Absorption and Inertness in Coarse-Grained Arithmetic: A Heuristic Application to the St. Petersburg Paradox

This paper proposes a heuristic resolution to the St. Petersburg paradox by introducing a coarse-grained addition framework where repeated aggregation of rewards eventually reaches a state of "inertness," thereby preventing unbounded growth without relying on traditional assumptions like diminishing marginal utility.

The Gist

The Big Problem: The "Infinite Money" Game

Imagine a game called the St. Petersburg Paradox. Here is how it works:

• You flip a coin.
• If it lands on Heads, you win a prize, and the game continues.
• If it lands on Tails, the game ends, and you get paid.
• The catch? Every time you get a Head, the prize doubles.
  • 1st Tail: You win $1.
  • 2nd Tail (after 1 Head): You win $2.
  • 3rd Tail (after 2 Heads): You win $4.
  • 4th Tail: You win $8.
  • And so on...

Mathematically, the "expected value" (the average amount you should win if you played this game a billion times) is infinity. According to strict math, you should be willing to pay any amount of money—even your house, your car, and your future earnings—to play this game once.

But in real life, nobody does that. If someone offered you this game, you'd probably only pay $10 or $20. Why? Because our brains don't work like infinite calculators. We have limits.

The Paper's New Idea: "Coarse-Grained" Thinking

Most economists try to solve this by saying, "Money isn't worth as much to you when you have a lot of it" (diminishing utility) or "Money in the future is worth less than money today" (discounting).

Takashi Izumo's paper takes a different approach. He suggests the problem isn't how we value money, but how we add it up.

He proposes that our brains don't add numbers like a supercomputer (1 + 1 = 2, 2 + 1 = 3). Instead, we add them like a bucket with a fuzzy ruler.

The Analogy: The Fuzzy Bucket

Imagine you are filling a bucket with water, but the bucket has a very strange, fuzzy ruler painted on the side.

• The ruler doesn't show every single drop. It only has big zones: "Empty," "Low," "Medium," "High," "Full."
• When you pour a cup of water in, you don't see the exact new level. You just see which zone the water lands in.
• Once the water is in the "High" zone, the ruler says, "Okay, we are in the High zone."

Now, imagine you keep pouring tiny cups of water (representing the small, steady gains of the game).

1. Normal Math: If you keep pouring, the water level rises forever. Eventually, the bucket overflows.
2. Coarse Math (The Paper's Idea): Because the ruler is fuzzy, once you are in the "High" zone, pouring in a tiny cup doesn't move the water enough to cross the line into the "Very High" zone. The water level stays stuck in the "High" zone.

The paper calls this "Inertness." It means that after a certain point, adding more stuff doesn't change the result anymore. The system becomes "inert" (stiff, unresponsive).

How the Math Works (The "Grains")

The author breaks the number line (0, 1, 2, 3...) into chunks called "Grains."

• Grain 1: Numbers 0–2.
• Grain 2: Numbers 3–5.
• Grain 3: Numbers 6–10.
• And so on.

Inside each grain, there is a Representative (a "boss" number). Let's say the boss of Grain 2 is the number 4.

The Rule of Addition:
When you add two numbers, you don't just add the raw numbers. You:

1. Find which Grain they belong to.
2. Grab their "Boss" numbers.
3. Add the Boss numbers.
4. Crucial Step: Look at the result. Which Grain does that result fall into? That Grain becomes your new total.

The Magic of "Absorption":
Sometimes, a Grain is so big that it can "swallow" a smaller addition without changing its identity.

• Imagine you are in a huge Grain (say, numbers 100 to 200).
• You add a tiny number (like 1).
• The result (101) is still inside the same huge Grain.
• Because the result is still in the same Grain, the system says, "Nothing changed!"
• If you keep adding 1s, you might stay in that same Grain forever. The total stops growing.

Applying it to the Paradox

In the St. Petersburg game, the "expected gain" at every step is a tiny, constant amount.

• Standard Math: Tiny amount + Tiny amount + Tiny amount... = Infinity.
• Coarse Math: Tiny amount + Tiny amount... = Stuck in a Grain.

