RECOGNITION
PHYSICS INSTITUTE

Our Science / Universe Formation

If the universe began, it began from balance, not from nothing.

Balance means two sides equal: two different things cost zero to compare only when equal in size, and books balance when every account’s debits equal its credits. Under one premise, the start is the empty books, every account at zero because nothing is written yet; balanced, but not nothing, since the derivation starts from accounts that exist and things that can be told apart. The cheapest change that only adds to the books is one entry in the record: one unit written on one account. Entries only add, so the record never returns, even when the books balance again: counted on the record, time runs the way the record grows.

Film · Universe FormationNarration and captions

The premise. One premise beyond the posting laws gives the books a first moment: the books are the tally of the record of events, so every count on them is a whole number of entries made and none is ever negative. The posting laws allow an endless past just as well; the premise picks the histories that start from the empty books. Other steps adopt premises, rules and readings of their own, each named where it is used.

  • Proved: only ratio one costs nothing; the record never returns, even when every balance does
  • Proved given stated premises: the first event’s price on each finite network; the light-cone geometry of the first record cell
  • Not derived yet: the hot, expanding universe and every cosmological value

Figure 1 · Only balance is freeLive · computed in your browser

A comparison at ratio x costs J(x) = ½(x + 1/x) − 1. J(1) = 0; J(2) = J(1/2) = 1/4; J(1/100) = 49.005. Only ratio one costs nothing. (The live figure needs JavaScript.)

If comparisons combine consistently and the cost curves by one unit at a perfect match, the cost of comparing two different things A and B is J(x) = ½(x + 1/x) − 1, where x = A/B is the ratio of their sizes. It is zero at x = 1 and positive at every other ratio, and it rises without bound as either side shrinks toward zero compared with the other. A void, a world of things that are all the same, hosts no distinction at all. Balance, x = 1, is the only ratio that costs nothing; the empty record, in a world with things to compare but nothing compared yet, costs the same zero. Drag the ratio, or press a button: three nothings (no elements at all, one side shrinking toward zero, the void) and balance (ratio one, the one comparison that adds nothing).
What is live, what is drawn

Live: the curve is drawn in your browser from J(x), and your point on it, the cost and ratio readouts and the two bars are recomputed each time the ratio changes. The ratio runs from 0.000001 to 1,000,000 in steps of a hundredth of a power of ten, so x = 1 is a setting, not an approximation; at powers of ten the readouts are exact decimals. Drawn: the frame, the arrows at the two ends where the curve leaves the chart and keeps rising, and the layout of the bars. The “no elements” and “void” settings draw a definition: there is no ratio to compute in either.

The question

The beginning has been argued from both sides for more than two thousand years.

From Aristotle to the theory of cosmic inflation, the same question kept returning: can reasoning about nature prove that the world began, and what, if anything, came before?

  1. c. 350 BCE

    Aristotle argues that every moment has an earlier one.

    “since the moment is both a beginning and an end, there must always be time on both sides of it.”

    He counts time by change, as this page does, and holds that every now ends a past. That endless past is a history the posting laws allow: each is blind to a common shift of every count and leaves all six entries (a debit or a credit on any one of the three accounts) open from any books; under the premise, the empty record begins the count and ends nothing.

    Aristotle, Physics VIII.1; translated by R. P. Hardie and R. K. Gaye (Oxford, 1930)

  2. 412–426

    Augustine writes that the world was made with time, not in time.

    “then assuredly the world was made, not in time, but simultaneously with time.”

    Counted on the record, nothing comes before the empty record, which matches a world made with time. The match rests on counting time on the record and on the premise stated at the top (examined in step 3, below).

    Augustine, The City of God XI.6 (completed 426); translated by Marcus Dods (1887)

  3. 1260s

    Aquinas holds that no demonstration proves the world began.

    “By faith alone do we hold, and by no demonstration can it be proved, that the world did not always exist.”

    A thing’s nature holds everywhere and always, he argued, so it cannot show that the world began.

    Thomas Aquinas, Summa Theologiae I, q. 46, a. 2; translated by the Fathers of the English Dominican Province (1920)

  4. 1714

    Leibniz asks why there is anything at all.

    “Why does something exist rather than nothing? For ‘nothing’ is simpler and easier than ‘something.’”

    He asks it right after stating that nothing takes place without sufficient reason, a reason that settles why things are so and not otherwise.

    Gottfried Wilhelm Leibniz, Principles of Nature and Grace, section 7 (1714); translated by Robert Latta (1898)

  5. 1781

    Kant sets a proof of a beginning against a proof of none.

    Thesis: “The world has a beginning in time, and is also limited in regard to space.” Antithesis: “The world has no beginning, and no limits in space, but is, in relation both to time and space, infinite.”

    Each comes with its proof. Kant’s verdict is that both proofs take the world to be a whole existing in itself; it is not, so it is neither a finite whole nor an infinite one.

    Immanuel Kant, Critique of Pure Reason (1781), the first antinomy; translated by J. M. D. Meiklejohn (1855)

  6. 1895

    Boltzmann closes a paper with his assistant’s idea: our world inside an eternal equilibrium.

    “We assume that the whole universe is, and rests for ever, in thermal equilibrium.”

    The sentence opens an idea he credits to his old assistant, Dr. Schuetz: in a great enough universe, a world as far from equilibrium as ours is no longer improbable.

    Ludwig Boltzmann, “On Certain Questions of the Theory of Gases,” Nature 51, 413 (1895)

  7. 1928

    Eddington names time’s arrow.

