Akita PCS Visualizer — a step-by-step visualizer of an Akita commitment and its opening proof

Akita PCS Visualizer

Akita is a lattice-based polynomial commitment scheme: a prover commits to a multilinear polynomial with Ajtai commitments, binding under Module-SIS, and later proves its value at any point. An opening folds the committed blocks with short ring challenges into one response, turns the fold's unknowns into a new witness of digits, checks the fold and the size of every digit with sum-checks, and repeats. This page runs one commitment and its opening end to end at toy size and prints the arithmetic of every step.

Launch the visualizer → Opens in a new tab · runs in the browser · needs 1024 × 600 or wider

Scope. A polynomial of 16 values over Z229[X]/(X4 + 1), opened with the paper's recursive shape at its smallest: one committed fold, then the terminal level, which reveals its response. Nothing at this size is secure.

Four phases, 51 steps
PhaseWhat the page shows
SetupThe ring and why q ≡ 5 (mod 8) makes challenge differences invertible; balanced base-4 digits; the single setup vector S that every matrix reads; the 24 short challenges
CommitThe values packed into ring elements and two blocks, their digits, the inner images ti = A·si, and the commitment u = B·t̂: four numbers
Open, level 0The claim f̃(r) = v; partial evaluations; the challenges, with a nonce retry; the response z; the next witness w of 128 digits; five relation rows reduced to field rows without quotients; the range and relation sum-checks round by round
Open, terminalPartials sent in the clear, four challenges, the revealed response z′ and its three checks
what one fold does
z = c0·s0 + c1·s1  →  w = (ẑ | ê | t̂)  →  w̃(r2) = vW
The response stands in for both blocks in every linear equation. Its digits, the partials and the inner images form the next witness, and two sum-checks reduce everything to one evaluation claim of the original kind.

Sixteen cornerstones carry the protocol, from the ring and the two-tier commitment to the halved range check and the terminal checks. Each is marked where it is established. A spec tray lists them with their purpose, a map shows every object with what it is made from, and a legend explains every colour and cell state.

Four provers
ProverWhat changesOutcome
HonestNothingAccepted
False valueClaims v + 1The scalar row fails; the relation sum-check's first round does not add up; rejected
Digit out of rangeWrites one response coefficient with a digit outside {−2, −1, 0, 1}Every equation still holds; the range sum-check's first round does not add up; rejected
Swaps wOpens a different w at the terminal than the one it boundLevel 0 passes; AT·z′ differs from the bound combination; rejected

Challenges come from a field of 229 elements, so a cheat sometimes passes: in about 1 run in 100 for the false value and 3 in 100 for the bad digit, a random weight lands on 0. The page says so when it happens.

What is real, and what is small
  • Real. The two-tier commitment, partials bound by an opening commitment, nonce retries, the next witness ẑ | ê | t̂, the relation of the paper's Figure 5 checked by transpose convolution, the digit-pairing range check, the fused sum-check and the terminal level.
  • Small on purpose. q = 229 and ring dimension 4, challenges in the base field, 16 values in 2 blocks, a family of 24 challenges, and one fold where the paper's running example has six.
  • Left out. Setup offloading (the setup's share of the last check is computed and shown, not proved), commitment compression, slicing, extension fields, quotient lifting, the Euclidean norm certificate, batching and distributed proving.
  • Visibly small. The proof, 126 values, is longer than the polynomial. Akita's sizes pay off at scale: at 230 values its proofs are 58–68 KB.
How it was checked
  • A separate Python implementation matches the page value for value on 40 seed pairs in all four prover modes: digits, images, transcript draws, round messages and verifier checks.
  • Its test suite runs 9,042 checks, including all 552 challenge differences inverted and 300 random runs in which each cheat is caught at the check the walk names; each of 126 single-value changes to a proof is rejected.
  • A click-driven probe walks every step of all four provers at six screen sizes, and a self-test reruns the main checks in the browser.
Launch the visualizer → Keys: → step · ← back · Enter run · K new polynomial · N new point · V prover · S spec · M map · L legend · ? parameters

Paper: Quang Dao, Omid Bodaghi, Amirhossein Khajehpour, Giuseppe Vitto, Mohammadtaghi Badakhshan, Markos Georghiades, Fengrun Liu, Jiapeng Zhang and Justin Thaler, Akita: A High-Performance Lattice-Based Polynomial Commitment Scheme, ePrint 2026/1983.