The paper shows that if you choose the right "fuzzy ruler" (the right way to group numbers), the infinite stream of tiny gains from the St. Petersburg game will eventually hit a point where the brain (the coarse system) stops noticing the growth. The total value freezes.

Why This Matters

This doesn't mean the game actually pays out less money. It means that human perception of value has a limit.

• Discounting (Old Idea): "I don't care about money I get 100 years from now." (Time-based).
• Coarse Addition (New Idea): "I don't care about a tiny bit of extra money when I already have a huge pile." (Size-based).

It's like holding a grain of sand. If you have 10 grains, adding one more feels significant. If you have a mountain of sand, adding one grain feels like nothing. The paper formalizes this "numbness" to big numbers using math.

The Catch (Non-Associativity)

There is a weird side effect. In normal math, $(A + B) + C$ is the same as $A + (B + C)$.
In this "Coarse" math, order matters.

• If you add small numbers first, you might get stuck in a small grain.
• If you add a big number first, you might jump to a huge grain that swallows the small numbers.
  This mimics how humans are sometimes inconsistent: we might react differently to a series of small gains depending on how we group them in our heads.

The Bottom Line

The paper argues that the St. Petersburg Paradox isn't a bug in human logic; it's a feature of coarse cognition. We aren't bad at math; we are just using a "low-resolution" calculator. When you add up infinite tiny rewards using a low-resolution system, the total eventually stops growing, making the "infinite" game feel like it has a reasonable, finite price tag.

Technical Summary

1. Problem Statement

The paper addresses the St. Petersburg Paradox, a classic problem in decision theory where a game offers an infinite expected monetary value ($E[X] = \infty$), yet rational agents are unwilling to pay a large finite amount to play.

• The Paradox: The expected payoff is calculated as $\sum_{n=1}^{\infty} \frac{1}{2^n} 2^{n-1} = \sum_{n=1}^{\infty} \frac{1}{2} = +\infty$.
• Limitations of Existing Solutions: Traditional resolutions rely on:
  1. Diminishing Marginal Utility: Replacing money with a concave utility function.
  2. Temporal Discounting: Applying a discount factor $\delta^n$ to future payoffs.
  3. Unbounded Utility Theory: Reordering divergent series within richer preference frameworks.
  4. Prospect Theory: Using probability weighting.
• The Gap: The author argues that these approaches modify the value or probability of outcomes. This paper proposes a different approach: modifying the aggregation rule (addition) itself. It posits that human cognition may not perform "fine-grained" arithmetic but rather "coarse-grained" aggregation, where small increments eventually fail to alter the perceived total.

2. Methodology: Coarse-Grained Arithmetic

The paper constructs a formal mathematical framework for Coarse-Grained Arithmetic on the set of non-negative integers $U = \mathbb{N}_0$.

A. Structural Setup

1. Partition ($\pi$): The set $U$ is partitioned into a countable family of finite, ordered intervals called grains ($G_{\pi, i}$).
   • $G_{\pi, i} = \{a_i, a_i+1, \dots, b_i\}$.
2. Cell Map ($\psi_\pi$): A function mapping any element $x \in U$ to the specific grain it belongs to.
3. Internal Representative Map ($\phi$): A function assigning a single representative value $\phi(G) \in G$ to each grain. Common choices include the minimum, maximum, or median of the grain.

B. Operations

The paper defines two new addition operations that replace standard arithmetic:

1. Coarse Representative Addition ($\oplus_{\pi, \phi}$): For $x, y \in U$:
   $$x \oplus_{\pi, \phi} y := \phi(\psi_\pi(\phi(\psi_\pi(x)) + \phi(\psi_\pi(y))))$$
   Process: Map inputs to grains $\to$ select representatives $\to$ add normally $\to$ map sum back to a grain $\to$ select new representative.
2. Coarse Cell Addition ($\boxplus_{\pi, \phi}$): An operation on grains themselves, defined similarly.

3. Key Concepts and Structural Properties

A. Absorption

A grain $G$ absorbs a grain $H$ if adding $H$ to $G$ results in $G$ remaining unchanged.

