skip to content
Naveen Sahu

Ethereum Yellow Paper Mathematics Deciphered | Part 1: World State

/ 8 min read

This part of the series will cover section 2 of the Yellow Paper, which explains world state.

I assume you’ve already gone through conventions defined in Part 0. If you haven’t yet, I highly recommend you to go check it out first.

World State

Ethereum can be considered as a transaction based state machine. It begins with a genesis state and by executing transactions, it goes to a new state - a valid one enforced by a set of rules. The state of this state machine is termed - “world state” and denoted by $\boldsymbol{\sigma}$. The $\boldsymbol{\sigma}$ is, simply put, a mapping between addresses and corresponding account states as you’ll see more later.

    | WORLD STATE |

Address     AccountState
----------------------------------------------------------------
address1 -> accountState1(nonce, balance, storageRoot, codeHash)
address2 -> accountState2(nonce, balance, storageRoot, codeHash)
address3 -> accountState3(nonce, balance, storageRoot, codeHash)
        .
        .

That brings us to the first equivalence from paper -

$$ \boldsymbol{\sigma}_{t+1} \equiv \Upsilon(\boldsymbol{\sigma}_t, T) \tag{1} $$

where,

  • $\boldsymbol{\sigma}_t$ is the world state at some time $t$
  • $T$ is a single transaction represented by a tuple
  • $\Upsilon$ is the state transition function

When given the current state $\boldsymbol{\sigma}_t$ and a transaction $T$, the state transition function $\Upsilon$ is defined to be a function that will generate a valid next state $\boldsymbol{\sigma}_{t+1}$.

Keep in mind that $\Upsilon$ provides a new state given a transaction tuple, $T$, i.e. next state that’ll result from executing a single transaction $T$.

Now, comes the block. Isn’t it that Ethereum jumps to a new state, when a block is mined? Where does the block fit in the equation?

Well, yes a new state is indeed generated when a block in finalized by mining. In fact, the individual state transitions corresponding to each transaction in the block ($T_0, T_1, T_2…$) happen through $\Upsilon$ function.

$$ \boldsymbol{\sigma}_{t+1} \equiv \Upsilon(\boldsymbol{\sigma}_{t}, T_0) \\ \boldsymbol{\sigma}_{t+2} \equiv \Upsilon(\boldsymbol{\sigma}_{t+1},T_1) \\ \boldsymbol{\sigma}_{t+3} \equiv \Upsilon(\boldsymbol{\sigma}_{t+2},T_2) \\ $$

and so on…

Above three can be combined to be written as: $$ \boldsymbol{\sigma}_{t+3} \equiv \Upsilon( \Upsilon( \Upsilon(\boldsymbol{\sigma}_t, T_0) , T_1) , T_2) \tag{i} $$

Now, we introduce another state-transition function $\Pi$ from paper, $$ \boldsymbol{\sigma}_{t+1} \equiv \Pi(\boldsymbol{\sigma}_t, B) \tag{2} $$

where,

  • $\boldsymbol{\sigma}_t$ is the world state at time $t$
  • $B$ is a single block, represented as a tuple
  • $\Pi$ is the state transition function at block level

$\Pi$ may look similar to $\Upsilon$, but notice that $\Upsilon$ transitions state at transaction level. While $\Pi$ transitions state at block level. It transitions the world state, $\boldsymbol{\sigma}$, to a new state by including and finalizing all the transactions in the block $B$.

The block is a tuple with multiple components, one of which is a list of transactions (precisely, a list of transaction tuples), $\mathbf{T} = (T_0, T_1. T_2, \ldots)$

$$ B \equiv (\ldots, \mathbf{T}, \ldots) \
$$ which is same as- $$ B \equiv (.., (T_0, T_1. T_2,…), ..) \tag{3} $$ (Note: ellipsis before and after the list of transactions, $\mathbf{T}$ abbreviate other components of the block tuple. Other components of $B$ will be introduced later in series.)

Paper defines another function $\Omega$ to be equivalent to $\Pi$ as, $$ \Pi(\boldsymbol{\sigma}, B) \equiv \Omega(B, \Upsilon( \Upsilon(\boldsymbol{\sigma}, T_0) , T_1) \ldots ) \tag{4} $$

where,

  • $\Omega$ is the block finalization state transition function
  • $\Pi$ is the block level state transition function
  • $B$ is a block tuple

This might seem a bit confusing, but remember that all of the above are equivalences ($\equiv$) not equalities ($=$).

