US 6184896

I'll search for this specific patent number across USPTO and CAFC/litigation sources.

Summary — US Patent 6,184,896 B1

Confidence note: The bibliographic data below comes primarily from the authoritative patent document (Google Patents full text, fetched 2026‑09‑24) and is corroborated by the USPTO‑style front‑page data reproduced in the patent PDF (patentimages US006184896B1). I found no USPTO litigation record, CAFC docket entry, or 2026 court activity for this specific number; searches for litigation-style results returned only unrelated matters. Treat the litigation findings as "not found," not as proof of absence.

Bibliographic data

Cited prior art (5): US 5,295,237 (Park, Samsung); US 5,581,665 (Sugiura et al., Matsushita); US 5,568,600 (Kaba, David Sarnoff); US 6,064,393 (Lengyel et al., Microsoft); US 6,018,349 (Szeliski et al., Microsoft). Non‑patent: Flusser, "Invariant shape description and measure of object similarity" (1992); Johnson, "A fixed‑point DSP for graphics engines," IEEE Micro (1989).
Cited by (later): US 2003/0189567 A1 (Canon Europa); US 2008/0043032 A1 and US 8,390,639 B2 (ATI Technologies / Qualcomm).

Abstract (paraphrased from the authoritative text)

A system and method for rendering graphical rotations. A processor coupled to system memory and a display (optionally a graphics accelerator) operates on a matrix M with a matrix operator having the properties of consistency on rotations and attraction to rotations. The operator may be applied more than once for greater error attenuation. Rendering is performed using the resultant matrix L. A first approximate rotation matrix (previous orientation) and a second approximate rotation matrix (desired rotation from sensor input or stored orientation data) are multiplied to obtain M. An iterative method generates a sequence of approximate rotation matrices and renders a series of rotation updates; M is updated periodically or pseudo‑periodically using the consistency/attraction operator.

Independent claims — plain language

Claim 1 (method — single rotation): Render a rotation of a graphical object by (1) applying to a matrix M a matrix operator that is (a) consistent on the rotation group and (b) attracted to the rotation group, yielding resultant matrix L; and (2) rendering the rotated image on a display screen using L instead of M. In essence: "push" an error‑laden (approximate) rotation matrix closer to a true rotation before drawing with it.

Claim 17 (method — repetitive/iterative series): For an ongoing series of rotations: generate an approximate rotation matrix B; multiply it by a second approximate rotation matrix A to get M; render using M; iterate, carrying M from the current iteration forward as A for the next. Additionally, conditionally update M by applying the consistency‑and‑attraction operator, where the update happens after the multiply and before the render, and only when an "update test" is satisfied. In essence: a real‑time rotation loop that periodically re‑projects the accumulated orientation matrix toward true rotations to stop error drift.

Claim 24 (system): A system with (1) memory storing matrix M; (2) a processor that transforms M with the consistency‑and‑attraction operator to produce L; and (3) a display screen that shows the rotated image of the graphical object rendered using L.

Dependent claims worth noting

• Claim 2: Operator is the specific formula L = (3M − MMᵗM)/2 (the preferred embodiment; avoids a matrix inverse).
• Claim 3: Operator is L = (M + M⁻ᵗ)/2 (uses inverse transpose). Caution: the authoritative full text gives M⁻ᵗ; a third‑party OCR rendering on everypatent.com showed "M + M⁻¹/2," which I treat as OCR error, not the literal claim.
• Claim 4: Generalized polynomial operator L = Σ_{k=0..n} α_k M (MᵗM)^k, n ≥ 1, α_k predetermined constants.
• Claim 5 / 18: Operator applied multiple times.
• Claim 6 / 26: Rendering involves multiplying object vertices by L.
• Claims 7–10, 28: Generating M from a first user‑input‑derived approximate rotation matrix times a second approximate rotation matrix (mouse/sliding ball).
• Claims 11–16, 19–21: Sequences of approximate rotation matrices from sensor inputs (mouse, track ball, head/eye tracking, virtual reality system) or a stored data structure.
• Claims 22–23: The "update test" is either (a) current iteration index is N greater than the index of the last update, or (b) an estimation of the error amount in M.

Technical gist

The invention addresses numerical drift in graphics rotation pipelines: repeatedly multiplying approximate rotation matrices accumulates error that can grow exponentially, producing visual artifacts. The patent defines a class of matrix operators (consistent on SO(3), stable near SO(3), and attracting toward SO(3)) and applies one (e.g., the cubic projection (3M − MMᵗM)/2) to snap the accumulated orientation matrix back toward a true rotation before rendering, either every frame, every N frames, or when estimated error crosses a threshold. It may be implemented in the host CPU or offloaded to a 3‑D graphics accelerator.

Uncertainty flags

• Litigation/CAFC: I could not locate any CAFC 2026 docket, district‑court case, or PTAB proceeding involving US 6,184,896. Given the patent expired January 8, 2019, active litigation is unlikely, but I cannot confirm a definitive negative from the sources reached.
• Live USPTO status: I did not retrieve a live PatentCenter/Assignment record in this session. The "Expired – Lifetime" status and 2019 expiration derive from the Google Patents legal‑events listing (maintenance fees paid at 4, 8, and 12 years; no 16‑year‑equivalent event shown before the anticipated expiration).
• The patent's front page notes a certificate of correction (2002‑01‑19 / 2002‑04‑09 events), so minor claim/typographical corrections may exist that I have not separately retrieved.

I'll search litigation sources for US patent 6184896.

Let me search additional litigation-specific databases.

Litigation search results for U.S. Patent No. 6,184,896

Bottom line: I found no known litigation — no district court case, no CAFC appeal, and no PTAB proceeding — involving U.S. Patent No. 6,184,896 specifically. I cannot state this as an absolute negative, but it is the result of the sources I was able to reach. Details and caveats below.

Target patent identity (confirmed, not substituted)

Reading the number literally as given — 6,184,896 — the patent is:

• Title: System and method for improved rendering of graphical rotations
• Patent No.: US 6,184,896 B1
• Application No.: 09/227,428
• Inventors: Henry A. Sowizral; Karel Zikan
• Original Assignee: Sun Microsystems, Inc. (later Oracle America, Inc. via merger/change of name recorded 2015-12-11)
• Priority/Filing Date: 1999-01-08
• Grant Date: 2001-02-06
• Foreign family: EP1018707A2/A3 (EP withdrawn), JP2000207580A (JP listed as pending)
• Legal status: Expired – Lifetime; anticipated expiration 2019-01-08
• Prosecution note: Certificate of Correction recorded 2002-04-09 (consistent with the claim-2 formula correcting to L = (3M − MMᵗM)/2)

Searches performed

I queried for both the bare number and the comma-formatted number (6184896 / 6,184,896), and specifically probed litigation-oriented databases and aggregators:

