Intermediate values (interpolation) after Runge-Kutta calculation

I have a numerical ODE simulation that I computed at fixed time step $h$ using a 4-th order Runge-Kutta method (RK4), producing a series of results $(x_1,y_1), (x_2,y_2), (x_3,y_3) \dots (x_N,y_N)$.

If I want to find an approximate solution $y$ at a location $x$ in between my intervals, I could use

1. linear interpolation (just kidding, I wouldn't use this)
2. cubic splines (current solution, whats my error estimate ?)
3. a new RK4 step with $h=x-x_i$ (is error still $O(h^4)$ ?)
4. some other appropriate method

What are the recommended methods for interpolating Runge-Kutta results and what is their error order?

• The link provided by @J.M. eludes to the solution that to have the same order as the method additional function evaluations are needed. This is fine by me. – ja72 May 24 '13 at 12:36
• Yes, and what you can do if you're willing to do more function evaluations (bootstrapping) is in that reference, too. – J. M. May 24 '13 at 12:47

You're asking how to produce dense output from your Runge-Kutta method. There are a number of ways to do this (see e.g. Hairer/Nørsett/Wanner). As noted in that reference, if you don't want to do more function evaluations aside from those already done by your fourth-order Runge-Kutta method, the best you can hope for is a third-order interpolant. This is fine, since it can also be shown that for a $p$-th order Runge-Kutta method, you can get by with dense output of order $p-1$.

The easiest third-order dense output you can construct is of course the cubic Hermite interpolant. Recall that given two function values and two derivative values, you can always build a unique cubic: the Hermite interpolant. Thus, you are guaranteed a $C^1$ interpolating function.

• Yes, and this is what I am currently doing ( I called them cubic splines ) but it is the same thing. – ja72 May 24 '13 at 12:23
• @ja, no, they're not splines. Cubic splines necessarily have $C^2$ continuity, which is not the case here. – J. M. May 24 '13 at 12:24
• You are saying that cubic splines will deviate more from the dense output due to the additional constraints to reach $\boldsymbol{C}^2$ ? – ja72 May 24 '13 at 12:33
• I didn't say that they'll deviate. What I said is that cubic splines and cubic Hermite interpolants are different; in particular, cubic splines are a special case of the cubic Hermite interpolants that have $C^2$ instead of just $C^1$ continuity. You don't need to build a spline; a simple Hermite interpolant should suffice. – J. M. May 24 '13 at 12:41

The RK4 method implicitly constructs a degree 3 polynomial interpolant, using the data $f(x_i)$, $f(x_{i+1})$, $f'(x_i)$, and $f'(x_{i+1})$ in each interval.

This interpolant can be constructed rather easily and efficiently using a linear combination of shifted Hermite basis functions in each interval.