A stochastic integral defined via symmetric (midpoint) Riemann sum approximations for continuous semimartingales or rough paths, yielding integrals that transform under the ordinary chain rule and coincide with classical limits under smooth change of variables; often used when stochastic calculus is treated as a limit of physical smoothing.

Principle

Symmetric sampling of increments (midpoint rule) produces an integral formally consistent with ordinary calculus; conversion to Itô adds a compensating drift term equal to half the integrand's derivative times the integrator's quadratic variation.

Demonstration

For smooth σ(t,x) and Brownian motion W_t, the Stratonovich integral ∫_0^T σ(t,X_t) ∘ dW_t equals the limit of midpoint Riemann sums and relates to the Itô integral by ∫ σ ∘ dW = ∫ σ dW + ½ ∫ ∂_xσ · d[W,X].

Misapplication

Using Stratonovich calculus without checking regularization assumptions — e.g., treating a limit of physically smoothed noises as Stratonovich when the noise lacks the necessary temporal correlation structure.

Consequence

Equations written in Stratonovich form follow classical calculus rules, making coordinate changes and geometric interpretations (e.g., stochastic differential geometry) more direct; conversion to Itô introduces explicit drift corrections.

Reversal

Itô integral employs nonanticipative left-point sampling and yields martingale properties and an explicit quadratic-variation correction in the chain rule, differing in interpretation and analytical properties.

Boundary

Appropriate for integrators with continuous semimartingale structure or for limits of smoothed noises; not generally equivalent to pathwise integrals for highly irregular signals without a chosen rough‑path lift.

Semantic Tension

Tension arises with the Itô integral and with purely pathwise definitions (Young, rough path): the same SDE written in different conventions changes drift interpretation and statistical estimation.

Synthesis

A midpoint‑sampled stochastic integral consistent with ordinary change‑of‑variable calculus; when compared to Itô it differs by a half‑quadratic‑variation drift that encodes the sampling convention.