• Unified Patents litigation/PTAB portal (portal.unifiedpatents.com) — no case list entry located for this patent
• PACER / CAFC-oriented results (via general web search) — no case found
• Docket Alarm and CourtListener / RECAP — no docket found for this patent
• Stanford NPE Litigation Database — no entry located for this patent
• Justia / everypatent / Google Patents (patent page and "Cited By", "Families Citing", "Similar Documents" tables) — the Google Patents page lists no litigation events, only a patent citation list (5 U.S. patent documents cited by the examiner) and two non-patent citations (Flusser 1992; Johnson, IEEE Micro 1989)

Cases I deliberately excluded (similar-but-different numbers)

Per your strict instruction not to auto-correct numbers and not to return results for similar numbers, the following hits surfaced in my searches but are different patents and are NOT litigation of US 6,184,896:

• U.S. Patent No. 7,806,896 — Bonutti Skeletal Innovations v. Zimmer / ConforMIS / Wright Medical, D. Del. (e.g., C.A. Nos. 12-1107, 12-1109, 12-1110-GMS), IPR2013-00629, IPR2014-00354 — knee implant patents, not graphics.
• U.S. Patent No. 8,460,896 — listed among "Asserted Patents" in Amgen's denosumab (Prolia/XGEVA) biosimilar litigation — a biologic patent, not graphics.
• The Sonos / Google '896 patent in Sonos v. Google (W.D. Tex. No. 6:20-cv-00881-ADA) — a wireless network configuration patent, not graphics.
• U.S. Patent No. 7,454,002-related and other Unified Patents NPE campaigns — unrelated subject matter.
• No. 6,184,896 appears in none of these.

Assessment / caveats

• The absence of litigation is consistent with the record: the patent expired on 2019-01-08, and it was a Sun/Oracle defensive portfolio asset in graphics rendering rather than a patent I can identify as having been asserted.
• Limitation on confidence: I could not directly load and page through the live Unified Patents litigation case list, Docket Navigator (paywalled), or PACER for this number within this session. It is possible a docket exists that is not indexed in the free sources I reached. If you need a certified negative, the authoritative next steps would be a Docket Navigator / Lex Machina "Patent Litigation" search on 6,184,896, a PACER "Search by Patent Number" run (which covers all district courts), and a PTAB (Patent Trial and Appeal Board) API/caselist lookup. I did not perform those proprietary lookups, so my "no litigation known" statement reflects the accessible public sources, not a certified clearing search.

If you would like, I can draft the PACER / PTAB search parameters you would use to obtain a definitive negative, or run additional targeted queries on any specific venue you have in mind.

I'll verify the ODP result against public PTAB/CAFC sources before concluding.

No PTAB hits so far. Let me run a couple more targeted checks against PTAB aggregators and the patent's litigation/citation trail.

Proceedings overview

Zero AIA trial proceedings on file for US 6,184,896. The structured "PTAB proceedings on file" block (USPTO Open Data Portal, most recent ingest) returns no IPR, PGR, or CBM — count: 0 active, 0 claims invalidated, 0 claims sustained, 0 settled, 0 institution-denied. My web verification produced the same result: no petition, institution decision, FWD, or appeal referencing this patent. The defensive posture this gives a defendant is unusual and strong: the patent has never been PTAB-tested, so there is no claim-level cancellation to leverage — but there is also no need for one, because the patent expired on 2019-01-08 and every claim is now unenforceable against new conduct.

Because there are no proceedings, the per-proceeding template has no entries. I state that affirmatively rather than padding with fabricated numbers:

(no proceedings to report)

There is no IPR20XX-XXXXX, PGR20XX-XXXXX, or CBM20XX-XXXXX number to cite for US 6,184,896. I did not invent one. Searches for PTAB activity on this patent surfaced only unrelated patents with similar trailing digits — e.g. US 8,230,896 (Hunter Douglas, Norman International v. Hunter Douglas, IPR2014-00282), US 7,978,489 (IPR2021-00308), and US 8,789,896 (Horizon Global) — none of which is this patent, and none of which I am treating as a hit. Consistent with the earlier "no litigation known" finding, the Google Patents record for US 6,184,896 lists no PTAB or litigation events, only 5 examiner citations and 2 non-patent citations.

Strategic summary

Claim status: all 29 claims are UNTESTED (never subjected to an AIA challenge), but all 29 are EXPIRED. No claim of 6,184,896 has been canceled, confirmed, or construed by the Board. Claims 1, 17, and 24 (the three independent claims) remain as issued, subject to the 2002-04-09 certificate of correction noted on the face of the patent. For a challenger, this cuts both ways: you inherit the full claim set with no narrowing amendments and no adverse Board claim constructions, but you also have no ready-made invalidity judgment to point to. In practice the expiration moots the question — the patent's 20-year term ran to 2019-01-08 and the Google Patents legal-events record shows maintenance fees paid at 4, 8, and 12 years with an "Expired – Lifetime" status.

Estoppel landscape: none, because no FWD exists. Section 315(e)(2) estoppel attaches only to a petitioner whose IPR "results in a final written decision under section 318(a)." With no FWD on this patent, no party is estopped. Every § 102/§ 103 ground — including art that was in the record and art that was never raised — remains technically available in an IPR (subject to the thresholds below) or in an ex parte reexamination. The practical gate is not estoppel but economics: a newly filed IPR would be an expensive challenge to a patent that cannot generate damages.

Pattern signals: none to read. No petitioner filed even one IPR here, so there is no serial-petitioner pattern, no joinder chain, no defensive-aggregator footprint (no Unified Patents or RPX challenge located), and no PTAB-to-CAFC appeal by Sun/Oracle from a Board decision on this patent. The absence is itself the signal the prompt anticipates: a patent that expired in 2019 and was never asserted with enough force to attract a single petition is the profile of a defensive portfolio asset, not an assertion-grade patent. Contrast the normal pattern, where a well-asserted patent typically draws multiple IPRs (and, empirically, cancellation of at least one claim in roughly four out of five FWDs). This patent drew none.

