Splines are used broadly in simulations and 3D-modelling. For applications where abrupt changes occur in the data, ordinary B-splines introduce spurious overshoots and unwanted oscillation. In this thesis we explore cubic L1 splines, often referred to as Lavery splines after John E. Lavery, as an alternative to conventional B-splines. Cubic L1 splines, in contrast to conventional B-splines, do not enforce continuity in the second derivative. Instead cubic L1 splines allow freedom in the first derivatives and are defined by minimizing the L1 norm of the second derivative over a given spline. The resulting minimization problem is not linear in general, but a discretization of the problem is. The problem can therefore be solved with standard interior point methods.
After presenting the necessary mathematical background and definitions, uni-variate cubic L1 splines are explored in detail followed by bi-variate cubic L1 splines. With the mathematical theory behind the splines in place, a discretization of the problem is made to make it computationally solvable. The implementation of the splines was done in the Julia programming language using the JuMP package. Our main findings indicate that cubic L1 splines do in fact interpolate abruptly changing data better than conventional methods. L1 splines perform especially well when there are linear segments in the data. However, for naturally oscillative data, cubic L1 splines underperform in comparison to conventional B-splines. Additionally the computational complexity of L1 splines is significantly higher due to the optimization problem that must be solved in order to obtain the spline coefficients, unlike B-splines, which can be calculated by solving tridiagonal linear systems.