如何解决Verlet积分下多球体堆叠碰撞不稳定问题？

1. 增加碰撞迭代次数 #

Verlet积分的碰撞修正需要多次迭代才能让堆叠的球体逐步分离，单次迭代不足以处理多球间的复杂相互作用。在主循环中重复执行碰撞检测与修正逻辑（建议5-10次）：

// 替换原单次checkCollisions()调用
for (int iter = 0; iter < 6; ++iter) {
    checkCollisions();
}

2. 加入碰撞修正松弛因子 #

直接修正重叠量容易导致过度位移，加入松弛因子让修正更平缓，避免连锁抖动：

// 修改碰撞检测函数中的修正代码
float relaxation = 0.75f; // 取值0-1，越小修正越温和
float overlap = ballA->radius + ballB->radius - distance;
sf::Vector2f correction = normal * (overlap / 2.f * relaxation);

ballA->position -= correction;
ballB->position += correction;

3. 避免重复处理碰撞对 #

每个球对只处理一次，防止双向重复修正引发的异常位移：

void checkCollisions() {
    for (size_t i = 0; i < balls.size(); ++i) {
        Ball* ballA = balls[i];
        // 只处理i<j的球对，跳过重复检测
        for (size_t j = i + 1; j < balls.size(); ++j) {
            Ball* ballB = balls[j];
            sf::Vector2f delta = ballA->position - ballB->position;
            float distanceSq = delta.x * delta.x + delta.y * delta.y;
            float minDistance = ballA->radius + ballB->radius;
            
            if (distanceSq < minDistance * minDistance) {
                float distance = sqrt(distanceSq);
                sf::Vector2f normal = delta / distance;
                float overlap = minDistance - distance;
                float relaxation = 0.75f;
                sf::Vector2f correction = normal * (overlap / 2.f * relaxation);

                ballA->position -= correction;
                ballB->position += correction;
            }
        }
    }
}

4. 调整阻尼的应用方式 #

不要直接在位置上乘阻尼，转而在速度（前后位置差）上应用，避免干扰碰撞修正的准确性：

void Ball::update(float dt) {
    sf::Vector2f velocity = position - lastPosition;
    velocity *= 0.97f; // 阻尼因子，接近1更平滑
    lastPosition = position;
    position += velocity + gravity * dt * dt;

    // 边界碰撞修正（加入反弹阻尼）
    if (position.x - radius < 0) {
        position.x = radius;
        lastPosition.x = position.x + velocity.x * 0.8f;
    }
    if (position.x + radius > windowWidth) {
        position.x = windowWidth - radius;
        lastPosition.x = position.x + velocity.x * 0.8f;
    }
    if (position.y - radius < 0) {
        position.y = radius;
        lastPosition.y = position.y + velocity.y * 0.8f;
    }
    if (position.y + radius > windowHeight) {
        position.y = windowHeight - radius;
        lastPosition.y = position.y + velocity.y * 0.8f;
    }
}

5. 新增球体睡眠机制（可选） #

对接近静止的球体暂停更新与碰撞检测，减少微小抖动积累：

// 在Ball类中添加
bool isSleeping() {
    sf::Vector2f velocity = position - lastPosition;
    return (velocity.x * velocity.x + velocity.y * velocity.y) < 0.01f;
}

void update(float dt) {
    if (isSleeping()) return;
    // 原update逻辑
}

// 修改碰撞检测函数
void checkCollisions() {
    for (size_t i = 0; i < balls.size(); ++i) {
        Ball* ballA = balls[i];
        if (ballA->isSleeping()) continue;
        for (size_t j = i + 1; j < balls.size(); ++j) {
            Ball* ballB = balls[j];
            if (ballB->isSleeping()) continue;
            // 原碰撞检测逻辑
        }
    }
}

这些修改的核心是让碰撞修正更平缓充分，避免单次修正过度引发连锁位移，同时优化阻尼与碰撞对的处理逻辑，从根源减少堆叠时的飞散问题。

内容的提问来源于stack exchange，提问作者Lewis Bamford