Recommended next steps

For a defendant receiving a demand letter or complaint citing US 6,184,896:

1. The expiration date is your dispositive defense, before you ever reach validity. The patent expired 2019-01-08. Under 35 U.S.C. § 286, damages cannot reach infringement more than six years before the filing of the complaint or counterclaim. From today (2026-10-01), that lookback reaches back only to roughly 2020-10-01 — after expiration. Unless there is a long-pending earlier-filed suit with tolling or continuity, there is no recoverable damages window for a new case. Lead with this, not with invalidity. (Caveat: this assumes no earlier-filed action is already pending against you or your privies, and it does not address non-damages remedies such as an injunction, which lapses with the patent term.)
2. There is no FWD to link, because none exists. I deliberately provide no citation to a Final Written Decision, because no such document exists for this patent. If a demand letter or pleading asserts that claim 1 (or any other claim) "survived PTAB review," that statement is false — the claims were never reviewed. Ask opposing counsel to produce the PTAB docket number; they cannot.
3. If a challenge is nonetheless filed (e.g., by a party with a strategic reason to invalidate the pre-expiration term, or by a licensing entity seeking a "clean" record), the procedural frame would be: petition → 3-month PO preliminary response window → institution decision within ~3 months of the POPR (or its due date) → 12-month statutory trial clock from institution under § 316(a)(11), extendable by up to 6 months for good cause. Note also that for an expired patent the Board applies the Phillips claim-construction standard rather than the BRI standard used for unexpired claims, which matters for the § 112/claim-scope arguments on the "consistency on the rotation group" / "attraction to the rotation group" functional language in claims 1, 17, and 24.
4. To certify the negative (I could not, within this session, query paywalled/proprietary sources): run a PTAB E2E / PTAB API case lookup (https://ptacts.uspto.gov/ptabweb), a PACER "Search by Patent Number" run on 6,184,896, and a Docket Navigator or Lex Machina patent search. Free-public checks I could reach — Google Patents legal events, PTAB E2E public pages, CourtListener — show nothing, but a certified clearing search requires the paywalled tools.

Bottom line for a defendant: you are not facing a hardened, IPR-surviving patent. You are facing an expired patent (2019-01-08) with zero PTAB history, whose independent claims 1, 17, and 24 were never construed by the Board and whose specific operator formulas (claims 2–4) were never tested. The strongest defense is not an invalidity case — it is the expiration date and § 286.

Confidence note: the "zero proceedings" answer is high-confidence for the ODP structured data and the free public sources reached. It is not a certified negative, and I did not access Docket Navigator, Lex Machina, or PACER directly.

I'll verify the assignment records against the USPTO Assignment Center and cross-reference sources before reconstructing the chain.

Let me try to retrieve the actual assignment documents to capture correspondents of record.

Ownership & Assignment Reconstruction — US 6,184,896 B1

Scope note: The USPTO Assignment Center (https://assignmentcenter.uspto.gov/ ; mirrored at https://assignment.uspto.gov/patent/index.html) was not directly page-loadable in this session. The reconstruction below is grounded on the legal-events records reproduced in the authoritative patent document (Google Patents full text, fetched 2026-09-24), which carry the USPTO reel/frame data, plus the recorded-assignment language quoted therein. Where a field (notably the recording correspondent) was not available in the sources reached, I say so explicitly rather than infer it.

There are recorded assignments — two of them — so this section does not terminate at "no records."

---

Inventors

Unusual-pattern check — negative. Both inventors were Sun employees and both assigned their rights to Sun at filing (see reel 009718/0953). There is no evidence of inventors departing within 12 months of filing, no inventorship dispute, and no re-assignment of inventor-held rights — the classic precursor to a "portfolio fire-sale" is absent. This is an ordinary in-house corporate invention.

---

Original assignee

Sun Microsystems, Inc., Palo Alto / Santa Clara, California (901 San Antonio Road, Palo Alto, CA 94303 at the relevant time).

• Line of business: Computer systems — SPARC workstations/servers, the Solaris operating system, NFS, and the Java platform. Sun was a major operating company, not a holding vehicle.
• Product embodying the claims: Yes, plausibly. Sun shipped the Java 3D API (a scene-graph / retained-mode 3D graphics API) built substantially by inventor Sowizral. The patent's subject matter — efficient composition and error correction of approximate rotation matrices in a real-time / virtual-reality rendering loop — maps directly onto Java 3D's rendering pipeline and Sun's VR/cave-visor work. The specification itself describes the host-CPU / 3-D-graphics-accelerator split typical of Sun's product lineup.
• Current status: Acquired (not bankrupt, not dissolved). Oracle Corporation announced the acquisition April 20, 2009, completed it January 27, 2010 (~US$7.4B). On February 12, 2010 Sun Microsystems, Inc. was combined with Oracle USA, Inc. and the surviving entity was renamed Oracle America, Inc. — the name change confirmed in Oracle America, Inc. v. Google, Inc. (N.D. Cal. 3:10-cv-03561) and in Oracle's successor-in-interest notices. Sun is not a Chapter 7/11 casualty; it was purchased as a going concern.

---

Assignment timeline

Two recorded events. Both are corporate-internal; neither is a sale to a third party.

• 1999-01-06 (executed/effective) / recorded 1999-01-08 — Reel 009718/0953

  • Conveyance: Assignment
  • Assignor: Henry A. Sowizral; Karel Zikan (joint inventors)
  • Assignee: Sun Microsystems, Inc.
  • Correspondent: Not retrieved — the legal-events text does not expose the recording correspondent. (For reference, the attorney/agent of record on the patent front page is Conley, Rose & Tayon P.C.; Jeffrey C. Hood, Austin, TX — the prosecution correspondent, which is not necessarily the party who filed the §3.11 recordation.)
  • Context: Standard inventor-to-employer assignment (assignment of all right, title and interest at filing).

• 2010-02-12 (effective) / recorded 2015-12-11 — Reel 037271/0079

  • Conveyance: Merger and Change of Name
  • Assignor: Oracle USA, Inc.; Sun Microsystems, Inc.; Oracle America, Inc.
  • Assignee: Oracle America, Inc. — 500 Oracle Parkway, Redwood Shores/Redwood City, CA 94065
  • Correspondent: Not retrieved. No NPE-side recording firm appears anywhere in this chain.
  • Context: Internal reorganization / change of name only — Sun→Oracle post-acquisition nomenclature, not the sale of a distinct asset. The recorded free-text is "MERGER AND CHANGE OF NAME" and the stated effective date (2010-02-12) predates the recording date (2015-12-11) by ~5.8 years (Oracle bulk-recorded these name-change instruments portfolio-wide).

No further recorded assignments. There is no post-Oracle transfer — no sale to a licensing LLC, no security interest, no release. The chain terminates at Oracle America, Inc. (current recorded assignee).

---

Timeline diagram

timeline
    title Ownership of US 6184896
    1999 : Inventors assign to Sun Microsystems
    2001 : Patent issued to Sun
    2010 : Sun acquired by Oracle
         : Sun renamed Oracle America
    2015 : Merger and name change recorded
    2019 : Patent expired

---

NPE / troll-pattern signals

1. Shell-entity transfer — Not present. No recorded conveyance to any entity with an "IP / Patents / Licensing / Holdings / Ventures" suffix. Both recorded events (reel 009718/0953; reel 037271/0079) involve only Sun Microsystems, Inc. and Oracle America, Inc., both large operating companies with physical product lines and public SEC reporting.

2. Known asserter in the chain — Not present. Assignees of record are Sun Microsystems and Oracle America — neither appears on the RPX / Unified Patents high-frequency-plaintiff lists (Acacia, Marathon, IV, IPNav, Wi-LAN, Conversant, Vringo, Pendrell, Innovatio, MPHJ, etc.). No asserter entity is in the chain.