The RHS of $(4)$ seems to elaborate the $\Pi$ a bit more explicitly through $\Omega$. Notice the $(\Upsilon(\Upsilon(\boldsymbol{\sigma}, T_0), T_1) \ldots)$ here is same as described at $(i)$ above, which is equivalent to a state $\boldsymbol{\sigma}$. If you tried to substitute it in the $(4)$,

$$ \Pi(\boldsymbol{\sigma}, B) \equiv \Omega(B, \boldsymbol{\sigma}) \tag{ii} $$

the equivalence seems more obvious. Additionally, $\Omega$ (and hence $\Pi$) is defined to reward the nominated miner of the block $B$ apart from finalization of $B$.

Account State

As said before, the world state, $\boldsymbol{\sigma}$, is a mapping between addresses and their account states. The account state is a RLP-serialized data-structure. For an address $a$, the account state can be denoted by $\boldsymbol{\sigma}[a]$ - similar to accessing a value in a map by a key (here, $a$) in programming.

An account state $\boldsymbol{\sigma}[a]$ comprises of the following components -

  • $\mathbf{nonce}$ ($\boldsymbol{\sigma}[a]_\mathrm{n}$): Nonce is number of transactions sent by this address, if it is an EOA, or it is number of contract-creations made by this address if it is a contract.

  • $\mathbf{balance}$ ($\boldsymbol{\sigma}[a]_\mathrm{b}$): Number of Wei owned by this address.

  • $\mathbf{storageRoot}$ ($\boldsymbol{\sigma}[a]_\mathrm{s}$): A 256-bit hash of root node of a Merkle Patricia Trie data-structure. This MPT structure holds contents of this account. The contents in the MPT are encoded mapping of 256-bit hash of 256-bit keys (aka storage slot) to RLP-encoded 256-bit integers as values (representing the account content). In simple terms, you might know this “content” as contract storage. As expected, only contract accounts can have non-empty storage. EOAs always have empty storage.

  • $\mathbf{codeHash}$ ($\boldsymbol{\sigma}[a]\mathrm{c}$): The hash of the EVM code (aka bytecode) of this account. Remember that only contracts have bytecode, EOAs have empty bytecode. Thus, if $\mathbf{b}$ is the bytecode then, $\boldsymbol{\sigma}[a]\mathrm{c} = \mathtt{KEC}(\mathbf{b})$. And $\mathbf{b} = ()$, always for an EOA account - meaning $\mathbf{codeHash}$ for an EOA is always hash of an empty string.

A diagram showing the make up of an account

Paper defines a function $L_I$ that given a key, $k$ and a value, $v$ outputs suitable item for inserting into the account storage Merkle Patricia Trie. In this regard, $k$ must be keccak-256 hashed and $v$ must be RLP-encoded, as discussed earlier above in $\mathbf{storageRoot}$ component. Hence,

$$ L_I((k, v)) \equiv (\mathtt{KEC}(k), \mathtt{RLP}(v)) \tag{8} $$

where $k$ must be 32-byte (256-bit) long and $v$ must be a natural number. Formally this is same as,

$$ k \in \mathbb{B_{32}} \land v \in \mathbb{N} \tag{9} $$

According to convention then, $L_T^$ is a similar function to $L_I$ except $L_T^$ takes as input a series of key-value pairs $((k_0, v_0), (k_1, v_1), (k_2, v_2)…)$ and performs same operation but element-wise. That is,

$$ L_T^*(( (k_0, v_0), (k_1, v_1),…) ) = (L_T((k_0, v_0)), L_T((k_1, v_1)), …) \tag{iii} $$

Finally, using all the above the paper defines a convenient equivalence -

$$ \mathtt{TRIE}( L_I^*( \boldsymbol{\sigma}[a]_\mathbf{s} ) ) \equiv \boldsymbol{\sigma}[a]_{\mathrm{s}} \tag{7} $$

Note that in the equivalence above, the subscript $\mathbf{s}$ (bolder, on LHS) is different from subscript $\mathrm{s}$ (on RHS)

We already know that $\boldsymbol{\sigma}[a]_{\mathrm{s}}$ is the $\mathbf{storageRoot}$ defined earlier. If you look carefully, $\mathtt{TRIE}( L_I^*( \boldsymbol{\sigma}[a]_{\mathbf{s}} ) )$ is actually defines the same Merkle Patricia Trie whose root’s hash is $\mathbf{storageRoot}$. How?

