Consider the FARIMA model
$$(1-B)^d Y_t = ar_1 X_{t-1} + ... + ar_p X_{t-p}+ma_1 e_{t-1}+...+ma_q e_{t-q}+e_t,$$
where \(e_t\) are the innovations and where \(X_t=(1-B)^d Y_t\).
\(ar_i\), \(i=1, ..., p\), are the AR-coefficients to pass to the
argument ar, \(ma_j\), \(j = 1, ..., q\), are the MA-coefficients
to pass to the argument ma. \(d\) is the fractional differencing
coefficient. The function then returns the coefficients
from the corresponding infinite-order AR-representation
$$-e_t = c_0 Y_t + c_1 Y_{t-1}+c_2 Y_{t-2} + c_3 Y_{t-3} + ...,$$
where \(c_l\), \(l = 0, 1, 2, ...\), are the coefficients. Following this
notation, \(c_0 = -1\) by definition.