Across eight boundaries covering 14,000 nm², the average compositions clustered at 9.3 ± 2 at.%, with a standard deviation of 1.2 at.%. Their effective segregation energies were within ±3 kJ/mol of the -10 kJ/mol average. The largest differences occurred locally within these interfaces, not between them.
Enough Area Matters More Than More Boundaries
Some detailed APT datasets come from nanocrystalline materials in which an individual boundary may cover only about 100 nm². In the boundary examined here, an area of that size was not necessarily representative of the full interface.
The authors found that measuring the average concentration of an individual random Ni-Pd boundary required an interfacial area on the order of 1,000 nm². Once they included enough area, the data converged toward a more robust value.
The same principle carried over to the polycrystalline average. For this particular system, the determining factor was not simply the number of crystallographically different boundaries. A handful of random boundaries could provide the average, provided that they represented enough total interfacial area.
Testing the APT Measurements Against Known Boundaries
The simulations identified exactly which atoms belonged to a grain boundary, allowing the researchers to test the APT analysis procedure against known structural information.
The mesh located from chemical fluctuations closely followed the structurally identified boundary, agreeing to within 0.25 nm normal to the boundary plane. In a dilute Monte Carlo configuration, the ladder-diagram fitting error was 2.6%, equivalent to ±2 atoms. This independently validated the method applied to the APT measurements.
The simulations then added site-level context to the concentration maps. Individual simulated boundaries contained much of the segregation-energy spectrum found across the full polycrystal. Although differences between boundaries remained, they were modest compared with the spread of more than 100 kJ/mol among individual sites.
They also reproduced the dependence on sampling area. At the smallest areas, calculated concentrations could differ from the true polycrystalline average by ±5 at.%. Grouping boundaries to provide roughly 300 nm² brought the error to within 1 at.% of that average.
What Sampling Scale Changes
The comparison also places an important limit on how to interpret local segregation distributions. At small sampling sizes, the measured distribution of effective segregation energies reflects both statistical sampling and correlations in site-wise behavior. Because each local measurement averages multiple atomic sites, it cannot be treated as a direct measurement of the underlying site-wise energy spectrum.
APT showed where palladium accumulated, how strongly its concentration varied across individual interfaces, and how the reported average changed as the analysis area grew. The simulations validated the measurement framework and helped explain the broadening of local distributions.
For researchers using APT to characterize grain boundary segregation, the study demonstrates the value of combining high spatial resolution with large analysis volumes. The total interfacial area included in the measurement can determine whether the result describes a local patch or represents the boundary as a whole.