PulseCore

Chapter 9 · Section 15

The Changes of 18 November 2026

The day's changes, as they were made; where a later change in the day replaced an earlier one, the earlier entry says so and the last entry of the Part gives the state at the day's end.

The Data Matter given its own particles (18 November 2026, the author: "give the dark matter its own particles, it should always be coming from the math, and we already have the math from the data layer"). The Data Matter had been a prescribed field: laid each step from the first ripples, a·S·δ capped at six times the mean, never moving, never feeling its own pull, so the matter had only shallow troughs and the web stalled (late in the life the densest 7 % of space held 25–29 % of the matter, ours about 60 %). It is now as many collisionless particles as the matter's bodies, seeded at the Physical Nova from the same ripple field and the same Zel'dovich growing mode (zeldovichStart, their own jittered grid), each an equal share of the Data Matter's mass as it returns home; laid cloud-in-cell on the gravity grid each time the forces are read and moved by the same leapfrog, drag and a⁻³, with no stars, no merging, no light (book Part 3.13: collisionless). Measured on the late universe (level 202, headless): at a = 0.51, 0.76 and 0.90 the matter's count contrast 1.34, 1.50, 1.57 (it was 0.75–0.85; the random scatter about 0.43) and the densest 7 % of space held 38 %, 47 % and 48 % of the matter (25–29 % before); the Data Matter itself reached a contrast of 2.28 with 57 % of it in the densest 7 %. Drawn flat, the late universe shows strands of Data Matter and stars between the galaxies, which sit at their crossings, and wide voids: the web. New matter is laid by the Data Matter's density (DUST_WEB 1.5), which now follows the particles.

The Data Matter drawn as nothing (18 November 2026, the author: "draw the dark matter in the game as nothing, it should be completely dark but obvious that it exists"). (Superseded in part later on 18 November 2026: its particles are still never drawn, but the web's density, theirs and the gas's, now glows on the torus; see the last entry of this Part.) The Data Matter's particles are never drawn. Each frame they are laid, on a layer of their own, into a density image a quarter of the screen's size as seen from the camera, a few pixels each and weighed by their distance from the eye (a lens bends light in proportion to its distance from the observer, for sources far behind it; unweighted, the particles beside a close camera filled the view and twisted it into loops); the image is blurred to a smooth field S = d / (d + 0.35), saturating at the densest halos, and the finished frame, copied after it is drawn, is bent by the field's slope: each pixel shows the light from a little nearer the dark matter, the thin-lens equation β = θ − α with α pointing out of the mass, at most about 2 % of the screen's height. Where there is no dark matter the frame is exactly as drawn (checked at zero bend). Measured headless at level 202: no page errors; the frame rate on a real graphics card with the three extra passes is not yet measured.

The galaxies' turn follows the universe's clock (18 November 2026, the author: "the galaxies are spinning to fast, while the whole at the center would be moving fast, the actual galaxy for the universe would spin slower right?"). (Superseded in part later on 18 November 2026: the cap became a turn in 2.5 s ("the galaxies arent spinning enough now"), the centres at most twice the rate at 2R_d, a well's own disk a turn in 2 s; see the last entry of this Part.) Measured at level 202: every galaxy turned at one rate, 67 radians a flip, because the Disk Turn Rate's clock t was frozen when a galaxy's stars first gathered and every galaxy was counted at about flip 2; on a normal computer they all ran at the visual cap, a turn in 2 seconds at twice the scale length, with the centres four times faster, while a well's own disk turned once in 4.2 seconds, slower than the galaxy round it. Now (1) the rate is read at the universe's present age, Ω = √(Δ_v/2)·H/(√2·λ) with H = 2/(3t): disks grow from the inside out, so the outer stars gathered late, and over a life a galaxy turns (135/2π)·ln(T/t₀), about 64 times, the Sun's sixty laps of the Milky Way; (2) each galaxy has its own halo spin λ, log-normal about 0.035 with σ(ln λ) = 0.5 (Bullock et al. 2001), held to 0.015–0.08, so rates differ about threefold (measured: 14.7 to 42.8 radians a flip at the same moment); (3) each star keeps its rate relative to its galaxy's, and the galaxy accumulates its turn in radians at the present rate, so the whole disk, its old stars included, slows as the universe ages; (4) the visual cap is a turn in 4 seconds at twice the scale length (was 2), a faster spinner keeping up to 1.5 times that; (5) a well's own disk turns by the frame's real time (was 0.025 radians a frame), once a second at first and, the same evening, once in two seconds ("some of the wells are still spinning a little much"). The universe's life plays in about 100 seconds, in which strict physics would turn a galaxy about once a second; the cap holds the galaxies slower than that, a visual choice made on the author's call. Not yet seen on a real graphics card.