    “I shall use the phrase ‘time’s arrow’ to express this one-way property of time which has no analogue in space.”

    An arrow points to the future, he wrote, if the random element in the state of the world grows along it; physics knows no other distinction.

    Arthur Eddington, The Nature of the Physical World (Cambridge University Press, 1928), p. 69

  8. 1931

    Lemaître counts the quanta backwards, down to a few, or even one.

    “If this suggestion is correct, the beginning of the world happened a little before the beginning of space and time.”

    The number of distinct quanta only grows, he reasoned, so the earliest state has very few, and with so few the notions of space and time fade out.

    Georges Lemaître, “The Beginning of the World from the Point of View of Quantum Theory,” Nature 127, 706 (1931)

  9. 1982

    Vilenkin has a universe tunnel into being.

    “the universe is created by quantum tunneling from literally nothing into a de Sitter space”

    After that the model follows inflation. He wrote that it has no big-bang singularity and needs no initial or boundary conditions.

    Alexander Vilenkin, “Creation of universes from nothing,” Physics Letters B 117, 25 (1982)

  10. 1989

    Penrose estimates how special the Big Bang’s state had to be.

    “the Creator would have to aim for an absurdly tiny volume of the phase space of possible universes”

    His figure puts that volume at about 1/1010123 of the whole.

    Roger Penrose, The Emperor’s New Mind (Oxford University Press, 1989), p. 343

  11. 2003

    Borde, Guth and Vilenkin show inflation can’t reach back forever.

    “inflationary models require physics other than inflation to describe the past boundary of the inflating region of spacetime”

    A path of light or of falling matter along which the average expansion rate is positive can’t be extended back without end.

    Arvind Borde, Alan Guth and Alexander Vilenkin, “Inflationary spacetimes are incomplete in past directions,” Physical Review Letters 90, 151301 (2003)

  12. 2026

    If history began, it began from balance, and no posting law says whether it did.

    A void hosts no distinction and starts nothing; only ratio one costs nothing. Counted on the record, time starts at the empty record, if there is one, and runs the way the record grows. The posting laws are each proved blind to a common shift of every count, so none can tell the books from the same books with every count shifted, and an endless past obeys them all. Aquinas held that reason can demonstrate neither that the world began nor that it always existed. A narrower statement holds as a theorem for the posting laws: they decide neither, and that says nothing about which history happened.

    Recognition Physics Institute, working papers (below)

How it forms

Nine steps from balance to clocks and light, each proved on its own model under the premises it names.

Recognition Science, the theory behind this page, starts from reality as the record of events. Nothing in the record is erased: the past stays written. Each event comes with its reverse; their flows cancel, so the net flow through each thing compared stays zero. The books use three accounts; the steps here take that number as an input. It matches the three dimensions of space, which the foundation derives under its linking conditions; this page takes that derivation as given. An account’s balance is debits minus credits, a difference of two counts, not a position. Step 8 uses a separate cell of four accounts, read alike and with entries counted as time, both by adopted premises; with every history kept causal and positive intervals between records holding the same number of entries, that reading picks out one direction as time. Five words carry the steps.

A comparison
Setting one thing against another, at a ratio x of their sizes. It costs J(x) = ½(x + 1/x) − 1, and it goes in with its reverse, the same comparison read the other way round. One comparison is one recognition event. A comparison at any ratio but 1 tells its two sides apart; at ratio 1 it tells nothing apart and costs nothing. How comparisons enter the record is not derived: reading the first event, entered with its reverse, as exactly one entry is an adopted premise, used in step 5.
An entry
One unit written on one account, on the debit side or the credit side.
The books
The debit and credit counts on three accounts. Each entry adds one to one count; that every count is a number of entries made, never negative, is the premise (examined in step 3).
The record
The list of entries, in the order they were made. Entries are only ever added; nothing in it is erased.
A tick
One step in writing the record: one entry. Under the adopted law of how histories weigh (see the numbers), ticks are counted differently: a tick can pass with no entry. Ticks are counts, not seconds.

Each step is proved on its own exact model, under any premise the step names: the list of entries, the list of comparisons, the books of debit and credit counts, the field around the first entry, the four-dimensional record cell, or the lattice of clocks and light. Joining the models into one continuous history is work still open.

Three nothings, and the start

  1. Set aside

    No elements

    No things at all, so there is no ratio and the cost is undefined. The derivation starts from things that can be compared and sets this case aside.

  2. Not a state

    One side shrinking toward zero

    Compare A with B while A shrinks. Name any cost, and a small enough ratio costs more. That limit is not a state anything can be in.

  3. Starts nothing

    The void

    A world with at least one thing, where no two things differ: everything in it is one and the same. Nothing in it can be told apart, so no first event can change it.

  4. The start, under the premise

    Balanced and unwritten: the empty books

    The accounts exist and things can be told apart, so comparisons can be made. The empty record, with nothing compared yet, costs zero; under the premise, every account is at zero because nothing is written yet. Under the premise, no books come before the empty books; either way, the empty books are not nothing.

None of the three nothings is the start: the shrinking side is not a state, the void starts nothing, and a world with no things lies outside the derivation, which starts from things that can be compared. Under the premise, the start is the empty books: every account at zero because nothing is written yet. Aquinas reads “from nothing” as “after nothing” or as “not from anything”, never as made out of nothing as a material (Summa Theologiae I, q. 45, a. 1, reply 3); the empty record fits the first reading, since nothing comes before it. “Not from nothing” here denies only a start from a state of nothing.