• Condition: $G$ absorbs $H$ if $\phi(G) + \phi(H) \in G$.
• Absorption Margin ($\mu$): Defined as $\mu_\phi(G) = \max(G) - \phi(G)$.
• Criterion: If the representative of the increment $\phi(H)$ is less than or equal to the absorption margin of the current grain ($\phi(H) \le \mu_\phi(G)$), then $G$ absorbs $H$.

B. Inertness

Inertness is the phenomenon where a sequence of repeated additions eventually stabilizes at a fixed value (or grain), despite the underlying standard sum diverging to infinity.

• Mechanism: If a sequence of increments is repeatedly absorbed by the current state, the coarse partial sums become constant after a finite number of steps.
• Sufficient Condition: If a sequence of grains $(H_n)$ reaches a state $G^*$ such that for all subsequent $n$, $\phi(H_n) \le \mu_\phi(G^*)$, the process becomes inert at $G^*$.

C. Non-Associativity

Unlike standard addition, coarse addition is generally non-associative.

• Reason: The intermediate projection to a grain representative discards "fine-grained" information. The order of operations (bracketing) determines which intermediate representatives are selected, leading to different final outcomes.
• Exception: Associativity is recovered only under highly specific conditions (e.g., partitions with constant odd widths and median representatives).

4. Application to the St. Petersburg Paradox

The author applies this framework heuristically to the St. Petersburg game.

• Rescaling: Instead of the divergent series of probabilities, the paper considers the sequence of expected increments. The expected increment at each step is constant ($1/2$). To fit the integer domain $\mathbb{N}_0$, the sequence is rescaled to a constant integer sequence: $(1, 1, 1, \dots)$.
• Triangular Partition: The author constructs a specific partition where the size of the $n$-th grain is $n$ (Triangular numbers).
  • $G_n = \{T_{n-1}, \dots, T_n - 1\}$, where $T_n = n(n+1)/2$.
• Representative Choice: The minimum element of each grain is chosen as the representative ($\phi(G_n) = T_{n-1}$).
• Result:
  • The absorption margin for grain $G_n$ is $n-1$.
  • The increment is always $1$.
  • For any grain $G_n$ where $n \ge 2$, the condition $1 \le n-1$ holds.
  • Conclusion: Once the running total enters $G_2$ (or any subsequent grain), the constant increment of $1$ is absorbed. The coarse partial sums stabilize immediately at the representative value of $G_2$ (which is 1).
  • The divergent series $1+1+1+\dots$ becomes inert (constant) under this coarse arithmetic.

5. Key Contributions

1. New Aggregation Rule: Introduces "Coarse Addition" as a distinct algebraic mechanism that differs from standard arithmetic by incorporating projection and loss of precision.
2. Formalization of Inertness: Defines and proves conditions under which divergent sequences become constant (inert) due to the "absorption" of small increments by large totals.
3. Structural Insight into the Paradox: Demonstrates that the St. Petersburg Paradox can be "resolved" not by changing utility or time preferences, but by modeling bounded numerical cognition. It suggests that if human aggregation is coarse, infinite expected value does not necessarily imply an unbounded perceived total.
4. Distinction from Discounting: Clarifies that this mechanism is distinct from temporal discounting. Discounting reduces the weight of future payoffs; coarse addition reduces the sensitivity to increments relative to the current total size.

6. Significance and Limitations

• Significance: The paper provides a structural explanation for bounded rationality. It suggests that the "paradox" may stem from the assumption that humans perform exact, infinite-precision arithmetic. If cognition is coarse-grained, the "infinite" expectation is mathematically capped by the absorption properties of the cognitive scale.
• Limitations:
  • The result is heuristic, not a solution within standard decision theory (the classical expectation remains infinite).
  • The outcome depends heavily on the specific choice of partition and representative map. Different coarse structures could yield non-inert results.
  • The model is restricted to countable discrete partitions of $\mathbb{N}_0$ and left-associative sums.

Conclusion

The paper concludes that divergence in the classical sense does not translate to unbounded growth in a coarse-grained system. By formalizing the mechanism of absorption, the author offers a mathematical model for how an agent with limited cognitive precision might perceive a divergent reward structure as finite and stable, providing a novel perspective on the St. Petersburg Paradox and bounded aggregation in behavioral economics.