allow skipping derivative computations

src/algorithms.jl

@@ -182,6 +182,7 @@ the interval/interval box.
 The algorithm also looks at the sign of the derivative / gradient to see if the range can be
 computed directly. By default, the derivative / gradient is computed using `ForwardDiff.jl`,
 but a custom value can be passed via the `df` keyword argument to [`enclose`](@ref).
+Pass `df = nothing` to avoid computing any derivatives.
 
 ### Examples
 

src/branchandbound.jl

@@ -13,9 +13,11 @@ end
 function _branch_bound(bab::BranchAndBoundEnclosure, f::Function, X::Interval_or_IntervalBox, df;
                        initial=emptyinterval(first(X)),
                        cnt=1)
-    dfX = df(X)
-    range_extrema, flag = _monotonicity_check(f, X, dfX)
-    flag && return hull(range_extrema, initial)
+    if !isnothing(df)
+        dfX = df(X)
+        range_extrema, flag = _monotonicity_check(f, X, dfX)
+        flag && return hull(range_extrema, initial)
+    end
 
     fX = f(X)  # TODO: allow user to choose how to evaluate this (mean value, natural enclosure)
     # if tolerance or maximum number of iteration is met, return current enclosure

test/univariate.jl

@@ -72,4 +72,8 @@ end
     x = enclose(f, dom, BranchAndBoundEnclosure())
     rleft, rright = relative_precision(x, xref)
     @test rleft ≈ 0 && 2.04e-14 ≤ rright ≤ 2.05e-14
+
+    x = enclose(f, dom, BranchAndBoundEnclosure(), df=nothing)
+    rleft, rright = relative_precision(x, xref)
+    @test rleft ≈ 0 && 2.04e-14 ≤ rright ≤ 2.05e-14
 end