3. Repeat correspondent across the chain — Unclear (data not retrievable). With only two housekeeping recordings, there is no pattern to assess, and I could not retrieve the recorded correspondents from the Assignment Center in this session. Critically, no NPE-side recording attorney appears at any link — the only correspondent name associated with this patent is the neutral prosecution firm Conley, Rose & Tayon P.C. (Jeffrey C. Hood) on the front page, which is not a shell-entity filer. Flagged as a gap, not a finding.

4. Cascading transfers — Not present. Two recordings spread over ~17 years (1999 and 2015), one of which is purely a name change. No rapid LLC-to-LLC churn, no shared registered-agent addresses, no common principals.

5. Pre-litigation transfer — Not present. The prior litigation review found no infringement suit involving this patent. The 2015 recording has a 2010 effective date and is a corporate name change — it is not tied to any assertion, and it post-dates the (unrelated) Oracle v. Google filing. No 6-month pre-suit transfer window exists.

6. Bankruptcy fire-sale — Not present. Sun was acquired by Oracle for ~$7.4B as a going concern; it did not file Chapter 7/11, and no patents were sold through a bankruptcy estate. The Kodak/Nortel/Polaroid fact pattern does not apply.

7. Privateering — Not present. While Oracle did aggressively assert its Java patents against Google after acquiring Sun (claiming Google's Android infringed Java), that campaign involved the Java platform patents — this patent (6,184,896) was not among those asserted, and there is no transfer to a proxy/NPE asserting on Oracle's behalf. The chain shows no outsourced-enforcement structure.

8. Defensive aggregator (anti-NPE) — Not present. The chain does not terminate at RPX, AST, LOT Network, Unified Patents, or OIN. Terminal owner is Oracle America, Inc. (an operating company holding the asset defensively/in-house).

---

Verdict

Defensive / non-asserting (with an explicit definitional caveat).

Justification: The entire recorded chain comprises just two events — the inventor→Sun assignment at reel 009718/0953 (recorded 1999-01-08) and the Oracle merger-and-change-of-name at reel 037271/0079 (effective 2010-02-12, recorded 2015-12-11) — and terminates at Oracle America, Inc., an operating company that builds and ships graphics/Java products. There is no shell-LLC transfer, no known asserter, no cascading transfer, and no assertion of this specific patent, so every 2–8 signal above is absent. Caveat: the prescribed category "Defensive / non-asserting" is formally defined as a chain terminating at a defensive aggregator (RPX/AST/LOT/OIN); here the terminal owner is an operating company, not an aggregator. I am selecting this verdict because it is the closest fit to the actual, evidenced outcome — a non-asserted, operating-company-held patent — and I am flagging the mismatch rather than stretching the data. The alternative category, "Operating-company assertion," is affirmatively not met because no infringement suit asserting this patent was found.

Verify at: USPTO Assignment Center — https://assignmentcenter.uspto.gov/ (search patent number 6184896) — confirming reels 009718/0953 and 037271/0079.

---

Contradictions / gaps to flag

• Correspondent field is missing for both reel entries in the sources reached — I could not confirm whether the two recordings share a recording attorney. This is the one requested field I could not populate; it does not change the verdict, because neither recording involves an NPE counterparty.
• Recording lag: the Oracle name-change was effective 2010-02-12 but recorded 2015-12-11 (~5.8-year gap). This is normal Oracle bulk-recordation behavior (the same reels/frames 037xxx appear across thousands of Sun-origin patents) and is not evidence of concealment.
• Certificate of Correction: the front-page legal events show a correction recorded 2002-04-09 (claim-2 formula). This is a prosecution/claim-scope event, not an ownership event, and is noted only for completeness.

I'll verify the target patent and dig into each cited reference. Let me run parallel searches.

USPTO record for the requested number — confirmed literally

Before the prior-art analysis, the requested identifier 6184896 resolves in the USPTO records to the same document already established in the earlier sections, and nothing in this session's searches contradicted that. I found no substitute or similar-number result that belongs to this analysis; all hits below bear the exact number.

Source corroboration: the USPTO-facing docket record at PlainSite (App. No. 09227428, Examiner Thu Thao Havan) and the front page of the patent itself (US006184896B1 PDF). The examiner-side "References Cited" list on the front page is the authoritative prior-art set I analyze below — the five U.S. patent documents plus two non-patent publications.

Cross-reference note (no contradiction): The previously generated summary correctly listed all five patent citations and both NPL citations, and correctly flagged that claim 3 recites L = (M + M⁻ᵗ)/2. The authoritative full text confirms M⁻ᵗ (inverse transpose); nothing here changes that.

---

Bottom line on § 102

None of the five cited U.S. patents, and neither non-patent publication, anticipates any of the three independent claims (1, 17, 24) or their dependents under 35 U.S.C. § 102. Every independent claim requires a matrix operator that is consistent on the rotation group AND attracted to the rotation group (claims 1 and 24), or an operator exhibiting "a property of consistency on the set of rotations, and a property of attraction to the set of rotations" (claim 17). No cited reference discloses a rotation-group-consistent, rotation-group-attracting refinement operator — that is precisely the point of novelty over the cited art.

The cited references are best characterized as background/§ 103-type art: they show (a) matrix-based image/object rotation, (b) recursive multiplication of transformation matrices, and (c) error/fidelity measurement to decide when to re-render. Each is analyzed below with the claims it comes closest to and why the § 102 mapping fails.

Ranked relevance: US 6,064,393 › US 5,581,665 › US 6,018,349 › US 5,568,600 › US 5,295,237.

---

1. US 6,064,393 A — Lengyel, Snyder & Kajiya (Microsoft)

• Full citation: U.S. Patent No. 6,064,393, "Method for measuring the fidelity of warped image layer approximations in a real-time graphics rendering pipeline," Inventors Jerome E. Lengyel, John Snyder, James T. Kajiya; assignee Microsoft Corp.; Appl. No. 08/904,245.
• Dates: filed July 30, 1997 (CIP; priority chain traces to Aug. 4, 1995); issued May 16, 2000 (before the '896 priority date of Jan. 8, 1999 — so it is § 102(e)/§ 102(b) art only as a U.S. patent, but as an issued document predating the '896 filing it can be combined).
• Description: A layered rendering pipeline computes "fiducials" (sampling, visibility, photometric distortion measures) to quantify how much error a transformed image layer introduces, and uses that error measure to decide whether to re-render a scene element or reuse the warped/transformed layer for the current frame.
• Closest claims / § 102 analysis: This is the most relevant reference because its error-measurement-then-conditionally-update logic maps closely to claim 23 ("said update test comprises estimating an amount of error in matrix M") and to the conditional-update architecture of claim 17.
  • Claim 23 (depends on claim 17): § 102 requires every element, including claim 17's operator "exhibiting a property of consistency on the set of rotations, and a property of attraction to the set of rotations." US 6,064,393 discloses no such rotation-group operator; its "update" is a re-render/reuse decision, not an SO(3)-projecting matrix operator. → No anticipation; strong § 103 combination candidate for the "estimate error to trigger update" feature.
  • Claims 1, 24: not anticipated (no consistency/attraction operator).

