# IVP Piecewise ODE: Changing the Domain outside region of interest changes result dramatically

Good afternoon,

I am attempting to solve the following ODE:

d[h_] := Piecewise[{{0, Abs[h] > 110}, {0.1,
110 >= Abs[h] > 100 && h < 0}, {-0.1,
110 >= Abs[h] > 100 && h >= 0}}]

S = NDSolve[{h''[t] == d[h[t]], h'[-120] == 1, h[-120] == -120}, h, {t, -120, 120}]


From the definition of d[h_], we would expect h'[t] to be constant in the regions where d[h_] = 0, and otherwise linearly increasing/decreasing depending on whether h[t] is positive or negative. And indeed, using the bounds -120 to 120, we see that that is the case:

Plot[h[t] /. S, {t, -120, 120}, AxesLabel -> {"t", "h[t]"}, PlotLabel -> "tf=120"]


Plot[h'[t] /. S, {t, -120, 120}, AxesLabel -> {"t", "h'[t]"}, PlotLabel -> "tf=120"]


From my knowledge of the analytical solution, this appears to be the correct answer.

Now, let me take tf=120->300, but examine h[t] and h'[t] only in the region t=-120 to 120, the same as before. The results dramatically change:

S2 = NDSolve[{h''[t] == d[h[t]], h'[-120] == 1, h[-120] == -120}, h, {t, -120, 300}]

Plot[h[t] /. S2, {t, -120, 120}, AxesLabel -> {"t", "h[t]"}, PlotLabel -> "tf=300"]


Plot[h'[t] /. S2, {t, -120, 120}, PlotRange -> All,


AxesLabel -> {"t", "h'[t]"}, PlotLabel -> "tf=300"]

The answer is now incorrect!

It appears that NDSolve is somehow not changing h'[t] when h''[t] is non-zero (for t such that 100<=h[t]<110). However, I'm confused why this is the case when all I did was change tf.

My question is that is there a way to prevent this from occurring? This is a simplified version of a much more complicated problem I am working on, which does the same thing.

I have tried making the following change to no avail:

S2 = NDSolve[{h''[t] == d[h[t]], h'[-120] == 1, h[-120] == -120}, h, {t, -120, 300}, InterpolationOrder -> All, WorkingPrecision -> 22]


and have also played around with the option:

S2 = NDSolve[{h''[t] == d[h[t]], h'[-120] == 1, h[-120] == -120}, h, {t, -120, 300}, InterpolationOrder -> All, WorkingPrecision -> 22, Method -> {"DiscontinuityProcessing" -> False}]


but neither change the last two plots at all. I'm not sure what else to try/any potential workarounds.

Thanks!

NDSolve gets excited when the estimated error is zero, which happens when solutions are of low degree. Rein it in with MaxStepSize or MaxStepFraction:
S2 = NDSolve[{h''[t] == d[h[t]], h'[-120] == 1, h[-120] == -120},

• (+1) just for NDSolve gets excited ;) – Nasser Mar 25 at 23:59