## Abstract

Data exchange is the problem of transforming data structured under a source schema into data structured under a target schema, taking into account structural relationships between the two schemas, which are described by a schema mapping. Existing schema-mapping languages lack the ability to express arithmetic operations, such as addition and multiplication, which naturally arise in data warehousing, ETL applications, and applications involving scientific data. We initiate the study of data exchange for arithmetic schema mappings, that is, schema mappings specified by source-to-target dependencies and target dependencies that may include arithmetic formulas interpreted over the algebraic real numbers (we restrict attention to algebraic real numbers to maintain finite presentability).

We show that, for arithmetic schema mappings without target dependencies, the existence-of-solutions problem can be solved in polynomial time, and, if a solution exists, then a universal solution (suitably defined) exists and can be computed in polynomial time. In the case of arithmetic schema mappings with a weakly acyclic set of target dependencies, a universal solution may not exist, but a finite universal basis exists (if a solution exists) and can be computed in polynomial space. The existence-of-solutions problem turns out to be NP-hard, and solvable in PSpace. In fact, we show it is ∃R-complete, which means that it has the same complexity as the decision problem for the existential theory of the real numbers, or, equivalently, the problem of deciding whether or not a quantifier-free arithmetic formula has a solution over the real numbers. If we allow only linear arithmetic formulas in the schema mapping and in the query, interpreted over the rational numbers, then the existence-of-solutions problem is NP-complete. We obtain analogous complexity results for the data complexity of computing the certain answers of arithmetic conjunctive queries and linear arithmetic conjunctive queries.