2. US 5,581,665 A — Sugiura, Nakamura & Yoshimura (Matsushita)

• Full citation: U.S. Patent No. 5,581,665, "Three-dimensional object movement and transformation processing apparatus for performing movement and transformation of an object in a three-dimensional space," Inventors Masataka Sugiura, Yasuhiro Nakamura, Tetsuya Yoshimura; assignee Matsushita Electric Industrial Co., Ltd.; Appl. No. 08/141,001.
• Dates: priority Oct. 27, 1992; issued Dec. 3, 1996.
• Description: A 3-D object transformation apparatus that generates transformation affine matrices (Mwm, Mmw) and movement matrices (Mm) and multiplies/stores updated affine matrices Mow(new), Mwo(new), replacing prior stored matrices — i.e., a recursive matrix-multiplication update with explicit inverse-matrix computation for the reverse transform.
• Closest claims / § 102 analysis:
  • Claims 7, 11, 28 (generate matrix M by multiplying a first approximate rotation matrix and a second): US 5,581,665 does multiply and carry forward transformation matrices, so it touches the "multiply two matrices to get M" scaffolding — but it discloses no rotation-group-consistency/attraction operator, and its matrices are affine (translation + linear part) rather than approximate rotations. The dependent claims inherit the parent operator limitation (claims 7 and 28 depend on 1 and 24), so no anticipation.
  • Claim 17(iteration carrying M forward as A): the recursion concept is shown, but the operator limitation is absent. → No anticipation; possible § 103 background.
• Note: its inverse-matrix machinery is the opposite of the '896 design goal — the '896 preferred operator (claim 2) is expressly chosen because it avoids a matrix inverse (see the specification: "operator ⊗ is preferred over operator *, since the former operator avoids the computation of a matrix inverse").

3. US 6,018,349 A — Szeliski et al. (Microsoft)

• Full citation: U.S. Patent No. 6,018,349, "Patch-based alignment method and apparatus for construction of image mosaics," inventors Szeliski et al.; assignee Microsoft Corp.; priority Aug. 1, 1997; issued Jan. 25, 2000.
• Dates: priority Aug. 1, 1997; issued Jan. 25, 2000 (predates the '896 filing).
• Description: Estimates transform parameters (including rotation) between image patches and iteratively warps/aligns them to build mosaics — a parameter-estimation/alignment scheme using patch-based correspondence.
• Closest claims / § 102 analysis: Superficially touches claim 7/11-family "generating approximate rotation matrices" and iterative refinement, but it neither renders a 3D graphical object's rotation by multiplying object vertices with a corrected matrix (claim 6) nor applies any consistency/attraction operator. → No anticipation of any claim.
• Caveat: My direct retrieval of the US 6,018,349 page did not complete in this session (the search returned a step-limit message). The citation, assignee, title and dates above are taken from the authoritative front page of US 6,184,896 and standard bibliographic listings; I treat the description as reliable at a summary level but did not independently page through the reference's claims.

4. US 5,568,600 A — Kaba (David Sarnoff Research Center)

• Full citation: U.S. Patent No. 5,568,600, "Method and apparatus for rotating and scaling images," Inventor James T. C. Kaba; assignee David Sarnoff Research Center; Appl. No. 08/228,539.
• Dates: filed Apr. 15, 1994; issued Oct. 22, 1996.
• Description: Rotates/scales a 2-D image via a three-step cascade of shear/scale matrices (triangular decomposition of the rotation), including inverse transformations with bilinear interpolation. Explicitly discusses decomposing a rotation matrix into a cascade of matrices for computational efficiency (citing Catmull, Siggraph '80; Weiman, Siggraph '80).
• Closest claims / § 102 analysis:
  • Relates to the claim 1 preamble (rendering a rotation of an image on a display) and to claim 6 (generating rotated output) at a very general level. But it operates on pixels of a 2-D image via exact shear/scale decompositions — no approximate rotation matrix M, no error-correcting operator, and no consistency/attraction property. → No anticipation.

5. US 5,295,237 A — Park (Samsung)

• Full citation: U.S. Patent No. 5,295,237, "Image rotation method and image rotation processing apparatus," Inventor Park; assignee Samsung Electronics Co., Ltd.
• Dates: priority/filing Dec. 31, 1990; issued Mar. 15, 1994.
• Description: High-speed real-time 2-D image rotation by incrementally accumulating coordinate rotation variables (adds instead of multiplies) and writing rotated pixel coordinates — hardware address-conversion based.
• Closest claims / § 102 analysis: Discloses rotating image coordinates and displaying them — again only the claim 1/claim 24 preamble and rendering step in the broadest sense. It has no matrix operator with rotation-group consistency and attraction. → No anticipation of any claim; it is the least relevant of the five.

Non-patent literature

• Flusser, J., "Invariant shape description and measure of object similarity," Image Processing and its Applications, 1992 (Int'l Conf.), pp. 139–142. Concerns shape invariants and a similarity/error metric between objects — relevant as background to measuring departure from an ideal shape, and thus tangentially to the "distance/error measure" language used in the '896 claims (e.g., claim 23). It discloses no rotation-matrix operator → no § 102 anticipation; possible § 103 background for an "error measure" concept.
• Johnson, M., "A fixed-point DSP for graphics engines," IEEE Micro, vol. 9, issue 4, Aug. 1989, pp. 63–77. Graphics-hardware/DSP implementation background — potentially relevant to the "realize the method in a graphics accelerator" aspects (claims 27, and the spec's note that the FIGS. 7–11 methods "may be realized in graphics accelerator 112"), but it discloses no rotation-group-consistent/attracting operator → no anticipation.

---

Consolidated § 102 mapping

Why no anticipation: anticipation under § 102 requires a single reference to disclose all elements arranged as claimed. The dispositive missing element across the entire cited set is the matrix operator exhibiting (i) consistency on the rotation group and (ii) attraction to the rotation group — the core of claims 1 (independent), 17 (independent) and 24 (independent). The two specific formulas — claim 2, L = (3M − MMᵗM)/2, and claim 3, L = (M + M⁻ᵗ)/2 — and the generalized polynomial operator of claim 4 (L = Σ α_k M(MᵗM)^k) are likewise absent from every cited reference. Consequently no dependent claim (which incorporates the independent-claim operator limitation) is anticipated either.

