Grandient calculation fails when using a parameter-dependent `SciMLOperator`

[albertomercurio]

Describe the bug 🐞

All the examples in the SciMLSensitivity.jl Documentation use a user-defined function for the ODEProblem. I need instead to define a parameter-dependent SciMLOperator (e.g., a MatrixOperator, AddedOperator, ...), but it fails.

Expected behavior

Returning the correct gradient without errors.

Minimal Reproducible Example 👇

using LinearAlgebra
using OrdinaryDiffEq
using SciMLOperators
using Zygote
using Enzyme
using SciMLSensitivity

const T = Float64
const N = 10
const u0 = ones(T, N)
H_tmp = rand(T, N, N)
const H = H_tmp + H_tmp'

coef(a, u, p, t) = - p[1]


function my_f(γ)
    # U = MatrixOperator(-1im * H - γ * Diagonal(H))
    U = ScalarOperator(one(γ), coef) * MatrixOperator(Diagonal(H))
    tspan = (0.0, 1.0)
    # prob = ODEProblem{true}(U, u0, tspan, [γ], sensealg = InterpolatingAdjoint(autojacvec=false))
    prob = ODEProblem{true}(U, u0, tspan, [γ])
    sol = solve(prob, Tsit5())
    return real(sol.u[end][end])
end

my_f(1.9) # 0.25621142049273665

Error & Stacktrace ⚠️

Zygote Error

Zygote.gradient(my_f, 1.9)

ERROR: MethodError: no method matching Float64(::ForwardDiff.Dual{ForwardDiff.Tag{ODEFunction{…}, Float64}, Float64, 10})
The type `Float64` exists, but no method is defined for this combination of argument types when trying to construct it.

Closest candidates are:
  (::Type{T})(::Real, ::RoundingMode) where T<:AbstractFloat
   @ Base rounding.jl:265
  (::Type{T})(::T) where T<:Number
   @ Core boot.jl:900
  Float64(::IrrationalConstants.Invπ)
   @ IrrationalConstants ~/.julia/packages/IrrationalConstants/vp5v4/src/macro.jl:112
  ...