1 · Before the first event

Balance is the one free ratio, and the empty record costs exactly zero.

A comparison sets one thing against another at ratio x and costs J(x) = ½(x + 1/x) − 1. At ratio 2, one side twice the other, it costs J(2) = 1/4. Read the other way round, at ratio 1/2, it costs 1/4 again. At ratio 1/100 it costs 49.005. At ratio 1 the two sides are equal, nothing is told apart, and J(1) = 0.

Two conditions leave one cost, J. First, costs combine consistently: J(xy) + J(x/y) = 2J(x)J(y) + 2J(x) + 2J(y). Second, the cost curves at unit rate near ratio one. No other cost meets both; no continuity is assumed beyond the curvature condition. J is zero only at ratio one and passes every bound as one side shrinks toward zero, so no start sits at that limit. A void, where no two things differ, hosts no distinction at all: nothing in it can be told apart, so no first event can change it. The empty record costs exactly zero: nothing has been compared yet. A comparison at ratio 1 keeps the total at zero, since J(1) = 0; every other comparison adds to it. The void can’t even hold two things that differ; the empty record sits in a world that can, before any comparison has been made. Under the premise, the empty books are the start; step 3 shows that no posting law supplies the premise.

Proved.

Standard mathematics

J ≥ 0 is the inequality of arithmetic and geometric means; the law for combining costs is d’Alembert’s functional equation once 1 is added. Unit curvature is a limit condition at ratio one; together with the law it makes the cost zero there and continuous there, and the classical theory of that equation (Aczél, Kannappan) leaves only J. New: a short direct proof that assumes no continuity beyond the curvature condition.

Working paper: Before the First Event: The Empty Record, sections 2 to 5. Paper page: the price of telling two things apart is forced. Essay: Why Is There Something Rather Than Nothing? Published: Uniqueness of the Canonical Reciprocal Cost, Mathematics 14(6), 935 (2026)

2 · The origin of time

Counted by the record’s entries, time starts at the empty record if the record has one, and no relabelling reads the count backwards.

Start with an empty list and add entries one at a time, each one unit on one account’s debit or credit side. Old entries never change. The length reads 0, 1, 2, 3. Nothing sits below 0, every list can take another entry, and no relabelling of the count reads it backwards. A signed clock is different: a bank balance can be read backwards by flipping every sign, and 3 < 5 becomes −3 > −5.

With the count of entries as its clock, the empty record is an origin of time: nothing comes before it, every record can grow, and no relabelling of the count reads it backwards. No clock that reads signed numbers, whole or real, is an origin: negation reverses it. Another candidate clock reads time from cost. A configuration is a list of ratios, each with a cost; the total is the reading. Under the adopted rule for how a configuration changes, the total falls to its least value and stays there; from then on a tick burns nothing. The configuration is a fuel gauge; the record is the clock.

Proved.

Working paper: Before the First Event: The Empty Record, section 6; The Formation of the Universe from Recognition, section 2

3 · The beginning

The posting laws fix what can happen, not whether history began.

An analogy for the split between the laws and the record: the rules of bookkeeping say how every entry is posted, never whether the books were ever opened; the journal of entries shows that. The books have the same split, and one account is enough to see it.

Take one account. Its book is a pair of counts, debits and credits, and the empty book is (0, 0). Subtract one from both and you get (−1, −1): the same balance, zero; the same two moves out of it, one more debit or one more credit; the same cost for each. It obeys every posting law. Yet no history that starts at (0, 0) ever writes it, because entries only add, and an endless chain of books lies below it: (−2, −2), (−3, −3) and on without end, each of which reaches it.

The posting laws are each proved blind to a common shift of every count, so they can’t tell the empty book from (−1, −1), and an endless past obeys them just as well. If no count is ever negative, the empty books are the unique first moment. No posting law, blind as each is to that shift, can supply this premise, and we prove it; deriving it would take a law that reads absolute counts, or a reason why our books are the tally of a record; neither is derived yet. So the posting laws, all blind to that shift, fix what can happen and which histories are allowed, not whether history had a first moment. No law blind to that shift can pick out the books that start from empty; that is proved for every such law.

Proved: the first moment inside the premise, and that no shift-blind law supplies the premise.

Figure 2 · Run the books backwardLive · computed in your browser

Shift every count by the same number: each posting law is blind to that shift, so it accepts the shifted books exactly as it accepts the unshifted ones. With counts that can be negative, every state of the books has an earlier one, and the steps back never run out. If no count is ever negative, the empty books are the first moment; no posting law supplies that premise. (The live figure needs JavaScript.)

Run the books backward. A step back is not an entry: it shows books that one entry, the one you pick, would turn into the books you see. Switch on the premise, “every count is a number of entries made”: every count starts at zero and grows only by entries posted to it. Now compare the same books with every count shifted by one number c. In the shifted books each entry still adds one to one count, and every balance is the same as in yours, yet the shifted copy of the empty books has every count at c instead of zero.
What is live, what is drawn

Live: the books, every step back, the shift and the comparison list are computed in your browser from the counts each time you act; the switch applies the premise exactly as stated. Drawn: the layout of the tables.

Working paper: Before the First Event: The Empty Record, sections 7 and 8. Essay: Did Time Have a Beginning? works the shift and time-reversal arguments step by step.

4 · The first event

The cheapest change that only adds to the books is one entry, and every comparison that tells two things apart costs more than nothing.

