecutive pairs $(i, j)$ from $\bar{V}$, - AND $w_i\geq w_j$ for all $i\in\bar{V}$, for all $j\in V\setminus\bar{V}$, - AND $\bar{W}_\text{total}-w_{\min(\bar{V})} \leq W_\min \leq \bar{W}_\text{total}$. where $\bar{W}_\text{total} = \sum_{i\in\bar{V}}w_i$. Then let $\bar{x}$ be an action profile such that $$ \bar{x}_i = \begin{cases} 1 & \text{if } i\in \bar{V} \\ 0 & \text{otherwise} \end{cases} $$ If $\pi_i(\bar{x})\geq 0$ for all $i\in V$, then there exists a Nash equilibrium $x$ such that $s(x)=1$. *Proof:* TBD $\square$ ### Making sure `DON'T SEND` is not a dominant strategy When $s(x)=0$, `DON'T SEND` yields a higher payoff for all $x$. So it is not possible to have a game where `SEND` is the dominant strategy for all cases. However, we can make sure `SEND` yields a higher payoff than `DON'T SEND` in optimal conditions so that `DON'T SEND` is not a dominant strategy. For a validator to choose `SEND`, there must exist an action profile where it gives a higher payoff than `DON'T SEND`. We compare two limiting cases where a validator with weight $w$ evaluates his options: #### Case A Validator's action causes the round to switch from successful to unsuccesful. We compare the payoffs when $W_p(x)=W_\min$ (i.e. $s(x)=1$) and $W_p(x')=W_\min-w$ (i.e. $s(x')=0$). Here, the only difference between $x$ and $x'$ is the given validator's action. **Lemma:** For a given validator with weight $w$, if $wW_\min R-c(w)\geq 0$ is satisfied, then `SEND` is the rational choice (has a higher payoff) for all $x$ where $W_p(x)\geq W_\min$. *Proof:* Using the definition of payoffs above and substituting values, we arrive at the inequality above. #### Case B The round is successful regardless of the validator's choice. We compare the payoffs when $W_p(x)=W_\min+w$ (i.e. $s(x)=1$) and $W_p(x')=W_\min$ (i.e. $s(x')=1$). **Lemma:** For a given validator with weight $w$, if $w^2R-c(w)\geq 0$ is not satisfied, then the validator's dominant strategy is `DON'T SEND`. *Proof:* Let $s(x)=1$ and $W_p(x)=\bar{W}\geq W_\min$. For the validator to choose `SEND`, corresponding payoff needs to be higher: $$ w\bar{W}R-c(w) \geq w(\bar{W}-w)R $$ Simplifying the expression, we arrive at the inequality above. If it is not satisfied, `DON'T SEND` yields a higher payoff in both cases and becomes the dominant strategy. $\square$ In the given setting, weight limit will always be smaller than $W_\min$, because $W_\min\geq 0.5$ and setting the weight limit above that would allow for only one validator in the game. For that reason, case B always yields the more critical condition, that is, if the inequality in case B is satisfied, then the inequality in case A is also always satisfied. ### Linear cost function Let us assume that the cost function is linear: $c(w) = c_vw+c_f$, where $c_v$ is the variable cost and $c_f$ is the fixed cost of sending a message. Substituting it in the inequality from case B, we get $$ Rw^2-c_vw - c_f \geq 0. $$ We solve the inequality for $w$ to obtain $$ w \geq \frac{\sqrt{c_v^2+4Rc_f}+c_v}{2R} =: w_\text{honest} $$ This defines the threshold at which validators switch from freeloading to being honest. The subset of validators with weight smaller than $w_\text{honest}$ are expected to freeload regardless of others' strategies, *in the given game context*. If the maximum validator weight at which freeloading can be tolerated is $w_\text{honest}$, then the base reward should be set to a minimum of $$ R_\min := \frac{c_vw_\text{honest}+c_f}{w_\text{honest}^2}. $$ *Example:* The game can be generalized to multiple rounds that occur over the course of a year. Let us assume that the information of previous rounds is not retained---we can then treat the whole year as a single game. Let there be a only a fixed cost of participating, equal to 500 USD per month, amounting to 6000 USD per year. Let us also assume that we cannot tolerate freeloading of validators with weight greater than 1%. Then, the minimum base reward that should be emitted every year should be equal to $6000/0.01^2=60\text{m}$ USD. Similarly, we can calculate the minimum weight that should be set in-protocol for a given base reward. For a yearly issuance rate of 2%, total token supply of 100m and token valuation at 0.1 USD. ### Setting a minimum weight to ensure honesty of all validators In designing participation rules, it might be wise to set a minimum validator weight $w_\min = w_\text{honest}$, so that all validators are incentivized not to freeload. In that case, the maximum possible number of validators would be equal to $N_\max = 1/w_\min$. If costs were incurred in fiat and rewards were paid out in tokens, then $N_\max$ would depend on the token price $P$, namely $N_\max \propto P$. As the price increased, the network would be able to support more validators of smaller weight. Now that we introduced the additional dimension of token price, we can move on to add more complexity into our game. A single round game is too simplistic to model participation in a Proof of Stake protocol, because the protocol governs a continuous process of proposing and finalizing new blocks, where each block corresponds to one game round. Thus, it makes more sense to model freeloading as a repeated game, where the outcome of previous rounds feed back into the token price. To explain the approach in one sentence: if no blocks get finalized, then the token price drops to zero, eliminating the prospect of *any* future reward. Therefore, freeloading might be even less rational after all. ## Repeated games We assume that multiple rounds of the game are played consecutively, and that the aggregate success of rounds feed back into the token price. Let us define a shorthand $s_t:=s(x^t)$ where $x^t$ is the action profile of round $t$. Then let $S_t = (s_0,s_1,\dots,s_{t-1})$ be the tuple of results of rounds preceding round $t$. We assume that price at round $t$ is given by a function $P_t(S_t)$. We assume that rewards are paid out in tokens and costs are incurred in fiat. Then the payoff of validator $i$ for a single round $t$ with an action profile $x^t$ is given by $$ \pi_i(x^t) = \begin{cases} w_iW_p(x^t)P_t(S_t) R-c(w_i) & \text{if } s_t = 1 \text{ and } x^t_i=1 \\ w_iW_p(x^t)P_t(S_t) R & \text{if } s_t = 1 \text { and } x^t_i = 0 \\ -c(w_i) & \text{if } s_t = 0 \text{ and } x^t_i=1 \\ 0 & \text{if } s_t = 0 \text{ and } x^t_i=0 \end{cases} $$ We assume that the price function admits the following property: For all $S_t\in\{0,1\}^{t-1}$ and for all $t'\in[0,t-1]$, $$P(S_t)\geq P(\tilde{S}_t)$$ where ($s_{t'}=1$) AND ($\tilde{s}_{t'}=0$) AND ($s_u=\tilde{s}_u$ for all $u\in[0,t-1]\setminus\{t'\}$). In other words, comparing the price yielded by two sequences of rounds where the only difference is the result of a round at a certain index, the sequence where that round is successful yields a greater or equal price than the one where it is not. Example: $P_5(1,0,\underline{1},1,0)\geq P_5(1,0,\underline{0},1,0)$. **Corollary:** $P_t$ is maximized when $s_0=\dots=s_{t-1}=1$, namely all rounds up until that point were successful. An example for such a price function could be $$ P_t(s) = \frac{\sum_{u=0}^{t-1}s_u}{t} $$ TBD
Last changed by  
 