Stacktrace:
  [1] convert(::Type{Float64}, x::ForwardDiff.Dual{ForwardDiff.Tag{ODEFunction{…}, Float64}, Float64, 10})
    @ Base ./number.jl:7
  [2] setproperty!(x::ScalarOperator{Float64, typeof(coef)}, f::Symbol, v::ForwardDiff.Dual{ForwardDiff.Tag{…}, Float64, 10})
    @ Base ./Base.jl:52
  [3] update_coefficients!(L::ScalarOperator{…}, u::Vector{…}, p::Vector{…}, t::Float64; kwargs::@Kwargs{})
    @ SciMLOperators ~/.julia/packages/SciMLOperators/Q5dkx/src/scalar.jl:193
  [4] update_coefficients!(L::ScalarOperator{…}, u::Vector{…}, p::Vector{…}, t::Float64)
    @ SciMLOperators ~/.julia/packages/SciMLOperators/Q5dkx/src/scalar.jl:192
  [5] update_coefficients!(L::SciMLOperators.ScaledOperator{…}, u::Vector{…}, p::Vector{…}, t::Float64)
    @ SciMLOperators ~/.julia/packages/SciMLOperators/Q5dkx/src/basic.jl:251
  [6] (::SciMLOperators.ScaledOperator{…})(du::Vector{…}, u::Vector{…}, p::Vector{…}, t::Float64; kwargs::@Kwargs{})
    @ SciMLOperators ~/.julia/packages/SciMLOperators/Q5dkx/src/interface.jl:116
  [7] (::SciMLOperators.ScaledOperator{…})(du::Vector{…}, u::Vector{…}, p::Vector{…}, t::Float64)
    @ SciMLOperators ~/.julia/packages/SciMLOperators/Q5dkx/src/interface.jl:115
  [8] (::ODEFunction{…})(::Vector{…}, ::Vararg{…})
    @ SciMLBase ~/.julia/packages/SciMLBase/hJh6T/src/scimlfunctions.jl:2355
  [9] (::ODEFunction{…})(::Vector{…}, ::Vararg{…})
    @ SciMLBase ~/.julia/packages/SciMLBase/hJh6T/src/scimlfunctions.jl:2355
 [10] initialize!(integrator::OrdinaryDiffEqCore.ODEIntegrator{…}, cache::OrdinaryDiffEqTsit5.Tsit5Cache{…})
    @ OrdinaryDiffEqTsit5 ~/.julia/packages/OrdinaryDiffEqTsit5/DHYtz/src/tsit_perform_step.jl:175
 [11] __init(prob::ODEProblem{…}, alg::Tsit5{…}, timeseries_init::Tuple{}, ts_init::Tuple{}, ks_init::Tuple{}, recompile::Type{…}; saveat::Vector{…}, tstops::Tuple{}, d_discontinuities::Tuple{}, save_idxs::Nothing, save_everystep::Bool, save_on::Bool, save_start::Bool, save_end::Nothing, callback::Nothing, dense::Bool, calck::Bool, dt::Float64, dtmin::Float64, dtmax::Float64, force_dtmin::Bool, adaptive::Bool, gamma::Rational{…}, abstol::Nothing, reltol::Nothing, qmin::Rational{…}, qmax::Int64, qsteady_min::Int64, qsteady_max::Int64, beta1::Nothing, beta2::Nothing, qoldinit::Rational{…}, controller::Nothing, fullnormalize::Bool, failfactor::Int64, maxiters::Int64, internalnorm::typeof(DiffEqBase.ODE_DEFAULT_NORM), internalopnorm::typeof(opnorm), isoutofdomain::typeof(DiffEqBase.ODE_DEFAULT_ISOUTOFDOMAIN), unstable_check::typeof(DiffEqBase.ODE_DEFAULT_UNSTABLE_CHECK), verbose::Bool, timeseries_errors::Bool, dense_errors::Bool, advance_to_tstop::Bool, stop_at_next_tstop::Bool, initialize_save::Bool, progress::Bool, progress_steps::Int64, progress_name::String, progress_message::typeof(DiffEqBase.ODE_DEFAULT_PROG_MESSAGE), progress_id::Symbol, userdata::Nothing, allow_extrapolation::Bool, initialize_integrator::Bool, alias_u0::Bool, alias_du0::Bool, initializealg::OrdinaryDiffEqCore.DefaultInit, kwargs::@Kwargs{})
    @ OrdinaryDiffEqCore ~/.julia/packages/OrdinaryDiffEqCore/NnA60/src/solve.jl:528
 [12] __init (repeats 5 times)
    @ ~/.julia/packages/OrdinaryDiffEqCore/NnA60/src/solve.jl:11 [inlined]
 [13] #__solve#75
    @ ~/.julia/packages/OrdinaryDiffEqCore/NnA60/src/solve.jl:6 [inlined]
 [14] __solve
    @ ~/.julia/packages/OrdinaryDiffEqCore/NnA60/src/solve.jl:1 [inlined]
 [15] solve_call(_prob::ODEProblem{…}, args::Tsit5{…}; merge_callbacks::Bool, kwargshandle::Nothing, kwargs::@Kwargs{…})
    @ DiffEqBase ~/.julia/packages/DiffEqBase/frOsk/src/solve.jl:612
 [16] solve_call
    @ ~/.julia/packages/DiffEqBase/frOsk/src/solve.jl:569 [inlined]
 [17] #solve_up#53
    @ ~/.julia/packages/DiffEqBase/frOsk/src/solve.jl:1092 [inlined]
 [18] solve_up
    @ ~/.julia/packages/DiffEqBase/frOsk/src/solve.jl:1078 [inlined]
 [19] #solve#51
    @ ~/.julia/packages/DiffEqBase/frOsk/src/solve.jl:1015 [inlined]
 [20] (::SciMLSensitivity.var"#333#342"{…})()
    @ SciMLSensitivity ~/.julia/packages/SciMLSensitivity/XCu1T/src/concrete_solve.jl:1073
 [21] unthunk
    @ ~/.julia/packages/ChainRulesCore/6Pucz/src/tangent_types/thunks.jl:205 [inlined]
 [22] wrap_chainrules_output
    @ ~/.julia/packages/Zygote/NRp5C/src/compiler/chainrules.jl:110 [inlined]
 [23] map
    @ ./tuple.jl:357 [inlined]
 [24] map (repeats 3 times)
    @ ./tuple.jl:358 [inlined]
 [25] wrap_chainrules_output
    @ ~/.julia/packages/Zygote/NRp5C/src/compiler/chainrules.jl:111 [inlined]
 [26] ZBack
    @ ~/.julia/packages/Zygote/NRp5C/src/compiler/chainrules.jl:212 [inlined]
 [27] (::Zygote.var"#294#295"{…})(Δ::ODESolution{…})
    @ Zygote ~/.julia/packages/Zygote/NRp5C/src/lib/lib.jl:206
 [28] (::Zygote.var"#2169#back#296"{…})(Δ::ODESolution{…})
    @ Zygote ~/.julia/packages/ZygoteRules/M4xmc/src/adjoint.jl:72
 [29] #solve#51
    @ ~/.julia/packages/DiffEqBase/frOsk/src/solve.jl:1015 [inlined]
 [30] (::Zygote.Pullback{…})(Δ::ODESolution{…})
    @ Zygote ~/.julia/packages/Zygote/NRp5C/src/compiler/interface2.jl:0
 [31] #294
    @ ~/.julia/packages/Zygote/NRp5C/src/lib/lib.jl:206 [inlined]
 [32] #2169#back
    @ ~/.julia/packages/ZygoteRules/M4xmc/src/adjoint.jl:72 [inlined]
 [33] solve
    @ ~/.julia/packages/DiffEqBase/frOsk/src/solve.jl:1005 [inlined]
 [34] (::Zygote.Pullback{…})(Δ::ODESolution{…})
    @ Zygote ~/.julia/packages/Zygote/NRp5C/src/compiler/interface2.jl:0
 [35] my_f
    @ ~/GitHub/Research/Undef/Autodiff QuantumToolbox/autodiff.jl:135 [inlined]
 [36] (::Zygote.Pullback{Tuple{…}, Tuple{…}})(Δ::Float64)
    @ Zygote ~/.julia/packages/Zygote/NRp5C/src/compiler/interface2.jl:0
 [37] (::Zygote.var"#78#79"{Zygote.Pullback{Tuple{…}, Tuple{…}}})(Δ::Float64)
    @ Zygote ~/.julia/packages/Zygote/NRp5C/src/compiler/interface.jl:91
 [38] gradient(f::Function, args::Float64)
    @ Zygote ~/.julia/packages/Zygote/NRp5C/src/compiler/interface.jl:148
 [39] top-level scope
    @ ~/GitHub/Research/Undef/Autodiff QuantumToolbox/autodiff.jl:143
Some type information was truncated. Use `show(err)` to see complete types.

Enzyme Error

autodiff(Reverse, my_f, Active(1.9))

ERROR: Constant memory is stored (or returned) to a differentiable variable.
As a result, Enzyme cannot provably ensure correctness and throws this error.
This might be due to the use of a constant variable as temporary storage for active memory (https://enzyme.mit.edu/julia/stable/faq/#Runtime-Activity).
If Enzyme should be able to prove this use non-differentable, open an issue!
To work around this issue, either:
 a) rewrite this variable to not be conditionally active (fastest, but requires a code change), or
 b) set the Enzyme mode to turn on runtime activity (e.g. autodiff(set_runtime_activity(Reverse), ...) ). This will maintain correctness, but may slightly reduce performance.
Mismatched activity for:   store atomic {} addrspace(10)* addrspacecast ({}* inttoptr (i64 138286044220400 to {}*) to {} addrspace(10)*), {} addrspace(10)* addrspace(11)* %48 release, align 8, !dbg !410, !tbaa !336, !alias.scope !340, !noalias !343 const val: {} addrspace(10)* addrspacecast ({}* inttoptr (i64 138286044220400 to {}*) to {} addrspace(10)*)
 value=[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0] of type Vector{Float64}
 llvalue={} addrspace(10)* addrspacecast ({}* inttoptr (i64 138286044220400 to {}*) to {} addrspace(10)*)