Three accounts, each with a debit side and a credit side, give six places for the first entry. Separately, take the first comparison, say at ratio 2 (any ratio but 1 works the same way): it costs J(2) = 1/4, and it goes in with its reverse, at ratio 1/2, which costs 1/4 too. Together they cost 1/2, and their flows cancel, so each thing’s net flow is zero.

On the books, the cheapest change that only adds is one unit entry; every change that only adds and is more than one entry costs more. On the list of comparisons, the first event goes in with its reverse and costs 2J(r), more than zero at every ratio r except 1. The cost law leaves the ratio free, so J alone does not set the size of the price. All six places cost the same and the posting laws don’t pick: 6 histories after one entry, 36 after two, 1,679,616 after eight.

Proved.

Working papers: The First Recognition Event and Its Price; Every Lawful History, Carried at Once. Place the first entry yourself on the topic page How Did the Universe Begin? Paper page: every event is paired with its reverse.

5 · The price

Given two premises, the first event’s price depends on which network carries its field.

Suppose the first event, with its reverse, is one entry, and an entry is a unit source of a field whose drop between the ends of the entry’s link sets the price: one unit of flow goes in at one end and out at the other, spreading through the network much as current does in a circuit but under a different law. Given these two premises, when the entry’s link has unit strength (weight 1) in a finite network, the price is √5 − 2 if that link lies on no loop and less if it lies on a loop.

Proved, given the two premises. Which network carries the field is not derived.

The example, Figure 3 and the exact statement

Two premises, adopted and not derived: the first event, with its reverse, is one entry; and an entry is a unit source of a field, like a unit charge, whose drop between the entry’s two ends sets the price. If that link has unit strength and lies on no loop, the drop is log φ, giving the ratio φ = 1.618… (the golden ratio) and the price √5 − 2 = 0.236…. On a loop some of the flow goes round, so the price falls: on the 3 × 3 × 3 wrap-around cubic lattice, with a unit link between every pair of nearest neighbours, it lies between 0.025705 and 0.025743.

On every finite network that joins the link’s two ends, the field exists, and every solution gives the same drop between the link’s ends. Given the two premises, on a single loop of unit links the longer the loop, the higher the price, always below √5 − 2. Like an extra road that slows the traffic (Braess’s paradox), strengthening or adding a link can raise the drop. That rise is mathematics of the field and needs neither premise, and given the two premises it raises the price too; with a unit source and the entry’s link of unit strength or more the rise is at most ½ − log φ (about 0.0188) and never lifts the drop above log φ, so, given the two premises, the price never passes √5 − 2. On the record of the first entry alone, its link is a unit link on no loop, so, given the two premises, that network gives √5 − 2; later entries can close a loop through it. Which network carries the field is not derived, so which price the first event pays is not settled.

Figure 3 · The first event’s price on a network you can rewireLive · computed in your browser

On a single link of unit strength that lies on no loop, the entry pays √5 − 2 = 0.236…, given the two premises of step 5; on a loop it pays less, and on the Braess network adding the cross link raises the price. (The live figure needs JavaScript.)

Under the two premises of step 5, one entry puts one unit of source at a and takes it out at b. The field is any set of values for which, at every site, the flow out minus the flow in equals the source there. The entry then pays 2J(ed) = 2 cosh d − 2, the cost of a comparison at ratio ed and its reverse, for the drop d from a to b. When its link (weight 1) lies on no loop, all of the flow crosses it and, given those premises, the entry pays the golden price √5 − 2. When the link lies on a loop, part of the flow goes around, the drop is smaller than log φ and the price is below √5 − 2; on a single loop of unit links, the longer the loop, the higher the price. More links don’t always mean a lower price: on the Braess network (the last button), which lacks the entry’s own link, the price is far above √5 − 2, and adding the cross link raises it further. When the entry has its own link of weight 1 or more, adding or strengthening links can still raise the drop, but by at most ½ − log φ (about 0.0188), and never above log φ.
What is live, what is drawn

Live: every number except the exact closed form shown for the Braess network without its cross link, which matches the value your browser computes. Each time a link changes, your browser solves the field equation on the network and prints the drop, the price, the entry’s own link’s share of the flow, each link’s flow and the change the click caused. The golden case is recognized exactly and printed as √5 − 2; the Braess network without its cross link prints the exact price from its proof, 606,841/3,920 (drop log(784/5), about 5.05). The loop chart solves the single-loop equation for loops of 3 to 16 links. The Braess network uses the weights of the proved example, scaled to one unit of source, which leaves the drop and the price unchanged. Drawn: the positions of the sites and the layout.

The exact statement

Premises: the first event, entered with its reverse, is one entry; an entry is the unit source of a field whose flows obey Gauss’s law at every site (flow out minus flow in equals the source there) and a hyperbolic-sine flow law (a link of weight w carries flow 2w sinh d for a drop d), and its price is set by the field’s drop d between the entry’s two ends: 2J(ed) = 2 cosh d − 2. On every finite network joining the link’s two ends the field exists, and every solution gives the same drop (the field itself is fixed only up to a constant on each connected piece). Make each link of weight w a resistor of conductance w, and let Γ be the conductance between the entry’s ends: then arsinh(1/(2Γ)) ≤ d < 1/(2Γ). With a link of weight one and a unit source, d = log φ exactly when the link lies on no loop (price √5 − 2), and d is smaller on a loop. On a ring of unit links the price rises with the ring’s length toward √5 − 2 and never reaches it; a parallel branch added to a series-parallel network lowers it. Networks don’t rank by inclusion: adding or strengthening a cross link can raise the drop (Braess); with a unit source and the entry’s own link of weight at least one, the rise is at most ½ − log φ (about 0.0188) and never lifts the drop above log φ. A loop of links of weight one or more through the link caps the drop at its value on a ring of unit links of the same length, and such loops that share no link besides the entry’s cap it together, at its value on those loops of unit links alone. On the 3 × 3 × 3 wrap-around cubic lattice, with a unit link between every pair of nearest neighbours, every link lies on a loop and the drop lies between 0.160157 and 0.160274, about a third of log φ.

