Large-N vector-model saddles

Consider an N-component real vector with a positive quartic coupling u and real quadratic coefficient r. Define rho as vector length squared divided by N. Rotational symmetry allows integration over directions, leaving a radial integral. The Jacobian is proportional to rho raised to N over two minus one. Taking its logarithm produces an order-N term, so ignoring the measure would miss the leading saddle even though the original action looks simple.

Up to normalization independent of rho, the effective radial function is r rho over two plus u rho squared over four minus one half log rho. The remaining inverse-rho factor is subleading in the exponent at large N. The physical domain is positive rho, inherited from the real vector integral. For positive u the original finite-dimensional integral converges even when r is negative; a positive radial saddle should be selected using that domain rather than an arbitrary root of a quadratic equation.

Setting the derivative of f to zero gives r rho plus u rho squared equal to one. Define a positive effective squared-mass parameter M squared as inverse rho. The same condition becomes M squared equal to r plus u divided by M squared. This resembles the field-theory relation between a self-consistent mass and the component variance. In zero dimensions M is an algebraic parameter, not the frequency of a propagating excitation or a physical correlation length.

The second derivative of f is u over two plus one over twice rho squared, positive for the specified coupling and domain. An isolated nondegenerate saddle therefore has width of order inverse square root N. This does not mean every field-theory fluctuation is small in every limit. Near critical points, infrared modes or competing saddles can invalidate a naive fixed-order expansion, and large N and infinite-volume limits may need to be taken in a specified order.

In a Euclidean O(N) field theory with potential r phi squared over two plus u times phi squared squared over four N, an auxiliary field can make the component integration Gaussian. For positive u a Gaussian auxiliary representation uses an imaginary coupling to phi squared; equivalently, a constraint field has an imaginary-axis contour. The effective action contains a trace logarithm after integrating components. Its saddle can lie away from the original real auxiliary coordinate, requiring a justified contour deformation rather than ordinary unconstrained real minimization.

In a translation-invariant symmetric phase, the field-theory gap equation replaces the zero-dimensional inverse M squared by a momentum integral of one over momentum squared plus M squared. A cutoff and renormalization prescription are required in dimensions where that integral diverges. Broken-symmetry saddles require additional order-parameter terms. These qualifications explain both the power and the scope of large N: it organizes a controlled expansion in a specified model and regime, without turning one algebraic toy equation into an exact solution of every interacting field theory.