$\boldsymbol{\sigma}[a]_{\mathbf{s}}$ (bolder $\mathbf{s}$) seems to represent a list/series of contract’s storage key-value pairs (raw integer key and raw storage content) corresponding to an account, $a$. The $L_I^*$ then hashes each of the keys and RLP-encodes each of the values. Just as defined at $(iii)$. And with these encoded inputs the Merkle Patricia Trie is constructed with $\mathtt{TRIE}$ function.

The equivalence $(7)$ is made just for convenience since we typically refer to the trie’s underlying set of key-value pairs and NOT just the hash of storage trie root $\boldsymbol{\sigma}[a]{\mathrm{s}}$. It is assumed from now on that the latter is equivalent to former i.e. $\boldsymbol{\sigma}\mathrm{s}$ maybe used to refer to the MPT and not just the simple hash value.

A non-existent account is defined as:

$$ \boldsymbol{\sigma}[a] = \varnothing $$

Whereas an empty account i.e. it has no code (bytecode), zero nonce and zero balance, given as:

$$ \mathtt{EMPTY}(\mathbf{\sigma}, a) \equiv \boldsymbol{\sigma}[a]\mathrm{c} = \mathtt{KEC}(()) \land \boldsymbol{\sigma}[a]\mathrm{n} = 0 \land \boldsymbol{\sigma}[a]_\mathrm{b} = 0 \tag{14} $$

An account is dead if it is either non-existent or empty. Mathematically it is same as,

$$ \mathtt{DEAD}(\boldsymbol{\sigma}, a) \equiv \boldsymbol{\sigma} = \varnothing \lor \mathtt{EMPTY}(\boldsymbol{\sigma}, a) \tag{15} $$

Now, paper defines the world-state collapse function $L_\mathrm{S}$ for existent accounts, $a$ as,

$$ L_\mathrm{S} \equiv \{ p(a): \mathbf{\sigma}[a] \neq \varnothing \} $$

where,

$$ p(a) \equiv \big(\mathtt{KEC}(a), \mathtt{RLP}\big( (\boldsymbol{\sigma}[a]_{\mathrm{n}}, \boldsymbol{\sigma}[a]_{\mathrm{b}}, \boldsymbol{\sigma}[a]_{\mathrm{s}}, \boldsymbol{\sigma}[a]_{\mathrm{c}}) \big) \big) \tag{11} $$

Remember, that the world-state is a mapping from account to account states. $L_\mathrm{S}$ function is basically squishing everything belonging to an account state and yields a set of these values for all accounts (i.e. $\forall a$) except non-existent accounts (i.e. $\mathbf{\sigma}[a] \neq \varnothing $).

This set of items, yielded by $L_\mathrm{S}$ function, when put into the Merkle Patricia Trie (using $\mathtt{TRIE}$ function), provides a short identity for the world-state. This identity is nothing but the hash of root of this trie. Keep in mind that this MPT is not the same one as defined for $\mathbf{storageRoot}$ for a single account. This MPT encompasses all the accounts with it’s contents. You might know this as “State Trie”.

It is assumed that account state can either be non-existent or valid if existent. Mathematically, $$ \forall a: \boldsymbol{\sigma}[a] = \varnothing \vee (a \in \mathbb{B}_{20} \wedge v(\boldsymbol{\sigma}[a])) \tag{12} $$

where, $v$ is a validity function which just sets some restrictions on what values an account’s components can take (e.g. nonce and balance must be natural number less than $2^{256}$):

$$ \quad v(x) \equiv x_{\mathrm{n}} \in \mathbb{N}_{256} \wedge x_{\mathrm{b}} \in \mathbb{N}_{256} \wedge x_{\mathrm{s}} \in \mathbb{B}_{32} \wedge x_{\mathrm{c}} \in \mathbb{B}_{32} \tag{13} $$

And that concludes this part of the series! Next part will be all about transactions. Stay tuned!

Resources