Using this linear approximation, we can estimate {{MATH_12}} at {{MATH_13}} as {{MATH_14}} This estimate is an {{CMD_15}}{{CMD_16}}underestimate{{CMD_17}}{{CMD_18}}overestimate{{CMD_19}}{{CMD_20}} because the second derivative of {{MATH_21}} is {{CMD_22}}{{CMD_23}}negative{{CMD_24}}{{CMD_25}}positive{{CMD_26}}{{CMD_27}} on the interval {{MATH_28}}.