# SDE Work-Precision Diagrams

In this notebook we will run some simple work-precision diagrams for the SDE integrators. These problems are additive and diagonal noise SDEs which can utilize the specialized Rossler methods. These problems are very well-behaved, meaning that adaptive timestepping should not be a significant advantage (unlike more difficult and realistic problems). Thus these tests will measure both the efficiency gains of the Rossler methods along with the overhead of adaptivity.

using StochasticDiffEq, Plots, DiffEqDevTools, DiffEqProblemLibrary
using DiffEqProblemLibrary.SDEProblemLibrary: importsdeproblems; importsdeproblems()
gr()
const N = 1000
1000

In this notebook, the error that will be measured is the strong error. The strong error is defined as

:$E = \mathbb{E}[Y_\delta(t) - Y(t)]$

where $Y_\delta$ is the numerical approximation to $Y$. This is the same as saying, for a given Wiener trajectory $W(t)$, how well does the numerical trajectory match the real trajectory? Note that this is not how well the mean or other moments match the true mean/variance/etc. (that's the weak error), this is how close the trajectory is to the true trajectory which is a stronger notion. In a sense, this is measuring convergence, rather than just convergence in distribution.

dX{t}=\left(\frac{\beta}{\sqrt{1+t}}-\frac{1}{2\left(1+t\right)}X{t}\right)dt+\frac{\alpha\beta}{\sqrt{1+t}}dW{t},\thinspace\thinspace\thinspace X{0}=\frac{1}{2} where $\alpha=\frac{1}{10}$ and $\beta=\frac{1}{20}$. Actual Solution: X{t}=\frac{1}{\sqrt{1+t}}X{0}+\frac{\beta}{\sqrt{1+t}}\left(t+\alpha W_{t}\right).

First let's solve this using a system of SDEs, repeating this same problem 4 times.

prob = remake(prob,tspan=(0.0,1.0))

reltols = 1.0 ./ 10.0 .^ (1:5)
abstols = reltols#[0.0 for i in eachindex(reltols)]
setups = [Dict(:alg=>SRIW1())
Dict(:alg=>EM(),:dts=>1.0./5.0.^((1:length(reltols)) .+ 1))
Dict(:alg=>SRA1())
]
names = ["SRIW1","EM","RKMil","SRIW1 Fixed","SRA1 Fixed","SRA1"]
wp = WorkPrecisionSet(prob,abstols,reltols,setups;numruns=N,names=names,maxiters=1e7,error_estimate=:l2)
plot(wp)
prob = remake(prob,tspan=(0.0,1.0))

reltols = 1.0 ./ 10.0 .^ (1:5)
abstols = reltols#[0.0 for i in eachindex(reltols)]
setups = [
Dict(:alg=>SRA1())
Dict(:alg=>SRA2())
Dict(:alg=>SRA3())
Dict(:alg=>SOSRA())
Dict(:alg=>SOSRA2())
]
wp = WorkPrecisionSet(prob,abstols,reltols,setups;numruns=N,maxiters=1e7,error_estimate=:l2)
plot(wp)

Now as a scalar SDE.

prob = remake(prob,tspan=(0.0,1.0))

reltols = 1.0 ./ 10.0 .^ (1:5)
abstols = reltols#[0.0 for i in eachindex(reltols)]

setups = [Dict(:alg=>SRIW1())
Dict(:alg=>EM(),:dts=>1.0./5.0.^((1:length(reltols)) .+ 1))
Dict(:alg=>SRA1())
]
names = ["SRIW1","EM","RKMil","SRIW1 Fixed","SRA1 Fixed","SRA1"]
wp = WorkPrecisionSet(prob,abstols,reltols,setups;numruns=N,names=names,maxiters=1e7,error_estimate=:l2)
plot(wp)
prob = remake(prob,tspan=(0.0,1.0))

reltols = 1.0 ./ 10.0 .^ (1:5)
abstols = reltols#[0.0 for i in eachindex(reltols)]
setups = [
Dict(:alg=>SRA1())
Dict(:alg=>SRA2())
Dict(:alg=>SRA3())
Dict(:alg=>SOSRA())
Dict(:alg=>SOSRA2())
]
wp = WorkPrecisionSet(prob,abstols,reltols,setups;numruns=N,error_estimate=:l2)
plot(wp)