Smooth between steps, and an alignment pass (18 November 2026, the author: "some of the wells are still spinning a little much but also the steps are stoppy goey would be nice if we could make everything smooth, we could actually probably use to do a full alignment ... also, the filaments are working but there arent enough stars to show them well").

  • Smooth. Every mover between physics steps was drawn at its latest state plus its rate times frac = min(1, time since the step / the last gap): it slid for one gap and stood still until the next step, and the steps come unevenly because the physics is sliced across frames. Now the picture runs one step behind, as a video player buffers. A display clock k advances by the frame's time over the average gap (an 0.8/0.2 running mean), 1.6 times faster when it is behind, and every mover is drawn at state + rate × (k − 1). Past the latest step it runs on along the velocities, easing to a stop a further step on (k ≤ 1.98), so a late step is waited for in motion. When a step lands, k drops by one and the motion continues where it was. The stars have their own clock, stepped back when their new places land (app.js queues setMatter to the next frame), so they never jump back a step in between. Measured headless, where a step came about once a second: before, the median star did not move from one frame to the next; after, it moved every frame, with 9 near-still frames in 119.
  • The wells' spin. A galaxy's inner stars turned up to four times its rate at twice the scale length (solid body inside half a scale length). An exponential disk's rotation curve rises to its peak near 2.2 R_d (Freeman 1970), so the body is solid inside one scale length, at most twice that rate. A well's own disk turns once in two seconds.
  • Stars along the dark filaments. The Data Matter's particles lay out the web, so a field body's round puff is stretched along the filament it sits in. At its block the 3×3-smoothed dark density's Hessian has two axes: the filament runs along the one it curves least. Filamentarity e = (λ₊ − λ₋)/(|λ₊| + |λ₋|) where the density falls across (λ₋ < 0), and 0 in a void or round knot. The puff is drawn (1 + 2e) times longer along and (1 − 0.6e) as wide across. No points are added, since the frame rate on a real graphics card is unmeasured. The bridges to neighbours stay (FIL_SHARE 0.6 of a field body's stars, to its two nearest bodies within FIL_REACH 8 cells).
  • Fixed in the alignment pass.
    • The dark particles now turn and grow with the matter, so the lens no longer drifts off the matter as the frame turns; they glide between steps on the stars' clock.
    • The lens fades in again with each new universe, and its material is freed when it regrows.
    • Dwarf galaxies take their host's spin cap.
    • The infall streams curl with their galaxy's capped turn, limited by the frame's time.
    • Star clusters turn by the summed time of every step that lands in a frame.
    • The sky's background turn is by time: 0.018 rad a second, as 0.0003 a frame did at 60 fps.
    • The matter worker's field lists carry the dark particles, in case the worker is turned on.
    • The prescribed Data Matter field is labelled a fallback.
  • Display constants, written down.
    • Sky: the sky's star room is one point per 24 pixels; the galaxies' room is one per 8 pixels (GAL_DRAW 0.5 of a galaxy's size drawn); the vacuum halo is VOID_HALO 3 times the void's radius.
    • Lens: DARK_INK 0.12, DARK_PT 4 px, DARK_NEAR 0.4 to DARK_FAR 3.0 (the eye-distance weight), DARK_DOWN 4 (a quarter-size density image), DARK_SAT 0.35, DARK_BEND 0.02. The pixel shift is 10 × DARK_BEND × the saturated field's slope per texel, in units of the screen's height: about 2 % at the steepest slopes seen.
    • Turn cap: a turn in four seconds at twice the scale length, a faster spinner up to 1.5 times that.
  • Not yet seen on a real graphics card: the smoothness at real step rates, the filaments by eye, and the frame cost. The sky flooded with stars (18 November 2026, the author: "the filaments still need more stars ... it needs to be just so full of stars ... cant you just flood it with stars to see what happens?"). (Superseded in part later on 18 November 2026: the web's point stars were raised to millions and then brought back to a sprinkle laid by the density (the bias); the web itself is the glow; see the last entry of this Part.) Two causes had kept the sky thin. First, the field's stars were placed by hand on the main thread every frame, which held their count down. Now the graphics card places them (matterMaterial: each star's eight numbers once a step). Second, they were shared by mass, and the mass sits in the knots, so every rise in the count went to the knots. Three changes followed. The sky's room went from a point per 24 pixels to one per 4. The web's own stars were added: up to WEB_PER 24 faint, fine stars round each Data Matter particle, spread WEB_SPREAD 0.45 cells, moving with it on the card. They stand for the faint stars and dwarf galaxies that line real filaments, and because the particles trace the web, the strands fill. The stars shrink with the universe's drawn size while it is young, so a small universe reads as a miniature sky, not a white ball. Measured headless at 1100 × 800, level 202: about 219,000 field, 197,000 web and 144,000 galaxy stars, drawing about 14 ms a frame. Not yet measured on a real graphics card at full HD, where the room is about 520,000.

