Kozai (1959): Motion of a Close Earth Satellite

Source

Kozai, Y., “The Motion of a Close Earth Satellite,” The Astronomical Journal, Vol. 64, No. 1274, pp. 367-377, November 1959. DOI: 10.1086/107957. Published in the same issue as Brouwer’s companion paper (pp. 378-396).

Download PDF · DOI: 10.1086/107957

Why This Matters for SGP4

Kozai and Brouwer published simultaneously in the same issue of the Astronomical Journal, solving the same problem — second-order perturbations of close earth satellites in an oblate gravitational field — using closely related but independently developed methods. Both derive secular rates, long-period corrections, and short-period corrections for all six orbital elements due to $J_2$, $J_3$, and $J_4$.

The two papers diverged in their operational adoption:

Model	Theoretical Basis	Drag Coupling	Status
SGP	Kozai (1959)	Added later	Superseded
SGP4	Brouwer (1959)	Lane & Hoots	Operational standard
SGP8	Hoots (1981)	Direct on Brouwer mean elements	Available, not operational

SGP4 won the adoption competition not because Brouwer’s theory was more accurate — the secular rates agree to machine precision — but because Brouwer’s von Zeipel framework proved more natural for coupling atmospheric drag. Lane & Hoots built STR#2 on top of Brouwer, and that became the basis of the SGP4 FORTRAN in STR#3.

Understanding Kozai’s paper still matters. The SGP4 code contains “Kozai mean motion” variables because drag is first applied to Kozai mean elements, then converted to Brouwer mean elements for the gravitational perturbation calculations. The conversion between the two mean-element definitions appears in the SGP4 initialization code.

The Method of Averaging

Where Brouwer constructs an explicit generating function via von Zeipel’s canonical transformation, Kozai works directly with the averaged equations of motion. The two approaches are mathematically equivalent to the relevant order but differ in formalism.

The procedure:

1. Express the geopotential disturbing function $R$ in Delaunay canonical variables:

$L = \sqrt{\mu a}, \quad G = \sqrt{\mu a(1-e^2)}, \quad H = G \cos i$

with conjugate angles $l$ (mean anomaly), $g$ (argument of perigee), $h$ (node).

2. Average $R$ over the fast variable $l$ to eliminate short-period terms.
3. Solve the averaged system for secular and long-period motion.
4. Recover short-period perturbations from the difference between the full and averaged disturbing functions.

No Long-Period Terms in the Semi-Major Axis

Section 3 contains an elegant proof that first-order long-period perturbations in $L$ (and therefore in $a$) vanish identically. Since $L$ is a canonical momentum conjugate to $l$, and the averaging eliminates $l$-dependent terms from $R$, the averaged $\dot{L}$ contains no $g$-dependent terms at first order.

This theorem — the mean semi-major axis changes only secularly, never with long-period oscillations — is one of the results that makes analytical propagation practical. Without it, the secular drag fit that SGP4 relies on would be contaminated by gravitational long-period energy fluctuations.

Secular Rates

The first-order secular rates for the node and argument of perigee (Eqs. 13-14):

$\dot{\Omega} = -\frac{3}{2} n_0 J_2 \frac{a_E^2}{p^2} \cos i$

$\dot{\omega} = \frac{3}{4} n_0 J_2 \frac{a_E^2}{p^2} (5\cos^2 i - 1)$

where $p = a(1-e^2)$ is the semi-latus rectum, $a_E$ is the equatorial radius, and $n_0$ is the unperturbed mean motion.

These are the same expressions Brouwer derives by a different route. The numerical agreement is exact — both reduce to the same function of the orbital elements and $J_2$.

The second-order ($J_2^2$) secular contributions to all three rates (node, perigee, mean anomaly) are also given.

Long-Period Perturbations

The first-order long-period terms arise from $J_3$ and $J_4$ and depend on $g$ (argument of perigee). They vary on the timescale of apsidal precession — months to years for typical LEO orbits.

The $J_3$ long-period term in eccentricity:

$\delta e_{lp} \sim J_3 \frac{a_E^3}{p^3} \sin i \sin g$

represents the pear-shaped asymmetry of the geopotential. This term and its companions in the other elements are the same terms that Hoots (1981) exhibits explicitly in his reformulation — and the same terms that contain the singularities at $e = 0$ and $i = 0$.

The Critical Inclination

The secular rate of the argument of perigee vanishes when $5\cos^2 i - 1 = 0$, yielding the critical inclination:

$i \approx 63.4\degree \quad \text{or} \quad i \approx 116.6\degree$

At these inclinations, the argument of perigee freezes — perigee and apogee stay fixed relative to the Earth’s equator. Kozai identifies this mathematical property without discussing its engineering implications. Molniya orbits, which deliberately exploit this effect to keep apogee over the northern hemisphere, would not fly until 1965.

In SGP4, the critical inclination requires special-case handling to avoid division by zero in the long-period perturbation terms.

Singularities and Their Later Resolution

Section 5 addresses the coordinate singularities that appear when $e \to 0$ (argument of perigee undefined) and $i \to 0$ (ascending node undefined). Kozai introduces combined angle variables — $l + g$ for the near-circular case, $g + h$ for the near-equatorial case — to absorb the singularities.

This treatment anticipates Lyddane’s (1963) more systematic solution for Brouwer’s theory. Lyddane uses Poincare canonical variables to produce a complete, non-singular reformulation that is directly implemented in the SGP4 FORTRAN. Kozai’s approach works but does not carry through a full re-derivation in the new variables.

Kozai Mean vs. Brouwer Mean Elements

TLEs encode Brouwer mean elements, not Kozai mean elements. The distinction matters: Brouwer means have both short-period and long-period terms removed, while Kozai means have only short-period terms removed. Using Kozai-mean propagation formulas with a Brouwer-mean TLE (or vice versa) introduces systematic error because the perturbation corrections no longer match what was removed during element fitting. The SGP4 initialization code converts between the two definitions internally.

The Brouwer-Kozai Relationship

Kozai’s acknowledgment at the end of the paper — “The author expresses his deep appreciation to Dr. Dirk Brouwer” — reflects a close working relationship at Yale Observatory. The two attacked the same problem in parallel, with enough communication to publish in the same journal issue but enough independence that their methods differ in formalism while converging on identical physical results.

The community eventually needed both: Brouwer’s framework for SGP4 propagation, and Kozai’s mean-element definitions for the intermediate drag coupling step that links atmospheric decay to the gravitational perturbation theory.

---

Brouwer (1959): Mean Element Theory The companion paper from the same AJ issue -- the direct basis of SGP4

Lyddane (1963): Singularity Fix Completes the singularity treatment Kozai began in Section 5

Hoots (1981): Brouwer Reformulation Exhibits the long-period terms where Kozai's singularities appear

SGP4 Theory Overview Full document lineage showing where Kozai and Brouwer diverge