### Diagonal Noise

We will use a 4x2 matrix of indepdendent linear SDEs (also known as the Black-Scholes equation)

dX{t}=\alpha X{t}dt+\beta X{t}dW{t},\thinspace\thinspace\thinspace X{0}=\frac{1}{2} where $\alpha=\frac{1}{10}$ and $\beta=\frac{1}{20}$. Actual Solution: X{t}=X{0}e^{\left(\beta-\frac{\alpha^{2}}{2}\right)t+\alpha W{t}}.

prob = prob_sde_2Dlinear
prob = remake(prob,tspan=(0.0,1.0))

reltols = 1.0 ./ 10.0 .^ (1:5)
abstols = reltols#[0.0 for i in eachindex(reltols)]

setups = [Dict(:alg=>SRIW1())
Dict(:alg=>EM(),:dts=>1.0./5.0.^((1:length(reltols)) .+ 1))
]
names = ["SRIW1","EM","RKMil","SRIW1 Fixed"]
wp = WorkPrecisionSet(prob,abstols,reltols,setups;numruns=N,names=names,maxiters=1e7,error_estimate=:l2)
plot(wp)
prob = prob_sde_2Dlinear
prob = remake(prob,tspan=(0.0,1.0))

reltols = 1.0 ./ 10.0 .^ (1:5)
abstols = reltols#[0.0 for i in eachindex(reltols)]

setups = [Dict(:alg=>EM(),:dts=>1.0./5.0.^((1:length(reltols)) .+ 2))
Dict(:alg=>SRI())
Dict(:alg=>SRIW1())
Dict(:alg=>SRIW2())
Dict(:alg=>SOSRI())
Dict(:alg=>SOSRI2())
]
wp = WorkPrecisionSet(prob,abstols,reltols,setups;numruns=N,maxiters=1e7,error_estimate=:l2)
plot(wp)

Now just the scalar Black-Scholes

prob = prob_sde_linear
prob = remake(prob,tspan=(0.0,1.0))

reltols = 1.0 ./ 10.0 .^ (1:5)
abstols = reltols#[0.0 for i in eachindex(reltols)]

setups = [Dict(:alg=>SRIW1())
Dict(:alg=>EM(),:dts=>1.0./5.0.^((1:length(reltols)) .+ 1))
]
names = ["SRIW1","EM","RKMil","SRIW1 Fixed"]
wp = WorkPrecisionSet(prob,abstols,reltols,setups;numruns=N,names=names,maxiters=1e7,error_estimate=:l2)
plot(wp)
setups = [Dict(:alg=>EM(),:dts=>1.0./5.0.^((1:length(reltols)) .+ 2))
Dict(:alg=>SRI())
Dict(:alg=>SRIW1())
Dict(:alg=>SRIW2())
Dict(:alg=>SOSRI())
Dict(:alg=>SOSRI2())
]
wp = WorkPrecisionSet(prob,abstols,reltols,setups;numruns=N,maxiters=1e7,error_estimate=:l2)
plot(wp)

Now a scalar wave SDE:

dX{t}=-\left(\frac{1}{10}\right)^{2}\sin\left(X{t}\right)\cos^{3}\left(X{t}\right)dt+\frac{1}{10}\cos^{2}\left(X{t}\right)dW{t},\thinspace\thinspace\thinspace X{0}=\frac{1}{2} Actual Solution: X{t}=\arctan\left(\frac{1}{10}W{t}+\tan\left(X_{0}\right)\right).

prob = prob_sde_wave
prob = remake(prob,tspan=(0.0,1.0))

reltols = 1.0 ./ 10.0 .^ (1:5)
abstols = reltols#[0.0 for i in eachindex(reltols)]

setups = [Dict(:alg=>SRIW1())
Dict(:alg=>EM(),:dts=>1.0./5.0.^((1:length(reltols)) .+ 1))
]
names = ["SRIW1","EM","RKMil","SRIW1 Fixed"]
wp = WorkPrecisionSet(prob,abstols,reltols,setups;numruns=N,names=names,maxiters=1e7,error_estimate=:l2)
plot(wp)