The web's stars by type, in streaks (18 November 2026, the author: "I am still not seeing variations of stars like dwarfs red giants, blue giants etc, even though the math is there for them"; "they are clumping together in circles too much"). (Superseded in part later on 18 November 2026: the types stand; the streaks are now set by the density (strands streaked, voids and knots round), WEB_ALONG 1.5 and WEB_ACROSS 0.7; see the last entry of this Part.) The web's stars, now most of the drawn sky, had been one warm white, and they swamped the field's stars, which were coloured by type. Now each web star draws its type from the universe's own mix as stars.js counts it (byType): brown dwarfs; M, K, G, F, A, B and O stars (with Population III); giants; white dwarfs; neutron stars. It takes that type's colour (STAR_TYPES, REMNANT_RGB, giants orange-red), and a brightness and size by luminosity, compressed (b ≈ 0.55 + 0.12·log10 L). So the few bright blue O and B stars and the giants stand out over the many faint red dwarfs, as in a photograph of the sky. Neutron stars barely shine. Early in a life only massive stars and their remnants exist, so the web is blue; the yellow and red stars come with time. Each particle's stars were a round cloud, so the web read as circles. They are now spread 2.6 times WEB_SPREAD along the particle's own motion, which runs along its filament, and 0.45 times across it, so they lie in streaks along the strands.

The decay paced by the universe itself (18 November 2026, the author: "it should not be a hard limit, it should be based on bpt physics right?"). The Full decay's clock grew by DECAY_GROW = 2^(1/6) a tick whatever happened, so it swept the early decay in a fraction of a second. A first fix, a fixed stop at 13.8 billion years, was a hard limit and is gone. Now each tick covers the time in which about DECAY_STEP = 1 % of the universe changes. The change is the share of the matter taken into the wells, plus the stars born or dead against those alive, measured over the last tick as a running rate. The pace is held between DECAY_GROW a tick and a DECAY_SLOW = 40th of it; those are playback bounds, not physics. While stars still form and die, the clock moves slowly; in the quiet eras it runs. "Today" is not a law: it is when we happen to look, 13.8 billion years in our universe. It is said, without stopping, only in a universe at our level (⚚♁ = 202), for about four seconds on the decay's card. "The vacuum takes over", the dark-energy turn, comes from the physics at every level.

The web's gas and the dark share (18 November 2026, the author: "split the web layer by the gas fraction but also in the beginning of the universe is there 80% dark matter like we observe?"). (Superseded in part later on 18 November 2026: the dark share stands; the web's gas is drawn by the glow, not as web points; see the last entry of this Part.) Most of the ordinary matter along real filaments is gas, the warm-hot intergalactic medium, which holds about 40 % of all ordinary matter (Nicastro et al. 2018). So each web point is now gas with the universe's own share of its ordinary matter still gas (stars.js gasMass over the matter's mass): a soft, dim violet haze about twice a star's size, lying in the same streaks. The rest are stars by type. Measured live at level 202, the dark share is 84.3 % of all matter at the big bang (a ratio of 5.363, the sky's 5.36 ± 0.06) and 84.7 % at a = 0.044, as some ordinary matter falls into wells; the gas fraction is 97.7 % at a = 0.044. So the strands glow as gas early, and the stars take over as the gas forms them.

The web as a glowing density, the stars by bias, and a cleanup after a three-part review (18 November 2026, the author: "the filaments are just being like lines of stars, it all looks pretty weird actually ... this needs a big update"; "they still arent forming the filaments, is there something wrong with the math"; "lets take a step back and go through everything and make sure that its properly setup").

