Descriptions may be stale. Consult the linked Lean sources.
Full-Range Cohypomonotone
Inclusions
We seek a zero of on a
finite-dimensional real inner product space . The map is -Lipschitz with , the set-valued operator is monotone, and
is -cohypomonotone with . A given point lies within distance of a solution. The oracle provides
unbiased estimates of with
bounded variance using independent, identically distributed samples, or
exact resolvent values of at
positive step sizes. For , the goal is a
measurable pair
with for
every realization and ,
averaging over oracle and algorithm randomness.
In terms of a point and
a vector , the inclusion is
to find The algorithm receives
, the measurable sample space
, the starting point , and the parameters , where bounds the root mean squared
oracle error.
Our assumptions:
Assumption 1 (Operator inequalities). For all , For every
and , write and . Monotonicity of and cohypomonotonicity of mean that
Assumption 2 (Initial distance). There exists a
solution of the inclusion
such that .
Assumption 3 (Stochastic evaluations and resolvent
access). At call , a query returns , where is jointly
measurable. The samples are independent random
variables with common law on
, and the sequence is
independent of the algorithm’s randomness. For each and , the vector is Bochner integrable, the
squared error
is integrable, and Alternatively, a query returns
. For every and , this reply satisfies the
resolvent identity Both query types draw on the same
call budget.
After the allotted calls, the output must be
measurable and satisfy for every realization. Under the joint law of the samples and the algorithm’s
randomness, the residual norm must be integrable and satisfy