Stacktrace:
 [1] _
   @ ~/.julia/packages/SciMLBase/hJh6T/src/problems/ode_problems.jl:118
 [2] ODEProblem (repeats 2 times)
   @ ~/.julia/packages/SciMLBase/hJh6T/src/problems/ode_problems.jl:111
 [3] #ODEProblem#318
   @ ~/.julia/packages/SciMLBase/hJh6T/src/problems/ode_problems.jl:194
 [4] ODEProblem
   @ ~/.julia/packages/SciMLBase/hJh6T/src/problems/ode_problems.jl:193
 [5] _
   @ ~/.julia/packages/SciMLBase/hJh6T/src/problems/ode_problems.jl:140
 [6] ODEProblem
   @ ~/.julia/packages/SciMLBase/hJh6T/src/problems/ode_problems.jl:136
 [7] my_f
   @ ~/GitHub/Research/Undef/Autodiff QuantumToolbox/autodiff.jl:134

Stacktrace:
  [1] _
    @ ~/.julia/packages/SciMLBase/hJh6T/src/problems/ode_problems.jl:118 [inlined]
  [2] ODEProblem (repeats 2 times)
    @ ~/.julia/packages/SciMLBase/hJh6T/src/problems/ode_problems.jl:111 [inlined]
  [3] #ODEProblem#318
    @ ~/.julia/packages/SciMLBase/hJh6T/src/problems/ode_problems.jl:194 [inlined]
  [4] ODEProblem
    @ ~/.julia/packages/SciMLBase/hJh6T/src/problems/ode_problems.jl:193 [inlined]
  [5] _
    @ ~/.julia/packages/SciMLBase/hJh6T/src/problems/ode_problems.jl:140 [inlined]
  [6] ODEProblem
    @ ~/.julia/packages/SciMLBase/hJh6T/src/problems/ode_problems.jl:136 [inlined]
  [7] my_f
    @ ~/GitHub/Research/Undef/Autodiff QuantumToolbox/autodiff.jl:134 [inlined]
  [8] diffejulia_my_f_49982wrap
    @ ~/GitHub/Research/Undef/Autodiff QuantumToolbox/autodiff.jl:0
  [9] macro expansion
    @ ~/.julia/packages/Enzyme/BRtTP/src/compiler.jl:8137 [inlined]
 [10] enzyme_call
    @ ~/.julia/packages/Enzyme/BRtTP/src/compiler.jl:7703 [inlined]
 [11] CombinedAdjointThunk
    @ ~/.julia/packages/Enzyme/BRtTP/src/compiler.jl:7476 [inlined]
 [12] autodiff
    @ ~/.julia/packages/Enzyme/BRtTP/src/Enzyme.jl:491 [inlined]
 [13] autodiff
    @ ~/.julia/packages/Enzyme/BRtTP/src/Enzyme.jl:537 [inlined]
 [14] autodiff(mode::ReverseMode{false, false, FFIABI, false, false}, f::typeof(my_f), args::Active{Float64})
    @ Enzyme ~/.julia/packages/Enzyme/BRtTP/src/Enzyme.jl:504
 [15] top-level scope
    @ ~/GitHub/Research/Undef/Autodiff QuantumToolbox/autodiff.jl:147

Environment (please complete the following information):

• Output of using Pkg; Pkg.status()

Status `~/GitHub/Research/Undef/Autodiff QuantumToolbox/Project.toml`
  [6e4b80f9] BenchmarkTools v1.5.0
  [7da242da] Enzyme v0.13.12
  [1dea7af3] OrdinaryDiffEq v6.89.0
  [6c2fb7c5] QuantumToolbox v0.20.0 `~/.julia/dev/QuantumToolbox`
  [1ed8b502] SciMLSensitivity v7.69.0
  [53ae85a6] SciMLStructures v1.5.0
  [e88e6eb3] Zygote v0.6.72

• Output of using Pkg; Pkg.status(; mode = PKGMODE_MANIFEST)

