The Problem That Led to fullrmc describes a system that could not be reduced to one well-defined structure: mixtures of phases, grain sizes, local domains, defects, and competing structural hypotheses that all had to be represented and refined at once. fullrmc’s multiframe concept addresses exactly this: each candidate structure or coexisting phase is represented as a subframe, and multiple subframes are refined together against the same experimental data.
How those subframes relate to one another physically is not always the same, so fullrmc offers three run modes, each built on a different assumption about how much subframes influence each other.
Note
Run modes are part of fullrmc’s Cloud Services, powered by softgrid, and are not included in the free, locally installable fullrmc package.
Every subframe is refined as a totally independent structure. No assumption is made about how the candidate structures relate to one another; each is simply fit on its own against the same experimental data. This is the right starting point when nothing is yet known about whether or how the candidate structures coexist or interact, for example when screening a broad set of structural hypotheses before narrowing them down.
Subframes are assumed to coexist within the same sample but to be far enough apart that any cross-structural correlation between them can be neglected. Rather than being fit independently, each subframe contributes to the overall experimental fit weighted by its estimated population fraction, and that weighting can itself be refined automatically as the fit progresses. This models situations such as a distribution of nanoparticle sizes, coexisting local structures, or grains far enough apart to be structurally decoupled, where the sample is genuinely a statistical mixture of distinguishable structures rather than one connected system.
Subframes are treated as parts of one connected physical system, and correlations between them are explicitly computed rather than neglected. This is the appropriate mode when structural domains physically interact, such as adjoining grains, interfaces, or coupled defects, where the structure of one subframe genuinely affects its neighbors and the fit must capture that coupling rather than treating each piece in isolation.
The three modes form a progression in how much inter-structure correlation is assumed: none in statistical, partial in mesoscopic, full in nanoscopic. Which one is appropriate depends on the physics of the system being modeled, not on convenience. Assuming independence between structures that are physically coupled will bias the fit just as much as forcing correlation onto structures that do not actually interact.
For parameter-level detail on launching each run mode, see fullrmc.MultiframeUtils.