Amos Uderzo
6 min
Abstract
In the present paper, a systematic study is made of quantitative semicontinuity (a.k.a. Lipschitzian) properties of certain multifunctions, which are defined as a solution map associated to a family of parameterized ``split" feasibility problems. The latter are a particular class of convex feasibility problems with well recognized applications to several areas of engineering and systems biology. As a part of a perturbation analysis of variational systems, this study falls within the framework of a line of research pursued by several authors. It is performed by means of techniques of variational analysis, which lead to establish sufficient conditions for the Lipschitz lower semicontinuity, calmness, isolated calmness, Lipschitz upper semicontinuity and Aubin property of the solution map. Along with each of these properties, a quantitative estimate of the related exact bound is also provided. The key elements emerging on the way to achieving the main results are dual regularity conditions qualifying the problem behaviour, which are expressed in terms of convex analysis constructions involving problem data. The approach here proposed tries to unify the study of the aforementioned properties.
Alex: So the merit function turns the abstract set into a number you can analyze, and the condition keeps corrections reliable.
Sam: For global versions, if the condition holds uniformly, solutions stay within a predictable bubble as parameters change. The paper shows this unifies calmness and other traits without separate proofs. It's a notable step for reliable solvers in noisy tasks.
Alex: How do they check if this dual regularity condition holds in practice?
Sam: They define it through two quantities, each as the smallest size of certain direction vectors near the sets. These directions come from a normal cone—imagine the set as a fenced yard; the normal cone lists arrows pointing straight out from the boundary. The condition requires those arrow lengths, after mapping back, to stay above zero nearby.
Alex: So it's measuring if those outward-push directions are always strong enough.
Sam: Yes. Convexity lets them compute the slope using these cones via chain rules. This shows the slope is at least the minimum value, satisfying the Basic Lemma. With continuity on the sets and mapping, it guarantees solutions exist nearby and the error bound holds.
Alex: Those continuity assumptions ensure the zones and mapping don't jump suddenly as parameters tweak?
Sam: Precisely. The theorem delivers local solvability and a linear error bound.
Alex: And solutions to perturbed problems converge back to the original as noise goes to zero?
Sam: That's a corollary: for parameters approaching the nominal, you can pick solutions converging too, by the bound and continuity of the merit function. Examples show why the condition is needed: one where solutions vanish without it, like a yard shrinking to empty; another where errors grow faster than linear.
Alex: In image reconstruction, noisy pixels shift the target zone predictably, without the solver diverging.
Sam: Yes. For Hilbert spaces like pixel grids, the cones simplify to projection directions. The paper suggests it provides a unified tool for stability, propagating data regularity to solution behavior via that error bound.
Alex: How does this give specific guarantees for those stability traits?
Sam: Under the condition plus continuity-like properties on the zones and mapping, the solution map inherits similar bounds. For example, they prove a bound ensuring solutions don't vanish too quickly: the distance to the nearest new solution is at most data stretch factors divided by the slope minimum.
Alex: So the solution set can't suddenly empty out; it overlaps a shrinking ball around the original.
Sam: Yes. For calmness—keeping new solutions close to the old ones—if the zones are calm and the mapping Lipschitz continuous, the solution map's calmness is bounded by those over the slope. For the stronger Aubin property, tighter assumptions give a bound like the sum of data Lipschitzians divided by the slope.
Alex: One condition plus data regularity gives quantitative bounds across traits—no separate analysis needed.
Sam: Exactly. The paper notes counterexamples where some stability holds without it fully, and it requires complete spaces and lower semicontinuous data. For global stability, a uniform version ensures solutions stay controlled even far from the reference.
Alex: Practical limits but solid when assumptions fit. What does this mean for sensor data or optimization?
Sam: It aids robust signal processing, where approximate solvers for polyhedral sets converge under these bounds. It supports reformulations ensuring stability in bi-level optimization. Overall, it unifies perturbation analysis for better methods in split setups.
Alex: This quantifies how data wiggles propagate, helping solvers handle noisy inputs.
Sam: Exactly. The work ties data behaviors to solution stability through that slope condition, offering exact bounds without case-by-case proofs.
Alex: Well put. Thanks for joining ResearchPod.