Change the particle diameter and grain size: smaller particles have more surface atoms and melt earlier; smaller grains are stronger until they get too small.
Why small is different. As a particle shrinks, a growing share of its atoms sit on the surface where they have fewer neighbours. In a face-centred-cubic (FCC) crystal a bulk atom has 12 nearest neighbours; this lab builds the actual lattice of a sphere of diameter D and counts atoms with fewer than 12, so the surface fraction fs comes out of the geometry (roughly fs ≈ 1 − (1 − 2 datom/D)3 ≈ 6 datom/D for large D, datom ≈ a/√2).
Melting-point depression (Gibbs-Thomson / Pawlow form). Tm(D) = Tbulk [1 − 4 σsl / (ΔHf ρs D)], where σsl is the solid-liquid interface energy, ΔHf the latent heat per kg and ρs the solid density. Under-coordinated surface atoms are easier to mobilise, so melting starts earlier.
Grain-size strengthening (Hall-Petch). In a polycrystal, grain boundaries block dislocations: σy = σ0 + k d−1/2. Below a critical grain size (about 10-20 nm for copper) strength stops rising and falls again because boundaries slide (the "inverse Hall-Petch" effect); here that regime is drawn with a simple linear decline, only a qualitative guide.
Grain picture. A random (Poisson) Voronoi tessellation with a fixed number of grains; the field of view scales with d (see the scale bar). For such a tessellation the mean linear intercept equals about 0.785 d, which the check below measures from the image.
Honest simplifications. No facets, no surface relaxation or reconstruction, no size distribution of the particles, 2D slice drawing of a 3D count, single-phase pure metals. The constants (σsl, σ0, k, critical grain size, slope of the inverse regime) are typical rounded literature-style values and are illustrative; they are not a substitute for measured data.