Onur
796

Read more

[ICortex](https://github.com/textcortex/icortex): Giving Programming Superpowers to Non-Programmers

Introducing ICortex, our new Jupyter kernel that allows you to generate Python code from natural language prompts: {%youtube AtVCBNq48oc %} ICortex is a drop-in replacement for the IPython kernel that powers Jupyter Notebooks. It is open source, fully extensible, and can connect to multiple code generation services, including TextCortex API, OpenAI Codex API, locally-running HuggingFace models, and so on. Whether you are ...

10/13/2022
AutoPrice

Background Control Mechanisms for Trustless Self-Governing Protocols It is a feature of public blockchains, where a protocol parameter needs to adapt to certain conditions, but should not be controlled by a specific participant. The solution is to implement a mechanism where it auto-adjusts, according to quantifiable and objective conditions. The primary example is Bitcoin’s difficulty adjustment. There, variations in average block time are used to dynamically adjust Bitcoin’s block difficulty, every 2016 blocks. Owing to difficulty adjustment, Bitcoin blocks are mined every 10 minutes on average without any external control. Shifts in block time feed back into the difficulty parameter, which then dampens said shifts, creating a negative feedback loop. Here, Bitcoin's designer made use of elementary control theory, setting mean block time as a process variable and difficulty adjustment as the control action: Gas Price Volatility Blockchains have a limited supply, and a vertical supply curve. For that reason, blockchain resources are subjected to very high price volatilities, once user demand surpasses platform capacity. Let's consider Ethereum. Demand for gas isn’t distributed equally around the globe. Ethereum users exist in every inhabited continent, with the highest demand seen in East Asia, primarily China. Europe+Africa and the Americas seem to be on par in terms of demand. This results in predictable patterns that follow the peaks and troughs of human activity in each continent. The correlation between gas usage and price is immediately noticeable, demonstrated by a 5 day period from March 2019. The grid marks the beginnings of the days in UTC, and the points in the graph correspond to hourly averages, calculated as:

6/29/2020
Lazy Strategies

In Proof of Work We investigate the strategy where miners mine empty blocks to reduce the uncertainty of mining a block which is invalid because it contains a transaction that is already included by another block. In this case, the miner is aware of the new block, already has its hash and/or header, but has not downloaded the rest of the block which contains the transactions. Constant reward per mined block: $R$ Time required to mine an empty block: $T_e$ Time required to mine a full block: $T_f$ Gas price: $P$ Gas in a block: $G$ Block gas limit: $G_\max$ Average tx fee collected from a block: $F(G) = PG$

5/18/2020
Compensation Fees in Transaction Fee Distribution

Users need to specify a fee when submitting their transactions. When it is a validator’s turn to propose a block, the validator collects transactions from the transaction pool, executes them in a certain order, and publishes them in a new block. Fees in a block are distributed to all validators pro rata. This is to weaken the incentive to steal transactions from other blocks, in case the block proposer receives the whole fee. The protocol has the following parameters: $P$: Average gas price of the previous era. This value is updated at the end of each era, calculated by dividing total collected fees by total consumed gas. $G$: Gas in a block. $G_{\text{max}}$: Block gas limit. $G \leq G_{\text{max}}$ for every block. $F$: Total fee collected from transactions in a block. $N$ is the current number of validators, $w_i$ is the normalized weight of validator $i$.

5/15/2020
Read more from Onur
Published on HackMD