Named for the pectinate muscles, the comb-like ridges inside the heart’s atria where blood is received and passed on. Its teeth are the three studies whose acronyms spell it: PEGKi, TENKi and ATEKi.
A materials campaign rarely starts from nothing. Before the first furnace run or hybrid-functional calculation, it can draw on band gaps from the Materials Project generalised gradient approximation (MP GGA), on in-house HSE06 calculations, on an experimental compilation, or, for an Al–Mg–Si heat-sink alloy, on formation energies from density functional theory (DFT) and from several interatomic potentials. These sources disagree on values, and their fidelity is known only qualitatively: everyone expects HSE06 to be closer to measurement than GGA, but nobody can say by how much for a given chemistry family. PECTINATE asks a narrower question that can be answered: are the relative differences between candidates (which compound leads, and by how much) preserved from one source to another, and what does reusing them save?
PECTINATE groups the studies that answer this question. The name joins their three acronyms: PEGKi, TENKi and ATEKi. Ratings in these studies are quantitative but relative: the margin between two compounds matters, not only their order, and no source is treated as ground truth.
PEGKi: rating whose knowledge to reuse¶
Evaluating transfer across modalities in autonomous materials systems
PEGKi infers relative ratings of candidates within each source from comparison outcomes, then scores every donor-to-receiver route with a robustness certificate or a refusal before any compute or instrument time is spent. On 261,053 archived band-gap records from eight sources, orderings alone recover the accuracy ladder of the DFT levels, and transfer follows the database of origin more than the functional. For the Al–Mg–Si alloy, where every query runs a DFT or molecular-dynamics calculation, the score selects each potential’s room-temperature run as donor at heat-treatment temperatures and discounts the two potentials that predict MgSi to be unstable.
Use for a campaign: decide which source to borrow from, in which direction, and when to refuse one, before committing compute.
TENKi: counting the saving in measurements¶
TENKi: counting the value of borrowed knowledge to a materials campaign in measurements saved
TENKi states the value of a borrowed ordering in evaluations saved. A donor ordering enters as a prior whose trust is learned from incoming results, and the agreement-versus-budget curve is recorded for each route. A GGA ordering cuts hybrid-functional compute about sixfold for band gaps; for Al–Mg–Si, a foundation-potential ordering lowers the DFT compute needed to find eight of the ten most stable compounds from 19% to 5%. An archived ordering that lacks the alloys gives no saving at fixed trust, so the saving depends on direction and coverage.
Use for a campaign: decide which source to compute first, how far to trust it, and when further evaluations stop improving the ranking.
ATEKi: dividing the outcome¶
A third study divides a campaign’s outcome among its members, meetings and borrowed knowledge when their contributions overlap and their costs differ. It is under double-blind review; its page will be added to Articles when the review ends.
Strengths and limitations¶
The evidence combines archived records (eight band-gap sources) with calculations run for these studies (DFT and molecular dynamics on Al–Mg–Si, band gaps computed on demand for 171 semiconductors). Each article states which is which.
Ratings are relative. They say which source’s ordering to reuse and how far, not what a property’s absolute value is.
Fixed choices in the methods (weights, priors, routing defaults) are cold-start defaults that a campaign can learn from its own results.
About a dozen compounds shared between two sources are needed before a route can be rated; a source that covers none of the candidates gives no saving.
Code and data¶
Each article has a public repository under Sparks