Status `~/GitHub/Research/Undef/Autodiff QuantumToolbox/Manifest.toml`
  [47edcb42] ADTypes v1.9.0
  [621f4979] AbstractFFTs v1.5.0
  [1520ce14] AbstractTrees v0.4.5
  [7d9f7c33] Accessors v0.1.38
  [79e6a3ab] Adapt v4.1.1
  [66dad0bd] AliasTables v1.1.3
  [ec485272] ArnoldiMethod v0.4.0
  [4fba245c] ArrayInterface v7.16.0
  [4c555306] ArrayLayouts v1.10.4
  [a9b6321e] Atomix v0.1.0
  [6e4b80f9] BenchmarkTools v1.5.0
  [e2ed5e7c] Bijections v0.1.9
  [62783981] BitTwiddlingConvenienceFunctions v0.1.6
  [fa961155] CEnum v0.5.0
  [2a0fbf3d] CPUSummary v0.2.6
  [7057c7e9] Cassette v0.3.14
  [082447d4] ChainRules v1.71.0
  [d360d2e6] ChainRulesCore v1.25.0
  [fb6a15b2] CloseOpenIntervals v0.1.13
  [861a8166] Combinatorics v1.0.2
  [38540f10] CommonSolve v0.2.4
  [bbf7d656] CommonSubexpressions v0.3.1
  [f70d9fcc] CommonWorldInvalidations v1.0.0
  [34da2185] Compat v4.16.0
  [b0b7db55] ComponentArrays v0.15.17
  [b152e2b5] CompositeTypes v0.1.4
  [a33af91c] CompositionsBase v0.1.2
  [2569d6c7] ConcreteStructs v0.2.3
  [187b0558] ConstructionBase v1.5.8
  [adafc99b] CpuId v0.3.1
  [9a962f9c] DataAPI v1.16.0
  [864edb3b] DataStructures v0.18.20
  [e2d170a0] DataValueInterfaces v1.0.0
  [2b5f629d] DiffEqBase v6.158.3
  [459566f4] DiffEqCallbacks v4.0.0
  [77a26b50] DiffEqNoiseProcess v5.23.0
  [163ba53b] DiffResults v1.1.0
  [b552c78f] DiffRules v1.15.1
  [a0c0ee7d] DifferentiationInterface v0.6.18
  [b4f34e82] Distances v0.10.12
  [31c24e10] Distributions v0.25.112
  [ffbed154] DocStringExtensions v0.9.3
  [5b8099bc] DomainSets v0.7.14
  [7c1d4256] DynamicPolynomials v0.6.0
  [4e289a0a] EnumX v1.0.4
  [7da242da] Enzyme v0.13.12
  [f151be2c] EnzymeCore v0.8.5
  [d4d017d3] ExponentialUtilities v1.26.1
  [e2ba6199] ExprTools v0.1.10
⌅ [6b7a57c9] Expronicon v0.8.5
  [7a1cc6ca] FFTW v1.8.0
  [7034ab61] FastBroadcast v0.3.5
  [9aa1b823] FastClosures v0.3.2
  [29a986be] FastLapackInterface v2.0.4
  [a4df4552] FastPower v1.1.1
  [1a297f60] FillArrays v1.13.0
  [6a86dc24] FiniteDiff v2.26.0
  [1fa38f19] Format v1.3.7
  [f6369f11] ForwardDiff v0.10.36
  [f62d2435] FunctionProperties v0.1.2
  [069b7b12] FunctionWrappers v1.1.3
  [77dc65aa] FunctionWrappersWrappers v0.1.3
  [d9f16b24] Functors v0.4.12
⌅ [0c68f7d7] GPUArrays v10.3.1
⌅ [46192b85] GPUArraysCore v0.1.6
  [61eb1bfa] GPUCompiler v1.0.1
  [14197337] GenericLinearAlgebra v0.3.14
  [c145ed77] GenericSchur v0.5.4
  [86223c79] Graphs v1.12.0
  [3e5b6fbb] HostCPUFeatures v0.1.17
  [34004b35] HypergeometricFunctions v0.3.24
  [7869d1d1] IRTools v0.4.14
  [615f187c] IfElse v0.1.1
  [40713840] IncompleteLU v0.2.1
  [d25df0c9] Inflate v0.1.5
  [18e54dd8] IntegerMathUtils v0.1.2
  [8197267c] IntervalSets v0.7.10
  [3587e190] InverseFunctions v0.1.17
  [92d709cd] IrrationalConstants v0.2.2
  [82899510] IteratorInterfaceExtensions v1.0.0
  [692b3bcd] JLLWrappers v1.6.1
  [682c06a0] JSON v0.21.4
  [ccbc3e58] JumpProcesses v9.14.0
  [ef3ab10e] KLU v0.6.0
  [63c18a36] KernelAbstractions v0.9.29
  [ba0b0d4f] Krylov v0.9.8
  [929cbde3] LLVM v9.1.3
  [b964fa9f] LaTeXStrings v1.4.0
  [23fbe1c1] Latexify v0.16.5
  [10f19ff3] LayoutPointers v0.1.17
  [5078a376] LazyArrays v2.2.1
  [2d8b4e74] LevyArea v1.0.0
  [87fe0de2] LineSearch v0.1.4
  [d3d80556] LineSearches v7.3.0
  [7ed4a6bd] LinearSolve v2.36.2
  [2ab3a3ac] LogExpFunctions v0.3.28
  [bdcacae8] LoopVectorization v0.12.171
  [d8e11817] MLStyle v0.4.17
  [1914dd2f] MacroTools v0.5.13
  [d125e4d3] ManualMemory v0.1.8
  [bb5d69b7] MaybeInplace v0.1.4
  [e1d29d7a] Missings v1.2.0
  [46d2c3a1] MuladdMacro v0.2.4
  [102ac46a] MultivariatePolynomials v0.5.7
  [d8a4904e] MutableArithmetics v1.5.2
  [d41bc354] NLSolversBase v7.8.3
  [2774e3e8] NLsolve v4.5.1
  [872c559c] NNlib v0.9.24
  [77ba4419] NaNMath v1.0.2
  [8913a72c] NonlinearSolve v3.15.1
  [d8793406] ObjectFile v0.4.2
  [6fe1bfb0] OffsetArrays v1.14.1
  [429524aa] Optim v1.9.4
  [3bd65402] Optimisers v0.3.3
  [bac558e1] OrderedCollections v1.6.3
  [1dea7af3] OrdinaryDiffEq v6.89.0
  [89bda076] OrdinaryDiffEqAdamsBashforthMoulton v1.1.0
  [6ad6398a] OrdinaryDiffEqBDF v1.1.2
  [bbf590c4] OrdinaryDiffEqCore v1.9.0
  [50262376] OrdinaryDiffEqDefault v1.1.0
  [4302a76b] OrdinaryDiffEqDifferentiation v1.1.0
  [9286f039] OrdinaryDiffEqExplicitRK v1.1.0
  [e0540318] OrdinaryDiffEqExponentialRK v1.1.0
  [becaefa8] OrdinaryDiffEqExtrapolation v1.2.1
  [5960d6e9] OrdinaryDiffEqFIRK v1.2.0
  [101fe9f7] OrdinaryDiffEqFeagin v1.1.0
  [d3585ca7] OrdinaryDiffEqFunctionMap v1.1.1
  [d28bc4f8] OrdinaryDiffEqHighOrderRK v1.1.0
  [9f002381] OrdinaryDiffEqIMEXMultistep v1.1.0
  [521117fe] OrdinaryDiffEqLinear v1.1.0
  [1344f307] OrdinaryDiffEqLowOrderRK v1.2.0
  [b0944070] OrdinaryDiffEqLowStorageRK v1.2.1
  [127b3ac7] OrdinaryDiffEqNonlinearSolve v1.2.1
  [c9986a66] OrdinaryDiffEqNordsieck v1.1.0
  [5dd0a6cf] OrdinaryDiffEqPDIRK v1.1.0
  [5b33eab2] OrdinaryDiffEqPRK v1.1.0
  [04162be5] OrdinaryDiffEqQPRK v1.1.0