Note that in this last problem, the adaptivity algorithm accurately detects that the error is already low enough, and does not increase the number of steps as the tolerance drops further.

setups = [Dict(:alg=>EM(),:dts=>1.0./5.0.^((1:length(reltols)) .+ 2))
Dict(:alg=>SRI())
Dict(:alg=>SRIW1())
Dict(:alg=>SRIW2())
Dict(:alg=>SOSRI())
Dict(:alg=>SOSRI2())
]
wp = WorkPrecisionSet(prob,abstols,reltols,setups;numruns=N,maxiters=1e7,error_estimate=:l2)
plot(wp)

### Conclusion

The RSwM3 adaptivity algorithm does not appear to have any significant overhead even on problems which do not necessitate adaptive timestepping. The tolerance clearly In addition, the Rossler methods are shown to be orders of magnitude more efficient and should be used whenever applicable. The Oval2 tests show that these results are only magnified as the problem difficulty increases.

using DiffEqBenchmarks
DiffEqBenchmarks.bench_footer(WEAVE_ARGS[:folder],WEAVE_ARGS[:file])

## Appendix

These benchmarks are a part of the DiffEqBenchmarks.jl repository, found at: https://github.com/JuliaDiffEq/DiffEqBenchmarks.jl

To locally run this tutorial, do the following commands:

using DiffEqBenchmarks
DiffEqBenchmarks.weave_file("NonStiffSDE","BasicSDEWorkPrecision.jmd")

Computer Information:

Julia Version 1.1.0
Commit 80516ca202 (2019-01-21 21:24 UTC)
Platform Info:
OS: Linux (x86_64-pc-linux-gnu)
CPU: Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz
WORD_SIZE: 64
LIBM: libopenlibm
LLVM: libLLVM-6.0.1 (ORCJIT, haswell)

Package Information:

Status: /home/crackauckas/.julia/environments/v1.1/Project.toml
[c52e3926-4ff0-5f6e-af25-54175e0327b1] Atom 0.8.5
[bcd4f6db-9728-5f36-b5f7-82caef46ccdb] DelayDiffEq 5.2.0
[bb2cbb15-79fc-5d1e-9bf1-8ae49c7c1650] DiffEqBenchmarks 0.1.0
[459566f4-90b8-5000-8ac3-15dfb0a30def] DiffEqCallbacks 2.5.2
[f3b72e0c-5b89-59e1-b016-84e28bfd966d] DiffEqDevTools 2.8.0
[78ddff82-25fc-5f2b-89aa-309469cbf16f] DiffEqMonteCarlo 0.14.0
[77a26b50-5914-5dd7-bc55-306e6241c503] DiffEqNoiseProcess 3.2.0
[055956cb-9e8b-5191-98cc-73ae4a59e68a] DiffEqPhysics 3.1.0
[a077e3f3-b75c-5d7f-a0c6-6bc4c8ec64a9] DiffEqProblemLibrary 4.1.0
[41bf760c-e81c-5289-8e54-58b1f1f8abe2] DiffEqSensitivity 3.2.2
[0c46a032-eb83-5123-abaf-570d42b7fbaa] DifferentialEquations 6.3.0
[b305315f-e792-5b7a-8f41-49f472929428] Elliptic 0.5.0
[7f56f5a3-f504-529b-bc02-0b1fe5e64312] LSODA 0.4.0
[c030b06c-0b6d-57c2-b091-7029874bd033] ODE 2.4.0
[54ca160b-1b9f-5127-a996-1867f4bc2a2c] ODEInterface 0.4.5
[09606e27-ecf5-54fc-bb29-004bd9f985bf] ODEInterfaceDiffEq 3.2.0
[1dea7af3-3e70-54e6-95c3-0bf5283fa5ed] OrdinaryDiffEq 5.6.0
[2dcacdae-9679-587a-88bb-8b444fb7085b] ParallelDataTransfer 0.5.0
[65888b18-ceab-5e60-b2b9-181511a3b968] ParameterizedFunctions 4.1.1
[91a5bcdd-55d7-5caf-9e0b-520d859cae80] Plots 0.24.0
[d330b81b-6aea-500a-939a-2ce795aea3ee] PyPlot 2.8.1