{
  "status": "research",
  "advantages": [],
  "limitations": [],
  "supersedes": [],
  "supersededBy": [],
  "firstSeen": null,
  "lastSeen": null,
  "updated": "2026-11-17",
  "slug": "lorenzo-predictor",
  "term": "Lorenzo predictor",
  "aliases": [
    "First-order Lorenzo predictor"
  ],
  "definition": "The Lorenzo predictor estimates a regular-grid sample from an alternating sum of previously reconstructed corners of its adjacent hypercube.",
  "note": "Stencil: In three dimensions it uses seven neighbors: add the three face-adjacent values, subtract the three edge-diagonal values, then add the opposite corner. The 2D case is the parallelogram rule. A codec still needs residual encoding and a boundary convention. [lorenzo]\n\nExactness: With exact arithmetic, the n-dimensional stencil reproduces every polynomial of total degree below n. This is sufficient, not necessary: a 3D x^4 field is also exact because it ignores two coordinates. On the unit cube, xyz instead predicts zero at (1,1,1), whose actual value is one. [lorenzo] [probe]\n\nDimensions: Adding a weakly correlated axis can worsen compression. The 2003 fluid-mixing example retained every hundredth time frame and quantized values to one byte; 3D slice coding used 304,937,058 bytes versus 318,871,620 for 4D coding. These are that preprint's results, not binary64 fpzip measurements. [lorenzo]\n\nReconstruction: A lossy predictive codec must synchronize the values used by encoder and decoder. The preprint quantizes residuals and feeds reconstructed values back into prediction to control accumulation. Floating-point arithmetic requires an implementation-specific contract beyond the polynomial theorem. [lorenzo] [api]",
  "related": [
    "fpzip"
  ],
  "citations": [
    {
      "id": "lorenzo",
      "title": "Ibarria et al. (2003): Out-of-core compression and decompression of large n-dimensional scalar fields, sections 3-8",
      "url": "https://digital.library.unt.edu/ark:/67531/metadc1404787/m2/1/high_res_d/15004618.pdf",
      "accessed": "2026-11-17"
    },
    {
      "id": "api",
      "title": "fpzip 1.3.0: API precision range, headers and floating-point portability",
      "url": "https://github.com/LLNL/fpzip/blob/4a539c06d98b1c029b08324a086d4b75689a2b72/include/fpzip.h",
      "accessed": "2026-11-17"
    },
    {
      "id": "probe",
      "title": "Hesela: fpzip precision ledger and exact-arithmetic Lorenzo examples",
      "url": "https://hesela.dev/research/fpzip-precision/README.md",
      "accessed": "2026-11-17"
    }
  ],
  "provenance": [
    "The cited text is the UCRL-JC-151934 preprint submitted to Eurographics 2003, by Lawrence Ibarria, Peter Lindstrom, Jarek Rossignac and Andrzej Szymczak.",
    "Source review covers the stencil, polynomial proof, boundary scan, memory footprint and two 4D evaluations. The original simulation data and compression experiments are not reproduced.",
    "Published examples use exact integer arithmetic and test monomials of total degree below n for n=1 through 4; native fpzip results are a separate experiment."
  ]
}