  [af6ede74] OrdinaryDiffEqRKN v1.1.0
  [43230ef6] OrdinaryDiffEqRosenbrock v1.2.0
  [2d112036] OrdinaryDiffEqSDIRK v1.1.0
  [669c94d9] OrdinaryDiffEqSSPRK v1.2.0
  [e3e12d00] OrdinaryDiffEqStabilizedIRK v1.1.0
  [358294b1] OrdinaryDiffEqStabilizedRK v1.1.0
  [fa646aed] OrdinaryDiffEqSymplecticRK v1.1.0
  [b1df2697] OrdinaryDiffEqTsit5 v1.1.0
  [79d7bb75] OrdinaryDiffEqVerner v1.1.1
  [90014a1f] PDMats v0.11.31
  [65ce6f38] PackageExtensionCompat v1.0.2
  [d96e819e] Parameters v0.12.3
  [69de0a69] Parsers v2.8.1
  [e409e4f3] PoissonRandom v0.4.4
  [f517fe37] Polyester v0.7.16
  [1d0040c9] PolyesterWeave v0.2.2
  [f27b6e38] Polynomials v4.0.11
  [85a6dd25] PositiveFactorizations v0.2.4
  [d236fae5] PreallocationTools v0.4.24
  [aea7be01] PrecompileTools v1.2.1
  [21216c6a] Preferences v1.4.3
  [27ebfcd6] Primes v0.5.6
  [43287f4e] PtrArrays v1.2.1
  [1fd47b50] QuadGK v2.11.1
  [6c2fb7c5] QuantumToolbox v0.20.0 `~/.julia/dev/QuantumToolbox`
  [74087812] Random123 v1.7.0
  [e6cf234a] RandomNumbers v1.6.0
  [c1ae055f] RealDot v0.1.0
  [3cdcf5f2] RecipesBase v1.3.4
  [731186ca] RecursiveArrayTools v3.27.2
  [f2c3362d] RecursiveFactorization v0.2.23
  [189a3867] Reexport v1.2.2
  [ae029012] Requires v1.3.0
  [ae5879a3] ResettableStacks v1.1.1
  [37e2e3b7] ReverseDiff v1.15.3
  [79098fc4] Rmath v0.8.0
  [47965b36] RootedTrees v2.23.1
  [7e49a35a] RuntimeGeneratedFunctions v0.5.13
  [94e857df] SIMDTypes v0.1.0
  [476501e8] SLEEFPirates v0.6.43
  [0bca4576] SciMLBase v2.58.0
  [19f34311] SciMLJacobianOperators v0.1.0
  [c0aeaf25] SciMLOperators v0.3.11
  [1ed8b502] SciMLSensitivity v7.69.0
  [53ae85a6] SciMLStructures v1.5.0
  [6c6a2e73] Scratch v1.2.1
  [efcf1570] Setfield v1.1.1
  [727e6d20] SimpleNonlinearSolve v1.12.3
  [699a6c99] SimpleTraits v0.9.4
  [ce78b400] SimpleUnPack v1.1.0
  [a2af1166] SortingAlgorithms v1.2.1
  [9f842d2f] SparseConnectivityTracer v0.6.8
  [47a9eef4] SparseDiffTools v2.23.0
  [dc90abb0] SparseInverseSubset v0.1.2
  [0a514795] SparseMatrixColorings v0.4.8
  [e56a9233] Sparspak v0.3.9
  [276daf66] SpecialFunctions v2.4.0
  [aedffcd0] Static v1.1.1
  [0d7ed370] StaticArrayInterface v1.8.0
  [90137ffa] StaticArrays v1.9.8
  [1e83bf80] StaticArraysCore v1.4.3
  [10745b16] Statistics v1.11.1
  [82ae8749] StatsAPI v1.7.0
  [2913bbd2] StatsBase v0.34.3
  [4c63d2b9] StatsFuns v1.3.2
  [789caeaf] StochasticDiffEq v6.70.0
  [7792a7ef] StrideArraysCore v0.5.7
  [09ab397b] StructArrays v0.6.18
  [53d494c1] StructIO v0.3.1
  [2efcf032] SymbolicIndexingInterface v0.3.34
  [19f23fe9] SymbolicLimits v0.2.2
  [d1185830] SymbolicUtils v3.7.2
  [0c5d862f] Symbolics v6.17.0
  [3783bdb8] TableTraits v1.0.1
  [bd369af6] Tables v1.12.0
  [8ea1fca8] TermInterface v2.0.0
  [8290d209] ThreadingUtilities v0.5.2
  [a759f4b9] TimerOutputs v0.5.25
  [9f7883ad] Tracker v0.2.35
  [d5829a12] TriangularSolve v0.2.1
  [781d530d] TruncatedStacktraces v1.4.0
  [3a884ed6] UnPack v1.0.2
  [a7c27f48] Unityper v0.1.6
  [013be700] UnsafeAtomics v0.2.1
  [d80eeb9a] UnsafeAtomicsLLVM v0.2.1
  [3d5dd08c] VectorizationBase v0.21.70
  [19fa3120] VertexSafeGraphs v0.2.0
  [e88e6eb3] Zygote v0.6.72
  [700de1a5] ZygoteRules v0.2.5
⌅ [7cc45869] Enzyme_jll v0.0.157+0
  [f5851436] FFTW_jll v3.3.10+1
  [1d5cc7b8] IntelOpenMP_jll v2024.2.1+0
  [dad2f222] LLVMExtra_jll v0.0.34+0
  [856f044c] MKL_jll v2024.2.0+0
  [efe28fd5] OpenSpecFun_jll v0.5.5+0
  [f50d1b31] Rmath_jll v0.5.1+0
  [1317d2d5] oneTBB_jll v2021.12.0+0
  [0dad84c5] ArgTools v1.1.2
  [56f22d72] Artifacts v1.11.0
  [2a0f44e3] Base64 v1.11.0
  [ade2ca70] Dates v1.11.0
  [8ba89e20] Distributed v1.11.0
  [f43a241f] Downloads v1.6.0
  [7b1f6079] FileWatching v1.11.0
  [9fa8497b] Future v1.11.0
  [b77e0a4c] InteractiveUtils v1.11.0
  [4af54fe1] LazyArtifacts v1.11.0
  [b27032c2] LibCURL v0.6.4
  [76f85450] LibGit2 v1.11.0
  [8f399da3] Libdl v1.11.0
  [37e2e46d] LinearAlgebra v1.11.0
  [56ddb016] Logging v1.11.0
  [d6f4376e] Markdown v1.11.0
  [a63ad114] Mmap v1.11.0
  [ca575930] NetworkOptions v1.2.0
  [44cfe95a] Pkg v1.11.0
  [de0858da] Printf v1.11.0
  [9abbd945] Profile v1.11.0
  [9a3f8284] Random v1.11.0
  [ea8e919c] SHA v0.7.0
  [9e88b42a] Serialization v1.11.0
  [1a1011a3] SharedArrays v1.11.0
  [6462fe0b] Sockets v1.11.0
  [2f01184e] SparseArrays v1.11.0
  [4607b0f0] SuiteSparse
  [fa267f1f] TOML v1.0.3
  [a4e569a6] Tar v1.10.0
  [8dfed614] Test v1.11.0
  [cf7118a7] UUIDs v1.11.0
  [4ec0a83e] Unicode v1.11.0
  [e66e0078] CompilerSupportLibraries_jll v1.1.1+0
  [deac9b47] LibCURL_jll v8.6.0+0
  [e37daf67] LibGit2_jll v1.7.2+0
  [29816b5a] LibSSH2_jll v1.11.0+1
  [c8ffd9c3] MbedTLS_jll v2.28.6+0
  [14a3606d] MozillaCACerts_jll v2023.12.12
  [4536629a] OpenBLAS_jll v0.3.27+1
  [05823500] OpenLibm_jll v0.8.1+2
  [bea87d4a] SuiteSparse_jll v7.7.0+0
  [83775a58] Zlib_jll v1.2.13+1
  [8e850b90] libblastrampoline_jll v5.11.0+0
  [8e850ede] nghttp2_jll v1.59.0+0
  [3f19e933] p7zip_jll v17.4.0+2