Standard mathematics monotone networks (Duffin, Minty) give the field and its one set of flows on every finite network joining the link’s ends, and Thomson’s principle gives the linear network’s conductance used in the bracket. Braess found the paradox in traffic networks, and Cohen and Horowitz built electrical and mechanical analogues (1991); in networks of linear conductors strengthening a conductor never raises the resistance (Rayleigh’s monotonicity law), so the rise here comes from the sinh law. New: reading the drop as the first event’s price, the loop caps and the ½ − log φ bound on the rise.

Working paper: The First Recognition Event and Its Price

6 · The first complete pass

With three accounts, the first complete pass takes at least eight ticks.

Ask of each account: is its balance odd? Three yes-or-no answers make 2 × 2 × 2 = 8 patterns, and each entry flips one answer; a complete pass touches all eight patterns and returns home.

Proved.

The example, Figure 4 and the standard mathematics

Give each of the three accounts one yes-or-no: is its balance odd? That makes 2 × 2 × 2 = 8 patterns, the corners of a cube, and each entry flips one answer: one step along one edge. A complete pass touches all eight corners and returns home, the first complete cycle of the parities, so it takes at least eight steps. Exactly 12 eight-step passes leave the empty record, and 3,072 of the 1,679,616 eight-entry histories follow one, about 1 in 547 when each history is counted once; the posting laws themselves set no odds and pick no history.

No complete pass takes fewer than eight ticks. A walk that keeps covering every corner, with each next corner set by the present one alone, repeats every eight ticks and no sooner. First completions happen only at even ticks from eight on: 8, 10, 12, never 9. “At least eight” is the exact claim: the posting laws allow a complete pass at tick 8 and don’t say it happened then. Ticks are counts, not seconds.

Figure 4 · The eight-tick passLive · computed in your browser

Three accounts, one entry every tick: the three parities walk the eight corners of a cube, one edge per tick, and no complete pass takes fewer than eight ticks. Of the 1,679,616 eight-entry histories, 3,072 complete a pass at tick 8. These are counts, not chances. (The live figure needs JavaScript.)

Each entry flips whether one account’s balance is even or odd, so the three parities walk the edges of a cube, one edge per tick. Each tick offers 6 entries, so n ticks give 6n histories, which end in (n + 5 choose 5) different sets of books (the books keep the six totals and forget the order).
What is live, what is drawn

Live: your walk and every readout, the twelve tours (found by search when the figure loads) and every count: 6n and (n + 5 choose 5) in whole numbers, first completions for ticks 1 to 24 by an exact walk over all 2,048 states, balanced histories by an exact sum, and the tick-8 classes by running all 3,072 tour histories. Drawn: the cube’s layout and the positions of its labels.

Standard mathematics

The 12 tours are the directed Hamiltonian cycles of the cube from one corner, the cyclic three-bit Gray codes. New: the exact first-completion counts by tick. Shown here: a walk that keeps covering every corner, its next corner set by the present one alone, has least period 8.

Working paper: The First Complete Eight-Tick Pass; the live counts run on the simulation page. Paper pages: why the cheapest kept act is a loop in three dimensions; the eight-tick register. Published: The Process Dimension of Recognition, Entropy 28(11), 1167 (2026)

7 · The arrow

Counted on the record, time runs one way; balances can come back.

Balances can come back; the record can’t, because every entry lengthens it. Counted on the record, time runs the way the record grows; which way a record grows along a physical motion is open (step 9).

Post a debit on one account, then a credit on the same account. The balances are exactly where they began, but the record holds two more entries, and entries are never removed. The books can balance again at any even tick: 44,730 of the 1,679,616 eight-entry histories (counts, not chances) end balanced, and none does after an odd number of entries. The record never comes back, because each entry lengthens it.

Cost alone can’t point time: a comparison and its reverse cost the same, J(x) = J(1/x). The direction of time sits in the record, which only grows: nothing in it is ever erased, so the past is still written, and with an entry every tick the record tells you which tick it is. The empty record is the only record that nothing precedes.

Proved.

Figure 5 · The record never returnsLive · computed in your browser

Add a debit and then a credit on one account: every balance comes back, and the record holds two more entries. No history ever returns to a record it held before. (The live figure needs JavaScript.)

Three accounts, each with a debit count and a credit count. Here the books start empty at tick 0. Every entry adds one to one count. At every even tick some history has every balance back at zero; after an odd number of entries, no history does.
What is live, what is drawn

Live: the books, the record length, the balances, the strip chart and the counts for all histories of up to 24 entries (6n histories and how many end with every balance back at zero) are computed in your browser in whole numbers; the 0 that return to a record they held earlier is printed with its reason. Drawn: the layout of the chart and tables.

The exact statement

The record never returns to a value it has left. With an entry every tick it fixes the tick; if a tick can pass without an entry, it only bounds the tick. Balance can be reached again at every even tick, so the state alone can’t order moments. A debit followed by a credit restores the balances exactly while the record grows by two. The empty record is the only record with no predecessor. Cost can’t point time: it charges a change and its reverse the same.

