# Callback problem with DifferentialEquations

**URL:** <https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057>\
**Category:** Modelling & Simulations\
**Tags:** diffeq\
**Created:** [September 4, 2020, 12:46pm UTC](https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057 "2020-09-04T12:46:25Z")\
**Posts on this page:** 11\
**Page:** 1

<div class="post-metadata">

**Author:** ![moesphere](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/moesphere/32/3348_2.png) [@moesphere](https://discourse.julialang.org/u/moesphere)\
**Post date:** [September 4, 2020, 12:46pm UTC](https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057/1 "2020-09-04T12:46:25Z")

</div>

I am stuck with a problem considering a `SavingCallback` from the `DifferentialEquations` suite.

I could not create a MWE, thus I will explain the problem.  
I have an `ODEProblem` where I use a `SavingCallback` for the calculation of some internals. In the calculation I need the `du` vector, thus I have created a callback function (I have seen that on stackoverflow) similar to

```julia
f = (u,t,integrator) -> calculate_internals(get_du(integrator), u, t)

```

If I solve this, I get an `UndefRefError: access to undefined reference` error, because it tries to access the `fsallast` property that apparently does not exist. However, if I check with `propertynames(integrator)` it does appear in the list (I checked this in the Juno debugger). Is this a bug?

Note that I can solve the problem without providing the callback or if I instead of providing the `du` vector in the callback function, I provide a random vector with the same size.

 ![image](https://global.discourse-cdn.com/julialang/original/3X/a/3/a3e0507564fe447a6d36ff5c3b8816b9cf032fb3.png)

---

<div class="post-metadata">

**Author:** ![ChrisRackauckas](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/chrisrackauckas/32/77_2.png) [@ChrisRackauckas](https://discourse.julialang.org/u/ChrisRackauckas)\
**Post date:** [September 4, 2020, 4:34pm UTC](https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057/2 "2020-09-04T16:34:05Z")

</div>

Are you calling this in initialize? I need a bit more information than this. Share some code.

---

<div class="post-metadata">

**Author:** ![moesphere](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/moesphere/32/3348_2.png) [@moesphere](https://discourse.julialang.org/u/moesphere)\
**Post date:** [September 4, 2020, 4:49pm UTC](https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057/3 "2020-09-04T16:49:29Z")

</div>

Thanks for getting back to me!

What do you mean with initialize?

Originally, the `calculate_internals` function is called within the ode function. If I want to extract the internals, I use the `calculate_internals` function in a SavingCallback. It worked when I did not use the `get_du` function previously and as said it works if I supply a random vector with the length of the states.

I don’t have access to the code atm.

I was thinking about augmenting the states and have all the internals as additional states by setting `du=0` and setting the states `u` within the ode function. Is there any speed compromise that comes with it, or any other disadvantages?

---

<div class="post-metadata">

**Author:** ![ChrisRackauckas](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/chrisrackauckas/32/77_2.png) [@ChrisRackauckas](https://discourse.julialang.org/u/ChrisRackauckas)\
**Post date:** [September 4, 2020, 5:02pm UTC](https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057/4 "2020-09-04T17:02:28Z")

</div>

Yeah I think the issue is just when it’s called at the initial point, If you do `save_start = false` in the SavingCallback it should be fine though.

---

<div class="post-metadata">

**Author:** ![moesphere](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/moesphere/32/3348_2.png) [@moesphere](https://discourse.julialang.org/u/moesphere)\
**Post date:** [September 8, 2020, 6:49am UTC](https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057/5 "2020-09-08T06:49:46Z")

</div>

`save_start = false` does not work.

I am using now `integrator(t, Val{1})` as you suggested in [https://github.com/SciML/OrdinaryDiffEq.jl/issues/785#issuecomment-504962062](https://github.com/SciML/OrdinaryDiffEq.jl/issues/785#issuecomment-504962062). That works.