• Output of versioninfo()

Julia Version 1.11.1
Commit 8f5b7ca12ad (2024-10-16 10:53 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 32 × 13th Gen Intel(R) Core(TM) i9-13900KF
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, alderlake)
Threads: 16 default, 0 interactive, 8 GC (on 32 virtual cores)
Environment:
  JULIA_EDITOR = code
  JULIA_NUM_THREADS = 16

[albertomercurio]

I think that the Zygote problem is related to the update of the ScalarOperator here. Where the struct contains a Float64, but during the differentiation it is trying to convert its field val to a Dual number, thus changing the type of the structure.

I found a possible partial fix for the out-of-place case. If I consider the out-of-place ODEProblem prob = ODEProblem{false}(U, u0, tspan, [γ]), and I update of the ScalarOperator is updated in the out-of-place case

function SciMLOperators.update_coefficients(L::ScalarOperator, u, p, t; kwargs...)
  return ScalarOperator(L.update_func(L.val, u, p, t; kwargs...), L.update_func)
end

instead of the current implementation

function update_coefficients!(L::ScalarOperator, u, p, t; kwargs...)
    L.val = L.update_func(L.val, u, p, t; kwargs...)
    nothing
end

function update_coefficients(L::ScalarOperator, u, p, t; kwargs...)
    update_coefficients!(L, u, p, t; kwargs...)
    L
end

Then it works

Zygote.gradient(my_f, 1.9) # (-0.17161488226273966,)

However, it still fails for the Enzyme case and for the in-place version of Zygote (which would be much more efficient I guess).

I will make a PR to SciMLOperators.jl for fix at least this case.

[albertomercurio]

I found that the in-place version works when using ComplexF64 types

using LinearAlgebra
using SparseArrays
using OrdinaryDiffEq
using SciMLOperators
using Zygote
using Enzyme
using SciMLSensitivity

##

T = ComplexF64
const N = 10
const u0 = ones(T, N)
# H_tmp = rand(T, N, N)
H_tmp = sprand(T, N, N, 0.5)
const H = H_tmp + H_tmp'
const U = ScalarOperator(one(params[1]), coef) * MatrixOperator(Diagonal(H)) + MatrixOperator(Diagonal(H))

coef(a, u, p, t) = - p[1]


function my_f(params)
    tspan = (0.0, 1.0)
    # prob = ODEProblem{true}(U, u0, tspan, [γ], sensealg = InterpolatingAdjoint(autojacvec=false))
    prob = ODEProblem{true}(U, u0, tspan, params)
    sol = solve(prob, Tsit5())
    return real(sol.u[end][end])
end

params = T[1]
my_f(params) # 0.25621142049273665

##

Zygote.gradient(my_f, params)

But I get the following warnings during the differentiation

┌ Warning: Potential performance improvement omitted. ReverseDiffVJP tried and failed in the automated AD choice algorithm. To show the stack trace, set SciMLSensitivity.STACKTRACE_WITH_VJPWARN[] = true. To turn off this printing, add `verbose = false` to the `solve` call.
└ @ SciMLSensitivity ~/.julia/packages/SciMLSensitivity/ME3jV/src/concrete_solve.jl:67

┌ Warning: Reverse-Mode AD VJP choices all failed. Falling back to numerical VJPs
└ @ SciMLSensitivity ~/.julia/packages/SciMLSensitivity/ME3jV/src/concrete_solve.jl:207
(ComplexF64[-1.9743163253371472 + 0.0im],)

First, how can I know the sensealg (and its options) that is automatically chosen?

Then, although I think the problem is related to ReverseDiff.jl, I followed this page and run

tspan = (0.0, 1.0)
# prob = ODEProblem{true}(U, u0, tspan, [γ], sensealg = InterpolatingAdjoint(autojacvec=false))
prob = ODEProblem{true}(U, u0, tspan, params)

u0 = prob.u0
p = prob.p
tmp2 = Enzyme.make_zero(p)
t = prob.tspan[1]
du = zero(u0)

if DiffEqBase.isinplace(prob)
    _f = prob.f
else
    _f = (du, u, p, t) -> (du .= prob.f(u, p, t); nothing)
end

_tmp6 = Enzyme.make_zero(_f)
tmp3 = zero(u0)
tmp4 = zero(u0)
ytmp = zero(u0)
tmp1 = zero(u0)


Enzyme.autodiff(Enzyme.Reverse, Enzyme.Duplicated(_f, _tmp6),
    Enzyme.Const, Enzyme.Duplicated(tmp3, tmp4),
    Enzyme.Duplicated(ytmp, tmp1),
    Enzyme.Duplicated(p, tmp2),
    Enzyme.Const(t))

which returns nothing for every variable

((nothing, nothing, nothing, nothing),)

[albertomercurio]

Now that I better understand how Enzyme.jl works, this is not a bad thing. It means that Enzyme can differentiate the function, right? So I don't understand why I see a warning on ReverseDiffVJP.