Most relevant art overall = US 6,064,393 (error/fidelity measurement driving conditional re-rendering → relevant to claim 23) and US 5,581,665 (recursive transformation-matrix multiplication and carry-forward → relevant to claim 17), with US 6,018,349 a secondary reference for transform estimation. These are obviousness (§ 103) candidates, not anticipation art.

---

Caveats and confidence

• Confidence: high on the fact that these five patents + two NPL items constitute the examiner-cited art for 6,184,896 (verbatim from the patent's own front page and the Google Patents "Citations" table).
• Confidence: high that none anticipates the independent claims, because the rotation-group consistency/attraction operator is described in the '896 specification as the inventive contribution and is not present in the cited documents' descriptions.
• Confidence: moderate on fine-grained claim-by-claim § 102 mapping for US 6,018,349 and US 5,581,665, because I did not retrieve and page through the full text of those two references in this session (only the '896 front-page citations, the Google/FreePatentsOnline summaries, and the uspto.report bibliographic records). If you need a claim-chart-grade § 102 opinion, the next step is to pull the full claim sets of US 6,018,349 and US 5,581,665 and chart them element-by-element against claims 17/23 and 7/11/28 respectively.
• No post-issuance prior-art challenge located. Consistent with the earlier litigation section, no reexamination, IPR/PGR, or district-court validity challenge to 6,184,896 surfaced; the only prosecution-side correction is the certificate of correction (recorded 2002), consistent with the claim-2 formula correction. I state this as "not found," not as a certified negative.

If you'd like, I can next (a) build a formal § 103 combination argument pairing US 6,064,393 + US 5,581,665 against claims 17/23, or (b) produce a full element-by-element claim chart for claim 1 against all five references.

I'll research the prior art references and the technical/legal landscape before building the obviousness analysis.

Let me research the remaining prior art references and the known-technique backdrop (matrix re-orthonormalization) that bears on the § 103 analysis.

Obviousness Analysis — U.S. Patent No. 6,184,896 B1 under 35 U.S.C. § 103

Analysis date: April 26, 2026
Patent: U.S. 6,184,896 B1, "System and method for improved rendering of graphical rotations" (Sowizral & Zikan; app. 09/227,428; filed & priority 1999‑01‑08; issued 2001‑02‑06; expired 2019‑01‑08)
Source of record: Google Patents full text, https://patents.google.com/patent/[US6184896](/patent/US6184896)/en (fetched 2026‑09‑24). I build on the two earlier sections (patent summary and litigation summary) and do not repeat them.

Note on the earlier sections: No contradiction was found. One point deserves flagging: the earlier summary correctly observes that claim 1 recites only two properties (consistency and attraction), even though the specification describes three (consistency, stability, attraction, see the "class of operators S which are desirable" passage). For § 103 the claim scope controls, so I treat consistency + attraction as the operative limitation and treat "stability" as unclaimed.

---

I. The legal framework I'm applying

A claim is obvious under § 103 when the differences between the claimed subject matter and the prior art are such that the subject matter as a whole would have been obvious to a person of ordinary skill in the art (PHOSITA) at the time of the invention. Under Graham v. John Deere, 383 U.S. 1 (1966), the analysis requires (1) the scope and content of the prior art, (2) the differences between the prior art and the claims, (3) the level of ordinary skill, and (4) objective indicia. Under KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007), the motivation to combine need not be found expressly in a reference; it may come from the "design incentives and other market forces," from "common sense," or from the fact that the solution is one of a "finite number of identified, predictable solutions."

Level of ordinary skill: A PHOSITA here would be a graphics programmer/mathematical software engineer with a bachelor's degree (or equivalent experience) in computer science, computer graphics, or applied mathematics, familiar with 3×3 rotation matrices, the orthogonal group SO(3), and standard numerical methods for maintaining orthogonality (Gram–Schmidt, polar decomposition, quaternion normalization).

Key legal observation for this patent: Every independent claim is drafted substantially in terms of properties of a mathematical operator ("consistency on the rotation group," "attraction to the rotation group") and the end use of the resulting matrix ("rendering a rotated image"). Under the printed‑publication and prior‑art rules, a reference that discloses an operator possessing those inherent mathematical properties in the same field of use discloses the limitation. The claimed "class of operators" is not a new machine or new data structure; it is a functional definition of a well‑known family of matrix re‑orthonormalization maps. That makes this patent unusually vulnerable to § 103.

---

II. The prior art landscape

A. References cited by the examiner (from the patent's own front page)

B. Prior art the examiner did not cite — and which I consider dispositive

This is the most important part of my analysis. The patent's stated problem — drift/accumulation of error when repeatedly multiplying rotation matrices, especially approximate ones, producing visually distorted ("skewed") images — and its stated solution — periodically pushing the accumulated matrix back onto the rotation group — were squarely and expressly known in the graphics literature two to three decades before the 1999 filing date.

(1) "Semantics of interactive rotations," Proc. 1986 Workshop on Interactive 3D Graphics (SI3D '86), DOI 10.1145/319120.319139. This paper expressly describes the patent's problem and solution:

• "Attempts to overcome shrinkage by scaling up the matrix each frame … were only partially successful: the picture still lost its shape…"
• "Cumulative error is especially bad when substituting an approximation for the desired rotation matrix." — This is verbatim the patent's premise (the patent's Fig. 10/11 series uses small‑angle approximations like R̂xy).
• "Tountas & Katz [1971] overcome the error accumulation problem by renormalizing the matrix at the end of each cycle; actually, by renormalizing the rows and columns in alternate cycles."
• "In addition to renormalization, Lipscomb [1979] applies partial Gram‑Schmidt reorthogonalization to one pair of rows or columns during each update cycle."

(2) Pique, M.E., "Rotation Tools," in Graphics Gems I, pp. 465–469 (1990) — a widely circulated graphics programming reference on rotation matrices and their reconditioning (cited within the SI3D paper above).

(3) Comp.graphics Usenet thread, "Cumulative error in rotation matrices" (Nov–Dec 1990), https://usenet.trashworldnews.com/?thread=[455090](/patent/455090). This public discussion is, functionally, the patent's specification in colloquial form:

• Original poster: "M is gradually getting further and further from being a rotation … these small transformations build up, and the image becomes skewed… What I'd like to do is apply some sort of process to M every so often that brings it back into line, i.e. that finds the rotation which M is closest to and changes it into that." → This is precisely the patent's claimed "attraction to the rotation group" applied periodically.
• Reply (John Schultz): "find the matrix that is closest to the current matrix and is also orthonormal. The methods … are described completely in 'Graphics Gems'." → "closest orthonormal matrix" = the orthogonal polar factor; the claimed operators are approximations of exactly this.
• Reply (David Baraff): "Try using unit quaternions … After each multiplication, the quaternion is renormalized, preventing error from building up" (citing Barzel & Barr, SIGGRAPH '88). → Normalization‑after‑each‑multiply, the iterative form of claim 17.

