Let π:P→M\pi:P\to M be a over a MM, where GG is a with g\mathfrak g.

The of PP is the

ad⁡(P):=P×Ad⁡g  ⟶  M,\operatorname{ad}(P):=P\times_{\operatorname{Ad}}\mathfrak g \;\longrightarrow\; M,

where GG acts on g\mathfrak g by the adjoint representation Ad⁡\operatorname{Ad}.

For k≥0k\ge0, a g\mathfrak g-valued ω∈Ωk(P;g)\omega\in\Omega^k(P;\mathfrak g) is called tensorial of type Ad⁡\operatorname{Ad} if it satisfies:

  1. Horizontality: ωp(v1,…,vk)=0\omega_p(v_1,\dots,v_k)=0 whenever at least one vi∈ker⁡(dπp)v_i\in\ker(d\pi_p) (i.e., a vertical tangent vector).
  2. Equivariance: for every g∈Gg\in G,
    (Rg)∗ω=Ad⁡(g−1) ω,(R_g)^*\omega = \operatorname{Ad}(g^{-1})\,\omega,
    where Rg:P→PR_g:P\to P is the right action.

Write Ωtensk(P;g)\Omega^k_{\mathrm{tens}}(P;\mathfrak g) for the vector space of such tensorial forms.

There is a natural vector space isomorphism

Ωtensk(P;g)  ≅  Ωk ⁣(M;ad⁡(P)).\Omega^k_{\mathrm{tens}}(P;\mathfrak g)\;\cong\;\Omega^k\!\bigl(M;\operatorname{ad}(P)\bigr).
Explicit correspondence

More explicitly:

  • Given ω∈Ωtensk(P;g)\omega\in\Omega^k_{\mathrm{tens}}(P;\mathfrak g), define α∈Ωk(M;ad⁡(P))\alpha\in\Omega^k(M;\operatorname{ad}(P)) by
    αx(u1,…,uk)  :=  [p,  ωp(u~1,…,u~k)]∈ad⁡(P)x,\alpha_x(u_1,\dots,u_k)\;:=\;\bigl[p,\;\omega_p(\tilde u_1,\dots,\tilde u_k)\bigr]\in \operatorname{ad}(P)_x,
    where p∈Pp\in P satisfies π(p)=x\pi(p)=x, and u~i∈TpP\tilde u_i\in T_pP are any tangent vectors with dπp(u~i)=uid\pi_p(\tilde u_i)=u_i. This is well-defined because vertical components do not affect ω\omega (horizontality) and changing pp in the fiber changes ω\omega by Ad⁡(g−1)\operatorname{Ad}(g^{-1}) (equivariance), exactly matching the associated-bundle identification.
  • Conversely, given α∈Ωk(M;ad⁡(P))\alpha\in\Omega^k(M;\operatorname{ad}(P)), define ω∈Ωk(P;g)\omega\in\Omega^k(P;\mathfrak g) by requiring that for p∈Pp\in P and vi∈TpPv_i\in T_pP,
    απ(p)(dπpv1,…,dπpvk)=[p,  ωp(v1,…,vk)]∈ad⁡(P)π(p).\alpha_{\pi(p)}(d\pi_p v_1,\dots,d\pi_p v_k)=\bigl[p,\;\omega_p(v_1,\dots,v_k)\bigr]\in\operatorname{ad}(P)_{\pi(p)}.
    This forces ω\omega to be horizontal, and the associated-bundle relation forces Ad⁡\operatorname{Ad}-equivariance.

These constructions are inverse to each other and are natural with respect to bundle morphisms covering the identity on MM.

A key application is that the of a is tensorial of type Ad⁡\operatorname{Ad}, hence canonically corresponds to an ad⁡(P)\operatorname{ad}(P)-valued 2-form on the base.

Examples
  1. Curvature descends to the base. If Θ\Theta is a principal connection on PP with curvature FΘ∈Ω2(P;g)F_\Theta\in\Omega^2(P;\mathfrak g), then FΘF_\Theta is horizontal and Ad⁡\operatorname{Ad}-equivariant. The theorem identifies FΘF_\Theta with a unique element of Ω2(M;ad⁡(P))\Omega^2(M;\operatorname{ad}(P)).
  1. Trivial bundle case. On the P=M×GP=M\times G, the adjoint bundle is canonically isomorphic to M×gM\times\mathfrak g. Under this identification, a tensorial ω\omega corresponds to an ordinary g\mathfrak g-valued form on MM (written in a chosen ), and the correspondence is given by pullback along the section and associated-bundle identification.
  1. Scalar “basic” forms as a special case. If one replaces g\mathfrak g by a trivial GG-representation (so equivariance becomes GG-invariance), then horizontal GG-invariant forms on PP correspond exactly to ordinary differential forms on MM via pullback along π\pi.