The Strange Math That Predicts (Almost) Anything

The “strange math” referenced in the title refers to Markov chains, a probabilistic model that predicts future states based solely on the current state, ignoring the full history—a property known as the Markov property. Despite their simplicity, these chains underpin technologies ranging from Google’s search algorithm to nuclear physics simulations and AI language models.

Origins in a Russian Mathematical Feud

In 1906, Russian mathematician Andrey Markov challenged Pavel Nekrasov, who claimed that the Law of Large Numbers (LLN) required independent events and thus proved free will. Markov refuted this by analyzing Alexander Pushkin’s Eugene Onegin, showing that dependent sequences (like vowels and consonants) still obeyed statistical laws.

He found:

43% of letters were vowels, 57% consonants.
If independent, vowel-vowel pairs should occur at 0.43×0.43=18.5%.
In reality, they occurred only 6% of the time—proof of dependence.

Markov then built a two-state model:

From a vowel: 13% chance next is vowel, 87% consonant.
From a consonant: 33% vowel, 67% consonant.

Simulating this chain over time converged to the empirical 43%/57% split—proving dependence does not break statistical predictability.

This became the first Markov chain, laying the foundation for modeling dependent systems.

Core Principles of Markov Chains

A Markov chain consists of:

A set of states (e.g., weather conditions, webpage IDs).
A transition matrix P where each entry pij is the probability of moving from state i to state j.
The Markov property: P(Xt+1=j∣Xt=i,Xt−1=xt−1,…)=P(Xt+1=j∣Xt=i).

For an ergodic chain (irreducible and aperiodic), the system converges to a stationary distribution π such that π=πP, representing long-term probabilities.

Key Concepts:

Memoryless: Only the present matters.
Stationary distribution: Long-run equilibrium.
Mixing time: Steps needed to reach randomness.

Example: Shuffling cards. Each shuffle is a transition between permutations. Seven riffle shuffles are sufficient to randomize a 52-card deck—proven using Markov chain analysis by Persi Diaconis.

10 Real-World Examples of Markov Chains

#	Application	How Markov Chains Are Used
1	Google’s PageRank	Models web surfing as a Markov process: pages are states, links are transitions. Importance = stationary visit frequency. Includes a damping factor (85% follow links, 15% jump randomly).
2	Nuclear Chain Reactions	Neutron behavior in uranium-235 modeled as states (scatter, absorb, fission). Transition probabilities depend on energy and position. Simulated via Monte Carlo methods.
3	Text Prediction (Autocomplete)	Predicts next word based on current word or n-gram. Claude Shannon pioneered this using letter/word sequences. Basis for early language models.
4	Weather Forecasting	Tomorrow’s weather (sunny, rainy, cloudy) depends only on today’s state. Transition matrix built from historical data.
5	Stock Market Modeling	Price levels or trends treated as states. Transitions based on volatility, volume, and current trend. Used for short-term forecasting.
6	DNA Sequence Analysis	Nucleotides (A, T, C, G) modeled as states. Transition probabilities reveal genetic patterns and mutation risks.
7	Queueing Systems	Number of customers in line is a state. Transitions occur with arrivals and service completions. Used in call centers, traffic flow.
8	Card Shuffling	Each deck permutation is a state. Riffle shuffles define transitions. Seven shuffles achieve near-uniform distribution.
9	Epidemiology (Disease Spread)	States: Susceptible, Infected, Recovered (SIR model). Transition rates model infection and recovery probabilities.
10	Speech Recognition	Phonemes (speech sounds) are states. Acoustic models use Markov chains to predict likely word sequences from sound input.

20-Question Test on Markov Chains (With Answers)

Questions

Who developed the concept of Markov chains?
What literary work did Markov use to demonstrate his theory?
What percentage of letters in Eugene Onegin were vowels?
If letters were independent, what would be the expected frequency of vowel-vowel pairs?
What was the actual frequency of vowel-vowel pairs in Onegin?
What is the Markov property?
What is a transition matrix?
What is the stationary distribution in a Markov chain?
How many riffle shuffles are needed to randomize a 52-card deck?
What is the name of the method that uses random sampling to solve complex problems, inspired by solitaire?
Which mathematician used solitaire to inspire a computational breakthrough?
What does the damping factor in PageRank represent?
In PageRank, what do webpages represent in the Markov model?
What problem arises when AI-generated text is used to train future models?
What is mixing time in a Markov chain?
Can dependent events follow the Law of Large Numbers?
What was Pavel Nekrasov’s argument about free will?
What machine was used to simulate neutron behavior in the Manhattan Project?
What are the three possible fates of a neutron in a uranium-235 chain reaction?
What field uses Markov chains to model customer arrivals and service times?

Answers

Andrey Markov
Eugene Onegin by Alexander Pushkin
43%
18.5% (0.43 × 0.43)
6%
The future depends only on the present state, not the full history
A matrix where each entry pij is the probability of transitioning from state i to state j
The long-run probability distribution that remains unchanged over time ( π=πP )
Seven
Monte Carlo method
Stanislaw Ulam
The probability (15%) that the random surfer jumps to a random page instead of following a link
States
Feedback loops leading to repetitive, meaningless content
The number of steps required for the chain to get close to its stationary distribution
Yes, Markov showed this with dependent letter sequences
That independent human decisions (e.g., marriage, crime) following LLN prove free will
ENIAC
Scatter, be absorbed, or cause fission
Queueing theory (or operations research)

Note: We do use YouTube Video’s under the “Fair Use” Act under the Copyright Law:

“Fair use is a doctrine in the United States copyright law codified in Section 107 of the Copyright Act of 1976.¹ It provides for the legal, non-licensed citation or incorporation of copyrighted material in another author’s work without requiring permission from the rights holders, such as for commentary, criticism, news reporting, research, teaching or scholarship.⁰¹ The U.S. Copyright Office Fair Use Index should prove helpful in understanding what courts have to date considered to be fair or not fair but it is not a substitute for legal advice.²“