Statement

Let f:MNf:M\to N be a between finite-dimensional without boundary, and let SNS\subseteq N be an . If ff is , then f1(S)f^{-1}(S) is an embedded submanifold of MM. For every xf1(S)x\in f^{-1}(S),

Txf1(S)=(dfx)1(Tf(x)S),T_xf^{-1}(S)=(df_x)^{-1}(T_{f(x)}S),

and

codimMf1(S)=codimNS.\operatorname{codim}_M f^{-1}(S)=\operatorname{codim}_N S.

Thus the inverse image has the codimension predicted by the normal directions to SS.

Proof idea

Choose local coordinates on NN in which SS is the zero set of a submersion hh to Rk\mathbb R^k, where k=codimNSk=\operatorname{codim}_N S. Transversality says that d(hf)xd(h\circ f)_x is surjective along f1(S)f^{-1}(S). The regular-level-set theorem applied to hfh\circ f gives the claimed smooth structure, codimension, and tangent-space kernel. See Hirsch, Chapter 3.

Special cases and examples

When S={y}S=\{y\}, transversality is exactly the assertion that yy is a , so this theorem recovers the regular-value theorem. Every submersion is transverse to every embedded submanifold; hence the inverse image of SS under a submersion has the same codimension as SS.

A constant map whose value lies in a positive-codimension SS is not transverse, and the theorem makes no regularity claim about its inverse image.

Scope

If f1(S)f^{-1}(S) is empty, it is an embedded submanifold and the tangent statement is vacuous. Manifolds with boundary or corners need additional conditions controlling boundary strata. The theorem also does not assert that an arbitrary nontransverse inverse image is singular; it says only that transversality is a sufficient regularity hypothesis.

References
  1. Morris W. Hirsch, Differential Topology, Springer, 1976. DOI record. Relevant: Chapter 3, transversality and inverse images.
  2. John M. Lee, Introduction to Smooth Manifolds, 2nd ed., Springer, 2012. DOI record. Relevant: Chapter 6, the transversality theorem.