Standard mathematics the 44,730 balanced histories are the closed walks of length 8 on the cubic lattice, counted by Pólya. Placing the arrow in records is old (Reichenbach, 1956). New: the exact statements on these books. Shown here: cost alone cannot orient time, since J(x) = J(1/x).

Working papers: Before the First Event: The Empty Record, sections 2 and 7; Every Lawful History, Carried at Once. Essays: Why Your Coffee Never Un-Stirs; The Universe Has No Delete Key. Paper page: the cost of heat has no direction.

8 · The first record cell

Under four conditions on its intervals, two of them adopted, the first record cell has one time and three space directions.

Every reading of the cell’s intervals that treats the four accounts alike and counts entries as time (both adopted), keeps every history inside or on the light cone and is positive between records holding the same number of entries is Lorentzian. No result yet joins cells into a spacetime.

Proved, given a reading that treats the four accounts alike and counts entries as time. Calling single entries light is an identification.

The example, Figure 6 and the exact statement

The record cell is four-dimensional: four accounts, all treated alike, each taking at most one entry. That makes 16 records, each named by which accounts hold an entry, from the empty record to the full one. A history climbs from empty to full one entry at a time, through five records, and there are exactly 24 climbs: four choices for the first entry, three for the next, then two, then one. Any tick clock that reads 0 on the empty record and adds 1 per entry reads the count of entries.

Our modelling premise: we read the cell’s intervals treating the four accounts alike, with the count as time. Among such readings, the ones that keep every history inside or on its light cone (no step outside the cone) and give a positive interval between records holding the same number of entries all have Lorentzian signature: one time direction, along the main diagonal where all four accounts advance together, and three directions of space across it. The cell has four accounts, as many as the three directions of space plus one; four is an input, not derived. No positive-definite reading makes any history causal. Calling single entries light is an identification, not a derivation, and no result yet joins cells into a spacetime.

Figure 6 · The first record cellLive · computed in your browser

Sixteen records, from the empty record 0000 to the full one 1111, and the 24 histories that climb between them one entry at a time. For an interval that treats the four accounts alike (an adopted premise) and is positive between records with the same tick, the form is λ(Σx² − κ(Σx)²) with λ > 0, and every history runs inside or on the cone exactly when κ is at least 1. (The live figure needs JavaScript.)

A record cell has four accounts, each with one entry or none, so it has 16 records, from the empty record to the full one, each named by which accounts hold an entry, in the order A B C D: 1010 means A and C each hold an entry, and B and D hold none. A history adds one entry per step: there are 24 histories, one for each order of the four entries, and a record’s tick is the number of entries it holds. Here x is a displacement between two records: +1 for each account that gains its entry, −1 for each that loses it. Read the cell’s interval without naming any account and require differences between records with the same tick to be space-like: it then takes the form λ(Σx² − κ(Σx)²) with λ > 0; λ only sets the scale, and the figure uses λ = 1. Slide κ. Below ¼ every displacement has a positive interval, so there is no time direction. Above ¼ the interval is negative along the diagonal from 0000 to 1111, where every account gains its entry, and stays positive on every difference between records with the same tick: one time direction and three of space. Every step of every history is a single entry, with interval 1 − κ, so the requirement that every history run inside or on the cone holds exactly when κ is at least 1: on it at κ = 1, inside it above. These requirements do not fix κ: κ = 1 and κ = 2 both meet them. Reading single entries as light is an identification; it fixes κ = 1. The form rests on one adopted premise, not derived: the interval treats the four accounts alike.
What is live, what is drawn

Live: the diagonal and single-entry intervals, the count of histories inside or on the cone and the table’s intervals are computed in your browser from the form with λ = 1 each time κ or the history changes; every slider value is exact. Drawn: the layout of the 16 records, the two thresholds on the bars (κ = ¼ and κ = 1) and the table’s set counts.

The exact statement

Every tick clock that reads 0 at the empty record and rises by 1 per entry equals the number of entries (the rank). Among readings of the cell’s intervals that treat the four accounts alike and count entries as time (both adopted premises), make every entry step causal and are positive on the differences between records holding the same number of entries, the readings are exactly the Lorentzian ones λ(Σx² − κ(Σx)²) with λ > 0 and κ ≥ 1: time along the cell’s main diagonal, space the differences between records holding the same number of entries. Single entries are light-like exactly when κ = 1; that light moves by single entries is an identification. No positive-definite reading makes any history causal. No result yet joins cells into a spacetime.

Standard mathematics the 24 climbs are the maximal chains of the Boolean lattice of subsets of a four-element set (the 16 records). Readings that treat the four accounts alike form a two-parameter family (Schur’s lemma); those positive between records holding the same number of entries are λ(Σx² − κ(Σx)²) with λ > 0. Shown here: among them, the ones that keep histories causal are exactly those with κ ≥ 1.

Working paper: The Formation of the Universe from Recognition (2026), not peer reviewed

9 · A first interacting model

With inputs supplied, every clock reading on the infinite lattice of clocks and light has a well-defined motion.

On the supplied lattice of clocks and light, the motion alone picks no first moment; a record that only grows, empty at the blank (every clock at zero energy, the light field in its vacuum) and counting something whenever a clock is excited, marks one; which way it grows along the lattice’s evolution stays open.

Proved, given the supplied lattice and, for the first moment, that record.