(4) Numerical linear algebra: the polar decomposition and the Newton–Schulz iteration. The claimed operators in dependent claims 2 and 3 are textbook methods for computing the orthogonal polar factor:

• Claim 3 formula, L = (M + M⁻ᵗ)/2, is literally the Newton iteration step for the orthogonal polar factor of M (M = U·P, U orthogonal), published in N. J. Higham, "Computing the Polar Decomposition—with Applications," SIAM J. Sci. Stat. Comput. 7 (1986) 1160–1174, and in Golub & Van Loan, Matrix Computations. It converges to a matrix that (a) equals M when M is already orthogonal (consistency) and (b) pulls a nearby M onto the orthogonal group (attraction).
• Claim 2 formula, L = (3M − MMᵗM)/2, is the Newton–Schulz (Schulz) iteration for the orthogonal polar factor — the inverse‑free variant, X_{k+1} = ½X_k(3I − X_kᵗX_k) ≡ (3X − XXᵗX)/2. It likewise converges to the orthogonal polar factor and is quadratically/cubically convergent.

In other words, both exemplary operators are conventional inverse‑free algorithms for exactly the "closest orthonormal matrix" the 1990 Usenet thread and the 1986 SI3D paper were already discussing.

Confidence caveat: I verified the existence and substance of the 1986 SI3D paper and the 1990 Usenet thread via search results. I did not retrieve the full text of Higham (1986) in this session; the equivalence of the two formulas to the Newton/Newton–Schulz polar iterations is my own mathematical analysis plus my training knowledge of the numerical‑linear‑algebra literature, and I rate it high but not verified‑against‑primary‑source confidence. The examiner's failure to cite any of these references (Pique, Tountas & Katz, Lipscomb, Higham) is itself notable.

---

III. The central § 103 theory

Primary combination: (Sugiura US 5,581,665 in view of Park US 5,295,237 / Kaba US 5,568,600) + (Pique Graphics Gems I (1990) / the 1986 SI3D paper / Tountas & Katz (1971) / Lipscomb (1979) / comp.graphics 1990) + (Higham 1986 / Newton–Schulz iteration).

Why a PHOSITA would combine them:

1. Same field of endeavor and same problem. All references concern 3‑D graphics rendering where an object's orientation is represented by a matrix that is repeatedly multiplied/accumulated and then used to transform vertices for display. The renormalization references expressly diagnose the exact failure mode the patent describes: cumulative error (worsened by approximation matrices) turning a rotation into a skew (the patent's "unwanted visual artifacts … manifestly inconsistent with a physical rotation").

2. The motivation is stated in the prior art itself. The 1986 SI3D paper and the 1990 Usenet thread articulate both the incentive ("the picture still lost its shape") and the remedy ("renormalize … at the end of each cycle"; "reorthogonalization"; "find the rotation which M is closest to"). No hindsight reconstruction is needed — the references supply the why explicitly.

3. The claimed operators are a recognized, finite set of known solutions. Once one decides to re‑project the matrix onto SO(3), the available techniques are a small, well‑known menu: Gram–Schmidt orthogonalization (Lipscomb 1979), row/column renormalization (Tountas & Katz 1971), quaternion normalization (Baraff 1990; Shoemake 1985), and the polar decomposition via Newton/Newton–Schulz (Higham 1986). KSR teaches that where there are "a finite number of identified, predictable solutions," a PHOSITA's selection among them is obvious, and that the patent's recitation of the properties of the chosen solution ("consistency," "attraction") does not make the selection inventive.

4. The results are predictable. Re‑orthonormalizing a near‑rotation matrix necessarily reduces distortion in the rendered image. The patent claims no measured/unexpected result; it simply asserts the expected one ("the displayed image … has a more accurate and visually satisfying appearance").

---

IV. Claim‑by‑claim analysis

Independent Claim 1 (method: operate on M with an operator having consistency + attraction; render using L)

Likely obvious. The combination discloses every element:

• representing orientation as a matrix and repeatedly multiplying/accumulating (Sugiura '665; Park '237);
• the drift/distortion problem from those multiplications, especially with approximations (1986 SI3D paper; 1990 comp.graphics);
• the remedy of applying, periodically, an operator that returns the accumulated matrix to the rotation group (Pique Graphics Gems I 1990; Tountas & Katz 1971; Lipscomb 1979; and the "closest orthonormal matrix" discussion);
• the two claimed operator properties are inherent to those known maps: re‑orthonormalization leaves an already‑orthonormal matrix unchanged (consistency) and maps a near‑orthonormal matrix onto the orthogonal group (attraction);
• rendering the transformed object (Sugiura '665 graphics display; Park '237; Kaba '560).

Claim 1 does not even require a specific operator — it recites a genus defined by outcome. A genus whose every practical member is a known re‑orthonormalization technique is obvious as a matter of law and, independently, its breadth invites a § 112 written‑description/enablement challenge (the specification enables only two concrete operators).

Claim 2 — L = (3M − MMᵗM)/2

Likely obvious. This is the Newton–Schulz iteration for the orthogonal polar factor (Higham 1986 body of work; standard inverse‑free orthogonalization). The patent itself touts the only arguable distinction — that it "avoids the computation of a matrix inverse" — which is a known advantage of the Newton–Schulz family over the Newton family, i.e., the raison d'être of that classical method. Motivation: inverse‑free, low‑cost orthogonalization for real‑time graphics (cf. Park '237's emphasis on avoiding floating‑point multiplies; Johnson 1989 on fixed‑point graphics engines).

Claim 3 — L = (M + M⁻ᵗ)/2

Likely obvious — and nearly anticipated. This is the Newton polar iteration step (Higham 1986). A numerical‑linear‑algebra reference disclosing this iteration in the context of computing the closest orthogonal matrix, combined with any graphics reference identifying the drift problem, renders it obvious. (As the earlier section notes, the OCR "M⁻¹/2" rendering on a third‑party site is an error; the authoritative text is M⁻ᵗ, which is precisely the Newton step.)

Claim 4 — L = Σ_{k=0..n} α_k M(MᵗM)^k, n ≥ 1

Likely obvious. This is the general polynomial (Padé‑style) family of orthogonalization iterations to which both the Newton and Newton–Schulz iterations belong; selecting coefficients α_k to trade accuracy for cost is routine optimization squarely within Higham's framework and the "finite number of predictable solutions" rationale.

Claim 5 / Claim 18 (apply the operator a plurality of times)

Likely obvious. Newton–Schulz/Newton iterations are by definition iterated to convergence; the references teach repetition (and "every so often," "each cycle"). Iterating a known convergent method is the ordinary use of that method.

Claim 6 / Claim 26 (multiply vertices by L to render)

Likely obvious. Matrix‑vertex multiplication to render a rotated object is the universal graphics operation (Sugiura '665; Park '237; Kaba '560). The patent specification concedes this is conventional ("each vertex P_i … is multiplied by a rotation matrix R").