The web. The web was millions of point stars streaked along the Data Matter's motion: hard lines, and no filament stood out. Now it is drawn as a simulation draws it, as density (setWebGlow). Each step the Data Matter's particles and the field's gas are laid cloud-in-cell on a grid GLOW_SUB times the matter's. The grid is smoothed by three box passes and mapped by the logarithm of density over its mean to a palette: black voids, blue strands, violet and magenta filaments, orange and white knots. It is drawn on a thin skin over the torus.

The web's stars. The web's point stars are laid by that density. A particle keeps min(1, ρ/(WEB_REF ρ̄))^WEB_BIAS of its stars, so below the strands the stars go as ρ², the bias with which galaxies trace the dark matter (Kaiser 1984); a particle inside a drawn galaxy lays none. The first version thinned by crowding, and that was the mistake that made the stars even: it kept every star of a lone void particle and thinned the strands. They are drawn WEB_BATCH to an instance on the graphics card, at most 4 px.

The review. A three-part review followed: the drawing, the physics, and consistency. It found, and this entry records as fixed:

  • pulsars had stopped blinking, the blink left after the graphics card's early return;
  • stars inside galaxies were drawn twice, at the least point size;
  • the web's streaks were inverted by the bias;
  • galaxies past the 1,024th dropped out;
  • the web, the lens and the glow were drawn after the decay;
  • every glow showed through the torus, because a shared material was changed;
  • stars jumped when several steps landed in one frame;
  • wells snapped back between readings;
  • the lens read particles mid-step, and the point layers turned a frame behind the torus;
  • a fixed 8 ms physics budget starved the physics when the graphics card made the frames long. It now takes 40 % of a slow frame's smoothed gap, at most 30 ms, and 8 ms on quick frames;
  • clump wells kept a stale feeding;
  • the quasars' feeding was read against the busiest drawn well. It is now the top of the Eddington-ratio distribution the wells flicker by, ξ ≈ 223 times the mean, for a drawn well standing for N ≫ 1;
  • the decay's pace mixed the matter's share a tick with the stars' share a unit of τ, and counted births twice;
  • the Bondi capture of the Data Matter added mass without taking it from the Data Matter;
  • 'today' was not gated to our level in one place;
  • two number formatters printed 10.00 × 10ⁿ.

Checked live at level 202, a = 0.138: no errors, about 2,500 pulsars blinking, and 58 quasars among the 59 wells feeding.

Left open, for the author:

  • The review computes that the star formation law turns most gas that passes its gate into stars over a life, roughly 50–80 %, against the sky's 6–10 % (Fukugita & Peebles 2004). Each drawn cloud's real mass is the whole ensemble it stands for, so every cloud passes the gate.
  • A disk's turn is set by the present age, where a real disk's is set when it assembled.
  • The decay's spans past the outward expansion are in the game's units, not the real 10⁶⁷-year evaporation of a Sun-mass black hole.

The display constants at the day's end (scene3d.js):

Constant Value Meaning
starRoom, galRoom a point per 6 px each the field's and the galaxies' rooms
GAL_DRAW 0.7 a galaxy drawn at this share of its size
MAX_TURN a turn in 2.5 s at 2R_d a galaxy's turn on screen, a faster spinner up to 1.5×; centres at most 2×; a well's own disk a turn in 2 s
FIL_SHARE, FIL_REACH 0.7, 8 cells a field body's stars on bridges to its neighbours
GLOW_SUB, GLOW_BLUR, GLOW_TOP, GLOW_VOID, GLOW_AMT 2, 2, 18, 0.45, 1.15 the web's glow
WEB_PER, WEB_ROOM, WEB_BATCH 2048, 16, 256 the web's stars: at most this many a particle, within this many skies' rooms, drawn this many an instance
WEB_SPREAD, WEB_ALONG, WEB_ACROSS, WEB_LUM 0.6 cells, 1.5, 0.7, 0.65 their spread, stretch at the strands, brightness
WEB_REF, WEB_BIAS 3, 1 their bias: all kept at 3ρ̄, (ρ/3ρ̄) below
FLASH_GROW, FLASH_GAIN 1.8, 1.7 supernova and merger flashes over the sky
BH_SPREAD_DEX, BH_MAX_SUNS 0.5, 10¹⁰ the biggest member a drawn well shows
DARK_* as above (the lens) unchanged
DECAY_STEP, DECAY_SLOW, DECAY_TODAY_TICKS 0.01, 40, 45 the decay's pace