The following is a brief introduction to the harmonic balance equations.
The general dynamic equation of motion in the time domain is:
(1–1)  | 
where:
Periodic excitation and response may be assumed and enforced by the following equations [1].
(1–2)  | 
(1–3)  | 
(1–4)  | 
where:
NH on
                            HROPT,HBM,NH). | 
| (¯) represents the complex Fourier coefficient of a quantity. | 
Substituting Equation 1–2 –
                    Equation 1–4 into Equation 1–1 and equating coefficients of
                     gives the balance equations describing the dynamics of a system in
                the frequency domain for time harmonic 
:
(1–5)  | 
(1–6)  | 
Collating the time harmonics in Equation 1–6 formulates the set of algebraic equations:
(1–7)  | 
The expanded version of the multiharmonic Equation 1–7 expressed with complex quantities is:
(1–8)  | 
The corresponding expanded version of Equation 1–7 expressed with real quantities is:
(1–9)  | 
where:
(1–10)  | 
(1–11)  | 
(1–12)  | 
and  represents the imaginary part of a complex quantity.
Newton-Raphson Method and Numerical Continuation
Equation 1–9 can be solved to obtain
                the forced response at a given frequency . A Newton-Raphson procedure is used to iterate from an initial
                guess (specified via HBMOPT,UINIT) [5, 6]. The nonlinear equation solver uses a
                trust-region algorithm based on a gradient-descent method [5,8]. 
The residual of the balance equation (Equation 1–9) is given by:
(1–13)  | 
The solver tries to satisfy the condition  iteratively making guesses of 
 using a Jacobian described as [1,
                2]:
(1–14)  | 
A numerical continuation algorithm may be optionally used with the Newton-Raphson type solver. In this case, the frequency is treated as an additional solution variable, and the solver tries to satisfy the following condition:
 
where:
|   | 
Numerical continuation enables automatic determination of the frequency step during solution [7, 8].
You can improve the convergence rate by scaling the augmented multiharmonic solution vector using the HBMOPT,SCAL command. The following scaling strategies are available:
Scale by a scalar value (
UScalwithType= 2 or VALU). All degrees of freedom (DOFs) of the solution vector are scaled with the same value:Scale by a value per solution DOF (
UScalwithType= 1 or APDL™, where:UScalis a 1D APDL array of the same length as the solution vector):where:
is a vector whose components are the inverse of the components of
UScaland ∘ is the Hadamard (term by term) product.
You can also scale the circular frequency  by a scalar value (
FSCAL) using the
                HBMOPT,SCAL command:
Scaling of the multiharmonic solution vector only, of the frequency only, and of both (augmented multiharmonic vector) are supported.
At any given iteration of the nonlinear equation solver, the solution is
                considered to be converged if the norm of the residual is lower than a tolerance
                criterion :
(1–15)  | 
The solver also stops at the kth
                iteration, irrespective of convergence if the norm of the difference between
                consecutive solution estimates is below a tolerance criterion :
(1–16)  | 
where:
 is the solution estimate at the
                    kth iteration, and 
 for continuation.
The solver also stops at the kth
                iteration, irrespective of convergence if the norm of the gradient estimate is below
                a tolerance criterion :
(1–17)  | 
where:
 is the gradient estimate at the
                    kth iteration, 
. 
These conditions on the norm of the difference between consecutive solution
                estimates (step size) , residual norm 
, and gradient norm 
 are stopping criteria. Specify their values as
                
UTOL, RTOL, and
                GTOL on HBMOPT,TRTOL. 
The numerical continuation algorithm stops if one of the following conditions occurs:
The number of iterations exceeds the maximum value (
MAXITERon HBMOPT,NR).The number of successful continuation steps exceeds the maximum value (
MAXSTEPSon HBMOPTCONTTERM).The continuation step size [7,8] is below the minimum value (
DSMINon HBMOPT,CONTSET).The minimum or maximum limit (
FREQMINandFREQMAXon HBMOPT,CONTTERM) of the simulation frequency range is encountered.
 and 
 in Equation 1–13 and
                    Equation 1–14 are obtained using an
                    alternating frequency-time (AFT) procedure.
                At any given iteration step, the solution estimates and parameters 
 are used to calculate the equivalent periodic time domain
                displacements 
 using backward fast Fourier transforms (backward FFTs). Note that
                this is equivalent to enforcing Equation 1–2. The initial velocity at the
                first time step of the period may also be calculated using similar equations.
A transient analysis is performed on a nonlinear submodel with the calculated quantities applied as boundary conditions (BCs) and initial conditions (ICs). The ordinary differential equation representing the dynamics of the nonlinear submodel is given by:
(1–18)  | 
where: , 
, 
 are the mass, damping, and stiffness matrices of the nonlinear
                submodel obtained by assembling the nonlinear elements only.
The internal forces at nonlinear elements are assembled to calculate the internal
                force vector  at each time step. This calculation is repeated for time points
                spanning multiple periods until periodic convergence is achieved or some
                user-specified limit is reached. Then, forward fast
                    Fourier transforms (FFTs) are used to convert back the periodic time
                domain internal forces to individual harmonics 
 in the frequency domain. These harmonic terms are the components
                of the force vector 
 on the right-hand side of Equation 1–8. Similar calculations are
                performed to obtain the Jacobian 
.