Claims 7–10, 28 (user input → approximate rotation matrix; multiply with orientation matrix; mouse/sliding ball)

Likely obvious. Incrementally mapping a mouse/trackball displacement to a small rotation and multiplying it into the accumulated orientation matrix is standard interactive‑graphics practice; Sugiura '665 discloses mouse/keyboard input to drive object transformation, and Park '237 discloses incremental rotation from input. Claim 10 (input "represents a desired rotation") is a statement of intended use.

Claims 11–16, 19–21 (sequence of approximate rotation matrices from sensor inputs; mouse, track ball, head/eye tracking, VR)

Likely obvious. US 6,018,349 (Szeliski) discloses generating a sequence of rotation matrices and iteratively aligning/correcting accumulated error — the structural counterpart of the claimed sequence. Head/eye tracking and VR head‑mounted displays were well known by 1999; reciting a known input modality does not impart patentability.

Claim 17 (iterative rendering loop with conditional consistency/attraction update)

Likely obvious over Sugiura '665/Szeliski '349 in view of Lengyel '393. The loop (generate B; multiply by A to get M; render; carry M forward as A) is conventional incremental transformation and is disclosed by Sugiura '665 (matrix updated from previous matrix) and Szeliski '349 (rotation matrix carried across an image sequence with periodic error correction). The conditional/periodic update is the only added element; periodic renormalization ("at the end of each cycle," "every so often") is expressly disclosed in the 1986 SI3D paper and the 1990 Usenet thread, and error‑triggered gating is disclosed by Lengyel '393 (compute a distortion measure such as singular values, compare to a threshold, decide whether to re‑render).

Claims 22–23 (update test = every N iterations; update test = estimate error in M)

Likely obvious. "Every N cycles" is routine throttling (the patent itself says any N ≥ 1), and is the natural implementation of "every so often." Error‑based triggering is disclosed by Lengyel '393's fidelity/singular‑value distortion measure and threshold logic. The patent's alternative formulations do not add anything beyond these two known scheduling mechanisms.

Claims 24–27 (system: memory storing M; processor applying the operator; display rendering L; optional graphics accelerator)

Likely obvious. Sugiura '665, Park '237, and Kaba '560 each disclose a processor + memory + display graphics system that transforms an object matrix and renders it. The graphics accelerator of claim 27 was, by 1999, conventional off‑the‑shelf hardware that the patent's own specification describes as a standard component ("a three dimensional (3-D) graphics accelerator is a specialized graphics rendering subsystem … designed to off‑load the 3-D rendering functions"). Reciting known hardware to perform the claimed method does not render a claim non‑obvious.

Claim 29 (second matrix = initial orientation; first matrix = desired rotation)

Likely obvious/Clerk's limitation. Pure description of what the two operands represent, consistent with any accumulated‑orientation system (Sugiura '665).

---

V. Motivation‑to‑combine, framed for the KSR record

---

VI. Secondary considerations (objective indicia)

I found no evidence of unexpected results, commercial success attributable to the claimed subject matter, industry praise, copying, or long‑felt but unsolved need. To the contrary, the "long‑felt need" is undercut by 20+ years of published renormalization practice (1971–1990). The relationship between the claimed operators and the asserted benefit is exactly the expected one, which weighs against non‑obviousness.

---

VII. Anticipated patentee rebuttals and my assessment

1. "The renormalization art teaches generic renormalization, not a consistent‑and‑attracting operator, and certainly not these specific formulas." — Weak. Consistency and attraction are inherent mathematical properties of any correct re‑orthonormalization; and claims 2 and 3 are the classical Newton/Newton–Schulz iterations, not new mathematics. The patent's contribution over the art appears to be naming properties and selecting two known methods.
2. "The prior art renormalizes rows/columns (Gram–Schmidt), not the whole matrix via polar decomposition." — Weak under KSR. This is a difference in the choice of known algorithm within a finite set; the patent offers no evidence that the polar‑based choice was unexpectedly superior.
3. "The claims require attraction to the rotation group, which Gram–Schmidt does not guarantee (determinant sign)." — Partially fair as to the general Gram–Schmidt genus, but resolved by combining with the polar‑decomposition references, whose orthogonal factor is the closest orthogonal matrix and (for matrices near SO(3), det ≈ +1) lies in SO(3). This actually strengthens the motivation to select the polar/Newton family — a reason to prefer the Newton/Newton–Schulz operators.
4. "Non‑analogous art" objection to the numerical‑linear‑algebra references — Fails: computing the closest orthogonal matrix was already being discussed within the graphics community (Pique, Graphics Gems I 1990; the 1990 Usenet thread citing Graphics Gems), so the numerical‑methods art is at minimum reasonably pertinent and combinable.
5. § 112 overlay — Independently of § 103, claims 1, 4, and 17's functional/property language raise written‑description and enablement questions, because the specification enables only two concrete operators and the "class" is defined by outcome rather than structure. This does not itself create § 103 invalidity but reinforces that the claims are broad and commensurately easy to invalidate.

---

VIII. Bottom line

• Strongest single theory: Claims 1, 2, 3, 4, 5, 18, 22, 23, 24 are likely obvious over the graphics‑renormalization prior art (Pique, Graphics Gems I 1990; Tountas & Katz 1971; Lipscomb 1979; the 1986 SI3D "Semantics of interactive rotations" paper; the 1990 comp.graphics thread) in view of the polar‑decomposition/Newton–Schulz numerical literature (Higham 1986), combined with a graphics rendering system reference (Sugiura US 5,581,665; Park US 5,295,237; Kaba US 5,568,600).
• Claims 17 and 22–23 additionally read on US 6,064,393 (Lengyel) for the error‑measured, conditional/periodic correction, and US 6,018,349 (Szeliski) for carrying a rotation matrix through a sequence and correcting accumulated error.
• Sensing/VR and hardware‑offload claims (7–16, 19–21, 26–29) are conventional in view of the same graphics system references plus the ordinary knowledge of mouse/trackball/head‑tracking input and graphics accelerators.
• The examiner appears to have missed the most on‑point art (Pique 1990 and the 1971/1979 renormalization lineage, and the polar/Newton–Schulz numerical methods). That gap, not any weakness in the patent, is the main reason the claims issued.

Confidence: High that a valid § 103 rejection can be constructed across all 29 claims using the references above; moderate‑to‑high on my specific identification of claim 2/claim 3 formulas with the Newton–Schulz/Newton polar iterations (mathematically certain, but I did not retrieve Higham (1986) in this session). I flag that I did not retrieve the full texts of Pique (Graphics Gems I) or the Tountas & Katz / Lipscomb primary sources; those are cited here via the 1986 SI3D paper and should be pulled from their originals for a formal office action.