---

<div class="post-metadata">

**Author:** ![moesphere](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/moesphere/32/3348_2.png) [@moesphere](https://discourse.julialang.org/u/moesphere)\
**Post date:** [September 9, 2020, 1:43pm UTC](https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057/6 "2020-09-09T13:43:09Z")

</div>

Unfortunately, the approach does not give the same derivative values as the `get_du` approach.

However, I have managed to create a MWE (partly taken from the official docs):

```julia
using DifferentialEquations

function lorenz!(du,u,p,t)
 du[1] = 10.0*(u[2]-u[1])
 du[2] = u[1]*(28.0-u[3]) - u[2]
 du[3] = u[1]*u[2] - (8/3)*u[3]
end

u0 = [1.0;0.0;0.0]
tspan = (0.0,100.0)
prob = ODEProblem(lorenz!,u0,tspan)

integ1 = init(prob, Tsit5())
get_du(integ1) # works
integ2 = init(prob, Rodas4P())
get_du(integ2) # throes the mentioned error

```

---

<div class="post-metadata">

**Author:** ![moesphere](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/moesphere/32/3348_2.png) [@moesphere](https://discourse.julialang.org/u/moesphere)\
**Post date:** [September 9, 2020, 1:49pm UTC](https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057/7 "2020-09-09T13:49:07Z")

</div>

Seems to work with all the other `Rodas` implementations apart from the ones with 4th order.

---

<div class="post-metadata">

**Author:** ![moesphere](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/moesphere/32/3348_2.png) [@moesphere](https://discourse.julialang.org/u/moesphere)\
**Post date:** [September 9, 2020, 2:01pm UTC](https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057/8 "2020-09-09T14:01:01Z")

</div>

And a MWE in combination with a callback where it fails for all solvers (also taken from the official docs, see [here](https://diffeq.sciml.ai/stable/features/callback_library/#Example-2):

```julia
using DiffEqCallbacks, OrdinaryDiffEq, LinearAlgebra
prob = ODEProblem((du,u,p,t) -> du .= u, rand(4,4), (0.0,1.0))
saved_values = SavedValues(Float64, Tuple{Float64,Float64})
cb = SavingCallback((u,t,integrator)->(get_du(integrator)[1],norm(u)), saved_values)
sol = solve(prob, Tsit5(), callback=cb) # throws mentioned error

```

---

<div class="post-metadata">

**Author:** ![moesphere](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/moesphere/32/3348_2.png) [@moesphere](https://discourse.julialang.org/u/moesphere)\
**Post date:** [September 9, 2020, 8:10pm UTC](https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057/9 "2020-09-09T20:10:02Z")

</div>

You were right, specifying `save_start = false` in the `SavingCallback` did solve the issue if another solver than one from the `Rodas4` family is chosen.

I do have another short question:  
The time values in the `saved_values` array do not need to be necessarily in ascending order, is that correct (happens in my case)?

---

<div class="post-metadata">

**Author:** ![ChrisRackauckas](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/chrisrackauckas/32/77_2.png) [@ChrisRackauckas](https://discourse.julialang.org/u/ChrisRackauckas)\
**Post date:** [September 9, 2020, 11:39pm UTC](https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057/10 "2020-09-09T23:39:04Z")

</div>

> [@moesphere](#):
>
> You were right, specifying `save_start = false` in the `SavingCallback` did solve the issue if another solver than one from the `Rodas4` family is chosen.

Could you open an issue on OrdinaryDiffEq? This likely comes down to being an issue at the initial point for some customized interpolants.

---

<div class="post-metadata">

**Author:** ![moesphere](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/moesphere/32/3348_2.png) [@moesphere](https://discourse.julialang.org/u/moesphere)\
**Post date:** [September 11, 2020, 12:40pm UTC](https://discourse.julialang.org/t/callback-problem-with-differentialequations/46057/11 "2020-09-11T12:40:24Z")

</div>

Done, see [here](https://github.com/SciML/OrdinaryDiffEq.jl/issues/1266)