Thomas Nowak

A grouping.

Thomas Nowak -- with Matthias Függer, Ulrich Schmid and colleagues at TU Wien, later CNRS and Paris-Saclay. Five records, no single-author paper among them, and together they are one sustained argument carried from an impossibility proof to an oscilloscope and back to a usable model. The corpus holds every step as a primary source and every one is readable.

2014 fuegger2014faithful involution channels proposed: delay functions that are their own inverse 2015 najvirt2015experimental the model's central hypothesis put on a chip and measured 2016 fugger2016unfaithful the impossibility: no bounded single-history binary model propagates glitches faithfully 2019 fugger2019faithful the journal account, folding the measurements into the theory 2021 maier2021composable the repair made composable, with independent discretisation thresholds

**Read the dates, not the citation order.** The proposal is 2014 and the impossibility theorem is 2016, so this is not a negative result followed by a rescue. The group published the candidate solution first and then proved what it had to escape from -- which is why the impossibility result names its own loophole rather than being punctured later by someone else. fugger2016unfaithful rules out **bounded single-history** channels, and the assumption doing the work is that every channel has some minimum delay; drop it and faithfulness returns, which is precisely what an involution channel does.

**What the programme is for.** A binary model of a circuit is not merely approximate, it can be wrong in both directions: licensing circuits that cannot be built and forbidding circuits that exist. The test the group chooses is **short-pulse filtration**, and faithfulness means the model's solvability boundary coincides with the boundary for physical circuits. That is a sharper standard than accuracy, and the papers say so -- faithfulness and accuracy are different virtues, and a non-faithful model may fit measurements better in some regimes.

**The reason this person set is worth having is the 2015 and 2019 pair.** najvirt2015experimental is, by the reckoning in the-switch-or-the-curve, the only member of that set that measures silicon rather than arguing about models: a custom UMC-90 ASIC carrying an inverter chain with on-chip high-speed amplifiers. Measured delay functions fit involutions; the incumbent exponential delay function cannot be fit across the whole pulse-width range, and the misfit is worse at 65 nm than at 90 nm, so the model's advantage grows as the process shrinks. And the honesty that bounds the claim is theirs, not a later critic's: the short-pulse region where involution does its work lies **below what the amplifiers resolve**, and its values were extrapolated using the involution property itself. The fit is measured above that range and assumed below it.

**Why the line did not stop in 2019.** maier2021composable answers two shortcomings that sit exactly on the seam this corpus cares about. Involution channels need matching discretisation thresholds, which are hard to determine and differ between interconnected gates -- so the abstraction each gate is licensed under is not the one its neighbour assumes. And metastability and high-frequency oscillation could be invisible in the predictions, which is a faithful model concealing the thing marino1981general proves cannot be designed away. The Composable Involution Delay Model allows independent thresholds and exposes cancelled pulse trains at the interconnect, and faithfulness survives as a consequence of the mathematical properties of involutions rather than of the original construction.

**A caution for anyone adding to this set.** Two duplicate records of the 2021 paper exist in the corpus, maier2021composableb and maier2021composablec, both created by the record create command failing to notice a work already held and both awaiting deletion by a person. Only maier2021composable belongs here. See the flags on those records.

Overlaps the-switch-or-the-curve, which holds the question these papers answer, and the metastability line running from marino1981general through friedrichs2018metastabilitycontaining -- Függer appears in both, Nowak does not, and the distinction is worth keeping: this group works on whether a *model* is faithful to the device, while the Lenzen-Friedrichs-Bund line works on what a *circuit* can be built to guarantee in spite of it.