The example and the exact statement

A cubic lattice of cells, each holding four clocks of eight states, shares one field of light. Each clock couples to the light through a unit current along its own edge. Supplied, not derived: a positive coupling strength, the speed of light in lattice units, the clock strength, which edges carry currents, and no background field. Every clock reading and every test of the light field has a well-defined motion on the whole infinite lattice, at every clock strength. Its blank state: every clock holds no energy, its eight readings equally likely, and the light field is in its vacuum.

About the blank state the motion’s past is the exact mirror image of its future, so it picks no first moment and the clocks give no arrow. A record decides: one that never shrinks, is empty at the blank and counts something whenever a clock is excited makes the blank the first moment, at every clock strength, though which way along the lattice’s evolution parameter the record grows stays open. When only the clocks of a finite block drive the light field, the lattice is never again exactly in the blank and no clock of the block is ever again exactly at rest; with every clock on the whole lattice driving it, this is open. With the clocks’ own ticking switched off, for each clock whose direction has a current on every edge (the sample arrangement gives such clocks), the phase relation between that clock’s two halves, the half that drives the light and the half that doesn’t, dies away on long-time average, while each half keeps its chance of one half (the lattice started at the blank); on the whole lattice, with the sample arrangement of currents (all parallel, one on every edge of one direction), the light the clocks give off settles into a steady background that never escapes, while the radiation of a finite block’s clocks alone does leave. No temperature fits the background’s photon counts, so it isn’t a thermal glow. The afterglow astronomers measure is thermal; reaching it is part of the frontier. So even with interaction, this lattice’s motion picks no first moment by itself; a record that only grows, empty at the blank and counting something whenever a clock is excited, marks one.

The exact statement

The lattice: cubic cells, four eight-state clocks per cell, one shared field of transverse light, each clock coupled through a unit current on its own edge; the coupling, the light speed in lattice units, the clock strength and the placement of currents are supplied; no background field. Energy in every finite region is bounded below, and the motion extends to the whole infinite lattice at every clock strength (for clock observables and test displacements of the field). With the clocks’ ticking switched off (clock strength zero) and currents on every edge of a clock’s direction, the motion from the reference state fades that clock’s occupation coherences on long-time average while its occupations keep their values, one half each; on the sample arrangement (all currents parallel, one on every edge of one direction), on the whole lattice, the radiation settles into a steady background that does not escape (the radiation of a finite region’s clocks alone does leave), different by direction, and thermal at no temperature. With the clocks on, dephasing and the long-time state are open. The motion has an exact time reversal that fixes the reference state (on every local observable), in which every clock holds no energy, its eight readings equally likely, and the light field is in its vacuum, so no rule built from the motion and that state alone makes it a beginning. A record that never decreases, is empty at the reference moment and is positive wherever a clock is excited makes that moment the first moment, on every placement and at every clock strength; the direction stays open. With a positive coupling, when only the clocks of a finite region drive the light field, the lattice is never again exactly in the reference state and no clock of the region is ever again exactly at rest; with every clock of the infinite lattice driving it, this is open.

Standard mathematics with the ticking off, each pattern of clock occupations is a fixed source of the light field, the mechanism of van Hove’s exactly solvable model; the motion on the infinite lattice uses locality estimates of Lieb-Robinson type and Bratteli and Robinson’s operator-algebra dynamics.

Working paper: The Formation of the Universe from Recognition (2026), not peer reviewed, with its focused papers on the Zenodo record; the steady background and its photon counts are in focused working papers there, and the time-reversal and record results are in a focused working paper added in the record’s newest version; the no-return results are proved; the written proof is not yet released.

Where it stands

The main results on this page, sorted by status.

Proved

  • Ratio one is the only free ratio; a void, a world of things that are all the same, hosts no distinction.
  • With entries as the clock, time starts at the empty record if the record has one; no record precedes it.
  • No law blind to a common shift of every count supplies the premise, and an endless past obeys every posting law.
  • The cheapest change that only adds to the books is one entry; the first comparison with its reverse costs 2J(r).
  • No complete pass takes fewer than eight ticks; 3,072 of 1,679,616 eight-entry histories make one at tick 8.
  • The record never returns: each entry lengthens it, even when the balances come back.
  • On the lattice of clocks and light, with inputs supplied: every clock reading and every test of the light field has a well-defined motion on the whole infinite lattice; with the ticking off and the lattice started at the blank, each clock whose direction carries a current on every edge dephases, and on the whole lattice with the sample arrangement the clocks’ radiation forms a steady background that isn’t thermal; when only the clocks of a finite block drive the light field, the lattice is never again exactly in the blank.

Proved, given stated premises

  • Given the premise (so no count is ever negative): the empty books are the unique first moment.
  • Given that the first event is one entry and an entry is a unit field source whose drop sets its price: √5 − 2 on a unit link on no loop.
  • Given the adopted law of how histories weigh and the premise: the relative entropy separating the odds of a recognizer’s state (read as its three balances) from its long-run odds is 9 log φ (about 4.33 nats; φ = 1.618…, the golden ratio) at the empty record and falls at every tick under this law (by the H-theorem for Markov chains it can never rise).
  • Given a reading that treats the four accounts alike and counts entries as time: the causal readings that are positive between records holding the same number of entries are exactly those with κ ≥ 1 (between two records, κ weighs the square of the net change in entries against the number of accounts that change), all Lorentzian (one time direction, three of space).
  • Given that the golden weights of the adopted law (1/φ per unit of imbalance) are Boltzmann factors, with ε the energy of one unit of imbalance: one temperature, kBT = ε/log φ (kB is Boltzmann’s constant); ε is not derived, so no temperature value follows, and no comparison with the measured afterglow is made before the derivations listed under “Not derived yet”.