I guess that I can try using autojacvec=EnzymeVJP(), but I don't know what is the used sensealg.

Moreover, can I use Enzyme instead of Zygote to directly differentiate the my_f function? What are the advantages and the disadvantages?

[albertomercurio]

If I try sensealg = BacksolveAdjoint(autojacvec=EnzymeVJP()), I get the error

ERROR: MethodError: no method matching augmented_primal(::EnzymeCore.EnzymeRules.RevConfigWidth{…}, ::Const{…}, ::Type{…}, ::Duplicated{…}, ::Duplicated{…}, ::Duplicated{…}, ::Const{…}, ::Const{…})
The function `augmented_primal` exists, but no method is defined for this combination of argument types.

Closest candidates are:
  augmented_primal(::EnzymeCore.EnzymeRules.RevConfig, ::Const{typeof(mul!)}, ::Type{RT}, ::Annotation{<:StridedVecOrMat}, ::Const{<:Union{SparseArrays.AbstractSparseMatrixCSC{Tv, Ti}, SubArray{Tv, 2, <:SparseArrays.AbstractSparseMatrixCSC{Tv, Ti}, Tuple{Base.Slice{Base.OneTo{Int64}}, I}} where I<:(AbstractUnitRange{<:Integer})} where {Tv, Ti}}, ::Annotation{<:StridedVecOrMat}, ::Annotation{<:Number}, ::Annotation{<:Number}) where RT
   @ Enzyme ~/.julia/packages/Enzyme/azJki/src/internal_rules.jl:732
  augmented_primal(::Any, ::Const{typeof(QuadGK.quadgk)}, ::Type{RT}, ::Any, ::Annotation{T}...; kws...) where {RT, T}
   @ QuadGKEnzymeExt ~/.julia/packages/QuadGK/BjmU0/ext/QuadGKEnzymeExt.jl:6
  augmented_primal(::Any, ::Const{typeof(NNlib._dropout!)}, ::Type{RT}, ::Any, ::OutType, ::Any, ::Any, ::Any) where {OutType, RT}
   @ NNlibEnzymeCoreExt ~/.julia/packages/NNlib/CkJqS/ext/NNlibEnzymeCoreExt/NNlibEnzymeCoreExt.jl:318

[ChrisRackauckas]

It looks like you're hitting a missing rule in enzyme for sparse mul! EnzymeAD/Enzyme.jl#2013

[wsmoses]

what's the full types of the method match failure? And what version of enzyme are you using (it is the latest)?

[albertomercurio]

LoadError: MethodError: no method matching augmented_primal(::EnzymeCore.EnzymeRules.RevConfigWidth{1, false, false, (false, false, false, false, false, false), false}, ::Const{typeof(mul!)}, ::Type{Const{Vector{ComplexF64}}}, ::Duplicated{Vector{ComplexF64}}, ::Duplicated{SparseMatrixCSC{ComplexF64, Int64}}, ::Duplicated{Vector{ComplexF64}}, ::Const{Bool}, ::Const{Bool})
The function `augmented_primal` exists, but no method is defined for this combination of argument types.

Closest candidates are:
  augmented_primal(::EnzymeCore.EnzymeRules.RevConfig, ::Const{typeof(mul!)}, ::Type{RT}, ::Annotation{<:StridedVecOrMat}, ::Const{<:Union{SparseArrays.AbstractSparseMatrixCSC{Tv, Ti}, SubArray{Tv, 2, <:SparseArrays.AbstractSparseMatrixCSC{Tv, Ti}, Tuple{Base.Slice{Base.OneTo{Int64}}, I}} where I<:(AbstractUnitRange{<:Integer})} where {Tv, Ti}}, ::Annotation{<:StridedVecOrMat}, ::Annotation{<:Number}, ::Annotation{<:Number}) where RT
   @ Enzyme ~/.julia/packages/Enzyme/azJki/src/internal_rules.jl:732
  augmented_primal(::Any, ::Const{typeof(QuadGK.quadgk)}, ::Type{RT}, ::Any, ::Annotation{T}...; kws...) where {RT, T}
   @ QuadGKEnzymeExt ~/.julia/packages/QuadGK/BjmU0/ext/QuadGKEnzymeExt.jl:6
  augmented_primal(::Any, ::Const{typeof(NNlib._dropout!)}, ::Type{RT}, ::Any, ::OutType, ::Any, ::Any, ::Any) where {OutType, RT}
   @ NNlibEnzymeCoreExt ~/.julia/packages/NNlib/CkJqS/ext/NNlibEnzymeCoreExt/NNlibEnzymeCoreExt.jl:318`

And this is my versioninfo

Status `~/GitHub/Research/Undef/Autodiff QuantumToolbox/Project.toml`
  [6e4b80f9] BenchmarkTools v1.5.0
  [13f3f980] CairoMakie v0.12.16
  [b0b7db55] ComponentArrays v0.15.19
  [7da242da] Enzyme v0.13.15
  [f6369f11] ForwardDiff v0.10.38
  [1dea7af3] OrdinaryDiffEq v6.90.1
  [33c8b6b6] ProgressLogging v0.1.4
  [6c2fb7c5] QuantumToolbox v0.21.5 `~/.julia/dev/QuantumToolbox`
  [731186ca] RecursiveArrayTools v3.27.3
  [37e2e3b7] ReverseDiff v1.15.3
⌃ [0bca4576] SciMLBase v2.61.0
⌃ [1ed8b502] SciMLSensitivity v7.71.1
  [5d786b92] TerminalLoggers v0.1.7
  [e88e6eb3] Zygote v0.6.73
Info Packages marked with ⌃ have new versions available and may be upgradable.

[albertomercurio]

Hello, any updates on this?

I get the warning

┌ Warning: Potential performance improvement omitted. ReverseDiffVJP tried and failed in the automated AD choice algorithm. To show the stack trace, set SciMLSensitivity.STACKTRACE_WITH_VJPWARN[] = true. To turn off this printing, add `verbose = false` to the `solve` call.
└ @ SciMLSensitivity ~/.julia/packages/SciMLSensitivity/7UEpc/src/concrete_solve.jl:67

┌ Warning: Reverse-Mode AD VJP choices all failed. Falling back to numerical VJPs
└ @ SciMLSensitivity ~/.julia/packages/SciMLSensitivity/7UEpc/src/concrete_solve.jl:207
(ComplexF64[-0.06308827054069974 + 0.0im],)

with the following code

using LinearAlgebra
using SparseArrays
using OrdinaryDiffEq
using SciMLOperators
using Zygote
using Enzyme
using SciMLSensitivity

# %%

const T = ComplexF64
const N = 10
const u0 = ones(T, N)
H_tmp = rand(T, N, N)
const H = H_tmp + H_tmp'

coef(a, u, p, t) = - p[1]


function my_f(params)
    γ = params[1]

    # U = MatrixOperator(-1im * H - γ * Diagonal(H))
    U = ScalarOperator(one(γ), coef) * MatrixOperator(Diagonal(H))
    tspan = (0.0, 1.0)
    # prob = ODEProblem{true}(U, u0, tspan, [γ], sensealg = InterpolatingAdjoint(autojacvec=false))
    prob = ODEProblem{true}(U, u0, tspan, params)
    sol = solve(prob, Tsit5())
    return real(sol.u[end][end])
end

params = [1.9 + 0im]
my_f(params) # 0.25621142049273665

# %%

Zygote.gradient(my_f, params)

But if I use the sensealg = BacksolveAdjoint(autojacvec=EnzymeVJP()) argument, I don't get any warning and everything seems to work. Is there a way to know what is the sensealg used by default?

Furthermore, differentiating it using Enzyme returns errors

Enzyme.gradient(Enzyme.Reverse, my_f, params)

ERROR: AssertionError: Base.allocatedinline(actualRetType) returns false: actualRetType = Any, rettype = Active{Any}
Stacktrace:
  [1] create_abi_wrapper(enzymefn::LLVM.Function, TT::Type, rettype::Type, actualRetType::Type, Mode::Enzyme.API.CDerivativeMode, augmented::Nothing, width::Int64, returnPrimal::Bool, shadow_init::Bool, world::UInt64, interp::Enzyme.Compiler.Interpreter.EnzymeInterpreter{…})
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/QsaeA/src/compiler.jl:1948
  [2] enzyme!(job::GPUCompiler.CompilerJob{…}, mod::LLVM.Module, primalf::LLVM.Function, TT::Type, mode::Enzyme.API.CDerivativeMode, width::Int64, parallel::Bool, actualRetType::Type, wrap::Bool, modifiedBetween::NTuple{…} where N, returnPrimal::Bool, expectedTapeType::Type, loweredArgs::Set{…}, boxedArgs::Set{…})
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/QsaeA/src/compiler.jl:1728
  [3] codegen(output::Symbol, job::GPUCompiler.CompilerJob{…}; libraries::Bool, deferred_codegen::Bool, optimize::Bool, toplevel::Bool, strip::Bool, validate::Bool, only_entry::Bool, parent_job::Nothing)
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/QsaeA/src/compiler.jl:4550
  [4] codegen
    @ ~/.julia/packages/Enzyme/QsaeA/src/compiler.jl:3353 [inlined]
  [5] _thunk(job::GPUCompiler.CompilerJob{Enzyme.Compiler.EnzymeTarget, Enzyme.Compiler.EnzymeCompilerParams}, postopt::Bool)
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/QsaeA/src/compiler.jl:5410
  [6] _thunk
    @ ~/.julia/packages/Enzyme/QsaeA/src/compiler.jl:5410 [inlined]
  [7] cached_compilation
    @ ~/.julia/packages/Enzyme/QsaeA/src/compiler.jl:5462 [inlined]
  [8] thunkbase(mi::Core.MethodInstance, World::UInt64, FA::Type{…}, A::Type{…}, TT::Type, Mode::Enzyme.API.CDerivativeMode, width::Int64, ModifiedBetween::NTuple{…} where N, ReturnPrimal::Bool, ShadowInit::Bool, ABI::Type, ErrIfFuncWritten::Bool, RuntimeActivity::Bool, edges::Vector{…})
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/QsaeA/src/compiler.jl:5573
  [9] thunk_generator(world::UInt64, source::LineNumberNode, FA::Type, A::Type, TT::Type, Mode::Enzyme.API.CDerivativeMode, Width::Int64, ModifiedBetween::NTuple{…} where N, ReturnPrimal::Bool, ShadowInit::Bool, ABI::Type, ErrIfFuncWritten::Bool, RuntimeActivity::Bool, self::Any, fakeworld::Any, fa::Type, a::Type, tt::Type, mode::Type, width::Type, modifiedbetween::Type, returnprimal::Type, shadowinit::Type, abi::Type, erriffuncwritten::Type, runtimeactivity::Type)
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/QsaeA/src/compiler.jl:5758
 [10] autodiff
    @ ~/.julia/packages/Enzyme/QsaeA/src/Enzyme.jl:485 [inlined]
 [11] autodiff
    @ ~/.julia/packages/Enzyme/QsaeA/src/Enzyme.jl:524 [inlined]
 [12] macro expansion
    @ ~/.julia/packages/Enzyme/QsaeA/src/sugar.jl:324 [inlined]
 [13] gradient(::ReverseMode{false, false, FFIABI, false, false}, ::typeof(my_f), ::Vector{ComplexF64})
    @ Enzyme ~/.julia/packages/Enzyme/QsaeA/src/sugar.jl:263
 [14] top-level scope
    @ ~/GitHub/Research/Undef/Autodiff QuantumToolbox/autodiff_scimloperators.jl:41
Some type information was truncated. Use `show(err)` to see complete types.

[albertomercurio]

So apparently the error is partially fixed, at least for ComplexF64. But I don't know why.

However, the gradient calculation with Enzyme doesn't work.

[ChrisRackauckas]

Can you try to isolate that? I don't think we've tested operators much with Enzyme, and if Enzyme is having issues with the differentiation of the operator code then it should be able to be isolated to not have any ODE solver at all in there, and that would help Billy get a clean situation to debug.