Not derived yet

  • The hot Big Bang and its heating.
  • The expansion of space, and inflation or whatever replaces it.
  • The excess of matter over antimatter, the first nuclei and the cosmic microwave background.
  • Every cosmological value: no cosmological number on this page is derived.
  • The derivation stops before four pieces: the law by which recognizers interact; a law that adds recognizers or distance; a derived space with a light cone across cells; the unit of energy in physical units.
  • Which network carries the first event’s field, and so whether the first entry’s link is a unit link on no loop, where the two premises give the price √5 − 2.

If the universe began, it began from balance, not from nothing. The steps answer a narrower question than Leibniz’s “Why does something exist rather than nothing?”: under the premise, the first moment is the empty books, none of the three nothings. Why there are things to compare at all is where the derivation begins, not something it derives. On the lattice of clocks and light, with the ticking switched off, no state the motion reaches from the blank is thermal at any temperature; astronomy traces the expansion back to a hot, dense early state, and this derivation does not yet reach it.

The numbers

Six numbers.

None of them is a cosmological value. Five count, price or measure something in the steps above; the other, 9 log φ, comes from the adopted law of how histories weigh, under which a tick can pass with no entry.

  • Cost of the empty record

    0

    Nothing written yet and no comparison made: a sum with no terms, zero by definition. A comparison costs nothing only at ratio one.

    Step 1 · The zero by definition; ratio one proved

  • Places for the first entry

    6

    Three accounts, two sides each: 3 × 2 = 6 possible first entries on the empty books. The posting laws don’t pick between them.

    Step 4 · Proved

  • Price on a unit link on no loop

    √5 − 2

    The first event’s price, 0.236…, when its link has unit strength and lies on no loop, given two premises: the event with its reverse is one entry, and an entry is a field’s unit source whose drop sets the price.

    Step 5 · Proved, given the premises

  • Least ticks, complete pass

    8

    Three accounts, a yes-or-no each: 2 × 2 × 2 = 8 corners of a cube. No complete pass, visiting every corner and returning, takes fewer.

    Step 6 · Proved

  • Information at the empty record

    9 log φ

    Under the adopted law, a recognizer’s state is read as its three balances (debits minus credits), which the law moves tick by tick. At the empty record, under the premise, every balance is zero, so that state is certain. Its odds then sit 9 log φ, about 4.33 nats, from the long-run odds, measured in relative entropy (how far one set of odds lies from another; nats are natural units of information; φ is the golden ratio). Under the adopted law the gap falls at every tick (by the H-theorem for Markov chains it can never rise), and the first tick removes 35.7% of it.

    Proved, given the adopted law of how histories weigh: four conditions and three readings (among them, that the state is the three balances), all adopted, not derived; under it a tick can pass with no entry

  • Climbs through the first cell

    24

    Four accounts, 16 records: exactly 24 ways to climb from the empty record to the full one, one entry at a time.

    Step 8 · Proved

Go deeper

The essays take the steps further.

  • The first entry · with the film

    How Did the Universe Begin?

    The topic page and its documentary (8 min 50 s): from Lemaître and Hubble to the afterglow, then the first entry in the record, written in one of six places that no law picks between.

  • The beginning

    Did Time Have a Beginning?

    Why the posting laws don’t fix whether time had a first moment, and what a record that only grows adds, worked on the books and on a lattice of clocks and light.

  • Nothing, sense by sense

    Why Is There Something Rather Than Nothing?

    What people mean by “nothing”, sense by sense, why none of the three senses can start anything, and why the empty record, in a world where things can be told apart, is something rather than nothing.

  • The arrow

    Why Your Coffee Never Un-Stirs

    From Carnot to Eddington: time points the way the record grows.

  • No erasure

    The Universe Has No Delete Key

    Erasing is a merge, and the difference it removes leaves as heat (Landauer’s principle, confirmed in experiment). In the record, nothing written is deleted.

  • Counting without end

    Is the Universe Infinite? Only the Way Counting Is.

    Every record can take one more entry, so no page has to be the last; whether the writing goes on forever is not derived.

  • Gauge and clock

    How Will the Universe End?

    Under the rule adopted for how configurations change, a configuration’s cost falls to its floor and stays there. The record can always take one more line.

The papers

The working papers carry the proofs, except two step 9 results.

One working paper covers steps 1 to 9 at length, with focused working papers for step 9 on its Zenodo record. Five shorter working papers come from the simulation page: four prove steps 1 to 7; the fifth gives the frontier and the information fall. Two results of step 9 hold with a positive coupling when only the clocks of a finite block drive the light field: the lattice is never again exactly in the blank (every clock at zero energy, the light field in its vacuum), and no clock of the block is ever again exactly at rest. Both are proved; their written proof is not yet released. Four published journal papers give the foundations the steps stand on; the working papers are not peer reviewed.

Steps 1 to 9 · working paper, not peer reviewed

The Formation of the Universe from Recognition

The First Record, Time, the First Interacting Dynamics and the Gap to the Hot Universe · Jonathan Washburn · Recognition Physics Institute, 2026

The start as balance, the empty record as the origin of time and the beginning as a premise no shift-blind law supplies; the first event and its price under two adopted premises; the arrow in the record; the first cell; the lattice of clocks and light with its supplied data; and an exact ledger of what stops before the hot universe.

Read the PDFZenodo recordPith record